Assessment of 18F-PI-2620 as a Biomarker in Progressive Supranuclear Palsy

This cross-sectional study investigates the potential of novel tau radiotracer 18F-PI-2620 as a biomarker in patients with clinically diagnosed progressive supranuclear palsy.

PET image post-processing An 18 F-PI-2620 PET template was generated to allow inclusion of patients not eligible to an MRI (because of pacemakers, metal implants etc.). 20 randomly chosen datasets of PSP patients (five PSP-RS and two PSP-non-RS from Munich, three PSP-RS and one PSP-non-RS from Leipzig, two PSP-RS from New Haven), disease controls (two AD from Munich and one AD from New Haven, one PD and one MSA from Munich) and healthy controls (two from New Haven) who all had a T1w 3D MRI sequence were automatically processed by the PNEURO pipeline 3 to obtain 18 F-PI-2620 images in the Montreal Neurology Institute (MNI) space. Frames between 30 and 60 minutes p.i. were summed and scaled by the global mean, followed by averaging all 20 cases to an 18 F-PI-2620 template in the MNI space. All 90 dynamic 18 F-PI-2620 PET datasets were transformed to the MNI space using the non-linear brain normalization of the summed 30-60 min frames to the established 18 F-PI-2620 template. Automatized brain normalization settings in PMOD included nonlinear warping, 8 mm input smoothing, equal modality, 16 iterations, frequency cutoff 3, regularization 1.0, and no thresholding. The transformation was saved and applied to the full dynamic 18 F-PI-2620 PET datasets to minimize interpolation.

Visual read
Three expert readers independently assessed the 18 F-PI-2620 DVR maps in 3D mode using standardized settings (lower/upper DVR threshold 1.0/1.5; spectrum color, overlay on a MRI standard template in the MNI space with 50% transparency) as presented by the Papaya viewer online platform and blinded to the identity of the subject. Five PSP (three PSP-RS and two PSP-non-RS; all scanned in Munich after data lock in October 2019) and five different control cases (three scanned in Munich after data lock in October 2019 and two scanned in New Haven with not matching age) were used as a training-set (see eFigure 2). 20% randomly chosen cases of each diagnosis category were repeatedly presented to the readers to assess test-retest variability. Each reader had to score the overall pattern of the scan as positive or negative for a PSP-typical 18 F-PI-2620 binding. In particular, the reader was instructed to evaluate binding in putamen, globus pallidus, subthalamic nucleus, substantia nigra and the dentate nucleus and the final decision of positivity or negativity of a PSP-typical pattern was dichotomous for the entire scan. Readers were allowed to take AD-like cortical binding into consideration. The majority judgment from the three readers defined positivity and negativity for a PSP-like 18 F-PI-2620 PET scan. Fleiss κ was determined as a measure of intra-reader agreement, and Cohens κ as a measure of agreement between semiquantitative and visual analyses, and for test-retest evaluation.

eResults Tracer kinetics
Time-activity-ratio-curves between 0-60 minutes showed inverted U-shapes for both parts of the globus pallidus, the putamen, and the subthalamic nucleus, whereas a plateau or a slightly increasing shape was observed for the substantia nigra, the dorsal midbrain and the dentate nucleus. The frontal cortex target regions of Alzheimer's disease patients indicated increasing time-activity-ratios.

Visual read
The inter-reader agreement was moderate to substantial (Fleiss κ=.594, p<.001). The agreement between semiquantitative and visual classification was substantial (Cohens κ=.642, p<.001). Test-retest agreement was excellent (Cohens κ=.810, p<.001), with congruent negative reading in 20 (95%) scans and congruent positive reading in 29 (88%) scans. Representative basal ganglia slices of all subjects and slices throughout the brain of all PSP-non-RS subjects are provided in eFigures 7&8

eDiscussion PSP Phenotype
Our analysis entails a preliminary comparison of 18 F-PI-2620 binding in different PSP phenotypes. Keeping limited sample sizes among the more rare PSP subtypes in mind, we observed a roughly matching magnitude of tau-PET signal in different PSP phenotypes when compared with their amount of tau found in autopsy 4 . Furthermore, we found a positive 18 F-PI-2620 signal in more than half of the PSP-non-RS patients, potentially allowing early identification of underlying PSP pathology in these underdiagnosed cases 5 .

Frontal cortex
Findings related to 18 F-PI-2620 binding in the frontal cortex of PSP patients indicated differences among the different applied analyses in vivo and in vitro. In vitro autoradiography showed a clear and co-localized cortical 18 F-PI-2620 binding in two cases with high AT8-positive tau load in the frontal cortex. In contrast, there was no elevated 18 F-PI-2620 signal in the in vivo PET data in the PSP groups and healthy controls at the group level when using predefined frontal cortical regions of interest. By visual inspection of the PET data of single patients, we observed a clearly defined frontal cortex signal only in some PSP cases (see eFigure 4), but no relevant frontal cortex 18 F-PI-2620 binding in the majority of the investigated PSP patients. This was also reflected by the results of the semi-quantitative analysis, indicating elevations of DVR above the mv ± SD threshold in the dorsolateral and the medial prefrontal cortex in only three PSP cases. Interestingly, the PSP-F patients studied did not show a higher PET signal in frontal cortical target regions when compared to other PSP phenotypes (see eFigure 8). Nonetheless, it needs to be considered that the cortical tau load of the studied in vivo cases was likely below the tau load of (usually more advanced) autopsy cases. Thus, we speculate that the abundance of tau in this brain area might be too low for the in vivo detection threshold of the tracer. Partial volume effects due to pronounced atrophy of frontal cortical regions in PSP patients could be another explanation of the missing group differences. Negative correlations of the 18 F-PI-2620 signal with age in the dorsolateral and the medial prefrontal cortex in the limited sample of healthy controls may support this hypothesis as partial volume effects due to age-related volume loss would fit to the observed associations. Future studies need to address if individual standardized assessments (i.e. by 3-dimensional Z-score maps) or partial volume effect correction of frontal cortical 18 F-PI-2620 binding have a value for tau PET imaging in PSP.

Binding characteristics
Binding characteristics in subcortical PSP target regions appeared different when compared to previously reported cortical Alzheimer's disease (AD) binding 6 . Due to different 18 F-PI-2620 binding characteristics to AD (--log10 of the half maximal inhibitory concentration: 8.5±0.1) and PSP (--log10 of the half maximal inhibitory concentration: 7.7±0.1) brain tissue 7 , it might be the case that even low deposition of 3/4R tau in the basal ganglia of AD could cause a higher PET signal as strong 4R tau deposition in the basal ganglia of PSP cases. Except for the substantia nigra and the dorsal midbrain we observed decreasing time-activity ratio curves until 60 minutes scan time after a peak within the first 30 minutes. The dentate nucleus even showed differences against healthy controls only in the first 30 minutes but revealed converging binding ratios in the later acquisition phase. In contrast, cortical AD timeactivity ratio curves were reported stable or even increasing with time until 180 minutes 6 . A potential factor explaining the different uptake dynamics could be the already discussed lower affinity of 18 F-PI-2620 to 4-repeat (4R) PSP tau when compared to 3/4R AD tau. Consequently, there might be more dissolution of 18 F-PI-2620 from 4R tau in contrast to 3/4R tau. Taking the fast wash out of the tracer into consideration this could explain targetto-reference-tissue uptake differences in the late PET image acquisition phase. Second, the subcortical localization of PSP target regions in contrast to cortical AD target regions could also contribute to differences in tracer delivery. A direct comparison of cortical 18 F-PI-2620 tau binding between 4R and 3/4R tauopathies might allow elucidating this phenomenon, and the contrast between β-amyloid positive and β-amyloid negative patients with corticobasal syndrome could serve for this purpose 8 . Differences in affinity of 18 F-PI-2620 to different tau isoform is also supported by our autoradiography findings, showing a discernible but lower signal in the basal ganglia of PSP tissue when compared to earlier reported autoradiography signal in cortical AD tissue 7 .

eFigure. Definition of target and reference regions
Definition of target (red) and reference (blue) regions in the Montreal Neurology Institute (MNI) space. Outlined volumes-of-interest are projected upon an MRI template. The Atlas of the basal ganglia 9 was used for definition of most subcortical volumes-of-interest (globus pallidus externus and internus, putamen, subthalamic nucleus, substantia nigra). The dorsolateral prefrontal cortex and the medial prefrontal cortex were defined by the Brainnetome atlas 10,11 . The cerebellum, the dentate nuclues and the dorsal midbrain were defined by the Hammers atlas 12 . For delineation of subcortical volumes-of-interest by the Atlas of the basal ganglia, a filter of 9x9x10 mm and a border threshold of 0.1 were applied. The standard grey matter threshold of 0.3 in PMOD was set for definition of voxels in the dorsolateral and medial prefrontal cortices by the Hammers atlas. Superior and posterior layers (1.5 cm) of the cerebellum were excluded manually in the MNI space. The dentate nucleus and the central cerebellar white matter were also excluded. Ventral parts of the midbrain and the lower parts of the brainstem were excluded for definition of the dorsal midbrain. GPe = globus pallidus externus; GPi = globus pallidus internus; PUT = putamen; STN = subthalamic nucleus; SN = substantia nigra; DMB = dorsal midbrain; DN = dentate nucleus; DLPFC = dorsolateral prefrontal cortex; MPFC = medial prefrontal cortex; CBL = cerebellum controls; α-syn = probable α-synucleinopathies; SUVr = standard-uptake-value-ratio; min = minutes eFigure 6. Agreement between dynamic and static imaging Agreement between distribution volume ratios (DVR) and standardized-uptake-value-ratios (SUVr) from 30 to 60 minute post injection. GPe = globus pallidus externus; GPi = globus pallidus internus; PUT = putamen; STN = subthalamic nucleus; SN = substantia nigra; DMB = dorsal midbrain; DN = dentate nucleus; DLPFC = dorsolateral prefrontal cortex; MPFC = medial prefrontal cortex; MRTM2 = multilinear reference tissue modelling 2. R 2 derive from the correlation of the full dataset of 90 subjects eFigure 9. Regional associations of 18F-PI-2620 binding with age, disease severity and disease duration