The study was approved by the Danish Medicines Agency, the Ethics Committee of Aarhus Municipality, and the Committee for Good Clinical Practice of Aarhus University Hospital. We used five males (ranges 37 to 66 years old, 70 to 94 kg) who gave informed consent to participate in the study after receiving a written and oral account of the project. They were currently in good general health with no indication of past or present mental illness.
For brain imaging, we used an ECAT EXACT HR PET camera (CTI/Siemens, Knoxville, TN, USA) with a radiation shield located on each side of the neck (NeuroShield®, Scanwell Systems, Montreal, Canada). After a transmission scan, subjects received an intravenous injection of [N-methyl-11C]mirtazapine (ranges: radioactivity injected = 175 to 413 MBq, specific activities = 13 to 67 GBq/μmol, stable mirtazapine dosage = 15 to 50 μg) at the start of a 60-min dynamic PET scan of 28 frames (6 × 10 s, 4 × 30 s, 7 × 60 s, 5 × 120 s, 4 × 300 s, 2 × 600 s) recorded in 3D mode. PET data were reconstructed using filtered backprojection and a Hanning filter with a cutoff frequency of 0.5 per cycles, resulting in a special resolution (FWHM) of about 5 mm. Correction for attenuation was based on a transmission scan. The dynamic PET data were decay-corrected to the scan start.
Radiochemistry, blood chemistry, and metabolite analysis
[N-methyl-11C]Mirtazapine was prepared from (±)-N-desmethyl mirtazapine (Z)-2-butenedioate, and analytical high-performance liquid chromatography [HPLC], determination of radiochemical purity, and product identity were done as described elsewhere [9, 11]. Thirty-five blood samples (18 × 10 s, 4 × 30 s, 5 × 1 min, 7 × 5 min, 1 × 15 min) were obtained manually from an antecubital artery and were decay-corrected to the scan start. The fraction of unchanged [N-methyl-11C]mirtazapine in the plasma was determined with radiodetection by integration of the peak corresponding to the radiopharmaceutical identity and was expressed as a percentage of the total of all radioanalytes recovered by HPLC. Seven radiochemical fractionations of extracts of plasma samples were measured at 1, 2.5, 5, 15, 25, 40, and 60 min. A double-exponential function was fitted to these measurements and was used to estimate the continuous time-course of the radiochemical fractions of [N-methyl-11C]mirtazapine needed to calculate the metabolite-corrected arterial input function.
The data of the dynamic [N-methyl-11C]mirtazapine scan were summed for each subject, and each summed image was coregistered automatically using a software based on the medical image NetCDF [MINC] programming package developed at the Montreal Neurological Institute [MNI]. Briefly, the summed PET scans were converted into the MINC format and were linearly registered to the MNI/International Consortium for Brain Mapping [ICBM] 152 T1 brain template . The transforms were concatenated to produce the transformation used for bringing the dynamic PET images into the MNI/ICBM 152 common standardized space.
Representative regions of interest were obtained automatically from each subject's data by a custom-made software and a segmented atlas of the human brain . Time-activity curves [TACs] were generated from the dynamic PET study for five regions: the cerebellum (region 1), striatum (region 2), hippocampus (region 3), frontal lobe (region 4), and thalamus (region 5).
Time-activity curves for each subject were analyzed using six kinetic methods: (A) single-tissue compartment model with uncorrected and metabolite-corrected arterial plasma input functions, (B) two-tissue compartment model with uncorrected and metabolite-corrected arterial plasma input functions, (C) graphical plasma input model with metabolite-corrected arterial plasma input function , (D) graphical reference tissue model with a cerebellum TAC , (E) reference tissue model with a cerebellum TAC , and (F) simplified reference tissue model with a cerebellum TAC . Methods A and B use metabolite-corrected arterial plasma curves as input function to the kinetic model, and uncorrected arterial plasma curve including metabolites for the blood volume. The reference tissue models, namely methods D, E, and F, use a cerebellum TAC instead of plasma input functions and do not require blood sampling. Method D can be applied with or without a k
2 correction ; we excluded the correction to maintain a linear method without assumptions about the k
All models can be described in terms of microparameters: K
1 (ml ml-1 min-1) denotes the influx rate constant of the parent compound from the plasma to the free tissue compartment; k
2 (min-1) is the rate constant of transfer from the free to the plasma compartment; k
3 (min-1) is the rate constant for transfer from the free to the bound compartment; k
4 (min-1) is the rate constant for transfer from the bound to the free compartment; and V
0 (ml ml-1) is the fractional blood volume in the brain. In methods A and B, we assumed a fixed fractional blood volume of 7% . Estimates of microparameters may be uncertain due to noise. However, data analyses of receptor studies focus on physiologic macroparameters, such as distribution volumes [V
T] (the ratio at equilibrium of the tracer concentration in the tissue to that in the plasma) and binding potentials [BPND] (the ratio at equilibrium of a specifically bound tracer to that of a non-displaceable tracer in the tissue), which are more stable and can be derived in terms of the microparameters. Method A provides estimates of the distribution volumes (V
T = K
2). In addition, indirect estimates of binding potentials were calculated for the binding regions by relating fitted values for the distribution volume in the binding region to that of the reference region, assuming that distribution volume of the non-displaceable compartment [V
ND] in the receptor-deficient reference region and in the receptor-rich binding region are equal:
Method B provides estimates of the distribution volumes (V
T = (K
2) (1 + k
4)) and binding potentials (BPND = k
4). Method C provides estimates of V
T as the slope of a linear regression to the late linear part of the Logan representation. For method C, BPND can be indirectly calculated according to Equation 1. Method D provides estimates of the distribution volume ratio V
ND as the slope of a linear regression from which the binding potential (BPND = V
ND - 1) is derived (Equation 1). Methods E and F directly include BPND = k
4 as a model parameter.
It has been shown for neuroreceptor modeling that weights should not be based on noisy TACs and that uniform weighting is recommended if nothing is known about the noise of the measurements . We tested two simple weighting schemes by comparing kinetic parameters estimated by nonlinear regression with uniform weighting and with weighting by frame duration. Goodness-of-fit was measured by the Akaike criterion . Parameter estimates may fluctuate considerably when fitted by the nonlinear methods A, B, E, and F. We report the best fits and their corresponding parameter estimates that represent the best mathematical representation of the data as found by an automatic optimization routine . For noisy data, the resulting parameters can depend on the initial guess due to local minima, which may be unphysiologic and even include negative microparameters that are not compatible with the kinetic model. In these cases, the data analysis is less straightforward since quality control of the fits is needed. In this study, our only exclusion criterion was the negative parameters, and those fits were remade with a different initial guess. Otherwise, we report parameters from the fits that yielded the lowest Akaike value. Except for the non-negativity constraint, we did not introduce subjective upper or lower limits for parameter estimates. In reality, we only had problems with local minima when using Method B that led to estimates of k
3 and k
4 that were particularly unreliable; its sensitivity to noise was systematically dealt with by making 20 fits using randomized initial guesses and reporting the parameters from the fit with the lowest Akaike value with non-negative parameters. For the other methods, we would get the same physiologically reasonable parameter estimates using any reasonable initial guess. Thus, the extensive procedure using 20 fits was not necessary for the other methods.
We used nonparametric tests (chi-square test, Kruskal-Wallis H test, Mann-Whitney U test, and Spearman's rho) with Bonferroni correction for multiple comparisons for determining the statistical significance of the results.