Neuroinflammation and Tau Colocalize in vivo in Progressive Supranuclear Palsy

We examined the relationship between tau pathology and neuroinflammation using [11C]PK11195 and [18F]AV‐1451 PET in 17 patients with progressive supranuclear palsy (PSP) Richardson's syndrome. We tested the hypothesis that neuroinflammation and tau protein aggregation colocalize macroscopically, and correlate with clinical severity.

has not addressed whether in vivo tau pathology and microglial activation are related in PSP. Answering this question would shed new lights in the pathophysiological mechanisms underlying PSP and may facilitate the development of therapeutic strategies that synergistically target neuroinflammation and tau pathology in PSP.
We used [ 11 C]PK11195 PET to measure of microglial activation and [ 18 F]AV-1451 PET as an index of tau burden (see discussion regarding limitations of these ligands and their mitigation 19,20 ). The latter binds to aggregated tau in Alzheimer's disease (AD) and, with lower affinity, to non-AD tauopathies. 21 It also does not distinguish tau-from TDP43-pathololgies. However, this lack of specificity does not undermine its utility to test our hypotheses because: (1) the clinical-pathological correlation in PSP-Richardson's syndrome is very high 22 ; (2) significant TDP43 pathology is exceedingly rare in PSP; and (3) [ 18 F]AV-1451 displays a specific anatomic pattern of binding that clearly distinguishes PSP from AD. 8 Moreover, to test our hypothesis, it is the distribution not relative affinity of binding that is critical. [ 11 C] PK11195 is widely used as a marker of microglial activation in neurodegenerative diseases, 23 including PSP. 17,18 It binds the translocator protein (TSPO) on mitochondrial membranes in activated microglia. Although its sensitivity may be affected by its relatively low signal-to-noise ratio and low brain penetration, [ 11 C]PK11195 is not significantly influenced by common genetic polymorphisms that affect second generation TSPO ligands.
We test the hypothesis that neuroinflammation and tau protein aggregation colocalize and correlate with clinical severity in PSP. We assessed the topography of this relationship with [ 11 C]PK11195 PET and [ 18 F]AV-1451 PET using: (1) a region of interest (ROI) approach to study correlations across the brain; and (2) a set of spatial patterns determined by principal component analysis (PCA).

Participants
As part of the Neuroimaging of Inflammation in Memory and Related Other Disorders (NIMROD) study, 24 we recruited 17 patients with a clinical diagnosis of probable PSP according to the Movement Disorder Society (MDS) 1996 criteria. All patients also met the later MDS-PSP 2017 criteria for PSP-Richardson's syndrome. 4 Patients underwent PET scanning with both [ 11 C]PK11195 and [ 18 F]AV-1451 to respectively assess neuroinflammation and tau pathology. To minimize radiation exposure in healthy people, 2 groups of control participants were enrolled: n = 16 underwent [ 11 C]PK11195 PET and n = 15 underwent [ 18 F]AV-1451 PET. At the first visit, demographic and cognitive measures (ie, the revised Addenbrooke's Cognitive Examination [ACE-R]) were collected in all participants. Disease severity of patients was measured at the baseline visit and at 6 monthly intervals, using the PSP rating scale (PSP-RS) (Fig 1). Participants had the mental capacity to take part in the study and provided written informed consent. The protocol was approved by the National Research Ethic Service East of England Cambridge Central Committee and the UK Administration of Radioactive Substances Advisory Committee.
PET and Magnetic Resonance Imaging Data Acquisition and Pre-Processing Full details of the imaging protocol are published elsewhere. 8,18 In brief, patients underwent 3 T magnetic resonance imaging (MRI) together with [ 11 C]PK11195 and [ 18 F]AV-1451 PET, using dynamic imaging for 75 and 90 minutes, respectively. The MRI scans used Siemens Magnetom Tim Trio and Verio scanners (Siemens Healthineers, Erlangen, Germany), whereas PET used a GE Advance and a GE Discovery 690 PET / computed tomography (CT; GE Healthcare, Waukesha, WI). The use of identical emission data acquisition protocols and image reconstruction algorithms on the 2 scanners meant that the differences were effectively limited to the attenuation correction method (rotating rod 68 Ge/ 68 Ga transmission scan vs a low dose CT scan) and the axial spatial resolution (6.8 mm full width at half maximum [FWHM] vs 5.1 mm FWHM). Regarding the attenuation correction, the CT (Hounsfield unit) to 511 keV linear attenuation coefficient transformation used on GE PET / CT systems is that of Burger et al, 25 which was determined from data acquired with a GE Discovery LS PET / CT, the PET part of which is identical to the GE Advance, thereby enhancing the correspondence between GE PET and PET / CT systems. With respect to differences in spatial resolution, the primary data given in the paper are for large ROIs. This will limit the impact of any spatial resolution differences, which for brain imaging on the scanners used mainly occur in the axial dimension. Furthermore, patient motion, together with resolution losses in image processing steps, such as realignment of dynamic image series and co-registration to MR, will reduce these differences. The interval between [ 11 C] PK11195 and [ 18 F]AV-1451 PET scans had mean and standard deviation (SD) of 1.2 AE 1.7 months. Eleven patients underwent [ 11 C]PK11195 and [ 18 F]AV-1451 PET scans using a GE Discovery scanner, whereas 6 patients underwent scans using a GE Advance scanner. These 2 groups did not differ in demographic or clinical characteristics (p > 0.05).
For each subject, the aligned dynamic PET image series for each scan was rigidly coregistered to the T1-weighted MRI image. BP ND was calculated in 83 cortical and subcortical ROIs using a modified version of the n30r83 Hammersmith atlas (www.brain-development. org), which includes brainstem parcellation and the cerebellar dentate nucleus. Prior to kinetic modeling, regional PET data were corrected for partial volume effects from cerebrospinal fluid by dividing by the mean regional greymatter plus white-matter fraction determined from SPM segmentation. For [ 11 C]PK11195, supervised cluster analysis was used to determine the reference tissue timeactivity curve and BP ND values were calculated in each ROI using a simplified reference tissue model with vascular binding correction. 26 For [ 18 F]AV-1451, BP ND values were quantified in each ROI using a basis function implementation of the simplified reference tissue model, with superior cerebellar cortex grey matter as the reference region. This cerebellar region was selected as reference region given postmortem evidence showing minimal tau pathology in PSP. 8 The same data acquisition and analysis approach was applied for the 2 control groups. In the control groups, 10 individuals were scanned using the GE Discovery scanner (N = 3 with

Statistical Analyses
Age, years of education, and ACE-R total and fluency scores were compared between patients and controls with independent samples t-tests, whereas sex was compared with the chi-square test. Statistical analysis proceeded in 4 steps.
First, to test whether microglial activation and tau pathology colocalized across the whole brain, we estimated the Pearson correlation of corresponding [ 11 C]PK11195 and [ 18 F]AV-1451 BP ND group average values across all 83 ROIs. We also applied a linear mixed effects model that takes into account the variability between subjects in both intercept and slope of the relation between the 2 tracers' regional binding. We compared 3 models: (1) an initial model with only a random intercept term for patients, (2) a model with also the fixed effect of regional [11C]PK11195 PET values (x variable) on regional [18F] AV-1451 PET (y variable), and (3) a model with uncorrelated random intercept and random slope terms. The R function (permlmer) was used to perform permutation tests on the terms of interest in each model comparison.
Second, the number of ROIs was reduced from 83 to 43, averaging left and right regional BP ND values, as in previous studies. 8,18 This step reduces the degrees of freedom, increasing power, and is justified in PSP in which the motor syndrome is essentially symmetric. Differences between PSP and controls in the ROIs were tested for each ligand with independent t-tests and false discovery rate (Benjamini-Hochberg False Discovery Rate [FDR]) correction for multiple comparisons.
Third, in patients with PSP, BP ND values in the 43 bilateral ROIs were included in separate PCAs for [ 11 C]PK11195 and [ 18 F]AV-1451. This reduces the data dimensionality further, identifying a small set of components that best explain the data variance. The resulting component revealed that anatomic patterns covary in terms of neuroinflammation or tau pathology. The orthogonal varimax rotation was applied on the single PCA, separately for each ligand. This rotation serves to maximize the dispersion of loadings within components and to facilitate their interpretability (ie, anatomic patterns of neuroinflammation and tau pathology). We selected components with eigenvalues > 1, up to a cumulative total of > 80% of variance explained. PET components were visualized with BrainNet Viewer. 27 Finally, to test for colocalization of microglial activation and tau pathology in specific neuroanatomic patterns of ligand binding, we performed Pearson correlations between individual scores of each ligand-specific component extracted. Bonferroni's method was used to correct for multiple comparisons. We also report the analyses adjusted for age, education, and sex, or for variability in the time interval between PET scans, included as covariates of no interest. For each ligand, we tested for significant associations between regionally specific intermodality correlating PCA clusters (ie, anatomic patterns of neuroinflammation and tau pathology) and disease severity. For severity, we imputed PSP-RS scores for the time of each scan (Fig 1), using multivariate imputation by chain equations ("mice" function in R software). Patient ID number, months from baseline to each followup visit, and all available PSP-RS scores were included in the multiple imputation. From this, 100 sets of PSP-RS scores were imputed for each PET scan, with 50 iterations to generate each estimated dataset. The average value per participant across all 100 ligand-specific estimated PSP-RS scores (Fig 1, blue and red dots), was retained for correlation analyses with imaging components. These correlation analyses were repeated using a single raw PSP-RS score as clinical severity index, identified as the closest clinical assessment to both PET scans.

Data Availability
Anonymized data may be shared by request to the senior author from a qualified investigator for noncommercial use (data sharing may be subject to restrictions according to consent and data protection legislation).

Discussion
This study suggests that microglial activation and tau protein aggregation are colocalized in PSP, macroscopically. The relationship between neuroinflammation and tau pathology is observed across widespread brain regions, although it is most evident in a subset of cortical regions (ie, insula and temporo-parietal junction) and subcortical structures (ie, brainstem and cerebellum). This colocalization resembles that observed between protein aggregation and microglial activation in both tau-and TDP43-associated forms of frontotemporal lobar degeneration observed by postmortem immunohistochemistry and in vivo with the ligands used in this PET study. 7 The in vivo measures of neuroinflammation and tau burden in the brainstem and cerebellum were both associated with disease severity.
Before considering the implications of our findings for the pathogenesis of PSP, we discuss the caveats associated with the PET ligands. Although [ 18 F]AV-1451 shows strong in vivo and postmortem binding to tau pathology in AD, it displays variable affinity in healthy aging and non-AD tauopathies. [28][29][30] The tracer also binds to nontau proteinopathies (ie, is positive in TDP43 pathology with C9orf72 mutations and semantic dementia 31 ), and other targets, such as neuromelanin, 29 choroid plexus, 28 and monoamine oxidase (MAO). 32 However, "off-target" binding can neither fully account for [ 18 F]AV-1451 signal in striatum or cortexas previous postmortem analyses reveal that no neuromelanin accumulates there 8,33nor in the choroid plexus, as histological analysis revealed tangle-like inclusions in its epithelial cells. 34 The [ 18 F]AV-1451 "off-target" binding to MAO 32 expressed on reactive astrocytes and activated microglia, 32,35 could in principle contribute to correlated binding of the 2 ligands. However, a previous report of a carrier of a microtubule associated protein tau (MAPT) genetic mutation described high [ 11 C]PK11195 binding in the absence of significant [ 18 F] AV-1451 binding, 36 which suggests that despite the potential for a common target, the 2 ligands are not equivalent in their binding. Furthermore, affinity of [ 18 F] AV-1451 for monoamine oxidase is weak and the use of MAO inhibitors does not significantly displace [ 18 F]AV-1451 binding in patients with tauopathies. 37 Nevertheless, we acknowledge that the affinity of [ 18 F]AV-1451 to the 4-repeat tau in non-AD tauopathies as PSP is lower than its affinity to 3/4-repeat AD-related tau pathology. 29,30 In PSP, higher [ 18 F]AV-1451 binding has been found in subcortical rather than cortical regions (the reverse for AD), consistent with the well-established cortical versus subcortical distribution of PSP-Richardson's syndrome. [8][9][10]13,14,16 This supports the use of [ 18 F]AV-1451 PET to quantify and localize tau pathology in different tauopathies with clear and known pathologic substrates, such as PSP.
The [ 11 C]PK11195 ligand is selective for activated microglia over quiescent microglia or reactive astrocytes, 38 but it has been criticized for its relatively low signal-tonoise ratio and low brain penetration, which may affect its sensitivity to activated microglia. Nevertheless, this would reduce effect sizes and increase type II errors, rather than leading to false positive findings. Second-generation PET radioligands for TSPO are characterized by higher signalnoise ratio than [ 11 C]PK11195 but their binding is markedly affected by single nucleotide polymorphisms (rs6971), which cause heterogeneity in PET data and require genetic screening. The [ 11 C]PK11195 binding is less affected by this polymorphism, and has well established methods of kinetic analysis. 39 Hence, [ 11 C] PK11195 PET was the ligand of choice for this study of PSP.
With these caveats in mind, we turn to our principal results. To study the in vivo colocalization between microglial activation and tau pathology in PSP, we applied correlation analyses between binding of the 2 ligands: (1) across all brain regions, and (2) between principal components of a set of bilateral brain regions, extracted to reduce complexity of the imaging data. With the first approach, we found a positive correlation at group level between [ 11 C]PK11195 and [ 18 F]AV-1451 binding across the whole brain (see Fig 2A), which remained significant also after accounting for the variability between patients (see Fig 2B). This indicates a close association between microglial activation and tau pathology in PSP that extensively involves subcortical and cortical regions. This finding aligns with in vivo correlation between neuroinflammation and tau aggregation in AD and frontotemporal dementia. 7,40 Collectively, these multitracer PET studies support previous in vitro evidence of the association between microglial activation and tau aggregation in different tauopathies. 41 The spatial distribution of this in vivo association between microglial activation and tau pathology in PSP mirrors previous findings showing that tau pathology affects not only subcortical but also cortical regions in PSP Richardson's syndrome. 42,43 When assessing the ligand-specific components from principal component analysis, we found a positive correlation between [ 11 C]PK11195 and [ 18 F]AV-1451 binding in brainstem and cerebellar regions, loaded into anatomically overlapping components of both ligands (see Fig 4,  right panel). This association occurs in motor-related regions that are involved in the neuropathology and symptomatology of PSP (ie, functional deficits, postural instability, and supranuclear gaze palsy) associated with PSP Richardson's syndrome. 44 Of note, both tau pathology and microglial activation in the brainstem-cerebellar component correlated with disease severity, using either the imputed PSP-RS scores for time of PET scans (see Fig 5) or the score of the closest PSP-RS assessment to PET scans. This finding adds relevant information to the literature as only a few previous studies have explored how [ 11 C]PK11195 and [ 18 F]AV-1451 binding relate to disease severity in PSP. 8,9,13,14,16,18 In our sample, although the 2 components individually correlate with clinical severity, they did not interact in their association with PSP-RS. This suggests an additive and partially independent effect of the 2 pathological processes on clinical progression rather than a synergistic effect, although larger sample sizes and longitudinal designs will need to explore this relationship further.
The [ 11 C]PK11195 and [ 18 F]AV-1451 binding were also correlated in a cortical component (see Fig 4, left panel), which for both ligands was weighted toward regions of the medial temporal lobe, insula, and temporo-parietal junction.
Previous studies have implicated the medial temporal lobe and limbic structures in basic emotional recognition, which is in turn found to be impaired in PSP, alongside theory of mind and social cognition. 45,46 Although recognition of happiness was reported to be preserved in patients with PSP, their recognition of negative emotions (ie, anger, disgust, surprise, fear, and sadness) has been reported as affected. 45 The basal ganglia, insula, and amygdala have been reported to be implicated in the recognition of predominately negative emotions, and these neuropsychological functions are abnormal in PSP. 43,46 The association between microglial activation and tau pathology we found in limbic regions may complement biological explanations of emotion-related and social deficits in PSP. However, longitudinal studies are needed to clarify the timing of these interacting effects on pathological and clinical disease progression.
Our finding of tau and neuroinflammation colocalization in the cortex of patients with PSP Richardson's syndrome is in keeping with previous postmortem evidence showing tau pathology and atrophy not only in subcortical and limbic regions, but also in the parietal lobe. 43 Specifically, the supramarginal gyrus has been described as the most affected brain region in 2 independent pathological cohorts of patients with PSP Richardson's syndrome. 43 The absence of in vivo evidence about supramarginal atrophy in the literature may enhance the importance of the demonstrated association between neuroinflammation and tau accumulation in this region as an early biomarker of later-stage neuronal loss.
Our study has limitations. First, we acknowledge the limited power of the analyses related to the relatively small size of our sample. Although our cohort is larger than many previous multi-tracer PET studies on rare neurodegenerative diseases like PSP, larger and independent replication samples would be helpful. Second, the recruitment was based on clinical diagnosis, which was confirmed at each follow-up visit; however, postmortem pathological confirmation was available only in 8 patients. However, all 8 patients had PSP, and over 95% of patients with a clinical diagnosis of PSP-Richardson's syndrome have PSP pathology or related 4R-tauopathy. Third, our results are based on a cross-sectional design, which cannot be used to infer causal relationship between tau and microglial activation. A longitudinal assessment of tau burden, microglial activation, and clinical progression alongside mediation analyses are necessary next steps to clarify the interplay between the 2 pathological processes and their effect on disease severity across time.
To conclude, our results confirm the relevance of neuroinflammation to PSP-Richardson's syndrome and a close association with tau pathology. Our findings indicate a macroscopic anatomic relationship between neuroinflammation and tau pathology. Although we cannot infer the causal direction in the relationship between pathological mechanisms, we speculate that microglial activation may be activated by an initial tau misfolding and contribute to tau pathology and propagation. The latter, in turn, may lead to further neuroinflammation, as previously suggested in AD (see review in ref. 41). Preclinical research suggests that microglial activation may precede the formation of tangles 47 and then promote the spreading of pathological tau. 48 Our findings suggest that the colocalization of neuroinflammation and tau pathology is an important pathogenetic mechanism in PSP, and both processes may be involved in defining PSP clinical severity. A better understanding of the interaction between the pathological substrates in PSP and its effects on disease progression may crucially contribute to improving patients' stratification and clinical trials. Specifically, our results encourage the application of [ 11 C]PK11195 and [ 18 F]AV-1451 PET as markers of colocalized pathological mechanisms in PSP to develop new targeting therapies and empower clinical trials.

Diseases
Research Network (DeNDRoN) for help with subject recruitment, and Drs Istvan Boros, Joong-Hyun Chun, and other WBIC RPU staff for the manufacture of the radioligands. We thank Avid (Lilly) for supplying the precursor for the production of [ 18 F]AV-1451 used in this study. We also thank Dr Thomas Cope and Dr Kamen Tsvetanov for discussion.

Financial Support
This study was co-funded by the National Institute for Health Research (