Test-retest variability of adenosine A2A binding in the human brain with 11C-TMSX and PET

Background The goal of the present study was to evaluate the reproducibility of cerebral adenosine A2A receptor (A2AR) quantification using 11C-TMSX and PET in a test-retest study. Methods Five healthy volunteers were studied twice. The test-retest variability was assessed for distribution volume (VT) and binding potential relative to non-displaceable uptake (BPND) based on either metabolite-corrected arterial blood sampling or a reference region. The cerebral cortex and centrum semiovale were used as candidate reference regions. Results Test-retest variability of VT was good in all regions (6% to 13%). In the putamen, BPND using the centrum semiovale displayed a lower test-retest variability (3%) than that of BPND using the cerebral cortex as a reference region (5%). The noninvasive method showed a higher or similar level of test-retest reproducibility compared to the invasive method. Conclusions Binding reproducibility is sufficient to use 11C-TMSX as a tool to measure the change in A2AR in the human brain. Electronic supplementary material The online version of this article (doi:10.1186/s13550-014-0076-9) contains supplementary material, which is available to authorized users.


Background
The regional cerebral binding of adenosine A 2A receptor (A 2A R) antagonists, [7-methyl-11 C]-(E)-8-(3,4,5trimethoxystyryl)-1,3,7-trimethylxanthine ( 11 C-TMSX) [1] and 11 C-KW-6002 [2,3], were quantitatively investigated in vivo in healthy human. 11 C-TMSX has been evaluated in human brain studies in not only healthy human controls [4][5][6][7] but also in drug-naïve Parkinson's disease patients before and after therapy [8]. An aging effect on A 2A R was also evaluated using 11 C-TMSX PET [9]. Several outcome measures, such as distribution volume (V T ), distribution volume ratio (DVR), and binding potential (BP ND ), can be used to detect changes of A 2A R binding due to disease progression or therapeutic treatment. The reproducibility of these measures is important to conduct studies to detect change in A 2A R. Given that A 2A R distribution is heterogeneous, with only a very small amount of extrastriatal-specific binding, either the frontal cortex in the rat [3], the cerebellum in the monkey [10,11], or centrum semiovale [7] or cerebral cortex [8,9] in human, was used as a reference region to estimate non-displaceable binding of 11 C-TMSX. Since only a few postmortem human brain studies and blocking studies are available, it is not clear which region is a suitable reference region.
The aim of this paper was to assess the test-retest reproducibility of PET outcome measures, V T and BP ND , with the centrum semiovale and cerebral cortex as candidate reference regions.

Human subjects
All studies were performed under a protocol approved by the Ethics Committee of the Tokyo Metropolitan Institute of Gerontology. Five healthy, male subjects participated in this study (mean age ± SD, 22.4 ± 2.6 years old, range: 21 to 27 years old). Subjects were all right-handed and screened for history of neurological, psychiatric, and physical diseases. All subjects did not have a history of alcoholism and not on any medications to affect brain function. Caffeine intake was not allowed for at least 12 h prior to PET scanning. Written informed consent was obtained from all subjects after receiving an explanation of the study. Magnetic resonance (MR) images were acquired on all subjects to eliminate those with any brain abnormalities and to place regions of interest (ROIs) on PET images. The MR imaging was conducted with three-dimensional spoiled gradientrecalled echo (SPGR) imaging on a SIGNA 1.5 Tesla machine (General Electric, Waukesha, WI, USA) [6].

Radiochemistry
Radiosynthesis of 11 C-TMSX followed the literature procedure [12]. All procedures were conducted under dim light to prevent photoisomerization of 11 C-TMSX. The radiochemical purity of 11 C-TMSX was >99%.

PET acquisition
Each subject underwent two 11 C-TMSX brain PET scans on two different days, and time of scanning was identical for test and retest scans of each individual subject, in order to remove the influence of circadian rhythm. The inter-scan interval was 28 to 35 days. Dynamic PET images were acquired in the Positron Medical Center, Tokyo Metropolitan Institute of Gerontology with the SET-2400 W PET scanner (Shimadzu, Kyoto, Japan), which acquires 63 slices (3.125-mm slice separation) with a spatial resolution of 4.4 mm full width at half maximum (FWHM) and a z-axis resolution of 6.5 mm FWHM [13]. Prior to the scan, a 5-min 68 Ga/ 68 Ge transmission scan was conducted for attenuation correction. 11 C-TMSX was injected intravenously over 60 s. Emission data were collected in two-dimensional mode for 1 h in 27 frames of increasing duration (6 × 10 s; 3 × 30 s; 5 × 1 min; 5 × 2.5 min; 8 × 5 min). Head movement was minimized with an air cushion. The dynamic images were reconstructed by the filtered back-projection method using a Butterworth filter (second-order low-pass filter, cutoff frequency was 1.25 cycles/cm) with corrections for scatter and randoms.

Input function measurement
In advance of each scan, an arterial catheter was inserted into the radial artery for blood sampling. After radiotracer injection, arterial blood samples were manually collected every 10 s for the first 2 min and thereafter at longer intervals, 2. 25, 2.5, 3, 5, 7, 10, 15, 20, 30, 40, 50, and 60 min post-injection. A total of 24 samples were obtained per scan. Whole blood and plasma were counted in a cross-calibrated well-type gamma-counter (BSS-1, Shimadzu, Kyoto, Japan). An additional venous blood sample was taken before 11 C-TMSX administration, which was used for the in vitro assessment of the fraction of 11 C-TMSX in plasma bound to plasma proteins (f P ). Arterial blood sampling was not available in one subject. Thus, a total of five subjects were included in reference region analyses and four subjects were also analyzed using plasma data.

Plasma metabolite and protein binding analysis
The fraction of intact radioligand to total plasma activity was determined from blood samples collected at 3, 10, 20, 30, 40, and 60 min after injection by high-performance liquid chromatography (HPLC). The blood was centrifuged at × 7,000 g for 1 min at 4°C to obtain the plasma, which was denatured with an equivalent volume of acetonitrile in an ice-water bath. The suspension was centrifuged under the same conditions and divided into soluble and precipitable fractions. The precipitate was resuspended in 2 vol. of 50% aqueous acetonitrile followed by centrifugation. The recovery yield of the radioactivity in the two soluble fractions was 98.7%. Two soluble fractions were combined, and into this solution, an equivalent volume of a solution of 50-mM aqueous acetic acid and 50-mM aqueous sodium acetate (pH 4.5; 50/50, v/v) was added. After centrifugation of the samples as described above, the supernatant was loaded onto a Nova-Pak C8 column equipped in an RCM 8 × 10 module (8 mm diameter × 100 mm length; Millipore-Waters, Milford, MA, USA). The mobile phase was a mixture of acetonitrile, 50-mM aqueous acetic acid and 50-mM aqueous sodium acetate (pH 4.5; 4/3/3, v/v/v) at a flow rate of 2 mL/min. The elution profile was detected with a radioactivity monitor (FLO-ONE 150TR; Packard Instrument, Meriden, CT, USA). The retention time of 11 C-TMSX was 6.2 min. The recovery in the eluate of the injected radioactivity was essentially quantitative. The six measured parent fractions were fitted by a sum of exponential functions. The metabolite-corrected plasma curve was generated as the product of the total plasma activity and the fitted parent fraction curve.
Individual f P values were determined by ultrafiltration. Prior to administration of 11 C-TMSX, approximately 6 mL of blood was taken from each subject. A reference blood sample was created by adding 22.9 ± 15.7 MBq (at the time of administration, range: 10.1 to 49.5 MBq of 11 C-TMSX in approximately 60 μL to this blood sample and incubated for 10 min at 37°C). Following centrifugation (2,000 g at room temperature for 3 min), triplicates of 400 μL aliquots of plasma sample were pipetted into ultrafiltration tubes (Microcon-30, 30 kDa, Merck Millipore, Billerica, MA, USA), and centrifuged at room temperature (14 min at 14,000 g). The free fraction f P was calculated as the ratio of activity in the ultrafiltrate to the total plasma. The amount of nonspecific binding of 11 C-TMSX to the filter was also determined by applying the same procedure to a sample created by addition of 11 C-TMSX to saline.

Image analysis
Regions of interest were defined by manually drawing circles using the registered MR images as additional reference. The details are written in [6,14]. Time-activity curves (TACs) were generated for eight ROIs: anterior putamen, posterior putamen, putamen, caudate head, thalamus, cerebellum, centrum semiovale, and cerebral cortex. The putamen ROI consists of the anterior and posterior putamen subregions. The cerebral cortex ROI included the frontal, temporal, and occipital cortices.
In the present study, the cerebral cortex and centrum semiovale were chosen as candidate reference regions. For 11 C-TMSX kinetic analysis, the cerebellum was not used as a reference region, because A 2A R binding in our previous human study [7] was higher in the cerebellum than in neocortical regions. In a previous human autoradiographic study [15], the density of A 2A Rs in the frontal cortex was found to be low, as that in the temporal and occipital cortices.

Outcome measures
The DVR has been used in our previous study on an aging effect of A 2A R in human brain [8,9]. In this study, the two additional outcome measures, V T and BP ND , were estimated. The definition of the outcome measures is described in [16]. Regional TACs were analyzed using the Logan graphical analysis (LGA) with input function and reference tissue (two-parameter version) [17,18] to estimate the outcome parameters of V T and BP ND . Starting time (t*) was set to 10 min post-injection [7].

Statistical analyses
The test-retest reproducibility was statistically evaluated according to the following three criteria: signed testretest variability (TRV), absolute test-retest variability (aTRV), and intra-class correlation coefficient (ICC). TRV was calculated as the difference between the test and retest measurements, divided by the mean of the test and retest values (2 × (p test − p retest )/(p test + p retest )). aTRV was calculated as the absolute value of TRV (2 × | p test − p retest |/(p test + p retest )). TRV indicates whether there is a systematic trend between the test and retest scans. The test-retest reliability of the two parameter measurements was the ICC calculated using the following equation [19]: where BSMSS and WSMSS are the mean sum of squares between subjects and within subjects, respectively. In the test-retest study, an ICC value ranges from −1 (no reliability) to 1 (maximum reliability) [20,21]. Sample sizes were calculated to detect a 20-percent difference in BP ND between independent groups (two-tails t-test) using the software G*power 3.1 [22]. The confidence level was set to be 5% (P < 0.05) and statistical power to 0.8. The mean of the test scans was used as the mean of baseline scans, and the SDs of the baseline and blocking scans were assumed to be same as the SDs of the test scans. All statistical parameters except for power analysis were calculated with MATLAB Version 7.
Arterial input function Figure 1A shows the averaged radioactivity in plasma with metabolite correction for test and retest scans (n = 4). The metabolism speed of 11 C-TMSX was slow in both scans ( Figure 1B): the unchanged fraction was still 85% ± 5% in test scan and 82% ± 6% in retest scan at 60 min postinjection. The free fraction of 11 C-TMSX in plasma was 2.40% ± 0.96% for test scans and 2.40% ± 0.47% for retest scans. There are no significant differences in f P between test and retest scans (paired t-test, P = 0.72). The ultrafiltrate-to-saline ratio was 46% ± 3% in test scans and 46% ± 3% in retest scan, indicating a high retention on the filter.

Quantitative analysis
Brain activity in all regions reached the peak around 5 min post-injection of 11 C-TMSX, and then gradually decreased. The average tissue-to-plasma ratio was shown in Figure 2. Typical parametric images of BP ND were displayed in Figure 3 using a centrum semiovale as a reference region. The ratios in most regions became constant around 20 min post-injection of 11 C-TMSX. The ratios in the putamen were decreased slightly throughout the scan. The values for V T and BP ND were summarized in Tables 1, 2, and 3. For each outcome parameter, the mean of the test and retest scans, the TRV (mean ± standard deviation), the aTRV, and the ICC were listed.
For V T values, the mean TRV was smaller than 5%, and always smaller than its standard deviation, indicating that there is no systematic trend between test and retest scans. The absolute TRV was ≤10% except for the thalamus. The ICC values were moderate (>0.65) except for the thalamus (0.27). We also calculated the normalized V T (V T /f P ). Global mean aTRV values were 8% and 15% for V T and V T /f P , respectively, indicating that normalizing by the plasma free fraction f p increased the variability of the outcome measure for 11 C-TMSX. For BP ND values, the mean TRV was between ±10% using either cerebral cortex or centrum semiovale as a reference region, and always smaller than its standard deviation, further indicating that there was no systematic trend between test and retest scans.
Since BP ND of the cerebellum from one subject was close to 0 in the test and retest scans with reference LGA with the cerebral cortex as a reference region, TRVs of the subject were different from the other subjects. Those values were removed from Table 2.
The mean BP ND values were larger, and the TRV and aTRV were smaller when using the centrum semiovale as reference instead of the cerebral cortex. Global mean aTRV values were 8% and 15% for BP ND using the centrum semiovale and cerebral cortex, respectively, as a reference region. Both TRVs using LGA with input function were comparable to those values using reference LGA. The BP ND estimates using LGA with input function were in excellent agreement with those from reference LGA (BP ND, reference LGA = 1.00 BP ND, LGA + 0.01, R 2 = 1.00 with the cerebral cortex as a reference region, BP ND, reference LGA = 1.05 BP ND, LGA + 0.01, R 2 = 0.98 with the centrum semiovale).
Power analysis (two-tails t-test, statistical power 0.8) was conducted to estimate the samples sizes to detect a 20percent difference in BP ND . Using the cerebral cortex as a reference region, sample sizes ranged from 4 (thalamus) to 21 (caudate head) per group. Using the centrum semiovale as a reference region reduced the required sample sizes: 3 (putamen and thalamus) to 9 (caudate head).

Discussion
The plasma free fraction (f P ) was measured in this study, allowing for correction of V T values. This correction by f P is useful if f P can be measured reliably and if there is  substantial intra-subject variation. In our measurements, the f P was consistently low (<3%), with evidence that 11 C-TMSX stuck to the ultrafiltration tubes, which may lead to underestimation of f p . However, the f P value measured in [23] was 9.1% ± 0.4% (n = 6, human) by the ultrafiltration method. Such a discrepancy might be attributable to high stick factor in our data. Note that the stick factor was not reported in [23]. Another possibility is the difference in the preparation of the injection solution. Finally, the signed and absolute TRVs were larger for V T /f P compared to those of V T . Hence, the normalization of V T by f P did not reduce variability in this case.
Inter-subject variability (% coefficient of variation (COV)) of V T at retest scans were lower (approximately 7%) than that of the test scans (approximately 16%), while no significant difference was observed in the injected dose, specific activity, and f P of 11 C-TMSX. Another possibility to explain the difference in the intersubject variability is a difference in the metabolism of the tracer. The subjects were controlled for caffeine intake, but not for smoking habituations. Nicotine consumption might change the metabolism of 11 C-TMSX as seen in the study with adenosine A 1 receptor ligand 18 F-CPFPX [24]. In a retrospective investigation, it turned out that subjects consisted of a nonsmoker, a smoker (blood sampling was not available), and three subjects with unknown status. However, the parent fraction of 11 C-TMSX was very high and well reproducible ( Figure 1B). Therefore, we concluded that a change in the metabolism speed was not a reason to increase the intersubject variability. In contrast to V T , such a difference in the inter-subject variability did not exist in BP ND . The difference in the %COV between test and retest scans might come from errors included in the input function measurement.
The test-retest variability and reliability of V T were good (aTRV ≤10%, ICC >0.6) across regions except for the thalamus (aTRV: 13% and ICC: 0.27). For BP ND , a good absolute TRV was seen in the high A 2A regions (putamen and caudate). However, lower-binding regions (BP ND < 0.4) showed high aTRV (>15%) and low ICC values; this is not surprising, since BP ND is small in those regions. We examined the test-retest variability data of V T and BP ND from a number of radioligands. The aTRV of V T of 11 C-TMSX (8% averaged across all regions) was comparable to that of other radioligands used to study dopamine and adenosine receptors. The reported aTRV values of V T were 5% to 11% (average: 7%) with 11 C-FLB457 [25] for dopamine D 2/3 receptor and 12% to 14% (average: 13%) with 18 F-CPFPX [26] for   adenosine A 1 receptor. The aTRV of BP ND with 11 C-TMSX was comparable to that with 11 C-FLB457 (6% to 15%) and larger than that with 18 F-FPFPX (3% to 9%). Given the good reproducibility of V T , 11 C-TMSX should be suitable for use in receptor occupancy studies with input function. The range of V T values was not wide across regions (0.70 to 1.46 mL/cm 3 ). However, using the occupancy plot [27] is feasible using the regions with a narrow range of V T values with 11 C-GSK931145 for glycine type 1 transporter (0.43 to 0.79 mL/cm 3 ) [28] and 18 F-CPFPX for adenosine A 1 receptor (0.42 to 0.82 mL/cm 3 ) [29]. Note that the occupancy plot assumes that the receptor occupancies are uniform in all regions of interest. Previous reports [1,30] suggest that some regions might have an 'atypical' binding. Therefore, we need to carefully choose regions used for the occupancy plot with 11 C-TMSX. Another possible way for estimating receptor occupancy is to estimate a relationship between blocking dose (or plasma level) and V T for each region [31]. This second method can be used even if all regions have the same baseline V T .
The test-retest variability of BP ND values using the cerebral cortex as a reference region was larger than those using the centrum semiovale. In the striatum, the high A 2A R-binding region, the aTRVs of BP ND were 5% in the putamen and 19% in the caudate head using the cerebral cortex as a reference region. On the other hand, the aTRVs of BP ND were 3% in the putamen and 13% in the caudate head using the centrum semiovale as a reference region. This is partly because the BP ND value was smaller using the cerebral cortex as a reference region.
The thalamus showed a low reproducibility of both V T and BP ND values. Moreover, while the mean distribution volume in the thalamus was high, a postmortem study with 3 H-SCH58261 [15] showed that A 2A R density is low. The uptake in the thalamus is considered to be 'atypical' binding [1,30], which is different from classical A 2A R binding. This low reproducibility in the thalamus may be partly due to such an 'atypical' binding. Thus, given the low reproducibility and 'atypical' binding, the thalamus should be carefully considered in further clinical research.
Using either the cerebral cortex or centrum semiovale as a reference region, reference LGA and LGA with input function provided similar BP ND values. The TRV and aTRV of BP ND were slightly smaller using the reference LGA. Not surprisingly, the reference tissue model is not affected by errors in the measurement of input function. This suggests that the reference LGA can be useful for further studies.
There are two limitations of this study: unknown optimal reference region for 11 C-TMSX and small sample Table 2 Test-retest variability and reproducibility of binding potential (BP ND ) using cerebral cortex as a reference region

Regions
Logan graphical analysis with input function (t* = 10 min, n = 4) Reference Logan graphical analysis (t* = 10 min, n = 5)  size. As far as we know, the only available A 2A R blocking study using an antagonist radiotracer in vivo in human brain is a 11 C-KW-6002 PET study with varying dose of cold KW-6002 [3] However, blocking results in the centrum semiovale and neocortical regions were not included in the report. Thus, the suitability of the cerebral cortex or central semiovale as a reference region has yet to be determined by blocking or occupancy studies. Due to a lack of blocking study and postmortem study in the regions with low A 2A R density, the region with lowest V T was chosen. For the SPECT A 2A R tracer 123 I-MNI-420, while a reference region is not yet validated, a testretest reproducibility of BP ND was evaluated to facilitate the comparison between 123 I-MNI-420 and other A 2A R radiotracers [32]. We also took an exploratory approach to calculate BP ND values using candidate reference regions in order to evaluate BP ND reproducibility. However, the determination of the reference region is most desirable in order to establish the utility of 11 C-TMSX for PET imaging. In this study, we evaluated outcome measures with input function in four subjects. We examined sample sizes for test-retest human studies using other radioligands. As far as we know, the minimum sample size is three subjects for test-retest protocol ( 18 F-MK-6577 [33] for glycine transporter type 1 and 123 I-MNI-420 [32] for A 2A R).

Conclusions
The quantification of 11 C-TMSX imaging was reproducible for PET studies of A 2A R. The LGA with input function achieved good reproducibility for V T in all regions. The results support the use of PET and 11 C-TMSX as a suitable tool for receptor occupancy studies. The use of the cerebral cortex or centrum semiovale as a reference region with invasive and reference LGA produced good or moderate reproducibility of the BP ND in high A 2A R regions. While the centrum semiovale showed higher reproducibility of the BP ND , blocking studies are required to determine the optimal reference region conclusively. BP ND : binding potential relative to non-displaceable uptake; DVR: distribution volume ratio; f P : plasma free fraction; HPLC: high-performance liquid chromatography; ICC: intra-class correlation coefficient; V T : distribution volume; TACs: time-activity curves; TRV: signed test-retest variability.