Seven healthy, nonsmoker, male volunteers (72 ± 9.5 kg, age 26 ± 4.5 years) from a previous clinical protocol were included in this study 7 . All subjects were free of current medical or psychiatric illnesses and had no history of drug or alcohol abuse. Subjects were instructed to avoid monoamine-rich food 3 days before and alcohol consumption 1 week before imaging. Vital signs (blood pressure, heart rate) were recorded before [¹¹C]befloxatone injection and at the end of the scan. The study was approved by the regional ethics committee for biomedical research at the Bicêtre Hospital. Each subject gave a written informed consent.

PET scans were performed on an ECAT EXACT HR + scanner (Siemens Medical Solutions, Knoxville, TN, USA). One transmission scan was acquired before intravenous bolus injection of [¹¹C]befloxatone (290 ± 24.4 MBq). The specific activity was 35.2 ± 14.9 GBq/μmol, which is within the range of the values usually obtained from the synthesis of [¹¹C]befloxatone 11 and sufficient to rule out any mass effect. Even when considering a conservative B _max value of 99 pmol/mL (as found in vitro in the human cerebellum 14 ), the B _max occupancy is well below 1% in each subject. Dynamic three-dimensional images were acquired over 60 min in 29 frames, whose duration ranged from 10 s to 5 min. Images were reconstructed with filtered back projection.

Each subject also underwent brain T1-weighed three-dimensional magnetic resonance imaging (MRI) using a 1.5 T Signa scanner (GE Healthcare, Milwaukee, WI, USA). MRI images were acquired with a voxel size of 0.938 × 0.938 × 1.3 mm³ and a thickness of 1.3 mm in axial slices.

Arterial blood samples were drawn from the radial artery of each subject in 10-s intervals until 2 min, followed by samples at 3, 4, and 5 min and then every 5 min until the end of the scan at 60 min. About 27 samples were drawn for each subject. For each blood sample, the decay-corrected concentration of [¹¹C]befloxatone was measured in whole blood and plasma. Plasma was obtained by centrifugation. Plasma-free fraction was not measured. Venous blood samples were simultaneously drawn from the cubital vein of the opposite arm, using the same number of samples and time schedule as the arterial ones. As for arterial blood, plasma was separated from total venous blood. Arterialization of venous blood was achieved by heating the arm with hot water bags, starting 30 min before ligand injection until the end of the scan. The bags were replaced every 20 to 30 min. Four or five times during each scan, blood gases (pO₂ and pCO₂) were measured in five subjects.

The whole plasma time-activity curves were taken as input functions because in a previous series of (yet unpublished) four human scans from our laboratory, chromatograms obtained with high-performance liquid chromatography (HPLC) did not show any detectable radiometabolites in the plasma, even at late time points. The technique we used to measure the amount of unchanged radiotracer in the plasma was similar to that described in 14 . Specifically, after deproteinization with acetonitrile, the samples were centrifuged and the supernatant was injected directly into the HPLC column. A reverse-phase μBondapak C18 column (300 × 7.8 mm, 10 μm; Waters, Milford, MA, USA) was eluted applying a gradient from 20% acetonitrile in 0.01 M phosphoric acid up to 80% in 5.5 min, to 90% at 7.5 min, and returning to 20% at 7.6 min with a total run length of 10 min. The flow rate of the eluent was maintained at 6 mL/min. Befloxatone was eluted with a retention time of 6 min. Data acquisition and analysis were carried out using Winflow software (version 1.21, JMBS Developments, Grenoble, France).

The summed PET image, obtained by averaging all frames, was first coregistered to the individual MRI using SPM5 (Wellcome Department of Cognitive Neurology, London, UK). MRI and PET images were then normalized to the Montreal Neurologic Institute (MNI) space using the transformation parameters from the MRI images. A template of regions of interest (ROIs) 15 was used to extract brain time-activity curves for the following 13 regions: anterior and posterior cingulate cortices, caudate, putamen, thalamus, pons, cerebellum, hippocampus, parahippocampus, and the prefrontal, parietal, temporal, and occipital cortices. Each region was obtained from the weighted average of the left and right region. Image analysis and kinetic modeling were performed with Pmod 3.1 (Pmod Technologies, Zurich, Switzerland).

Because [¹¹C]befloxatone is a reversible MAO-A inhibitor 11 , kinetic analysis was performed using one- (1TCM) and two-tissue compartment (2TCM) model to calculate the total distribution volume (V _T), which equals the ratio at equilibrium of the concentration of radioligand in the brain to that in the plasma 16 . V _T is the sum of the binding in the specific compartment (V _S) and in the nondisplaceable compartment (V _ND). In theory, V _T may change not only as a function of V _S but also of V _ND. However, in kinetic modeling studies, V _ND is commonly assumed to be constant across different brain regions and across different subjects of the same species. Thus, variations of V _T are considered to reflect variations of V _S values. For [¹¹C]befloxatone in particular, in vivo studies in primates showed that, after pharmacological blockade, the residual radioactive concentrations were very low and identical in all brain structures 11 . The 2TCM analysis was performed in two ways, (1) without constraining the rate constants and (2) fixing the ratio K ₁/k ₂ to the value obtained in the whole brain, to improve the identifiability of parameters. The individual rate constants were estimated with the weighted least square method and with the Marquardt optimizer. Brain activity was corrected for its vascular component using the measured whole blood concentrations, assuming that the cerebral blood volume is 5% of the total brain volume 17 . The delay of radiotracer arrival between the radial artery and the brain was corrected by fitting the whole brain. The plasma input function was modeled using a linear interpolation of the [¹¹C]befloxatone concentrations before the peak and a tri-exponential fit of concentrations after the peak. Kinetic modeling was also performed at the regional level using the graphical Logan plot (Logan_voi), Ichise's multilinear analysis (MA1_voi), and a standard spectral analysis (SA_voi).

Goodness of fit by nonlinear least squares analysis was evaluated using the Akaike information criterion (AIC) and model selection criterion (MSC). The most appropriate model is that with the smallest AIC and the largest MSC score. Goodness of fit by the one- and two-compartment models was compared with F statistics 23 . A value of P < 0.05 was considered significant. The identifiability of kinetic variables was expressed as a percentage of the rate constant and calculated as standard error obtained from the diagonal of the covariance matrix 24 . Standard error for V _T was calculated from the covariance matrix using the generalized form of error propagation equation 25 , where correlations among parameters are taken into account. Smaller values indicate better identifiability.

For each subject, regional V _T values obtained with 2TCM at the regional level were considered as the gold standard. Comparisons to the V _T calculated with Logan_voi, MA1_voi, SA_voi, Logan_voxel, MA1_voxel, and SA_voxel were performed with a repeated measures analysis of variance (ANOVA) for V _T values in the various regions. Statistical analyses were performed with SPSS (version 17 for Windows; SPSS Inc., Chicago, IL, USA).

[¹¹C]Befloxatone concentrations in arterial plasma peaked about 30 s after injection at 21 ± 7.1 standard uptake value (SUV) and then rapidly decreased. Both the plasma and whole blood arterial curves were well fitted by a tri-exponential function. Venous concentrations peaked about 50 s after injection with an SUV of 6.7 ± 2.6.

Pairs of arterial and venous samples were sometimes drawn with a few seconds of delay in either direction. Therefore, in order to compare arteriovenous concentrations of the radioligand, we fitted the plasma time-activity curves with a tri-exponential function and interpolated the values at the same grid of standard times. The average pO₂ measured in venous blood was 77% ± 8% of the pO₂ measured in the arterial blood. The average pCO₂ in the vein was 101% ± 1% compared to the arterial value. A stable arteriovenous equilibrium of [¹¹C]befloxatone concentrations was found only a few minutes after injection (Figure 1). Average arteriovenous differences were already at about 5% or less within 2.5 min after injection and remained unchanged until the end of the scan (Figure 2). After the peak, in each individual subject the arteriovenous difference was almost always less than 10% at each time point. One subject, however, had differences between 10% and 20% at each time point from 40 min after injection to the end of the scan. A good correlation was found between arterial and venous concentrations at each time point from 3 to 60 min (n = 13 pairs). R ² ranged from 0.64 at 3 min to 0.99 at 30 min.

Radioactivity showed a rapid uptake in the brain, progressively increased during the first 25 min (mean peak value, 1.5 to 2 SUV), and then slowly decreased (Figure 3). These time-activity curves were very similar to those described for [¹¹C]befloxatone in nonhuman primates 12 .

Radioactivity uptake was widespread and distributed fairly homogeneously across the brain (Figure 4). Consistent with the known distribution of MAO-A 2 , a lower uptake was noted in the cerebellum of all subjects (the cerebellar area under the curve was about 35% lower than the mean value of the whole brain). However, because low amounts of MAO-A are expressed in the cerebellum 26 , analysis with a reference region was not used.

The brain kinetic analysis had two major results: (1) [¹¹C]befloxatone binding in the brain was better described by a two-tissue compartment model but the 2TCM with a constraint of K ₁/k ₂ did not provide a better goodness of fit for V _T, and (2) both at the regional and voxel level, SA provided better V _T estimates than Logan and MA1.

As regards the first finding, unconstrained one- and two-compartment fitting converged in all regions and in all scans. Fitting was visually quite similar using either model; however, the two-tissue compartment model showed lower mean AIC (-22.3) and higher mean MSC (4.2) scores than the one-compartment model (AIC = 1.1, MSC = 3.4). The superiority of 2TCM was confirmed by F statistics. In addition, SA showed at least two equilibrating components in all regions for all subjects. The average identifiability across brain regions for two-compartment fitting was 4.7%. The average V _T value across all brain regions was 14.3 ± 3.0 mL/cm³. The highest binding was found in the thalamus (average V _T of 19.3 ± 1.4 mL/cm³), and the lowest in the cerebellum (6.7 ± 0.9 mL/cm³) (Table 1). The constrained two-compartment model fitting converged in all regions and in all scans but provided a similar goodness of fit (AIC = -20.8 and MSC = 4.1) as the unconstrained model. The average identifiability across brain regions was 4.4% (versus 4.7% for the unconstrained fitting).

Values are presented as mean ± SD from seven human subjects. For each brain region, standard errors are listed in parentheses and are expressed as the percentage of the variable itself.

As regards the second point, Logan_voi and MA1_voi underestimated V _T values by about 10% (Figure 5A), and these differences were statistically significant (p < 0.05; factorial repeated measures ANOVA using Bonferroni adjustment, F _1.33,8.01 = 57.93). In contrast, V _T values obtained with SA_voi had an average bias of only +1% (range -7% to +6%) and were not statistically different from the reference values obtained with 2TCM (p > 0.05). Logan_voi, MA1_voi, and SA_voi correlated well with the reference V _T values obtained with the 2TCM: R ² was >0.90 for all subjects, except for one subject in whom R ² for SA_voi was 0.73. Logan_voxel severely underestimated V _T, with an average difference of -40% (p < 0.05) (Figure 5B). Moreover, linear regression analysis showed that Logan_voxel was significantly correlated to 2TCM V _T values in only four of seven subjects. The parametric images obtained with MA1 (MA1_voxel) contained many noisy voxels that prevented reliable quantification. These noisy voxels were located predominantly, but not exclusively, in areas where blood flow (K ₁) is low, such as the white matter. Attempts to eliminate these noisy voxels by adjusting K ₁ values were not successful. Therefore, MA1_voxel was not considered for further analysis. Compared to 2TCM, V _T values obtained with SA_voxel showed a mean bias of only +2% (range, -10% to +7%) and were not statistically different (p > 0.05) from the reference values. Predictably, SA_voxel values were higher than Logan_voi, Logan_voxel, and MA1_voi (p < 0.05). Finally, SA_voxel significantly correlated with 2TCM in all subjects: R ² for brain regions ranged from 0.81 to 0.95, except in one subject who had an R ² of 0.60.

Logan-V _T results obtained with a full venous input function were consistent with Logan-V _T results obtained with an arterial plasma input function in six subjects (with an error <10% compared to the reference values obtained with the arterial input function). However, in one subject, Logan-V _T was overestimated by 27%.