Main

The current state of knowledge suggests that the pathological processes underlying Alzheimer’s disease (AD) develop along a continuum1,2, with a prolonged preclinical phase of pathological buildup preceding detectable neurodegeneration, and cognitive symptoms manifesting in the later disease stages3,4,5. The defining histopathological hallmarks of AD are the accumulation of amyloid-β (Aβ) plaques and fibrillary tangles of hyperphosphorylated tau proteins in the brain4,6,7,8. Aβ plaques deposition can begin up to two decades before symptom onset1,4,9, initially accumulating in cortical areas with high metabolic baseline activity, such as the precuneus, medial orbitofrontal and posterior cingulate cortices10,11, before spreading to the entire neocortex, brainstem and subcortical nuclei12,13. Tau deposition also follows a relatively stereotypical distribution pattern, accumulating first in the entorhinal cortex before spreading to limbic areas and eventually the neocortex6,14,15. The joint accumulation of Aβ and tau triggers a cascade of deleterious events from synaptic loss, neuronal death, to brain atrophy5,16,17,18, which are thought to underlie the cognitive deficits that characterize the disease4,19. However, the exact mechanisms behind the disruption of human neural function and cognition related to AD proteinopathy remain to be elucidated.

A plausible mechanism of these noxious effects is that the joint accumulation of Aβ and tau alters neurophysiological signaling, triggering further pathological processes in a chain reaction that promotes disease progression20,21. Animal models show that Aβ deposition induces a toxic regime of neuronal hyperactivity, which then contributes to aggravating Aβ pathology itself22,23. This positive feedback process would then promote the accumulation and spread of tau pathology20. Consequently, the joint, colocalized accumulation of Aβ plaques and tau neurofibrillary tangles can be conceived as synergistic, potentially driving shifts in neuronal activity toward a regime of relative hypoactivity20,24. At this stage, neuronal activity is considerably diminished, eventually leading to cell death, tissue degeneration and severe behavioral changes20,21,25.

Whether this hypothetical two-pronged shifting regime of neural dynamics induced by AD proteinopathy occurs at the macroscopic scale in the human brain is still debated. In the prodromal stage of AD, increased blood-oxygen-level-dependent (BOLD) activations in hippocampal and middle temporal lobe circuits have been reported in memory tasks using functional magnetic resonance imaging (fMRI)26,27,28. In contrast, BOLD responses are typically diminished during later disease stages26,29. Considering the magnitude of BOLD signals as a proxy of brain activation, these observations are aligned with the expected gradual shift of brain activity related to the synergistic effects of Aβ and tau pathology, along an inverted U-shape trajectory—from baseline to hyperactive then hypoactive levels across the AD continuum (Fig. 1).

Fig. 1: Time-course of pathological changes along the AD continuum.
figure 1

The traces represent the possible evolution of pathological changes across the AD continuum, with the new hypothesis of an early shift in neurophysiological activity from acceleration to slowing during the preclinical phase. Aβ deposition begins early in the preclinical stage of the disease, preceding tau deposition, subsequent neurodegeneration and presentation of cognitive symptoms. According to the proposed model, early Aβ deposition is related to a transient acceleration of neurophysiological activity, while the later joint deposition of Aβ and tau would be associated with a shift towards neurophysiological slowing. The left axis in blue corresponds to the acceleration and slowing of neurophysiological activity, while right black axis depicts more generally the nature of AD pathological changes from normal to abnormal. The double arrows highlight the uncertainty around the onset of the neurophysiological acceleration and slowing with respect to other early pathological changes.

Full size image

Direct measurements of human brain electrophysiology also indicate aberrant macroscopic neurophysiological signaling in patients with AD-like symptoms30,31,32,33,34. A consistent observation with electroencephalography (EEG) and magnetoencephalography (MEG) is that low-frequency neurophysiological activity (delta–theta bands; 2–8 Hz) is increased and higher frequency activity (mainly in the alpha band; 8–12 Hz) is reduced in the symptomatic and prodromal stages of the disease31,35,36,37,38,39,40.

A recent computational model of the neurophysiology of AD proteinopathies points to a possible causal effect of early Aβ and tau depositions via the disruption of the excitatory/inhibitory balance of local neuronal populations, which in turn alters the macroscopic neurophysiological frequency spectrum across brain networks in the later stages of the disease41. According to this model, the accumulation of Aβ and tau induces a gradual shift in neuronal activity from hyperactivity to hypoactivity with respect to baseline healthy variants, notably manifesting at the macroscopic scale with levels of alpha-band activity following an inverted U-shaped trajectory across the AD continuum. For this reason, in the present study, we map with MEG the levels of cortical alpha-band activity with respect to other typical frequency bands as a noninvasive, macroscopic proxy of relative neurophysiological acceleration/slowing effects related to AD proteinopathy.

There is early evidence of these model predictions from human empirical data, albeit still relatively limited in scope. The alpha-band activity in participants with greater Aβ deposition is increased compared to normative healthy controls, but it is reduced in patients presenting mild cognitive impairment (MCl)42. The same study also reported increased, slower delta-band (1–4 Hz) activity in MCI patients, which together with decreased alpha-band activity, is aligned with the prediction of an acceleration effect on oscillatory neurophysiological activity (increased alpha-band activity) followed by a shift towards neurophysiological slowing (decreased alpha-band power and increased delta-band activity) during the prodromal stage of AD, as cognitive decline starts to manifest. These observations were independently confirmed recently in reports of Aβ33,40 and tau32,33 accumulations, which are related to decreased alpha-band and increased delta-band activity—that is neurophysiological slowing—in the later stages of AD. Whether a more subtle neurophysiological shift from enhanced fast-frequency activity to neurophysiological slowing related to AD proteinopathy and cognitive outcomes is detectable macroscopically in the preclinical stages of the disease, when potential therapeutic interventions may have the greatest impact39, was the overarching question of the present study.

We used task-free MEG source imaging of millisecond brain activity and whole-brain quantitative PET imaging of Aβ and tau in 104 asymptomatic older adults with familial history of sporadic AD dementia (Fig. 2a). As anticipated, we found that early Aβ deposition is associated with macroscopic neurophysiological manifestations of hyperactivity, expressed as an acceleration of neurophysiological oscillatory activity reflected by increased alpha-band and decreased delta-band activity. As predicted from disease models, we also found that these effects shift from oscillatory acceleration to slowing in individuals presenting early temporal lobe tau pathology, as denoted by decreased alpha-band and increased delta-band activity (Fig. 1). Finally, we demonstrate that the magnitude of the shift from neurophysiological acceleration to slowing is associated with longitudinal cognitive decline, and that the magnitude of the changes in neurophysiology and proteinopathy predicted by our model aligns with published observations in patients in the later stages of the AD continuum (that is MCI and AD).

Fig. 2: Multimodal data processing and analysis pipeline.
figure 2

a, Neurophysiological data were preprocessed, source imaged, frequency-transformed and averaged per canonical frequency bands (delta, 2–4 Hz; theta, 5–7 Hz; alpha, 8–12 Hz and beta, 15–29 Hz) and within each of the 68 cortical regions-of-interest (ROI) from the Desikan–Killiany atlas. PET images were preprocessed to obtain standardized uptake value ratio (SUVR) maps, which were then averaged within each cortical region of the same atlas across the whole brain (Aβ) or within temporal lobe ROIs (entorhinal cortex and temporal meta-ROI) for tau. b, Nested LME models were used to test for the main effect of Aβ deposition on neurophysiological activity, as well as the interactive between-participants effect of temporal tau deposition on this association. c, To assess impacts on cognitive performance, the correlation between Aβ and neurophysiological spectral power was estimated to test whether the interaction of this relationship with tau deposition across individuals related to longitudinal declines in attention and memory.

Full size image

Results

Neurophysiological shifts linked to Aβ and tau

Participant demographics and summary descriptive statistics of the PET data analyses and the screening and MEG-PET visit cognitive testing are provided in Table 1. We used ANCOVAs to compare the mean relative spectral power of whole-brain cortical neurophysiological activity across the following three PET-defined subgroups of individuals (Table 2): (1) those with no Aβ nor tau burden (Aβ−/Tau−), (2) those with high levels of global Aβ but no entorhinal tau (Aβ+/Tau−) and (3) those with both high global Aβ and high entorhinal tau (Aβ+/Tau+). In general, participants with greater levels of Aβ and tau exhibited higher levels of slow activity (delta–theta band), and lower levels of faster activity (alpha–beta band; Extended Data Fig. 1). Repeating the analysis after removing the influential cases resulted in similar associations, with significant statistical differences observed for the delta and alpha bands (Extended Data Table 1).

Table 1 PREVENT-AD participant characteristics
Full size table
Table 2 Whole-brain averaged between-participants spectral power analysis
Full size table

Proteinopathy-related shifts in neurophysiological activity

Considering the sample size imbalance between PET subgroups and the limitations of using a positivity cutoff that ignores potentially meaningful, subthreshold levels of pathology, we implemented nested linear mixed effects models (LME) to better incorporate intraparticipant spatial variability in our modeled associations between Aβ, entorhinal tau and neurophysiological activity (Fig. 2b). We found that the regional deposition of Aβ is related to an enhancement of fast-frequency neurophysiological activity, scaling with increased alpha-band activity (t(6967) = 14.23, PFDR = 0.016) and decreased delta-band activity (t(6967) = −14.69, PFDR = 0.016). Notably, this effect was reduced in individuals with greater tau pathology in both the alpha (t(6966) = −5.46, PFDR < 0.001) and delta (t(6966) = 4.33, PFDR < 0.001) frequency bands, indicating a tau-related shift towards neurophysiological slowing (Fig. 3). The aperiodic component of the neurophysiological power spectrum did not impact the observed associations between neurophysiological activity and Aβ deposition (alpha band—t(6965) = 14.25, PFDR = 0.008; delta band—t(6965) = −15.76, PFDR < 0.001), nor their synergistic effects with entorhinal tau deposition (alpha band—t(6964) = −4.52, PFDR < 0.001; delta band—t(6964) = 5.54, PFDR < 0.001).

Fig. 3: Interactive effects of AD proteinopathies on neurophysiological activity in asymptomatic adults.
figure 3

LME models show that Aβ was negatively associated with relative delta-band (t(6967) = −14.69, PFDR = 0.016) and positively associated with alpha-band (t(6967) = 14.23, PFDR = 0.016) power, suggesting an accelerating effect of Aβ on neurophysiological activity. Higher tau deposition in the entorhinal cortex was related to a shift towards neurophysiological slowing in these Aβ–neurophysiology associations (delta band—t(6966) = 4.33, PFDR < 0.001; alpha band—t(6966) = −5.46, PFDR < 0.001). Group average delta- and alpha-band relative power maps are shown on the left and average whole-brain Aβ and entorhinal SUVR maps are shown on top. Plots to the left indicate the nested relationship between Aβ and neurophysiological activity in the delta band (top) and alpha band (bottom), with lines of best fit for each participant indicated by thin colored lines representing entorhinal tau SUVR values and a thick black line representing the group main effect. Shaded region around the line of best fit across all participants represents the 95% confidence intervals. Plots to the right indicate the moderation of these nested Aβ–neurophysiological relationships by entorhinal tau deposition, with lines of best fit for individuals classified as tau-positive (red) and tau-negative (black), based on the group definition from the supplementary between-participant analysis. Shaded regions indicate 95% confidence intervals. Note that entorhinal tau SUVR values were treated as a continuous variable in all models and that the Tau−/Tau+ separation is only for visualization of the interaction effect. Corresponding t statistic and permuted, one-sided FDR corrected P values for the main effect of Aβ or the Aβ × Tau interaction are overlaid on each plot.

Full size image

To ensure that the moderating effects of tau deposition on the Aβ–neurophysiological relationships are not biased by assessing tau burden from a single region-of-interest (ROI) (entorhinal cortex), we repeated the analyses after assessing tau burden across an expanded set of temporal cortical regions (temporal meta-ROI) which are widely used for assessing early tau deposition43. Higher temporal meta-ROI tau values also related to a shift in the association between Aβ and neurophysiological activity in the alpha (t(6966) = −4.27, PFDR < 0.001) and delta (t(6966) = 3.46, PFDR = 0.004) frequency bands, in the same opposite directions as observed with the entorhinal cortex as tau ROI (Extended Data Fig. 2). Once again, the aperiodic components of the neurophysiological power spectrum did not affect the synergistic effects of Aβ and tau on neurophysiological activity (alpha band—t(6964) = −3.31, PFDR = 0.010; delta band—t(6964) = 4.8, PFDR < 0.001). The same pattern of tau effects remained after removing the entorhinal cortex from the temporal meta-ROI (alpha band—t(6966) = −3.74, PFDR < 0.001; delta band—t(6966) = 3.06, PFDR = 0.008). A similar trend for the interactive effect of Aβ and tau on alpha-band power was observed when using whole-brain tau standardized uptake value ratio (SUVR) values as a nested variable, although these associations were no longer significant after permutation and FDR correction (alpha band—t(6965) = −6.42, PFDR = 0.106; delta band—t(6965) = 2.67, PFDR = 0.296; Extended Data Fig. 2).

Proteinopathy-related neurophysiological changes and cognition

In terms of longitudinal cognitive changes evaluated by repeated administration of the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) neuropsychological assessment battery, we observed that individuals with high levels of both Aβ and tau showed more pronounced declines in the composite score calculated from the attention and immediate/delayed memory domains (F(2, 93) = 8.81, PFDR = 0.001; Fig. 4a). We then used linear models to test whether the interactive effects of continuous Aβ and tau SUVR values on neurophysiological activity were related to longitudinal cognitive performance (Fig. 2c). We found that individuals with a stronger moderating influence of tau deposition on the association of Aβ deposition with alpha-band activity have experienced a more pronounced cognitive decline, indicated by steeper negative slopes in the attention/memory composite score (t(92) = 3.37, PFDR = 0.002; Fig. 4b). When assessing each cognitive domain separately, we found the same interaction effect on the attention scores (t(92) = 3.58, PFDR = 0.002; Extended Data Table 2). Such associations remained significant after removing influential cases from each of the regression models, also extending to the immediate memory domain. The cognition analysis results were also replicated when using the temporal meta-ROI instead of the entorhinal cortex tau SUVR, both for the attention–memory composite score (t(92) = 2.48, PFDR = 0.038) and the attention scores (t(92) = 2.37, PFDR = 0.038). We did not find such associations with the Aβ–delta-band correlation values.

Fig. 4: Neurophysiological impacts of AD proteinopathy are associated with longitudinal cognitive declines.
figure 4

a, Box plots show the median, quartiles, minima and maxima, while violin plots show the density distribution of the annual rate of change calculated for the RBANS attention–memory composite score (attention, immediate and delayed memory) of the participants, separated according to Aβ and tau positivity. The two-sided, one-factor ANCOVA analysis (F(2, 93) = 8.81, PFDR = 0.001) revealed that individuals expressing both Aβ and tau pathology (Aβ+/Tau+, n = 11) showed steeper cognitive decline compared to individuals with no pathology (Aβ−/Tau−, n = 49) and individuals with Aβ but no entorhinal cortex tau load (Aβ+/Tau−, n = 42). Tukey HSD P values adjusted for multiple comparisons are shown for the difference between the Aβ+/Tau+ and the Aβ−/Tau− and Aβ+/Tau− groups. b, Linear regression models demonstrated an interaction effect (t(92) = 3.37, PFDR = 0.002), where a more pronounced shift in the Aβ–alpha-band activity relationship for individuals with higher entorhinal tau was associated with longitudinal decline in attention–memory composite scores. The thick colored lines represent the line of best fit across participants in the Aβ−/Tau− (blue, n = 49), Aβ+/Tau− (green, n = 42) and Aβ+/Tau+ (red, n = 11) groups. The shaded area represents the 95% confidence intervals. Note that whole-brain Aβ and entorhinal tau SUVR values were treated as continuous variables for this analysis, and that participants were divided according to Aβ/Tau +/− groups only for visualization purposes. Corresponding t statistic and one-sided, FDR corrected P values are overlaid on the scatter plot.

Full size image

Predictive analysis of later-stage neurophysiological changes

Finally, we tested whether the magnitude of neurophysiological changes associated with the higher Aβ (whole brain) and tau (temporal meta-ROI) concentrations observed in later disease stages can be predicted from the interactive effects of Aβ and tau on alpha-band activity we observed in the preclinical data reported herein. To that end, we first retrieved the Aβ (cortical meta-ROI) and tau (temporal meta-ROI) deposition values reported in ref. 44, and the relative alpha-band power measurements reported in ref. 34 from independent but age-matched samples of participants presenting with MCI and probable AD. We used relative alpha-band power measures because our data showed this was the neurophysiological feature associated with longitudinal cognitive decline. We found that our model predictions overlapped substantially with these independent, empirical observations (Fig. 5). This analysis therefore supports the notion that the early synergistic interaction we observed between AD proteinopathies and neurophysiological activity in asymptomatic adults is consistent with and might be predictive of the related effects observed in the more advanced stages of the disease. We also emphasize this finding is robust against possible effects caused by different study sites, participant samples, PET tracers and MEG instruments.

Fig. 5: Independent validation of the synergistic association of proteinopathy with alpha-band neurophysiological activity in participants with MCI and early AD.
figure 5

We used the linear mixed-effects model coefficients of the interactive effect of Aβ and temporal meta-ROI tau on relative alpha-band power to infer the values of neurophysiological activity alpha-band power (colored heatmap) across increasing levels of global Aβ and temporal meta-ROI tau deposition along the AD continuum. The red cross (marked with ‘a’) represents the median and inter-quartile range of the empirical measures of global Aβ and temporal meta-ROI tau deposition of the present PREVENT-AD data in asymptomatic participants. The purple cross (marked with ‘b’) indicates the independent values reported in ref. 44 in MCI and probable AD participants. The black solid and dashed lines represent the median and inter-quartile range for the adjusted alpha-band power values independently reported in ref. 34. The heatmap shows that the model predictions from asymptomatic participant data are consistent with the empirical measures of global Aβ, temporal meta-ROI tau and alpha-band power reported in these two independent samples of participants with MCI and probable AD.

Full size image

Discussion

We report the first observations of interactive effects between early Aβ and tau pathology on cortical neurophysiological activity in asymptomatic adults with a family history of sporadic AD, and their associations with longitudinal cognitive decline. In line with predictions from in vivo and computational disease models, our data show that Aβ deposition and neurophysiological enhancement of fast-frequency activity colocalize in the human cortex, before the emergence of clinical symptoms. We propose that when tau also starts to accumulate in the temporal lobe, the initial acceleration of neurophysiological activity related to Aβ deposition shifts towards neurophysiological slowing, accompanying cognitive decline. Although our interpretations are limited by the cross-sectional nature of our study, this body of results provides empirical evidence in humans for the hypothesized analogous hyperactive and hypoactive effects of Aβ and tau accumulation manifested as macroscopic spectral power changes in neurophysiological activity along the AD continuum, which has been suggested by decades of animal, late-stage human and computational modeling research. Our model estimates are also consistent with previously-reported levels of proteinopathy and neurophysiological changes observed later in the AD continuum (that is MCI and AD). As such, we anticipate that the present findings will inform more accurate models of AD progression and aid in the future identification of new prognostics and preventative targets.

Our study points at possible nonlinear interactions between AD proteinopathy and human cortical neurophysiology in asymptomatic individuals at risk of developing the disease. Our results are aligned with fMRI studies of prodromal AD26,28,29 and confirm the predictions from recent computational models of preclinical AD41. We show that, as predicted from these models, early Aβ deposition is associated with expressions of neurophysiological enhancement of fast-frequency activity, as indexed by increased alpha-band and decreased delta-band activity41,42. Early tau deposition in medial–temporal regions can be hypothesized to synergistically trigger further pathological processes with Aβ41,45, which our data suggest would manifest as a shift in neurophysiological activity. This shift occurs from a regime of accelerated oscillatory activity in the preclinical stage of the disease to a generalized slowing of activity in the subsequent prodromal and clinical stages of AD31,32,33,34,39. This effect was not present when using whole-brain tau measures, which emphasizes the pathological significance of early tau deposition in medial–temporal regions. However, our data also suggest that the shift in neurophysiological activity does affect the cortex broadly and beyond medio-temporal regions, as a possible precursor of the widespread slowing observed in the later phases of AD39,40. We found that this shift in neurophysiological dynamics is essentially driven by rhythmic brain activity, and not by broadband arrhythmic background activity.

Our study also provides direct evidence from human neurophysiology of analogous phenomena previously observed only in animal models of AD. Notably, our data support model predictions of how early, microscopic alterations at the neuronal circuit level also manifest as aberrant patterns at the macroscopic scale of neurophysiological activity. Computational models of AD pathophysiology predict that Aβ deposits induce toxic effects on the dynamics of neuronal circuits, by simultaneously increasing excitation and decreasing inhibition, leading to augmented levels of alpha-band activity41. According to these models, subsequent tau deposition damages axonal connections and reduces excitatory transmission, driving the neuronal dynamics towards a hypoactive regime, which manifests with decreased alpha-band activity at the macroscopic level. Our results support these predictions in human empirical data and are also in agreement with recent findings showing that macroscopic alterations of the neurophysiological frequency spectrum might be attributable to differential effects on the time constants of excitatory and inhibitory neuronal populations46. It is important to note that at least one modeling study has suggested that increased neuronal excitability in AD would instead be expressed as neurophysiological slowing at the macroscopic level47. Future experimental and computational modeling research elucidating the link between micro- and macroscopic expressions of aberrant neurophysiological activity is needed to fully contextualize our results.

Previous studies in patients with MCI and probable AD have reported that delta-band neurophysiological activity tracks disease progression along the AD continuum33,34,42,48, and that the accumulation of Aβ42 and postmortem tau pathology32 are related to the magnitude of alpha-band activity, showing reduced alpha-band activity in later-stage AD32,33,42. Our results provide the first observations of this type in asymptomatic participants with a family history of AD, before they show cognitive symptoms. Notably, we demonstrate that in this group, there is a synergistic association of Aβ and tau pathologies with alpha-band neurophysiological activity, which is related to cognitive decline in attention and memory32,33,42. Reduced levels of alpha-band activity may precede subsequent increases in slower brain activity in the delta and theta bands, which are linked to impairments in additional cognitive domains, such as processing speed34. We further show that this model is also consistent with independent observations in the later stages of Aβ and tau pathology. These observations support the notion that early neurophysiological expressions of enhanced fast-frequency activity might presage the subsequent shift towards neurophysiological slowing and cognitive decline, as Aβ and tau continue to accumulate. However, the cross-sectional nature of our study limits our ability to draw definitive conclusions regarding the direction of these relationships. Future studies with longitudinal collection of neuroimaging data will further shed light on the temporality of AD pathological processes and changes in neurophysiological activity. We also look forward to future translational studies building on these new observations and proposed mechanisms, for new diagnostics, prognostics, outcome monitoring tools and ideally the definition of modifiable therapeutic targets. Our research agenda is to extend the present body of work towards the identification of robust neurophysiological and protein-pathological features to predict the possible progression to clinical AD, as well as to track the evolution of neurophysiological changes with the progressive accumulation of Aβ and tau, as we continue to follow the present cohort longitudinally.

We report synergistic associations between the early pathological deposition of Aβ and tau, and macroscopic neurophysiological activity in asymptomatic humans at risk of developing sporadic AD. By combining time-resolved MEG cortical mapping with Aβ and tau PET imaging, we demonstrate that Aβ depositions parallel an enhancement of fast-frequency neurophysiological activity in the asymptomatic human brain, and that tau deposition promotes a shift towards expressions of neurophysiological slowing related to longitudinal cognitive declines of attention and memory. These results support the long-standing hypothesis that Aβ and tau differentially affect neural activity in the earliest stages of the AD continuum and provide foundations for future research on models and predictors of AD-related neurophysiology.

Methods

Participants

The participants were from the PRe-symptomatic EValuation of Experimental or Novel Treatments for Alzheimer’s Disease (PREVENT-AD) cohort49, a large sample of asymptomatic middle-aged and older individuals with elevated familial risk of sporadic AD, defined as having one parent or multiple siblings affected by the disease50,51. As part of the inclusion criteria for the PREVENT-AD study, participants were required to (1) be at least 60 years of age or between 55 and 59 years if their age was 15 years younger than their first-affected relative’s age at dementia onset, (2) have no history of major psychiatric or neurological disorders and (3) be rated as cognitively normal at the time of enrollment. Normal cognition at enrollment was determined by having a score ≥ 26 in the Montreal Cognitive Assessment52 and a score of 0 in the Clinical Dementia Rating scale53. Normal cognition at the time of the MEG–PET sessions was determined by having a score ≥ 24 in the mini-mental state examination54. Participants whose scores exceeded the thresholds were further examined by a neuropsychologist from the PREVENT-AD group to verify their cognitive status. We retrieved data from a subsample of 124 PREVENT-AD participants, who all underwent Aβ and tau PET and resting-state MEG between 2017 and 2019 (ref. 49). No statistical methods were used to predetermine sample size, but our sample size is similar55 or larger34,42 to that reported in previous publications including multimodal imaging data (that is MEG, PET) in similar populations. From this subsample, the data from 20 participants were discarded, either due to issues with MEG data integrity (1 participant), PET data quality (1 participant) or unusable MEG data due to high-amplitude noise caused by ferromagnetic dental implants (18 participants). The final sample was therefore 104 participants (mean age = 67.4 years, s.d. = 4.9; 74 females; mean years of education = 15, s.d. = 3.1), who all provided informed consent that their de-identified data be used for research purposes reviewed and approved by the Institutional Review Board at McGill University, as the present study, with protocols compliant with the Declaration of Helsinki (protocol A05-B16-11B and A05-M55-11B). Participants received compensation to cover travel expenses and time. MEG data collection was performed blind to the Aβ/tau status of the participants. Data analysis was not performed blind to the conditions of the experiment, but group allocation was only implemented in the supplementary between-participants analysis.

Neuropsychological testing

The participants underwent yearly cognitive assessments using one of the four versions of the RBANS56. The RBANS is a practical neuropsychological tool to characterize abnormal cognitive decline in older adults. It yields scaled scores subdivided into five cognitive domains comprising immediate and delayed memory, attention, visuospatial constructional abilities and language. Longitudinal changes in cognition were assessed by estimating the linear slopes from the RBANS scores, computed using all available cognition timepoints obtained either before or after the MEG and PET visits (mean number of longitudinal cognitive assessments per participant = 6.6, s.d. = 2). We then normalized these slopes based on the average number of days between visits, and scaled the values to represent the annualized change in cognition. Considering that our participants were cognitively unimpaired at the time of the scans, and that the earliest cognitive impairments in AD are typically observed in the memory and attention domains40,57,58, we averaged the estimated linear slopes from the attention and memory RBANS scores to derive a composite measure of cognitive decline. As a follow-up analysis, we also assessed each cognitive domain (attention, immediate and delayed memory) separately. To ensure that the effects reported are not biased by the variability of when the MEG visit occurred along this longitudinal trajectory, the relative temporal distance between the dates of the first cognitive assessment and the MEG visit (mean number of days = 1282.4, s.d. = 569.6) was included as a nuisance covariate in all statistical models involving longitudinal cognition.

Neuroimaging data acquisition and analysis

MRI

Structural MRIs were acquired at the Brain Imaging Center of the Douglas Research Centre (Montreal, Quebec, Canada) using a 3T Siemens TIM Trio Scanner equipped with a 12 or 32-channel coil (Siemens Medical Solutions). T1-weighted images were acquired using a MPRAGE sequence with the following parameters: 176 slices (1-mm slice thickness), repetition time = 2,300 ms, echo time = 2.98 ms, flip angle = 9°, FoV = 256 × 240 × 176 mm and voxel size = 1 mm3. In the PREVENT-AD protocol, MRI data are collected annually as part of the longitudinal follow-up51. Here we used the structural scan acquired the closest to the MEG–PET visit (mean difference in days = 318.6, s.d. = 203.7).

We obtained the cortical surface from the T1-weighted structural MRI volumes of each individual using the FreeSurfer software package (version 5.3)59. We then parcellated the cortical surfaces according to the Desikan–Killiany atlas, comprising 34 × 2 bilateral cortical parcels60. This parcellation of individual cortical surfaces was used for MEG source mapping and SUVR quatification of the Aβ and tau PET data.

PET

PET imaging data were collected at the McConnell Brain Imaging Centre of the Montreal Neurological Institute and Hospital (Montreal, Quebec, Canada) using a Siemens HRRT head-only, high-resolution PET camera. Aβ scans were performed 40–70 min after injection of ~6 mCi of [18F] NAV4694 (Navidea Biopharmaceuticals), and tau scans were obtained 80–100 min after injection of ~10 mCi [18F] AV-1451 (FTP; Eli Lilly and Company). An attenuation scan was also acquired. Images were reconstructed using a 3D OP OSEM61,62 (10 iterations, 16 subsets) algorithm, and were decay and motion corrected. Scatter correction was performed using a 3D scatter estimation method62. For most participants (n = 96), the two PET scans were acquired one day apart, and all but one were collected less than 5 months apart (mean = 6.9 days, s.d. = 29.3 days).

PET images were preprocessed using an in-house pipeline from the Villeneuve Lab (https://github.com/villeneuvelab/vlpp), as described in ref. 49. Briefly, the 4D image files (six frames of 5 min for NAV and four frames of 5 min for FTP) were realigned, averaged, and registered to their corresponding structural MRI. SUVR maps were generated by using the cerebellum gray matter as a reference region for Aβ scans, and the inferior cerebellum gray matter for tau scans. The PET signal was then averaged across each parcel from the Desikan–Killiany atlas. Considering the characteristics of our cohort, we report Aβ uptakes across the whole-brain and tau PET signal in the entorhinal cortex, because of evidence of early AD-related tau depositions in that region6,7,43. To account for the limitations of using a single small cortical region such as the entorhinal cortex for tau quantification, we repeated all analyses using an early tau-sensitive temporal meta-ROI (Extended Data Fig. 2), obtained by averaging SUVR values over the entorhinal cortex, fusiform gyrus, parahippocampal gyrus, lingual gyrus, and inferior and medial–temporal gyri43. Considering the high correlation between entorhinal and temporal meta-ROI tau SUVR values, we also repeated these analyses after excluding the entorhinal cortex from the temporal meta-ROI.

MEG

For most participants (n = 80) MEG was collected on the same day as one of the two PET visits using a 275-channel CTF system located inside a three-layer magnetically shielded room. The average time between MEG and PET data collection was 23 days for Aβ and 24.6 days for tau (s.d. = 57.4 and 60, respectively). Two 5-min runs (that is a total of 10 min) of resting state, eyes-open MEG data were collected from the participants, to derive robust measures of ongoing brain activity across the frequency spectrum63. We elected to use an eyes-open resting-state paradigm because, during eyes-open recordings, participants are less prone to experiencing drowsiness, which would confound the interpretation of neurophysiological slowing. Further, eye movements do occur behind the lids with the eyes closed, which are less stereotyped than eye blinks and therefore cause MEG artifacts that are more challenging to correct. Participants were instructed to sit upright while looking at a fixation cross displayed on a screen. The data sampling rate was 2,400 Hz, with a hardware anti-aliasing low-pass filter set at 600 Hz. Built-in third-order gradient compensation was applied to attenuate environmental noise. We collected about 100 scalp points in each participant with a Polhemus Fastrak device, including anatomical landmarks at the nasion and the left and right preauricular points. We monitored head movements during MEG data collection with head position indicators attached to the participants’ forehead and the left and right mastoids. We also collected reference signals for heartbeats and eye movements with concurrent electrocardiographic and electrooculographic recordings. Empty-room recordings (about 2 min) were obtained at the beginning of all MEG visits to characterize environmental noise at the time of each individual session.

We used Brainstorm3 for MEG data preprocessing and analysis64,65 and followed recommended good-practice procedures66. The data were notch filtered to remove power-line noise and harmonics (60, 120, 180, 240 and 300 Hz), and high-pass filtered above 0.3 Hz to remove low-frequency drifts and MEG-sensor DC-offset. We inspected all MEG time series for bad channels and segments with prominent movement artifacts, which were marked and excluded from further analysis. Cardiac and eyeblink events were detected from their respective reference signals using an automated procedure64. We derived specific signal space projectors to attenuate eyeblink and heartbeat contaminants, and other stereotyped artifacts when necessary. We time segmented the MEG recordings into 4-s epochs, excluding segments that still contained major artifacts, as identified from the union of two standardized thresholds of ±3 median absolute deviations from the median—one for peak-to-peak signal amplitude and one for signal gradients.

MEG data were registered to individual structural T1-weighted scans using the scalp’s digitized points. Using Brainstorm3, we obtained a MEG forward model (overlapping spheres) from 15,000 cortical locations for each participant. We estimated individualized sensor noise covariances from the empty-room recordings, which we then used to derive the participants’ imaging kernels of depth-weighted dynamic statistical parametric mapping of cortical current flows oriented perpendicularly to the cortex67. We then derived the power spectrum density for each 4-s epoch of the cortical time series using Welch’s estimator (2-s windows, 50% overlap). We then averaged all available epochs from both MEG runs to produce cortical maps of frequency band-specific activity in canonical frequency bands—delta (2–4 Hz), theta (5–7 Hz), alpha (8–12 Hz) and beta (15–29 Hz), each scaled relatively to the total power across all bands. Finally, we extracted the average relative signal power values in each frequency band of interest, and for each of the 68 cortical regions of the Desikan–Killiany atlas.

Statistics and reproducibility

Joint MEG–PET analysis

We first performed a between-participants analysis to test for potential whole-brain changes of neurophysiological (MEG) activity related to the pathological deposition of cortical Aβ and entorhinal tau. We derived the mean signal power across all cortical regions for each frequency band in the following three groups of participants: (1) participants with no Aβ nor entorhinal tau pathology (Aβ−/Tau−, n = 50), (2) participants with significant global Aβ burden but no entorhinal tau (Aβ+/Tau−, n = 42) and (3) participants with both global Aβ and entorhinal tau burden (Aβ+/Tau+, n = 11). The only participant (male) who was classified as Aβ−/Tau+ was excluded from the analysis. Aβ positivity was defined based on the average SUVR computed over a set of early Aβ sensitive regions (including the precuneus, posterior cingulate, parietal, frontal and lateral temporal cortices)11. Individuals with a global Aβ SUVR >1.18 (13.3 in Centiloid scale) were rated as Aβ+. This threshold was derived by estimating two s.d. above the mean global Aβ SUVR calculated from a normative subset of 11 young individuals who underwent the same PET protocol. Similarly, tau positivity was defined using a threshold of entorhinal SUVR >1.25, corresponding to two s.d. above the entorhinal SUVR values obtained for the young participants. We used the Levene test to assess variance homogeneity between our PET-defined groups, which showed that all of our models fulfilled this assumption (all P > 0.05). We ran an analysis of covariance (ANCOVA) to test whether the neurophysiological spectrum differed across groups while accounting for age, sex, years of education, hippocampal volume and the number of MEG epochs included as nuisance variables. We applied FDR corrections to account for the multiple (four) statistical tests performed across frequency bands. We then used bootstrapped resampling to derive confidence intervals around the mean statistics for each participant group while keeping the unbalanced proportion of individuals. To assess the robustness of these associations considering the small number of individuals within each group, we evaluated the presence of outliers by calculating the Cook’s distance, a summary metric of how each observation influences the regression model. We repeated the analysis for each frequency band after removing the influential cases, defined as those having a Cook distance above three times the mean distance of all data points included in the model.

Considering that the residuals for some frequency bands (that is theta and alpha) were not normally distributed, and the issues around the imbalance in sample sizes between the PET subgroups defined in our between-participants analysis approach, we designed nested LME models to test the within-participants association between band-specific neurophysiological activity and PET markers of AD pathology, accounting for potential confounds of age, sex, years of education, hippocampal volume and number of MEG epochs included per participant. Specifically, we tested for consistent within-participant relationships between Aβ deposition and neurophysiological activity across all cortical regions (neurophys. ~Aβ + age + sex + education + hippocampal volume + number of MEG epochs, random = ~1|participant), as well as the interactive effect of entorhinal tau deposition on these relationships across individuals (neurophys. ~Aβ × Tau + age + sex + education + hippocampal volume + number of MEG epochs, random = ~1 | participant). Both Aβ and entorhinal tau were treated as continuous variables for this analysis. We repeated our Aβ × Tau models by using the temporal meta-ROI instead of the entorhinal cortex tau SUVR values, and also by incorporating whole-brain estimates of tau deposition across all cortical regions as a nested variable of interest (neurophys. ~Aβ × regional Tau + age + sex + education + hippocampal volume + number of MEG epochs, random = ~1 | participant).

To visualize the synergistic interaction of tau deposition with the Aβ-neurophysiology relationships, we plotted the association between Aβ and neurophysiological activity separately for participants in the Tau+ and Tau− groups, based on the threshold defined for the between-participant analysis. For the models that included whole-brain regional tau as a nested variable, we showed the interaction effect by plotting the line of best fit obtained by separating the cortical regions exhibiting the highest and lowest (that is upper and lower quartiles) levels of tau within each participant. To ensure robustness, we performed 1,000 nonparametric permutations by randomly shuffling the Aβ SUVR values across the 68 cortical regions of each individual while keeping the MEG maps and all other covariates intact. For the models including whole-brain tau as a nested variable, we ran the permutations by shuffling between pairs of Aβ and tau estimates to preserve any interactive association between data points. From these permuted models, we then built a null distribution of the t statistic obtained for the relevant main effect (that is Aβ) or the interaction of Aβ × Tau and calculated the probability (P value) of obtaining the comparable statistic from the original model. We implemented FDR corrections to adjust the permuted P values accounting for the multiple statistical comparisons across the four frequency bands of interest. To test whether these relationships were confounded by global spectral changes of aperiodic neurophysiological activity, we repeated all analyses with LME models that included the region-wise aperiodic slope and offset of the neurophysiological power spectrum as a nuisance covariate68,69. We parameterized the power spectrum data obtained at each atlas region by using the specparam module in Brainstorm3 (Matlab, version R2021b) using the recommended default parameters69—frequency range 1–40 Hz, Gaussian peak model with peak width limits between 0.5 and 12 Hz, and a maximum number of three detectable peaks, a minimum peak height of 0.3 dB and a proximity threshold equivalent to two s.d. from the largest peak, with a fixed aperiodic and no guess weight. As mentioned above, we also repeated all LME analyses using the tau SUVR values from the temporal meta-ROI instead of those from the entorhinal cortex and the nested tau estimates across all cortical regions. All statistical analyses were computed using R (version 4.1.1)70. LME models were computed using the nlme package (version 3.1.152)71, ggplot2 (version 3.4.0)72 and ggseg (version 1.6.5)73 for visualization.

Association with cognitive decline

We first assessed whether longitudinal cognition differed across Aβ−/Tau−, Aβ+/Tau− and Aβ+/Tau+ individuals, by running an ANCOVA and posthoc Tukey’s Honestly Significant Differences (HSD) tests. We then tested whether the strength of any significant relationship between neurophysiological activity and AD proteinopathy was associated with longitudinal changes in cognitive performance. For each participant, we calculated the Fisher-transformed (atanh function in R) Pearson correlation coefficient between Aβ concentrations and neurophysiological signal power across all cortical regions. We restricted the analysis to the alpha and delta frequency bands because they showed significant associations with AD pathology. We fitted a linear model in R between the longitudinal cognitive scores described in neuropsychological testing above and the continuous Aβ–neurophysiology correlation coefficients, with tau concentrations as a moderating factor (lm function; cognitive slope ~(Aβ-neurophys. correlation) × Tau + age + sex + education + hippocampal volume + number of MEG epochs + time between MEG and the first cognition timepoint). We performed these analyses with 102 individuals, after excluding the Aβ−/Tau+ participant and the participant (female) who had completed only one cognitive assessment. We then applied FDR corrections to account for the four statistical comparisons across the composite attention–memory scores and the individual cognitive domains. Once again, we identified influential cases for each regression model using Cook’s distance approach described above and recomputed the analyses after removing individuals identified as potential outliers.

Predicting neurophysiological changes in later disease stages

Finally, we tested whether our linear mixed-effects model estimates matched the levels of neurophysiological changes reported in later disease stages. We used the coefficients from our linear mixed-effect model relating region-wise Aβ, temporal meta-ROI tau deposition and relative alpha-band power to estimate whole-brain relative alpha-band power values across a wider range of increasing levels of Aβ and tau deposition while keeping all other covariates constant (by multiplying their model coefficients by the median values calculated across all participants). To facilitate direct comparison with other studies, we expressed our global Aβ deposition estimates in Centiloid units for visualization, which was achieved by converting our NAV4694 SUVR values following the procedures described in ref. 74 to validate and adapt the NAV4694 SUVR to Centiloid formula proposed in ref. 75. We then compared these estimates against the median global Aβ, temporal meta-ROI tau SUVR and whole-brain relative alpha-band power values reported in two independent studies of participants with MCI and AD34,44. The participant samples were similar to our cohort in terms of age (PREVENT-AD—median = 66.3, IQR = 64.2, 70.6; ref. 44—median = 72.7, IQR = 65.8, 80.1; ref. 34—median = 70, IQR = 64, 74). For Aβ and tau, we used the estimates reported in ref. 44, calculated across 272 cognitively impaired patients. We calculated the weighted mean of the global Aβ Centiloid and tau meta-ROI flortaucipir SUVR values reported for the MCI and AD groups used in this work. For the relative alpha-band power estimates, we used the median and quartiles from 38 MCI and mild probable AD patients reported in ref. 34. Because these measures were from eyes-closed resting-state data, and our data were collected with eyes-open, we adjusted the values from ref. 34 with the multiplicative eyes-open to eyes-closed ratio reported in ref. 76 (median = 0.21, IQR = 0.16, 0.26). We then overlayed the median and quartile estimates of Aβ and tau from ref. 44 and the adjusted relative alpha-band power from ref. 34 on top of a heatmap produced from our model estimates (Fig. 5). We also verified that the alignment between the three independent datasets was similar when using a more conservative eyes-open to eyes-closed correction ratio77 (median = 0.16, IQR = 0.11, 0.21) and when using no correction (median = 0.24, IQR = 0.19, 0.29).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.