PET imaging was carried out as part of a phase 1 study of the experimental drug elacestrant, which aimed to determine ER availability [10]. Elacestrant was developed as a selective estrogen receptor degrader that crosses the blood–brain barrier for the treatment of estrogen receptor positive breast cancer brain metastases. The drug was shown to competitively bind to ERs [11] in the same binding pocket as estrogen and [18F]FES [12]. The study was approved the independent ethics committee of the foundation “evaluation of ethics in biomedical research” (CCMO code: NL49312.056.14) and was performed in accordance with standards for Good Clinical Practice, in full compliance with the principles of the 1964 Declaration of Helsinki. Seven healthy post-menopausal women (age 61.5 ± 9.7) were included in the study. Informed consent was obtained from all individual participants included in the study. Exclusion criteria were the use of any concomitant medication, smoking or any other substance dependence. At baseline, a 3D T1-weighted MRI and a dynamic [18F]FES PET scan were acquired. Subjects were treated daily with an oral dose of elacestrant (500 mg) for 7 days, to reach steady-state levels in plasma, without hormonal replacement therapy. In 4 subjects, [18F]FES PET was repeated 4 h after the last drug dose.

A structural 3D T1-weighted MRI sequence (matrix size 256 × 256 × 3, voxel size 0.97 × 0.97 × 20, repetition time 11.12 ms, echo time 4.60 ms) on a 3 Tesla Ingenuity TF system (Philips, Netherlands) was acquired for each subject to be used as individual anatomical reference for spatial normalization and co-registration of the PET scans.

A catheter was placed in a brachial vein for intravenous administration of the tracer, and a cannula was inserted into the radial artery of the opposite wrist for blood sampling. PET/CT images were acquired with a Biograph mCT system (Siemens, Knoxville, USA). After a low-dose CT was acquired, a bolus (8.3 ml) of [18F]FES (baseline 199 ± 6 MBq; post-dose 209.0 ± 13 MBq) was intravenously injected (0.5 ml/s) and a 90-min dynamic PET scan of the brain was started. The dynamic PET data were reconstructed using a time-of-flight version of the 3D ordered-subsets-expectation–maximization algorithm (3 iterations, 24 subsets) and corrected for decay, attenuation scatter and random coincidences. List-mode data were reconstructed into 33 temporal frames: 6 × 5 s; 4 × 10 s; 4 × 15 s; 3 × 30 s; 3 × 60 s; 4 × 150 s; 3 × 300 s; 6 × 600 s. The final images had a matrix size of 400 × 400 × 111 and a voxel size of 2.03 × 2.03 × 2 mm.

The radioactivity concentration in arterial blood was continuously measured during the first 30 min of the scan, using an automatic blood sampling system (Veenstra Instruments, Joure, Netherlands). In addition, seven manual samples were taken at approximately 5, 10, 20, 30, 40, 60 and 90 min after tracer injection for calibration of the automated sampler and metabolite analysis. In each sample, the radioactivity concentration in 250 µl of whole-blood and plasma were measured with an automated gamma-counter (Wizard2480, PerkinElmer, USA). The radioactivity concentration was expressed as standardized uptake values (SUV). SUV values were calculated by dividing the measured radioactivity concentration (kBq/mL) by the ratio of the injected dose (kBq) and body weight (g) of the subject. It was assumed that 1 g equals 1 mL.

For metabolite analysis, 50 μl aliquots of the plasma samples were diluted with 100 μl of acetonitrile and centrifuged (5 min, 15000 g). A 2.5 µl aliquot of the supernatant, was analyzed by thin-layer chromatography, using a silica gel 60 F254 TLC plate (Merck, Germany) and n-hexane/ethyl acetate (7/3) as the mobile phase. A phosphor storage screen (PerkinElmer, USA) was exposed to the TLC plate for approximately 18 h. The phosphor storage screen was scanned with a Cyclone Imaging System (PerkinElmer, USA) and analyzed with OptiQuant Software version 3.0 to determine the percentage of intact [18F]FES in plasma. A one-exponential function was fitted to the metabolite data and was used to generate a metabolite-corrected plasma input function.

PET images were co-registered to the corresponding T1-weighted MRI scan of the same subject, using PMOD version 4.1 (PMOD Technologies LLC, Zürich, Switzerland). Head motion correction was applied to the PET data, if necessary. MRI scans were spatially normalized to Montreal Neurological institute (MNI) space [13] and the transformation matrix was used to align the co-registered PET images. Volumes-of-interest (VOIs) for individual brain regions were obtained from the Hammers atlas [14]. The 83 available brain structures within the Hammers atlas were aggregated into 15 brain regions, as no differences between left and right were expected and no differences within cortical regions. Regions with expected high ER expression, e.g., thalamus, hippocampus and amygdala, were analyzed as separate structures. VOIs for white matter, grey matter, and the whole brain were segmented from the MRI dataset in SPM12 (Wellcome Trust Center for Neuroimaging, UK) with a probability map threshold of 0.5. A VOI for the pituitary gland was drawn manually for each subject in PET-space, using a 3D iso-contour at 40% of the maximum uptake. TACs for each VOI were generated for kinetic modeling.

Pharmacokinetic modeling was performed using the metabolite-corrected plasma TAC as the input function and the TAC of whole-blood for blood volume correction. To improve the accuracy of the fits, frame duration and frame mid-time decay were applied as weighting factors for all evaluations. The one-tissue compartment model (1T2k), irreversible two-tissue compartment model (2T3k) and reversible two-tissue compartment model (2T4k) were assessed for fitting the regional TACs, using a fitted fractional volume of blood (VB). The 2T4k model was further explored using the VB as a fixed value of 0.05 [15] and by fixation of the influx-efflux rate constant ratio (K1/k2) to the K1/k2 ratio of the whole-brain. In addition, Logan graphical analysis was performed with a starting time (t*) of 20 min.

The Akaike Information Criterion (AIC) was used to select the most appropriate model. The standard error in the parameter estimated by the compartment models was used to determine the reliability of in the total volume of distribution (VT) and non-displaceable binding potential (BPND) estimates, using an arbitrary cut-off value of 25%. The accuracy of SUV and Logan graphical analysis derived VT was determined via correlational analyses with the macro-parameters of the optimal compartment model. The change in ER availability by the drug was calculated using the Lassen plot [16].

All statistical analyses were performed using IBM SPSS Statistics (Version 23, Armonk, NY, USA). Differences in kinetic parameter estimates between the baseline and post-dose scans were assessed by generalized estimated equation, using the main effects “brain region” and “treatment” and the interaction “brain region × treatment”. A Bonferroni post-hoc analysis was used for multiple comparisons correction. Results are reported as mean ± standard deviation (SD) and were considered statistically significant if the null hypothesis was rejected at a probability of 95% (p < 0.05). The correlation between parameters from different models were examined by Pearson linear regression analysis.