Kinetic Modelling and Test–Retest Reproducibility for the Dopamine D1R Radioligand [11C]SCH23390 in Healthy and Diseased Mice

Purpose Our aim in this study was to compare different non-invasive pharmacokinetic models and assess test–retest reproducibility of the radioligand [11C]SCH23390 for the quantification of dopamine D1-like receptor (D1R) in both wild-type (WT) mice and heterozygous (HET) Q175DN mice as Huntington’s disease (HD) model. Procedures Adult WT (n = 9) and HET (n = 14) mice underwent a 90-min [11C]SCH23390 positron emission tomography (PET) scan followed by computed tomography (CT) to evaluate the pharmacokinetic modelling in healthy and diseased conditions. Additionally, 5 WT mice and 7 HET animals received a second [11C]SCH23390 PET scan for test–retest reproducibility. Parallel assessment of the simplified reference tissue model (SRTM), the multilinear reference tissue model (MRTM) and the Logan reference tissue model (Logan Ref) using the striatum as a receptor-rich region and the cerebellum as a receptor-free (reference) region was performed to define the most suitable method for regional- and voxel-based quantification of the binding potential (BPND). Finally, standardised uptake value ratio (SUVR-1) was assessed as a potential simplified measurement. Results For all models, we measured a significant decline in dopamine D1R density (e.g. SRTM = − 38.5 ± 5.0 %, p < 0.0001) in HET mice compared to WT littermates. Shortening the 90-min scan duration resulted in large underestimation of striatal BPND in both WT mice (SRTM 60 min: − 17.7 ± 2.8 %, p = 0.0078) and diseased HET (SRTM 60 min: − 13.1 ± 4.1 %, p = 0.0001). Striatal BPND measurements were very reproducible with an average test–retest variability below 5 % when using both MRTM and SRTM. Parametric BPND maps generated with SRTM were highly reliable, showing nearly perfect agreement to the regional analysis (r2 = 0.99, p < 0.0001). Finally, SRTM provided the most accurate estimate for relative tracer delivery R1 with both regional- and voxel-based analyses. SUVR-1 at different time intervals were not sufficiently reliable when compared to BPND (r2 < 0.66). Conclusions Ninety-minute acquisition and the use of SRTM for pharmacokinetic modelling is recommended. [11C]SCH23390 PET imaging demonstrates optimal characteristics for the study of dopamine D1R density in models of psychiatric and neurological disorders as exemplified in the Q175DN mouse model of HD. Supplementary Information The online version contains supplementary material available at 10.1007/s11307-020-01561-1.


Introduction
Dopamine D 1 -like receptors (D 1 R) are post-synaptic G protein-coupled receptors widely distributed in the central nervous system [1]. They are primarily expressed in the caudate and putamen nucleus with lower levels in limbic and cortical structures [2,3]. Under physiological condition, dopamine D 1 R are involved in the modulation of the reward system, motor control and spatial working memory [4,5]. However, alterations in dopamine release and dopamine D 1 R have been associated with the phenotype of different neurological and neuropsychiatric disorders, including Parkinson's disease [6], schizophrenia [7], drug addiction [5,8] and Huntington's disease (HD) [9].
The value of [ 11 C]SCH23390 as a radiotracer to measure dopamine D 1 R using PET imaging in the putamen and the caudate nucleus has been largely demonstrated in clinical settings. In larger animals and humans, [ 11 C]SCH23390 is commonly quantified using a 50-90-min dynamic PET scan with reference region-based kinetic modelling with either the simplified reference tissue model (SRTM) or the multilinear reference tissue model (MRTM) given their high test-retest reliability [14][15][16][17][18]. Nonetheless, kinetic modelling and testretest reproducibility of [ 11 C]SCH23390 in mice has not yet been investigated, an important limitation for its application to preclinical drug development. Indeed, dopamine D 1 R PET imaging is a potential phenotypical readout for therapeutic efficacy in neurological and neuropsychiatric disorders. For instance, dopamine D 1 R is markedly reduced in individuals with HD, as demonstrated in vivo using [ 11 C]SCH23390 PET imaging [19][20][21][22]. This phenotype was also confirmed in vitro in the transgenic R6/2 and BACHD mouse models of HD using [ 3 H]SCH23390 autoradiography [23] as well as in vivo in the knock-in Q175DN mouse model of HD using [ 11 C]NNC-112 PET imaging [24]. Since the performance of a radioligand can vary with receptor density, we focused on the methodological characterisation of [ 11 C]SCH23390 PET imaging using both wild-type (WT) mice as well as heterozygous (HET) Q175DN littermates [25,26]. Our aims in the present study were threefold: firstly, investigate the capability of [ 11 C]SCH23390 PET imaging to quantify dopamine D 1 R changes in Q175DN mice; secondly, compare radioligand performance, including time stability of outcome parameters, following regionaland voxel-based kinetic modelling using three different reference-based methods in both genotypes and assess possible semi-quantitative approaches and thirdly, measure the test-retest reproducibility of [ 11 C]SCH23390 PET imaging in both genotypes.

Animals
Adult 10-month old heterozygous (HET, n = 14) male knock-in Q175DN mice (C57BL/6J background and same disease progression as the parental Q175 model [25,26] with the removal of the neo-cassette used for the insertion of the expanded CAG sequence) and age-matched wildtype (WT, n = 9) littermates from Jackson Laboratories (Bar Harbour, Maine, USA) were included in the study. Given the sporadic congenital portosystemic shunt occurring in C57BL/6J mice [27], all animals were screened at Jackson Laboratories before shipment in order to avoid this variable as a confounding factor. Upon arrival, animals were group-housed in individually ventilated cages under a 12-h light/dark cycle in a temperatureand humidity-controlled environment. Food and water were provided ad libitum and more than one week of habituation was allowed before the start of the procedures. [ 11 C]SCH23390 PET imaging was performed for all animals (HET, n = 14; WT, n = 9) for evaluation of the pharmacokinetic modelling in both healthy and disease mouse brains. For assessment of the [ 11 C]SCH23390 testretest reproducibility, 7 HET Q175DN and 5 WT littermates underwent a second [ 11 C]SCH23390 PET scan 5.6 ± 1.6 days following the first scan. The Netherlands) based on the one-pot strategy [11] via common N-methylation of the desmethyl precursor. Briefly, [ 11 C]MeI was added to a precooled (− 20°C) reaction vessel containing N-desmethyl-SCH23390 (1.0 mg ± 10 %) and aqueous NaOH (1 M, 5 μl) in anhydrous DMF/DMSO (ratio 50/50, 300 μl) at room temperature. The reaction lasted for 8 min at 50°C to synthesise [ 11 C]SCH23390. The product was subsequently collected using a reverse phase semipreparative HPLC column (Phenomenex Luna C18, 250 × 10 mm, 10 μm) with a biocompatible mobile phase (NaOAc 0.05 M pH 5.5/EtOH 96 %, 50/50, v/v) at a flow rate of 3.0 ml/min. Finally, the collected product was diluted (1 in 5) with saline solution through a sterile membrane filter in order to obtain an intravenously injectable solution. The radiochemical purity of the produced [ 11 C]SCH23390 was determined using an isocratic HPLC method (Phenomenex Luna C18, 150 × 4.6 mm, 5 μm) with NaOAc 0.05 M pH 5.5/ACN, 70/30 (v/v) as a mobile phase, flow rate 1 ml/ min and UV absorption at 280 nm. Molar activity at the end of the synthesis was 72.4 ± 4.7 GBq/μmol, with an average radiochemical purity greater than 99 %.

PET Acquisition and Reconstruction
MicroPET/computed tomography (CT) images were acquired using two virtually identical Siemens Inveon PET/ CT scanners (Siemens Preclinical Solution, Knoxville, USA). Animal preparation was performed as previously described [28,29]. A bolus of radioligand was injected using an automated pump (Pump 11 Elite, Harvard Apparatus, USA) over a 12-s interval (1 ml/min) immediately after the start of the 90-min dynamic PET scan. [ 11 C]SCH23390 was injected in a trace dose with WT mice receiving an average of 1.18 ± 0.28 μg/kg and HET littermates an average of 1.44 ± 0.37 μg/kg (p = 0.17) keeping the cold mass within 2.0 μg/kg to avoid potential mass effect. On the scan day, body weight was 30.9 ± 2.2 g and 26.9 ± 1.0 g for the WT and HET mice (p = 0.0015), respectively, with an injected activity of 4.6 ± 0.9 MBq for WT animals and 5.5 ± 1.5 MBq for HET Q175DN mice (p = 0.19). A significant reduction in body weight in this animal model of HD is commonly observed starting at 6 months of age [30,31]; however, since we are performing dynamic acquisition and pharmacokinetic modelling, alterations in body weight are taken into account, and therefore, they were not expected to affect the quantification.
PET data were acquired in a list mode format and followed by a 10-min 80 kV/500 μA CT scan performed on the same gantry for attenuation correction and coregistration purposes. One WT animal received an injection that extravasated for the retest scan; therefore, it was omitted from the test-retest analysis. Acquired PET data were histogrammed and reconstructed into 39 frames of increasing length (12 × 10 s, 3 × 20 s, 3 × 30 s, 3 × 60 s, 3 × 150 s and 15 × 300 s) using a list mode iterative reconstruction with proprietary spatially variant resolution modelling in 8 iterations and 16 subsets of the 3D ordered subset expectation maximisation (OSEM 3D) algorithm [32]. Normalisation, dead time and CT-based attenuation corrections were applied. PET image frames were reconstructed on a 128 × 128 × 159 grid with 0.776 × 0.776 × 0.776 mm 3 voxels.

Image Analysis and Processing
Image analysis and processing of the PET data were performed in PMOD 3.6 software (Pmod Technologies, Zurich, Switzerland). Based on our previous observation that the use of magnetic resonance imaging (MRI) templates for spatial normalisation and VOI definition improves the accuracy of the regional quantification of PET data with focal uptake (as we previously investigated with the radioligand [ 18 F]MNI-659 for phosphodiesterase 10A [31]), an MRI template for each genotype was obtained from another independent cohort of age-matched Q175DN WT (n = 6) and HET (n = 6) mice. WT and HET MR images were rigidly aligned to the space of the first animal and averaged to generate genotypic-specific MR templates. Since all animals were aligned to the same animal, both MRI templates are in the same space. PET registration was achieved by the rigid spatial normalisation of the individual CT images to the MR templates and then apply the same rigid transformation to the PET images. All images were visually checked for accuracy following spatial transformation. The volumes of interest (VOIs) were manually delineated on the genotype-specific MRI templates, and regional timeactivity curves (TACs) were extracted for the striatum and whole cerebellum in order to perform kinetic modelling. No volumetric difference in brain structures was observed between the WT and HET MRI templates. The final volumes were as follows: the striatum 0.0215 cm 3 for WT and 0.0208 cm 3 for HET mice, while the cerebellum was the same for both genotypes (0.0507 cm 3 ). The former was considered the receptor-rich region, while the latter was used as the receptorfree region [33]. Cortical structures were not considered given fivefold lower receptor density and low selectivity over serotoninergic 5-HT 2A receptors [18].

Kinetic Modelling
We measured the non-displaceable binding potential (BP ND ) analysing 3 different pharmacokinetic models. We compared the SRTM [34], the MRTM [35] and the Logan reference tissue model (Logan Ref) [36] in order to determine the most appropriate for estimation of [ 11 C]SCH23390 BP ND in the brain of both healthy WT animals and diseased HET Q175DN mice. When applying SRTM and MRTM, the relative tracer delivery R 1 was also measured, while for the Logan Ref, the linear phase (t * ) was fixed at t * = 15 min with k 2 ' derived with MRTM. MRTM-based k 2 ' was preferred over the SRTM-based k 2 ' given the lower % standard error (SE) of the k 2 ' estimation observed with the former model (3.3 ± 0.3 % SE) compared to the latter (5.6 ± 1.9 % SE).
The relative performance of each model to fit the regional PET data was assessed by calculating the goodness-to-fit of the models using the Akaike Information Criterion (AIC) [37].
Time stability of the estimated striatal BP ND (and R 1 if applicable) was analysed for each investigated model by repeatedly excluding the last 5 min of PET acquisition from 90 min down to 45 min. The 90-min BP ND and R 1 were considered the reference outcome, and all the values obtained with shorter acquisitions were compared to the 90-min values. Variation in the estimation of BP ND based on a shorter acquisition was considered acceptable only if the average percentage difference was lower than 10 % with an inter-individual standard deviation below 5 % when compared to the 90-min PET acquisition as previously applied [28,29].
Parametric BP ND maps were generated using SRTM [38], MRTM [35] and Logan reference tissue model [36] with the k 2 ' as calculated with MRTM. SRTM2 [39] and MRTM2 [35] were also explored. Nonetheless, they did not improve the reliability of the parametric maps as SRTM was already accurate and MRTM2 was still presenting failed voxels; thus, we report SRTM and MRTM in order to investigate the agreement between parametric maps and regional analysis. Besides, parametric R 1 maps were generated using SRTM and MRTM. For all models, the striatum was considered the receptor-rich region, while the cerebellum represented the receptor-devoid region (reference region). Parametric images were cropped using the brain mask of the MRI template, represented as group averages and overlaid onto the genotype-specific MRI templates for anatomical reference.
Additionally, we wanted to relate striatal BP ND values obtained using the regional-and voxel-based analyses to determine the reliability of the parametric maps within each pharmacokinetic model. To this end, following the generation of the parametric BP ND maps, we applied the VOI generated for the regional analysis in order to average the BP ND of each striatal voxel. Next, we compared striatal BP ND values calculated using the voxel-based maps and regional analysis to assess their agreement within each pharmacokinetic model.
Finally, we explored the applicability of a simplified approach for the quantification of striatal [ 11 C]SCH23390 binding by measuring the ratio of the striatal standardised uptake values (SUV) over the cerebellar SUV (denoted as SUVR) based on the scan intervals 40-60 min as well as 70-90 min. The resulting measurement, SUVR-1, was compared to BP ND .

Statistical Analysis
All the data were normally distributed as assessed with the Shapiro-Wilk test; therefore, parametric analyses were performed. Unpaired T-tests were performed to compare scan parameters, BP ND , SUVR-1 and R 1 between genotypes during both VOI-based and voxel-based analyses. Given the sample size of WT mice in the test-retest study (n = 4), a comparison of the test-retest scan parameters was performed using paired T-test under the assumption of normality in the distribution. All correlations between variables were investigated with Pearson's correlation tests and linear regression analyses. Bland-Altman plots, reported as bias and 95 % limits of agreement (1.96 × SD), were used to assess agreement between test-retest scans in the estimation of striatal BP ND and R 1 . In addition, the reproducibility of the test-retest data was determined by the intraclass correlation coefficient (ICC), relative test-retest variability (TRV) and absolute TRV (aTRV). A mixed-model reliability analysis for absolute agreement was performed for the assessment of ICC with genotype included as the fixed effect in the model.
TRV was calculated as follows: while aTRV was measured as follows: Finally, the mean ± standard deviation (SD) of the intraanimal coefficient of variation (COV) was calculated as follows: where G represents the group, N is the number of animals in the group, x G i and SD G i are respectively the mean and standard deviation of the test and retest values for animal i. All aforementioned statistical analyses were performed using GraphPad Prism (v8.4) statistical software except for ICC, calculated in JMP Pro 14 software (SAS Institute Inc., USA). The effect size d, determined using G * Power software (http://www.gpower.hhu.de/), was calculated using the mean and variance of each experimental group (WT and HET). Data are represented as mean ± SD, unless specified otherwise. All tests were two-tailed and significance was set at p G 0.05.

Time Stability of the Striatal [ 11 C]SCH23390 BP ND Estimates
Next, in order to assess the time stability of striatal BP ND , we investigated the effect of the duration of the PET acquisition on its estimation for all the kinetic models (SRTM, MRTM and Logan reference). As depicted in Fig. 2a, when normalising shorter scan durations to the 90-min BP ND for each subject, a large

Test-Retest Reproducibility of Striatal [ 11 C]SCH23390 BP ND Estimates
For the test and retest scans, no significant methodological confounding factors were observed (Suppl.    3b and Table 2).

Parametric [ 11 C]SCH23390 BP ND Maps
Average voxel-based parametric BP ND maps for WT mice and HET Q175DN animals are shown in Fig. 4a. Visually, parametric BP ND maps generated with SRTM resulted in reliable maps showing a nearly perfect agreement to the regional analysis (r 2 = 0.99, p G 0.0001) and no deviation from the identity line (slope = 1.03) (Fig. 4b). On the contrary, parametric BP ND maps obtained using MRTM were characterised by many failed voxels randomly scattered across different animals (Fig. 4a). Besides, even though the MRTM-based striatal BP ND values obtained using the parametric maps agreed well with the VOI-based analysis (r 2 = 0.96, p G 0.0001), they deviated from the identity line (slope = 1.15) (Fig. 4b). The Logan reference method produced an agreement between striatal BP ND values based on parametric maps and regional analysis similar to MRTM (r 2 = 0.96, p G 0.0001) (Fig. 4b).

Applicability of Simplified [ 11 C]SCH23390 Measurements
Given the extensive 90-min dynamic acquisition recommended for [ 11 C]SCH23390 BP ND estimation, we explored the applicability of SUVR-1 based on the scan intervals 40-60 min as well as 70-90 min. As shown in Suppl.

Discussion
Our study compared three reference region-based pharmacokinetic models to quantify [ 11 C]SCH23390 PET imaging to determine the optimal methodology for striatal BP ND and R 1 estimation in the mouse brain. We also evaluated this radioligand in a diseased condition characterised by the reduction of dopamine D 1 R density in the Q175DN HD mouse model, which exhibits several HD phenotypic hallmarks [24-26, 31, 40], including impairment of the dopaminergic system [24,41,42]. Reduction of striatal dopaminergic D 1 and D 2/3 receptors has been largely documented in patients with premanifest and manifest HD compared to healthy controls [19][20][21][22]43]. Here, we measured an in vivo dopamine D 1 R reduction of 38 % in HET Q175DN at 10 months of age using [ 11 C]SCH23390, similar to the 34 % reduction previously reported at 9 months of age using [ 11 C]NNC-112 [24]. The relevance of these findings is further substantiated by previous clinical reports describing a subtle reduction in presymptomatic patients [22] followed by a 35-40 % decline in D 1 R with the progression of the disease [19] as measured in vivo with [ 11 C]SCH23390 PET imaging. To assess [ 11 C]SCH23390 kinetic modelling, we focused on three reference methods, namely, SRTM, MRTM and Logan reference, using the cerebellum as a reference region. When evaluating the VOI-based analysis, MRTM resulted in the best performance, as supported by the AIC, with SRTM being a valid alternative, while Logan Ref displayed the lowest accuracy in striatal BP ND estimation. However, when conducting voxel-based analysis, BP ND [ 11 C]SCH23390 maps obtained with MRTM displayed failed voxels in nearly half of the WT animals in the proximity of high uptake structures as depicted in Suppl. Fig. 3 (see ESM). Additionally, MRTM and Logan Ref had similar suboptimal performances (r 2 = 0.96 and r 2 = 0.96, respectively), while SRTM was the most accurate (r 2 = 0.99). Even though from a strictly VOI-based kinetic modelling perspective, MRTM proved to be slightly more accurate than SRTM; for studies where voxel-based parametric maps are also of interest, SRTM represents the optimal balance for both VOI-and voxel-based analyses. Previous studies investigating the kinetics in humans and baboons have reported using SRTM to estimate BP ND [18,34,44,45], while more recently, the simplified MRTM method, MRTM2, was applied in a human study [14]. Additionally, reported radioligand kinetics in humans and larger animals suggest a faster striatal washout when compared to mice, with PET scan acquisition ranging from 50 to 90 min [14][15][16][17][18].
Evaluation of time stability of striatal BP ND showed that shortening the acquisition to less than 90 min led to large underestimations of striatal BP ND in both healthy and diseased mice for all assessed kinetic models, which was expected given the slow washout of [ 11 C]SCH23390 in the striatum, especially in healthy animals with higher receptor density. Thus, based on a 90-min acquisition, a scan duration of at least 80 min is necessary to accurately estimate striatal BP ND in mouse brain. Longer scan acquisition might be potentially required to characterise even more precisely the dynamics of the receptor. Nonetheless, within the 90-min acquisition [ 11 C]SCH23390 kinetics appeared to be stable and, due to the short half-life of the radioisotope (20.3 min), a longer acquisition may likely introduce noise hampering the quantification.
Since reliable quantification of the receptor density is fundamental to conduct pharmacological, interventional and longitudinal studies in animal models, we investigated both healthy WT and diseased HET Q175DN mice, in which the dopamine D 1 R density is reduced [24], to evaluate testretest reproducibility of [ 11 C]SCH23390 striatal BP ND in mouse brain. BP ND measurements had an extremely low test-retest variability with absolute TRV ranging from 0 to 11 % (on average below 6 %) for the WT and HET mice with both SRTM and MRTM. Striatal BP ND quantification was also reproducible with both SRTM (ICC = 0.748) and MRTM (ICC = 0.810). [ 11 C]SCH23390 test-retest reproducibility has not been assessed before in rodents, but human studies have reported high test-retest stability with ICC values ranging from 0.81 to 0.94 [14,44,46]. Estimation of striatal [ 11 C]SCH23390 BP ND recommends a 90-min dynamic acquisition. To avoid such extensive acquisition, we evaluated static time intervals quantified with SUVR-1 as an alternative since it has been previously reported a robust approach with other radioligands [47][48][49]. Nevertheless, it did not prove sufficiently reliable when compared to BP ND with either time interval (40-60 min: r 2 = 0.609; 70-90 min: r 2 = 0.651). A possible reason for the lack of reliability may be associated to the noise levels in the cerebellum as a consequence of the low uptake in parallel to the rapid decay of the 11 C radioisotope.
Finally, R 1 could be reliably estimated using the SRTM and MRTM methods for both regional-and voxel-based approaches. Noteworthy, estimation based on SRTM was more reliable than MRTM with both regional-and voxelbased analyses. Striatal R 1 estimation using SRTM was extremely stable down to 45-min acquisition for both WT mice (− 2.84 %) and HET Q175DN animals (− 1.98 %) compared to 90 min, with extremely good test-retest reproducibility (bias = 2.5 %).
In the brain, dopamine D 1 R are found mainly in the terminal structures of the dopaminergic system (striatum) with lower density in the cortical areas [3,50], so dopamine D 1 R PET imaging could potentially be applied to cortical structures to study psychiatric disorders such as schizophrenia [51]. However, when considering the cerebral binding affinity, in vitro studies with SCH23390 reported a dissociation constant (K D ) of 0.14-0.37 nM for dopamine D 1 R and K D of 19.9-37 nM for 5-HT 2A R in the rodent brain [18,[52][53][54]. Similarly, an in vitro K D of 0.18 nM for dopamine D 1 R and K D of 18 nM for 5-HT 2A R have been described for NNC-112 [12,18]. Thus, the selectivity of both [ 11 C]SCH23390 and [ 11 C]NNC-112 towards the serotoninergic 5-HT 2A receptors are negligible (circa 100-fold lower) compared to the dopamine D 1 R. Accordingly, in vivo PET studies in baboons and humans demonstrated that approximately a quarter of cortical BP ND of [ 11 C]SCH23390 and [ 11 C]NNC-112 is driven by 5-HT 2A R binding [18,55,56]. Although cortical selectivity has not been investigated in rodents, the combined contribution of dopamine D 1 R and 5-HT 2A R to the cortical signal represents a shortcoming of the current dopamine D 1 R radioligands and should be addressed in future studies. Consequently, caution is warranted before interpreting cortical-binding changes with the currently available dopamine D 1 R radioligands. Without more selective radioligands, coinjection of a 5-HT 2A R blocker and the radioligand might enable accurate quantification of D 1 R cortical binding in vivo, as previously postulated [56].

Conclusion
We recommend a 90-min acquisition and the use of SRTM for pharmacokinetic modelling of [ 11 C]SCH23390 in healthy and diseased mice in order to achieve reproducible values and reliable parametric BP ND and R 1 maps. Our findings demonstrate the utility of [ 11 C]SCH23390 PET imaging for the study of dopamine D 1 R density in psychiatric and neurological disorders as exemplified in the Q175DN HD mouse model.
Acknowledgements. Antwerp University funded the work through a partial assistant professor position for JV and LW and a full professor position for SStr and SSta. LW and SStr are also supported by Antwerp University Hospital through a departmental position. The authors thank Philippe Joye, Caroline Berghmans, Silvia Incardona and Eleni Van der Hallen of the Molecular Imaging Center Antwerp (MICA) for their valuable assistance.
Compliance with Ethical Standards. All applicable institutional and/or national guidelines for the care and use of animals were followed. Experiments were performed according to the European Committee Guidelines (decree 2010/63/CEE) and the Animal Welfare Act (7 USC 2131), and they were approved by the Ethical Committee for Animal Testing (ECD 2018-82) at the University of Antwerp (Belgium).