Sage Journals: Discover world-class research

Abstract

Adenosine is an important neuromodulator. Basic cerebral effects of adenosine are exerted by the A₁ adenosine receptor (A₁AR), which is accessible in vivo by the novel ligand [¹⁸F]8-cyclopentyl-3-(3-fluoropropyl)-1-propylxanthine ([¹⁸F]CPFPX) and positron emission tomography (PET). The present study investigates the applicability of kinetic models to describe the cerebral kinetics of [¹⁸F]CPFPX in order to quantify A₁AR density in vivo. Six healthy volunteers underwent dynamic PET scanning and arterial blood sampling after bolus injection of [¹⁸F]CPFPX. For quantitative analysis, a standard two-tissue compartment model (2TCM) was compared with a one-tissue compartment model (1TCM) and Logan's graphical analysis (GA). The 2TCM described the cerebral kinetics of [¹⁸F]CPFPX significantly better than the 1TCM (in all regions and subjects examined). The estimated values of the regional total distribution volumes (DV_t) correlated strongly between the 2TCM and GA (linear regression r² = 0.99, slope: 1.007). The DV_t correlation between the 2TCM and the 1TCM was comparably high, but there was a significant bias towards lower DV_t estimates given by the 1TCM (r²: 0.99, slope: 0.929). It is concluded that a 2TCM satisfactorily accounts for the cerebral kinetics of [¹⁸F]CPFPX. GA represents an attractive alternative method of analysis.

Keywords

Positron emission tomography Kinetic modeling

The purine adenosine is an important neuromodulator, beside its central role in cellular energy metabolism and genetic encoding. Four adenosine receptors (A₁, A_2A, A_2B, and A₃) have been cloned and expressed (Dunwiddie and Masino, 2001; Fredholm et al., 2001). The G_i/o protein coupled A₁ adenosine receptor (A₁AR) is pre- and postsynaptically localized and expressed throughout the human brain with a dense and heterogeneous distribution (Fastbom et al. 1987a; Svenningsson et al., 1997; Bauer et al., 2003a). It acts through a number of effectors that inhibit adenylate cyclase activity, reduce Ca²⁺ influx and promote a hyperpolarizing outward K⁺ current. Collectively, these actions are tonically neuromodulatory and reduce neurotransmitter release and neuronal exciility.

There are numerous reports on chronic alterations of A₁AR density in neurological and psychiatric disorders (for review see Dunwiddie and Masino, 2001). A₁AR activation is postulated to be protective against glutamate excitotoxicity, as extracellular adenosine levels increase during hypoxia, ischemia, and excessive neuronal activity (Dunwiddie and Masino, 2001; Haas and Selbach, 2000). A₁AR density is reduced in the temporal cortex of patients with temporal lobe epilepsy (Glass et al., 1996) as well as in the hippocampus of patients suffering from dementia (Ulas et al., 1993; Deckert et al., 1998). Thus, its wide distribution and involvement in a variety of physiological and pathological conditions make the A₁AR an attractive target for in vivo imaging and quantification.

The fluorine-18 labeled A₁AR antagonist [¹⁸F]8-cyclopentyl-3-(3-fluoropropyl)-1-propylxanthine ([¹⁸F]CPFPX) was recently developed as a PET ligand for in vivo imaging of cerebral A₁ARs (Holschbach et al., 1998; Holschbach et al., 2002). [¹⁸F]CPFPX shows nanomolar in vitro affinity to cloned human A₁ARs (K_D = 1.26 nmol/L) and a high A₁AR selectivity over other cerebral receptors including the A_2AAR (K_D = 940 nmol/L). [¹⁸F]CPFPX can be synthesized with high radiochemical yield (45%), purity (> 98%) and specific radioactivity (> 270 GBq/μmol [> 7.3 Ci/μmol]) (Holschbach et al., 2002). The cerebral distribution of [¹⁸F]CPFPX found ex vivo in mice (Holschbach et al., 2002) and rats (Bauer et al., 2003b) as well as in in vivo PET studies in baboons (Boy et al., 1998) corresponded to the known distribution of A₁ARs. In mice, rats, baboons, and humans, at least two polar, species-specific metabolites are rapidly formed, which are highly unlikely to the cross the blood-brain barrier due to their polarity. In fact, brain homogenate extracts from mice (60 minutes post-injection) demonstrated >98% unchanged ligand (Holschbach et al., 2002). The first A₁AR PET study in humans showed a cerebral distri-bution pattern of [¹⁸F]CPFPX that is in agreement with human post mortem A₁AR autoradiography (Bauer et al., 2003a).

The present study is intended to evaluate the use of PET kinetic compartment models (one- and two-tissue compartment model) (Mintun et al., 1984) and of Logan's graphical analysis method (Logan et al., 1990) to describe the cerebral kinetics of [¹⁸F]CPFPX and to derive quantitative measures of A₁AR density after a single bolus injection of [¹⁸F]CPFPX.

MATERIALS AND METHODS

Subjects

All procedures were approved by the Ethics Committee of the Medical Faculty of the University of Düsseldorf, Germany, and the German Federal Office for Radiation Protection. Six healthy male volunteers (mean age = 31.89 ± 8.46 years, range = 21.9–40.5 years) were included after giving written informed consent. Volunteers were screened for history of neurological and psychiatric diseases, head injury, alcohol and substance abuse. A review of current medications excluded the intake of substances interfering with the adenosinergic system (e.g., theophylline). Volunteers avoided caffeine intake for at least 24 hours prior to the PET studies.

To exclude structural brain abnormalities and to define regions of interest (ROI), individual high-resolution magnetic resonance imaging (MRI) data sets were acquired (Magnetom VISION 1.5 T, Siemens, Erlangen, Germany) using a 3D T1-weighted magnetization-prepared rapid-acquisition gradient-echo sequence (voxel size = 1×1×1 mm³).

PET Acquisition

Radiosynthesis and formulation of [¹⁸F]CPFPX was performed as previously described (Holschbach et al., 2002). At time of injection the mean specific radioactivity was 72.17 ± 58.74 GBq/μmol (range = 18.66–159.80 GBq/μmol) (1.950 ± 1.587 Ci/μmol [range = 0.504 ± 4.319 Ci/μmol]).

All PET measurements were performed in 3D mode on a Siemens ECAT EXACT HR+ scanner (Siemens-CTI, Knoxville, TN, USA) equipped with a circular lead shield (Neuroinsert) to reduce scatter radiation from outside the field of view. Scanning took place as the subjects laid supine in a quiet ambience with their heads immobilized in canthomeatal orientation by means of polyurethane foam. A 10 minutes ⁶⁸Ge/⁶⁸Ga-transmission scan was acquired to correct for attenuation. Sixty-three slices of 2.425 mm slice thickness (pixel size = 2 × 2 mm²) were reconstructed per time frame by filtered back-projection (Shepp filter, filter width = 2.5 mm) after FORE rebinning (Defrise et al., 1997) into 2D sinograms. Data sets were fully corrected for random coincidences, scatter radiation and attenuation. Head positions were permanently monitored and, if necessary, manually corrected using a video system and reference skin marks.

A venous catheter in a forearm vein served for tracer injection. An arterial catheter was inserted into the left distal radial artery under local anesthesia and used for arterial blood sampling. Dynamic PET acquisition (90 minutes) in list mode began with injection of [¹⁸F]CPFPX over 20 seconds. Mean injected radioactivity was 273.5 ± 12.0 MBq (range = 251.6 −284.9 MBq) (7.39 ± 0.32 mCi [range = 6.80–7.70 mCi]) in a volume of 10 mL sterile saline. The mass of injected CPFPX was 6.54 ± 4.76 nmol (range = 1.69–14.7 nmol, equal to 0.54–4.73 μg CPFPX). During the first 90 seconds after [¹⁸F]CPFPX injection, heparinized arterial blood samples (4 mL) were withdrawn every 6 seconds. Thereafter, blood samples were collected at 1:45, 2, 3, 4, 6, 8, 10, 15, 20, 30, 45, 60, 75 and 90 minutes post-injection. An additional blood sample was taken before [¹⁸F]CPFPX administration to assess the fraction of free (unbound to plasma proteins) [¹⁸F]CPFPX in plasma (denoted by f₁), the extraction fraction of pure [¹⁸F]CPFPX (E₀) and the plasma caffeine level. List mode data was reframed into a dynamic sequence of 6 × 10 seconds, 3 × 20 seconds, 3 × 30 seconds, 4 × 60 seconds, 3 × 150 seconds, and 15 × 300 seconds frames.

Metabolite Analysis

Centrifugation (1000 × g, 3 minutes) separated plasma from whole blood. Radioactivity in whole blood and plasma samples (500 μL), respectively, was measured in an automated gamma counter (1480 WIZARD, Wallac-ADL GmbH, Freiburg, Germany), that is weekly cross-calibrated with the PET scanner using a ⁶⁸Ge/⁶⁸Ga-phantom. Whole blood and plasma time-activity curves were decay-corrected for the time of injection.

Metabolite analyses were performed in arterial plasma samples taken at 1 minute, 2 minutes and all sampling times thereafter. Aliquots (0.1 mL) of plasma samples were diluted with methanol / acetonitrile (50/50, v/v, 0.1 mL), vortexed for 20 seconds at room temperature, and centrifuged for 1 minute at 20,000 × g. Aliquots (5 μL) of the clear supernatant were spotted onto thin-layer chromatography (TLC) plates (SIL G-25 UV₂₅₄ glass plates, Macherey-Nagel, Düren, Germany) and developed with ethyl acetate / hexane 75 / 25 (v / v). An InstantImager (Canberra-Packard GmbH, Dreieich, Germany) was used for TLC analysis.

The total radioactivity extraction fraction of each plasma sample (E_i) was determined by spotting 5 μL (in triplicate) of the clear supernatant and of the untreated plasma onto a TLC plate, which was read out with the same system. The same procedure was applied to plasma samples obtained before tracer injection to assess the extraction fraction of added [¹⁸F]CPFPX (E₀).

The time course of the fraction of total plasma activity representing the intact ligand (L_TLC(t)) was described by fitting the parameters p₁ to p₄ in Eq. 1 (modified from Watabe et al. (2000)) to the fraction of intact [¹⁸F]CPFPX in the plasma samples given by TLC:

L_{T L C} (t) = \frac{1}{(1 + p_{2} \cdot (t - p_{1})^{p 3})^{p 4}}

(1)

For all t <p₁, L_TLC(t) was set to 1 (i.e., 100% unchanged ligand in plasma, usually during the first 1–2 minutes). The additional inclusion of the delay parameter p₁ (restricted to p₁ > 0 minute) and a variable p₃ instead of using a fixed value (as done by Watabe et al. (2000) for [¹¹C]MDL100,907) led to a considerably improved fit of the TLC data in case of [¹⁸F]CPFPX.

There was a only slight decrease of the total radioactivity extraction fraction over time (falling from virtually 100% to approximately 85–90% at 90 minutes), which precluded to identify all four parameters when fitting Eq. 1 to the time course of the ratio E_i/E₀ (Fig. 1). Thus, parameters p₁ and p₃ were empirically set constant (p₁ = 0 minute and p₃ = 2), which allowed reasonably good fits (see Fig. 1). To correct the TLC data for the time dependent decrease in extraction efficiency, the fits of the TLC data were multiplied with the respective fits to the E_i/E₀ ratio. The product was subsequently used to generate metabolite corrected plasma input functions.

FIG. 1.

Representative Metabolite Analysis. Closed diamonds (left y-axis) show the time course of the fraction of the intact [¹⁸F]CPFPX measured by thin-layer chromatography in arterial plasma of a representative healthy volunteer (same subject as in Fig. 2, Fig. 3 and Fig. 4). The solid line represents the respective non-linear regression fit. Open diamonds (right y-axis) denote the corresponding total radioactivity extraction at each time point in relation to the total radioactivity extraction in a blood sample, which was in vitro incubated with [¹⁸F]CPFPX. The broken line represents the respective non-linear regression fit. The time course of intact ligand in arterial plasma, corrected for the slight decrease in extraction efficacy, is given by the product of both fits.

FIG. 2.

Cerebral Distribution of [¹⁸F]CPFPX in Vivo. Individual MRI (upper panel), summed PET (5.5–60 minutes post-injection, accumulated activity in percent injected dose per liter) (middle panel) and PET-MRI-overlay (lower panel) of a representative healthy volunteer (same subject as shown in Fig. 1, Fig. 3, and Fig. 4). The five representative transaxial slices depict the high [¹⁸F]CPFPX accumulation in cortical and subcortical grey matter regions. Note the low accumulation of [¹⁸F]CPFPX in cerebellum, brainstem and white matter. (%ID/L, percent injected dose per liter)

FIG. 3.

Plasma input function and whole brain uptake of [¹⁸F]CPFPX. Closed diamonds (left y-axis, logarithmic scale) denote the metabolite corrected arterial plasma input function of [¹⁸F]CPFPX in a representative healthy volunteer (same subject as in Fig. 1, Fig. 2 and Fig. 4). Open diamonds (right y-axis) show the corresponding time course of the whole brain [¹⁸F]CPFPX uptake (not corrected for blood volume).

FIG. 4.

Representative Time-Activity Curves, Compartment Model Fits and Graphical Analyses Plots. Scatter plots on the left depict the time-activity curves (without blood volume correction) for cerebellum (top), lateral temporal cortex (middle) and putamen (bottom) of a representative healthy volunteer (same subject as in Fig. 1, Fig. 2 and Fig. 3). The broken line represents the curve fit given by a one-tissue compartment model, the solid line represents the curve fit of the two-tissue compartment model. Plots on the right show the corresponding graphical analyses. The slopes of the linear regression lines equal the total distribution volume of [¹⁸F]CPFPX in the respective region (after blood volume correction assuming a fractional blood volume of 5%).

Plasma caffeine levels were assessed by HPLC (in triplicate). Individual f₁ values were determined by ultrafiltration (in triplicate).

Image Analyses

An interactive three-dimensional image registration software (MPI-Tool version 3.35, ATV, Erftstadt, Germany) (Pietrzyk et al., 1994) was used to align the individual MRI data sets according to the anterior commissure/posterior commissure line. In each subject, a summed PET image (5.5 to 60 minutes post-injection) was manually co-registered to the realigned MRI. Registration parameters were subsequently applied to each dynamic frame. Errors of manual co-registration using the MPI-Tool software were reported to be below 2 mm (Pietrzyk et al., 1994), which compares favorable with the spatial resolution of the employed PET scanner (6 mm FWHM). Volumes of interest (VOI) were defined by freehand drawing of polygonal ROI onto the individual MR images and subsequently superimposed onto the PET frames using a dedicated software package (PMOD, version 2.45, PMOD Group, Zurich, Switzerland, www.pmod.com) (Burger and Buck, 1997; Mikolajczyk et al., 1998). Side-averaged, decay-corrected time-activity curves were generated for the following cerebral regions: dorsal prefrontal cortex (mean total volume of left and right side = 188.9 ± 22.7 mL), oribitofrontal cortex (40.4 ± 6.5 mL), lateral temporal cortex (129.2 ± 14.7 mL), mesial temporal lobe including amygdala, hippocampus and mesial temporal cortex (19.4 ± 2.0 mL), parietal cortex (87.9 ± 8.1 mL), occipital cortex (94.5 ± 9.9mL), precentral gyrus (35.0 ± 3.7 mL), postcentral gyrus (30.5 ± 3.2 mL), cingulate gyrus (32.0 ± 2.7mL), insula (18.6 ± 2.5 mL), thalamus (10.7 ± 0.6 mL), putamen (8.2 ± 0.8 mL), caudate nucleus (7.9 ± 0.7 mL), cerebellum (112.3 ± 15.6 mL), pons (14.9 ± 1.9 mL) and whole brain (1431.7 ± 88.9 mL).

Kinetic analyses

Compartmental analyses

Two-tissue compartment model

A standard two-tissue compartment model (2TCM) was used for the kinetic analyses of [¹⁸F]CPFPX binding to cerebral A₁ARs after a single bolus injection. The compartment model consists of an arterial plasma input function C_p and two tissue compartments. The first tissue compartment represents the combined compartment of the free and the non-specifically bound ligand, which are assumed to equilibrate rapidly so that they can be treated as a single compartment C_f+ns (Mintun et al., 1984). The second tissue compartment C_s consists of the ligand specifically bound to A₁ARs. The unidirectional first-order rate constants K₁ [mL·mL⁻¹·min⁻¹] and k₂ [min⁻¹] describe the transfer from C_p to C_f+ns and backwards, respectively. Likewise, k₃ [min⁻¹] and k₄ [min⁻¹] describe the transfer from C_f+ns to C_s and backwards, respectively. Eqs. 2 and 3 define k₃ and k₄:

k_{3} = k_{o n} \times {B_{max}}^{'} \times f_{2}

(2)

k_{4} = k_{o f f}

(3)

Parameters k_on [(nmol/L)⁻¹·min⁻¹] and k_off [min⁻¹] are the association and dissociation constants of the ligand-receptor complex, respectively. B_max′ [nmol/L] is the maximum concentration of available receptors (non-occupied) and f₂ [unitless] is the fraction of free ligand in the tissue (Mintun et al., 1984; Laruelle et al., 1994). The regional activity concentration measured by PET (C_PET) summarizes both tissue compartments and a fractional blood volume (V_B) component. This can be expressed as a function of time by:

C_{P E T} (t) = (1 - V_{B}) (C_{f + n s} (t) + C_{s} (t)) + V_{B} C_{a} (t)

(4)

C_a(t) represents the activity concentration in arterial whole blood at a time t. Differential equations (5) and (6) describe the change of ligand concentrations over time in the tissue compartments (Mintun et al., 1984):

(1 - V_{B}) \frac{d C_{f} + n s (t)}{d t} = K_{1} C_{p} (t) - (k_{2} + k_{3}) C_{f + n s} (t) + k_{4} C_{s} (t)

(5)

(1 - V_{B}) \frac{d C_{s} (t)}{d t} = k_{3} C_{f + n s} (t) - k_{4} C_{s} (t)

(6)

At time of equilibrium t_e, i.e., when net transfer between compartments equals zero, the left hand terms of Eq. 5 and Eq. 6 equal zero. Rearrangement of Eqs. 5 and 6 gives the following relationships between the compartmental rate constants and the equilibrium volumes of distribution of both tissue compartments C_f+ns (DV_f+ns) and C_s(DV_s) (Laruelle et al., 1994):

D V_{f + n s} = \frac{C_{f + n s} (t_{e})}{f_{1} C_{p} (t_{e})} = \frac{K_{1}}{k_{2} f_{1}}

(7)

D V_{s} = \frac{C_{s} (t_{e})}{f_{1} C_{p} (t_{e})} = \frac{K_{1} k_{3}}{k_{2} k_{4} f_{1}}

(8)

The sum of DV_f+ns and DV_s is equal to the total volume of distribution of the ligand in tissue (DV_t). Parameter f₁ denotes the fraction of free ligand in plasma.

With f₂ being equal DV⁻¹_f+ns (Abi-Dargham et al., 1994), insertion of Eqs. 2, 3 and 7 into Eq. 8 yields the definition of the binding potential (BP), which represents the ratio of B_max' to the equilibrium dissociation rate constant of the ligand-receptor complex (K_D [nmol/L]) (Mintun et al., 1984):

D V_{s} = \frac{K_{1} k_{3}}{k_{2} k_{4} f_{1}} = \frac{k_{o n} {B_{max}}^{'}}{k_{o f f}} = \frac{{B_{max}}^{'}}{K_{D}} = B P

(9)

Using the regional time-activity curves, the metabolite-corrected arterial plasma input curve, and the arterial whole blood time-activity curve, the individual rate constants K₁, k₂, k₃, and k₄ and the regional V_B-term were fitted by weighted non-linear regression analyses for each VOI in each subject using the Levenberg-Marquart least squares minimization algorithm as implemented in the PMOD software. The residuals were weighted by the inverse of the variance of the total count rate in each frame. The variance was calculated as the squared standard deviation of the total number of trues within the respective sinogram. The standard deviation of the count rate in each frame i (SD(R_i)) was calculated according to the following formula (Eq. 10) (Lammertsma, 2003), which incorporates an appropriate decay correction factor (F_i) as decay-corrected data was fitted:

S D (R_{i}) = \frac{F_{i}}{L_{i}} \cdot \sqrt{T_{i}}

(10)

where T_i represents the total number of trues in a frame of the length L_i.

Before fitting the individual rate constants, a term for the delay of the arterial input function relative to the tissue response was fitted for the respective subject's whole brain time-activity curve and kept constant throughout the following analyses. Parameter estimates generated by fitting each subject's whole brain data were used as an initial parameter set for the fitting of the regional data. It was found, that the fitting algorithm converged very reproducibly to the same or a very closely related parameter set, even if, for instance, k₂ and k₃ were varied inversely by one order of magnitude. The standard error of each parameter at convergence (taken from the respective diagonal of the covariance matrix, multiplied by the square root of the degree of freedom adjusted chi square), was taken as a measure of the precision of the estimate and expressed as percentage of the mean (coefficient of variation, %CV). Since the concentration of the total (bound and free) ligand in arterial plasma was used as an input function, binding measures not corrected for the plasma protein bound fraction of the ligand were gained (BP′ and DV_t′). BP and DV_t were calculated by dividing BP′ and DV_t′, respectively, by f₁.

One-tissue compartment model

C_f+ns and C_s may be kinetically indistinguishable if rapid equilibration occurs between both compartments (Koeppe et al., 1991). Thus, from an analytical point of view the 2TCM collapses into an one-tissue compartment model (1TCM) containing free, non-specifically and specifically bound ligand in a single tissue compartment. Under this condition, the rate constants k₂, k₃, and k₄ can be lumped together to form a composite rate constant (k₂′ [min⁻¹]) (Koeppe et al., 1991):

{k_{2}}^{'} = \frac{k_{2}}{1 + k_{3} ∕ k_{4}}

(11)

Thus, DV_t can be calculated as:

D V_{t} = \frac{K_{1}}{{k_{2}}^{'} f_{1}}

(12)

As for the 2TCM, the individual rate constants K₁ and k₂′ and the V_B-term were fitted for each VOI in each subject using the PMOD software.

Graphical analysis

In addition to the compartment analyses Logan's graphical analysis (GA) (Logan et al., 1990) was used to determine DV_t. In GA, the DV_t is given by the slope of the linear part of the following equation, which describes the dynamic of reversible ligand binding:

\frac{\int_{0}^{t} C_{P E T} (τ) d τ}{{C_{P E T}}^{(t)}} = D V_{t} \frac{\int_{0}^{t} C_{p} (τ) d τ}{{C_{P E T}}^{(t)}} + a

(13)

with C_p(t) being the input function of the free ligand (accounting for f₁) in arterial plasma. The time t∗ determines the time lag after which the intercept term a becomes constant and the relationship between the left hand and the right hand fractional expression linear. t∗ was determined for each region in each volunteer separately. Prior to GA regional time-activity curves were corrected for the fractional blood volume within each VOI assuming a V_B of 5% (Leenders et al., 1990).

Comparison of the different analyses

The goodness of fit achieved by the 2TCM and the 1TCM was compared by calculating the Akaike Information Criterion (AIC) (Akaike, 1974) and an F-test as described by Landaw and DiStefano (1984). The DV_t values provided by the 2TCM, 1TCM and GA were compared by linear regression analyses. A possible association between the regional binding estimates (DVt' and DVt) and mass of injected CPFPX and plasma caffeine levels, respectively, was explored by Spearman's non-parametric correlation coefficient.

RESULTS

Metabolite analyses

Fig. 1 shows a representative example of the time course of the [¹⁸F]CPFPX degradation as given by TLC analysis of arterial plasma samples. Subjects' metabolite data could all be well fitted by Eq. 1, the mean parameter being:p₁ = 2.1 ± 0.9 minute, p₂ = 0.07 ± 0.02, p₃ = 0.98 ± 0.28, and p₄ = 1.78 ± 1.03. The E_i/E₀ ratio fell slightly during the 90 minute scan to 0.87 ± 0.03 (Fig 1). The time course of the E_i/ E₀ ratio could also be well fitted by Eq. 1 with p₁ and p₃ being constrained to 0 minute and 2, respectively, the mean parameters being:p₂ = 0.0014 ± 0.0011 and p₄ = 0.054 ± 0.014. The mean fraction of intact ligand in arterial plasma was 50% at approximately 11 mintues post-injection and less than 10% after approximately 60 minutes. The average f₁ value was 1.18 ± 0.28%.

Cerebral uptake and distribution of [¹⁸F]CPFPX

Fig. 2 depicts the typical distribution of [¹⁸F]CPFPX in the human brain (summed image, 5.5–60 minutes post-injection), which is in agreement with the known in vitro distribution of cerebral A₁ARs (Fastbom et al. 1987a; Svenningsson et al., 1997; Bauer et al., 2003a). Fig. 3 shows a representative whole brain time-activity curve and the corresponding arterial input function. [¹⁸F]CPFPX showed a rapid brain uptake reaching a maximum of 2.3 ± 0.4% of the injected dose (%ID) (1.6 ± 0.22%ID/L) at 4.2 ± 1.1 minutes post-injection (corrected for an assumed cerebral blood volume of 5%). Depending on the regional A₁AR density, regional time-activity curves (V_B corrected) peaked approximately 2 to 8 minutes post-injection (later in regions with higher A₁AR densities), followed by a rapid washout (Fig. 4).

Kinetic analyses

Two-tissue compartment model

The results of the 2TCM are summarized in Table 1, representative curve fits are given in Fig. 4. The regional distribution and magnitude of V_B are in good agreement with the known distribution of the regional cerebral blood volume in healthy subjects (Leenders et al., 1990). The average %CV of the regional V_B estimates range 7 to 12% in cortical structures, being higher in the smaller subcortical nuclei (18–20%; exceeding 20% only in five subcortical regions). Similarly, the regional distribution of K₁ values is in good agreement with the regional cerebral perfusion pattern in healthy controls (Leenders et al., 1990). The average %CV of the K₁ and k₂ estimates range 2–7% (highest in pons; exceeding 10% in no single region) and 7–12%, respectively. Particularly k₂ was less precisely defined in subcortical structures (average %CV = 14–17%; exceeding 20% 4 times in subcortical regions and 3 times in cortical regions [insula and mesiotemporal]).

TABLE 1.

Results of the two-tissue compartment model (2TCM) analyses

Region	V_B %CV	K₁ %CV	k₂ %CV	k₃ %CV	k₄ %CV	BP′ %CV	DV′_t%CV	BP	DV_t	AIC
FRO-D	0.041 ± 0.009	0.061 ± 0.018	0.109 ± 0.016	0.016 ± 0.006	0.042 ± 0.009	0.20 ± 0.06	0.77 ± 0.22	17.9 ± 4.9	67.3 ± 18.6	−63.4 ± 20.3
	8.18 ± 3.37	2.54 ± 0.97	5.72 ± 2.45	24.98 ± 6.37	15.64 ± 2.76	9.27 ± 3.04	1.34 ± 0.31
FRO-O	0.042 ± 0.010	0.061 ± 0.018	0.108 ± 0.027	0.018 ± 0.012	0.039 ± 0.013	0.24 ± 0.06	0.82 ± 0.24	21.0 ± 6.6	71.6 ± 20.3	−45.0 ± 13.9
	10.42 ± 1.98	3.26 ± 1.02	7.28 ± 2.49	30.01 ± 7.04	20.61 ± 7.48	10.44 ± 2.52	2.00 ± 0.80
TEMP-L	0.046 ± 0.010	0.062 ± 0.017	0.101 ± 0.018	0.018 ± 0.009	0.046 ± 0.011	0.23 ± 0.06	0.85 ± 0.24	20.4 ± 5.3	73.8 ± 18.7	−53.4 ± 20.6
	8.51 ± 2.53	2.95 ± 1.17	7.25 ± 3.22	29.63 ± 6.81	17.04 ± 2.31	11.26 ± 3.41	1.38 ± 0.29
PAR	0.044 ± 0.011	0.061 ± 0.019	0.104 ± 0.014	0.014 ± 0.005	0.041 ± 0.008	0.20 ± 0.06	0.80 ± 0.24	17.8 ± 4.9	69.0 ± 17.6	−58.5 ± 19.7
	8.41 ± 3.37	2.65 ± 0.89	5.99 ± 2.19	27.94 ± 7.11	17.69 ± 3.11	10.37 ± 3.35	1.43 ± 0.25
OCC	0.046 ± 0.011	0.067 ± 0.020	0.113 ± 0.019	0.016 ± 0.006	0.039 ± 0.010	0.24 ± 0.06	0.84 ± 0.23	20.7 ± 5.8	73.5 ± 20.4	−51.7 ± 24.4
	8.65 ± 3.03	2.77 ± 1.09	6.03 ± 2.71	24.63 ± 6.15	16.21 ± 2.33	8.86 ± 2.79	1.53 ± 0.31
TEMP-M	0.047 ± 0.012	0.050 ± 0.015	0.113 ± 0.041	0.038 ± 0.027	0.052 ± 0.013	0.28 ± 0.07	0.75 ± 0.20	24.4 ± 5.6	65.2 ± 15.6	−36.6 ± 24.5
	11.13 ± 5.23	5.48 ± 2.64	15.15 ± 8.16	38.79 ± 8.42	18.21 ± 2.61	14.87 ± 4.30	1.88 ± 0.26
PREC	0.036 ± 0.008	0.059 ± 0.019	0.115 ± 0.016	0.015 ± 0.005	0.038 ± 0.008	0.20 ± 0.06	0.72 ± 0.22	17.6 ± 5.0	62.4 ± 17.1	−52.5 ± 14.1
	11.01 ± 3.48	3.08 ± 1.01	6.45 ± 2.22	27.18 ± 6.57	18.44 ± 3.57	9.59 ± 3.15	1.82 ± 0.39
POSTC	0.039 ± 0.007	0.061 ± 0.019	0.118 ± 0.020	0.015 ± 0.005	0.039 ± 0.008	0.20 ± 0.06	0.73 ± 0.23	17.5 ± 5.5	62.9 ± 17.0	−62.0 ± 25.3
	9.05 ± 3.32	2.74 ± 1.06	5.80 ± 2.28	24.41 ± 6.98	15.89 ± 3.03	8.96 ± 3.04	1.52 ± 0.37
CING	0.054 ± 0.009	0.061 ± 0.020	0.116 ± 0.026	0.018 ± 0.013	0.045 ± 0.013	0.18 ± 0.05	0.72 ± 0.20	16.1 ± 4.5	62.4 ± 16.0	−50.5 ± 16.6
	7.38 ± 1.38	3.29 ± 1.31	7.41 ± 3.49	32.28 ± 4.67	20.01 ± 2.93	12.13 ± 2.67	1.63 ± 0.21
INS	0.065 ± 0.007	0.062 ± 0.018	0.117 ± 0.042	0.027 ± 0.033	0.058 ± 0.029	0.18 ± 0.08	0.75 ± 0.21	16.3 ± 7.3	65.2 ± 17.7	−30.1 ± 20.7
	8.03 ± 2.21	4.71 ± 2.51	12.08 ± 8.04	54.25 ± 17.35	31.36 ± 11.12	22.17 ± 9.28	2.05 ± 0.38
PUT	0.037 ± 0.011	0.078 ± 0.021	0.138 ± 0.080	0.050 ± 0.080	0.067 ± 0.036	0.26 ± 0.08	0.92 ± 0.27	24.5 ± 15.6	79.3 ± 19.2	−17.4 ± 22.2
	18.09 ± 7.30	5.05 ± 2.15	13.87 ± 8.23	50.76 ± 16.04	26.53 ± 11.41	20.53 ± 7.95	1.73 ± 0.47
NCD	0.033 ± 0.007	0.062 ± 0.021	0.108 ± 0.022	0.022 ± 0.016	0.056 ± 0.027	0.21 ± 0.06	0.80 ± 0.28	18.3 ± 5.0	69.9 ± 22.1	−19.6 ± 28.4
	19.51 ± 7.01	5.28 ± 2.40	14.14 ± 8.68	59.11 ± 24.19	32.30 ± 9.51	23.98 ± 12.33	2.37 ± 0.81
THAL	0.040 ± 0.012	0.061 ± 0.018	0.115 ± 0.034	0.038 ± 0.029	0.060 ± 0.020	0.29 ± 0.10	0.85 ± 0.22	26.9 ± 14.2	74.1 ± 19.4	−14.6 ± 14.1
	18.53 ± 6.87	6.09 ± 1.86	17.17 ± 6.71	53.00 ± 5.81	26.11 ± 9.42	20.85 ± 3.03	2.15 ± 0.50
CBL	0.040 ± 0.010	0.063 ± 0.020	0.190 ± 0.027	0.012 ± 0.004	0.029 ± 0.006	0.14 ± 0.03	0.47 ± 0.13	12.1 ± 3.5	40.8 ± 10.1	−44.1 ± 27.4
	12.08 ± 6.29	4.36 ± 1.44	6.80 ± 2.27	28.01 ± 7.82	25.49 ± 7.49	11.63 ± 2.60	3.58 ± 0.96
PONS	0.038 ± 0.007	0.046 ± 0.015	0.149 ± 0.023	0.022 ± 0.005	0.035 ± 0.008	0.19 ± 0.06	0.50 ± 0.15	16.8 ± 5.4	43.9 ± 12.7	−25.8 ± 27.2
	15.75 ± 3.63	7.08 ± 2.31	13.32 ± 4.26	39.05 ± 16.47	28.12 ± 16.34	13.87 ± 5.75	4.36 ± 2.45

FRO-D, dorsal frontal cortex; FRO-O, orbitofrontal cortex; TEMP-L, lateral temporal cortex; PAR, parietal cortex; OCC, occipital cortex; TEMP-M, mesiotemporal region; PREC/POSTC, pre-/postcentral gyrus; CING, cingulate gyrus; INS, insula; PUT, putamen; NCD, caudate nucleus; THAL, thalamus; CBL, cerebellum; PONS, pons; V_B, fractional blood volume [%}; %CV, coefficient of variation of the parameter estimate [%]; AIC, Akaike Information Criterion [unitless]. For definition of K₁ mL · mL⁻¹ · min⁻¹], k₂ [min⁻¹], k₃ [min⁻¹], and k₄ [min⁻¹], BP′ [unitless], DV′_t [unitless], BP [unitless], DV_t [unitless] see text. All values are given as mean ± standard deviation.

The parameter estimations of k₄ and particularly k₃ suffer from low to very low precision (average %CV ranging 16 to 32% and 24 to 59%, respectively), being worst defined in subcortical structures. Neither does k₃ reflect the distribution of cerebral A₁ARs, nor is the magnitude of k₄ homogeneous across different regions, as might be expected based on their postulated physiological meaning (see Eq. 2 and 3). Similarly, the macro parameter BP′ is associated with a high %CV, being on average 9 to 15% in cortical structures (exceeding 20% twice in the insula and once in the mesiotemporal region) and 20 to 24% in subcortical nuclei (exceeding 20% in 7 cases, but remaining always below 40%). The regional BP (and BP′) estimates show no clear relationship to the distribution of cerebral A₁ARs. In fact, the BP (being equivalent to DV_s) accounts for only 25 − 38% (mean = 29%) of the DV_t.

In contrast, the %CV of an individual DV_t′ never exceeds 5% except in one case (pons, 9%), mostly ranging from 1–3%. Furthermore, the mean regional estimates of DV_t′ (and consequently DV_t) correspond to the known distribution of cerebral A₁ARs.

One-tissue compartment model

The results of the 1TCM are given in Table 2, representative curve fits are displayed in Fig. 4. The regional distribution of K₁ and V_B values is in good agreement with the known distribution of regional cerebral perfusion and regional cerebral blood volume (Leenders et al., 1990). In absolute terms, V_B (ranging 6–8%) was found to be higher than expected (5%). There is a strong correlation between the 1TCM and the 2TCM in regard of the regional K₁ (linear regression: slope = 0.86, intercept = −0.003, r² = 0.96) and V_B values (linear regression: slope = 0.94, intercept = 0.023, r² = 0.83). However, the regional K₁ and V_B values yielded from the 1TCM are on average 19.3 ± 5.2% lower and 51.5 ± 19.4% higher, respectively, than the corresponding values given by the 2TCM. The individual estimates of V_B given by the 1TCM are associated with a comparable %CV as in the 2TCM (on average ranging 8–13% in cortical regions and 14–16% in subcortical regions, never exceeding 20% in any region). The average %CV of the K₁ and k₂′ estimates range from 3 to 6% and from 3 to 7% (always below 10%, exceeding 5% only in cerebellum and pons), respectively. Likewise, the estimates of the DV_t' are associated with high estimate precision (%CV always below 5%, exceeding 3% only in cerebellum and pons).

TABLE 2.

Results of the one-tissue compartment model (1TCM) analyses

Region	V _B %CV	K ₁ %CV	k′₂ %CV	V′_t %CV	DV _t	AIC
FRO-D	0.062 ± 0.011	0.050 ± 0.015	0.071 ± 0.007	0.70 ± 0.20	61.4 ± 16.5	−5.9 ± 14.0
	11.02 ± 2.74	3.14 ± 0.43	3.31 ± 0.46	1.82 ± 0.19
FRO-O	0.064 ± 0.013	0.049 ± 0.015	0.067 ± 0.007	0.74 ± 0.20	64.3 ± 17.2	3.6 ± 9.0
	12.33 ± 2.66	3.51 ± 0.49	3.71 ± 0.53	2.05 ± 0.30
TEMP-L	0.066 ± 0.012	0.050 ± 0.015	0.064 ± 0.006	0.78 ± 0.22	68.1 ± 16.8	−4.0 ± 13.0
	10.77 ± 2.40	2.95 ± 0.37	3.12 ± 0.40	1.73 ± 0.24
PAR	0.065 ± 0.012	0.051 ± 0.016	0.070 ± 0.008	0.73 ± 0.21	63.0 ± 15.3	−6.1 ± 16.1
	10.69 ± 2.99	3.06 ± 0.47	3.23 ± 0.49	1.77 ± 0.21
OCC	0.071 ± 0.013	0.055 ± 0.016	0.072 ± 0.008	0.75 ± 0.20	66.1 ± 17.5	5.9 ± 14.9
	11.64 ± 3.33	3.49 ± 0.52	3.69 ± 0.56	2.00 ± 0.23
TEMP-M	0.065 ± 0.013	0.038 ± 0.012	0.055 ± 0.006	0.69 ± 0.18	59.9 ± 13.6	−3.1 ± 17.5
	11.55 ± 4.01	3.53 ± 0.63	3.75 ± 0.67	2.15 ± 0.32
PREC	0.058 ± 0.010	0.048 ± 0.016	0.074 ± 0.007	0.64 ± 0.19	55.8 ± 15.0	−0.6 ± 14.4
	12.61 ± 3.04	3.70 ± 0.47	3.90 ± 0.50	2.17 ± 0.21
POSTC	0.061 ± 0.009	0.050 ± 0.017	0.076 ± 0.009	0.65 ± 0.20	56.5 ± 14.9	−3.2 ± 15.8
	11.49 ± 2.93	3.54 ± 0.63	3.73 ± 0.67	2.06 ± 0.31
CING	0.073 ± 0.009	0.050 ± 0.018	0.076 ± 0.009	0.66 ± 0.19	57.3 ± 14.5	−8.9 ± 16.1
	8.83 ± 1.87	3.22 ± 0.55	3.40 ± 0.59	1.88 ± 0.30
INS	0.080 ± 0.005	0.053 ± 0.016	0.075 ± 0.008	0.70 ± 0.19	61.1 ± 16.0	−8.9 ± 15.7
	7.90 ± 1.44	3.03 ± 0.50	3.20 ± 0.54	1.76 ± 0.32
PUT	0.057 ± 0.011	0.064 ± 0.020	0.074 ± 0.009	0.87 ± 0.24	74.9 ± 17.8	6.6 ± 17.1
	14.42 ± 3.70	3.07 ± 0.54	3.24 ± 0.58	1.71 ± 0.23
NCD	0.051 ± 0.011	0.052 ± 0.018	0.070 ± 0.010	0.75 ± 0.25	65.0 ± 19.6	3.6 ± 18.4
	15.82 ± 4.13	3.41 ± 0.54	3.61 ± 0.57	1.99 ± 0.24
THAL	0.059 ± 0.014	0.049 ± 0.015	0.061 ± 0.006	0.79 ± 0.20	69.4 ± 18.1	6.1 ± 14.9
	14.43 ± 4.16	3.41 ± 0.58	3.62 ± 0.63	2.00 ± 0.30
CBL	0.064 ± 0.010	0.050 ± 0.019	0.127 ± 0.016	0.39 ± 0.12	33.7 ± 8.2	−3.1 ± 19.4
	10.60 ± 3.32	5.42 ± 0.83	5.59 ± 0.78	3.17 ± 0.58
PONS	0.061 ± 0.010	0.033 ± 0.011	0.079 ± 0.012	0.42 ± 0.13	36.7 ± 10.1	8.1 ± 12.8
	13.50 ± 2.15	6.40 ± 0.74	6.78 ± 0.77	3.95 ± 0.57

For abbreviations see Table 1.

Graphical analysis

The results of the GA are summarized in Table 3, representative GA plots are shown in Fig. 4. The time lag t∗ after which the Logan plot turned linear ranged between 25 and 35 minutes, except in the cerebellum where t∗ was 40 minutes in one case and in the caudate nucleus where t∗ always fell between 15 and 25 minutes.

TABLE 3.

Results of the graphical analyses (GA)

Region	t∗	DV′_t	DV _t
FRO-D	29.2 ± 2.0	0.76 ± 0.21	66.2 ± 18.7
FRO-O	30.8 ± 2.0	0.80 ± 0.22	70.2 ± 20.6
TEMP-L	33.3 ± 2.6	0.84 ± 0.24	73.7 ± 19.2
PAR	30.0 ± 3.2	0.78 ± 0.23	68.2 ± 17.5
OCC	33.3 ± 4.1	0.83 ± 0.22	73.0 ± 20.7
TEMP-M	28.3 ± 4.1	0.74 ± 0.19	65.1 ± 16.3
PREC	30.0 ± 0.0	0.69 ± 0.21	60.2 ± 16.8
POSTC	30.8 ± 2.0	0.70 ± 0.22	61.3 ± 16.9
CING	30.8 ± 2.0	0.72 ± 0.20	63.1 ± 16.4
INS	28.3 ± 2.6	0.77 ± 0.20	67.3 ± 18.1
PUT	25.0 ± 3.2	0.90 ± 0.26	77.7 ± 18.4
NCD	21.7 ± 4.1	0.77 ± 0.27	67.5 ± 21.3
THAL	27.5 ± 4.2	0.83 ± 0.22	73.1 ± 19.8
CBL	34.2 ± 3.8	0.45 ± 0.13	39.2 ± 10.5
PONS	26.7 ± 2.6	0.47 ± 0.14	41.4 ± 12.4

For abbreviations see Table 1; t^∗, start of linear segment of the GA plot [min].

Comparison between different kinetic analyses

As shown by the AIC (Table 1 and Table 2) and in Fig. 4, the 2TCM fits the cerebral kinetics of [¹⁸F]CPFPX considerably better than the 1TCM. Of 90 VOI tested (6 subjects × 15 VOI per subject), this difference is significant in all regions (F test, all p < 0.001, except insula VOI in one volunteer: p < 0.003).

The results of the linear regression analyses between the individual DV_t′ and DV_t values, respectively, given by the different analyses are summarized in Table 4. As shown in Fig. 5, the 1TCM analysis considerably underestimates DV_t′ (and consequently DV_t) compared to the 2TCM. Depending on the absolute value of DV_t′ and DV_t, the mean regional underestimation ranges between 5.4% (putamen; mean absolute DV_t underestimation = 4.4) and 17.5% (cerebellum; mean absolute DV_t underestimation = 7.2). In contrast, the linear regression fits between the DV_t′ and DV_t values, respectively, given by the GA and the 2TCM show only a minimal deviation from the line of identity (see Table 4: barely significantly y-axis intercept in the regression DV_t GA vs. DV_t 2TCM, while the regression slope is not different from 1). Consequently, the maximum mean regional deviations between the DV_t values given by the GA and the 2TCM are rather small, ranging between −5.9% [-2.5 DV_t units] in the pons and 3.3% [2.1 DV_t units] in the insula.

FIG. 5.

Correlation Between DV_t Values Provided by 2TCM, 1TCM and GA. The scatter plot at the top shows the correlation between for the DV_t values provided by the 1TCM and the 2TCM analysis. The corresponding correlation between GA and 2TCM is shown below. The dotted lines represent unity. For regression parameters see Table 4.

TABLE 4.

Results of the linear regression analyses between different kinetic analyses

Regression	Parameter	y-Axis intercept	Slope	r²
1TCM vs. 2TCM	DV′_t	−0.014 [−0.029; 0.001]	0.929 [0.910; 0.949]	0.991
	DV _t	−1.204 [2.662; 0.254]	0.929 [0.907; 0.950]	0.988
GA vs. 2TCM	DV′_t	0.006 [−0.007; 0.019]	0.974 (0.959; 0.992]	0.993
	DV _t	−1.392 [−2.780; −0.004]	1.007 [0.986; 1.027]	0.991

1TCM, one-tissue compartment model; 2TCM, two-tissue compartment model; GA, Logan's graphical analysis. For definition of DV′_t [unitless] and DV_t [unitless] see text. All regressions p < 0.0001, numbers in squared brackets denote 95%-confidence limits.

The mass of injected CPFPX had no effect on the regional binding parameters DVt ′ and DVt, neither in any region nor in any analysis method. The average plasma caffeine level of the included normal volunteers after requested 24h caffeine abstinence was 0.26 ± 0.14 mg/L (range: 0.13–0.44 mg/L) (for comparison: resorption of 100 mg caffeine, corresponding to one cup of coffee, would lead to a peak plasma caffeine level of approximately 2.5 mg/L). No correlation was found between plasma caffeine levels and binding parameters.

DISCUSSION

The present study is intended to evaluate the applicability of standard compartment models (1TCM and 2TCM) and of Logan's GA (Logan et al., 1990) to quantitatively describe the cerebral kinetics of [¹⁸F]CPFPX. [¹⁸F]CPFPX is the first PET ligand that has successfully been applied to image A₁ARs in the human brain in vivo (Bauer et al., 2003a). It shows several characteristics that make it a promising ligand for the quantitative assessment of cerebral A₁ARs: Its cerebral uptake is reversible, peaks rapidly and is in accordance with the in vitro distribution of A₁ARs (Fastbom et al. 1987a; Svenningsson et al., 1997; Bauer et al., 2003a). [¹⁸F]CPFPX metabolites are considerably more polar and highly unlikely to cross the blood-brain barrier. Brain homogenate extracts in mice demonstrated >98% unchanged ligand at 60 minutes post-injection (Holschbach et al., 2002). The maximum brain uptake of [¹⁸F]CPFPX, approximately 2.3%ID, permits dynamic PET studies with adequate image characteristics. The long physical half-life of fluorine-18 (110 minutes) allows a reliable determination of the metabolite-corrected plasma input function at later time points.

The 2TCM fitted the experimental data significantly better than the 1TCM in all regions of all subjects examined. However, due to the high correlation between individual rate constants the 2TCM suffered from high standard errors of the individual parameter estimates, particularly with regard to k₃ and k₄, and the derived macro parameter BP′. The relative long scan duration of 90 minutes as compared to 60 minutes and the application of residual weighting had only a moderate beneficial impact on estimate precision. A common method to improve the identifiability of individual parameters is to constrain parameters (e.g., k₄ (=k_off) or the ratio K₁/ k₂, equal to f₁DV_f+ns) in all brain regions to a reasonable constant value, assuming that they are homogenous throughout the brain. Since at this stage there is no reliable estimate of k₄, the 2TCM cannot be constrained by using a fixed k₄.

In vitro animal studies in commonly used laboratory animals (Fastbom et al. 1987b) and PET studies in primates using the A₁AR ligands [¹⁸F]CPFPX (Boy et al., 1998) and [¹¹C]KF15372 (Wakabayashi et al., 2000) showed a high cerebellar A₁AR density. In contrast, human post mortem autoradiographic studies (Fastbom et al. 1987a; Svenningsson et al., 1997; Bauer et al., 2003a) and PET studies (Bauer et al., 2003a) displayed an overall considerably lower density of A₁ARs in the cerebellum. Nevertheless, own preliminary bolus/infusion displacement studies in humans showed a specific [¹⁸F]CPFPX binding even in regions with lowest binding, which is in agreement with A₁AR PET studies in monkeys and cats using [¹¹C]KF15372 (Wakabayashi et al., 2000) and [¹¹C]MDPX (Shimada et al., 2002), respectively. From these studies it can estimated that 25–30% of the cerebellar DV_t constitutes specific binding. Under similar circumstances, Watabe et al. (2000) were able to successfully quantify the binding of [¹¹C]MDL100,907 to 5-HT_2A receptors by using a constrained 2TCM (called 2T′ in their study), in which the K₁/ k₂ ratio was constrained to the estimate of the DV_f+ns of the cerebellum. The latter estimate has been generated by fitting a 2TCM to the data of the cerebellum, where a low level of displaceable binding was found. In case of [¹⁸F]CPFPX, DV_f+ns′ (=K₁/ k₂) accounted for approximately 70% of the cerebellar DV_t′ (Table 1), which could be in accordance with our preliminary displacement studies. However, this finding does most likely not reflect a reliable estimate of the cerebellar DV_f+ns′, since a relationship of approximately 30:70 of specific:non-specific binding was given by the 2TCM in merely all regions, regardless of the known regional A₁AR density. This clearly indicates that the BP′ estimates and the derived BP value do not reflect the amount of specific binding. Alternative approaches to achieve a quantification of the BP include a dual-injection protocol with subsequent injections of high and low specific radioactivities and a bolus/infusion protocol with displacement. The pharmacological and radiochemical characteristics of [¹⁸F]CPFPX appear to be particularly well suited to the latter approach.

The 2TCM analyses resulted in highly precise estimates of the DV_t′ (and consequently DV_t). The same was true for the 1TCM, which, however, gave significantly negatively biased estimates due to its poor fit. In contrast, the GA provided DV_t values, that were least biased in comparison to the 2TCM. The largest mean regional deviations occurred in pons (−5.9%) and insula (+3.3%). This might be explained by two reasons. Firstly, all regions were corrected for an assumed homogenous V_B component of 5%, which differs from the invasive estimates of 3.8% for pons and 6.5% for insula. Thus, the data of pons and insula might be slightly over- and under-corrected, respectively. Neglecting V_B in GA leads to an overestimation of the DV_t compared to the 2TCM of approximately 5–12% (depending on regional V_B and DV_t). Region-specific V_B values could be assumed in order to minimize this bias. Secondly, it is well recognized that the Logan's GA method itself is affected by a negative bias in regions with high noise levels (Logan et al., 2001; Slifstein et al., 2000), which additionally might explain the negative bias in small and/or low-binding regions (e.g., pons: −5.9%, cerebellum: −4.3%, caudate nucleus: −3.5%). Several attempt have been made to minimize this bias (Logan et al., 2001; Ichise et al., 2002). Despite these limitations, GA (or a less biased method (Logan et al., 2001; Ichise et al., 2002) is the favored approach because it is computational simple and also applicable on a pixel-level, which allows it to generate parametric images that are amenable to pixel-by-pixel statistical mapping.

At this stage, DV_t appears to be the most valid measure to assess the A₁AR density in vivo with [¹⁸F]CPFPX PET. Because DV_t summarizes both specific and non-specific binding, it reflects relative changes of the A₁AR density in an attenuated manner, whereby the degree of attenuation is proportional to the regional DV_s/ DV_t ratio. Preliminary [¹⁸F]CPFPX displacement studies in humans demonstrate that specific binding accounts for at least 60% of the total cortical binding. A 10% change of receptor density would then translate into a 6% change of DV_t.

Theoretically, DV_t is superior to DV_t′ as an outcome measure, because of its independence from between-subject or between-condition f₁ variations. However, the determination of f₁ in the magnitude of 1–2% (as in case of [¹⁸F]CPFPX) is technically challenging and threatened by a significant measurement inaccuracy. Therefore, at this stage it is difficult to infer, if a f₁ correction is advantageously, e.g. in a diagnostic scenario: We found that there is a weak relationship between DV_t′ and f₁ in the present study group (r ≈ 0.45–0.59) which leads to a slight decrease of the variation of the regional DV_t compared to DV_t′ (2TCM, mean regional coefficient of variation, DV_t′: 29.0%, DV_t: 26.8%). This does not necessarily imply, that the slightly lower variation of the DV_t enables a better discrimination between two subject groups with different A₁AR densities. In fact, the higher variation of the DV_t′ might reflect the true variation caused by unknown factors more closely. Thus the exploration of confounding variables as for example age and caffeine consumption might help to decide, whether a f₁ correction is favorably, since the A₁AR density can be expected to vary with both age (Cheng et al., 2000; Cunha et al., 2001) and chronic caffeine consumption (Shi and Daly, 1999).

Moreover, the use of DV_t′ and DV for inter-individual comparisons relies on the assumption, that DV_f+ns is reasonably constant across subjects. Although this can be hardly verified in every subject studied, it can be shown by representative pre-blocking or displacement studies.

In conclusion, [¹⁸F]CPFPX shows promising in vivo characteristics as a PET ligand for quantification of cerebral A₁ARs. A 2TCM satisfactorily accounts for the cerebral kinetics of [¹⁸F]CPFPX, although individual tissue compartments could not be separated using a standard 2TCM. Logan's GA is an attractive alternative method of analysis.

Footnotes

Acknowledgments

The authors gratefully acknowledge the excellent technical assistance of Lutz Tellmann, Elisabeth Theelen, Suzanne Schaden, Hans Herzog, Markus Lang, Barbara Elghahwagi, and Jon Shah (Research Center Juelich) and the valuable advice of Cyrill Burger (PMOD Group, Department of Nuclear Medicine, Universitaetsspital Zurich, Switzerland) on the use of the PMOD software package.

References

Abi-Dargham

Laruelle

Seibyl

Rattner

Baldwin

Zoghbi

Zea-Ponce

Bremner

Hyde

Charney

Hoffer

Innis

(1994) SPECT measurement of benzodiazepine receptors in human brain with iodine-123-iomazenil: Kinetic and equilibrium paradigms. J Nucl Med 35:228–238

Akaike

(1974) A new look at the statistical model identification. IEEE Trans Atomat Contr AC19:716–723

Bauer

Holschbach

Meyer

Boy

Herzog

Olsson

Coenen

Zilles

(2003a) In vivo imaging of adenosine A1 receptors in the human brain with [18F]CPFPX and positron emission tomography. Neuroimage 19:1760–1769

Bauer

Holschbach

Cremer

Weber

Boy

Shah

Olsson

Halling

Coenen

Zilles

(2003b) Evaluation of 18F-CPFPX, a Novel Adenosine A(1) Receptor Ligand: In Vitro Autoradiography and High-Resolution Small Animal PET. J Nucl Med 44:1682–1689

Boy

Holschbach

Mühlensiepen

Herzog

Shah

Ollson

Coenen

Müller-Gärtner

H-W

(1998) A1-adenosine receptor (A1AR) mapping in primate brain using the novel A1AR-antagonist [18F]CPFPX [abstract]. Neuroimage 7:A39

Burger

Buck

(1997) Requirements and implementation of a flexible kinetic modeling tool. J Nucl Med 38:1818–1823

Cheng

Liu

Juang

Jou

(2000) Decrease of adenosine A-1 receptor gene expression in cerebral cortex of aged rats. Neurosci Lett 283:227–229

Cunha

Constantino

Fonseca

Ribeiro

(2001) Age-dependent decrease in adenosine A1 receptor binding sites in the rat brain. Effect of cis unsaturated free fatty acids. Eur J Biochem 268:2939–2947

Deckert

Abel

Kunig

Hartmann

Senitz

Maier

Ransmayr

Riederer

(1998) Loss of human hippocampal adenosine A1 receptors in dementia: Evidence for lack of specificity. Neurosci Lett 244:1–4

10.

Defrise

Kinahan

Townsend

Michel

Sibomana

Newport

(1997) Exact and approximate rebinning algorithms for 3-D PET data. IEEE Trans Med Imaging 16:145–158

11.

Dunwiddie

Masino

(2001) The role and regulation of adenosine in the central nervous system. Annu Rev Neurosci 24:31–55

12.

Fastbom

Pazos

Probst

Palacios

(1987a) Adenosine A1 receptors in the human brain: A quantitative autoradiographic study. Neuroscience 22:827–839

13.

Fastbom

Pazos

Palacios

(1987b) The distribution of adenosine A1 receptors and 5′-nucleotidase in the brain of some commonly used experimental animals. Neuroscience 22:813–826

14.

Fredholm

IJzerman

Jacobson

Klotz

Linden

(2001) International Union of Pharmacology. XXV. Nomenclature and classification of adenosine receptors. Pharmacol Rev 53:527–552

15.

Glass

Faull

Bullock

Jansen

Mee

Walker

Synek

Dragunow

(1996) Loss of A1 adenosine receptors in human temporal lobe epilepsy. Brain Res 710:56–68

16.

Haas

Selbach

(2000) Functions of neuronal adenosine receptors. Naunyn Schmiedebergs Arch Pharmacol 362:375–381

17.

Holschbach

Fein

Krummeich

Lewis

Wutz

Schwabe

Unterlugauer

Olsson

(1998) A1 adenosine receptor antagonists as ligands for positron emission tomography (PET) and single-photon emission tomography (SPET). J Med Chem 41:555–563

18.

Holschbach

Olsson

Bier

Wutz

Sihver

Schuller

Palm

Coenen

(2002) Synthesis and evaluation of no-carrier-added 8-cyclopentyl-3-(3-[(18)F]fluoropropyl)-1-propylxanthine ([(18)F]CPFPX): A potent and selective A(1)-adenosine receptor antagonist for in vivo imaging. J Med Chem 45:5150–5156

19.

Ichise

Toyama

Innis

Carson

(2002) Strategies to improve neuroreceptor parameter estimation by linear regression analysis. J Cereb Blood Flow Metab 22:1271–1281

20.

Koeppe

Holthoff

Frey

Kilbourn

Kuhl

(1991) Compartmental analysis of [11C]flumazenil kinetics for the estimation of ligand transport rate and receptor distribution using positron emission tomography. J Cereb Blood Flow Metab 11:735–744

21.

Lammertsma

(2003) Receptor kinetics - modelling and practical approach. In: PET Pharmacokinetic course manual ( Maguire

Leenders

eds), University of Groningen, Groningen, The Netherlands and McGill University, Canada, pp71–89

22.

Landaw

DiStefano

3rd (1984) Multiexponential, multicompartmental, and noncompartmental modeling. II. Data analysis and statistical considerations. Am J Physiol 246:R665–677

23.

Laruelle

al-Tikriti

Zea-Ponce

Zoghbi

Baldwin

Charney

Hoffer

Kung

Innis

(1994) In vivo quantification of dopamine D2 receptor parameters in nonhuman primates with [123I]iodobenzofuran and single photon emission computerized tomography. Eur J Pharmacol 263:39–51

24.

Leenders

Perani

Lammertsma

Heather

Buckingham

Healy

Gibbs

Wise

Hatazawa

Herold

Beaney

Brooks

Spinks

Rhodes

Frackowiak

RSJ

Jones

(1990) Cerebral blood flow, blood volume and oxygen utilization. Normal values and effect of age. Brain 113:27–47

25.

Logan

Fowler

Volkow

Wolf

Dewey

Schlyer

MacGregor

Hitzemann

Bendriem

Gatley

Christman

(1990) Graphical analysis of reversible radioligand binding from time-activity measurements applied to [N-11C-methyl]-(−)-cocaine PET studies in human subjects. J Cereb Blood Flow Metab 10:740–747

26.

Logan

Fowler

Volkow

Ding

Wang

Alexoff

(2001) A strategy for removing the bias in the graphical analysis method. J Cereb Blood Flow Metab 21:307–320

27.

Mikolajczyk

Szabatin

Rudnicki

Grodzki

Burger

(1998) A JAVA environment for medical image data analysis: Initial application for brain PET quantitation. Med Inform (Lond) 23:207–214

28.

Mintun

Raichle

Kilbourn

Wooten

Welch

(1984) A quantitative model for the in vivo assessment of drug binding sites with positron emission tomography. Ann Neurol 15:217–227

29.

Pietrzyk

Herholz

Fink

Jacobs

Mielke

Slansky

Wurker

Heiss

(1994) An interactive technique for three-dimensional image registration: Validation for PET, SPECT, MRI and CT brain studies. J Nucl Med 35:2011–2018

30.

Shi

Daly

(1999) Chronic effects of xanthines on levels of central receptors in mice. Cell Mol Neurobiol 19:719–732

31.

Shimada

Ishiwata

Kiyosawa

Nariai

Oda

Toyama

Suzuki

Ono

Senda

(2002) Mapping adenosine A(1) receptors in the cat brain by positron emission tomography with [(11)C]MPDX. Nucl Med Biol 29:29–37

32.

Slifstein

Laruelle

(2000) Effects of statistical noise on graphic analysis of PET neuroreceptor studies. J Nucl Med 41:2083–2088

33.

Svenningsson

Hall

Sedvall

Fredholm

(1997) Distribution of adenosine receptors in the postmortem human brain: An extended autoradiographic study. Synapse 27:322–335

34.

Ulas

Brunner

Nguyen

Cotman

(1993) Reduced density of adenosine A1 receptors and preserved coupling of adenosine A1 receptors to G proteins in Alzheimer hippocampus: A quantitative autoradiographic study. Neuroscience 52:843–854

35.

Wakabayashi

Nariai

Ishiwata

Nagaoka

Hirakawa

Oda

Sakiyama

Shumiya

Toyama

Suzuki

Senda

(2000) A PET study of adenosine A1 receptor in anesthetized monkey brain. Nucl Med Biol 27:401–406

36.

Watabe

Channing

Der

Adams

Jagoda

Herscovitch

Eckelman

Carson

(2000) Kinetic analysis of the 5-HT2A ligand [11C]MDL 100,907. J Cereb Blood Flow Metab 20:899–909

Quantification of Cerebral A 1 Adenosine Receptors in Humans using [ 18 F]CPFPX and PET

Abstract

Keywords

MATERIALS AND METHODS

Subjects

PET Acquisition

Metabolite Analysis

Image Analyses

Kinetic analyses

Compartmental analyses

Two-tissue compartment model

One-tissue compartment model

Graphical analysis

Comparison of the different analyses

RESULTS

Metabolite analyses

Cerebral uptake and distribution of [18F]CPFPX

Kinetic analyses

Two-tissue compartment model

One-tissue compartment model

Graphical analysis

Comparison between different kinetic analyses

DISCUSSION

Footnotes

Acknowledgments

References

Cerebral uptake and distribution of [¹⁸F]CPFPX