Quantification of aromatase binding in the female human brain using [11C]cetrozole positron emission tomography

Aromatase, the enzyme that in the brain converts testosterone and androstenedione to estradiol and estrone, respectively, is a putative key factor in psychoneuroendocrinology. In vivo assessment of aromatase was performed to evaluate tracer kinetic models and optimal scan duration, for quantitative analysis of the aromatase positron emission tomography (PET) ligand [11C]cetrozole. Anatomical magnetic resonance and 90‐min dynamic [11C]cetrozole PET‐CT scans were performed on healthy women. Volume of interest (VOI)‐based analyses with a plasma‐input function were performed using the single‐tissue and two‐tissue (2TCM) reversible compartment models and plasma‐input Logan analysis. Additionally, the simplified reference tissue model (SRTM), Logan reference tissue model (LRTM), and standardized uptake volume ratio model, with cerebellum as reference region, were evaluated. Parametric images were generated and regionally averaged voxel values were compared with VOI‐based analyses of the reference tissue models. The optimal reference model was used for evaluation of a decreased scan duration. Differences between the plasma‐input‐ and reference tissue‐based methods and comparisons between scan durations were assessed by linear regression. The [11C]cetrozole time–activity curves were best described by the 2TCM. SRTM nondisplaceable binding potential (BPND), with cerebellum as reference region, can be used to estimate [11C]cetrozole binding and generated robust and quantitatively accurate results for a reduced scan duration of 60 min. Receptor parametric mapping, a basis function implementation of SRTM, as well as LRTM, produced quantitatively accurate parametric images, showing BPND at the voxel level. As PET tracer, [11C]cetrozole can be employed for relatively short brain scans to measure aromatase binding using a reference tissue‐based approach.


| INTRODUC TI ON
Aromatase is the enzyme that converts androgens into estrogens (Thomas & Potter, 2013). Besides its role in reproduction, and its presence in the ovary, testis, and adipose fat tissue, aromatase is present in the brain and involved in neuronal survival, emotional behavior and cognition (Brocca & Garcia-Segura, 2019). Moreover, altered levels of aromatase have been found in different brain regions of patients with Alzheimer's disease and depression (Ishunina et al., 2005;Wu et al., 2017). However, the association between aromatase and brain functions is not fully understood.
Fine mapping and quantification of regional brain expression of aromatase in vivo are important to better understand physiological and pathological conditions. To assess aromatase in vivo, positron emission tomography (PET) combined with carbon-11 labeled reversible aromatase inhibitors has been employed, using tracers like [ 11 C]vorozole (Biegon et al., 2010 and, more recently, (Takahashi et al., 2018). Further quantification of aromatase binding can be performed by applying suitable mathematical models on the data both on the regional level, in specific volumes of interests (VOIs), and on a voxel level, to produce parametric images. In humans, the highest binding to aromatase has been found in the thalamus and hypothalamus, followed by the medulla, amygdala, nucleus accumbens, caudate nucleus, putamen, hippocampus, and cortex (Biegon, 2015;Biegon et al., 2010Biegon et al., , 2015Takahashi et al., 2018), thus in brain regions of relevance for reproductive as well as mental functions. On the other hand, the cerebellum has been selected as reference region because of low aromatase expression (Biegon, 2015;Biegon et al., 2010Biegon et al., , 2015Takahashi et al., 2018). This is in line with postmortem analyses demonstrating mRNA levels being highest in the hypothalamus and thalamus, intermediate in the amygdala and hippocampus followed by the frontal cortex, and lowest in the cerebellum (Sasano, Takashashi, Satoh, Nagura, & Harada, 1998).
The novel aromatase radioligand [ 11 C]cetrozole has been shown to bind specifically to aromatase, with higher signal-to-noise ratio than [ 11 C]vorozole in nonhuman primates and without accumulation of radiolabeled metabolites in the brain of rats (Takahashi et al., 2014). In humans, [ 11 C]cetrozole binding in men was found to be higher in the left hypothalamus than in women (Takahashi et al., 2018); additionally, [ 11 C]cetrozole binding in the left amygdala correlated positively with aggressiveness in the women. In previous studies with [ 11 C]cetrozole, both in humans and nonhuman primates, aromatase was quantified using plasma-input function as well as reference region approaches (Takahashi et al., 2014(Takahashi et al., , 2018, but no extensive validation of tracer kinetic models or methods for generating parametric images has been performed for this tracer. Thus, the aims of the present study were to (a) evaluate tracer kinetic models (Gunn, Gunn, & Cunningham, 2001) for quantitative analysis of [ 11 C] cetrozole in humans, based on plasma-and reference tissue input data, (b) to evaluate methods for producing parametric images of aromatase binding, and (c) to investigate the effect of scan duration on the above outcome parameters. The identification of optimal modeling to quantify [ 11 C]cetrozole uptake and scanning protocol are presented.

| Participants
As part of the "Brain Sex Hormones" (BSH) project, 13 healthy women were recruited by public announcements for participation in the present study. The study was conducted at the Department  (Sheehan et al., 1998). All participants received monetary compensation.
The study procedure was in accordance with ethical standards for human experimentation (Helsinki declaration) and was approved by the Regional Ethical Review Board of Uppsala (2014/393), and the Medical Radiation Ethics Committee of Uppsala University Hospital.

K E Y W O R D S
aromatase, brain, cetrozole, kinetic modeling, positron emission tomography, women Significance Estrogens synthesis is regulated by aromatase, an enzyme whose level can be assessed in the brain by use of a biochemical imaging technique. Optimal steps for aromatase analysis were identified by employing this neurotechnique on healthy women using the selective aromatase inhibitor cetrozole. As the association between aromatase and brain functions is not fully understood, quantitative studies in living humans are of importance to gain knowledge on the relation between aromatase expression and physiological and pathological processes.
Cyclotron-produced [ 11 C]carbon dioxide was converted to [ 11 C]methyl iodide by the "wet method" and transferred to the reagent solution in a stream of nitrogen gas. The resulting mixture was heated at 65°C for 2 min, diluted with an acetonitrile-water mixture (40/60, 2 ml), and [ 11 C]cetrozole was then purified and isolated by semi-preparative HPLC (ammoniumcarbonate 8.1 mM:acetonitrile; Chromosil Cholester 5 µm, 10 × 250 mm, Nacalai Tesque). The HPLC eluent was removed by vortex evaporation and the product was reformulated in ethanol (0.5 ml), Kleptose HPB (hydroxypropyl-β-cyclodextrin in sodium chloride 9 mg/ml, 300 mg/ml, 0.8 ml), and phosphate buffer pH 7.4 (0.1 M, 5 ml) and filtered through a 0.22-μm sterile filter (Millex-GV, Millipore) into a sterile glass vial. Radiochemical purity, identity, and concentration of the labeled product were assessed by analytical HPLC (ammoniumcarbonate 8.1 mM:acetonitrile; Gemini NX C18, Phenomenex, 4.6 × 100 mm) equipped with radio and UV detectors and isotopically unmodified cetrozole was used as reference. The palladium concentration in the product solution was quantified by ICP-AES.

| Data acquisition
Participants were instructed to abstain from caffeine and alcohol intake within the previous 12 hr, as well as food intake 3 hr prior to the session. Pregnancy tests were checked negative. The PET scans were acquired using either a Discovery ST (10 scans), Discovery IQ (two scans), or Discovery MI (18 scans) PET/CT scanner (GE Healthcare, Milwaukee, WI, USA), at baseline and after two challenges in random order (that is three scans for each participant, given the fact that no effect of challenge was observed on kinetic modeling, as reported in the results section). After a low-dose computed tomography (CT), performed for attenuation correction, 4.0 ± 0.7 MBq/kg [ 11 C]cetrozole was administered as a bolus, simultaneously with the start of a 90-min dynamic PET scan of 26 frames of increasing length (6 × 10, 3 × 20, 2 × 30, 3 × 60, 2 × 120, 4 × 300, 6 × 600 s). Reconstruction settings were chosen to yield similar spatial resolution across the three scanner models used: Ordered Subsets Expectation Maximization with two iterations and 21 subsets and a 4-mm postprocessing filter for the Discovery ST, four iterations and 12 subsets and a 4-mm postprocessing filter for the Discovery IQ, and three iterations and 34 subsets and a 5-mm postprocessing filter for the Discovery MI, with a resulting spatial resolution of about 6 mm. All appropriate corrections resulting in quantitative images were performed on the PET data. During eight scans, online blood sampling (Veenstra Instruments, Joure, The Netherlands) from a radial artery cannula was performed during the first 10 min of the scan (3 ml/min). Discrete blood samples (5 ml) were also taken at 5,10,15,20,30,40,50,60,70,80, and 90-min postinjection for measurements of whole blood and plasma radioactivity and for metabolite analysis. In addition, all subjects underwent a T1weighted magnetic resonance image (MRI) scan on a 3T Achieva scanner (Philips Healthcare, Best, The Netherlands) at the Department of Radiology for anatomical information and for VOI definition.

| Radiometabolite analysis
Blood was collected in heparin tubes and plasma was separated by centrifugation. The percentage of intact [ 11 C]cetrozole in plasma was determined by high-performance liquid chromatography (HPLC).
The column was eluted at a flow rate of 6 ml/min with acetonitrile −50 mM and ammonium carbonate 8.1 mM (55:45, v/v). The outlet from the detector was connected to a switching valve on the arm of the liquid handler to enable automatic fraction collection. Three fractions were collected, where the third contained the parent compound. The radioactivity in each fraction was measured by a welltype scintillation counter. The continuous arterial sampling data were calibrated relative to the discrete samples using the overlapping samples at 5 and 10 min postinjection. The arterial input function was obtained by multiplying the measured whole blood data by a single exponential fit to plasma/whole blood ratios, and a sigmoid fit to the measured fraction of intact [ 11 C]cetrozole in plasma.

| Image analysis
The dynamic [ 11 C]cetrozole PET data were corrected for interframe subject movement using VOIager software (GE Healthcare, Uppsala, Sweden). The MRI scans were co-registered to the sum of the first 5 min of the PET scan based on a 6-parameter rigid transformation using SPM8 (Wellcome Trust Center for Neuroimaging, University College London, UK). The MR images were segmented into gray matter, white matter, and CSF using SPM8. Six VOIs were included in the data analysis: thalamus, amygdala, hypothalamus, putamen, raphe nuclei, and cerebellum, averaged over the left and right sides. All VOIs, except amygdala, were defined using an automated probabilistic VOI template as implemented in the software PVElab (Svarer et al., 2005).
The amygdala VOIs were defined using a 70% isocontour VOI, drawn on precalculated parametric images of the [ 11 C]cetrozole binding (see Figure 5). The acquired VOIs were projected over all frames of the dynamic PET data to obtain time-activity curves (TACs).

| Tracer kinetic analysis
Single-tissue (1TCM) and two-tissue (2TCM) reversible plasmainput compartment models were applied to all TACs, using nonlinear regression with blood volume as an additional fit parameter.
Nondisplaceable binding potential (BP ND ) was calculated both directly (for 2TCM) as well as indirectly as the ratio of the volumes of distribution (V T ) in the target and reference regions minus one (DVR-1), with cerebellum gray matter as reference region. The optimal plasmainput model was determined using the Akaike information criterion (AIC; Akaike, 1974). In addition, plasma-input Logan analyses (Logan et al., 1990) were performed, estimating the binding as DVR-1.
Reference tissue models were also applied using cerebellum gray matter as reference region (Takahashi et al., 2014). BP ND was calculated using the simplified reference tissue model (SRTM) (Lammertsma & Hume, 1996) and DVR-1 was calculated using Logan reference tissue model (LRTM) (Logan et al., 1996). LRTM analysis was done both with and without the efflux constant, k 2 , in the model, with this constant based on 1TCM k 2 values in cerebellum (Logan et al., 1996). BP ND was also estimated using the standardized uptake value ratio (SUVR) minus one (SUVR-1), calculated as the ratio of the radioactivity concentration in each VOI relative to the reference region, on an interval of 80-90 min (SUVR 80-90 -1). The reference tissue methods were validated by comparison of BP ND or DVR-1 values with DVR-1 values based on the optimal plasma-input compartment model and plasma-input Logan analyses by calculating the orthogonal regression the square of the Pearson's correlation coefficient (R 2 ) and the bias.

| Parametric images
Voxel-level analysis was performed to generate parametric images showing the [ 11 C]cetrozole binding for each voxel. The basis function implementation of SRTM (RPM) (Gunn, Lammertsma, Hume, & Cunningham, 1997) and the two-parameter version of SRTM (RPM2) (Wu & Carson, 2002) were used, with a set of 50 basis functions with clearance rate values ranging from 0.01 to 2.00, to calculate BP ND for each voxel. The number of estimated parameters was reduced to two for RPM2 by assuming a constant efflux rate constant from the reference tissue, k 2 ', in the whole brain. This value was determined by taking the median k 2 ' value from the previous RPM analysis in all voxels with a BP ND value > 0. In addition, LRTM DVR-1 was also estimated for each voxel. Quantitative evaluation of the parametric methods was performed by projecting the set of VOIs used for analysis onto the parametric images to retrieve regionally averaged BP ND and DVR-1 values. The values from the parametric images were compared to the optimal VOI-based reference method using orthogonal regression and correlation analysis. Plasma-input and reference tissue modeling, as well as generating the parametric images, were performed using inhouse developed software in Matlab (The Mathworks, Natick, USA).

| Scan duration
In addition to the 90-min scan data, shortened data sets of the initial 40 min (21 frames) and 60 min (23 frames) were extracted from the dynamic scans to evaluate the minimum scan duration needed for quantification of [ 11 C]cetrozole. BP ND was calculated using SRTM for both the 40-and 60-min data sets. DVR-1 was calculated using LRTM, at time intervals of 20-40 and 30-60 min, and SUVR-1 were estimated at 30-40 and 50-60 min. Absolute bias was calculated between BP ND , DVR-1, or SUVR-1 of the shorter scan durations and the full 90-min scan.

| Simulations
To assess the robustness of the outcome measures, three sets of 100 noisy TACs were simulated using the two-tissue compartment model, using typical rate constant values for high, medium, and low binding, representing thalamus, amygdala, and raphe as well as cerebellum as reference tissue. Noise was added with levels similar to those seen in the clinical data, and the simulated TACs were fitted with the compartment models described above. Accuracy and precision of V T , DVR-1, BP ND were calculated as the bias and coefficient of variation (COV).

| Statistics
Analyses for symmetrical regions were based on averages of left and right. Analysis of tracer kinetic models and parametric images was performed by the use of orthogonal linear regression and Pearson correlation tests in GraphPad Prism (GraphPad Software, San Diego, Ca, USA). In addition, a multilinear regression was done to assess the effect of each challenge on the correlation between plasma-input and reference tissue models.

| Subjects
Three subjects were excluded from the analysis due to incomplete PET data and subsequent dropout of the study, thus the final data set comprised of 30 scans, three for each woman. The 10 participants included in the analyses were healthy, naturally cycling women (mean age 26 ± 3 years, 22-33 y.o.); none presented any psychiatric symptoms. All were Caucasians, had education level equal or higher than high school corresponding to a mean of 14.9 ± 1.7 years of education, 70% of them were students, while 30% worked full time, and one of them was mother of a child.

| Radiosynthesis
Sterile [ 11 C]cetrozole ready for injection was produced with a radioactivity yield of 1.0 ± 0.6 GBq (n = 40), typically starting from 25 GBq [ 11 C]methyl iodide. The molar activity was 100 ± 60 GBq/ μmol and the radiochemical purity was 98.5 ± 1%. The palladium concentration in the product solution was 8.4 ± 0.6 ng/ ml. Figure 1 shows a typical whole blood activity curve from one subject as well as the mean plasma-to-whole blood ratio and mean fraction of intact tracer in the eight subjects with arterial blood samples. The mean plasma-to-whole blood ratio increased from 1.3 to 1.6 during the course of the scans and, on average, 26.6% of unmetabolized tracer was left in plasma at 90 min postinjection.

| Tracer kinetic analysis
Invasive kinetic models were evaluated: 1TCM, 2TCM, and plasmainput Logan. Typical TACs for the cerebellum, thalamus, and amygdala, from one representative subject, with the fits of 1TCM and 2TCM are given in Figure 2a-c. As expected, since the cerebellum has no or very low aromatase expression, both 1TCM and 2TCM were able to fit the cerebellum TAC. The 2TCM was able to fit the data well and was also preferred over 1TCM according the Akaike criteria for all TACs, except one TAC in one subject. However, it was not possible to robustly determine BP ND directly from the 2TCM as the standard errors frequently exceeded 25% of the BP ND value itself. The kinetic parameters from 1TCM and 2TCM are given in Table 1, and as expected, the highest V T was found in thalamus.  Table 2. A mean 1TCM k 2 -value in the cerebellum was calculated to 0.28 ± 0.015 and was used as the efflux constant in the LRTM DVR-1 calculations. However, omission of the k 2 term did not significantly change the DVR-1 values (R 2 = 1.00; slope = 1.00, CI = 0.99-1.01) and for the remainder of the study, only LRTM DVR-1 values without the k 2 term are reported.

Plasma-input
In Figure 4, the relationships between 2TCM DVR-1 and the reference models are illustrated and corresponding slope, bias, and 95% limit of agreement are given in Table 2. An overall high correlation was seen between the plasma model and both LRTM and SRTM (R 2 ≥ 0.85). LRTM DVR-1 values were overestimated (slope = 1.19, CI = 1.03-1.36), while SRTM BP ND values were not (slope 1.01, CI = 0.89-1.14). SUVR-1 values showed a poorer correlation and were overestimated compared to 2TCM DVR-1. Multilinear regression did not show a significant effect of challenge on the relationship between SRTM BP ND and 2TCM DVR-1.

| Parametric images
Parametric images from one subject ( Figure 5) Table 3, with the slopes of the orthogonal regression and R 2 values. In Table 3 is also slope and R 2 values given between SUVR-1 values and SRTM. All three parametric methods correlated well with VOI-based analysis of SRTM BP ND (R 2 ≥ 0.90). Agreement was good for RPM and LRTM with a slight overestimation compared to SRTM BP ND (slope = 1.07 and 1.08) but with a very high correlation (R 2 ≥ 0.99). Although RPM2 BP ND also showed a high correlation with the VOI-based analysis, it was overestimated compared to SRTM BP ND (R 2 = 0.90; slope = 1.48). The SUVR-1 values were also overestimated compared to SRTM BP ND and had a lower correlation than the other TA B L E 1 Regional kinetic parameters from the 1TCM and 2TCM analysis. Values are the mean for each region with the range of the parameter in parentheses  60-min data sets compared to the full scan duration (R 2 = 0.91-0.93) but was still highly overestimated (slope = 1.58-2.15). In all graphs in Figure 6 there are three points with a value around 1.5. These points correspond to the thalamus region, in the three different scans, of one subject. p value for all correlations was <0.0001.

| D ISCUSS I ON
In the present study, tracer kinetics for quantitation of the novel aromatase PET tracer [ 11 C]cetrozole were evaluated, using both plasma-input as well as reference tissue models. Different methods To identify the optimal model to describe the in vivo kinetics of [ 11 C]cetrozole PET, VOI-based analysis was performed using single-tissue and two-tissue reversible plasma-input compartment models and Logan graphical analyses (Akaike, 1974;Logan et al., 1990).
Additionally, simplified reference tissue model (Lammertsma & Hume, 1996), and Logan reference tissue model, analyses (Logan et al., 1996) were performed with cerebellum as reference region, as done by (Takahashi et al., 2014). The [ 11 C]cetrozole TACs were best described by the 2TCM, according to the AIC, compared to 1TCM.

DVR-1 values from plasma-input Logan analyses were overestimated
compared to the 2TCM DVR-1 values, while the V T values from the two models were in better agreement. Both SRTM and LRTM agreed well with the plasma-input binding results, but the highest agreement was observed between SRTM BP ND values and 2TCM DVR-1.
We did not correct for clearance from the reference tissue in the LRTM analyses, as omission of the k 2 term did not change the DVR-1 values. Further, substantiating this choice, we found excellent agreement between the plasma-input Logan and LRTM DVR-1 values.
Reference tissue models correlated well with plasma-input models, thus indicating that arterial blood sampling is not necessary. This is  and Logan reference tissue model based on the average k 2 ' (BP ND ) (Takahashi et al., 2014). Simulations showed that 2TCM DVR-1 and SRTM BP ND can be estimated with a high accuracy and precision in high-binding regions such as thalamus and amygdala. Accuracy and precision in absolute terms are comparable in low-binding regions such as the raphe, but relative accuracy and precision are poor due to the very low absolute values of DVR-1 and BP ND in these regions.

SRTM BP ND
Additionally, as a semi-quantitative measure, SUVR was analyzed to evaluate potential simplification of the data acquisition. SUVR overestimated the binding compared to both the plasma-input and the reference tissue models on an 80-90 min interval. Although The present results are expected to apply also to men, as sex is not expected to influence the outcome parameters of the present study.
The investigated VOIs (i.e., thalamus, hypothalamus, putamen, raphe nuclei, and amygdala) are putative brain regions of interest to behavior and mental health. A validated, quantitative PET measure for [ 11 C]cetrozole is therefore useful to investigate testosterone-estrogens dynamics in the healthy brain (Takahashi et al., 2014), as well as how these are influenced by pathophysiological conditions or pharmacological treatments.

| CON CLUS ION
SRTM using cerebellum as reference region appears to be the optimal method to noninvasively estimate [ 11 C]cetrozole binding.
SRTM BP ND values were highly correlated with the plasma-input models, with the highest agreement between SRTM BP ND and cetrozole can be employed as PET tracer for relatively short-dynamic brain scans and reference tissue-based analyses to image aromatase in the human brain. University for quantification of palladium by ICP-AES.

CO N FLI C T O F I NTE R E S T
No conflicting interests exist.

AUTH O R CO NTR I B UTI O N S
All the authors had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the