Tau-PET and in vivo Braak-staging as a prognostic marker in Alzheimer’s disease

INTRODUCTION: Tau pathology in Alzheimer’s disease tracks clinical status more closely than beta-amyloid. Thus, tau-PET may be a promising prognostic marker for cognitive decline. Here, we systematically compared tau-PET and Braak-staging vs. amyloid-PET as predictors of cognitive decline. METHODS: We included 396 cognitively normal to dementia subjects with 18 F-Flutemetamol/ 18 F-Florbetapir-amyloid-PET, 18 F-Flortaucipir-tau-PET and ~2-year cognitive assessments. Annual cognitive change rates were calculated via linear-mixed models. We determined global amyloid-PET, global tau-PET, and tau-PET-based Braak-stage (Braak 0 /Braak I+ /Braak I-IV+ /Braak I-VI+ /Braak atypical+ ). In bootstrapped linear regression, we assessed whether tau-PET outperformed amyloid-PET in predicting cognitive decline. Using ANCOVAs, we tested whether later Braak-stage predicted accelerated cognitive decline and determined Braak-stage-specific conversion risk to MCI or dementia. RESULTS: Global tau-PET was a better predictor of cognitive decline than global amyloid-PET (p<0.001). Advanced Braak-stage was associated with faster cognitive decline (p<0.001) and elevated clinical conversion risk. DISCUSSION: Tau-PET and Braak-staging show promise for predicting patient-specific risk of clinical AD progression. and an average of 2-year follow-up assessments of global cognitive and memory performance. We tested whether i) global measures of tau-PET are a better predictor of future cognitive decline and clinical AD progression than global measures of amyloid-PET and whether ii) advanced tau-PET-assessed Braak-stage predicts faster cognitive decline and risk of clinical AD progression. Conversion were using c 2 All analyses For (adjusted 0.05/3=0.017). Post-hoc Tukey were to ANCOVAs.


INTRODUCTION
Beta-amyloid (Ab) and tau pathology are pathognomonic brain changes of Alzheimer's disease (AD), ensuing neurodegeneration, cognitive decline and dementia. [1][2][3] The development of in vivo Ab and tau biomarkers has greatly facilitated diagnosing AD, 2,4 however, reliable prognosis of AD-related cognitive decline remains a critical yet unmet challenge. Tau biomarkers may offer particular potential for individualized risk prediction of AD progression. 5 While Ab plaques develop 20-30 years prior to symptom onset, 1,2 tau pathology emerges much closer to symptom onset as revealed by positronemission-tomography (PET), 6,7 biofluid biomarkers, 8 and post-mortem examinations in AD patients. 9 Supporting the view that tau is a key determinant of cognitive impairment in AD, PET-based tau assessments show a stronger association with cross-sectional cognition 5 and retrospective cognitive changes 10 than amyloid-PET. Similarly, post-mortem studies in AD found that the extent of neurofibrillary tau tangles showed a stronger association with ante-mortem clinical status than Abplaques. 11 Thus, tau is potentially more informative for forecasting future cognitive decline and clinical progression than Ab. Also, the spatio-temporal progression of tau pathology follows a relatively stereotypical spreading pattern, which is closely related to disease stage. 12,13 Specifically, tau emerges first in the entorhinal cortex, before spreading across the temporal lobe, association cortices, and eventually primary sensorimotor and visual cortices, as summarized in the Braak-staging scheme of progressing tau pathology. 12,13 Thus, tau-PET-based Braak-staging may be clinically useful for patientspecific risk estimation of future cognitive decline and clinical AD progression.
To address this, we systematically tested whether tau-PET and in vivo Braak-staging outperform established amyloid-PET markers as predictors of future cognitive decline and clinical AD progression.
To this end, we included 396 subjects ranging from cognitively normal to AD dementia, characterized by baseline 18 F-Flutemetamol/ 18 F-Florbetaben amyloid-PET, 18   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint

Participants
We included 396 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
Beyond ADNI inclusion criteria, the current study required availability of amyloid-PET (either 18F- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint we averaged SUVRs across cortical Freesurfer ROIs, excluding the cerebellum, hippocampus, thalamus, and basal ganglia (i.e. typical regions of 18 F-Flortaucipir off-target binding) following a previously described approach. 26 For Braak-stage specific tau-PET, we applied in vivo Braak-staging that allows application of the post-mortem established Braak tau staging system to tau-PET imaging (see Fig. 1B Subjects were classified as Braak 0 /tau -(n=293), when all Braak ROIs had an SUVR below 1.3 (Fig. 1A).
The Braak-staging scheme is illustrated in Fig. 1C. Note that exploratorily altering the tau-PET threshold between 1.2-1.4 yielded congruent results with those presented in the manuscript.

Statistical analysis
Differences between diagnostic groups were assessed using ANOVAs for continuous and chi-squared is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint controlled for age, sex, education, diagnosis (CN, MCI, dementia), and baseline performance on the respective cognitive test (i.e. MMSE, ADAS13, or ADNI-MEM). When assessing global tau-PET SUVR as a predictor of longitudinal cognitive changes, we ran additional models controlling also for global amyloid-PET (i.e. CL), to ensure that tau-PET explained additional variance in cognitive decline when controlling for amyloid-PET. To determine the variance that amyloid-PET (i.e. CL) or tau-PET explained in longitudinal cognitive changes, we calculated partial R 2 values for either amyloid-or tau-PET as predictors of cognitive changes. In order to assess whether tau-PET was a better predictor of future cognitive changes than amyloid-PET, we performed bootstrapping, repeating the above described regression models on 1000 bootstrapped samples and compared confidence intervals (CI) of amyloid-PET and tau-PET of the resulting partial R 2 values using paired t-tests. Next, we tested whether more advanced Braak-stage was associated with increased risk of cognitive decline. To this end, we ran ANCOVAs, using annual cognitive change rates as dependent variables (i.e. MMSE, ADAS13, and ADNI-MEM) and Braak-stage group (i.e. Braak 0 , Braak I+ , Braak I-IV+ , Braak I-VI+ , Braak atypical+ ) as independent variable, controlling for age, sex, education, diagnosis, global amyloid-PET, and the baseline score of the respective cognitive test. Standardized differences in cognitive decline between Braak-stage groups were determined using Cohen's d. Further, we exploratorily assessed whether Braak-stage specific associations with cognitive decline were consistent across diagnostic groups, i.e. we repeated the above described analyses stratified by diagnostic groups (CN, MCI, dementia). Lastly, we determined for each Braak-stage group the risk of clinical conversion, defined as the relative risk of a change in diagnosis from CN to MCI/dementia or from MCI to dementia during follow-up. Note that subjects with a baseline diagnosis of dementia were excluded from this analysis, since no further conversion can be diagnosed in these patients. Conversion rates were compared using c 2 tests.
All analyses were computed using R statistical software (r-project.org 29 ). For each analysis on cognitive measures (MMSE, ADAS13, ADNI-MEM), Bonferroni correction was applied (adjusted alpha level: 0.05/3=0.017). Post-hoc Tukey tests were applied to ANCOVAs. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

Data availability statement
All data used in this manuscript are publicly available from the ADNI database (adni.loni.usc.edu) upon registration and compliance with the data use agreement. The data that support the findings of this study are available on reasonable request from the corresponding author.
. CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

Global tau-PET is a better predictor of future cognitive decline than global amyloid-PET
First, we tested whether global tau-PET at baseline is a better predictor of future cognitive decline than is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint mean=0.030; tau-PET: CI=0.075-0.078, mean=0.077), providing non-parametric support of a significant difference between amyloid-PET and tau-PET-derived partial R 2 distributions.

Advanced Braak-stage is associated with faster cognitive decline
Next, we tested whether more advanced Braak-stage at baseline was associated with faster subsequent cognitive decline using ANCOVAs. We found the expected association between more advanced Braak-   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  (n=1/7) (Fig. 4). Again, these results illustrate that spatial expansion of tau pathology is strongly associated with the risk of future cognitive decline, while amyloid-PET is prognostically less conclusive. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

Advanced Braak-stage is associated with higher conversion risk
The copyright holder for this this version posted February 8, 2021. For our main finding, we show that tau-PET clearly outperforms amyloid-PET in forecasting future cognitive decline. This result pattern is in agreement with several previous studies suggesting a close association between tau accumulation and the development of cognitive deficits in AD. 6,7,9,30 A recent longitudinal PET-study in preclinical AD showed that increased Ab-burden mediates tau accumulation which in turn promotes cognitive decline. 30 This is congruent with the amyloid cascade hypothesis, suggesting that Ab is the initial trigger of pathological tau in AD 31,32 preceding symptom onset by decades, 1,2 while tau is the actual driver of neurodegeneration and cognitive decline. 33 Therefore, a focus on Ab biomarkers alone is likely insufficient for reliable prediction of clinical AD trajectories. This has important implications for clinical trial design in AD, since markers of tau pathology could be critical in addition to Ab-markers for matching progression risk among placebo vs. verum groups. Further, Ab and tau accumulation show striking differences in their spatial accumulation patterns: While Ab accumulates rather globally, 34 tau spreads in a relatively stereotypical spatio-temporal pattern 35 that is closely associated with clinical status. 13 By applying Braak-staging to tau-PET, we could confirm that more advanced Braak-stage was associated with gradually accelerated future cognitive decline.
Specifically, Braak 0 individuals (i.e. without evidence of elevated tau pathology) showed slowest annual . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. Several caveats should be considered when interpreting our results. First, 18 F-Flortaucipir shows considerable off-target binding in the hippocampus and basal ganglia, which may confound the assessment of tau pathology. 43 Therefore, we excluded regions which are known to be affected by off- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint target binding. However, influences of unspecific binding remain possible, hence our findings await further replication once sufficient data with second-generation tau-PET tracers (i.e. with a better offtarget binding profile) are available. Second, we classified PET using pre-established cut-offs, which is of high clinical use but arbitrarily binarizes a continuous biological process (i.e. Ab or tau accumulation). Although the currently used tau-PET 1.3 SUVR cut-off was selected based on recommendations for tau-PET, 26,28 we exploratory repeated our analyses by slightly altering the tau-PET SUVR thresholds (e.g. ranging between 1.2-1.4), revealing a consistent result pattern. Still, PET cutoffs may be replaced in the future by more advanced methods such as gaussian-mixed model-based transformation of tau-PET SUVRs to tau positivity probabilities. 44,45 A third limitation relates to the individual spatial variability in tau deposition patterns. Previous work found that the spreading patterns of tau pathology can be spatially heterogeneous across individual patients. 7,44 To address this, we applied a relatively simple tau staging scheme (i.e. Braak I+ , Braak I-IV+ , Braak I-VI+ , Braak atypical ) that does not take into account asymmetry in tau deposition 7 or fine-grained regional differences in the distribution of tau pathology. Only ~8% of subjects deviated from this staging-scheme, suggesting that the currently employed Braak-staging system is applicable to a majority of typical AD patients that are included in ADNI. Still, we caution that this Braak-staging scheme may not be applicable to patients with atypical AD, characterized by highly heterogeneous tau deposition patterns. Thus, further studies are necessary to determine the predictive accuracy of tau-PET for future cognitive decline in these rare atypical AD cases. 7 Finally, PET imaging comes with high costs, radioactive burden, and may not be available for each patient or within each country. Therefore, current investigations of plasma tau markers 46 are of high clinical importance for widespread screening for tau pathology. Plasma screening may be used to select subjects eligible for tau-PET and in vivo Braak-staging, which may allow more accurate risk prediction than single plasma-derived tau measures.
Together, we show that tau-PET outperforms amyloid-PET in predicting cognitive decline and clinical AD progression, supporting the notion that tau pathology is a key driver of cognitive decline in AD. 47 Importantly, we found that regional tau staging allows accurate risk estimation of future cognitive changes, which can be critical to stratify risk groups for clinical trials. From a clinical perspective, our is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint findings suggest that in vivo tau-PET-based Braak-staging may be a valuable tool to identify subjects at imminent risk of cognitive decline.

Acknowledgement/Funding:
The study was funded by grants from the LMU (FöFoLe, 1032, awarded to NF), the Hertie foundation  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted February 8, 2021. ; https://doi.org/10.1101/2021.02.04.21250760 doi: medRxiv preprint