3 receptors in rat brain using factor analysis corrected [18F]Fallypride images

NeuroImage 62 (2012) 1455–1468

Contents lists available at SciVerse ScienceDirect

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

Quantification of dopamine D2/3 receptors in rat brain using factor analysis corrected [ 18F]Fallypride images Philippe Millet a,⁎, Marcelle Moulin-Sallanon a, b, Benjamin B. Tournier a, c, Noé Dumas a, Yves Charnay d, Vicente Ibáñez a, Nathalie Ginovart a, c a

Clinical Neurophysiology and Neuroimaging Unit, Division of Neuropsychiatry, Department of Psychiatry; University Hospitals of Geneva, Switzerland INSERM, Unit 1039, J. Fourier University, La Tronche, France Department of Psychiatry, University of Geneva, Switzerland d Morphology Unit; Division of Neuropsychiatry, Division of Neuropsychiatry, Department of Psychiatry, University Hospitals of Geneva, Switzerland b c

a r t i c l e

i n f o

Article history: Accepted 26 May 2012 Available online 1 June 2012 Keywords: Factor analysis Fallypride Dopamine D2/3 receptors Kinetic modeling PET

a b s t r a c t The goal of this work is to quantify the binding parameters of [18F]Fallypride in the striatal and extrastriatal regions of the rat brain using factor analysis (FA) to correct small animal PET kinetic imaging for spillover defluorination radioactivity. Eleven rats were employed for YAP-(S)PET acquisitions and metabolite studies. All kinetic parameters including B′max and KdVR were estimated with a three-tissue compartment sevenparameter model (3T-7k) on the basis of all the FA-corrected data from the multi-injection protocol. Binding potential (BPND) was calculated with Logan's graphical analysis taking cerebellum as the reference region and using the first injection raw (BPND-RAW) and FA-corrected (BPND-FA) data. Three distinct factors corresponding to free + non-specific binding, specific binding and skull and gland accumulation were recovered from FA with their corresponding spatial distributions. The resulting reconstructed images without skull and gland accumulation were improved to provide a better contrast between specific and non-specific regions. Very bad fits were obtained when using time–activity curves (TACs) calculated from the raw [18F]Fallypride data, whereas all TACs were well fitted by the 3T-7k model after FA correction. FA-corrected data enables the cerebellar region to be used as reference for the Logan approach. The magnitude of the BPND-FA values was increased from 21% to 108% across regions and the rank order of BPND-FA values (Cx b Hip b MB ≈ Thal b VST b DST) matched those of B′max values. This [18F]Fallypride study in rats shows that all brain regions are contaminated by skull and gland radioactivity accumulation. We show that FA is a very effective method of correcting kinetic data for spillover activity. Moreover, the approach presented here with [ 18F]Fallypride data can be extended to other radioligands and also to human data which can be highly distorted by radiodefluorination as shown in the literature. © 2012 Elsevier Inc. All rights reserved.

Introduction During the last decade, D2/3 receptor imaging using the positron emission tomography (PET) radioligand [ 18F]Fallypride has engendered considerable interest in clinical and pharmacological research. A number of PET studies have demonstrated the efficiency of the high-affinity D2/3 antagonist [ 18F]Fallypride for imaging receptors in the striatal and extrastriatal regions in human and nonhuman primates (Christian et al., 2000; Mukherjee et al., 2002, 2004). More recent studies have shown that [ 18F]Fallypride was suitable to measure variations in endogenous dopamine concentrations caused by drug challenge or to determine DA D2/3 receptor occupancy in striatal and ⁎ Corresponding author at: University Hospitals of Geneva, Clinical Neurophysiology and Neuroimaging Unit, 2, chemin du Petit-Bel-Air CH-1225 Chêne-Bourg, Geneva, Switzerland. Fax: + 41 22 305 53 75. E-mail address: [email protected] (P. Millet). 1053-8119/$ – see front matter © 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2012.05.075

extrastriatal regions (Mukherjee et al., 2005; Riccardi et al., 2008; Slifstein et al., 2004, 2010). With the major technical evolution of small animal imaging with PET, recent studies have demonstrated that it is feasible to perform dynamic neuroreceptor imaging in rodents (Dupont and Warwick, 2009; Xi et al., 2011) and thus to study physiologic processes in small brain regions. In this context, many [ 18F]Fallypride studies are now performed in rodents which focus on the reliability of acquired data to assess quantitative or semi-quantitative binding parameters using kinetic modeling (Rominger et al., 2010a, 2010b; Vucckovic et al., 2010; Yoder et al., 2011). However, it is clear from these studies that several technical issues affect the accuracy of the output estimates. Injection of [ 18F] Fallypride induces the presence of [ 18F]Fluorine in the blood and thus a rapid accumulation in the skull and glands. The resulting signal in the brain corresponds to a mixing of tissue and skull and gland signal caused by spillover activity. Because of the small size of the rodent brain relative to the spatial resolution of the small animal PET scanners,

1456

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

partial volume effects (PVE) are relatively large in small structures which are distorted by the presence of neighboring activity (Rousset and Zaidi, 2006). For small animal [ 18F]Fallypride studies, one of the most contaminated regions is the cerebellum, considered as a reference tissue region in the quantitative approach used to estimate binding parameters such as binding potential (BPND) (Logan et al., 1996). Spillover activity is generally overcome in humans either by applying an algorithm to correct for PVE using anatomical information from MRI (Meltzer et al., 1999) or by including spillover correction directly in the compartmental model used to fit kinetic PET data (Carson et al., 2003). However, the first method requires accurate PET and MRI coregistration and tissue segmentation, only possible using individual MRI scans which are rarely available in rat studies. The second method seems to be more appropriate for animal studies, but it significantly increases errors in the estimated model parameters. In this study, we use factor analysis (FA) to correct the crosscontamination between skull and gland and brain activities in the decomposition of PET kinetic images into functions free of noise and spillover contamination from adjacent structures (Buvat et al., 1993; Di Paola et al., 1982; Frouin et al., 1999). A multi-injection approach is used to validate our FA-corrected data from which we estimate all binding parameters. Taking the multi-injection approach as a gold standard to estimate receptor density B′max and ligand affinity KdVR and thus their ratio, which defines binding potential BP, we evaluate the ability of Logan's graphical analysis to estimate BPND with the widely used single injection [ 18F]Fallypride approach. More generally, we propose a very effective method for PVE correction that can be used in human PET studies.

Material and methods The YAP-(S)PET small animal scanner The YAP-(S)PET system is a small animal scanner that combines PET and SPECT imaging (Del Guerra et al., 2006). The scanner is made up of four detector heads, each composed of 4 × 4 cm 2 of yttrium aluminum perovskite activated by a cerium matrix of 20 × 20 crystals, 2 × 2 mm wide and 25 mm deep. The matrix is directly coupled to a position-sensitive photomultiplier with an active area 50 mm in diameter with 0.5 mm intrinsic spatial resolution. The timing and energy windows are 8 ns and 50–850 keV respectively. The scanner has an axial field of view (FOV) of 4 cm and a transaxial FOV 4 cm in diameter. A 2D expectation maximization algorithm with single-slice rebinning was used for the reconstruction procedure (50 iterations). The resulting voxel size was 0.5 × 0.5 × 2 mm 3 (Del Guerra et al., 2006). No post-filtering was applied. Data were corrected

for dead time, radioisotope decay and random counts. No corrections were made for attenuation and scatter. Prior to each experiment, we employed a phantom to calibrate all measurement systems. The phantom was a [ 18F]-distributed source (7.4 MBq) 20 mm in diameter and 50.19 mm in height (V= 63.07 ml) which was acquired for 20 min and reconstructed with the protocol used in this study. Animal preparation Eleven male Sprague Dawley rats (Janvier Laboratories, Le GenetSt-Isle, France) weighing 370 to 400 g were employed for YAP-(S) PET acquisitions and metabolite studies (Table 1). The rats were anesthetized with 4% isoflurane vaporized in oxygen and placed in the YAP-(S)PET gantry with their brain positioned in the center of the field of view. Anesthesia was maintained at 2.5% during PET scanning and body temperature maintained at 37 ± 1 °C using a thermostatically controlled heating blanket. All surgical and experimental procedures were performed in accordance with the Swiss Federal Law on animal care under a protocol approved by the Ethical Committee on Animal Experimentation of the Canton of Geneva, Switzerland. Polyethylene catheters (i.d. = 0.58 mm, o.d. = 0.96 mm) were inserted into the left femoral vein for radiotracer injection and the left femoral artery for blood sampling. Experimental protocol [ 18F]Fallypride and [ 18F]Fluorine were obtained from Advanced Accelerator Applications (Thoiry, France). Similar experimental protocols using three injections were used for both tracers. For [18F]Fallypride imaging, the protocol included [18F]Fallypride injection (injection 1) at the start time (T0), a co-injection of [ 18F]Fallypride and unlabeled Fallypride (injection 2) at T0 + 180 min and a displacement with unlabeled fallypride at T0+ 240 min (injection 3). For [ 18F]Fluorine imaging, [ 18F]Fallypride was simply replaced by [18F]Fluorine. All injections were administered at a constant volume (0.6 ml) over a 1-min period using an infusion pump. The injected quantities (activity and mass) and specific activities are listed in Table 1. J⁎i and Ji correspond to [18F]Fallypride activity (activity/mass) and unlabeled fallypride mass respectively, injected during the ith injection (i =1, 2, 3). A 360-min dynamic acquisition was initiated with the first bolus injection of radiotracer using a set of 72 sequential scans. The data were initially acquired with short frames during the first 5 min after injection. Frames then increased to 10 min. We applied the following protocol: 4 × 0.5 min; 3 × 1 min; 11× 5 min; 12× 10 min; 4 × 0.5 min; 3 × 1 min; 11× 5 min; 4 × 0.5 min; 3 × 1 min; 11 × 5 min; and 6 × 10 min.

Table 1 Numerical values of the YAP-(S)PET and metabolite protocol parameters corresponding to the 11 experiments. Rats

1 2 3 4 5 6 7 8 9 10 11

Modality/tracer

TLC/[18F]fallypride TLC/[18F]fallypride TLC/[18F]fallypride TLC/[18F]fallypride TLC/[18F]fallypride PET/[18F]fallypride PET/[18F]fallypride PET/[18F]fallypride PET/[18F]fallypride PET/[18F]fluorine PET/[18F]fluorine

Duration (min)

180 180 180 180 180 360 360 360 360 360 360

Injection 1 (T = 0 min)

Injection 2

Injection 3

SAa (GBq/μmol)

J⁎ 1 (MBq/μg)

Time (min)

J⁎ 2 (MBq/μg)

J2 (μg)

Time (min)

J3 (μg)

123.49 203.64 443.44 100.84 87.77 53.03 132.96 169.58 117.20 – –

41.8/0.123 41.1/0.121 46.9/0.138 44.1/0.129 49.4/0.145 43.2/0.295 33.0/0.090 40.2/0.086 38.7/0.120 23.9/– 5.9/–

– – – – – 180 180 180 180 180 180

– – – – – 47.5/0.325 42.7/0.116 46.7/0.099 38.7/0.120 17.9/– 6.6/–

– – – – – 2.00 2.10 2.45 5.00 5.00 5.00

– – – – – 240 240 240 240 240 240

– – – – – 200.00 210.00 245.00 250.00 250.00 250.00

Ji* and Ji correspond to [18F]fallypride activity (activity/mass) and unlabeled fallypride mass respectively, injected during the ith injection. a Specific activity at the beginning of the experiment.

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

Arterial plasma analysis During YAP-(S)PET acquisitions, about 60 arterial blood samples (25 μl each) were withdrawn at regular time intervals, and immediately centrifuged for 5 min. Plasma 18F radioactivities were measured with the appropriate counting systems. After calibration, 18F plasma time–concentration curves were converted to kBq/ml. Metabolite measurements During YAP-(S)PET acquisitions, only the whole plasma concentration (CTotal plasma(t)) was measured from arterial blood samples obtained for each rat; metabolites were not measured individually. Metabolite correction was performed from the mean results (adjusted using a coupled fitting procedure) obtained from an independent series of experiments described below. Metabolites were analyzed in five rats. Injected quantities and specific activities are given in Table 1. Fifteen arterial blood samples (~400 μl each) were withdrawn at 0, 1, 2, 3, 4, 6, 8, 10, 15, 30, 45, 60, 90, 120 and 180 min post-radioligand injection. All samples were treated using the procedure developed by Christian et al. (2000). Briefly, arterial blood samples were collected on heparin-coated tubs, gently stirred and centrifuged at 5000 g for 5 min at 0 °C. For each sample, the plasma (10 μl) was measured for radioactivity using the gamma counting system. Plasma (190 μl) was alkalinized with 5% sodium bicarbonate (40 μl) and extracted three times with ethyl acetate (250 μl, 250 μl, and 190 μl respectively). The ethyl acetate extract was carefully removed with a syringe-driven filter unit (Millex-GV 4 mm, 0.22 μm, Millipore Corporation, USA). The filter protein residue and 10 μl aliquots of both the ethyl acetate extract and the aqueous plasma layer, containing respectively the hydrophobic and hydrophilic metabolites of [ 18F]Fallypride, were retained and measured for radioactivity. The remaining ethyl acetate extract (480 μl) was dried using an Eppendorf Concentrator plus (1400 rpm at room temperature) and the desiccated residue was dissolved in methanol (100 μl). A 10 μl fraction was measured for radioactivity and 5 μl samples were used for TLC analyses. TLC analyses were conducted with plastic sheets (silica gel 60F254) from Merck, Germany. All 5 μl fractions were applied to a plate and migration was conducted in a mixture of 10% methanol and 0.2% triethylamine in a methylene chloride solvent system for 30 min. The TLC plate was briefly air-dried and placed for 30 min in a cassette in contact with a phosphor imaging plate (BAS-IP MS2325; Fuji Photo Film Co, Ltd.). The distribution of radioactivity on the plate, corresponding to unchanged [ 18F]Fallypride and its metabolite, was visualized with the Fujifilm BAS-1800 II phosphorimager system and Image Reader v2.02 software (Raytest, Straubenhardt, Germany) and analyzed using Aida Mac Software V4.06 (Raytest Isotopenmessgeräte GmbH). A standard composed of three diluted spots was placed on each plate for calibration purposes as well as 5 μl aliquots of all plasma and methanol extract samples. The mean percentage of non-metabolized [ 18F]Fallypride in rats over time was fitted using the following models to obtain An and Bn parameters: Pnm1(t) = A1e − B1t + A2e − B2t + A3e − B3t, Pnm2(t) = A1e − B1t + A2e − B2t + A3, Pnm3(t) = A1e− B1t + A2e− B2t and the Hill function: Pnm4(t) = 1 − (A1t B1) / (tB1 + C1). The metabolite-corrected plasma input function was obtained using the following equation: Cp(t) = Pnm1/2/3/4(t) * CTotal plasma(t). Coupled fitting procedure for metabolite correction A coupled fitting procedure was used to adjust the mean correction values (A1, B1, A2, B2 and A3) obtained from the group of five rats (Millet et al., 2008). This procedure consists of global minimization using FA-corrected data obtained in seven brain regions (see

1457

below). The goal is to estimate jointly the ligand-receptor model parameters and metabolite correction model. Factor analysis FA is a method used for the extraction of a few elementary components from a series of dynamic images (Di Paola et al., 1982). It summarizes dynamic images into a finite set of curves (K) termed factors, fk, and their associated structures termed factor images, ak, which represent the spatial distribution of the factors fk. The original signal Si(t) associated with pixel i and time t (trixel) can then be expressed through a linear combination of K factors, fk,, weighted by K structures, ak, as demonstrated by the following equation: K

Si ðt Þ ¼ ∑k¼1 ak ðiÞf k ðt Þ þ ei ðt Þ

ð1Þ

ei(t) represents the error term for each pixel i at time t including both noise and modeling errors. It is the remaining part of the image that is not modeled by the factors and structures. One objective of FA is to minimize the following global error term (Frouin et al., 2004): Err ¼ ∑i ∑t ei ðt Þ:

ð2Þ

Each pixel from the raw data is composed of a mixing of signals resulting mainly from activity that is due to the type of tissue (specific or non-specific) and activity contamination from adjacent structures caused by the finite spatial resolution of the imaging system (spillover). In the context of neurotransmission PET images, FA attempts to recover components of the signal such as specific binding, free + non-specific binding and additional skull and gland accumulation as proposed in this study. The sum of these components should correspond to the measured signal without the term ei(t) as shown in Fig. 1. Some components can be removed from the sum to facilitate image interpretation, e.g., the spillover component is removed in the present study. Factor analysis estimates the factors fk(t) and the associated factor images ak(i) assuming that the number of factors K is known (K = 3 in our study). Constraints are added to select the solution with physiological meaning. The analysis requires four main stages as shown in Fig. 1: data preprocessing, orthogonal analysis, oblique analysis and factor image computation. A detailed description of each stage can be found elsewhere in the literature (Buvat et al., 1993, 1998; Di Paola et al., 1982; Frouin et al., 1999). Briefly, data preprocessing consists in selecting appropriate trixels using thresholding, clustering or masking. It reduces the influence of noise in dynamic images. In our study, we did not include in the factor analysis all PET data measured in the whole field of view of the camera (4 cm * 4 cm), but only pertinent information (i.e. activity obtained in brain, skull and glands). Indeed, the PET signal outside the rat head correlates with noise only. Therefore, out-of-interest regions have been removed using the co-registered PET images on the MRI brain template. In this way, only the relevant signal is included in FA. The orthogonal analysis is based for the most part on a statistical analysis. It identifies a low-dimensional subspace which represents the relevant part of the trixels without the noise. In the case of nuclear medicine data, the optimal orthogonal decomposition is obtained by correspondence analysis (Benali et al., 1993). In this stage, the trixels are decomposed into a set of orthogonal curves uk(t) and the associated images Vk(i). The K orthogonal curves, which account for most signal variance, associated with the coefficients of the decomposition, yield solution (1) that minimizes error (2) in a weighted least-square sense. The image sequence Si(t) can be approximated by S^ i ðtÞ, a linear combination of the K most important orthogonal components uk(t). However, orthogonal components

1458

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

Fig. 1. The factor analysis (FA) algorithm used to decompose kinetic PET data into factors and factor images. Data preprocessing consists in selecting appropriate trixels using thresholding and clustering. Orthogonal analysis is a statistical processing equivalent to a principal component analysis of the relevant trixels obtained in the previous step. Oblique analysis allows a further interpretation using a priori information and led to three main factors which are the linear combination of the K orthogonal curves that best satisfy our constraints. Factor images are computed in the initial spatial sampling by means of oblique projection. The reconstructed dynamic images were finally obtained using the first two components and the remaining error ei(t) defined in Eq. 1.

obtained by this analysis are difficult to interpret in physiological terms. Oblique analysis allows a further interpretation using a priori information not only based on statistical considerations. These constraints should be representative of the physiological process in question to ensure an accurate solution. In our study, we have selected the first three orthogonal components and introduced a positivity constraint on all factors and “an increasing function” on one of the three factors corresponding to 18F skull and gland accumulation. The resulting factors (factor 1, factor 2 and factor 3) are the linear combination of the K orthogonal curves that best satisfy our constraints.

Factor images were computed in the initial spatial sampling by means of oblique projection. In this stage, a normalization was carried out with L1 norm on compartments as well as the same metric used for orthogonal analysis, i.e. the correspondence analysis. The reconstructed dynamic images were finally obtained using the first two components without skull and gland accumulation (factor 3). Finally, the remaining factors (4 and more) obtained during the orthogonal analysis and excluded from the oblique analysis were added to the final volume to avoid modeling errors. These factors correspond to noise and modeling errors. The factor analysis was performed with customized commercially available software (Pixies, Apteryx, Issy-les-Moulineaux, France).

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

Data analysis The dynamic PET images were first averaged over time frames measured between 120 and 170 min after the first injection to enhance visualization of the different structures. The averaged PET images were then manually co-registered to the anatomical data of an MRI template for rat brain (Schweinhardt et al., 2003) using PMOD software (PMOD, version 3.3, 2012 PMOD Technologies Ltd, Zurich, Switzerland). The transformation parameters obtained from this co-registration were then applied to dynamic images. An inhouse region of interest (ROI) template was defined on the MRI rat brain atlas, using the Paxinos and Watson stereotaxic rat brain atlas (Paxinos and Watson, 1998) and previously described guidelines for analyzing rodent PET data (Dalley et al., 2007). The ROI template included brain regions defined as follows. Cylindrical ROIs measuring 2 mm in diameter were placed at the stereotaxic center of selected brain structures. Central coordinates (in mm, relative to bregma) and anterior/posterior (AP) length (in mm) for the different ROIs were: [AP +0.5, DV (dorsal/ventral) −5, ML (medial/lateral) ±3; AP length 2 mm] for dorsal striatum (DST); [AP +1.2, DV −7.5, ML ±1.5; AP length 1.5] for ventral striatum (VST); [AP −3, DV − 6, ML ±2; AP length 1.8] for Thalamus (Thal); [AP − 5.4, DV − 8, ML ±1.5; AP length 1] for Midbrain (MB); [AP − 5.3, DV −6, ML ± 5; AP length 1.5] for Hippocampus (Hip); and [AP −3, DV −1.5, ML ±2; AP length 1.8] for cortex (Cx). For the cerebellum (Cereb), a single elliptical ROI (2 mm × 1 mm) was placed over the central lobule of the cerebellar cortex [AP −11, DV − 5, ML 0]. A single ROI was placed over the skull. Time–activity curves (TACs) were finally generated from the defined ROIs applied on dynamic images. Different model configurations have been used for data processing. These models were either written in MATLAB R2010b codes (multiinjection approach) or directly obtained with the PMOD software for Logan's graphical analysis. The three-tissue compartment seven-parameter model (3T-7k) shown in Fig. 2 was applied to analyze the time–concentration curves obtained with the multi-injection approach (Delforge et al., 1999). The unchanged [ 18F]Fallypride measured in plasma was used as the input function. The tissue compartments correspond to the concentrations

1459

of free ligand, specifically bound ligand and non-specifically bound ligand. The following parameters were estimated from data obtained with the multi-injection protocol: B′max, K1, k2, kon/VR, koff, k5 and k6. K1 and k2 are associated with exchanges between plasma and the free ligand compartment, while k5 and k6 correspond to rate constants describing exchanges between non-specifically bound and free ligand compartments. B′max represents the concentration of receptor available for binding. The kon and koff parameters are the association and dissociation rate constants respectively, while VR is the volume of reaction, which accounts for tissue inhomogeneity. FV, representing the fraction of blood present in the tissue volume, was assumed to be 0.05 (Millet et al., 2008). Binding potential, BP was calculated from the model parameters as the B′max/KdVR ratio. The BPND parameter was obtained from Logan's graphical analysis using the following equation to determine the distribution volume ratio (DVR = BPND + 1) (Logan et al., 1996): 2T

T

∫ Ct ðtÞdt 0

Ct ðTÞ

6∫ Cref ðtÞdtþ 60 ¼ DVR 6 6 Ct ðTÞ 4

3

Cref ðTÞ k′2 7

7 7 þ int ′ 7 5

ð3Þ

where Cref(t) and Ct(t) denote, respectively, the activity concentration measured by PET in the reference region and target region and k′2 (in min − 1) is an average tissue-to-plasma clearance k2′ estimated in the reference region. Using the results of the multi-injection approach, k′2 value was fixed at 0.36 min − 1. The results of the linear regression give the slope, DVR, and an intercept, int′, which become constant after an equilibrium time. The fitting time periods were estimated to range from 60 to 180 min. FA simulation study Simulation studies were performed to evaluate the reliability of FA to correct rat PET data from spillover defluorination radioactivity. The simulated dynamic images were generated from real data obtained in different rat experiments (experiments 9 and 11, Table 1). The specific

Fig. 2. The 3T-7k model used in analysis of PET time–concentration curves obtained from the multi-injection protocol. The model at the top describes the kinetics of the radioligand (quantities denoted with a star superscript). The bottom part shows the same model for the unlabeled ligand. Parameters K1 and k2 are associated with the exchanges between the plasma and free ligand compartment; B′max represents the concentration of receptors available for binding; kon and koff are the association and dissociation rate constants respectively and VR is the volume of reaction, which accounts for tissue inhomogeneities; k5 and k6 are the rate constants associated with non-specific binding. All ligand transfer probabilities between compartments are linear except for binding probability, which depends on the bimolecular association rate constant (kon), local free ligand concentration Mf(t), and the local concentration of free receptor sites [B′max -M⁎s (t)-Ms(t)]. Mns(t) is the concentration of non-specific bound ligand. The PET experimental data correspond to the sum of the labeled ligand in the free and specific and non-specific bound ligand compartments and a fraction, Fv, of the blood compartment.

1460

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

and non-displaceable (ND = free + non-specific) components in the brain were obtained with the FA procedure described in the Factor analysis section. Briefly, three factors (specific, ND and skull and gland accumulation) were obtained from the raw PET data (experiment 9, Table 1) using FA. The skull and gland accumulation was then suppressed and the remaining two components were used to reconstruct FA-corrected images. These images were considered as the “pure” components (original images) in our simulations from which reference TACs were extracted in the ROIs defined in the data analysis described in the Data analysis section (Cereb, VST). The “pure” skull and gland accumulation images were obtained with another rat experiment using a multi-injection of [ 18F]fluorine, as described in Table 1 (experiment 11, Table 1). A thresholding procedure was used to remove radioactivity in the brain caused by spillover while keeping the radioactivity in the scalp and glands. In this way, we obtained a mask that was added to the original images to build the simulated images (simulated images = original images + Mask), the two volumes being co-registered to the same MRI template. Then, a 3D Gaussian kernel of increasing size (FWHM = 0.5, 1, 1.5, 2, 2.5, 3, 3.5 and 4 mm) was applied to the simulated images to obtain different filtered images. The FA procedure was then applied to the filtered images to retrieve the original data. Finally, TACs were obtained from the defined ROIs applied to the filtered and FA-corrected images and were compared to the reference TACs. Results [ 18F]Fallypride and [ 18F]Fluorine brain distribution Figs. 3(A–C) show an example of PET images averaged over 120 to 170 min post-injection 1 (labeled injection) and 90 to 120 min postinjection 3 (displacement procedure) of dynamic [ 18F]Fallypride and [ 18F]Fluorine data projected upon the MRI rat atlas in the coronal, sagittal and horizontal planes. For raw [ 18F]Fallypride data (Fig. 3A), high radioactivity levels are visible both in specific structures such as DST and in non-specific areas such as the skull and surrounding glands. The displacement procedure using a large mass of cold Fallypride (T = 240 min) produced a significant decrease in radioactivity in the striatal and extrastriatal regions but not in the skull and glands where it was accumulated. Using an [ 18F]Fluorine multi-injection protocol (Fig. 3B), similar patterns of accumulated radioactivity were observed in the skull and glands and their contaminations were clearly visible in the brain regions (DST, VST and Cereb for example). After FA correction (Fig. 3C), radioactivity surrounding the brain (skull and glands) was largely reduced, which improved the contrast between specific and non-specific regions. The displacement procedure made it possible to observe the non-displaceable signal across the brain regions. [ 18F]Fallypride and [ 18F]Fluorine TACs Figs. 3(D–F) show TACs obtained from ROIs defined in eight structures using kinetic data shown in Figs. 3(A–C), respectively. For raw [ 18F]Fallypride data (Fig. 3D), high uptakes were found after each tracer injection in all regions followed by a slow decrease in high D2/3 receptor density regions (DST, VST) or a rapid decrease in low D2/3 receptor density regions. At the displacement time (240 min), a rapid decrease was observed in the striatal regions. After initial uptake peak post-injections 1 and 2, a constant increase of radioactivity was observed in the skull region. As previously described in Fig. 3A, all [ 18F]Fallypride TACs were disturbed by the accumulated radioactivity in the skull and glands. This bias was clearly visible during the displacement period, especially in the VST region. For [ 18F]Fluorine (Fig. 3E), a slow uptake was observed in all regions without the initial peak observed on the skull TACs with the [ 18F]Fallypride data. All TACs were similar in shape; only the level of

radioactivity was different across regions with a maximum level in the skull or glands. Among the brain regions, the VST was the region most disturbed by the accumulated radioactivity in the skull and glands. Fig. 3F shows the TACs obtained after applying FA correction to raw [ 18F]Fallypride data shown in Figs. 3(A, D). As shown in Fig. 3C, the accumulation component associated with defluorination was removed not only from the skull and glands, but also from the brain structures, leading to more realistic TACs in all regions, especially the VST. The constant level of radioactivity observed in raw Cereb data between 50 and 180 min post-injection was transformed into the usual decrease using FA-corrected data. Factor analysis Figs. 4(A–C) show the detailed FA results obtained using the [ 18F] Fallypride PET volume in one rat. Three factors (i.e. curves) and the corresponding factor images were computed using the algorithm shown in Fig. 1. Factor 1 in Fig. 4A was visually identified as the specific binding with the DST region clearly observed on the corresponding factor images. The contribution of factor 1 to the whole PET signal was 27%. Factor 2 (Fig. 4B) was identified as the free + non-specific binding (26.6%) and factor 3 (Fig. 4C) as the skull accumulation that involves up to 46.3% of the brain PET signal. The skull and gland regions were clearly visible on the last factor images. Figs. 4(D–F) show the FA projection on three brain regions: DST, Cereb and Skull. The contribution of each factor to the original TAC of each ROI is given in the figure. For example, the contribution of factor 3 (skull and gland accumulation) to the original PET signal was only 12% in DST, 45.4% in Cereb and 79.7% in the skull region. The FA projection on the skull shows that the peak found after injections 1 and 2 corresponds to a factor 2 contribution (free + non-specific binding). FA simulation studies By using real data obtained in different rat experiments, a dynamic phantom was created to simulate [18F]Fallypride binding in the brain and [18F]fluorine accumulation in the skull and surrounding glands. (A–D) shows examples of original, simulated (original images+ mask), filtered (only FWHM=0.5 mm, 2 mm and 4 mm) and FA-recovered images for several brain slices. The images represent the radioactivity measured between 120 and 170 min after the first injection. Fig. 5C (FWHM = 2 mm) represents the simulation closest to the data obtained with the YAP-(S)PET scanner since the size of its point spread function is around 1.8 mm FWHM. Figs. 5(E, F) show the TACs obtained from the Cereb and VST regions defined in Fig. 5A and applied to original, simulated, filtered and FA-recovered dynamic images. Simulations show that the TACs (blue curves) are highly perturbed by spillover contamination when the size of the Gaussian kernel increases. The FA procedure using only three factors makes it possible to recover true activity since all TACs (black curves) are superposed on that obtained using the original data (red curve). If we increase the number of factors in the oblique analysis (4 and more), the resulting decomposition is difficult to interpret from a pharmacological point of view. Arterial input function Fig. 6A is the percentage of nonmetabolized [18F]Fallypride over time analyzed using a group of five rats. Data were initially fitted using a three-exponential model (Pnm1(t)= A1e− B1t + A2e− B2t + A3e− B3t); the estimated values were A1 = 0.62, B1 = 3.55, A2 = 0.34, B2 = 0.11, A3 = 0.077 and B3 = 0.002. Because of the low B3 value, we additionally applied a bi-exponential model plus constant (Pnm2(t) = A1e− B1t + A2e− B2t + A3). In this case, we estimated values of A1 = 0.35, B1 = 0.10, A2 = 0.62, B2 = 0.3.47, and A3 = 0.062. The use of simple biexponential without constant (Pnm3(t)) led to a very bad fit. Finally,

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

1461

Fig. 3. (A–C) Example of averaged YAP-(S)PET images obtained between 120 and 170 min post-injection 1 (columns 1, 2 and 3 corresponding to coronal, sagittal and horizontal orientation, respectively) and between 90 and 120 min post-injection 3 which is the displacement procedure using 0.250 mg of cold fallypride (columns 4, 5 and 6 using the same orientation). The first row (Fig. 3A) represents images obtained with the multi-injection protocol including [18F]Fallypride and/or cold fallypride injections (experiment 9 in Table 1). The second row (Fig. 3B) corresponds to a multi-injection protocol using [18F]Fluorine and/or cold fallypride injections (experiment 11 in Table 1). The third row (Fig. 3C) shows [18F]Fallypride images depicted in the first row but after FA correction. (D–F) TACs measured in different brain regions and the skull corresponding to images shown in Figs. 3 (A–C), respectively.

we applied the Hill function (Pnm4(t) = 1 − (A1tB1) / (tB1 + C1)). The estimated values were A1 = 0.98, B1 = 0.63, and A2 = 0.79. The Hill function (Pnm4(t)) was initially used in conjunction with the ligand–receptor model in a coupled fitting procedure to adjust the metabolite correction parameters for each rat used in YAP-(S) PET experiments. However, this model led to convergence problems for A1 and B1 parameters during the identification procedure. Therefore, the Pnm2(t) model was used during the coupled fitting procedure. The mean values obtained were A1 = 1.02 ± 0.36, B1 = 6.08 ± 2.91, A2 = 0.029 ± 0.015, B2 = 6.55 ± 1.98, and A3 = 0.033 ± 0.017 for all rats. Compartmental kinetic analyses The 3T-7k model was applied to fit [ 18F]Fallypride data. The use of raw data led to aberrant biological values. For example, high koff values greater than 0.2 min − 1 were estimated in the VST regions because of

bias during the displacement procedure. The use of FA-corrected data made it possible to obtain correct estimates for all model parameters. Two typical fits for DST and Cereb are shown in Figs. 6(B–C), with separation of free, specific and non-specific binding from the model. After each tracer injection, a high peak uptake was obtained in all brain regions followed by a decrease, the magnitude of which depends on receptor density. In DST, much of the signal represents specific binding (about 75% at 180 min), the remaining part being free and nonspecific binding. In Cereb, TAC mainly involves free and non-specific binding, but also a small amount of specific binding confirmed by a weak displacement at 240 min. Table 2 shows model parameter values obtained in all brain regions across four rats. The highest B′max values were obtained in the DST, specifically, 44.1 ±9.4 pmol/ml, and the lowest values in the cerebellum, specifically, 0.9 ±0.5 pmol/ml. High K1 values were estimated in all ROIs leading to DVfree (K1/k2) values larger than unity. The kon/VR

1462

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

Fig. 4. (A–C) Examples of factors and factor images obtained by FA with the whole data shown in the last row of Fig. 3. The PET signal at the pixel level is approximated to a linear combination of these three factors (specific binding, free + nonspecific binding (NS) and skull and gland accumulation) represented by the white curves. The corresponding associated factor images are computed for each factor. The percentage indicates the influence of the calculated factor compared to the whole signal. (D–F) Projection of the FA on three regions (DST, Cereb and Skull). The influence of each factor on the brain regions can be evaluated using these projections. For example, Cereb is particularly biased by factor 3 (45.4%), i.e. the skull accumulation.

and koff parameter values were relatively uniform across high D2/3 receptor density regions. Therefore, the resulting KdVR parameter values were uniform in these ROIs. A non-negligible non-specific binding was found in all regions with k5 and k6 values ranging from 0.025 to 0.048 min − 1 and 0.005 to 0.025 min − 1 respectively. Binding potential (BP) estimations Binding potential values were obtained with 1) Logan's graphical analysis (BPND) using both the raw and FA-corrected data, BPND-RAW and BPND-FA respectively and 2) directly with the 3T-7k model

parameters estimated with the multi-injection approach (BP = B′max/ KdVR) and FA-corrected data (Table 2). The rank order of BP and BPND-FA values (Cx b Hip b MB ≈ Thalb VST b DST) matched those of B′max values. High coefficients of correlation were found between both BP and BPND-RAW (r= 0.981, p b 0.05) and BP and BPND-FA (r = 0.999, p b 0.05). However, the BPND-RAW values were underestimated compared to those obtained with the FA-corrected data (BPND-FA) from 21% to 108% across brain regions at 180 min (Fig. 7). BPND-RAW values in Fig. 7A were approximately constant in all brain regions over acquisition times ranging from 60 to 180 min, whereas BPND-FA values (Fig. 7B) become stable after 90 min of PET acquisition.

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

1463

Fig. 5. Simulation of the spillover effects from radioactivity in the skull and glands on the [18F]Fallypride signal in the brain. Figs. 5 (A–D) show examples of original (nondisplaceable (ND) + specific), simulated (original images + mask), filtered (only FWHM = 0.5 mm, 2 mm and 4 mm) and FA-recovered images for several brain slices. These images represent radioactivity measured between 120 and 170 min after the first injection. Figs. 5 (E, F) show TACs obtained from Cereb and VST regions defined in Fig. 5A and applied to original, simulated, filtered and FA-recovered dynamic images. Simulations show that the TACs (blue curves) are highly perturbed by spillover contamination when the size of the Gaussian kernel increases. The FA procedure using only three factors makes it possible to recover true activity since all TACs (black curves) are superposed on that obtained using the original data (red curve).

Discussion This work describes full quantitative kinetic modeling of the [18F] Fallypride radiotracer for in vivo estimation of D2/3 receptors in the rat brain using FA-corrected PET kinetic data. With the increasing number of [18F]Fallypride imaging studies in small animals (Constantinescu

et al., 2011; Rominger et al., 2010a, 2010b; Tantawy et al., 2009; Yoder et al., 2011), it seemed important to check the reliability of the quantitative results obtained from this radiotracer. Indeed, recent studies have shown a significant skull and gland labeling after [ 18F]Fallypride injection, which disturbs the signal measured in all brain regions (Constantinescu et al., 2011; Rominger et al., 2010a; Tantawy et al.,

1464

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

2009; Yoder et al., 2011). At present, the intrinsic spatial resolution limitations of PET make small animal imaging a challenge. The accuracy of PET measurements is reduced by this limited resolution and the resulting PVE leads to quantitative under- and overestimations of the regional concentration of radioactivity depending on brain regions.

PVE consist of spillover effects between brain regions and “tissuefraction” effects reflecting underlying tissue heterogeneity (Aston et al., 2002). To overcome the spillover effects, a FA method was used to correct the dynamic images obtained with a multi-injection protocol. This protocol has already been applied with [ 18F]Fallypride to quantify B′max and KdVR in monkeys (Christian et al., 2004; Vandehey et al., 2009), but it has been adapted in this study to allow us to estimate both B′max and KdVR parameters from the whole experimental protocol and the BPND parameter with Logan's graphical approach using only the first 180 min data. Our raw data were consistent with those found in the literature, i.e. concordant with the well known distribution of [ 18F]Fallypride in the striatal and extrastriatal regions and with a high level of radioactivity in the skull and glands. Indeed, injection of [ 18F]Fallypride leads to the presence of [ 18F]Fluorine in the blood which rapidly accumulates in the skull and glands. Our kinetic analysis in Figs. 3A and D shows clearly that radioactivity in the skull exceeds that in the specific brain regions for late post-injection time. Therefore, radioactivity in all the brain regions was overestimated because of the activity spillover from neighboring structures (skull and glands). The consequences of spillover activity were particularly evident in VST during the displacement procedure. Mixing of the “true” and spillover radioactivities led to a decrease in PET signal followed by a non-pharmacological increase. At the displacement time, only a decrease should be obtained. Radiodefluorination has already been described with several PET tracers in humans (Carson et al., 2003; Kimura et al., 2010; Ryu et al., 2007; Shetty et al., 2008). For example, the 5-HT1A antagonist [ 18F] FCWAY showed an irreversible uptake of [ 18F]Fluorine in the skull after injection (Carson et al., 2003). To highlight the consequences of defluorination on the brain regions, we applied a multi-injection approach in replacing [ 18F]Fallypride by [ 18F]Fluorine. Using similar acquisition and reconstruction procedures for both [18F]Fallypride and [ 18F]Fluorine data, we have shown similar accumulations in skull and glands and no effect from the large mass of cold fallypride ligand on [18F]Fluorine accumulation. At late timepoints, the [18F]Fallypride and [ 18F]Fluorine images were similar because of displacement of the specific [18F]Fallypride signal (Fig. 3A) by a large mass of cold fallypride. In this case, only the 18F accumulation signal was still visible because of its greater magnitude compared to that of the nondisplaceable [ 18F]Fallypride components (free + non-specific binding). The main advantage of using a similar [ 18F]Fluorine multi-injection protocol was to quantify directly the effect of activity spillover from the skull and glands into the brain regions without the specific binding of [ 18F]Fallypride. Kinetic analysis in Figs. 3B and E shows a rapid accumulation of radioactivity in the skull and glands and a non-negligible activity spillover in all brain regions. Moreover, we also observed that the uptake of [ 18F]Fluorine at injection times differed from that obtained with the [ 18F]Fallypride. A true accumulation was observed with [ 18F]Fluorine in the skull and glands, whereas an additional high peak was obtained with [ 18F]Fallypride. This peak is explained by the spillover activity from free [ 18F]Fallypride into the tissue because of high tracer extraction from the blood at injection times (see discussion below about K1 values). Several methods are available to correct images for skull and gland radioactivity spillover. One approach is a partial volume correction (PVC) method based on anatomical information from MRI (Hayakawa Fig. 6. (A) Time-courses of the non-metabolized [18F]Fallypride measured from 5 experiments. Different models (Pnm(t)) were used to fit data. The model input function was defined by Cp(t) = Pnm1/2/3/4(t) * CTotal plasma(t). (B–C) Examples of DST and Cereb time–concentration curves measured in experiment 9. The solid line represents the 3T7k compartment model fit to the YAP(S)PET data (DST: B′max = 37.0 pmol/ml, K1 = 0.89 min− 1, k2 = 0.32 min− 1, kon/VR = 0.081 mL/(pmol.min), koff = 0.072 min− 1, k5 = 0.073 min− 1, k6 = 0.005 min− 1; Cereb: B′max = 1.3 pmol/ml, K1 = 0.93 min− 1, k2 = 0.34 min− 1, kon/VR = 0.142 ml/(pmol.min), koff = 0.076 min− 1, k5 = 0.041 min− 1, k6 = 0.011 min− 1). Free, specific and non-specific binding are the compartment components of the PET signal simulated by the model.

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

1465

Table 2 Mean parameter estimates obtained with the multi-injection approach and with Logan's graphical analysis.

DST SD VST SD Thal SD MB SD Hypo SD Cx SD Cereb SD

B′max pmol/ml

K1 min− 1

k2 min− 1

kon/VR mL/(pmol min)

koff min− 1

k5 min− 1

k6 min− 1

kdVR nM

BP (B′max/kdVR)

BPND-RAW

BPND-FA

44.1 9.4 22.4 2.6 9.9 3.0 8.7 3.2 5.6 2.1 2.5 1.3 0.9 0.5

1.13 0.35 1.11 0.39 1.61 0.60 1.56 0.62 1.28 0.54 1.09 0.42 1.25 0.57

0.36 0.09 0.35 0.03 0.50 0.19 0.45 0.20 0.32 0.16 0.28 0.10 0.36 0.16

0.086 0.040 0.065 0.014 0.060 0.020 0.060 0.012 0.042 0.015 0.101 0.065 0.113 0.022

0.107 0.053 0.079 0.015 0.079 0.022 0.075 0.013 0.062 0.013 0.065 0.030 0.054 0.031

0.033 0.029 0.034 0.013 0.043 0.020 0.044 0.023 0.025 0.014 0.043 0.016 0.048 0.019

0.012 0.017 0.005 0.004 0.008 0.008 0.010 0.010 0.007 0.006 0.018 0.011 0.021 0.012

1.248 0.258 1.243 0.208 1.364 0.258 1.284 0.300 1.547 0.326 0.907 0.541 0.494 0.294

36.02 7.85 18.19 1.81 7.68 3.55 7.31 4.25 3.70 1.49 3.34 1.37 2.54 1.43

3.21 1.23 2.13 0.63 0.50 0.23 0.74 0.28 0.36 0.21 0.15 0.12 – –

5.10 2.87 2.58 1.28 1.03 0.39 1.05 0.34 0.52 0.23 0.25 0.19 – –

Values are mean ± SD of four experiments. BP corresponds to the B′ max/KdVR ratio. BPND is estimated using Logan's Graphical analysis using raw and FA data. The seven regions correspond to dorsal striatum (DST), ventral striatum (VST), thalamus (Thal), MidBrain (MB), hippocampus (Hip), cortex (Cx) and cerebellum (Cereb).

Fig. 7. Time stability of BPND-RAW (A) and BPND-FA (B) parameters using the mean values obtained from the four experiments. The BPND-FA values are increased from 21% to 108% across regions at 180 min compared to BPND-RAW values. The time required to obtain stable BPND-FA values in the striatal and extra-striatal regions is estimated to be 90 min postinjection.

et al., 2000; Muller-Gartner et al., 1992; Rousset and Zaidi, 2006; Rousset et al., 2008). However, the effectiveness of this method relies mainly on the accuracy of PET and MRI co-registration and tissue segmentation. Therefore, individual variability in brain anatomy necessitates the use of an additional MRI scan for each individual rat rather than an MRI template as is the case in our study. The MRI-based method has thus not been investigated. The PVC VOI-based method introduced by Rousset et al. (1998, 2008) was also evaluated in this study. However, this method has not yielded satisfactory results because of the difficulty in defining precisely a ROI mask including skull and gland activity. Depending on the generated mask, the correction on dynamic PET images was very variable and the resulting TACs were aberrant with negative values. Another approach is based on the estimation of correction factors for spillover during mathematical modeling of kinetic data (Carson et al., 2003; Fang and Muzic, 2008). However, this approach has not yielded satisfactory results because of the large number of model parameters to be estimated, leading to a high covariance between the model and spillover parameters. In this study, the FA method was used as an alternative solution to correct the cross-contamination between skull (and glands) and brain activities. FA techniques have been mainly used in Cardiology for extracting the “pure” arterial TAC from kinetic PET data (Kim et al., 2003; Kim et al., 2006; Wu et al., 1995). These studies demonstrated that the FA method could extract accurate input functions and myocardial TACs from small-animal PET images of rodents. To date, FA has never been applied to neurotransmission PET studies to correct spillover effects from adjacent structures or to extract the radioligand time-courses for free, specific or non-specific binding in tissue. The FA-corrected images were improved with a better contrast between specific and non-specific regions. The resulting TACs were more consistent with the pharmacokinetics of a radioligand. For example, a constant decrease of the specific PET signal during the displacement procedure was observed. Three distinct factors were clearly recovered from FA with their corresponding spatial distributions. Each factor represented a clear statement of the radiotracer in tissue, namely free + non-specific binding, specific binding and radioactivity accumulation in the skull and glands. Only the non-specific binding was not clearly separable. The addition of a fourth factor to FA was not convincing, because of the variable resulting factors across rats, making it difficult to interpret from a pharmacokinetic standpoint. Knowing the biological interpretation of each factor, we have projected the FA results on three different regions, DST, Cereb and Skull, in order to evaluate the approximate percentage contribution of each factor to the regional PET signal. We found two interesting results.

1466

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

Firstly, as expected, the cerebellar region is highly contaminated by skull and gland radioactivity since the major part of the signal at 180 min corresponds to spillover activity. This result was predictable in view of our results and those published in the literature that show a constant level of radioactivity in the cerebellum for the late timepoints (Constantinescu et al., 2011; Rominger et al., 2010a; Yoder et al., 2011). The real washout is probably balanced by the additional signal caused by radioactivity spillover. The radiotracer kinetic in the cerebellum is of great importance for [ 18F]Fallypride studies because it is commonly used as a reference region to estimate the non-displaceable fraction. Secondly, the initial peak in the skull is a spillover activity from free ligand radioactivity into the skull. This is caused by a very rapid transfer of the ligand across the blood–brain barrier, resulting in mathematical terms in K1 ≫ k3 in most brain regions. The kinetics of specific binding were slower than those of free binding causing a sharp peak after each tracer injection. This high early activity contaminated the skull by spillover activity, but was also visible on brain kinetics (initial peak). Simulations (Fig. 5) have replicated the different artifacts obtained on our YAP-(S)PET data, i.e. the early peak in the skull and glands after each tracer injection, the contamination of the brain [ 18F] Fallypride signal by the [ 18F]fluorine accumulated in skull and glands. These simulations have also confirmed the efficiency of FA to recover signal accumulation in skull and glands with only the first three factors. Indeed, the FA-corrected TACs are very close to the “true” TACs. Kinetic analyses of FA-corrected TACs showed that a 3T-7k model provided stable results in all brain regions whereas very bad fits were obtained with raw multi-injection data. The magnitudes of B′max and KdVR values were close to those estimated in nonhuman primates with a similar multi-injection approach using [ 18F]Fallypride (Christian et al., 2004) or using FLB (Delforge et al., 1999). The resulting BP values (BP = B′max/KdVR) were also very close to those obtained in humans (Mukherjee et al., 2002; Slifstein et al., 2010). The delivery of the parent compound from the plasma to the free space of the tissue (K1) was higher than that identified in monkeys (Christian et al., 2004; Vandehey et al., 2009). We found K1 values larger than unity in all brain regions whereas they found a mean of K1 = 0.18 ± 0.04 min − 1 across all brain regions and animals. High K1 values were also reported in a previous rat study (Tantawy et al., 2009). These high K1 values explain the difference in kinetics between primate and rodent. Indeed, Christian et al. (2004) have shown much slower uptakes in the striatal and extra-striatal regions of nonhuman primates without peaks at the injection time. The use of a 2T-5k model without non-specific binding was not satisfactory for most ROIs. Fits were very poor, in particular for kinetic data obtained after injection of cold ligand (co-injection and displacement). Indeed, injection of a large mass of cold ligand during displacement made it possible to observe in our data a non-negligible part of non-displaceable signal. If we consider a uniform non-specific binding across the brain, radioactivity levels after full displacement should be similar in specific and non-specific regions. This was not the case for our data. Introducing a nonspecific binding compartment (3T-7k model) resolved the problems. In most neurotransmission studies, free and nonspecific binding concentrations are usually included in a single compartment because of a rapid equilibrium between the two components. In this study, it appeared that a kinetically distinguishable nonspecific binding was found in all brain regions. The nonspecific binding terms (k5, k6) were then added to the model yielding relatively uniform values across the brain regions. The long time period included in the experimental protocol made it possible to identify both k5 and k6 parameters. Binding potential is often used as the primary outcome measure to access receptor density in radioligand imaging studies. BP can be calculated directly from the model parameters, BP = B′max/KdVR as shown in this study, or by simplified methods based on reference tissue devoid of specific receptors (BPND). In the case of [ 18F]Fallypride

studies, Logan's graphical analysis with the cerebellum as the reference tissue is commonly used to derive the BPND parameter (Logan et al., 1996). The Logan analysis has been used in monkeys and humans (Christian et al., 2000; Mukherjee et al., 2002) and recently in rodents (Rominger et al., 2010a, 2010b; Tantawy et al., 2009). While the choice of cerebellar region (cerebellar lobes) for the [ 18F]Fallypride reference tissue seems to be correct in humans and monkeys, the use of such a reference in rodents led to underestimated BPND-RAW values because of much greater effects of spillover activity in small animals. Indeed, all rodent studies have highlighted the defluorination problem and the resulting accumulation of radioactivity in skull and glands, but without investigating the possible effects of spillover activity on the brain tissue. Therefore, the literature on rodents reported large differences in BPND values, which were certainly related to the difficulty in defining a reference region not contaminated by the spillover effects (Constantinescu et al., 2011; Yoder et al., 2011). In our study, we have also shown an important effect of spillover activity on all brain regions, mainly in Cereb and VST. This spillover effect does not produce a constant error in the BPND-RAW calculation. Indeed, BPND-RAW values include both a systematic bias caused by contamination of the reference region by the skull and gland accumulation and a non-systematic bias depending on the location of the target region to the skull or glands. Moreover, as the BPND values calculated with the Logan analysis using raw data are relative to the radioactivity level in the reference region (BPND = target/reference), the BPND value calculated in a non-perturbed region as DST is largely underestimated (Fig. 7). The resulting BPND-FA values are therefore increased from 21% to 108% across regions compared to those obtained with the raw data. However, comparison of BP values obtained with the multi-injection approach and with Logan's graphical analysis yielded high coefficients of correlation using both raw (r= 0.981) and FA-corrected data (r = 0.999). This demonstrates that BPND-RAW values are underestimated in all brain regions and that overestimation of radioactivity in the reference region is the main cause. An important variability in the BPND values (Table 1) has been obtained using Logan's graphical analysis. This is explained by the variability in the specific activity values across experiments. However, we have verified that the receptor occupancy caused by injected ligand is less than 10% as recommended in the literature (Skinbjerg et al., 2010). Only experiment 6 (Table 1) could be critical in our study (SA =53.03 GBq/μmol). However, we injected 2 nmol/kg [18F]Fallypride in this case (injection 1). Specific binding in striatum was 4.28 nM, calculated as striatum-cerebellum. If we consider a mean B′max value of 44 nM in rat striatum (Table 1), this amount represents ~9.73% receptor occupancy (4.28/44). The time required to obtain stable BPND estimates varies between 60 min and 120 min in the literature (Christian et al., 2000; Rominger et al., 2010b). The time stability for the BPND parameter has been studied with our data and shows that data obtained at least 90 min post-injection are necessary to obtain stable BPND-FA values in the striatal and extra-striatal regions. It was difficult to draw conclusions about this stability with raw data because the skull contamination completely distorts the time-course. Conclusions The PET tracer [18F]fallypride is increasingly used in small animal and human imaging. As with many other radiotracers, radiodefluroination is particularly problematic because it contaminates PET brain activity. In this study, we show that spillover effects are particularly important in small animal imaging because of the small size of the brain structures relative to the limited resolution of the scanner. The resulting BPND-RAW values calculated with the Logan reference approach are thus largely underestimated with a non-systematic bias which depends on the location of the structures in the brain. The FA-based method proposed in this study to correct kinetic data for spillover activity is very effective because it can easily distinguish

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

and eliminate the contaminating component in all brain structures without complex modeling or additional MRI. Moreover, the approach presented here with [ 18F]Fallypride data can be extended to other radioligands and also to human data which can be highly distorted by radiodefluorination as shown in the literature (Shrestha et al., 2012). Our results show that the specific to non-specific ratio has been increased in PET kinetic images after factor analysis based correction without bias of the brain PET signal. This tool is therefore entirely consistent with challenge studies which aim to measure endogenous neurotransmitter release after a pharmacological, cognitive, or motor task. Acknowledgments This work was supported by the Swiss National Science Foundation (no. 310030-120369). The authors are grateful for the contributions of the BioPark Platform in Archamps, the Fondation Caisse d'Epargne Rhône-Alpes, and the ABC laboratory of the European Scientific Institute (ESI). The authors thank Hervé Guillemet from the Apteryx company for his technological support and for providing the developer version of the Pixies software. Thanks also to Alberto Del Guerra for very helpful comments on acquisition and image reconstruction procedures. References Aston, J.A., Cunningham, V.J., Asselin, M.C., Hammers, A., Evans, A.C., Gunn, R.N., 2002. Positron emission tomography partial volume correction: estimation and algorithms. J. Cereb. Blood Flow Metab. 22, 1019–1034. Benali, H., Buvat, I., Frouin, F., Bazin, J.P., Di Paola, R., 1993. A statistical model for the determination of the optimal metric in factor analysis of medical image sequences (FAMIS). Phys. Med. Biol. 38, 1065–1080. Buvat, I., Benali, H., Frouin, F., Bazin, J.P., Di Paola, R., 1993. Target apex-seeking in factor analysis of medical image sequences. Phys. Med. Biol. 38, 123–138. Buvat, I., Benali, H., Di Paola, R., 1998. Statistical distribution of factors and factor images in factor analysis of medical image sequences. Phys. Med. Biol. 43, 1695–1711. Carson, R.E., Wu, Y., Lang, L., Ma, Y., Der, M.G., Herscovitch, P., Eckelman, W.C., 2003. Brain uptake of the acid metabolites of F-18-labeled WAY 100635 analogs. J. Cereb. Blood Flow Metab. 23, 249–260. Christian, B.T., Narayanan, T.K., Shi, B., Mukherjee, J., 2000. Quantitation of striatal and extrastriatal D-2 dopamine receptors using PET imaging of [(18)F]fallypride in nonhuman primates. Synapse 38, 71–79. Christian, B.T., Narayanan, T., Shi, B., Morris, E.D., Mantil, J., Mukherjee, J., 2004. Measuring the in vivo binding parameters of [18F]-fallypride in monkeys using a PET multiple-injection protocol. J. Cereb. Blood Flow Metab. 24, 309–322. Constantinescu, C.C., Coleman, R.A., Pan, M.L., Mukherjee, J., 2011. Striatal and extrastriatal microPET imaging of D2/D3 dopamine receptors in rat brain with [(18)F]fallypride and [(18)F]desmethoxyfallypride. Synapse 65, 778–787. Dalley, J.W., Fryer, T.D., Brichard, L., Robinson, E.S., Theobald, D.E., Laane, K., Pena, Y., Murphy, E.R., Shah, Y., Probst, K., Abakumova, I., Aigbirhio, F.I., Richards, H.K., Hong, Y., Baron, J.C., Everitt, B.J., Robbins, T.W., 2007. Nucleus accumbens D2/3 receptors predict trait impulsivity and cocaine reinforcement. Science 315, 1267–1270. Del Guerra, A., Bartoli, A., Belcari, N., Herbert, D., Motta, A., Vaiano, A., Di Domenico, G., Sabba, N., Moretti, E., Zavattini, G., Lazzarotti, M., Sensi, L., Larobina, M., Uccelli, L., 2006. Performance evaluation of the fully engineered YAP-(S)PET scanner for small animal imaging. IEEE Trans. Nucl. Sci. 53, 1078–1083. Delforge, J., Bottlaender, M., Loc'h, C., Guenther, I., Fuseau, C., Bendriem, B., Syrota, A., Maziere, B., 1999. Quantitation of extrastriatal D2 receptors using a very highaffinity ligand (FLB 457) and the multi-injection approach. J. Cereb. Blood Flow Metab. 19, 533–546. Di Paola, R., Bazin, J.P., Aubry, F., Aurengo, A., Cavailloles, F., Herry, J.Y., Kahn, E., 1982. Handling of dynamic sequences in nuclear medicine. IEEE Trans Nucl. Sci. NS29, 1310–1321. Dupont, P., Warwick, J., 2009. Kinetic modelling in small animal imaging with PET. Methods 48, 98–103. Fang, Y.H., Muzic Jr., R.F., 2008. Spillover and partial-volume correction for imagederived input functions for small-animal 18F-FDG PET studies. J. Nucl. Med. 49, 606–614. Frouin, F., De Cesare, A., Bouchareb, Y., Todd-Pokropek, A., Herment, A., 1999. Spatial regularization applied to factor analysis of medical image sequences (FAMIS). Phys. Med. Biol. 44, 2289–2306. Frouin, F., Delouche, A., Raffoul, H., Diebold, H., Abergel, E., Diebold, B., 2004. Factor analysis of the left ventricle by echocardiography (FALVE): a new tool for detecting regional wall motion abnormalities. Eur. J. Echocardiogr. 5, 335–346. Hayakawa, N., Uemura, K., Ishiwata, K., Shimada, Y., Ogi, N., Nagaoka, T., Toyama, H., Oda, K., Tanaka, A., Endo, K., Senda, M., 2000. A PET-MRI registration technique for PET studies of the rat brain. Nucl. Med. Biol. 27, 121–125.

1467

Kim, K.M., Watabe, H., Kawachi, N., Iida, H., 2003. Extraction of blood pool in dynamic O-15 water PET with slow-infusion using factor analysis. Nuclear Science Symposium Conference Record, 2003 IEEE, vol. 2644, pp. 2640–2641. Kim, J., Herrero, P., Sharp, T., Laforest, R., Rowland, D.J., Tai, Y.C., Lewis, J.S., Welch, M.J., 2006. Minimally invasive method of determining blood input function from PET images in rodents. J. Nucl. Med. 47, 330–336. Kimura, Y., Simeon, F.G., Hatazawa, J., Mozley, P.D., Pike, V.W., Innis, R.B., Fujita, M., 2010. Biodistribution and radiation dosimetry of a positron emission tomographic ligand, 18F-SP203, to image metabotropic glutamate subtype 5 receptors in humans. Eur. J. Nucl. Med. Mol. Imaging 37, 1943–1949. 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. Meltzer, C.C., Kinahan, P.E., Greer, P.J., Nichols, T.E., Comtat, C., Cantwell, M.N., Lin, M.P., Price, J.C., 1999. Comparative evaluation of MR-based partial-volume correction schemes for PET. J. Nucl. Med. 40, 2053–2065. Millet, P., Moulin, M., Bartoli, A., Del Guerra, A., Ginovart, N., Lemoucheux, L., Buono, S., Fagret, D., Charnay, Y., Ibanez, V., 2008. In vivo quantification of 5-HT1A-[18F] MPPF interactions in rats using the YAP-(S)PET scanner and a beta-microprobe. Neuroimage 41, 823–834. Mukherjee, J., Christian, B.T., Dunigan, K.A., Shi, B., Narayanan, T.K., Satter, M., Mantil, J., 2002. Brain imaging of 18F-fallypride in normal volunteers: blood analysis, distribution, test-retest studies, and preliminary assessment of sensitivity to aging effects on dopamine D-2/D-3 receptors. Synapse 46, 170–188. Mukherjee, J., Shi, B., Christian, B.T., Chattopadhyay, S., Narayanan, T.K., 2004. 11CFallypride: radiosynthesis and preliminary evaluation of a novel dopamine D2/ D3 receptor PET radiotracer in non-human primate brain. Bioorg. Med. Chem. 12, 95–102. Mukherjee, J., Christian, B.T., Narayanan, T.K., Shi, B., Collins, D., 2005. Measurement of D-amphetamine-induced effects on the binding of dopamine D-2/D-3 receptor radioligand, 18F-fallypride in extrastriatal brain regions in non-human primates using PET. Brain Res. 1032, 77–84. Muller-Gartner, H.W., Links, J.M., Prince, J.L., Bryan, R.N., McVeigh, E., Leal, J.P., Davatzikos, C., Frost, J.J., 1992. Measurement of radiotracer concentration in brain gray matter using positron emission tomography: MRI-based correction for partial volume effects. J. Cereb. Blood Flow Metab. 12, 571–583. Paxinos, G., Watson, C., 1998. The rat brain in stereotaxic coordinates, fourth edition. Academic Press, San Diego. Riccardi, P., Baldwin, R., Salomon, R., Anderson, S., Ansari, M.S., Li, R., Dawant, B., Bauernfeind, A., Schmidt, D., Kessler, R., 2008. Estimation of baseline dopamine D2 receptor occupancy in striatum and extrastriatal regions in humans with positron emission tomography with [18F] fallypride. Biol. Psychiatry 63, 241–244. Rominger, A., Mille, E., Zhang, S., Boning, G., Forster, S., Nowak, S., Gildehaus, F.J., Wangler, B., Bartenstein, P., Cumming, P., 2010a. Validation of the octamouse for simultaneous 18F-fallypride small-animal PET recordings from 8 mice. J. Nucl. Med. 51, 1576–1583. Rominger, A., Wagner, E., Mille, E., Boning, G., Esmaeilzadeh, M., Wangler, B., Gildehaus, F.J., Nowak, S., Bruche, A., Tatsch, K., Bartenstein, P., Cumming, P., 2010b. Endogenous competition against binding of [(18)F]DMFP and [(18)F]fallypride to dopamine D(2/3) receptors in brain of living mouse. Synapse 64, 313–322. Rousset, O.G., Zaidi, H., 2006. Correction for partial volume effects in emission tomography. In: Zaidi, H. (Ed.), Quantitative Analysis in Nuclear Medicine Imaging. Springer, New York, pp. 236–271. Rousset, O.G., Ma, Y., Evans, A.C., 1998. Correction for partial volume effects in PET: principle and validation. J. Nucl. Med. 39, 904–911. Rousset, O.G., Collins, D.L., Rahmim, A., Wong, D.F., 2008. Design and implementation of an automated partial volume correction in PET: application to dopamine receptor quantification in the normal human striatum. J. Nucl. Med. 49, 1097–1106. Ryu, Y.H., Liow, J.S., Zoghbi, S., Fujita, M., Collins, J., Tipre, D., Sangare, J., Hong, J., Pike, V.W., Innis, R.B., 2007. Disulfiram inhibits defluorination of (18)F-FCWAY, reduces bone radioactivity, and enhances visualization of radioligand binding to serotonin 5-HT1A receptors in human brain. J. Nucl. Med. 48, 1154–1161. Schweinhardt, P., Fransson, P., Olson, L., Spenger, C., Andersson, J.L., 2003. A template for spatial normalisation of MR images of the rat brain. J. Neurosci. Methods 129, 105–113. Shetty, H.U., Zoghbi, S.S., Simeon, F.G., Liow, J.S., Brown, A.K., Kannan, P., Innis, R.B., Pike, V.W., 2008. Radiodefluorination of 3-fluoro-5-(2-(2-[18F](fluoromethyl)thiazol-4-yl)ethynyl)benzonitrile ([18F]SP203), a radioligand for imaging brain metabotropic glutamate subtype-5 receptors with positron emission tomography, occurs by glutathionylation in rat brain. J. Pharmacol. Exp. Ther. 327, 727–735. Shrestha, S., Hirvonen, J., Hines, C.S., Henter, I.D., Svenningsson, P., Pike, V.W., Innis, R.B., 2012. Serotonin-1A receptors in major depression quantified using PET: controversies, confounds, and recommendations. Neuroimage 59, 3243–3251. Skinbjerg, M., Liow, J.S., Seneca, N., Hong, J., Lu, S., Thorsell, A., Heilig, M., Pike, V.W., Halldin, C., Sibley, D.R., Innis, R.B., 2010. D2 dopamine receptor internalization prolongs the decrease of radioligand binding after amphetamine: a PET study in a receptor internalization-deficient mouse model. Neuroimage 50, 1402–1407. Slifstein, M., Hwang, D.R., Huang, Y., Guo, N., Sudo, Y., Narendran, R., Talbot, P., Laruelle, M., 2004. In vivo affinity of [18F]fallypride for striatal and extrastriatal dopamine D2 receptors in nonhuman primates. Psychopharmacology (Berl) 175, 274–286. Slifstein, M., Kegeles, L.S., Xu, X., Thompson, J.L., Urban, N., Castrillon, J., Hackett, E., Bae, S.A., Laruelle, M., Abi-Dargham, A., 2010. Striatal and extrastriatal dopamine release measured with PET and [(18)F] fallypride. Synapse 64, 350–362. Tantawy, M.N., Jones, C.K., Baldwin, R.M., Ansari, M.S., Conn, P.J., Kessler, R.M., Peterson, T.E., 2009. [(18)F]Fallypride dopamine D2 receptor studies using delayed microPET scans and a modified Logan plot. Nucl. Med. Biol. 36, 931–940.

1468

P. Millet et al. / NeuroImage 62 (2012) 1455–1468

Vandehey, N.T., Moirano, J.M., Converse, A.K., Holden, J.E., Mukherjee, J., Murali, D., Nickles, R.J., Davidson, R.J., Schneider, M.L., Christian, B.T., 2009. High-affinity dopamine D2/D3 PET radioligands 18F-fallypride and 11C-FLB457: a comparison of kinetics in extrastriatal regions using a multiple-injection protocol. J. Cereb. Blood Flow Metab. 30, 994–1007. Vucckovic, M.G., Li, Q., Fisher, B., Nacca, A., Leahy, R.M., Walsh, J.P., Mukherjee, J., Williams, C., Jakowec, M.W., Petzinger, G.M., 2010. Exercise elevates dopamine D2 receptor in a mouse model of Parkinson's disease: in vivo imaging with [(1)F] fallypride. Mov. Disord. 25, 2777–2784.

Wu, H.M., Hoh, C.K., Choi, Y., Schelbert, H.R., Hawkins, R.A., Phelps, M.E., Huang, S.C., 1995. Factor analysis for extraction of blood time–activity curves in dynamic FDG-PET studies. J. Nucl. Med. 36, 1714–1722. Xi, W., Tian, M., Zhang, H., 2011. Molecular imaging in neuroscience research with small-animal PET in rodents. Neurosci. Res. 70, 133–143. Yoder, K.K., Mock, B.H., Zheng, Q.H., McCarthy, B.P., Riley, A.A., Hutchins, G.D., 2011. Assessment of IP injection of [(18)F]fallypride for behavioral neuroimaging in rats. J. Neurosci. Methods 196, 70–75.