Molecular pathology and synaptic loss in primary tauopathies: UCB-J PET study

The relationship between in vivo synaptic density and tau burden in primary tauopathies is key to understanding the impact of tauopathy on functional decline and in informing new early therapeutic strategies. In this cross-sectional observational study, we determine the in vivo relationship between synaptic density and molecular pathology, in the primary tauopathies of Progressive Supranuclear Palsy (PSP) and Corticobasal Degeneration (CBD), as a function of disease severity. Twenty three patients with PSP, and twelve patients with Corticobasal Syndrome (CBS) were recruited from a tertiary referral centre. Nineteen education, sex and gender-matched control participants were recruited from the National Institute for Health Research Join Dementia Research platform. Cerebral synaptic density and molecular pathology, in all participants, were estimated using PET imaging with the radioligands [11C]UCB-J and [18F]AV-1451, respectively. Patients with CBS also underwent amyloid PET imaging with [11C]PiB to exclude those with likely Alzheimer's pathology - we refer to the amyloid negative cohort as having CBD although acknowledge other pathologies exist. Disease severity was assessed with the PSP rating scale; regional non-displaceable binding potentials (BPND) of [11C]UCB-J and [18F]AV-1451 were estimated in regions of interest from the Hammersmith Atlas, excluding those with known off-target binding for [18F]AV-1451. As an exploratory analysis, we also investigated the relationship between molecular pathology in cortical brain regions, and synaptic density in connected subcortical areas. Across brain regions, there was a positive correlation between [11C]UCB-J and [18F]AV-1451 BPND ({beta}=0.4, t=4.7, p<0.0001). However, the direction of this correlation became less positive as a function of disease severity in patients ({beta} = -0.03, T = -4.0, p = 0.002). Between brain regions, cortical [18F]AV-1451 binding was negatively correlated with synaptic density in subcortical areas (caudate nucleus, putamen, and substantia nigra). Brain regions with higher synaptic density are associated with a higher [18F]AV-1451 binding in PSP/CBD, but this association diminishes with disease severity. Moreover, higher cortical [18F]AV-1451 binding correlates with lower subcortical synaptic density. Longitudinal imaging is required to confirm the mediation of synaptic loss by molecular pathology. However, the effect of disease severity suggests a biphasic relationship between synaptic density and tauopathy, with synapse rich regions vulnerable to accrual of pathology, followed by a loss of synapses in response to pathology. Given the importance of synaptic function for cognition, our study elucidates the pathophysiology of primary tauopathies and may inform the design of future clinical trials.


Introduction
Synaptic loss is a feature of many neurodegenerative disorders [1][2][3] . It is closely related to cognitive decline in symptomatic stages of disease 4,5 , but can begin long before symptom onset and neuronal loss 6 . Synaptic loss and dysfunction may be an important mediator of decline even where atrophy is minimal or absent 7,8 . Conversely, synaptic connectivity may facilitate the spread of oligomeric mis-folded proteins such as tau 9,10 . The relationship between synaptic loss and the accumulation of mis-folded proteins in primary tauopathies has yet to be determined in vivo. Preclinical models suggest early synaptotoxicity of oligomeric tau, leading to reduced plasticity and density 11,12 . In patients with mutations of microtubuleassociated protein tau (MAPT), there are deficiencies in many synaptic pathways including GABA-mediated signalling and synaptic plasticity 13 . The mechanisms of synapse loss following tau pathology include both direct and indirect pathways (reviewed in Spires-Jones et al. 2014 14 ). In the related tauopathy of Alzheimer's disease, there is differential expression of synaptic proteins in the early stages 15,16 ; this may be an attempt to maintain cellular physiology, which fails as the disease progresses, leading to loss of synaptic function and synapse numbers.
In clinical disorders, the in vivo pathologies of synaptic density and tau burden can be characterised by positron emission tomography (PET). In Alzheimer's disease for example, increased temporal lobe binding of the tau radioligand [ 18 F]MK-6240 is associated with decreased synaptic density measured by the radioligand [ 11 C]UCB-J 17 . However, the pathology of Alzheimer's disease is multifaceted with amyloid and tau aggregation, vascular changes and neuroinflammation 18 .
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 5, 2021. ; In this study, we use Progressive Supranuclear Palsy (PSP) and Corticobasal Degeneration (CBD) as models of human tauopathy, with relevance to other tau-mediated neurodegenerative disorders, and examine the relationship between synaptic density and tau burden 19 . An advantage of studying PSP is the very high correlation between the clinical syndrome, and the specific 4R-tauopathy at autopsy 20 . The clinical phenotype of Corticobasal Syndrome (CBS), may be caused by CBD, but can also be mimicked by Alzheimer's disease, and less commonly by other forms of frontotemporal lobar degeneration 20 . Here, we use the term CBD to refer to patients with CBS in whom Alzheimer's disease is excluded by [ 11 C]PiB PET, whereby in the absence of amyloid pathology there is a high clinicopathological correlation with 4R-tauopathy at post mortem. Both PSP and CBD demonstrate synaptic loss in vivo 7,8 and at post-mortem 1,2 . The distribution of tau pathology in both diseases is well characterised with cortical and subcortical involvement 21,22 . Animal models of tauopathy have illustrated the colocalisation of misfolded tau protein and synaptic loss at the synaptic bouton 14,23 but the tau-synapse association is yet to be determined in vivo. Figure 1 illustrates our hypotheses. Previous studies suggest that the strength of connectivity within a region and between brain regions can promote the spread of tau pathology, in humans as in preclinical models 10,24 . Therefore, we hypothesised that brain areas with higher synaptic density would develop more tau pathology. We predicted that the spatial distribution of pathology, as measured with the PET radioligand [ 18 F]AV-1451, would be correlated with synaptic density, as measured with the PET radioligand [ 11 C]UCB-J (which binds to the presynaptic vesicle glycoprotein SV2A that is ubiquitously expressed in all brain synapses 25,26 ). Since tauopathy in a region may impair efferent projections, a corollary hypothesis is that tau accumulation in one region (source region) leads to diaschisis characterised by reduced synaptic density in the areas to which it connects (target regions).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Alzheimer's disease, and the off-target binding of this ligand within the basal ganglia. We therefore refer to its binding target as 'molecular pathology', covering tau and non-tau targets.
A second part of the model describes the consequence of the pathology, which is to reduce synaptic density. The predicted result is a positive relationship between [ 18 F]AV-1451 binding and synaptic loss, negatively moderated by disease progression (Figure 1b).

Figure 1. Schematic diagram illustrating the predicted toxic effect of tau on synaptic density
At a regional level (A) synaptic density promotes the spread of tau from one region to another. Tau burden therefore depends on a region's baseline synaptic density, for example in B, region 3 would accumulate more tau given its high baseline synaptic density. However, this relationship between tau and synapses is further affected by stage of disease (B) such that in any given region, tau-induced synaptic loss ensues as the stage of disease progresses from mild to moderate to severe.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted March 5, 2021. ; https://doi.org/10.1101/2021.03.03.21252808 doi: medRxiv preprint

Participant recruitment and study design
Twenty three people with probable PSP-Richardson Syndrome 19 , and twelve people with probable CBS in whom Alzheimer's disease was excluded with [ 11 C]PiB PET 7 , were recruited from a regional specialist National Health Service clinic at the Cambridge University Centre for Parkinson-plus. We refer to our amyloid-negative CBS cohort as having CBD but acknowledge other pathologies are possible 20

PET data acquisition and kinetic analysis [ 11 C]UCB-J PET
The procedure for [ 11 C]UCB-J synthesis, PET data acquisition, image reconstruction and kinetic analysis was the same as in Holland et al. 2020 7 . In brief, dynamic PET data acquisition was performed on a GE SIGNA PET/MR (GE Healthcare, Waukesha, USA) for 90 minutes immediately after injection, with attenuation correction using a multi-subject atlas . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted March 5, 2021. ; https://doi.org/10.1101/2021.03.03.21252808 doi: medRxiv preprint method 27,28 and improvements to the MRI brain coil component 29 . Emission image series were aligned using SPM12 (www.fil.ion.ucl.ac.uk/spm/software/spm12/), and rigidly registered to the T1-weighted MRI acquired during PET data acquisition (TR = 3.6 msec, TE = 9.2 msec, 192 sagittal slices, in plane resolution 0.55 x 0.55 mm, interpolated to 1.0 x 1.0 mm; slice thickness 1.0 mm). The Hammersmith atlas (http://brain-development.org) with modified posterior fossa regions was spatially normalized to the T1-weighted MRI of each participant using Advanced Normalisation Tools (ANTs) software 30 . Regional time-activity curves were extracted following the application of geometric transfer matrix (GTM) partial volume correction (PVC 31 ) to each dynamic PET image. Regions of interest (ROIs) were multiplied by a binary grey matter mask (>50% on the SPM12 grey matter probability map smoothed to PET spatial resolution), with the exception of the subcortical grey matter regions pallidum, substantia nigra, pons and medulla. To assess the impact of PVC, time-activity curves were also extracted from the same ROIs without the application of GTM PVC (discussed below as "without partial volume correction").
To quantify SV2A density, [ 11 C]UCB-J non-displaceable binding potential (BP ND ) was determined using a basis function implementation of the simplified reference tissue model 32 , with the reference tissue defined in the centrum semiovale 33,34 . . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 5, 2021.

Statistical analyses
We compared demographic and clinical variables between the diagnostic groups using ANCOVA, and chi-square tests where appropriate. We used a linear mixed effects model to . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 5, 2021. Analyses were performed with and without partial volume correction, yielding similar results; we focus on partial volume corrected BP ND to limit the potential effect of atrophy on our ligand cross-correlation but present data without partial volume correction in the supplementary material (Supplementary Figure 1 and 2). Statistical analyses were implemented in R (version 3.6.2).

Data Availability Statement
The data that support the findings of this study are available from the corresponding author, upon reasonable request for academic (non-commercial) purposes, subject to restrictions required to preserve participant confidentiality.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Demographics
The patients (PSP and CBD) and control groups were similar in age, sex, education and injected activity of [ 11 C]UCB-J and [ 18 F]AV-1451 (Table 1). We observed typical cognitive profiles for people with PSP and CBD: impaired on verbal fluency, memory and visuospatial domains of the ACE-R and MMSE. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Practically identical findings were observed using BP ND derived from data without partial volume correction (Supplementary Figure 1). Of note, the significance of the overall model above did not change with the addition of scanning interval as a covariate of no interest.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 5, 2021. ; is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Cross-regional correlation between [ 18 F]AV-1451 BP ND and [ 11 C]UCB-J BP ND
Synaptic density in a region is proposed to be affected by both local tau pathology and tau burden in connected regions from which it receives afferent projections. As a result, despite a positive correlation at a regional level, the synaptic density in any given region may be negatively affected by remote insult, with diaschisis between anatomically connected regions (illustrated schematically in Figure 1A). As an exploratory analysis, we computed the asymmetric Pearson's correlation matrix shown in Figure 3  illustrating various direct tau-induced synaptic abnormalities see 40 ). Tau also directly affects the axon cytoskeleton and trafficking, as well as the soma 41 . Indirectly, hyperphosphorylated tau adversely affects the functioning of the neuronal support network, including glia cells and astrocytes 42,43 . These events are affected by the stage and severity of the disease process, and in relation to regional differences in connectivity which we discuss next (concepts schematically illustrated in Figure 1).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 5, 2021. ; https://doi.org/10.1101/2021.03.03.21252808 doi: medRxiv preprint We identified a positive relationship between the binding of [ 11 C]UCB-J and [ 18 F]AV-1451 such that areas of the brain with higher synaptic density develop higher pathology. This accords with preclinical and clinical models of tauopathy in which the strength of local network connectivity facilitates the transneuronal spread of tau pathology 10,24,44,45 . However, the relationship between tau accumulation and synaptic density changes with disease progression, at least as inferred from the cross-sectional moderation by disease severity ( Figure 2B). With increasing scores on the PSP rating scale, synaptic density becomes less dependent on local tau accumulation. In other words, in areas with relatively low tau accumulation synaptic density is minimally affected, whereas in areas with higher tau accumulation there is reduction of synaptic density as the disease progresses and this preferentially occurs in synapse rich areas. As the disease progresses, other pathological processes may contribute to synaptic loss, such as inflammation, another predictor of prognosis and mediator of synaptic loss 46  To understand the biphasic relationship between tau accumulation and synaptic density, one must consider other key players in synaptotoxicity in tauopathies, such as neuroinflammation 47 . Recent in vivo studies have confirmed the regional co-localisation of inflammation and [ 18 F]AV-1451 binding in PSP 48 , in line with previous in vivo 49,50 , and post mortem 51 reports of the tight interplay between neuroinflammation and tau accumulation in tauopathies. There . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 5, 2021. ; https://doi.org/10.1101/2021.03.03.21252808 doi: medRxiv preprint is growing evidence that these two pathological processes affect synaptic function both independently and synergistically 40,42 .
The relationship between tauopathy and synaptic density is even more intriguing when considering the change in synaptic density in one region as a function of pathology in another. There are strong correlations between [ 11 C]UCB-J binding within the basal ganglia and brain stem and [ 18  accumulating in synapse-rich areas of the brain, for example the neocortex. A second explanation is that loss of descending cortico-striatal axons due to cortical pathology, may cause diaschisis, affecting subcortical synaptic density even further. Previous analysis of diffusion tensor imaging in patients with PSP/CBD have revealed extensive white matter abnormalities (within the main association fibres) beyond the degree of cortical atrophy 52,53 resulting in loss of cortical afferents onto subcortical structures. A third, though not mutually exclusive, explanation is the weakening of cortical-subcortical functional connectivity resulting from dysfunctional synapses rather than synaptic loss 24 .
Although at a regional level there is a positive correlation between [ 11 C]UCB-J and [ 18 F]AV-1451 BP ND , we are not directly measuring either synaptic function or the synaptotoxic tau oligomers. This caveat must be borne in mind when interpreting PET data. It is the preclinical . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted March 5, 2021. ; models that have shown that oligomers of tau are toxic to synaptic function, even in the absence of tau polymers/fibrils 11,12 . By the time tau aggregates are established, oligomers of tau are expected cortically, and perhaps interfering with cortical function and the integrity of descending axons.
There are other limitations to our study. First, the low affinity of [ 18 F]AV-1451 for PSP and CBD 4R tau. Even where this radioligand recapitulates the distribution of post-mortem neuropathology in PSP and CBD, and binds PSP 4R tau, the affinity is very much lower than for 3R tau in Alzheimer's disease 21,39 . Second, there is well-established off-target binding of [ 18 F]AV-1451, particularly within subcortical structures where monoamine oxidase is present. Off-target binding is most prominent in the basal ganglia which we excluded before running our statistical analyses. We included these regions in the detailed descriptive correlation matrices in Supplementary Figure 3 for completeness sake, noting the strong negative correlations between cortical [ 18 F]AV-1451 BP ND and subcortical [ 11 C]UCB-J BP ND . Third, we note that in PET studies of neurodegeneration with atrophy, grey matter volume loss can affect the interpretation of PET signals. However, synaptic loss in PSP and CBD occurs even in areas of the brain without discernible atrophy on MRI 7,8 . Nonetheless, we used a stringent partial volume correction method (GTM) to minimise the effect of atrophy on our ligand cross-correlations. Of note, our data without partial volume correction yield similar results in all the main analyses ( Supplementary Figure 1 and 2). Lastly, the cross-sectional design of this study limits the interpretation of the dynamic relationship between tau accumulation and synaptic loss. Although we include patients at various stages of their illness, a longitudinal design is necessary to test the dynamic relationship we propose, and the mediation of synaptic loss by progressive tauopathy.
In conclusion, we demonstrate a widespread positive association between [ 18 F]AV-1451 and [ 11 C]UCB-J binding in patients with symptomatic PSP and amyloid-negative corticobasal . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted March 5, 2021. ; https://doi.org/10.1101/2021.03.03.21252808 doi: medRxiv preprint syndromes. Individual variability in this association correlates with disease severity. The complex relationship between tau accumulation and synaptic density in vivo may explain changes in cognitive and motor physiology. We hope that these insights will inform the design of new clinical trials to arrest PSP and CBD.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted March 5, 2021. ; https://doi.org/10.1101/2021.03.03.21252808 doi: medRxiv preprint