Spatial Extent of Amyloid-β Levels and Associations with Alzheimer’s Disease Biomarkers

Objective: To investigate the biological and clinical correlates of Aβ spatial extent deposition levels in cognitively unimpaired older adults. Methods: We included cognitively unimpaired older adults from three cohorts, totalling 529 participants (PREVENT-AD, n=129; ADNI, n=400 and HABS, n=288) who underwent Aβ positron emission tomography (PET). We used Gaussian-mixture models to identify region-specic thresholds of Aβ positivity in seven brain regions prone to early Aβ accumulation. Individuals were classied as having “widespread” Aβ deposition if they were positive in all seven regions, “regional” Aβ deposition if they were positive in one to six regions, or Aβ negative if negative in all regions. We compared demographics, genetics, tau-PET binding, and cognitive performance and decline between the three groups. Results: In all cohorts, most participants with regional Aβ-PET binding did not meet the cohort-specic criteria for Aβ-positivity (79% for PREVENT-AD, 57% for ADNI, and 100% for HABS). Regional Aβ groups had normal baseline cognition and relatively normal tau-PET binding, but a greater proportion of APOE ε4 carriers, decreased CSF Aβ 1-42 levels, and greater amount of longitudinal Aβ-PET binding accumulation (only available in ADNI and HABS) when compared with the Negative Aβ groups. Widespread Aβ groups had lower baseline cognitive performance (PREVENT-AD only), faster cognitive decline (all cohorts) and greater amount of longitudinal tau binding than the other groups (only available in ADNI and HABS). Conclusions: Individuals with regional Aβ deposition might be the best candidate for preventive trials since they do not yet have widespread tau and cognitive decline. Widespread levels of Aβ seem to be needed for tau spreading. testing


Introduction
Amyloid-beta (Aβ) and tau are the main pathological hallmarks of Alzheimer's disease (AD). The deposition of these pathological proteins is a continuous process that starts decades before the onset of AD symptoms 1,2 . While tau deposition may initiate prior to Aβ accumulation 3 , it is widely held that Aβ pathology is required to make tau spread out of the temporal lobe and start the pathological cascade leading to AD dementia, making it an ideal target for clinical trials 4,5,6,7 . Several clinical trials have now successfully reduced brain Aβ without slowing down AD clinical progression [8][9][10] . While the role of Aβ in the pathological cascade of AD has been questioned based on large-scale clinical trial failures 11 , one could argue that Aβ needs to be targeted before the spread of tau pathology 12 . The appropriate timing likely corresponds to a stage where there is a limited amount and spreading of Aβ pathology, hence making it challenging to identify 13,14 .
Accumulation of Aβ starts in a few distinct brain regions almost simultaneously, which makes it possible to characterize early regional Aβ deposition in vivo using positron emission tomography (PET) imaging, before it rapidly evolves to widespread distribution [15][16][17] . In general, most studies have used global brain load to classify individuals with intermediate or high levels of Aβ 15,18,19 . We took advantage of the spatial distribution of Aβ deposition to identify different groups of cognitively unimpaired individuals based on the extent of Aβ tracer uptake with the objective of identifying early Aβ deposition. In three cohorts of cognitively unimpaired older adults, including one with a family history of AD dementia, we sought to investigate the characteristics of the different groups based on various AD markers  (4) performed within education-adjusted norms on Logical Memory-delayed recall. Participants who had a score of 5 or more in Hachinski, history of stroke with residual de cits, and history of intercranial hemorrhage were excluded from the study. Data were obtained from the HABS data release 2.0 in October 2020 via habs.mgh.harvard.edu.
Note that all participants from the three different cohorts were cognitively normal at the time of Aβ PET to be included in the study.
Standard protocol approvals, registrations, and patient consents. All PREVENT-AD participants were fully briefed and gave their explicit consent for participation using procedures and consent forms approved by the Institutional Review Board of the McGill University Faculty of Medicine. Data collection and sharing in ADNI were approved by the Institutional Review Board of each participating institution and written informed consent was obtained from all participants. Participants from the HABS cohort provided written informed consent prior to study procedures, which used protocols approved by the Partners Healthcare institutional review board.

b. Neuropsychological Evaluation
In the three cohorts, participants underwent cognitive testing annually. We analyzed both baseline (corresponding to the time of Aβ PET scan) and longitudinal cognitive performance.
In PREVENT-AD, the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) 23  In ADNI, we used the available four composite scores re ective of memory, executive functions, language, and visuospatial functioning that have been previously described 24,25 . Longitudinal cognitive assessment was available for 393 (98%) subjects, with a median follow-up time of 6 [IQR: 1,14] years.
In HABS we used the preclinical Alzheimer's cognitive composite (PACC5) that is a composite score including memory, executive function and semantic processing 26 that was available through the HABS data release. All participants had a longitudinal cognitive assessment, with a median follow-up time of 6 [IQR: 1,9] years. c. APOE Genotyping Genomic DNA was extracted from whole blood, and the apolipoprotein E (APOE) genotype was determined 27 . The same procedure was done for all cohorts participants at their baseline visit.
Participants were classi ed as APOE ε4 carriers (i.e., those who had at least one ε4 allele) or noncarriers.

d. Cerebrospinal Fluid Biomarkers
In PREVENT-AD, a subsample of 77 participants underwent a lumbar puncture, with a maximum of up to 2 years before PET (mean difference to PET scan 10.42 ± 8.38 months). Cerebrospinal uid samples (CSF) were collected in the morning after an overnight fast and stored in cryovial tubes at −80°C. CSF Aβ 1-42 , p-tau 181 (phosphorylated at threonine 181) and total tau levels were assayed in duplicate with the INNOTEST ELISA (Fujirebio, Ghent, Belgium) 20,28 .
In ADNI, 276 participants had CSF biomarkers available, based on a two-year interval between PET and CSF (mean time from PET scan 0.40 ± 0.75 months). CSF samples were frozen within 1 hour after collection and were shipped overnight frozen on dry ice to the Penn AD Biomarker Fluid Bank Laboratory. Aliquots of 500 μL were stored in polypropylene tubes at −80°C. CSF Aβ 1-42 and p-tau 181 were measured using Elecsys immunoassays 29 . Acquisition of the multicentric MRI and PET imaging data in ADNI has been reported previously and is described in detail at adni.loni.usc.edu/methods/. Brie y, Aβ PET scans were acquired using Florbetapir ([18F] AV45) during a 50-to-70-minute interval following a 370 MBq bolus injection (≈10 mCi) and FTP scans were acquired during a 75-to-105-minute interval following a 370 MBq bolus injection (≈10 mCi). T1-weighted structural MRI data were acquired on 3T scanning platforms using sagittal 3D magnetization-prepared rapid gradient-echo sequences. The T1 sequence was the same as for the PREVENT-AD cohort. A subsample of 176 participants (44%) underwent tau-PET scans. Tau-PET was added later on in the course of the study, around 2016, and thus most tau scans were acquired 5 years after the Aβ-PET (median delay 5 [IQR: 0, 8] years in ADNI).
In HABS, PET data were acquired as described previously 22,30 . Aβ PET scans were acquired using PIB ( 11 C Pittsburgh Compound B) during a 60-minute dynamic acquisition starting directly after the injection and FTP scans were acquired from 80-100 minutes after a 9.0 to 11.0 mCi bolus injection. MRI scans were performed on a 3T Tim Trio (Siemens) with a 12-channel phased-array head coil. The imaging measures were typically collected every two years (mean delay between FTP and PIB scans 3.5 months).

f. PET Processing
In all cohorts, T1-weighted MRI were processed using FreeSurfer (version 5.3 or 6) and parcellated according to the Desikan-Killiany atlas 31 . PREVENT-AD PET images were processed with a standard, inhouse pipeline (available on Github: https://github.com/villeneuvelab/vlpp). Brie y, the 4D PET images were realigned, averaged, and registered to the corresponding T1-weighted MRI. Images were then masked to exclude CSF signal and smoothed with 6 mm 3 Gaussian kernels. Standardized uptake value ratios (SUVRs) were computed by dividing the tracer uptake in each voxel by the uptake in the whole cerebellum gray matter for NAV scans 32 and the inferior cerebellum gray matter for FTP scans 33 .
PET images for ADNI went through standardized preprocessing steps in order to increase data uniformity across the multicentric data acquisition 34 . Brie y, Florbetapir-PET frames were co-registered, averaged, reoriented into a standardized image and voxel size, and smoothed to produce a uniform resolution. FTP frames were co-registered and resliced to the structural MRI closest in time to the FTP-PET. The cerebellum gray matter was used as the reference region for Florbetapir scans and the inferior cerebellum gray matter was used for FTP scans. The Aβ (2019-12-04 version) and tau (2020-02-04) regional SUVR data were downloaded from the ADNI database.
In HABS, following the PET image acquisitions, a mean image was created (for PIB -the rst 8-minute post-injection), and PET images were co-registered to the corresponding T1-weighted MRI with 6 DoF rigid body registration using spm_coreg from the SPM12 package. Bilateral cerebellum grey matter was used as the reference region for SUVR measurements.
Regions of interest regarding Aβ are described subsequently. For tau-PET, SUVR from six bilateral regions that represent early tau-PET deposition were investigated, i.e. entorhinal cortex, amygdala, fusiform, parahippocampal, inferior temporal, and middle temporal cortex 35 . The hippocampus was not included given the off-target binding spillover from the choroid plexus 36,37 .

g. Global Threshold of Aβ Positivity
Each cohort had an available global threshold described previously to categorize participants into Aβ positive and negative. All such thresholds were derived from the average SUVR of lateral and medial frontal, cingulate, parietal and lateral temporal regions. In PREVENT-AD, the NAV threshold for positivity was 1.37 35 . In ADNI, the Florbetapir threshold was 1.1 38,39 , and the PIB threshold in HABS was 1.19 40,41 .

h. Regional Thresholds of Aβ Positivity
Aβ PET values were extracted across seven bilateral regions which were hypothesized to be sensitive to early Aβ accumulation: medial orbitofrontal, rostral anterior cingulate, posterior cingulate, precuneus, rostral middle temporal, superior frontal, inferior parietal 32 . Tracer uptake in those rst ve regions has also been found to be elevated in Aβ-negative individuals who subsequently had signi cant evidence of Aβ deposition 14 .
A Gaussian mixture modelling approach (GMM) was used to quantify region-speci c Aβ thresholds in the 7 bilateral regions hypothesized to be sensitive to early Aβ accumulation. Typically, Aβ follows a bimodal distribution and thus we tted two Gaussian distributions as commonly used to categorize Aβ positivity 18,32,42 . The two distributions acquired from GMM assigned each participant a probability of belonging to either the lower or higher distributions. We set a cut-off at the 90 th percentile of the lower distribution. Those who had higher SUVR values than the regional cut-off was classi ed as "positive" for that speci c region. According to the region-speci c positivity, individuals who were Aβ-positive in all 7 regions were classi ed as the "Widespread Aβ group"; those who were positive in at most 6 regions were included in the "Regional Aβ group"; those who were negative in all the regions were considered as the "Negative Aβ group".
As expected, the SUVR regional distribution of the data in all cohorts differed because of tracer differences. In PREVENT-AD and HABS (NAV and PIB tracer respectively), GMM analyses provided a clear distinction between distributions using thresholds for each region corresponding to a 90% probability of belonging to the low Aβ distribution (Figure e-1). The distinction between regional positive and negative binding was less evident with Florbetapir (ADNI). As shown in Figure e-1, participants from ADNI followed a more continuous distribution without a distinctive cut-off between lower and higher distributions which might partly be due to the different properties of the tracers. This interfered with using the same cut-off criteria in ADNI. Based on previous publications 41,43,44 , we therefore decided to use a 50% probability of belonging to the low-Aβ distribution as cut-off criteria instead of a 90% probability.

Statistical Analysis
We compared demographics, APOE4 status, baseline and longitudinal cognition, baseline and longitudinal tau-PET between the three Aβ groups in the three cohorts separately using analysis of covariance tests and chi-squared tests for normally distributed continuous variables and categorical variables, respectively. Tukey HSD post hoc test and Bonferroni correction were applied to examine differences between the three Aβ groups.
Linear mixed-effects models were used to investigate longitudinal Aβ and tau-PET accumulation (ADNI and HABS) and cognitive decline (all cohorts) between the three Aβ groups. Participants who had at least 2 assessments were included in the analysis. Models included random slope and intercept, where the time by subject interaction determined change in cognition or tau. The analyses were anchored at the participant's baseline visit. For Aβ and tau accumulation, age and sex were included as covariates in the models, and for cognitive decline, education was further included as a covariate. Post-hoc tests are reported only when there was a signi cant main effect.
All statistical analyses were conducted using RStudio, version 1.2.5001 45 . The cutoff and mixtools packages (github.com/choisy/cutoff) were used for GMM and lme4 46 for mixed-effects models. The criterion for statistical signi cance was α ≤ 0.05 after correction for multiple comparisons by Tukey's test.

Results a. De ning Amyloid Groups based on Aβ spatial extent
In each cohort, GMMs were t to the 7 bilateral early Aβ regions ( Figure 1). Region-speci c thresholds for all cohorts are presented in detail in Table e-1. In PREVENT-AD, 81 participants (62%) were in the Negative Aβ group, 28 participants were in the Regional Aβ group (22%) and 20 exceeded the positivity thresholds in all regions and were placed in the Widespread Aβ group (16%). Applying the thresholds to the ADNI cohort resulted in 202 (50.5%) in the Negative Aβ group, 108 (27%) individuals in the Regional Aβ group and 90 (22.5%) individuals in the Widespread Aβ group. In the HABS cohort, 139 participants (48%) were in the Negative Aβ group, 76 (26%) participants in the Regional group and 73 (25%) participants in the Widespread group. The results presented below were done on all participants. The analyses were repeated when only keeping individuals in the Regional group that would have been classi ed as Aβ-negative based on the cohort speci c global Aβ thresholds (n=22 for the PREVENT-AD and n=61 for ADNI). The results were unchanged when removing Aβ positive participants from the Regional group (see Table e The Widespread and Regional Aβ groups had larger proportions of APOE ε4 carriers than the Negative Aβ groups in both the PREVENT-AD (Widespread vs. Negative X 2 (1, N = 101) = 8.54, p < 0.01; Regional vs. Negative X 2 (1, N = 109) = 10.8, p < 0.01) and ADNI (Widespread vs. Negative X 2 (1, N = 292) = 29.11, p < 0.01; Regional vs. Negative X 2 (1, N = 310) = 5.24, p < 0.05) cohorts (Table 1). In ADNI, the Widespread Aβ group also had a larger proportion of APOE ε4 carriers compare to the Regional Aβ group [X 2 (1, N = 198) = 6.98, p < 0.01]. In HABS, only Widespread Aβ group had larger proportions of APOE ε4 carriers than both groups (Widespread vs. Negative X 2 (1, N = 207) = 40.59, p < 0.001; Widespread vs. Regional X 2 (1, N = 144) = 16.54, p < 0.001).

d. Longitudinal Aβ Trajectories
In ADNI, all groups showed Aβ accumulation rates signi cantly different from zero over up to 4 years, with a median follow-up time of 3 [interquartile range (IQR): 1, 4] years. The amount of Aβ accumulation, however, differed between the groups (e- Table 3). The Widespread Aβ group showed faster Aβ accumulation over time than the Negative Aβ group in all the 7 early regions of interest (e- Table 3). The Regional group also showed faster Aβ accumulation than the Negative Aβ group in all the 7 early regions of interest (e- Table 3). Interestingly, no difference was found between the Regional and Widespread Aβ group regarding Aβ accumulation over time in any of the 7 early regions of interest (e- Table 3).

e. Cross-sectional and Longitudinal Tau Trajectories
In PREVENT-AD, the Widespread Aβ group had elevated Tau-PET signal when compared with Negative and Regional Aβ groups across the ve regions investigated (Entorhinal, Amygdala, Fusiform, Inferior Temporal, and Parahippocampal) ( Table 2). The Regional Aβ group had elevated tau-PET binding only in the Entorhinal cortex (F (2, 128) = 19.21, p<0.05) and Middle Temporal gyrus (F (2.128) = 14.06, p<0.05) compared with the Negative Aβ group. In both ADNI and HABS, the Widespread Aβ group had elevated Tau-PET signal compared with Negative and Regional Aβ groups across all regions investigated (Table 2 Figure 5). However, only in Amygdala, Fusiform, Inferior Temporal and Parahippocampal, the Widespread group accumulated greater tau compare to both the Negative Aβ group and the Regional Aβ group ( Figure 5). In addition, in the Middle temporal, the Widespread group accumulated greater tau only compare to Negative group (β [SE], 0.01 [0.004]; p < 0.05). The Regional group did not show any difference from Negative Aβ regarding tau accumulation over time both in ADNI and HABS.

Discussion
Most Alzheimer's drugs are targeting single disease pathways. Removing Aβ when tau has already disease progression even if administer in preclinical individuals. One way to identify individuals with Aβ that do not yet have tau or cognitive decline could be to assess Aβ spatial extent severity. The hypothesis would be that individuals that have Aβ-PET binding restricted to few brain regions might not yet have tau and related cognitive decline and therefore be optimal candidates for anti-Aβ therapies.
We investigated the biological and clinical relevance of Aβ spatial extent severity on AD biomarkers in three independent cohorts of cognitively normal older adults. We focused on seven regions hypothesized to be early Aβ accumulating regions 14,32 and classi ed participants into Widespread (7 regions with signi cant Aβ-PET binding), Regional (1-6 regions with Aβ-PET binding) and Negative groups. Our results suggest that when Aβ is spread out throughout the cortex, which in most cases equal being Aβ positive on classical whole brain measures, tau is also already spread and cognitive decline is prevalent. Individuals with regional binding however do not yet have signi cant tau or cognitive impairment and are probably idea candidates for anti-Aβ clinical trials.
We found that individuals with Regional Aβ-PET binding have a higher proportion of APOE ε4 carriers when compared to individuals with no Aβ-PET binding in two cohorts out of three. This proportion of APOE ε4 carriers reached 64% in the PREVENT-AD cohort, a cohort of cognitively normal individuals with a rst-degree family history of AD dementia and therefore with a higher risk to develop the disease themselves. In PREVENT-AD and ADNI, we found a grading effect of (quasi-continuous) Aβ 1-42 levels such that participants in the Widespread and Regional groups had decreased CSF Aβ 1-42 levels when compared to the Negative group, and the Widespread group had decreased CSF Aβ 1-42 levels compared to the Regional group (this information was not available in the HABS cohorts). Furthermore, in ADNI and HABS, the Regional group had greater amount of Aβ-PET accumulation when compared to the Negative group (information that was not available in the PREVENT-AD cohort). Of interest, however, the Regional groups showed no or very little tau-PET binding and no baseline cognitive impairment. While these ndings suggest that the regional Aβ PET-binding signal is biologically relevant, they also suggest that widespread Aβ is necessary to detect tau-PET signals outside of the entorhinal cortex and signi cant cognitive impairment.
Continuous variables are often dichotomized in the clinic to provide a straightforward diagnosis or identify patients that would bene t from treatment 48 . They are also used in research to simplify the interpretation of results and provide empirical evidence for clinical practice. The most common approach to analyzing Aβ-PET is to classify individuals into two groups, Aβ-negative and Aβ-positive. This approach is not always optimal to detect individuals with early Aβ levels, mainly if Aβ has accumulated regionally but is not yet globally widespread 32,47 . There is a growing body of literature documenting the earliest topographical distribution of Aβ-PET binding in individuals with and without cognitive impairment 15,18,49 . We took advantage of this literature to identify seven regions with early detectable Aβ-PET binding, and instead of classifying our participants on a "global" Aβ index, we dichotomized Aβ positivity/negativity within each of these regions and counted the number of regions with positive Aβ-PET binding. While PREVENT-AD, most of the participants with regional binding (79%) would have been classi ed as negative using a "global" Aβ index, all the Regional Aβ group participants in HABS would have been classi ed as negative. This number was slightly lower in ADNI (57%), which can probably be explained by the fact that their global threshold is slightly lower than what was used in the PREVENT-AD (centiloid 26.7 vs 18.5) and HABS (centiloid 23.9 vs 18.5).
Anti-Aβ therapeutic trials have failed to improve or slow down cognitive symptoms 8,50,51 . The association of cognition and neuronal loss is stronger with the tau-PET signal compared to Aβ 52 . The failure of these trials could be partly due to the inclusion of individuals who already have elevated tau-PET signals. Our results showed that cognitively normal individuals with widespread Aβ, which would most have been considered Aβ-positive using a classical global threshold, have detectable tau-PET signals in several temporal brain regions. Interestingly, elevated tau PET-binding was absent (ADNI and HABS) or restricted to the entorhinal and the middle temporal cortices (PREVENT-AD) in individuals showing regional Aβ-PET.
Binding in the entorhinal cortex is common with advanced age 53 . Furthermore, the Widespread Aβ groups in ADNI and HABS accumulated greater amount of tau compared to the Negative and Regional group over longitudinal follow-ups. The CSF data corroborate these ndings in ADNI and PREVENT-AD cohorts, with CSF p-tau being higher in the Widespread group compared to the two other groups, with an absence of signi cant differences between the Regional and the Negative groups. Previous studies have shown that when Aβ pathology reaches "widespread" spatial distribution, tau-PET uptake increases faster compared to the individuals with lower Aβ 54 , and the rate of tau-PET change is associated with cognitive decline 55 . In line with these results, baseline cognitive impairment was only found in the Widespread groups, and cognitive decline was also mainly restricted to the Widespread groups. The ADNI regional group also showed a cognitive decline when compared to the Negative group after a decade of follow-up, by which time most Regional individuals probably developed Widespread Aβ binding 14 .
Furthermore, our ndings highlight the biological relevance of the Regional Aβ group. Our result showed that the Regional Aβ groups had intermediate CSF Aβ 1-42 levels between the Widespread (lower Aβ  and Negative (higher Aβ 1-42 ) Aβ groups, showing signs of incipient cerebral accumulation of Aβ 56 . Even though Aβ increases with older age, Regional Aβ group participants were in the same age range as the Negative Aβ group, which was younger than the Widespread Aβ group in all cohorts. Furthermore, the Regional Aβ groups in ADNI and HABS accumulated greater amount of Aβ compared to the Negative group. Another crucial difference between groups was marked by APOE ε4 carrier status; compared to the Negative Aβ group, both Regional and Widespread Aβ groups had higher percentages of APOE ε4 carriers in two cohorts, which places them at increased risk for developing the disease 57 . There is an increasing interest in the biological relevance of regional Aβ and the assessment of regional patterns 15,16 . Recent studies have shown decreased CSF Aβ 1-42 levels in participants with regional Aβ 18,58 , as well as higher proportions of APOE ε4 carriers, compared to Aβ negative participants 59 . Even in individuals categorized as Aβ negative, subthreshold Aβ predicted a slight memory decline 60 and the development of tau pathology over ve years 61 . APOE ε4 carriership has also been associated with increased Aβ load compared to non-carriers across all clinical diagnostic groups 62 . Taken together, our ndings highlight the biological relevance of the Regional Aβ group, for which tau and cognitive impairments are still minimal.
Therefore, most individuals with regional Aβ binding are at the earliest stage of the AD continuum and only a few years away from when cognitive decline is about to start.
There are several limitations to take into account. An important factor that may have impacted the current study results is the Aβ groups' disproportion due to the small sample size of the Widespread and Regional groups in PREVENT-AD. More than 63% of the cohort were in the Negative Aβ group, which led to the Regional and Widespread groups consisting of less than 30 individuals each. To address this limitation and to validate the results, the ADNI cohort with 400 participants and the HABS cohort with 288 participants were included in the study. ADNI, however, used the Florbetapir tracer, for which it might be more di cult to establish clear dichotomized values given the high variability related to white matter signal 63 . In addition, previous studies also reported that this tracer had shown a low correlation between tracer-speci c regional rankings compared to four other tracers 41,58 . Despite the differences in the study designs, it is nevertheless important to mention that most of the results across cohorts were comparable.
In conclusion, assessing the spatial Aβ burden could be a powerful way to identify the best candidates for preventive clinical trials. Assessing the presence of Aβ-PET binding in early accumulating regions can help identify individuals with biologically relevant signals that would have been classi ed as being negative using more established whole-brain thresholds for Aβ positivity. While these individuals accumulate Aβ over time, they do not yet have signi cant tau or cognitive decline. In individuals with widespread Aβ (most of whom would have been included in current clinical trials), tau pathology might be too advanced to stop the cognitive decline or disease progression after the removal of the Aβ plaques from the brain. Our results suggest that anti-Aβ trials should be performed in individuals with regional binding at the latest since even individuals classi ed as being Negative showed Aβ-PET binding  The values are reported as Mean (SD) except for Sex, APOE 4, and Subjective Cognitive Decline which are reported as the Number of participants (% of the group). BOLD text represents the groups between which there were signi cant differences: a = p<0.05 between Negative Aβ and Regional Aβ groups; b = p<0.05 between Negative Aβ and Widespread Aβ groups; c = p<0.05 between Regional Aβ and Widespread Aβ groups. *In PREVENT-AD, CSF samples were available for 46 Negative, 19 Regional, and 12 Widespread; in ADNI, CSF samples were available for 138 Negative, 78 Regional and 60 Widespread. APOE ε4: Apolipoproteinε4; Aβ: beta-amyloid; CSF: Cerebrospinal uid. Using ANCOVA and multiple comparisons corrected for age and sex, we test whether Tau-PET uptake in early tau regions signi cantly differed between the Aβ groups in the (A) PREVENT-AD cohort, (B) ADNI cohort and (C) HABS cohort. For post-hoc analysis, Bonferroni correction was applied when comparing the pair of group means. BOLD text represents the signi cant between-group differences. a = p<0.05 between Negative Aβ and Regional Aβ Groups; b = p<0.05 between Negative Aβ and Widespread Aβ Groups; c = p<0.05 between Regional Aβ and Widespread Aβ Groups. Table 3. Baseline Cognition gCognitive test scores were compared at the baseline visit corrected for age and sex; test scores are reported as Mean (SD). (A) As part of the PREVENT-AD battery, all participants undergo annual cognitive testing using the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS). (B) In ADNI, participants received detailed cognitive assessments from which composite scores are derived. All the composite scores have a mean of 0, and a standard deviation of 1. (C) HABS participants undergo annual cognitive testing with PACC5 to derive a cognitive composite score including memory, executive function and semantic processing. BOLD text represents the signi cant between-group differences: a = p<0.05 between Negative Aβ and Regional Aβ groups; b = p<0.05 between Negative Aβ and Widespread Aβ groups; c = p<0.05 between Regional Aβ and Widespread Aβ groups. Figure 1 De ning the Aβ Groups. Individuals were separated into three groups based on their Aβ status in seven cortical regions: rostral anterior cingulate, precuneus, medial orbitofrontal, rostral middle frontal, inferior parietal, superior frontal, and posterior cingulate. According to the region-speci c positivity, individuals who were Aβ-positive in all 7 regions were classi ed as the "Widespread Aβ group"; those who were positive in at most 6 regions were included in the "Regional Aβ group", while those negative in all the regions were considered as the "Negative Aβ group".  Tau-PET signal compared with Negative and Regional Aβ groups across all regions. One PREVENT-AD Regional participant and one PREVENT-AD Widespread participant were considered in uential cases based on their Cook's distance.

Figures
Removing these participants did not in uence the results. Analyses were corrected for age and sex. * p<0.05; ** p<0.01; ***p<0.001. SUVR: standardized uptake value ratio. (B) for ADNI; and (C) for HABS were represented over time in the three different groups. The Widespread Aβ group showed a greater decline in their cognition scores when compared with the two other groups in all cohorts. In both ADNI and HABS, the Regional group showed a greater cognitive decline compared to the Negative group. * p<0.05; ** p<0.01; ***p<0.001 Change in Aβ Uptake Over Time Between the three Aβ Groups in ADNI and HABS. Linear mixed-effect models investigating the effect of the groups on Aβ accumulation rate over time in ADNI and HABS cohorts corrected for age and sex. Plotted is the association between Aβ groups based on (A) Precuneus SUVR score and (B) Medial orbitofrontal SUVR score over the years from their rst scan. While both the Regional and Widespread Aβ groups accumulated Aβ at a faster rate compared to the Negative Aβ group in both cohorts; while only in HABS, the Widespread group accumulated Aβ at a faster rate compared to the Regional Aβ group in Precuneus. * p<0.05; ** p<0.01; ***p<0.001; etc. SUVR: standardized uptake value ratio.

Figure 5
Change in TAU Uptake Over Time Between the three Aβ Groups in ADNI and HABS. Linear mixed-effect models investigating the effect of the groups on tau accumulation rate over time in ADNI and HABS cohorts corrected for age and sex. Plotted is the association between Aβ groups based on Fusiform, Inferior Temporal and Middle Temporal SUVR scores over the years from their rst scan. In both cohorts, the Widespread group accumulated a greater amount of tau compare to the Negative group. However, only in Fusiform and Inferior Temporal (in HABS) and Middle Temporal (in ADNI) regions, compared to the Regional Aβ group, Widespread group accumulated a greater amount of tau. * p<0.05; ** p<0.01; ***p<0.001; etc. SUVR: standardized uptake value ratio.