High power transient 15-29Hz beta event features as early biomarkers of Alzheimer’s Disease conversion: a MEG study

2.1 Subject recruitment and neuropsychological assessment

Participants were recruited from Hospital Clínico Universitario San Carlos in Madrid, Spain. The study was approved by the Ethics Committee, and all participants provided written informed consent prior to participation. All participants were right-handed native Spanish speakers.

The study sample included 85 subjects diagnosed with mild cognitive impairment (MCI). Initially, participants were screened according to the diagnostic criteria of the National Institute on Aging-Alzheimer’s Association (NIA-AA) (Albert et al., 2013) and underwent a comprehensive neuropsychological assessment as previously described (López-Sanz et al., 2016; Pusil, Dimitriadis, et al., 2019), along with a MEG recording. They were cognitively and clinically followed-up in a temporal interval of 2.5 years and then subdivided in two groups considering the criteria for probable Alzheimer’s disease (McKhann et al., 2011): 41 subjects with mild cognitive impairment who converted to AD (CONV), and 44 subjects with mild cognitive impairment (MCI) which did not convert to AD (NOCONV). Subjects in CONV and NOCONV group were matched by age, sex, and education years, as it is reported in reported in Table 1.

The neuropsychological assessment included seven tests: four measures of memory recall, Immediate Logic Memory Units, Delayed Logic Memory Units, Immediate Logic Memory Themes and Delayed Logic Memory Themes (Wechsler Memory Scale,WMS-III) (Wechsler, 1955); one measure of working memory, Inverse Digits (WMS-III); one measure of cognitive flexibility, Trail Making Test-B (TMT-B) (Bowie & Harvey, 2006); two language measures, Semantic Fluency (Controlled oral Word Association Test, COWAT) (Benton et al., 1994) and Boston Naming Test (BNT) (Kaplan et al., 1983); and one global screening measure for cognitive impairment and dementia, the Clock-Drawing test (Agrell & Dehlin, 1998). Table 1 includes paired t-test and previously reported differences across groups (López-Sanz et al., 2016; Pusil, Dimitriadis, et al., 2019).

2.2 Magnetoencephalography data acquisition

The dataset was acquired using a 306-channel (102 magnetometers and 204 gradiometers) Vectorview MEG system (Elekta AB, Stockholm, Sweden) placed inside a magnetically shielded room (Vacuum-Schmelze GmbH, Hanau, Germany) located at the Laboratory of Cognitive and Computational Neuroscience (Madrid, Spain). MEG data consisted of 5 min eyes-closed resting-state recordings in a 60 min session, with a sampling rate of 1000 Hz and an online [0.1 - 330] Hz anti-alias band-pass filter. To allow further analysis, including subject-specific source reconstruction, MEG recordings were complemented by MRI scans acquired within a month after the MEG session, which were recorded at the Hospital Universitario Clínico San Carlos (Madrid, Spain) using a 1.5 T General Electric MRI scanner with a high-resolution antenna and a homogenization PURE filter (fast spoiled gradient echo sequence, with parameters: repetition time/echo time/inversion time = 11.2/4.2/450 ms; flip angle = 12°; slice thickness = 1 mm; 256×256 matrix; field of view = 256 mm).

The MEG recordings were preprocessed offline using a temporal-spatial filtering algorithm (tSSS) (Taulu & Hari, 2009) (Maxfilter Software v2.2, correlation limit of 0.9 and correlation window of 10 s) to eliminate magnetic noises and compensate for head movements during the recording. The continuous MEG data were imported into MATLAB (R2017b, Mathworks, Inc.) for pre-processing steps, carried out using the Fieldtrip Toolbox (Oostenveld et al., 2011) (https://www.fieldtriptoolbox.org/). Data were automatically scanned for ocular, muscle and jump artefacts using the Fieldtrip software. Artefacts were then visually confirmed by an MEG expert. The remaining artefact-free data were segmented into 4-s segments (epochs). An independent component analysis-based procedure was used to remove the heart magnetic field artefact. Source reconstruction was performed using minimum norm estimates (Hämäläinen & Ilmoniemi, 1994) with the software Brainstorm (Tadel et al., 2011). Current dipoles were constrained to be perpendicular to the individual’s cortical surface, to model the orientation of macro columns of pyramidal neurons (Tadel et al., 2011). Neural time series were finally averaged within regions of interest (ROI) of the Schaefer 100-17 network atlas (Schaefer et al., 2018). Two regions of interest were extracted for subsequent analysis: the anterior cingulate cortex (as the merge of left ACC and right ACC, corresponding to SalV entAttnB PFCmp 1 area of the mentioned atlas, respectively), and the precuneus (as the merge of left PC and right PC, corresponding to ContC pCun 1 and DefaultA pCunPCC 1 area of the mentioned atlas), see Figure 1a. The data was band-pass filtered between 0.5 and 45 Hz (broadband), using FIR filtering.

• Download figure
• Open in new tab

Figure 1:

a) Representation of the right side regions of interest using the Schaefer 100-17 network atlas: SalV entAttnB PFCmp 1 (corresponding to ACC), ContC pCun 1 (corresponding to PC) and DefaultA pCunPCC 1 (corresponding to PC); b) Example of resting-state beta activity from left-ACC and detection of transient burst of high-power activity (events), using the frequency-based algorithm of Shin et al. (2017). There are several possible features of such events that could contribute to increased power averaged across time and frequency, including event number (rate), event duration, event frequency span and event power.

2.3 Power spectral density (PSD)

We computed the power spectral density of each of the ROI time series by using the Welch’s periodogram method (Welch, 1967), with 1s window length and 50% overlap ratio. For each ROI signal, the normalised power was calculated by averaging the power spectral density obtained by each epoch and then normalising the value associated to each frequency by total power over the [1–30] Hz range (Figure 2).

• Download figure
• Open in new tab

Figure 2:

Normalized power spectral density plots (mean ± SEM) for regions of interest in CONV and NOCONV patients, in the range [2-30] Hz. Shaded grey areas represent statistically significant clusters of differences between CONV and NOCONV groups (non-parametric statistical test).

2.4 Time-Frequency Representations (TFRs) & Spectral Events Extraction

During resting-state alpha and beta activity, periods of transient high power can be quantified over time in unaveraged data. We use a time-frequency-based algorithm as in Shin et al. (2017) to capture these bursts.

Time-Frequency Representations (TFRs) were calculated using the MATLAB SpectralEvents Toolbox (find method = 1, as in Shin et al. (2017)) (https://github.com/jonescompneurolab/SpectralEvents). Each artefact-free 4-sec epoch was convolved with a 7-cycle Morlet wavelet. For all epochs, TFR across time and across each patient was calculated and finally averaged to obtain a representative TFR for each

2.5 Statistical analysis

To determine frequency ranges that represent oscillatory slowing in our cohort, we tested for significant differences in average power spectral density between AD CONV and NOCONV for each selected ROI by performing a non-parametric statistical test (Maris & Oostenveld, 2007) in the frequency domain (see light grey window in Figure 2).

In each frequency-band, we examined the relationship between event features and the averaged PSD using a linear regression analysis over all subjects. The regression β coefficients were calculated with a 95% CI and all p-values were corrected for multiple comparisons using the Benjamini–Hochberg (BH) step-up procedure (Benjamini & Hochberg, 1995) with a False Discovery Rate (Q) set at 0.05. Statistically significant p-values after BH correction are reported as ∗p < 0.05(Q = 0.05).

We also tested for group differences in spectral event features between CONV and NOCONV groups. High power spectral event features were detected in alpha and beta frequency bands for each resting state segment from four ROI’s (left and right ACC and PC, respectively). Event rates were averaged across epochs, and other features (duration, f-span, power) were averaged across all events, and then between hemispheres, within each subject. The final dataset consisted of 16 variables for each subject: four event features (averaged spectral event rate, duration, f-span and power); by two ROI’s (ACC, PC); by two frequency bands (alpha and beta). For each variable, t-tests were used to assess group differences in transient event features. The tests were conducted with a left tail for the beta band and a right tail for the alpha band. This approach was consistent with the observation that the average power spectra in the alpha band are higher for CONV compared to NOCONV, while the effect is reversed in the beta band. We hypothesized that differences in event features align with the directional trend observed in the averaged relative spectral power. All reported p-values were corrected for multiple comparisons across the number of tests applied using the Benjamini–Hochberg (BH) step-up procedure (Benjamini & Hochberg, 1995) with a False Discovery Rate (Q) set at 0.05. Statistically significant p-values after BH correction are reported as ∗ ∗ p < 0.05(Q = 0.05). Statistical tendencies are reported as significant if ∗p < 0.1. We computed effect size of the differences between groups with a robust variant of Cohen’s d (Algina et al., 2005).

For event features where significant difference across groups were found, we examined the relationship between these features and AD conversion and cognitive performance.

To assess the relationship with AD conversion, we fit beta event features from ACC and PC to a logistic regression model ( R glm() function, family = binomial argument) in order to predict each patient’s conversion label (CONV or NOCONV). The model’s output is the probability of the positive class (CONV group). We used a default 0.5 threshold value to transform the probability to a binary class. Thus, a subject is classified as class 1 (CONV) if the predicted probability is greater than or equal to 0.5, and class 0 if the probability is less than 0.5. The odds ratio in logistic regression is calculated by exponentiating the coefficient of the predictor variable. As an example, for predictor variable Number of events, if the coefficient is β = 0.58 the odds ratio is given by OR = e⁰.⁵⁸ ≈ 1.79. This means that the likelihood of the predicted outcome (AD conversion), is approximately 1.79 times greater than the likelihood of the outcome not occurring (no conversion to AD). Every reduction of 1 unit in Number of events increases by 79% the odds of AD conversion (CONV group).

To asses the relationship with cognitive performance, we calculated a Cognitive Performance Index and applied linear regression. Aiming to account for variability across the nine neuropsychological tests used, the Cognitive Performance Index was calculated by applying Principal Components Analysis (PCA) (R princomp() function) to the neuropsychological testing dataset. As a first step, the data from each of the nine tests was z-scored to prevent biases due to different test scales. Additionally, the data were transformed so that for each test higher values correspond to better cognitive performance. Then, we extracted the first principal component, which explained at least 55% of variance. The projection of individual data onto the new axes (principal component) represent the Cognitive Performance Index values for each subject, and this projection is achieved through linear transformation using the eigenvectors of the covariance matrix. All eigenvectors had a positive load in the first component. A subject with higher values of this Cognitive Performance Index correspond to better cognitive performance.

3.1 AD Converters (CONV) exhibit oscillatory slowing in average PSD

We confirm a slowing effect in power spectral density in CONV subjects for all ROIs of interest in ACC and PC, as observed in prior studies (Pusil, Dimitriadis, et al., 2019). In averaged PSD, the CONV group shows a statistically significant cluster of decreased relative power in beta frequency band compared to NOCONV (Figure 2, 17-30Hz ACC and 11-27Hz PC, blue underlines) and increased relative power in lower frequency bands (theta, low-alpha) (Figure 2, 5-7Hz ACC and 4-8Hz PC, red underlines). Based on our data, for all subsequent analyses, we utilized a [5-10] Hz alpha bands and [12-30] Hz beta-band (see Methods section 2.4 for further details).

3.2 Resting-state alpha and beta emerge as transient high-power events

The PSD plots of Figure 2 rely on Fourier analysis performed on averaged epochs of brain activity from the two different groups. As described previously, differences in averaged power across patient groups could emerge from several features in transient of high-power activity (i.e., events) in the unaveraged data.

Visual inspection of time-frequency representations in Figure 3 shows that seemingly continuous high average power in the alpha and beta bands (top panels) is the result of the accumulation of transient high power activity (’events’) across epochs of the unaveraged resting-state data (Figure 3 middle and bottom panels) for both groups.

• Download figure
• Open in new tab

Figure 3:

Time-Frequency representations of spectral events in PC (a), and ACC (b) for non-converters (NOCONV, left) and converters (CONV, right) groups. For each case, we show the representative averaged broadband TFR (top), as well as three TFR related to randomly chosen trials for low-alpha (middle), and beta (bottom) band.

As such, higher averaged power could be due to increased expression in several features, including a higher number of events (rate), longer duration events, increased frequency spans and or increased power of the event. To assess if these features contributed to averaged power in our sample, we performed a regression analysis between each feature and averaged alpha and beta power (Figure 4). Our results show a strong correlation with averaged power and each event features in the beta band (number: β = .28, R² = .069, p − value = .0088; duration: β = .45, R² = .192, p − value < .0001; max power: β = .41, R² = .156, p − value = .0001, frequency span: β = −.29, R² = .07, p − value = .0075). While similar relationships occurred between alpha event features and averaged alpha power, only alpha event frequency span significantly correlated with averaged alpha power after correction for multiple comparisons and the effect was weaker than in the beta band (frequency span: β = −.25, R² = .05, p − value = .025).

• Download figure
• Open in new tab

Figure 4:

Diminished expression of event features in ACC and PC is associated with reduced mean power spectral density (PSD). A) Regression analysis between alpha [5-10] Hz spectral events and averaged alpha power. B) Regression analysis between beta [12-30] Hz spectral events and averaged beta power. Statistically significant p-values after BH correction are reported as ∗p < 0.05(Q = 0.05).

3.3 Reduced beta event features as predictive biomarkers of AD conversion and cognitive decline

Given the event like nature of alpha and beta, we next tested the hypothesis that the slowing observed differences across groups in averaged PSD power in Figure 2 was due to across group differences in event features, namely rate, power, duration and/or frequency-span. We found effects only in the beta band that were more prominent in PC, such that CONV had a lower mean rate of beta events, lower power events, and a trend to shorter duration beta events in PC. Shorter duration beta events were also present in ACC in the CONV group (Figure 5 and Table 2; rate in PC p − value_BH = .003, power in PC p − value_BH = .002, duration in PC p − value_BH = .051; duration in ACC p − value_BH = .01, see also Supplementary Figure 2 and Figure 3 for further summary statistics). Notably, the effect size for differences between CONV and NOCONV in these features are of medium size (Cohen d > 0.4), see Table 2).

• Download figure
• Open in new tab

Figure 5:

Mean and distribution of event features for AD converters (CN) and non-converters (NC) in alpha [5-10] Hz and beta [12-30] Hz frequency bands. T-test statistically significant p-values ( p < 0.05) after BH correction are marked with the asterisk (**). Statistical tendency ( p < 0.1) are marked with the asterisk (*).

The fact that our method allows us to explain differences in beta band, but not in alpha band, could reflect the method’s sensitivity in the alpha band. Indeed, as shown in Figure 4, in this subject sample, individual alpha event features are not a strong predictor of averaged alpha power (see also Supplementary Tables 1 & 2 for analysis of other power thresholds) and the slowing effects in the alpha band may instead be due to a combination of transient event features and/or more stationary properties of the signal.

To investigate the further predictive potential of resting-state beta event features as a biomarker for conversion from MCI to AD, we examined the association between beta event rate, duration, and maximum power and the probability of AD conversion within a 2.5-year timeframe (See Figure 6A). The probability of conversion was calculated from a logistic regression (see Methods section 2.5). Consistent with the pooled results in Figure 5, lower beta event rates, shorter duration, and reduced event power are linearly associated with an increased probability of conversion. MCI subjects with fewer than 4.92 events in four seconds of resting-state data have 1.79 (95 %CI [1.1, 3.0]) times greater odds of converting to AD (R² = .10, p − value = .024). Those with event duration less than 159 ms have 1.06 (95 %CI [1.0, 1.1]) times greater odds of AD conversion (R² = .06, p − value = .008), and subjects whose events have maxima power (calculated as factors of median power) less than 8.41 have 2.01 (95 %CI [1.15, 3.76]) times greater odds of AD conversion (R² = .05, p − value = .019).

• Download figure
• Open in new tab

Figure 6:

Diminished expression of beta event features in ACC and PC (number of events, duration, maxima power) is associated with greater odds of future AD conversion, and reduced cognitive performance. A) Logistic regression between beta event features and AD conversion within 2.5 years. B) Linear regression between beta event features and Cognitive Performance Index separated by CONV (red) and NOCONV (blue).

Further examination of the relationship between beta event features and cognitive performance showed that, in the CONV group, a lower rate of beta events and lower power beta events were linearly associated with greater cognitive decline (Figure 6B, lower Cognitive Performance Index values, see Methods section 2.5 Statistical Analysis) (β = .26, R² = .07, p − value = .016 for number of events; β = .31, R² = .10, p − value = .0041 for maxima power). These relationships did not emerge for duration, nor appear in NOCONVs, suggesting beta event features and particularly beta event power (p − value = 0.0041) are predictive of cognitive decline only in MCI patients that will convert to AD withing 2.5 years.

4.1 Consistency with prior studies examining beta, ageing and cognitive control

Reductions in beta event expression yielded a lower PSD in the beta band [12-30] Hz averaged across trials, giving further insight into the underpinning of oscillatory slowing in AD (Brunã et al., 2023; Dauwels et al., 2011; Jelic et al., 2000). This trend is also consistent with a critical shift in beta activity at approximately 60 years of age in healthy patients (Brady et al., 2020; Power & Bardouille, 2021). Following this inflection point, resting-state relative source power, as well as beta event characteristics such as event rate, peak frequency, duration, peak power, etc., progressively begin to decline with age (Brady & Bardouille, 2022). Other studies have shown that averaged M/EEG spectral activity in the beta frequency range (13-30) Hz is a more powerful predictor of MCI-to-AD conversion than activity in other frequency bands, including in the slow alpha frequency range ((Gaubert et al., 2024; Poil et al., 2013). Thus, transient beta event features may be key factors in delineating differences and tracking the neurophysiology of healthy and pathological ageing.

We found that reductions in beta event features were associated with global cognitive decline, as measured by an index based on a battery of neuropsychological tests. However, this effect is only significant in the CONV group. MCIs who convert to AD 2.5 years later seem to be more reliant on beta event expression for cognitive robustness.

Several studies have shown that beta events throughout the cortex are a signature of inhibitory control. In sensory cortex, beta event expression can be manipulated with attention (increasing in non-attentive states), and an increase in beta event rates is associated with a decrease in perceptual salience (Shin et al., 2017), a process predicted to be mediated by an increase in inhibitory neuron activity (Law et al., 2022). In motor cortex, increased beta event expression is associated with inhibited motor control and Parkinson’s disease (Yu et al., 2021), and in frontal cortex beta events are a signature of stopping of movement and long-term memory retrieval (Schmidt et al., 2019; Wessel, 2020). Conversely, reduction in beta event rates in frontal cortex has been explicitly linked to encoding and decoding in working memory processes, where they have been suggested as a mechanism for volitional control and memory content reactivation (Lundqvist et al., 2016, 2018; Spitzer & Haegens, 2017). Likewise, decreased posterior parietal beta oscillatory activity predicts episodic memory formation and retrieval (Griffiths, Martín-Buro, Staresina, Hanslmayr, & Staudigl, 2021; Nyhus, 2018), and is correlated with enhanced memory performance (Griffiths, Martín-Buro, Staresina, & Hanslmayr, 2021). Transcranial stimulation of the prefrontal cortex at beta (18 Hz) has been found to induce memory encoding impairments (Hanslmayr et al., 2014), and during posterior parietal cortex stimulation, lower prestimulus beta power predicts higher phosphene ratings, reflecting increased neural excitability (Samaha et al., 2017).

The consistent relationship between beta event expression and inhibitory control in these myriad studies (Lundqvist et al., 2024) suggests that the ability to modulate beta events according to task demands is necessary for optimal function, and that the diminished resting state beta event expression and associated cognitive decline observed in the PC in CONV in our study may be directly related to lack of inhibitory cognitive control.

4.2 Why are differences in alpha bursts not present?

Despite the observation of significant across group differences in alpha band in the average PSD (Figure 2), we did not find a significant relationship between transient [5-10] Hz alpha event features and averaged power, nor did we find a difference in event features between MCI patients who will convert to AD and those who will not, which is consistent with previous longitudinal studies of AD progression (Gaubert et al., 2024; Poil et al., 2013). This was true for several event detection thresholds (see Supplementary Figure 2 and Supplementary Tables 1 and 2).

Since we did not find a relationship between spectral events in the lower 5-10 Hz alpha band and averaged alpha power in our sample, we did not examine the relationship to cognitive performance. In terms of components of slowing, other studies have found a relationship between bursts of slow activity (1-6 Hz) age and cognition in healthy adults, where older subjects and lower cognitive performance participants exhibited longer and slower events (Power et al., 2024).

The lack of alpha event effects in our sample could reflect the sensitivity limitations of the spectral event detection method or indicate that, although the alpha band contains transient components, the differences between the two groups involve a combination of transient feature or more stationary features than transient ones. A slowed occipital alpha rhythm is a common biomarker of several neurological or psychiatric disorders (Hughes & Crunelli, 2005; Samson-Dollfus et al., 1997). According to the “thalamocortical (TC) hypothesis” (Klimesch et al., 2007; S. M. Sherman, 2001) slow cortical rhythms could be generated by TC cells in tonic, single spike, firing, or in arhythmic bursting mode. Hugues (2005) argued that a slowing effect when shifting from alpha to theta waves could arise from the hyperpolarization of the thalamo-cortical neuron population, resulting in a deceleration of high-threshold bursting (HT) in individual cells, which could translate to a shift from bursty alpha to more continuous slower theta range oscillations. Thus, in this cognitively impaired population, while we still detect some alpha bursting, Alzheimer’s disease-related mechanisms may be disrupting thalamo-cortical connections (Eustache et al., 2016) and slowing the thalamic bursting activity, making it more stationary.

In the early stages of AD continuum, even before a mild cognitive impairment diagnosis, alpha disruption is a marker of cognitive decline (Babiloni et al., 2010; Brunã et al., 2023; Huang et al., 2000; López et al., 2020; López-Sanz et al., 2016). Still, when patients reach a more advanced stage of the disease, as it is mild cognitive impairment, beta band is more discriminant between AD converters and non-converters (Gaubert et al., 2024; Poil et al., 2013; Pusil, Dimitriadis, et al., 2019). Consequently, when measured in MCIs the spectral events method may not be sensitive to this transient-to-stationary shift in the alpha [5-10] Hz frequency band but can detect differences in the beta [12-30] Hz band.

4.3 A mechanistic link between reduced beta event expression and hyperexcitability in CONV

Animal studies have suggested Aβ-induced changes in E/I balance is a mechanism for hyperexcitability and cognitive decline in Alzheimer’s disease (Maestú et al., 2021). This notion has been supported by several computational neural modeling frameworks examining causal links between disease processes and increased neuron firing rates (Alexandersen et al., 2023; Cabral et al., 2014; de Haan et al., 2017; Hutt et al., 2023; Nakagawa et al., 2014; Stefanovski et al., 2019; Zimmermann et al., 2018), albeit without explicit consideration of the biophysical generators of MEG current sources or of beta frequency oscillations.

Modeling work by our group specifically designed to interpret the detailed cell and circuit origin of localized MEG (and EEG) current source signals (Neymotin et al., 2020) provides a hypothesized direct mechanistic link between beta event expression and hyperexcitability in CONV. Specifically, our prior modeling and cross-species empirical studies suggested neocortical beta events are generated by bursts of exogenous thalamocortical drive that targets excitatory synapses on the proximal and distal dendrites of pyramidal neurons in deep and superficial neocortical layers, such that the distal drive is stronger and last a beta period (i.e. ≈ 50ms) (Jones et al., 2009; M. A. Sherman et al., 2016). This thalamic burst drive induces current flow in pyramidal neuron dendrites that generates MEG beta event waveform characteristics consistent with those observed experimentally in sensory, motor and frontal cortices (Bonaiuto et al., 2021; M. A. Sherman et al., 2016), and when occurring rhythmically can produce multiple beta cycles and/or a complex of alpha and beta activity (Jones et al., 2009). Follow-up studies predicted further that the thalamic drive inducing a beta event also activates inhibitory neurons in supragranular layers, providing a causal mechanism for beta-associated inhibitory control (Law et al., 2022; Shin et al., 2017). Together with our current findings that CONV have reduced beta expression (namely lower event rates, power and duration), these prior studies suggest CONV have a reduction in thalamocortical burst drive to cortex, which in turn recruits less cortical inhibitory neuron activity leading to hyperexcitability.

Such a decrease in inhibitory activity could contribute to the observation of diminished Gabaeric terminals on cortical neurons near amyloid plaques, in addition to the toxic effects of amyloid oligomers on inhibitory terminals (Alexandersen et al., 2023; Garcia-Marin et al., 2009), particularly in PC which exhibits Aβ deposition in the early stages of AD. The predicted reduction in thalamic bursting is also synergistic with studies showing thalamic atrophy and reduced inhibitory thalamic tone (Abuhassan et al., 2014; Forno et al., 2023), as inhibition from the thalamic reticular nucleus is known to be a driver of rebound bursting mechanisms in thalamic relay cells (Destexhe & Sejnowski, 2002). Moreover, decreases in thalamocortical drive are consistent with theories of thalamocortical dysrhythmia (Llińas et al., 1999) and could contribute to loss of cortical signal complexity and biophysical heterogeneity (Szul et al., 2023) occurring with the disease and thought to be an important homeostatic control mechanisms capable of bolstering the network’s resilience to perturbations, such as the toxic effects of amyloid (Dauwels et al., 2011; Hutt et al., 2023).