Modeling [11C]yohimbine PET human brain kinetics with test-retest reliability, competition sensitivity studies and search for a suitable reference region

Previous work introduced the [11C]yohimbine as a suitable ligand of central α2-adrenoreceptors (α2-ARs) for PET imaging. However, reproducibility of [11C]yohimbine PET measurements in healthy humans estimated with a simplified modeling method with reference region, as well as sensitivity of [11C]yohimbine to noradrenergic competition were not evaluated. The objectives of the present study were therefore to fill this gap.


METHODS
Thirteen healthy humans underwent two [11C]yohimbine 90-minute dynamic scans performed on a PET-MRI scanner. Seven had arterial blood sampling with metabolite assessment and plasmatic yohimbine free fraction evaluation at the first scan to have arterial input function and test appropriate kinetic modeling. The second scan was a simple retest for 6 subjects to evaluate the test-retest reproducibility. For the remaining 7 subjects the second scan was a challenge study with the administration of a single oral dose of 150 µg of clonidine 90 minutes before the PET scan. Parametric images of α2-ARs distribution volume ratios (DVR) were generated with two non-invasive models: Logan graphical analysis with Reference (LREF) and Simplified Reference Tissue Method (SRTM). Three reference regions (cerebellum white matter (CERWM), frontal white matter (FLWM), and corpus callosum (CC)) were tested.


RESULTS
We showed high test-retest reproducibility of DVR estimation with LREF and SRTM regardless of reference region (CC, CERWM, FLWM). The best fit was obtained with SRTMCC (r2=0.94). Test-retest showed that the SRTMCC is highly reproducible (mean ICC>0.7), with a slight bias (-1.8%), whereas SRTMCERWM had lower bias (-0.1%), and excellent ICC (mean>0.8). Using SRTMCC, regional changes have been observed after clonidine administration with a significant increase reported in the amygdala and striatum as well as in several posterior cortical areas as revealed with the voxel-based analysis.


CONCLUSION
The results add experimental support for the suitability of [11C]yohimbine PET in the quantitative assessment of α2-ARs occupancy in vivo in the human brain. Trial registration EudraCT 2018-000380-82.


Introduction
Investigations of the noradrenergic system function in the brain have mainly emerged from animal studies so far. Indeed, the lack of suitable imaging tools has hampered its understanding in human. Nevertheless, we now have the possibility to overcome this difficulty as in vivo imag-towards the collection of missing data in the living human brain by direct quantification of regional 2-ARs availability. Indeed, 2-ARs play a key role in regulating noradrenergic neurotransmission ( Szabadi, 2013 ) as altered noradrenergic transmission with specific loss of 2-ARs is currently theorized to play a critical role in both symptoms and progression of some neurodegenerative and mood disorders ( Marien et al., 2004 ;Ordway et al., 2003 ). Human in vivo imaging of 2-ARs is therefore essential. Previous work showed that [ 11 C]yohimbine PET tissue data can be described by a 1 tissue compartment (1-TCM), and 2-ARs binding potential can be estimated using the corpus callosum (CC) as reference tissue . Simplified Reference tissue methods ( SRTM ) have been widely used to estimate neuroreceptor binding potential because they eliminate the invasive arterial input function (AIF) step. However, reproducibility of non-displaceable binding potential (BP ND ) in healthy humans, as well as displaceability with an 2-ARs agonist, were not evaluated. Yet, demonstration of a reproducible [ 11 C]yohimbine PET outcome measure is critical and preliminary to accurately determine the clinical significance of pharmacologic or pathophysiologic 2-ARs changes.
The objectives of the present study were therefore (1) to identify an optimal [ 11 C]yohimbine PET full kinetic modeling with AIF, (2) assess the validity of non-invasive kinetic modeling, (3) identify the best reference region, (4) estimate the reproducibility of [ 11 C]yohimbine PET measurements with test-retest scans and (5) examine the sensitivity of [ 11 C]yohimbine with a noradrenergic pharmacological competition using acute administration of clonidine, a drug known to decrease NA release through pre-synaptic 2-ARs activation.

Materials and methods
All procedures performed in this study were in accordance with the ethical standards of the institutional and national research committee and with the principles of the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study was approved by the Ethics Committee of Sud Mediterranée III (EudraCT: 2018-000380-82) and pre-registered before being conducted on the Clinical-Trials.gov database under the trial record number NCT03520543. The subjects gave written informed consent to participate in the study.

Participants
Over sixteen healthy men included, two were discontinued, and fourteen completed the study. The mean age was 25 years old (standard deviation, SD, 3 years) and the mean weight was 74. 4 ± 7.2 kg (range, 63-87). All subjects were free of medical or neuropsychiatric illness and none of them were smokers or under medication. Each subject underwent two [ 11 C]yohimbine PET scans separated by an interval of 7 to 21 days. For the first scan (test), an arterial catheter was inserted into the radial artery after completion of the Allen test and infiltration of the skin with 1% lidocaine. For the second scan, 7 subjects had a repeated baseline PET (retest), and 7 other subjects had a second PET 74 ± 11 min after administration of a 150 μg oral dose of clonidine (challenge). For all subjects, this second scan was performed without blood sampling. Subjects were also genotyped for the cytochrome P450 (CYP) system with regard to the CYP2D6 isoform, as this latter is involved in the metabolism of yohimbine in the liver, yielding two metabolites that may have some action at 2-ARs ( Le Corre et al., 1999 ).

PET procedures
[ 11 C]yohimbine was synthesized as previously described ( Jakobsen et al., 2006 ). The radiochemical purities of syntheses used for the study were greater than 95%, with corresponding molar activities of 53.4 ± 16.4 GBq/μmol at the end of synthesis.
All subjects received an intravenous bolus injection of 370 MBq ± 10% of [ 11 C]yohimbine ( Table 1 ). List-mode PET data were acquired, during 90 min from the injection of the tracer, simultaneously with 3T MRI data (Dixon T1, anatomic MPRAGE T1, veinous and arterial TOF), on a Siemens mMR Biograph system.

Input function measurement
AIF ( Ca ) and metabolite levels in the plasma were measured from 25 heparinized arterial blood samples manually collected with the following timing: every 5 s for the first minute, every 10 s until second minute, and at times 5, 10, 30, 45, 60, and 90 min post-injection. Eight whole blood aliquots (100 μL), at times of 1, 2, 5, 10, 30, 45, 60, and 90 min post injection, were counted in gamma counter (Perkin-Elmer) to evaluate whole blood to plasma ratio ( f wb ). Twenty-five blood samples were centrifuged for 3 min (4000 g) and plasma aliquots (100 μL) were counted in gamma counter to measure uncorrected plasma curve ( Cp ). For metabolite analysis, other plasma aliquots were filtrated using a 0.45 μm membrane filter. Two hundred micromilliliter plasma filtrates were injected in HPLC system with a C8 CAPCELL PAK MF column (Osaka Soda). Fractions were collected and counted for radioactivity in gamma counter. The activity of [ 11 C]yohimbine was divided by the total activity recovered from the gamma counter to give the plasma parent fraction of unmetabolized [ 11 C]yohimbine (PPf). Plasma-free fraction of [ 11 C]yohimbine freely diffusible to tissue ( fp ) was measured by ultrafiltration at 1, 2, 5, 10, 30, 45, 60, and 90 min post-injection as previously described ( Moore et al., 2003 ). Arterial blood samples were centrifuged, and 1 mL of plasma was placed in ultracentrifugation devices (Centrifree®, Millipore) and spun for 10 min at 2000 g. One hundred micromilliliter aliquots of whole plasma and ultrafiltrate were counted in gamma counter. After counting, all samples were weighed, and counts were corrected. The fp was determined from the ratio of concentrations in the ultrafiltrate and whole plasma. AIF was the plasma curve corrected from the plasma parent fraction curve AIF(t) = PPf(t).Cp(t). The whole blood curve was determined by the mean f wb and the plasma curve Cwb(t) = f wb .Cp(t).

Image processing
Raw PET data were motion corrected ( Reilhac et al., 2018 ), then rebinned into 24-time frames (variable length frames, 8 × 15 s, 3 × 60 s, 5 × 120 s, 1 × 300 s, 7 × 600 s) sinograms for dynamic reconstruction. Images were reconstructed using 3D ordinary Poisson-ordered subsets expectation maximization (OP-OSEM 3D), incorporating the system point spread function using 3 iterations of 21 subsets. Sinograms were corrected for scatter, randoms, normalization and attenuation ( Mérida et al., 2017 ). Reconstructions were performed with a zoom of 2, yielding a voxel size of 2.03 × 2.03 × 2.08 mm 3 in a matrix of 172 × 172 voxels with a 4 mm 3D post-reconstruction gaussian filtering. The mean PET image of the session 2 was coregistered (rigid transform) onto the mean PET image of the session 1. Individual MRI T1 of the first session was normalized to the MNI space (Montreal Neurological Institute template of the International Consortium for Brain Mapping Project) with the Segment function of SPM 12 ( Ashburner and Friston, 2005 ). Labeling of the structural brain regions was performed using the multi-atlas propagation with enhanced registration (MAPER) methodology ( Heckemann et al., 2010 ), and the 83-region Hammerssmith atlas ( Hammers et al., 2003 ). After projection of the atlas onto the two sessions, regional time activity curves (TAC) were extracted for a selection of brain regions including the amygdala, cerebellum, corpus

Kinetic modeling
Modeling of the PET TAC was performed with the AIF corrected for metabolite and plasma free fraction using the Turku PET center utilities library ( TPCCLIB , https://gitlab.utu.fi/vesoik/tpcclib ). Invasive models including the 1-TCM ( fitk2 ), the 2-TCM ( fitk4), and the Logan graphical analysis (LGA) ( Logan, 2000 ) were used to estimate the volume of distribution (V T , mL. g − 1 ) across regions. Fitting accuracy was evaluated with the Akaike information criteria (AIC) ( Akaike, 1974 ). Reference tissue modeling techniques were also evaluated using SRTM ( Lammertsma and Hume, 1996 ) and the non-invasive Logan reference tissue model ( LREF) ( Logan, 2000 ). The CC was chosen as reference region based on the recommendations of previous work . In addition, the cerebellar white matter (CERWM) and the frontal lobe white matter (FLWM) were also tested as potential reference tissue because white matter has previously been proposed to represent a region of non-specific binding ( Landau et al., 2012 ). The distribution volume ratio (DVR) was the parameter of interest of the simplified modeling techniques: DVR SRTM , DVR LREF , and DVR 1-TCM (DVR computed as the ratio of the V T of the target region to the V T in the reference region, derived from the 1-TCM with AIF).
Reference tissue modeling was also performed at the voxel level to compute intra-cerebral DVR parametric maps with SRTM ( Gunn et al., 1997 ) and with LREF ( Varga and Szabo, 2002 ). DVR parametric images were then transformed from the subject's space to the MNI space using the nonlinear deformation fields derived from the spatial normalization of the individual's MR image. Finally, normalized parametric images were smoothed using an 8 × 8 × 8-mm 3 full width at half maximum isotropic gaussian kernel to account for the interindividual anatomy variability, normalization uncertainties, and to improve the sensitivity of the SPM analysis. Additionally, regional mean DVR values were computed from the parametric maps in the subject's space.

Bias and variability
The test-retest bias was calculated as the difference between the test and retest DVRs, divided by the mean of the test and retest values, and the variability as the SD of the bias. These parameters were expressed as percentage.

Reliability
The reliability of the measurements was assessed by the intraclass correlation coefficient (ICC) calculated as (BSMSS-WSMSS)/(BSMSS + WSMSS) where BSMSS is the mean sum of square between subjects and WSMSS is the mean sum of square within subjects ( Shrout and Fleiss, 1979 ). This statistic estimates the relative contributions of between-and within-subject variability and assumes values from − 1 (i.e., BSMSS = 0) to 1 (identity between test and retest, i.e., WSMSS = 0).

Test-challenge study
Clonidine is highly selective and exerts potent agonistic effects at pre-synaptic 2-ARs ( Delaville et al., 2011 ). Regional changes induced by the challenge with clonidine were computed as the relative difference of DVR between scans in selected ROIs, expressed as percentage. Binding changes were also observed at the voxel level.

Statistical analysis
Statistical analysis was performed using Rstudio ( https://github. com/rstudio/rstudio ). Paired Student's t-tests were used to test for differences in the injected dose and molar activity of the [ 11 C]yohimbine between Scan 1 and Scan 2 and also applied to test whether DVRs after retest or clonidine challenge significantly differed from baseline condition. Results with p < 0.05 were considered statistically significant. SPM12 was used for voxel-wise comparison with spatially normalized smoothed DVR images. Statistical parametric maps of the t statistic were computed with a threshold of p < 0.05 uncorrected at the voxel level and an extent voxels threshold the "Expected number of voxel per cluster ". Significant clusters were selected at a corrected cluster level of p < 0.05 ('' False discovery rate ") ( Poline et al., 1997 ).

Subject demographics
In total, over the fourteen subjects scanned, only thirteen completed the study according to the protocol. One subject's retest scan (subject 1) was not evaluated because of PET camera malfunction during the acquisition and was excluded from the test-retest analysis. Additionally, over the thirteen subjects, two subjects did not have arterial sampling in scan 1 but completed the study. For three subjects, the metabolite data were non-exploitable. At the end, for the modeling study, seven datasets were available (PET scan, AIF, metabolite function and free fraction), for the test-retest, six datasets were available (six subjects, two PET scans), and for the test-challenge study, seven datasets were available (seven subjects, two PET scans) ( Table 1 ). Mean age for both groups was identical (25 ± 4 years old for the test-retest group and 25 ± 2 for the test-challenge group). Fig. 1 shows metabolization curve (mean and SD plasma parent fractions across subjects) and its modeling depending on the CYP2D6 activity. The best modeling curve of the mean plasma parent fraction curve was a 3-exponentials model:
Free plasma fraction was constant over time and had a mean value of 8.5 ± 1.7%. Note that intermediate metabolizers presented mean lower fp values compared to extensive metabolizers (7.4 ± 0.5% and 9.1 ± 1.8%, respectively) not statistically significant ( W = 22, p = 0.16 ). Whole blood to plasma ratio was constant over time with a mean value of 66.1 ± 3.3% ( Table 1 ) without significant differences between both types of genotypes ( W = 21, p = 0.46 ). Fig. 2 shows an example of AIF corrected for metabolites and whole blood concentration curve for one subject with intermediate CYP2D6 activity and one subject with extensive profile.
Overall, the Akaike information (AIC) showed slightly lower values for the 2-TCM (225 ± 21) compared to the 1-TCM (244 ± 18) approach, indicating better fits for the 2-TCM. However, as previously highlighted , the 1-TCM approach is much more robust and produced less non-physiological estimates of the kinetic parameters. Accordingly, the 1-TCM was judged as the most appropriate model for analysis of [ 11 C]yohimbine imaging data. Kinetic parameters and V T using 1-TCM and LGA in the investigated brain regions are reported in Table 2 . The V T were higher in almost all the cortical regions ( > 0.4). Lower V T were observed in the cerebellum and the striatum ( < 0.35) as well as in the three potential reference tissues (CERWM, CC, FLWM).
Although, the regression between V T values computed by the 1-TCM model and V T values computed with LGA were extremely well correlated (V T_LGA = 0.99 * V T_1-TCM + 0.04; r 2 = 0.98; p < 0.001), this latter model slightly but significantly overestimated V T in all regions compared to 1-TCM ( p < 0.0002).

Model with tissue reference regions
Regression of DVR LREF and DVR SRTM to DVR 1-TCM for the three tested reference regions are shown in Table 3 . Regression coefficients were over 0.8 for all regressions, with the best fit obtained for SRTM CC (0.94). Regression slopes were close to 1 for both methods (ranging from 0.81 for SRTM CERWM to 1.13 for SRTM CC ). Intercepts ranged from 0.03 ( SRTM CC ) to 0.24 ( SRTM CERWM ).

Test-retest study
There were no significant differences between test and retest scans in neither the amount of radioactivity injected (MBq mean ± SD: Scan 1: 361.8 ± 12.6; Scan 2: 374.7 ± 22.1; Student_t = − 1.3; p = 0.26) nor the molar activity of [ 11 C]yohimbine (GBq/ mol mean ± SD: Scan 1: 33.5 ± 9.5; Scan 2: 30.8 ± 9.2; Student_t = 0.71; p = 0.51). The biases, variabilities and ICC values of the [ 11 C]yohimbine binding parameters are presented in Table 3 (right part) for both compartmental models with the three potential reference tissues. Reliability was very good ( > 0.74) for both models using either the CC or the CERWM as reference region, while it was acceptable when using the FLWM (around 0.5).
The mean values of bias were low ( < 2%) for all models, and the average variability ranged from 3.8% for the SRTM CERWM to 6.3% for the LREF CC model. Test and retest DVR values were highly correlated (R 2 > 0.98). Test-retest performance based on averaged values across subjects are presented in detail, region by region, in Table 4 for the most reproducible methods, namely the LREF and the SRTM with CC and CERWM as reference regions. Using the CC, the test-retest DVR reliability was moderate to high, ranging from 0.67 (gyrus cinguli) to 0.93 (cerebellum) with the LREF and from 0.52 (striatum) to 0.98 (cerebellum) with the SRTM . Using the CERWM, the test-retest DVR reliability was also moderate to high, ranging from 0.59 (frontal lobe) to 0.92 (amygdala) with the LREF and from 0.70 (cerebellum) to 0.92 (amygdala) with the SRTM . The test-retest DVR variability was excellent ranging from 2.6% (cerebellum) to 7.1% (insula) using the SRTM with CC, while this Fig. 2. Illustration of arterial input function (AIF), whole blood plasma (C wb ), uncorrected plasma ( Cp ) curves for one intermediate metabolizer subject (Part A) and one extensive metabolizer (part B). f wb : plasma to whole blood fraction.

Table 2
Kinetic parameters and V T estimated with 1-TCM and LGA in 7 healthy volunteers. Data are presented as mean values ( ± SD). Data are presented as mean values ( ± SD). Data are presented as mean values ( ± SD). Data in bold indicate significant differences between sessions ( p < 0.05).
variability was slightly higher, ranging from 4.7% (cerebellum) to 7.9% (frontal lobe) with the LREF . Using the CERWM as reference region, the test-retest DVR variability ranged from 2.8% (cerebellum) to 5.3% (insula) with the LREF and from 2.6% (cerebellum) to 5% (insula) with the SRTM . The mean bias across all ROIs was trivial ( < 4%). Of note, using SRTM method, the R1 parameter in any ROI was not significantly different between test and retest using the CC or the CERWM as reference tissue ( p = 1 and p = 0.97 respectively).

Challenge study
There were no significant differences between test and challenge scans in neither the amount of injected radioactivity (Scan 1: 362 ± 14 MBq; Scan 2: 370 ± 13 MBq; Student_t 6 = − 0.98; p = 0.367) nor the molecular activity of [ 11 C]yohimbine (Scan 1: 27.7 ± 10.9 GBq/ mol; Scan 2: 25.3 ± 7.4 GBq/ mol; Student_t 6 = 0.47; p = 0.652). The mean systolic and diastolic arterial blood pressures were 120 and 76 mmHg, respectively at the time of administration. Ninety minutes after administration, clonidine produced a transient 24% decrease only in the mean diastolic blood pressure ( 2 (6) = 17.81, p = 0.007). This effect of clonidine on the cardiovascular system is in good agreement with the literature ( Talke et al., 2001 ). At the anatomical regional level, results of the challenge study are given in Table 4 (right part). Significant increase of DVR in the challenge compared to the baseline condition was only observed in the amygdala ( + 6%) and striatum ( + 4%) with SRTM CC . Im- portantly, using SRTM method, the R1 parameter in any ROI was not significantly different between test and challenge using the CC or the CERWM as reference tissue ( p = 0.99 and p = 0.91 respectively.

Statistical parametric mapping
Statistical parametric maps present the results of the comparison between challenge and test conditions, for the challenge group, with respect to the same comparison in test-retest control group ( t contrast of condition effect: (DVR 2 -DVR 1 ) Challenge -(DVR 2 -DVR 1 ) control ). Fig. 3 shows the thresholded maps for the SRTM CC (3a), the LREF CC (3b), and the SRTM CERWM (3c), and the LREF CERWM (3d). Table 5 gives the cluster parameters that elicited differences according to the parametric imaging method ( SRTM or LREF) and the reference region (CC or CERWM). Using SRTM and the CC, two clusters showed significant activities including bilaterally the temporal and occipital fusiform gyrus, the cerebellum, the lateral part of the occipital cortex, the middle temporal gyrus, as well as the right occipital pole, inferior temporal gyrus and angular gyrus. Same contrast with SRTM and the CERWM reference region or with the LREF model did not show any significant cluster. Yet, these contrasts showed activities mainly located in the posterior part of the brain encompassing the temporal and occipital cortex as well as the cerebellum, as observed significantly using the SRTM CC imaging method.

Discussion
The regional distribution of [ 11 C]yohimbine binding corresponds with the known distribution of 2-ARs in post mortem human brain studies ( Ordway et al., 1993 ;Vos et al., 1992 ). The most prominent uptake of the tracer was seen in cortical brain regions, especially in the cingulum, frontal, parietal and occipital cortices whereas uptake was less prominent in the striatum and the cerebellum ( Fig. 4 ).

The kinetic modeling study
In order to evaluate a reliable and suitable method for 2-ARs quantification, we compared various invasive and non-invasive models, often used for brain receptor quantification.
Using invasive models and direct fitting of the PET kinetics with tissue compartmental model and AIF, the 1-TCM was found to be sufficient for describing the tracer kinetics of [ 11 C]yohimbine in the healthy human brain . Testing the LGA alternative to 1-TCM as a simpler resolution method from an algorithmic point of view led to results quite similar despite a very limited higher V T values found with LGA (around 12%). This might be attributed to the fact that blood volume, Vb, is not taken into account in LGA model. However, since Vb is fairly stable over the regions, this overestimation does not exclude the use of LGA as an alternative method to 1-TCM for a reliable estimate of the volume of distribution.
With non-invasive models, our findings show, for the first time, the feasibility of using a simple acquisition protocol for kinetic modeling avoiding arterial blood sampling. In particular, DVR estimated using non-invasive kinetic models ( LREF and SRTM ) showed an excellent correlation to the invasive 1-TCM whatever the reference region (CC, CERWM or FLWM) with the best fit obtained for SRTM CC (R 2 = 0.94). Of note, the results indicated a slight tendency toward an overestimation with CC (slope equal to 1.07 using LREF and 1.13 with SRTM). However, since the correlation with the 1-TCM is excellent and stable across brain regions, the induced bias can be predicted and should not have an impact on comparative studies. In parallel, tendency toward a slight underestimation was reported with CERWM and FLWM, which was more pronounced with SRTM (slope equal to 0.81 and 0.94 respectively) than LREF (slope equal to 0.94 and 0.98 respectively). Using reference tissue methods with cerebral regions, underestimation of DVR values can be explained by a possible specific binding within the reference region as well as spill-over effect ( Salinas et al., 2015 ). Nonetheless, CERWM or FLWM are valuable reference regions in the event of the impossibility of using the CC due to lesion of this zone in a patient, for instance.

Test-Retest reproducibility
In addition to the accuracy of the reference tissue methods, we evaluated their test-retest performance. Using FLWM as reference region, our results showed poor test-retest reliability of DVR measurements with both parametric imaging methods. However, the SRTM with the CERWM as reference region showed excellent test-retest reproducibility of DVR measurements, with variability ranging from 2.6% to 5.0% and a small negative bias ( − 0.1%). Using the SRTM with the CC, test-retest reproducibility of DVR measurements were a bit more spread, with a slightly higher variability ranging from 2.6% to 7.0% and a slightly higher negative bias (2%). Of importance, whatever the reference region, R 1 values, also computed with SRTM, did not differ between test and retest across all ROIs. In addition, the LREF method showed comparable very good test-retest reliability of DVR measurements (ICC = 0.75) for both reference region with a slightly lower variability when using the CERWM (~4%) compared to the CC (~6%). This demonstration of

Challenge study
So far, very limited evidence has been presented to support PET imaging of noradrenaline neurotransmission in humans. The brain uptake and receptor binding of [ 11 C]yohimbine was found to be reduced by unlabeled yohimbine challenge, amphetamine administration or by acute vagus nerve stimulation in rats and/or Landrace pig brain ( Jakobsen et al., 2006 ;Landau et al., 2015Landau et al., , 2012Phan et al., 2017Phan et al., , 2015. In our study, regional analyses showed significant regional change induced by the challenge with clonidine within the amygdala and the striatum, only when using the SRTM with the CC as reference region. In parallel, we conducted a voxel-based analysis as this latter has several advantages over the ROI-based approach, including the use of SPM to identify alterations in receptor binding in all brain areas without anatomical a priori . Interestingly, the voxel-based analysis showed changes that were not discernable with the ROIs analysis, where delineation of regions is determinist and based on anatomo-functional considerations. In particular, statistical parametric maps of [ 11 C]yohimbine binding under clonidine administration showed significant specific increases in the posterior part of the brain including the temporal and occipital lobes, as well as the cerebellum. However, this increase was again reported only when using SRTM with the CC as reference region. Overall, the pattern of increased [ 11 C]yohimbine uptake observed with this voxel-based analysis is consistent with previous studies demonstrating that 2-ARs agonists changed regional cerebral blood flow (CBF) in posterior cortical regions including the cerebellum, the temporal cortex and the angular gyrus ( Bonhomme et al., 2008 ;Fu et al., 2001 ). This increase in [ 11 C]yohimbine binding is likely the consequence of clonidine acting at presynaptic 2-ARs sites resulting in reduced noradrenergic neurotransmission ( Dennis et al., 1987 ;Starke, 1981 ) and leading to a state of low tonic arousal. Indeed, all of our volunteers reported that time spent in the scanner during the challenge condition had seemed shorter than during the first baseline scan. Considering the known effect of clonidine on vigilance ( Coull et al., 2004 ;Hall et al., 2001 ), we can reasonably assume that our participants had a low level of consciousness during the challenge scan. Interestingly, many studies have previously highlighted the role of a posterior network in the deleterious effect of clonidine on attention and arousal ( Bonhomme et al., 2008 ;Coull et al., 2004 ;Fu et al., 2001 ). In particular, the results of the present study reinforce the nonmotor role of the cerebellum ( Strick et al., 2009 ) through 2-ARs modulation ( Schambra et al., 2005 ) as well as the role of the fusiform gyrus and lateral occipital cortex in attention ( Tallon-Baudry et al., 2005 ).

Limitations of the study
There are some limitations in this study that has to be mentioned. First, the experimental design did not provide for arterial blood sampling during the second scan session. In fact, the first objective of the present study was to validate the use of a simplified reference tissue model, and the second objective was to assess both the reproducibility and the sensitivity of [ 11 C]yohimbine PET measurements estimated with this simplified modeling method. To this end, we limited the use of the invasive sampling of arterial blood only to the first scan in order not to lose volunteers for the second scan who might not want to repeat this unpleasant experience. Yet, it would have made sense to have arterial input in the clonidine challenge to evaluate the impact of the challenge on peripheral metabolism, if any, as well as to quantify the true volume of non-displaceable volume of distribution in the potential reference regions. This point would merit further investigations. Then, another shortcoming is the proportion of clonidine that bounds to plasma protein. Indeed, free plasma fraction can change upon pharmacological challenge due to displacement of ligand binding to plasma proteins. For instance, Phan et al. (2017) have shown an elevation of fp in response to challenge with unlabeled yohimbine. However, it has to be highlighted that the bound fraction of yohimbine to plasma proteins is around 80% (Berlan et al., 1993) while this bound fraction for clonidine is register to be of 20% ( Khan et al., 1999 ). Although we cannot completely rule out the possibility of a change of fp upon clonidine challenge, we believe this is very unlikely since the reported protein binding is relatively low. Indeed, within the same range (around 20%, ( de la Torre et al., 2004 ) no change of fp has been observed with amphetamine challenge . In the same vein, the potential effects of clonidine on CBF should also be considered. Since [ 11 C]yohimbine was administered as a systemic bolus, potential changes in CBF might affect the kinetics of the tracer in the brain and influence its binding. In humans, clonidine is known to decrease blood pressure and reduce CBF ( James et al., 1970 ;Lee et al., 1997 ). In our study, administration of clonidine produced a transient 26% decrease in mean diastolic arterial blood pressure. A clonidine-induced decrease in CBF could decrease the delivery of the tracer, which would induce a smaller Vt, therefore an underestimation of the clonidine-induced increase in 2-ARs binding. In other terms, the effect of clonidine in the present study, might have been underestimated. Finally, the results of the present study cannot completely rule out a possible effect of genotype (see supplementary materials). Further studies are needed to confirm this observation. In the meantime, the authors recommend to genotype volunteers for the cytochrome P450 system with regard to the CYP2D6 isoform if between groups comparisons have to be performed.

Conclusion
Our results support the use of [ 11 C]yohimbine PET in the in vivo assessment of human brain 2-ARs. Rapid tracer uptake associated with low test-retest variability and good reproducibility was demonstrated in the regions with the highest densities of the 2-ARs. In particular, the authors recommend the use of [ 11 C]yohimbine parametric imaging of DVR by non-invasive SRTM using the CC as reference tissue for imaging 2-ARs. This method was able to evidence moderate occupancy with concurrent drug on 2-ARs. Alternatively, CERWM might be considered when measurements in the CC would not be reliable for structural or lesional reason. In other terms, simplified imaging protocols can be used for reliable [ 11 C]yohimbine PET quantification which opens the possibility to investigate, in large human samples, the role of 2-ARs in various neuropsychiatric disorders including schizophrenia, depression, Parkinson's disease and Alzheimer's disease.

Declaration of Competing Interest
The authors have no conflicts of interest to declare that are relevant to the content of this article.

Data/code availability statement
Public access to the data online is not permitted without the approval of the study's sponsor namely the Hospices Civils de Lyon (Protocol ID 69HCL17_0196). The authors will be happy to support request for a formal data sharing agreement.