Synaptic density affects clinical severity via network dysfunction in syndromes associated with frontotemporal lobar degeneration

There is extensive synaptic loss from frontotemporal lobar degeneration, in preclinical models and human in vivo and post mortem studies. Understanding the consequences of synaptic loss for network function is important to support translational models and guide future therapeutic strategies. To examine this relationship, we recruited 55 participants with syndromes associated with frontotemporal lobar degeneration and 24 healthy controls. We measured synaptic density with positron emission tomography using the radioligand [11C]UCB-J, which binds to the presynaptic vesicle glycoprotein SV2A, neurite dispersion with diffusion magnetic resonance imaging, and network function with task-free magnetic resonance imaging functional connectivity. Synaptic density and neurite dispersion in patients was associated with reduced connectivity beyond atrophy. Functional connectivity moderated the relationship between synaptic density and clinical severity. Our findings confirm the importance of synaptic loss in frontotemporal lobar degeneration syndromes, and the resulting effect on behaviour as a function of abnormal connectivity.

There is extensive synaptic loss from frontotemporal lobar degeneration, in preclinical models and human in vivo and post mortem studies.Understanding the consequences of synaptic loss for network function is important to support translational models and guide future therapeutic strategies.To examine this relationship, we recruited 55 participants with syndromes associated with frontotemporal lobar degeneration and 24 healthy controls.We measured synaptic density with positron emission tomography using the radioligand [ 11 C]UCB-J, which binds to the presynaptic vesicle glycoprotein SV2A, neurite dispersion with diffusion magnetic resonance imaging, and network function with task-free magnetic resonance imaging functional connectivity.Synaptic density and neurite dispersion in patients was associated with reduced connectivity beyond atrophy.Functional connectivity moderated the relationship between synaptic density and clinical severity.Our findings confirm the importance of synaptic loss in frontotemporal lobar degeneration syndromes, and the resulting effect on behaviour as a function of abnormal connectivity.
Frontotemporal lobar degeneration (FTLD) pathologies cause heterogenous syndromes with partially overlapping clinical features and highly variable correlations between neuropathology and phenotypic expression [1][2][3][4] .These conditions are associated with early loss of functional independence, considerable care burden and reduced life expectancy 5,6 .There is a pressing need for new therapeutic interventions, based on better characterisation and in vivo analysis of the pathogenic cascade leading from the molecular root causes to clinical presentation and progression 7,8 .This cascade includes severe synaptic loss 9,10 , which integrates the toxic effects of protein aggregation and inflammation [11][12][13][14] .
We propose that the severe synaptic loss arising from frontotemporal lobar degeneration would impair local and long-range functional connectivity, and thereby cognitive and behavioural change.We test this hypothesis with three clinical syndromes associated with different types of frontotemporal lobar degeneration; progressive supranuclear palsy (PSP), corticobasal syndrome (CBS) and behavioural variant frontotemporal dementia (bvFTD).These phenotypic entities show molecular heterogeneity, with 3-repeat and 4-repeat tauopathies and TDP-43 pathology.By studying these distinct syndromes from the FTLD-spectrum, we capture variations in anatomical distribution of pathology to better assess its consequence for connectivity and clinical severity.
The radioligand [ 11 C]UCB-J quantifies synaptic density through selective binding to the presynaptic vesicle glycoprotein 2 A (SV2A).When used in positron emission tomography (PET) imaging, it confirms the reduction of synaptic density in bvFTD 10 , CBS and PSP 9,15 .Reductions in [ 11 C]UCB-J non-displaceable binding potential (BP ND ), a metric of binding site density, correlate with clinical severity and are observed in the distributions also found in post mortem studies 16,17 .
[ 11 C]UCB-J BP ND is primarily a measure of synaptic density rather than synaptic function 18 and is directly related to changes in cortical neurophysiological generators in FTLD 19 .Disruption to functional connectivity and network integration in FTLD-disorders aligns closely with symptom onset and progression 20,21 , although uncertainty exists as to the cellular and molecular processes that determine functional connectivity as measured from resting-state functional MRI (fMRI) 22,23 .
In this work we show that synaptic density is related to large-scale brain connectivity and network organisation.We undertook a multimodal neuroimaging study to combine [ 11 C]UCB-J PET with resting state functional MRI in participants with three FTLD-associated disorders and similarly-aged healthy controls.We also used neurite orientation dispersion and density imaging (NODDI, a diffusion magnetic resonance imaging method 24 , to provide complementary evidence for the role of synaptic health on connectivity.We quantified local functional connectivity using the graph metric of weighted degree.We show that local synaptic loss is associated with reduced connectivity, and that reductions in synaptic density explain connectivity loss that is not accounted for by atrophy.To examine syndrome-specific differences in the effect of synaptic loss, we reduced data dimensionality by independent component analysis.We found that reduced synaptic density was associated with connectivity loss in such a way as moderates and explains additional variance in the individual differences in cognition.We further explored the effect of synaptic loss on connectivity, distinguishing the effect of synaptic loss in a region versus the region to which it is connected.

Clinical characteristics of participants
Demographic details and clinical characteristics of participants are set out in Table 1 and Supplementary Table 1.No significant group differences were observed for age or sex.Clinical and neuropsychological assessments showed impairment in all patient groups, with expected higher average scores on the Cambridge Behavioural Inventory-Revised (CBI-R) in bvFTD and greater impairment on the PSP-rating-scale (PSPRS) in PSP than bvFTD.Participants with bvFTD had increased in-scanner motion during fMRI acquisition, and therefore we included a metric of motion as a covariate of no interest throughout.
Group differences in synaptic density, neurite dispersion, connectivity, and grey matter There were widespread significant reductions in [ 11 C]UCB-J BP ND , the orientation dispersion index (ODI) determined with NODDI, grey matter volume, and weighted degree in patients compared to control participants in cortical and subcortical regions.The largest effect sizes were seen for [ 11 C]UCB-J BP ND (Supplementary Fig. 1).The distribution of group differences were similar for [ 11 C]UCB-J BP ND and ODI, with large between-group changes in these modalities spanning the frontal lobe, cingulate, insula, and basal ganglia.For grey matter volumes, there were prominent decreases in the frontal lobe and basal ganglia.The largest reductions in weighted degree in patients compared to controls were observed for peri-rolandic regions and the cerebellar dentate.The extensive differences between controls and patients in [ 11 C]UCB-J BP ND were observed even after local volume was included as a covariate, in both regional and voxelwise analyses (Supplementary Fig. 2).
For weighted degree and [ 11 C]UCB-J BP ND there were no regional differences between-patient groups, after correcting for multiple comparisons.Uncorrected reductions for [ 11 C]UCB-J BP ND were greater in bvFTD than CBS and PSP, in the frontal lobe, insula, cingulate, and anterior temporal lobe; and for PSP when compared to CBS in the pallidum (Supplementary Table 2).Uncorrected reductions in weighted degree in bvFTD versus CBS (Supplementary Table 3) were observed in the anterior temporal lobe and substantia nigra.Significantly lower ODI values for bvFTD than PSP and CBS were found for widely distributed regions across the brain (Supplementary Table 4), with the largest effect sizes in the frontal and temporal lobe and insula.There were smaller ODI values for PSP than CBS for subcortical structures.Reduced grey matter volumes in bvFTD compared to CBS and PSP (Supplementary Table 5) were found in the frontal and temporal lobes, the cingulate, insula, and caudate, with reductions for CBS compared to PSP in the precentral gyri.We found reduced total volumes for PSP compared to CBS in the midbrain and cerebellum dentate.Weighted degree is related to synaptic density in patients but not control participants We calculated regional [ 11 C]UCB-J BP ND and weighted degree z scores in patients standardised to control data.For cortical regions, mean weighted degree z scores correlated with mean [ 11 C]UCB-J BP ND z scores (Pearson's r = 0.39 P = 0.048, Fig. 1a).Parcels in the subgenual frontal cortex and subcallosal area lay off the main slope.The relationship was present even without these outliers, with a significant correlation in non-frontal cortical regions (Pearson's r=0.43P = 0.016).There was no such relationship in subcortical regions (Pearson's r = −0.1 P = 0.78).
We fitted linear mixed-effects models to account for individual variation in the regional relationship between [ 11 C]UCB-J BP ND and weighted degree.Inclusion of an effect of [ 11 C]UCB-J BP ND slope within each region improved model fit (χ 2 = 42, P = 8 × 10 −11 ).Weighted degree was associated with [ 11 C]UCB-J BP ND in patients (Standardised Beta 0.19, P = 6 × 10 −11 ) but not in control participants (Standardised Beta −0.01 P = 0.90).The group-by-[ 11 C]UCB-J BP ND interaction in a refitted model with all participants was significant ([ 11 C]UCB-J BP ND *Group Standardised Beta 0.067, P = 0.0008), with unchanged effect size and significance with scanning interval included as a covariate of no interest.In patients, assessing the [ 11 C]UCB-J BP ND effect within the region showed higher standardised betas for the weighted degree-[ 11 C] UCB-J BP ND relationships in temporal, parietal, cingulate, and superior frontal regions (Fig. 1b).The relationship between [ 11 C]UCB-J BP ND and weighted degree remained significant (Standardised Beta 0.18, P = 1 × 10 −9 , Supplementary Table 6) with the inclusion of regional grey matter volumes and total intracranial volume, and was also observed in each FTLD syndrome individually (bvFTD Standardised Beta 0.2 P = 1 × 10 −5 ; CBS Standardised Beta 0.12 P = 0.007; PSP Standardised Beta 0.23, P = 4 × 10 −9 ).Using the non-reparcellated Hammersmith atlas and the Brainnetome parcellation we similarly found an association between synaptic density and weighted degree, and that this relationship was stronger in patients than control participants (Supplementary Results).
To further explore the effect of synaptic health on connectivity, we considered the complementary metric ODI.In linear mixed-effects models, regional [ 11 C]UCB-J BP ND values were related to regional ODI in patients (Standardised Beta 0.54 P < 2 × 10 −16 ) and controls (Standardised Beta 0.39 P < 2 × 10 −16 ).ODI values were significantly associated with a regional weighted degree in patients (Standardised Beta 0.056 P = 0.007) but not in controls (Standardised Beta 0.028 P = 0.39).The group-by ODI interaction was not significant (ODI*Group Standardised Beta 0.014, P = 0.51).The relationship between ODI and weighted degree persisted with grey matter volumes and total intracranial volume as covariates (Standardised Beta 0.043 P = 0.037).For each cortical region, we tested the relationship between weighted degree and [ 11 C]UCB-J BP ND , ODI, and grey matter volume across participants.The measures of synaptic health, [ 11 C]UCB-J BP ND and ODI, were significantly associated with weighted degree in more regions (Fig. 1c, Supplementary Fig. 3) than was grey matter volume.Combining modalities with non-parametric combination testing, including adjustment for the number of contrasts, resulted in the most widespread significant relationships with weighted degree.We further tested if regional [ 11 C]UCB-J BP ND was related to weighted degree with grey matter atrophy and total intracranial volume included as covariates in the general linear model.We found post-correction significant relationships in the right orbitofrontal gyrus and bilateral superior temporal gyri (Supplementary Fig. 4).

An independent component analysis shows regional variation in synaptic density by clinical syndrome
To investigate how variation in synaptic density influences functional connectivity and cognition, we first performed an independent component analysis on concatenated participant [ 11 C]UCB-J BP ND partial volume corrected maps.One component was discarded as it was not robust to removal of an outlier (Grubb's test G 5.0 P = 3 × 10 −6 ).Two further components were not taken forward for further analysis as they incorporated regions particularly prone to artefact.Six of the remaining components differed between groups (Fig. 2a, b We repeated source-based synaptometry following regression of grey matter tissue probability maps from the partial volume corrected [ 11 C]UCB-J BP ND maps.We found participant loadings were strongly correlated for the matched components, with a highly similar distribution of between-group differences as noted in our primary analysis (Supplementary Results, Supplementary Fig. 6).
The functional connectivity of a region is associated with its synaptic density, both at the site of synaptic loss and remotely from it Dual regression identified participant-specific patterns of spatial covariance to each [ 11 C]UCB-J BP ND component map.Participant component-specific functional connectivity maps were taken forward to assess for voxel-wise associations with [ 11 C]UCB-J BP ND component loadings in a general linear model with permutation testing.There were significant associations between increased [ 11 C] UCB-J BP ND loadings and increased functional connectivity for five of the components (Fig. 2b, P < 0.05 with family-wise error correction across voxels).We found that connectivity differences were observed both at the site of synaptic loss and remotely from it.For instance, greater [ 11 C]UCB-J BP ND loadings in a component (component 1) with high values in the bilateral striatum were associated with increased functional connectivity across the cortex.
We extracted participant component scores by taking the mean beta, normalised to the residual within-subject noise, from participants' component-specific connectivity maps within a mask defined as significant areas of control functional connectivity with each component.Five of the six components showed associations between [ 11

Functional connectivity adds to and moderates the explanatory effect of UCB-J BP ND on clinical severity
We tested whether the inclusion of connectivity component scores to  3a and Supplementary Table 7), suggesting that greater cognitive impairment is associated with cortical synaptic loss, moderated by frontal and insular connectivity.For both components 4 and 5, we found that the relationship between [ 11 C]UCB-J BP ND loadings and ACE-R was stronger in those with higher connectivity scores.The winning model for the PSPRS included the striatal-[ 11 C]UCB-J BP ND component 1 and the fMRI-[ 11 C]UCB-J BP ND interaction term for component 5 (Fig. 3b and Supplementary Table 8).We found that the relationship between UCB-J BP ND loading and PSPRS was only present in those with higher connectivity scores.
In summary, reductions in synaptic density were associated with reduced functional connectivity, and that connectivity both adds to and moderates the explanatory effect of synaptic density on clinical severity.

Spatially remote functionally connected brain regions can moderate the effect of synaptic density on clinical severity
To test the effects of connectivity remote from synaptic loss, we considered connectivity scores from participants' component connectivity maps within connected regions but outside the [ 11 C]UCB-J BP ND component.We repeated the model selection process above using connectivity scores outside the [ 11 C]UCB-J BP ND mask.We found that for the final model predicting ACE-R (Supplementary Table 9), the fMRI-[ 11 C]UCB-J BP ND interaction was significant for component 4 for the ACE-R (Component 4 Interaction Std Beta 0.28, P = 0.005).The final model predicting PSPRS included only [ 11 C]UCB-J BP ND component 1 (Supplementary Table 10), although the fMRI-[ 11 C]UCB-J BP ND interaction was significant for component 5 with PSPRS as the dependent variable,

Discussion
There are three principal results of this study.First, lower synaptic density as estimated by [ 11 C]UCB-J BP ND is associated with lower functional connectivity.Second, functional connectivity augments and moderates the relationship between synaptic density and clinical severity.Third, the [ 11 C]UCB-J BP ND estimates of synaptic density predict individual differences in connectivity over and above grey matter volume.The profound reductions in synaptic density in frontotemporal lobar degeneration occur in partially overlapping distributions, but with some disease-specific effects.The corresponding reductions in functional connectivity are observed both at the site of the synaptic loss and remotely from it: as one of the potential mechanisms underlying diaschisis, functional diaschisis, and the recent formalisation of connectomal diaschisis 25 .Leveraging information from multiple modalities allows us to estimate the relative contribution of individual determinants of brain functional connectivity.
Our study therefore provides important mechanistic insights into the multiple and interacting pathogenic processes of FTLD that result in diverse clinical syndromes.
Progressive supranuclear palsy, corticobasal syndrome and behavioural variant frontotemporal dementia are severely disabling progressive conditions that significantly reduce life expectancy 5,6 .To improve patient outcomes we require models of the human neurodegenerative pathogenesis as a bridge between preclinical studies and experimental medicines in people.This approach complements mesoscale mechanistic models of cortical function in frontotemporal lobar degeneration 19,26 and macroscale whole-brain models of neurodegeneration [27][28][29] .Such transitional markers may facilitate the selection and design of trials of potential disease-modifying agents 7 .
Our primary finding of an association between lower synaptic density and reduced functional connectivity accords with the ubiquitous role of synaptic health and plasticity in generating neurophysiological connections, with synaptic change mediating information storage and contributing to learning 30,31 .In neurodegeneration, preclinical and neuropathological studies have shown that synaptic dysfunction and impaired plasticity are key determinants of impaired brain network organisation and cognitive dysfunction 14 , and occur before neuronal degeneration 32,33 .The in vivo associations between synaptic density and markers of connectivity derived are strengthened by the concordance across imaging and analytical methodologies.
The variances in functional connectivity explained by [ 11 C]UCB-J BP ND is independent of and in addition to that accounted for by grey matter atrophy, demonstrated through partial volume correction and by direct inclusion of grey matter volumes in the analysis models.This highlights the distinct contribution [ 11 C]UCB-J PET offers over and above T1-weighted MRI in capturing aspects of the neurodegenerative cascade.The histopathological processes that cause atrophy on structural imaging are incompletely characterised, with evidence for neuronal and synaptic loss, axon degeneration and cell death being important contributory factors 34,35 .Synaptic dysfunction and loss may occur without cell death or atrophy 12,32 , in keeping with our finding of regions of reduced [ 11 C]UCB-J BP ND but without atrophy.Quantifying dendritic complexity through the diffusion imaging method NODDI provides converging evidence for the role of synaptic loss in network dysfunction.Since [ 11 C]UCB-J BP ND and ODI are independent predicters of connectivity and function, measuring synaptic density and neurite dispersion has the potential to improve our understanding of the individual role of synaptic loss in cognitive symptoms.
The distinct proteinopathies underlying frontotemporal lobar degeneration show overlap in clinical syndromes and anatomical distribution, while the same pathological entity can result in heterogeneous clinical presentations 1,4 .Despite the heterogeneity and pleiotropy of FTLD-disorders, synaptic loss and dysfunction is a common feature 36,37 , associated with the direct and indirect toxic effects of multiple proteins' aggregates.We found patterns of synaptic density loss that were differentially expressed across the diagnostic labels in a manner in keeping with their typical neuropathological distributions described at post mortem 38,39 , albeit with regional synaptic loss occurring in a continuum across participants.For instance, we found lower [ 11 C]UCB-J BP ND loadings in CBS than PSP for posterior components, with frontal synaptic loss in all conditions with the largest effect sizes in bvFTD.Synaptic dysfunction results in altered cognition and behaviour through loss of connectivity, with synaptic loss strongly predicting cognitive function and decline 40,41 .We have shown that connectivity improves prediction and moderates the effect of synaptic density on clinical severity, with functional connectivity loss both overlapping with and occurring remotely from sites of reduced synaptic density.These remote connectivity differences can themselves moderate the relationship between synaptic density and cognition.
Our work advances on recent studies of fMRI and SV2A radioligand PET in psychiatric disease 42 and neurodegeneration 43 .Our work differs in investigating a cohort with a heterogeneous distribution of synaptic loss, through adopting whole-brain approach to demonstrate the anatomical effects of synaptic loss on connectivity, and by combining synaptic density, connectivity, and severity in a single model.Nonetheless, there are several limitations to our study.Interpreting connectivity from resting state fMRI is made challenging by the low signal-to-noise ratio, the impact of movement and other artefact on measures of connectivity 44 , the indirect neuronal interpretability the BOLD signal 23 and the small effect sizes of brainbehaviour studies in a healthy population 45 .Yet it is a scalable and widely available modality with good spatial localisation that provides a neural correlate closely connected to behaviour, with network integrity closely aligned with symptom development in neurodegeneration 20,21 .It is therefore a potential outcome marker of the consequences of upstream neuropathological change.We found that the relationship between connectivity and [ 11 C]UCB-J BP ND showed regional heterogeneity.That the variance in functional connectivity was not fully explained by synaptic density raises the possibility for a role for other factors, such as the neurochemical deficits observed in FTLD 46 .Combining PET, fMRI, and structural imaging modalities with clinical and other data would allow individualised modelling of pathophysiological pathways with implications for developing personalised treatments 47,48 .
We leverage the heterogeneity in FTLD to capture variation in the anatomical distribution of synaptic loss and explore its consequences but recognise that the small-moderate and varied group sizes mean that our study may be underpowered to investigate small differences between disease groups.Motion artifacts have a well-recognised confounding effect on functional connectivity 44,49 , with higher motion observed in participants with bvFTD during resting state fMRI scanning.Despite careful pre-processing, including ICA denoising and wavelet despiking, together with inclusion of an in-scanner movement related parameter as a covariate of no interest in relevant models, it remains possible that residual artefact influences our findings.We acknowledge that longitudinal and interventional studies are required to demonstrate causality between synaptic dysfunction, brain network organisation and clinical severity, and cannot be assumed from the statistical associations observed here.Although SV2A expression is closely related to synaptic activity and function 50 , [ 11 C]UCB-J BP ND is considered a measure of synaptic density rather than synaptic function 18 .Given that functional connectivity in fMRI is defined as a statistical dependency, with functional activation only indirectly related to neuronal activity, any inferences in our study made about synaptic function are necessarily implicit.Nonetheless our findings are as expected given the interplay between synaptic loss and dysfunction observed in preclinical studies 14 .We excluded participants with corticobasal syndrome and amyloid-positive PET imaging, but did not collect Alzheimer's disease biomarkers for patients with PSP and bvFTD.The clinicopathological correlation in PSP Richardson's syndrome is high 51,52 , while Alzheimer's disease pathology can occur in behavioural variant FTD 53 and may alter synaptic density and connectivity.Lastly, we used reference tissue modelling rather than arterial blood sampling given the challenges of scanning in an FTLDassociated patient cohort.We have previously performed sensitivity analyses to ensure that group differences cannot be explained by any bias introduced through reference tissue selection 9 .
To conclude, we report that reduced synaptic density in multiple frontotemporal lobar degeneration syndromes is associated with lower functional connectivity, with connectivity moderating the relationship between synaptic density and clinical severity.Synaptic density independently explains variance in connectivity beyond measuring atrophy from structural MRI.Our study provides in vivo support for preclinical findings and paves the way for individualised pathogenic models and personalised treatments.

Methods
Participants 29 participants with probable progressive supranuclear palsy, Richardson's syndrome 54 , 16 participants with probable corticobasal syndrome and probable corticobasal degeneration 55 , and 10 participants with behavioural variant Frontotemporal Dementia 56 were recruited from specialist clinics at the Cambridge Centre for Parkinson-plus, the Cambridge Centre for Frontotemporal Dementia, and National Hospital for Neurology and Neurosurgery at Queen Square, London 9,10 .24 healthy volunteers were recruited from the UK National Institute for Health Research Join Dementia Research (JDR) register.Participants were initially screened via telephone, using the following exclusion criteria: a current or recent history of cancer within the last 5 years, concurrent use of the medication levetiracetam, any contraindications to undergoing an MRI, a history of ischaemic or haemorrhagic stroke evident on the MRI from the clinic, and any severe physical illness or co-morbidity that could limit their ability to fully participate in the study.Twenty-three participants who passed initial screening were not included in this study due to (i) positive Alzheimer's biomarkers (all CBS, n = 9) or (ii) failure to complete all scanning sessions (n = 14: Control n = 8, bvFTD n = 1, CBS n = 1, PSP n = 4).Participants sex was determined based on participant self-report and monitored during recruitment to ensure balance between patient groups and controls.The research protocol (18/EE/0059) was approved by the Cambridge Research Ethics Committee and the Administration of Radioactive Substances Advisory Committee.All participants provided written informed consent in accordance with the Declaration of Helsinki.Participants were compensated for any travel, food, and accommodation expenses incurred as part of the study.
All participants underwent brain imaging with 3-Tesla MRI, including echo-planar imaging sequences sensitive to the bloodoxygen-level-dependent signal, and in a separate session PET scanning with 61 .All bar four participants (two PSP, one bvFTD, one control participant) additionally completed a diffusion sequence in the same session as echo-planar imaging (TE = 75.6 ms, TR = 2.4 s, slice thickness = 1.75 mm, 98 directions, 104 slices, bvals = 300, 1000, 2000).The median interval between scanning sessions was 58 days in patients (inter-quartile range 13-180 days) and 194 days in control participants (inter-quartile range 32-284 days).Participants with CBS also underwent amyloid PET using Pittsburgh compound B ([ 11 C]PiB).Cortical standardised uptake value ratio (SUVR; 50-70 minutes post injection; whole cerebellum reference tissue was determined using the Centiloid Project methodology 62 .Only participants with corticobasal syndrome and a negative amyloid status, as characterised by a cortical [ 11 C]PiB SUVR < 1.21 (obtained by converting the Centiloid cut-off of 19 to SUVR using the Centiloid-to-SUVR transformation 63 ) are included in the analysis.In the absence of Alzheimer's disease, corticobasal degeneration (CBD) is the most common pathological finding in CBS, although we acknowledge other pathologies are possible 64 .

Data acquisition and processing
We have previously reported our protocol for [ 11 C]UCB-J synthesis, data acquisition, image reconstruction and kinetic analysis 9 .Dynamic PET acquisition was performed on a GE SIGNA PET/MR (GE Healthcare, Waukesha, USA) for 90 minutes following [ 11 C]UCB-J injection, with attenuation correction including the use of a multisubject atlas method 65 and improvements to the MRI brain coil component.Each emission image series was aligned to a T1-weighted MRI acquired during the same session (TE = 3.6 ms, TR = 9.2 ms, 192 sagittal slices, inplane resolution 0.55 × 0.55 mm [interpolated to 1.0 × 1.0 mm]; slice thickness 1.0 mm).We derived a [ 11 C]UCB-J BP ND map for each participant from a dynamic image series corrected for partial volume effects using the iterative Yang method 66 .For regional analysis we used a modified version of the n30r83 Hammersmith atlas (http://braindevelopment.org) including segmentation of brainstem and cerebellar structures, with the atlas non-rigidly registered to the T1-weighted MRI of each participant.Regions of interest were multiplied by a binary grey matter mask determined from the SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) grey matter probability map smoothed to PET resolution and thresholded at >50%, unless masking eliminated the region in some subjects.For regional analysis, correction for cerebrospinal fluid partial volume error was applied to each image of the dynamic series.[ 11 C]UCB-J binding potential (BP ND ) was calculated both at the regional and voxelwise level using a basis function implementation of the simplified reference tissue model 67 , with centrum semiovale as the reference tissue 68 .We further calculated regional [ 11 C] UCB-J BP ND for parcels of the Brainnetome Atlas 69 to test whether the results were independent of parcellation.For independent component analysis (see below), [ 11 C]UCB-J BP ND maps were registered to the FSL MNI152 6 th generation atlas using the warps and linear transforms created in spatial normalisation of the T1-weighted images using Advanced Normalization Tools (ANTs) 70 version 2.1.0.Normalised maps were smoothed with a 6 mm Gaussian kernel.
Functional MRI was performed with a 3-Tesla Siemens Prisma (Siemens Healthcare using echo-planar imaging sensitive to the blood-oxygen-level-dependent signal (TR 2.5 secs, TE 30 ms, whole brain acquisition, 3 x 3 x 3.5 mm voxels, 200 volumes).High resolution T1-weighted Magnetisation Prepared Rapid Gradient Echo (MPRAGE structural images (TR 2, TE 2.93 ms, voxel size 1.1 mm isotropic) were acquired during the same session.fMRI preprocessing followed the FSL pipeline 71 (using version 6.0.4), with the addition of wavelet despiking 72 given that motion artefact may be more common in patients with neurodegenerative diseases.T1 structural images were cropped to remove non-brain tissue followed by brain extraction using FSL's Brain Extraction Tool (BET).We then used FSL's FEAT with the following steps: motion correction using MCFLIRT; spatial smoothing using a Gaussian kernel of 5 mm full-width at half maximum; grand-mean intensity normalisation of the 4D dataset by a single multiplicative factor; and 100 Hz high-pass temporal filtering.Structured artefacts were removed using independent component analysis denoising using FSL's MELODIC together with FIX.FIX was hand-trained using a set of 20 subjects.Registration to high resolution structural and/or standard space images was carried out using FLIRT.Registration from high-resolution structural to MNI space was then further refined using FNIRT nonlinear registration.We did not use global signal regression.Wavelet despiking was used for further removal of motion artefact 72 .For dual regression analysis data was further smoothed with a 6 mm FWHM Gaussian kernel.
From volumetric T1-weighted MRI images grey matter volumes and total brainstem volumes were extracted for the same regions of the Hammersmith Atlas using SPM12 (version 7771) segmentation.Total intracranial volume was calculated via direct segmentation using Sequence Adaptive Multimodal SEGmentation 73 .We followed our previously reported pipeline for analysis of diffusion-weighted MRI to obtain the neurite orientation dispersion and density imaging (NODDI) metric orientation dispersion index (ODI) 15 .Non-brain tissue was removed from diffusion datasets using BET.Eddy currents and inscanner head motion were corrected using FSL's eddy.We applied TOPUP to correct for correction of susceptibility induced distortions.Eddy was further used for quantitative identification of slices with signal loss, which were replaced by non-parametric predictions with the Gaussian process 74 .The b-matrix was reorientated by applying the rotation part of the affine transformation used during eddy correction 75 .ODI maps were derived using the Microstructural Diffusion Toolbox 76 .ODI maps were warped to MNI space using parameters from the spatial normalisation of the co-registered T1 image using ANTS.We then obtained regional ODI values for the modified Hammersmith Atlas within a grey matter mask using FSL's fslstats function.

Weighted degree
Participants' preprocessed fMRI was parcellated using the modified Hammersmith atlas, with cortical regions masked with a grey matter mask.The straight gyrus was excluded given limited coverage in some participants.Given the variation in parcel size in the Hammersmith atlas, we sub-parcellated the masked Hammersmith atlas with 161 regions of approximately equal volume, such that each sub-parcel could be uniquely identified with an atlas region.Pearson correlations were calculated between nodes, followed by Fisher's r-to-Z transformation.Weighted degree was derived from association matrices using the Maybrain software (https://github.com/RittmanResearch/maybrain) and Networkx.We then calculated mean weighted degree across the sub-parcellations for each Hammersmith atlas region to compare with regional [ 11 C]UCB-J BP ND .

Comparing modalities by group
We compared regional values for [ 11 C]UCB-J BP ND , weighted degree, grey matter volumes/total brainstem volumes, and ODI between participants with FTLD-associated syndromes and controls in a linear model with age and sex as covariates (plus mean DVARS for weighted degree and total intracranial volume for volumetric measures).We further compared differences in the four modalities between patient groups in a linear model with the same covariates.For both analyses P values were adjusted for the false discovery rate across regions 77 .We additionally tested for differences in [ 11 C]UCB-J BP ND with grey matter volume and total intracranial volume included as covariates of no interest in regional and voxelwise analysis (see Supplementary Methods).

Weighted degree and synaptic density
We tested the spatial correlation between regional z scores for weighted degree and [ 11 C]UCB-J BP ND using mean values and standard deviation from the control participants.P-values were calculated using a permutation test with 5000 spatial autocorrelation-preserving null models of the weighted degree parcellation 78 , using the neuromaps toolbox 79 .
To capture the effect of individual variability in the relationship between weighted degree and [ 11 C]UCB-J BP ND , we then derived mixed linear effects models for all patients using the lme4 package in R 80 with crossed random effects for region and participant and an effect of [ 11 C] UCB-J BP ND slope within each region.We compared models using the anova function in R to ensure that the inclusion of a random slope for [ 11 C]UCB-J BP ND per region improved model fit.We repeated the model with control participants alone, and for each clinical syndrome individually but without a random slope for [ 11 C]UCB-J BP ND given convergence failures in the CBS and bvFTD groups.We tested the difference between patients and controls with a further model using all participants including a group by [ 11 C]UCB-J BP ND interaction.Age at fMRI scan, sex and a marker of fMRI motion (mean DVARS) 44 were included as covariates of no interest.We also included a covariate denoting whether a region was cortical or subcortical, to ensure that the relationship between [ 11 C]UCB-J BP ND and weighted degree was not driven by differences due to the average lower [ 11 C]UCB-J BP ND in subcortical regions, and a group by cortical/subcortical interaction in the model with all participants.We performed a sensitivity analysis with scanning interval as an additional covariate.We further tested whether the inclusion of regional grey matter volume and total intracranial volume in the model in patients altered the [ 11 C]UCB-J BP NDweighted degree relationship.We repeated models assessing the relationship between [ 11 C]UCB-J BP ND and a weighted degree in patients and control participants using the Brainnetome parcellation and the non-reparcellated Hammersmith Atlas, with weighted degree calculated for regions of these atlases using the methodology outlined above.
To assess converging evidence for the role of synaptic health in determining functional connectivity, we considered the complementary NODDI derived metric ODI.We tested the relationship between weighted degree and ODI (and between [ 11 C]UCB-J BP ND and ODI) in the mixed linear effects models outlined above.In controls, the model comparing ODI to [ 11 C]UCB-J BP ND was fit without a slope for [ 11 C]UCB-J BP ND in region due to model singularity.
To inform the contribution of different modalities to functional connectivity we compared weighted degree to [ 11 C]UCB-J BP ND , ODI, and grey matter volumes, across participants and within each cortical region separately.Relationships were tested in a general linear model with age at fMRI, sex, and mean DVARS as covariates, plus total intracranial volume for grey matter volumes.We used permutation testing using FSL's Permutation Analysis of Linear Models 81 (PALM version alpha119) with 10000 permutations and correction for the false discovery rate across cortical regions.To test if the three modalities in combination would improve the prediction of weighted degree, we performed joint inference with non-parametric combination testing 82 , with correction across the number of contrasts and the same covariates.In addition, we tested whether the relationships between weighted degree and [ 11 C]UCB-J BP ND would persist with grey matter and total intracranial volume as covariates of no interest, both at regional and voxelwise (Supplementary Methods) levels.

Source-based synaptometry
We then proceeded to independent component analysis (referred to here as source-based synaptometry, analogous to "volumetry") to identify a small number of statistically independent components capturing spatial variation in [ 11 C]UCB-J BP ND .Spatially concatenated [ 11 C]UCB-J BP ND maps were submitted to sourcebased synaptometry using the GIFT toolbox 83,84 with a model order of 10, with regions with negative values in any participants masked out.Components were discarded if they represented artefact, captured regions known to be sensitive to artefact in fMRI acquisition or were driven by outliers identified using Grubbs' test 85 (i.e. the component did not persist with the outlier removed).Component loading values, which represent the degree to which an individual expresses a [ 11 C]UCB-J BP ND component map, were taken forward to estimate association with connectivity providing they were differentially expressed by participants with neurodegenerative diseases and controls after correction for false discovery rate with p < 0.05.Six components satisfying these criteria were included.In the primary analysis presented here partial volume corrected [ 11 C]UCB-J BP ND maps were used, with repeat sourcebased synaptometry with uncorrected maps performed to ensure robustness of spatial distribution of components to atrophy correction.Independent component analysis model order was chosen a priori, with additional analysis that components of interest with similar spatial distributions could be extracted at alternative model orders.

Connectivity of [ 11 C]UCB-J BP ND components
We then sought to investigate subject functional spatial covariance with the identified [ 11 C]UCB-J BP ND components, using a seed-based dual regression approach 86,87 .In the first stage of dual regression, we regressed the selected [ 11 C]UCB-J BP ND components maps into each participant's fMRI 4-dimensional dataset to give participants specific timecourses per component.These timecourses were taken to a second regression with the [ 11 C]UCB-J BP ND component maps to obtain participant spatial maps per component, termed here componentspecific functional connectivity maps.We assessed the association between [ 11 C]UCB-J BP ND loadings values and each voxel of the component-specific functional connectivity maps in a general linear model with age, sex, and mean DVARS as covariates of no interest using threshold-free cluster enhancement with 5000 permutations using FSL's randomise tool 81 with family-wise error significance level p < 0.05.We then calculated the mean beta, normalised to the residual within-subject noise, for each participant's component-specific functional connectivity maps from the second stage of dual regression (within a mask defined as regions in controls showing significant mean connectivity with each [ 11 C]UCB-J BP ND component map at family-wise error p < 0.01) using FSL's fslmeants function.We compared the association between [ 11 C]UCB-J BP ND loadings and functional connectivity scores for each component in the whole group and in patients alone using the same covariates of no interest.

Associations with clinical scores
We and others have previously shown that [ 11 C]UCB-J BP ND values are strongly related to markers of clinical severity in neurodegenerative diseases 10,88,89 .We sought to understand whether connectivity moderates this relationship and improves the prediction of total ACE-R and total PSPRS.We, therefore, performed model selection using stepwise regression with the Bayesian information criteria from a baseline model of: Age at fMRI, sex and mean DVARS were considered covariates of no interest and were not stepped out of the model.

Remote connectivity differences
We proceeded to explore the effects on cognition of connectivity differences at the site of synaptic loss and of the region connected to it.To do this we thresholded and binarised the [ 11 C]UCB-J BP ND component maps beyond the 97.5 th percentile point.We calculated connectivity scores, as outlined above, both within the component mask and within connected regions but outside the component mask.Then, on a post hoc basis, we repeated the stepwise regression for PSPRS and ACE-R using only the connectivity scores outside the [ 11 C]UCB-J BP ND component mask.

Statistical analysis
We tested the relationships between [ 11 C]UCB-J BP ND , ODI, functional connectivity, grey matter volume, and cognition in regional and voxelwise analyses.First, we tested whether regional [ 11 C]UCB-J BP ND and ODI explain regional variation in connectivity beyond that accounted for by grey matter volume.We then used voxel-wise [ 11 C] UCB-J BP ND maps to identify patterns of [ 11 C]UCB-J BP ND that were differentially expressed in neurodegeneration using an independent component analysis 76 .We identified participants-specific maps of functional spatial covariance with the [ 11 C]UCB-J components through seed-based dual regression.We tested whether variability in functional connectivity to regions showing group differences in synaptic density: a) improves modelling of clinical severity as assessed through the ACE-R and PSPRS; b) moderates the effect of [ 11 C] UCB-J BP ND differences on cognition.We further tested whether the effects of [ 11 C]UCB-J BP ND differences on cognition were influenced by connectivity remote from synaptic loss.Statistical analyses and visualisation were performed in R 90 version 4.1.2,unless stated otherwise.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
Derived imaging data, including those required to generate Figs.1-3 and supplementary figures, are provided in the Source Data file.These data have also been deposited in a Figshare repository (https://doi.org/10.6084/m9.figshare.24188580) 91.Access to raw imaging data and linked clinical data may be available via an agreed third party Secure Research Environment or by transfer under a material/data transfer agreement, subject to conditions required to comply with participant consent and data protection regulations.Requests should be addressed to the senior author in the first instance.Initial responses to queries or requests for data, determining the need for any material/ data transfer agreement, will be provided within a month of a request being made.The n30r83 Hammersmith atlas was modified for regional analysis and is publicly available at http://brain-development.org.The Brainnetome Atlas (https://atlas.brainnetome.org/)used for ensuring results are independent of parcellation in this study is publicly available.Source data are provided with this paper.

Fig. 1 |
Fig. 1 | Weighted degree and synaptic density in frontotemporal lobar degeneration syndromes.a Mean [ 11 C]UCB-J binding potential (BP ND ) in cortical regions in patients (n = 55) is associated with mean weighted degree, with z scores calculated relative to control values.P value calculated using two-sided spatial-autocorrelation preserving permutation testing.The distribution of null models is shown in the box-plot, with median, inter-quartile range, and whiskers extending from the hinge to the largest/smallest point at most 1.5 times the interquartile range from the hinge.b Variation in strength of the weighted degree-[ 11 C]UCB-J BP ND relationship in patients (n = 55) derived from the effect of [ 11 C]UCB-J BP ND slope within each region, adjusted by the [ 11 C]UCB-J BP ND fixed effect.Larger standardised estimates were observed across the cortex and away from inferior frontal, anterior temporal and striatal regions.c Regions with a relationship with weighted degree in a general linear model for all participants for [ 11 C]UCB-J BP ND , orientation dispersion index (ODI), grey matter (GM) volumes, and for the three modalities combined using non-parametric combination testing with correction across contrasts.Regions shown are significant with FDR-adjusted p-values < 0.05.Source data are provided as a Source Data file.
[ 11 C]UCB-J BP ND loadings improved modeling of clinical severity as measured by the ACE-R and PSPRS.Individual functional component scores were not significant predictors of either the ACE-R or the PSPRS, whereas [ 11 C]UCB-J BP ND component 2 (Standardised Beta 0.44 FDR P = 0.002), [ 11 C]UCB-J BP ND component 3 (Standardised Beta 0.49 FDR P = 0.001), UCB-J BP ND component 4 (Standardised Beta 0.64 FDR P = 6 × 10 −6 ), and UCB-J BP ND component 6 (Standardised Beta 0.47 FDR P = 0.0009) were significantly associated with total ACE-R.Lower loadings on striatal [ 11 C]UCB-J BP ND component 1 were significantly associated with higher PSPRS scores (Standardised Beta −0.50 FDR P = 0.001).Stepwise regression using Bayesian Information Criteria determined whether combinations of predictors improved model fit.The winning model for ACE-R included [ 11 C]UCB-J BP ND component 2 and the fMRI-[ 11 C]UCB-J BP ND interaction terms for components 4 and 5 (Fig.

Fig. 2 |
Fig. 2 | Spatial variation in synaptic density and associated connectivity loss.a [ 11 C]UCB-J BP ND independent component analysis loadings by group for the six components that show differential expression in neurodegeneration in a linear model with false discovery rate correction across multiple comparison.Box and whisker plots show median, inter-quartile range, with whiskers extending from the hinge to the largest/smallest point at most 1.5 times the interquartile range from the hinge.Significance brackets (* p < 0.05, ** p < 0.01, *** p < 0.001) represent P values from Tukey adjusted post-hoc pairwise comparisons.Component 1 bvFTD <Control P = 0.0001, CBS <Control P = 0.0015, PSP <Control P < 0.0001; component 2 bvFTD <Control P = 0.018, bvFTD <PSP P = 0.021, CBS <Control P = 0.029, CBS < PSP P = 0.024; component 3 bvFTD <Control P < 0.0001, bvFTD <CBS P = 0.003, bvFTD <PSP P = 0.010; PSP <Control P = 0.0014; component 4 bvFTD <Control P < 0.0001, bvFTD <CBS P < 0.0001, bvFTD <PSP P < 0.0001, PSP <Control P = 0.002; component 5 bvFTD <Control P < 0.0001, bvFTD <CBS P = 0.029, PSP <Control P = 0.003; component 6 bvFTD <Control P = 0.0001, CBS<Control P = 0.015, PSP <Control P < 0.0001.b [ 11 C]UCB-J BP ND component maps (in green) with areas of increased functional connectivity (in red-orange significantly associated with increased [ 11 C]UCB-J BP ND component loadings.P values calculated using a one-sided permutation test with family-wise error correction across voxels.c Connectivity scores, derived from participant-specific functional connectivity maps per component, are associated with [ 11 C]UCB-J BP ND independent component analysis loadings.Standardised beta calculated from linear regression with twosided testing, with false discovery rate adjusted P values.The error bands represent the 95% confidence interval.bvFTD behavioural variant frontotemporal dementia, CBS corticobasal syndrome, PSP progressive supranuclear palsy, FWE family-wise error, FDR false discovery rate.Source data are provided as a Source Data file.

Fig. 3 |
Fig. 3 | Functional connectivity, synaptic density, and clinical severity.[ 11 C] UCB-J BP ND components (in green) and distribution of functional connectivity (in blue) for components included in final model selected using stepwise regression for a Addenbrooke's Cognitive Examination-Revised (ACE-R) and b Progressive Supranuclear Palsy Rating Scale (PSPRS).The relationship between the fitted final model and true scores is shown with a linear regression with error bands representing the 95% confidence interval.For the interactions included in the final models the relationship between [ 11 C]UCB-J BP ND component loading and clinical measure of severity was seen only in those with higher connectivity scores.Standardised betas and P values were calculated from two-sided testing from linear regression for the winning model, with the direction of the interaction terms illustrated with the marginal effect of the fMRI component with three groups (mean and +/-one standard deviation) and 95% confidence intervals.bvFTD: behavioural variant frontotemporal dementia, CBS: corticobasal syndrome, PSP progressive supranuclear palsy, fMRI functional magnetic resonance imaging.Source data are provided as a Source Data file.

Table 1 |
Demographic and clinical characteristics for participants Scores are mean (Sd).Categorical and continuous variables were compared across groups using chi-squared tests and one-way analysis of variance respectively.Post-hoc pairwise comparisons used two-sided testing with Tukey method to adjust P values.fMRI functional magnetic resonance imaging, DVARS spatial standard deviation of successive images, ACE-R Addenbrooke's Cognitive Examination-Revised, PSPRS Progressive Supranuclear Palsy Rating Scale, CBI-R Cambridge Behavioural Inventory Revised.