In Vivo 18F-Flortaucipir PET Does Not Accurately Support the Staging of Progressive Supranuclear Palsy

Visual Abstract

Pr ogressive supranuclear palsy (PSP) is a severe neurodegenerative disorder resulting in diverse clinical phenotypes with restricted eye movements, an akinetic-rigid syndrome, falls, and cognitive and behavioral deficits (1). The neuropathology of PSP is characterized by intracellular aggregates of 4-repeat tau in neurons and glia (2)(3)(4)(5); these aggregates are distributed in a progressive sequence starting in the substantia nigra, globus pallidus and subthalamic nucleus, then pons, striatum and the precentral gyrus in the cerebral cortex, before reaching the cerebellum and frontal cortex (6). Later, the neuroglial pathology might extend to the occipital cortex (7).
Here, we test whether regional binding of the radioligand 18 Fflortaucipir (also known as 18 F-AV-1451), quantified using nondisplaceable binding potential, can be used to replicate the staging of PSP pathology in vivo. We validated the staging in 2 ways: correlation with clinical severity at the time of 18 F-flortaucipir PET and neuropathologic staging of a subset of participants postmortem.

Participants
We recruited 42 patients with a clinical diagnosis of probable PSP using Movement Disorder Society PSP 2017 criteria (1) ( Fig. 1A) (supplemental materials are available at http://jnm.snmjournals.org). The striatum and subthalamic nucleus were excluded because of 18 F-flortaucipir off-target binding or challenges in defining the PET signal. Regional values were quantified using a modified version of the n30r83 Hammersmith atlas (http://brain-development.org/brainatlases/adult-brain-atlases/adult-brain-maximumprobability-map-hammers-mith-atlas-n30r83-inmni-space/), which includes parcellation of the brain stem and cerebellum, and a basis function implementation of the simplified reference tissue model (19), with cerebellar cortex gray matter as the reference region. Before kinetic modeling, regional PET data were corrected for partial-volume effects from cerebrospinal fluid by dividing the regional PET value by the mean regional gray matter plus white matter fraction determined from Statistical Parametric Mapping (SPM12, https://www.fil.ion.ucl.ac.uk/spm/) segmentation. Left and right regional nondisplaceable binding potential values were averaged bilaterally. Using regional mean and SD values from controls, we calculated w-scores (zscores adjusted for the effect of covariates, Supplemental Fig. 1B), accounting for phenotypic and systematic differences, such as age and Step 1: in vivo stages are defined with cumulative evidence of absence (region 5 0) or presence (region 5 1) of pathology in each of 5 regions considered, as defined by region-specific thresholds (regional w-score . threshold 5 1; regional w-score # threshold 5 0). Step 2: in vivo substages are defined within each step 1 stage, considering 3-level pathology severity scale (0 5 none; 1 5 mild/moderate pathology; 2 5 moderate/severe pathology). Regions: cerebellum (CER; white matter and dentate nucleus), middle frontal gyrus (FR), globus pallidus (GP), and occipital lobe (OCC; lingual gyrus and cuneus). Data-Driven Severity Thresholds. To quantify pathology severity in each region, we used a conditional inference tree analysis to define in a data-driven manner region-specific 18 F-flortaucipir binding thresholds of w-scores, entering both patients and controls in the model. This method is similar to that used previously for imaging-based pseudo-Braak staging of Alzheimer disease (9). Specifically, region-specific thresholds were identified using nonparametric binary recursive partitioning with the function "ctree" in R (v. 4.0.0, R Core Team -R Foundation for Statistical Computing) and running this tree analysis on w-scores for each region separately. Using these region-specific thresholds, we assigned binary severity scores to individual regional w-scores (w-score # regional threshold: 0 or absent; w-score . regional threshold: 1 or present).
In Vivo Staging. First, using the staging system described by Kovacs et al. (7), which is based on cumulative and progressive pathology severity, we evaluated the combination of absent/present values across all 4 regions to assign each participant to stages I/II, III/IV, or V/VI (step 1 on Fig. 1). Second-in an explorative analysis-within each stage defined in the previous step, a 3-point pathology severity system was applied to each region (w-score # regional threshold: absent, coded as 0; w-score . regional threshold: mild/moderate pathology, coded as 1; w-score . 2 times the threshold: moderate/ severe pathology, coded as 2), and 1 of the 6 stages was assigned accordingly (stages I-VI; step 2 on Fig. 1). We repeated these staging analyses with a second analytic approach, using a preselected number of SD values from region-specific nondisplaceable binding potential control means to define pathology severity (Supplemental Fig. 2). ANOVA was applied to analyze differences among means of disease severity (PSPRS) between stages.

Postmortem Diagnosis and Staging Based on Immunohistochemistry
Tissue blocks of the left hemisphere were sampled according to National Institute of Neurological Disorders and Stroke standard guidance for neurodegenerative diseases from the brain stem, subcortical, and cortical areas. These were evaluated for the initial pathologic diagnosis of PSP (hyperphosphorylated tau; AT8, MN1020; Thermo Scientific, possible concomitant pathologies of amyloid b (clone 6F/3D, M0872; Dako), a-synuclein (SA3400; Enzo Life Sciences), and TDP-43 (TIP-PTD-P02; Cosmo Bio Co. Ltd.); and vascular pathology. Using the previously described staging scheme (7,8), we evaluated neuronal and oligodendroglial tau pathology in the globus pallidus, subthalamic nucleus, and cerebellar white matter and dentate nucleus and astrocytic tau pathology in the striatum, middle frontal gyrus, and occipital cortex. The regional cytopathologies were rated on a 4-level system (none, mild, moderate, and severe) using the guidelines proposed by Briggs et al. (8). In vivo staging results with both data-driven and SD approaches were compared with postmortem staging in the 9 patients who donated their brain.

RESULTS
The conditional inference tree analysis identified region-specific pathologic thresholds of 18 F-flortaucipir binding for the globus pallidus (w-score, .0.795), cerebellar white matter (w-score, .0.783) and dentate nucleus (w-score, .0.845), and middle frontal gyrus (w-score, .1.416). For the occipital lobe, the analysis did not identify the threshold, so we used 1.645 as the w-score critical value (P 5 0.05). A simple set of decision rules (Fig. 1) enabled plausible Kovacs stages to be estimated in 41 patients ( Fig. 2A): 10 patients were classified in stage I/II because of increased 18 F-flortaucipir binding limited to the globus pallidus, 26 were classified in stage III/IV because of additional increased 18 F-flortaucipir binding in the frontal or cerebellar regions, and 5 were classified in stage V/VI because of additional increased 18 F-flortaucipir binding in the occipital lobe; 1 patient could not be classified because no increased binding was found in the globus pallidus. The explorative substaging (6 stages) identified 2 patients in stage I (mild/moderate pathology in the globus pallidus), 8 in stage II (moderate/severe pathology in the globus pallidus), 9 in stage III (mild/moderate pathology in the frontal lobe or cerebellum), 17 in stage IV (moderate/severe pathology in the frontal lobe or cerebellum), and 5 in stage V (mild/moderate pathology in the occipital lobe). When the same approach was applied to controls, 31, 5, 1, and 2 participants were classified in no stage, stage I, stage II, and stage III, respectively. Four patients ( Fig. 2A,  patients 6, 35, 36, and 39) showed an atypical severity pattern that was discordant with the description of Kovacs et al. (7).
Across all patients, the estimated in vivo stages did not relate to clinical severity (P . 0.05 in an ANOVA) (Figs. 2B and 2C). In 8 of the 9 patients who donated their brains, pathology stage as determined by in vivo 18 F-flortaucipir PET was less than or equal to that determined postmortem (Fig. 3). In vivo staging and postmortem staging were not significantly correlated (Spearman r, 0.168; P 5 0.67). Correlation analyses were also used to test the residuals of each staging variable (in vivo and postmortem staging) after regressing out clinical severity (PSPRS scores) and the interval from PET to time of death. The correlation was not statistically significant (Spearman r, 0.150; P 5 0.70). Figure 4 shows examples of 18 F-flortaucipir nondisplaceable binding potential maps and corresponding postmortem staining data for patients who were classified in stage II (patient 4) and stage IV (patient 26) by both in vivo staging and postmortem staging.

DISCUSSION
The principal finding of the present study was that 18 F-flortaucipir PET does not provide accurate in vivo staging corresponding to neuropathologic staging for PSP. The nominal stage derived from 18 F-flortaucipir PET did not correlate with disease severity or relate to staging postmortem.
As a result of the data-driven in vivo staging system, compared with controls, we observed higher 18 F-flortaucipir binding in the globus pallidus in all but 1 patient, with a few patients showing increased 18 Fflortaucipir binding in the occipital cortex ( Fig. 2A). This regional distribution of 18 F-flortaucipir binding was in line with the pathologic description of PSP and with what was previously described for 18 Fflortaucipir in PSP (13,17,18,20). Whereas the 18 F-flortaucipir binding patterns allowed us to nominally apply PSP pathology staging in vivo, the in vivo staging was not systematically predictive of pathology staging postmortem. As expected because of the time interval between the PET scan and autopsy, in 8 of 9 cases with autopsy, the individual in vivo staging was less than or equal to the postmortem staging. However, 4 patients who were labeled as stage IV in vivo were then classified in 4 different stages postmortem (Fig. 3). Neither clinical severity nor the time interval between the PET scan and death was useful for predicting the individual postmortem stage from in vivo staging.
The number of patients with a positive signal for 18 F-flortaucipir in the cerebellum (n 5 29) exceeded the number of patients with a positive result for frontal 18 F-flortaucipir binding (n 5 10). Although this finding might reflect earlier involvement of the cerebellum in our cohort, regional differences in the density of tau aggregates and predominant cytopathologies could contribute to regional differences in tracer retention (11,13,21)-for example, neuronal and oligodendroglial tau predominates in the cerebellum, whereas astrocytic tau predominates in cortical regions.
Off-target binding for 18 F-flortaucipir is well characterized, but this problem alone would still leave open the possibility of quantifying tau pathology in areas without significant monoamine oxidase levels or neuromelanin, such as the cerebellum and medial frontal cerebral cortex (22). However, recent PET-toautopsy correlational studies suggested that 18 F-flortaucipir PET does not reliably correspond to postmortem tau pathology in non-Alzheimer tauopathies (13,23). This finding suggests that 18 Fflortaucipir lacks sensitivity in non-Alzheimer tau pathology. This characteristic might explain the underperformance of this tracer in defining an in vivo classification that systematically aligns with postmortem staging. Next-generation tau tracers might prove to be more useful for tracking in vivo PSP pathology progression because of a combination of good affinity for 4-repeat tau and lower off-target binding to monoamine oxidases (i.e., 18 F-PI-2620 (24)). However, evidence from PET-to-autopsy studies for these new ligands is needed, together with better segmentation and signal detection from small regions. These features would be particularly important for early-stage pathology detection and the classification of stage I/II of the system of Kovacs et al. (7).

CONCLUSION
We conclude that 18 F-flortaucipir PET is not a useful marker of the neuropathologic stage in PSP, despite increased binding and some regional concordance between tau pathology and ligand binding. This analytic approach, seeking to mirror in vivo neuropathology staging with PET-to-autopsy correlational analyses, could be applied to test next-generation tau PET tracers. However, comparisons with postmortem data are also required. Rowe serves as an associate editor to Brain and is a nonremunerated trustee of the Guarantors of Brain, Darwin College, and the PSP Association (United Kingdom). He provides consultancy to Asceneuron, Astex, SVHealth and Curasen, and UCB and has received research grants from AZ-Medimmune, Janssen, and Lilly as industry partners in the Dementias Platform U.K. John T. O'Brien has received honoraria for work as Drug and Safety Monitoring Board chair or member for TauRx, Axon, Eisai, and Novo Nordisk; has acted as a consultant for Biogen and Roche; and has received research support from Alliance Medical and Merck. Maria Grazia Spillantini is in the Scientific Advisory Board of the Tau Consortium supported by the Rainwater Charitable Foundation. Caroline H. Williams-Gray has received honoraria from Lundbeck and Profile Pharma Ltd. and consultancy fees from Modus Outcomes and Evidera, Inc./GlaxoSmithKline. Unrelated to this work, Timothy Rittman has received honoraria from Biogen, Oxford Biomedica, and the National Institute for Health and Clinical Excellence (NICE). No other potential conflict of interest relevant to this article was reported.