# Quantification of synaptic density with [11C]UCB-J PET

[11C]UCB-J binds to synaptic vesicle glycoprotein 2A (SV2A), which is ubiquitously and homogeneously located in synapses across the brain (Bajjalieh et al., 1994; Janz & Südhof, 1999). [11C]UCB-J could be displaced by levetiracetam and padsevonil, which are SV2A-selective drugs (Finnema et al., 2016; Koole et al., 2019); this confirms that [11C]UCB-J is specifically binding to SV2A. Increased neural firing does not affect [11C]UCB-J binding in the brain, despite vesicle exocytosis, and marked increase in blood flow (Smart et al., 2021). [11C]UCB-J can be used to assess occupancy of SV2A binding drugs (Nicolas et al., 2016; Finnema et al., 2019).

PET imaging of patients with temporal lobe epilepsy revealed unilateral decreased binding, suggesting that [11C]UCB-J can be used to detect synaptic loss in vivo in human subjects (Finnema et al., 2016). Binding is reduced in the seizure onset zone, with larger asymmetry index than observed with FDG PET (Finnema et al., 2020). [11C]UCB-J PET has revealed synaptic loss in brainstem nuclei in Parkinson's disease (Matuskey et al., 2020; Wilson et al., 2020), and in the frontal and anterior cingulate cortices in schizophrenia (Onwordi et al., 2020). In Alzheimer's disease (AD), widespread decrease in [11C]UCB-J uptake has been observed (Chen et al., 2018; Mecca et al., 2020), although with marked overlap between patients and control subjects (Tuncel et al., 2021). In AD, [11C]UCB-J and [18F]FDG show high inter-tracer regional correlations, especially in medial temporal regions (Chen et al., 2021). [11C]UCB-J PET has shown severe and disease severity dependent synaptic loss in the primary tauopathies of progressive supranuclear palsy and amyloid-negative corticobasal syndrome (Holland et al., 2020). In patients with mild cognitive impairment, the brain regions with higher uptake of tau protein tracer [18F]MK-6240 have shown lower uptake of [11C]UCB-J (Vanhaute et al., 2021).

[11C]UCB-J production is robust with high yield (Rokka et al., 2019) and it has favourable dosimetry (Nabulsi et al., 2016; Bini et al., 2020; Cawthorne et al., 2021). Binding is specific; only ∼20% of the volume of distribution (VT) of [11C]UCB-J is due to the nondisplaceable binding (VND); therefore VT can be used as outcome parameter (Rabiner, 2018), representing the SV2A concentration in the brain. The same-day test-retest variability is excellent in the brain imaging (Finnema et al., 2018). In 60-min PET scans of Alzheimer's disease patients and healthy controls, 28-day repeatability is adequate to reveal changes of ∼15% (Tuncel et al., 2021). Negative bias in 28-day retest scans, but not same-day retest scans, of healthy subjects has been observed, possibly caused by acute stress (Tuncel et al., 2021).

[18F]UCB-J is an analogue of [11C]UCB-J, developed by the same group (Li et al., 2019).

## Blood data

Plasma free fraction (fp), the fraction of tracer not bound to plasma proteins, was 0.32 (range 0.29-0.34) in the test-retest study of five human volunteers (Finnema et al., 2018). In twelve healthy subjects fp was 0.18-0.28, and appeared to decrease with age (Mansur et al., 2020). In rhesus monkeys fp is 0.43±0.05 for [11C]UCB-J and 0.42±0.06 for [18F]UCB-J (Li et al., 2019).

The fraction of non-metabolized radioligand drops to ∼0.6 already in 5 minutes (Tuncel et al., 2021), is ∼0.3 20 minutes after injection (Finnema et al., 2018; figure 1), and after that the fraction drops only slowly, being ∼0.2 at 55 min and up to 120 min (Tuncel et al., 2021; Finnema et al., 2018; Mansur et al., 2020). Individual variation in fractions seems to be high, probably preventing the usage of population average with [11C]UCB-J; with [18F]UCB-H individual variance was low, and population average was used to correct image-derived input function (Bastin et al., 2020). Finnema et al (2018) fitted an inverted integrated gamma function to the non-metabolized parent fraction data. The parent fractions in Göttingen minipigs (Thomsen et al., 2020) seem to be similar than in humans. In mice the metabolism is faster, being ∼0.2 already at 15 min p.i., and a sigmoid function was fitted to the ratio data (Bertoglio et al., 2020). Sigmoid function has been applied also to human data (Mansur et al., 2020). Radiometabolites do not seem to penetrate the blood-brain barrier (Finnema et al., 2016).

As an alternative to arterial blood sampling, extraction of image-derived input function is feasible even in mice studies (Bertoglio et al., 2020; Glorie et al., 2020), and has been applied to human studies with [18F]UCB-H (Bahri et al., 2017; Bastin et al., 2020). Conversion of concentration in blood to that in plasma is relatively simple, because plasma-to-blood ratio is stable at ∼1.3 (Mansur et al., 2020). The image-derived input method still requires blood samples for metabolite correction and scaling. In animal studies, a separate group of animals can be used for assessing the metabolism, and if group differences in metabolism are not observed the metabolite correction can be omitted (Bertoglio et al., 2020). Use of population-based input data was validated in Göttingen minipigs (Thomsen et al., 2020).

## Brain data analysis methods

The brain uptake of [11C]UCB-J is highest in the striatum and cortex, and moderate in the thalamus and cerebellum, and low in white matter (Finnema et al., 2016 and 2018). Distribution of specific [11C]UCB-J binding sites in human and non-human primate brain, based on autoradiographic mapping, has been reported by Varnäs et al (2020).

Peak radioactivity concentration in the plasma is about two times higher than the peak radioactivity in the brain; thus the vascular radioactivity can be neglected in compartmental model analysis, but it may have an impact in time delay correction (Finnema et al., 2018). Koole et al (2019) fixed VB to 5% in compartmental model fitting. In mice studies, VB has been fixed to 3.6% (Bertoglio et al., 2020). Tuncel et al (2021) fitted VB in their test-retest study.

Based on AIC, the two-tissue compartmental model (2TCM) fitted the regional data better than the one-tissue compartmental model (1TCM), but the difference in VT estimates was small; additionally, 2TCM did not provide reasonable parameter values for some data sets, and therefore 1TCM was selected for the final data analysis (Finnema et al., 2018; Mansur et al., 2020). Mean VT ranged from 5.3 ± 0.5 in the centrum semiovale to 22.4 ± 1.8 in the putamen (Finnema et al., 2018). Based on AIC and F-test Koole et al (2019) ended up using 1TCM, and also for the F-18 labelled UCB-J derivative [18F]SynVesT-1 ([18F]MNI-1126, [18F]SDM-8) the 1TCM performed better (Constantinescu et al., 2019). With [18F]UCB-H, 2TCM model is preferred over 1TCM, but a coupled fit with global K1/k2 must be used to avoid unstable fits (Goutal et al., 2021). In the occupancy studies, when the occupancy was highest, 2TCM performed better than 1TCM, probably because of the relatively increased contribution of nonspecific binding; yet, VT from 2TCM and 1TCM did not differ significantly, supporting the use of 1TCM in occupancy studies as well (Koole et al., 2019). Tuncel et al (2021) noted that 1TCM with VB as one of the model parameters is sufficient to fit 60-min [11C]UCB-J data.

In mice and rat studies, 1TCM and Logan plot have been applied (Bertoglio et al., 2020; Glorie et al., 2020; Thomsen et al., 2021). Logan plot has also been used with human [18F]UCB-H data (Bastin et al., 2020).

Parametric VT images can be calculated using 1TCM with basis functions method, with k2 limits set to 0.01-1.0 min-1 (Finnema et al., 2018). Highest VT, 34±4, was seen in parietal cortex (Chen et al., 2018).

Calculation of parametric VT image provides also a map of K1, representing perfusion and passage through blood brain barrier. Visual stimulation increases brain perfusion, which is seen as increased 1TCM K1, with no effect on VT and BPND (Smart et al., 2021). The pattern of K1 and R1 (K1 divided by cerebellar K1) reduction in AD patients is similar to the pattern of hypometabolism in AD seen with [18F]FDG (Chen et al., 2018 and 2021). A single [11C]UCB-J PET study can thus provide information on both neuronal activity and synaptic density (Chen et al., 2021).

Finnema et al (2018) assessed the time stability of the 1TCM , and noticed that the study length could be reduced from 120 to 60 min with no impairment in ICC or test-retest variability. In their study, shorter scan duration will lead to somewhat lower VT estimates (maximally about -5% when comparing 60 min and 120 min scan lengths), but there is also possibility that the longer scan duration may lead to overestimated VT due to uptake of radioactive metabolite(s) in the brain, especially in the low uptake regions. Shortening the scan duration further from 60 min to 45 and 30 min led to overestimation of VT and poor reliability (Tuncel et al., 2021). Also Mansur et al (2020) recommended the 60 min scan length. For mice studies, Bertoglio et al (2020) concluded that 60 min scan length is sufficient.

Correction of VT for plasma protein binding (VT/fp) worsened ICC and test-retest variability, and thus Finnema et al. (2018) recommended that VT/fp would be used as outcome parameter only in cross-sectional studies, if group differences or treatment effects in plasma protein binding could be expected.

In rat brain study with [18F]UCB-H, SUV correlated well with VT from Logan plot using population-based input function; static 20-40 min scan was recommended (Serrano et al., 2019). [11C]UCB-J brain-to-blood ratio (0-60 min) correlated strongly with VT (calculated using whole blood curve as input function) in mice models of PD and AD (Xiong et al., 2021).

Occupancy and displacement studies with [11C]UCB-J can be conducted using bolus plus constant infusion protocol (Finnema et al., 2019).

### Reference region

Ubiquitous distribution of SV2A means also that there is no true reference region in the brain that could be used as input function or to determine the SV2A binding potential. Therefore arterial input function must be measured, and VT is the only possible quantitative parameter; because of the relatively low nonspecific binding (VNS) of [11C]UCB-J in the brain VT can be assumed to well represent the SV2A density. Lassen plot can be used to assess receptor occupancy from VT in the absence of reference region (Koole et al., 2019; Finnema et al., 2019; Bertoglio et al., 2020; Rossano et al., 2020).

Yet, for comparative studies where strict quantification is not necessary, centrum semiovale (white matter underneath the cerebral cortex), that has low uptake of SV2A ligands, can be used as reference region in [11C]UCB-J data analysis, applying either SRTM, SRTM2, or one-tissue compartmental model based DVR (Finnema et al., 2016; Toyonaga et al., 2018; Rossano et al., 2020; Mertens et al., 2020; Mecca et al., 2020), or 60-90 min tissue-to-reference ratio (SUVR) (Naganawa et al., 2018; Koole et al., 2019; Delva et al., 2020; Naganawa et al., 2021; Vanhaute et al., 2021). In SRTM2, k'2 has been fixed to 0.027 min-1 (Smart et al., 2021). In minipigs the centrum semiovale appeared to contain specific binding, but it might have been artefact caused by spillover from cortex (Thomsen et al., 2020). Image reconstruction with increased OSEM iterations and careful ROI placement on centrum semiovale reduces the spill-in from grey matter but marked bias still remains (Rossano et al., 2020). Although VT in centrum semiovale was decreased in an occupancy study, the occupancy results were not statistically different between VT-based Lassen plot analysis and binding potential-based analysis using centrum semiovale as reference region (Koole et al., 2019). When centrum semiovale was used as reference region for UCB-J derivative [18F]SynVesT-1 ([18F]MNI-1126, [18F]SDM-8) the bias in occupancy studies was minimal (Constantinescu et al., 2019). VT in the white matter (centrum semiovale) was virtually identical between AD patients and age matched control subjects (Chen et al., 2018). Highest BPND, 6.09±0.33 in control subjects, was seen in parietal cortex (Chen et al., 2018).

Cerebellum has considerable SV2A-specific binding, and therefore cannot be used for computing true BPND. However, VT (based on 1TCM) difference is minimal (∼1%) between cerebellum of healthy and AD subjects (Mecca et al., 2020). Compared to cerebellum, centrum semiovale ROI is smaller and activity level lower, leading to greater variability. Thus, in practice, cerebellum may be superior to centrum semiovale as reference region in AD studies (Mecca et al., 2020).

In a mice study time range 30-60 min was found to be sufficient for SUVR (tissue-to-reference ratio) calculation when brain stem was used as the reference region (Toyonaga et al., 2019). In human study, using centrum semiovale as the reference region, SUVR images from time ranges 40-70 min or 50-80 min were found to provide good results as compared to reference input methods (Mertens et al., 2020). SUVR from 40-60 min correlated well with 1TCM-derived DVR (Tuncel et al., 2021). The best correlation between SUVR and BPND was obtained from 60-90 min scan, with only ‑1±7% difference between SUVR‑1 and BPND in the whole brain of healthy subjects, and ‑1±8% difference in neuropsychiatric subjects (Naganawa et al., 2021). SUVR is not affected by sex or healthy ageing, except for small age-related SUVR decrease in caudate nucleus (Michiels et al., 2021).

SRTM may require at least 70 min dynamic PET scan (Koole et al., 2019), but parametric BP images produced with SRTM2 using only the first 60 min of the dynamic data were not much different from the maps computed using 90 min of the data (Mertens et al., 2020). From 60-min scans, SRTM-derived BPND was underestimated by ∼25% as compared to DVR-1 calculated from 1TCM results, with adequate repeatability in AD patients and healthy control subjects (Tuncel et al., 2021). SRTM2 from 60 min data using centrum semiovale as reference has been used to analyse data in a study comparing PD patients and control subjects (Matuskey et al., 2020). K1 is much lower in white matter than in grey matter, leading to high R1 estimate from SRTM methods.

Logan plot with centrum semiovale as reference input performed clearly worse than SRTM2 and MRTM2 in calculation of parametric images (Mertens et al., 2020).

## References:

Chen M, Mecca AP, Naganawa M, Finnema SJ, Toyonaga T, Lin S, Najafzadeh S, Ropchan J, Lu Y, McDonald JW, Michalak HR, Nabulsi NB, Arnsten AFT, Huang Y, Carson RE, van Dyck CH. Assessing synaptic density in Alzheimer disease with synaptic vesicle glycoprotein 2A positron emission tomographic imaging. JAMA Neurol. 2018; 75(10): 1215-1224. doi: 10.1001/jamaneurol.2018.1836.

Finnema SJ, Nabulsi NB, Eid T, Detyniecki K, Lin S, Chen M-K, Dhaher R, Matuskey D, Baum E, Holden D, Spencer DD, Mercier J, Hannestad J, Huang Y, Carson RE. Imaging synaptic density in the living human brain. Sci Transl Med. 2016; 8: 348ra96. doi: 10.1126/scitranslmed.aaf6667.

Finnema SJ, Nabulsi NB, Mercier J, Lin SF, Chen MK, Matuskey D, Gallezot JD, Henry S, Hannestad J, Huang Y, Carson RE. Kinetic evaluation and test-retest reproducibility of [11C]UCB-J, a novel radioligand for positron emission tomography imaging of synaptic vesicle glycoprotein 2A in humans. J Cereb Blood Flow Metab. 2018; 38(11): 2041-2052. doi: 10.1177/0271678X17724947.

Koole M, van Aalst J, Devrome M, Mertens N, Serdons K, Lacroix B, Mercier J, Sciberras D, Maguire P, Van Laere K. Quantifying SV2A density and drug occupancy in the human brain using [11C]UCB-J PET imaging and subcortical white matter as reference tissue. Eur J Nucl Med Mol Imaging 2019; 46: 396-406. doi: 10.1007/s00259-018-4119-8.

Löscher W, Gillard M, Sands ZA, Kaminski RM, Klitgaard H. Synaptic vesicle glycoprotein 2A ligands in the treatment of epilepsy and beyond. CNS Drugs 2016; 30(11): 1055-1077. doi: 10.1007/s40263-016-0384-x.

Nabulsi NB, Mercier J, Holden D, Carré S, Najafzadeh S, Vandergeten MC, Lin SF, Deo A, Price N, Wood M, Lara-Jaime T, Montel F, Laruelle M, Carson RE, Hannestad J, Huang Y. Synthesis and preclinical evaluation of 11C-UCB-J as a PET tracer for imaging the synaptic vesicle glycoprotein 2A in the brain. J Nucl Med. 2016; 57(5): 777-784. doi: 10.2967/jnumed.115.168179.

Rabiner EA. Imaging synaptic density: a different look at neurological diseases. J Nucl Med. 2018; 59(3): 380-381. doi: 10.2967/jnumed.117.198317.

Tags: ,

Updated at: 2022-01-13
Created at: 2017-11-29
Written by: Vesa Oikonen