Association of Amyloid and Tau With Cognition in Preclinical Alzheimer Disease

1Alzheimer disease (AD) is a progressive cognitive disorder leading to dementia in which the brain gradually accumulates both amyloid-β (Aβ) and tau pathologies. Autopsy studies identified the early stages of Aβ and tau pathologies in individuals who were clinically normal during life, representing a preclinical stage of AD. Based on autopsy studies, the prevailing research hypothesis posits that Aβ precedes and accelerates neocortical tau pathology, which together precipitate cognitive decline. Molecular positron emission tomography (PET) tracers for Aβ and tau have made it possible to detect these pathologies in living individuals, including in clinically normal adults. However, previous longitudinal PET studies tracked either Aβ, or tau, accumulation; to our knowledge, a temporal sequence of Aβ and tau accumulation has not yet been evaluated. To observe the sequence of these pathologic events, we investigated the trajectories and temporal courses of longitudinal Aβ-PET and longitudinal tau-PET data. Improved understanding of these trajectories is needed to efficiently test therapeutic strategies designed to halt the progression of pathology and delay cognitive decline. While cognitive decline has been demonstrated in longitudinal studies of older adults with elevated levels of Aβ pathology at baseline,,,, tau-PET may be more closely linked to neuronal injury and cognition.,, We therefore conducted a prospective natural history study to determine whether serial Aβ and tau measures were associated with concurrent and subsequent, serial measures of cognitive performance.

2In this report, we analyzed data from the Harvard Aging Brain study, a longitudinal study of aging conducted at Massachusetts General Hospital, Boston. We reported prospective observations collected from January 1, 2010, to October 31, 2017, from 60 individuals who had normal cognition at study entry: global Clinical Dementia Rating of 0 and/or Mini Mental State Examination (MMSE) and Wechsler Logical Memory II delayed recall (LM) scores within normal range (MMSE ≥27 and LM ≥11 if ≥16 years of education and MMSE ≥25 and LM ≥7 otherwise). Exclusion criteria included drug or alcohol abuse, head trauma, and serious medical or psychiatric condition (Geriatric Depression Scale >10 of 30). Annual consensus meetings evaluated progression to mild cognitive impairment (MCI). The Partners Institutional Review Board has approved the Harvard Aging Brain study protocol, and participants provided written informed consent before undergoing any procedures. The participants analyzed in this report had multiple flortaucipir (FTP, also known as AV1451 or T807) and Pittsburgh compound B (PiB) PETs assessing tau and Aβ pathology, respectively.

3Longitudinal data (Figure 1) were acquired for PiB and cognition from 2010. Because FTP was not available before 2013, the initial FTP was defined as baseline (timet = 0, where subscript t indicates time in years), and the terms baseline FTPt = 0 and initial FTPt = 0 are equivalent. The initial PiB was acquired a median of 3 years before baseline and is termed initial PiBt = -3, with baseline PiBt = 0 referring to PiB at approximately the time of baseline FTPt = 0, with a median time difference of −1.1 months (range, −9.7 to 9.0).

4Final PiBt = 2 was performed at the same time as final FTP t = 2, with a median time difference of 0.0 months (range, −6.3 to 23.4). Baseline cognitiont = 0 was evaluated within 6 months of baseline FTPt = 0, with a median time difference of −1.2 months (range, −5.9 to 5.5). Final cognitont = 3 was evaluated 11.8 months (range, −3.9 to 26.2) after final FTPt = 2.

5Participants had 2 or 3 FTP observations (n = 9) over a median follow-up of 26.0 months (range, 13.1-36.4). Participants had 2 to 5 PiB and 4 to 8 cognitive sessions. Pittsburgh compound B and cognition were measured in 2 successive periods: before and after baseline FTPt = 0. Pittsburgh compound B changes were measured before baseline for 36.4 months (range, 15.9-63.5) and after baseline for 24.4 months (range, 16.6-49.3). Cognitive changes were measured before baseline for 34.6 months (range, 4.3-48.8) and after baseline for 38.2 months (range, 23.2-50.5). In 10 participants without prebaseline PiB data, PiB change was assessed from PiBt = 0.

6Participants in the Harvard Aging Brain study are evaluated annually with a battery of cognitive assessments, including tests of episodic memory, executive function, global cognition, and the Clinical Dementia Rating. For this study, we evaluated cognition using the Preclinical Alzheimer Cognitive Composite (PACC-96), a mean of z score performances on 4 tests sensitive to cognitive decline in at-risk individuals: MMSE, LM, Digit-Symbol Coding, and the Free and Cued Selective Reminding Test, which uses 3 versions with different items, each version repeated every 3 years.

7The 11C-PiB and F-18–FTP tracers were synthesized and administered onsite. Positron emission tomography images were acquired using a Siemens HR+ scanner. Both PiB and FTP measures were computed as standardized uptake value ratios (SUVr; 4 frames of 5 minutes: 80-100 minutes for FTP and 40-60 minutes for PiB) using cerebral white matter as the reference region,, because this reference provided more stable estimates of both PiB and FTP change (eAppendix in the Supplement). Positron emission tomography data were coregistered to each participant’s magnetic resonance imaging and segmented with Freesurfer, version 5 (Martinos Center for Biomedical Imaging). For each participant, we selected the magnetic resonance imaging closest to the midpoint between FTP sessions. Partial volume correction was applied using geometric transfer matrix. The PiB signal was extracted from a neocortical aggregate and FTP from inferior temporal, a region where tau is commonly observed in preclinical AD. Additional brain regions were investigated in eTables 1 and 2 in the Supplement; similar results were observed in the temporal neocortex and precuneus. High-PiB threshold was set at 0.724 SUVr using a Gaussian mixture model on the initial PiBt = -3 data. Some analyses focused on participants with low PiB to evaluate the contribution of subthreshold PET signal accumulation.

8We computed linear mixed models with random intercept and time slope per participant predicting PACC, FTP, and PiB over time, in separate models. Individual slopes of change were calculated by summing the estimated fixed and random effects of time. For PiB and PACC data, slopes were estimated for the entire follow-up (PiBt = -3 to PiBt = 2), and for shorter periods (referred to as PiBt = -3 to PiBt = 0 before baseline and PiBt = 0 to PiBt = 2 after baseline). Cross-sectional measures and slope data were entered as predictors or outcomes in linear regressions evaluating the associations between PACC, FTP, and PiB and their respective slopes. Five thousand–iteration bootstrapping procedures that accounted for the 2-stage estimation procedure generated 95% confidence intervals. Older ages were associated with steeper PACC slope but not with greater FTP or PiB slope (eFigure 1 in the Supplement); all models predicting PACC slope were therefore adjusted for age. Nine statistical models are displayed in Tables 1 and 2 and another 4 models are displayed in Figures 2 and 3. We did not correct for multiple comparisons., Results were summarized in a serial mediation model (Figure 3), providing evidence for sequential biomarker changes in preclinical AD. This model tested whether the association of initial PiBt = -3 with final PACCt = 3 was mediated by sequential changes in PiBt = -3 to PiBt = 0 and FTPt = 0 to FTPt = 2, adjusting for age, sex, education, and initial PACCt = -3. All possible indirect pathways between PiB, FTP, and final PACC t = 3 scores were tested. Total, direct, and indirect associations were tested using a 5000-iteration bootstrapping procedure. Models were fit in Matlab, version 9.0 (MathWorks), except mediation models, which used R, version 3.4.2, Lavaan package (the R Foundation). We report 2-sided P values with a significance of.05.

9Demographics, cognitive, and PET data of the 60 participants are provided in Table 1. Based on the initial PiBt = -3, 43 participants were classified as low PiB and 17 as high PiB. During the first observation period of the 3 years prior to baseline FTP, PiB increased and 3 participants progressed from low PiB to high PiB. The PACC scores also increased, presumably indicating practice effects. Participants with high PiB had faster PiB increase and lower practice effect than participants with low PiB. After the first period, no participant met the criteria for MCI.

10During the second observation period from baseline to 3 years after baseline, PiB increased, FTP increased, and PACC decreased. Although all these changes were greater than 0 in participants with low PiB (Table 1), participants with high PiB had faster PiB increase, FTP increase, and PACC decline. Six participants with high PiB and no participants with low PiB met clinical criteria for MCI at study end.

11We first observed that FTPt = 0 to FTPt = 2 changes were associated with contemporaneous PiBt = 0 to PiBt = 2 changes (Figure 2A and D), indicating a longitudinal association between AD pathologies. We then observed that an early PiBt = -3 to PiBt = 0 rise was associated with later FTPt = 0 to FTPt = 2 changes, regardless of initial PiBt = -3 (Table 2; model 1). However, we could not evaluate whether an early FTP rise was associated with later PiB changes because FTP was not measured in the first observation period. Therefore, we investigated whether FTP or PiB changes in the second observation period were associated with final PiB or FTP, respectively. Consistent with PiB increases preceding FTP, we observed that previous FTPt = 0 to FTPt = 2 changes were not associated with later PiBt = 2 (model 2), but previous PiBt = 0 to PiBt = 2 changes were associated with later FTPt = 2 (model 3). When measured over a longer period preceding the FTP measure, PiB changes were even more closely associated with final FTPt= 2 levels (model 4).

12To investigate whether baseline PiB and FTP were independently or synergistically associated with subsequent changes in PiB and FTP, we investigated their main and interactive effects. We observed that FTPt = 0 to FTPt = 2 change was associated with the interaction between baseline PiBt = 0 and FTPt = 0 (model 5), but PiBt = 0 to PiBt = 2 change was only associated with baseline PiBt = 0 (model 6), providing evidence that PiB changes occurred independently of baseline FTP levels, while FTP changes were contingent on baseline PiB levels.

13We next aimed to investigate a hypothetical sequence between PiB, FTP, and PACC. The PiBt = 0 was not significantly associated with PACCt = 0 at the cross-section, but FTPt = 0 was (−1.50; 95% CI, −2.90 to −0.10; P =.04). The PACC was more closely associated with FTP than with PiB at all times when PiB and FTP competed in the same models, suggesting that FTP signal is more proximal to PACC decline than PiB signal. Similarly, PiB slope and PACC slope were not significantly associated in both periods of observations. The PiB slope was associated with PACC slope for the entire study (Figure 2B and E), but this association did not survive adjusting for initial PiBt = -3. In contrast, FTP slope was associated with PACC slope (Figure 2C and F), including after covarying FTPt = 0.

14In a multiple regression estimating PACCt = 0 to PACCt = 3 change with baseline and change in PiBt = 0 to PiBt = 2 and FTPt = 0 to FTPt = 2, only baseline PiBt = 0 and FTPt = 0 to FTPt = 2 change were significant (Table 2; model 7). The interaction between PiBt = 0 and FTPt = 0 to FTPt = 2 change was also significant (model 8), such that FTP change had a greater association with PACC decline at higher PiB-SUVr. Remarkably, although the association of FTP change with PACC change was greater in high PiB, it was also marginally present in individuals with low PiB (−2.39; 95% CI, −6.05 to 0.29; P =.09).

15Flortaucipir change was associated with clinical progression from preclinical to prodromal AD. Despite the small sample size, the participants with high PiB who progressed to MCI (n = 6) had significantly greater FTPt = 0 to FTPt = 2 change (0.05 SUVr per year; 95% CI, 0.03-0.07; P =.001, eFigure 2 in the Supplement) than the stable participants with high PiB (n = 11). The stable participants with high PiB had similar FTPt = 0 to FTPt = 2 change compared with the participants with low PiB (n = 43), highlighting that only a subgroup of participants with high PiB had fast FTP increase, ie, those who progressed to MCI. The PiB change did not differ between those who progressed to MCI and stable participants with high PiB.

16The previous models pointed to FTP change as the strongest factor associated with PACC change, potentially because FTP change was closer in time to PACC change. Because FTP change was associated with previous PiBt = -3 to PiBt = 0 change (Table 2; model 1), we inquired whether a sequence of successive change in PiB and FTP could account for the association between initial PiBt = -3 and final PACCt = 3 scores (PACC SD, −1.50 per PiB-SUVr; P =.004). To this end, we modeled a serial mediation assessing different possible pathways between Aβ, tau, and cognition. This model demonstrated that initial PiBt = -3 was associated with sequential changes, first in PiBt = -3 to PiBt = 0, and then in FTPt = 0 to FTPt = 2, and this sequence was associated with final PACC t = +3 scores (Figure 3). After mediation, the direct association of initial PiBt = -3 with final PACCt = 3 became nonsignificant because it reduced from −1.50 to −0.82 (45%).

17Our results raise the possibility that halting tau accumulation would prevent cognitive decline. To evaluate the potential advantage of using serial FTP-PET measures in trials, we directly compared the association of FTPt = 0 to FTPt = 2 change and final FTPt = 2 with PACCt = 2 to PACCt = 3 change observed after the final FTPt = 2 measure. We found that cognitive decline had a greater association with the longitudinal measure of FTP change than with the cross-sectional measure of FTP (Table 2; model 9), indicating that trials would benefit from serial FTP-PET measures to better identify participants at risk of subsequent decline.

18In this prospective study, we followed up clinically normal older adults in the preclinical phase of AD for a period of 7 years and observed an antecedent rise in Aβ to be associated with subsequent tau accumulation in inferior temporal cortex. We found this sequence to be strongly associated with cognitive decline. All participants were clinically normal at baseline, but the subgroup of individuals with high Aβ with fast tau increase met clinical criteria for MCI at follow-up. Our results indicate a sequence of observable phenomena in preclinical AD:

19Amyloid-β increase was the initial event observed, including in those with low-Aβ levels. In the first observation period, Aβ increased but cognition did not decline until the second period. Tau measures were not available in the first period, but Aβ measures were associated with subsequent tau changes (model 1) and final tau levels (models 3-5). Two studies, also found that an antecedent rise in Aβ was associated with final tau regardless of Aβ levels, but they had no longitudinal tau-PET data to investigate sequential changes.

20Tau increase in inferior temporal neocortex, while measurable in low-Aβ individuals, was faster in those who were increasing Aβ. A longitudinal tau-PET study observed that tau increased faster in high-Aβ than in low-Aβ clinically normal adults and another did not, but neither provided longitudinal Aβ-PET data. Our data indicate that tau changes are more closely associated with the rate of Aβ change than by Aβ levels (model 1). A short delay between Aβ and tau increases is likely to occur in some individuals, as suggested by 3 participants initially classified as low PiB who had both PiB and FTP increase after crossing the threshold for PiB-PET positivity (Figure 2B and C).

21Cognitive decline was most closely associated with tau change, beyond baseline Aβ and tau. Model 9 indicated that tau changes were associated with subsequent cognitive changes beyond the final FTP scan.

22After 7 years of cognitive follow-up, criteria for MCI owing to AD were met in a subset of 6 participants with high PiB (35%). These observations, suggesting higher rates of tau accumulation with clinical progression, are consistent with previous studies showing higher rates of tau accumulation in patients with symptomatic AD than in clinically normal older adults., The sequence from subthreshold Aβ accumulation to MCI was not observed in any participant, suggesting it requires longer than 7 years. We did not observe any MCI owing to non-AD etiologies.

23Altogether, our findings indicate that Aβ-PET measures have a delayed and indirect, tau-mediated association with cognition. Previous longitudinal PiB data estimated that the threshold for Aβ positivity was reached many years before dementia onset. We observed that tau changed shortly after Aβ positivity; however, we also observed high variability in tau change among individuals with high Aβ; those with rates of tau change similar to the low-Aβ group had stable cognition, highlighting that the Aβ-cognition delay may be variable and emphasizing the value of measuring tau to track disease progression in preclinical AD.

24Cerebrospinal fluid studies also found that longitudinal Aβ and tau trajectories were associated, and the rate of tau changes, not the rate of Aβ changes, were associated with cognition, but sequential associations in different times were not investigated using cerebrospinal fluid. Both PET and cerebrospinal fluid data indicate that synergy between Aβ and tau is associated with brain dysfunction,, atrophy,,, and cognitive decline., We observed that Aβ and tau in inferior temporal neocortex interacted and potentiated tauopathy (model 5) and cognitive decline (model 8).

25The low-Aβ group demonstrated tau increase (Table 1) and an association between tau increase and cognition, albeit weak. In contrast, tau increase was not observed in participants with low PiB in a tau-PET study with a 14-month follow-up. Our findings indicate that some individuals classified as low Aβ may be on the same trajectory of tau-mediated memory decline as those with high Aβ., Autopsy studies indicate that PiB is not sensitive to prefibrillar or low levels of fibrillar Aβ, which may be biologically active. The observation that the subthreshold Aβ accumulation was associated with subsequent tau accumulation highlights the limits of cross-sectional Aβ-PET and advocates for using repeated PET measures to improve characterization of preclinical AD.

26Our results may inform prevention therapeutic trials. Tau measures showed greater rates and consistency of accumulation than measures of Aβ or cognition (Table 1). Tau-PET outcomes may thus permit more rapid assessment of pharmacodynamic effects and thereby facilitate early phase proof-of-concept trials. Furthermore, serial tau-PET measures could identify individuals at risk of rapid cognitive decline (Table 2; model 9). Probably because Aβ increases long before cognition declines, Aβ changes did not add information compared with a baseline Aβ-PET to predict cognition; however, Aβ changes were associated with tau changes, suggesting that the effect of Aβ accumulation could be more easily assessed with tau-PET outcomes than with cognitive measures. Lastly, because clinical progression is more closely associated with tau than with Aβ, drugs effectively reducing tau increase on a tau-PET outcome may be more likely to slow down the rate of decline when tested with clinical outcomes.

27We did not observe tau at study start because the FTP tracer was not yet available. Although we provided evidence in favor of Aβ preceding inferior temporal tau, we could not test the absence of association between early tau changes and later Aβ changes. Because our data set only included participants older than 65 years, we focused the current work on neocortical tau, but future studies will also need to focus on tau accumulation in the medial temporal lobe, a region in which tau may accumulate at younger ages and may precede Aβ accumulation according to autopsy studies. Future research will also determine the trajectory of structural and functional neurodegenerative markers respective to changes in Aβ, tau, and cognition as well as their spatial overlaps. A 2019 study suggested that tau accumulation and brain atrophy share a similar topography. Finally, the relatively modest sample size of this study prevents generalization. Some observations were based on a few individuals with high Aβ who progressed to MCI; we observed an association between tau change and cognitive change in the participants with low Aβ as well, but it was only trend level. All cases may not follow the same temporal progression, and larger studies are thus required to evaluate interindividual variations in biomarkers trajectories.

28In this longitudinal PET study, we observed that successive changes in Aβ and then tau were associated with lower cognition after a 7-year follow-up. Larger samples are needed to validate the proposed sequence. Additional observations will help estimate the delay separating the trajectories of Aβ, tau, and cognition.