Neural flexibility is higher in Alzheimer’s Disease and predicts Alzheimer’s disease transition

Abstract

Background:

Neural flexibility (NF), a measure of dynamic functional connectivity, was associated with psychiatric diseases but has not yet been studied in Alzheimer’s Disease (AD).

Objective:

We aim to evaluate whether AD is associated with alterations in NF and probe its predictive utility for AD conversion.

Method:

The study included 862 older adults (461 CN, 294 MCI, 107 AD) with valid resting-state fMRI data from the Alzheimer’s Disease Neuroimaging Initiative. We defined the NF of a node as the number of times that a node changed its community assignment across the sliding windows, normalized by the total number of possible changes. We computed global NF and 12 functional Network-specific NFs, then performed linear mixed models on NFs separately to explore the differences in these measures between our three groups. Finally, we evaluated the predictive utility of NF on dementia transition using survival analysis.

Results:

NF is significantly higher in AD than CN on global NF (β=0.002, 95% CI 0.001 to 0.004), and NF in six networks, and NF is significantly higher in MCI than CN in the visual network. Among n=617 non-demented participants at baseline, n=53 (8.6%) participants converted to dementia during the follow-up visits. Higher NF in the visual network was positively associated with AD transition (HR=1.323, 95%CI 1.002 to 1.747, p=0.049, per 1 SD in NF), controlling for age, gender, and education.

Conclusion:

We found that NF during rest was higher in AD patients, and predicted dementia transition. Thus, NF may be a valuable biomarker of AD, however more validation and mechanistic studies need to be performed.

Introduction

Alzheimer’s disease (AD) is characterized by extensive cortical neuronal loss and the breakdown of connections between brain systems.¹ This degeneration of neural pathways disrupts the functional coherence of brain activation, leading to altered brain functional connectivity, including a weakening of connectivity between hippocampus and the frontal lobes,² accelerated decreased connectivity and network integrity in default mode network (DMN),¹ decreased segregation between brain systems,³ and alterations in the connectivity of the default mode, salience, and limbic networks.⁴ There is great interest in identifying functional connectivity-based biomarkers of AD diagnosis and severity, with many pointing to the role of DMN connectivity in helping identify diagnostic subgroups. However, efforts in this area have yielded mixed results, suggesting cautious optimism for its potential future diagnostic utility.⁵

While static measures of functional connectivity reveal some alterations in AD, recent approaches have explored the relevance of time-varying or dynamic measures of functional connectivity to cognitive and disease processes. These dynamic functional connectivity measures aim to capture the flexible aspects of network organization and reorganization, a process which could illuminate the on-demand, evolving nature of network structure and function. While static functional connectivity analyses may illuminate information about general features of brain organization or network structure, dynamic connectivity analyses may reveal more about the ways in which the brain may be able to flexibly modulate this network architecture to accommodate an ever-changing set of cognitive demands. Previous studies have shown differences in dynamic functional connectivity strength in the left precuneus, DMN, and dorsal attention network (DAN) among cognitively normal (CN) controls, and those with mild cognitive impairment (MCI), and AD.⁶ Another study explored the effect of AD progression on both static and dynamic measures of functional connectivity, finding that even those with “very mild AD” begin to show changes in both static and dynamic measures of connectivity within the cognitive control network, and that these deviations are in some cases magnified in later stages of the disease.⁷ The authors suggest that dynamic measures of functional connectivity, particularly in the cognitive control network, may have better potential as a biomarker for AD staging.

One specific measure of time-varying connectivity, neural flexibility (NF), is defined as the frequency with which nodes change their functional network assignment over the course of a scan.^8–10 This measure is conceptually quite a promising method for estimating these dynamic connectivity features, as it explicitly indexes the degree to which each node in the brain changes network membership over a period of time (either during rest or during a task). Being able to know, at the node/network/brain level how frequently regions of the brain may “move” between networks could be informative in establishing the benefits vs. costs of this flexibility, and gaining a sense of the regional specificity of these changes. NF has previously been studied in the context of development,¹⁰ as well as in psychiatric¹¹ and healthy aging¹² populations. In our previous work, NF during a task was associated with both participant age and cognitive performance, while NF during rest was not.¹² However, this prior study only included cognitively healthy adults across the adult lifespan (ages 20–80). Additionally, while the presence of a cognitive task presents a “challenge” for the brains of healthy adults, it is unclear whether certain populations facing cognitive challenges as part of a disease process (e.g., AD) might begin to show differences in NF in the absence of an explicit cognitive “challenge”. Given prior studies employing static and/or dynamic resting state functional connectivity measures as biomarkers of AD staging, it remains to be seen whether this NF measure might similarly track with disease severity in this population. Further, given that past studies of dynamic functional connectivity in AD either conducted whole-brain analyses agnostic to brain network structure,⁶ or solely within one cognitive network,⁷ the present study employing NF as a measure of dynamic functional connectivity will add a level of whole brain network relevance in order to contextualize dynamic network properties with respect to functionally-defined networks.

Currently, the role of NF in AD remains largely unexplored. Our research aimed to fill this gap by evaluating whether AD severity is associated with NF, and whether it can be predictive of AD conversion. Based on prior literature generally showing a breakdown in network architecture in AD, we hypothesize that NF will generally be higher in AD group, and that early alterations in NF may be able to predict conversion to AD.

Methods

Participants

The Alzheimer’s Disease Neuroimaging Initiative (ADNI) was launched in 2003 as a public-private partnership and examines the progression of MCI and AD through longitudinal assessments of MRI, PET, and other biomarkers, along with clinical and neuropsychological assessments. A detailed description of the ADNI cohort has been previously published ¹³. Eligible MCI subjects had memory complaints, but no significant functional impairment, scored between 24 and 30 on the mini-mental status examination (MMSE), had a global clinical dementia rating (CDR) score of 0.5, a CDR memory score of 0.5 or greater and objective memory impairment on the Wechsler Memory Scale – Logical Memory II test. CN participants had MMSE scores between 24 and 30, a global CDR of 0 and did not meet criteria for MCI or AD. Inclusion and diagnostic criteria, as well as procedures and protocols, for the ADNI studies can be found on http://www.adni-info.org/Scientists/ADNIStudyProcedures.html.

For the current study, we only included CN, MCI, and AD individuals from ADNIGO/2/3 phase who had completed baseline resting state scans from the single band sequence for protocol consistency as available on March 13^th, 2021.

All participants gave informed consent through their local Institutional Review Boards prior to study participation. Within the ADNI protocol, all procedures involving human participants were conducted in accordance with the 1964 Helsinki Declaration and its later amendments (for details see www.adni.loni.usc.edu).

MRI data acquisition

T1 and fMRI images were acquired using Philips Medical Systems Scanners with a field strength of 3.0 Tesla. The T1 images were obtained with a 6.8 second repetition time (TR), a 3.1 second echo time (TE), a flip angle of 9°, matrix 256 × 256, 179 slices per volume, a slice thickness of 1 mm, and field of view (FOV) 204mm (right-left;RL), 240mm (anterior-posterior;AP) and 256mm (feet-head;FH). The fMRI images were obtained with a 3s TR, a 30 ms TE, a flip angle of 80°, matrix 64 × 64, 140 volumes, 48 slices per volume, a slice thickness of 3.3 mm, and field of view (FOV) 212mm (RL), 198.75mm (AP) and 159mm (FH). The voxel size was 3.3 × 3.3 × 3.3 mm³. Details on MRI acquisitions can be found in the “MRI scanner protocol” on the ADNI website and in the previous study.¹⁴

Rs-fMRI data analysis

Resting-state functional MRI (rsfMRI) data were preprocessed using fMRIPrep¹⁵; a previous study includes a detailed description of the preprocessing.¹⁶ Briefly, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep (v20.1.1). Head-motion parameters with respect to the blood oxygen level dependent (BOLD) reference (transformation matrices, and six corresponding rotation and translation parameters) were estimated before any spatiotemporal filtering using mcflirt (FSL 5.0.9, Jenkinson et al. 2002). BOLD runs were slice-time corrected using 3dTshift from Analysis of Functional Neuroimages (AFNI). The BOLD reference was then co-registered to the T1w reference using bbregister (FreeSurfer) which implements boundary-based registration.¹⁷ Co-registration was configured with six degrees of freedom. The BOLD time-series were resampled into standard MNI152NLin2009cAsym space.

Several confounding time-series were calculated based on the preprocessed BOLD: Frame-wise displacement (FWD) was calculated from the six motion parameters and root-mean-square difference (RMSD) of the BOLD percentage signal in the consecutive volumes. Contaminated volumes were then detected and classified as outliers by the criteria FWD > 0.5 mm or RMSD >0.3% and replaced with new volumes generated by linear interpolation of adjacent volumes. The three global signals were extracted within the cerebrospinal fluid (CSF), the white matter (WM), and the whole-brain masks. A bandpass filter with cut-off frequencies of 0.01 and 0.09 Hz was used. Finally, the covariates corresponding to head motion (6 realignment parameters), outliers, and the BOLD time series from the subject-specific WM and CSF masks were used in the connectivity analysis as predictors of no interest, and were removed from the BOLD functional time series using linear regression. The nodes for our functional connectivity analyses were defined using the Power atlas ¹⁸, and the mean time series of each node was extracted. Due to the fact that the networks of interest from power coordinate did not include cerebellum, concerns in distortion in resting state scans, and lack of whole cerebellum coverage (Supplementary Figure 4), only non-cerebellar ROIs were included in our analyses, resulting in a total of 256 ROIs (264 ROIs – 8 cerebellar ROIs).

Neural flexibility computation

The sliding-window based functional connectivity metrics were computed using Pearson’s correlation coefficients with a window width of 30 repetition time (TR) (90 sec) and a step size of 1 volume (3 sec). Within each window, matrices were first thresholded based on an FDR correction of p<0.05 across all correlations, then the absolute value of all correlations was computed. For dynamic community detection, we employed functions based on a generalized Louvain method (GenLouvain) with a sliding window approach as described in Yin et al.’s work in 2020.¹⁰ Briefly, the GenLouvain identifies the dynamic community structure in a time-dependent, multilayer, and/or multiplex network. The multilayer modularity quality function (Q) depends on two parameters: γ determines the resolution of each layer; ω represents the weight of the interlayer connections. The dynamic community detection can be affected by these two parameters and window sizes. In previous studies, γ = 1 and ω = 1 have been used estimating NF with similar time resolution and lengths.^8–10

We repeated this 30 times to get the optimal results, since the generalized Louvain method is not deterministic. At any given time point each node may have a different community assignment compared to those of the adjacent time points. Given the dynamic community detection results, we defined the NF of a node as the number of times that a node changed its community assignment across the sliding windows, normalized by the total number of possible changes. We computed global NF (GNF) as the average NF over 256 nodes, and network-level NF as the average NF of the nodes based on their membership in 12 pre-defined networks: Auditory (AUD), Cingulo-Opercular (CON), Default Mode (DMN), Dorsal Attention (DAN), Fronto-parietal (FPN), Memory Retrieval (MRN), Salience (SN), Sensorimotor hand (SMN-H), Sensorimotor mouth (SMN-M), Subcortical (SUB), Ventral Attention (VAN), and Visual (VIS).¹⁸

Scanner/site effect was adjusted using ComBat,^19,20 a statistical harmonization method based on empirical Bayes frameworks to model and adjust for batch (site/scanner) effects across imaging measures while preserving biological variability. For the primary analysis, AD group was adjusted as a covariate, and we performed sensitivity analysis by running ComBat with AD group, age at scan, gender, and education as covariates.

Statistical analysis

There were 956 samples with rsfMRI data at baseline in the ADNI study. Participants were excluded if any of the following was true: a participant had an outlier percentage > 30%; a participant had a repetition time (2.9, 3.1); or a participant was missing gender information. Overall, 862 participants with fMRI measurements remained for baseline analysis after sample filtering. The demographic and clinical characteristics across clinical diagnosis groups (CN, MCI, AD) were compared using analysis of variance (ANOVA) for continuous variables and the Pearson’s Chi-square test for categorical variables.

We first conducted baseline analyses in this sample to see if there was a significant difference in NF across CN, MCI and AD. We performed linear mixed models on GNF, based on NF in each of the 12 networks, to test the effect of group indicators on NF after controlling for the age at scan, gender, and education level, as well as the random intercept of research site. We performed F-tests to test the overall group effect and post hoc pair-wise contrast analyses to quantify the pair-wise group differences. The group difference estimation and their 95% confidence intervals for the global and all network levels are reported. The p-values of the F-tests for group were adjusted for multiple comparisons correction by controlling for false discovery rates (FDR).²¹

We then evaluated the likelihood of dementia transition over the follow-up (maximum 11.3 years) based on these estimates of NF using survival analysis. In the non-demented participants at baseline, we performed Cox-proportional hazard regression analysis with each NF measure at baseline, using age, gender, and education as fixed effects and random intercepts for scanner. Multiple comparisons correction controlling for FDR was then performed.

In accordance with prior studies using this measure of NF,^10,12 we also looked at nodal membership in the “flexible club” to see whether the nodes that changed network assignment more frequently across the scan were similar across our 3 groups. The flexible club was determined using a 20% threshold on the average Global NF within each group.¹²

Sensitivity Analyses

To ensure the reliability of the findings, we performed a series of sensitivity analyses. First, NF estimates might vary across different parameter settings, which are composed of 2 different possible window sizes (30 and 40 volumes), 3 values of gamma (0.75, 1, 1.5), and 3 values of omega (0.75, 1, 1.5). All three models were recalculated on global and network estimates, and the resulting p-values were compared.

To evaluate within-sample consistency, we evaluated split-half reliability. We divided the dataset randomly into two equal groups and repeated the regression analyses. We performed stratified random samples within the diagnosis group to ensure balanced sample sizes across the diagnosis group. We repeated this procedure 500 times. Since the reduction in the sample size, we focus on the parameter estimates rather than p-values.

Results

Demographic characteristics

Details regarding participant eligibility are presented in Figure 1. The mean age of the entire study sample (N = 862) was 73.7 (SD 7.8) years, 53% of the sample identified as female, and 89.3% identified as White. Demographic characteristics for participants included in the present analyses across the three diagnostic groups are reported in Table 1. Comparisons of the study sample demographics by group indicated that there was a significantly greater proportion of females in CN, compared to MCI or AD (P <0.001) (Table 1), and AD participants were older (p=0.006) and reported lower levels of education (p=0.003). There were no significant differences with respect to race (p=0.439).

Figure 1.

Flowchart for sample filtering based on inclusion criteria.

Table 1.

Demographics of the study sample

		CN (N=461)	MCI (N=294)	Dement (N=107)	Total (N=862)	p value
Age, years	Mean (SD)	73.18 (7.39)	73.80 (8.08)	75.86 (8.18)	73.72 (7.77)	0.006
Gender	M	187 (40.6%)	163 (55.4%)	55 (51.4%)	405 (47.0%)	< 0.001
	F	274 (59.4%)	131 (44.6%)	52 (48.6%)	457 (53.0%)
Education, years	Mean (SD)	16.74 (2.34)	16.41 (2.69)	15.87 (2.58)	16.52 (2.51)	0.003
Race	White	406 (88.1%)	267 (90.8%)	97 (90.7%)	770 (89.3%)	0.439
	Others	55 (11.9%)	27 (9.2%)	10 (9.3%)	92 (10.7%)
APOE4	N	280 (68.6%)	141 (57.3%)	33 (34.4%)	454 (60.5%)	< 0.001
	P	128 (31.4%)	105 (42.7%)	63 (65.6%)	296 (39.5%)
	N-Miss	53	48	11	112
AD transition	No	376 (99.2%)	188 (79.0%)	0	564 (91.4%)	<0.001
	Dement	3 (0.8%)	50 (21.0%)	0	53 (8.6%)
	N-Miss	82	56	107	245

Clinical group differences in neural flexibility

Baseline NF showed significant group differences in SMN-H, VAN and VIS networks, and those survived multiple comparisons correction controlling for FDR at the 5% level (Supplementary Table 1). Figure 2 shows pair-wise group comparison results of AD vs. CN, MCI vs. CN, and AD vs. MCI in the baseline analysis model. NF is significantly higher in the AD group than the CN group on Global (β = 0.002, 95%CI: [0.001,0.004]), CON (β = 0.003, 95%CI: [0,0.006]), MRN (β = 0.003, 95%CI: [0,0.007]), SMN-H (β = 0.005, 95%CI: [0.001,0.008]), SMN-M (β = 0.004, 95%CI: [0,0.007]), VAN (β = 0.004, 95%CI: [0.002,0.007]) and VIS (β = 0.003, 95%CI: [0.001,0.005]) networks; and NF is significantly higher in the MCI group than the CN group in VIS (β = 0.002, 95%CI: [0.000,0.003]) network.

Figure 2.

Pairwise group comparison results of the global and network-level neural flexibility.

Supplementary Table 2 reports the parameter estimates and their 95% CIs for all networks. The results showed similar patterns across different combinations of parameters (omega and gamma, Supplementary Figures 1 and 2). Also, we re-ran the same analysis with ComBat without covariates and adjusted for all covariates. The results were similar.

Dementia transition prediction

Among 617 non-demented participants (MCI and CN) at baseline, 53 (8.6%) participants converted to dementia at one of the follow-up timepoints. Higher NF in the VIS was positively associated with AD transition (HR=1.323, 95%CI 1.002 to 1.747, p=0.049, per 1 standard deviation in NF), controlling for age, gender, and education, although it did not survive multiple comparisons correction. For all models, the proportional hazard assumptions were satisfied (p>0.05). Figure 3 shows unadjusted Kaplan-Meier survival curves by high and low (median split) NF in VIS. Higher NF in VIS was associated with higher overall rate of AD conversion.

Figure 3.

Unadjusted Kaplan-Meier survival curves by high and low (median split) NF in Visual Network (VIS). Higher NF in VIS showed higher overall rate of AD conversion.

Flexible Club Membership Across Groups

The overlap between nodes exhibiting high NF across groups was also examined (Figure 4). Generally, nodes belonging to the flexible club appeared to be fairly overlapping, with 62.7% (42 of 67) of all flexible nodes belonging to the “flexible club” in more than one group.

Figure 4.

Overlap among the Flexible Clubs across AD Groups at 20% threshold. A showed van diagram of the flexible clubs and B. regional distribution of members of the Flexible Clubs at 20% threshold.

Sensitivity Analysis

First, the results did not vary across different community parameter choices. When we fix γ=1, the patterns of group comparison results (MCI vs. CN, AD vs. CN) did not differ by window sizes, but the estimates were larger when ω=0.75, while the statistical significance remained the same across parameter choices (Supplementary Figure 1A). When we fix ω=1, the patterns of group comparison results (MCI vs. CN, AD vs. CN) did not differ by window sizes or γ (Supplementary Figure 1B). For the dementia transition, the results were similar across parameter choices (Supplementary Figure 2).

For the half-split reliability, we found that the results were distributed similarly across the random sets for each group comparison, showing good within-sample consistency. We also observed similar patterns when we ran the full sample analysis, confirming that few participants do not drive the current results. In the revision, we included the figure below in Supplementary Figure 3 and the description in the Sensitivity Analysis section.

Discussion

In the present study, we found that neural flexibility during resting state is higher in AD patients than in CN or MCI, and predicts AD conversion. Previous studies have reported lower neural flexibility in psychiatric disorders such as ADHD in younger participants. Conversely, in our sample, neural flexibility increased from CN to MCI to AD at the global level, as well as in a variety of specific networks, including somatomotor/sensory networks (VIS, SMN-H, and SMN-M), and cognitive networks (CON, VAN, and MRN). Thus, individuals diagnosed with AD tend to show generally greater instability in network membership across a variety of cognitive and sensorimotor networks.

This may indicate loss of network integrity as pathology spreads, suggesting that nodes may be more likely to flexibly change their network membership as the networks they are members of slowly degrade (i.e., MRN, VAN, CON). As pathology takes more of these nodes “offline”, other nodes must begin to play a more adaptable role in attempting to maintain function in the face of these losses. Alternatively, this pattern could also mean that nodes in pathologically “spared” networks (i.e., VIS, SMN-H/M), may need to take up more of the heavy lifting as other networks become more heavily affected by pathology.

Another way past studies have used this NF measure is to examine which nodes exhibit the highest NF, and compare these “flexible club” members across diagnostic groups/categories. When members of this club differ across groups, it might suggest that groups employ different “strategies” or compensatory mechanisms when it comes to flexibly adapting to their environment, but when members are similar across groups, it may suggest that any macro-level differences in NF between groups are more likely due to a general shift in NF across all nodes. In the present study, we found that 62.7% of the “flexible club” nodes were the same across all groups. This may indicate that generally these nodes tend to be the most flexible in older age, however as AD pathology begins to spread and networks begin to break down, some nodes ramp up their flexibility in response to this neuronal insult.

We also found that higher neural flexibility in the visual network predicted transition to AD. These findings underscore the role of neural flexibility as a potential biomarker for early detection and progression of Alzheimer’s disease. Past studies exploring how AD affects resting state functional connectivity have found differences in functional connectivity between CN and AD in the DMN,^1,4,6 the cognitive control network,⁷ the salience network,⁴ the dorsal attention network,⁶ the limbic network,⁴ or generally between a variety of networks.³ However, the present study critically illuminated the role of changes in network assignment of nodes within the visual network as predictive of AD conversion. While this initially may appear to be at odds with findings from several past studies, it also highlights some of the different implications that can be drawn from static vs. dynamic connectivity analysis. Results from static connectivity analysis highlight differences in network structure or integrity between diagnostic groups, while NF analyses illuminate networks whose nodes frequently and flexibly modulate their network membership over time. In this vein, in a separate study by Sendi and colleagues utilized a different dynamic functional connectivity method that utilizes a sliding window approach to examine transitions between, and time spent in, “brain states”, and similarly found differences between individuals with very mild AD (vmAD) and CN in connectivity between the sensorimotor network and visual network.²² Specifically, they found that their vmAD group showed decreased connectivity between these two networks, and that CN spent more time in a state with higher connectivity between visual and sensorimotor networks. While the differences in methodology may limit the direct comparisons between our two sets of findings, it suggests that dynamic measures of FC in the visual network clearly play a role in distinguishing AD from other diagnostic groups. This may indicate that individuals with AD, or who are likely to progress to AD, have a visual network that may show dynamic functional changes. While the present study does not include analysis of neuropathology data, one thing to note is the relative pathological sparing of regions associated with somatomotor and visual networks in early AD. This may suggest that these “spared” networks must functionally re-organize or compensate in response to the spread of pathology throughout the rest of the brain.

Alternatively, one potentially speculative interpretation of the role of the visual network and somatomotor networks in the present study could relate to their delayed role in the progression of AD pathology.^23–25 Since we know amyloid plaques and neurofibrillary tangles tend to spread to visual and motor regions later in the disease staging, it could be that nodes in these areas/networks tend to be healthier and less disrupted by pathology, and thus better able to adapt to changing neural demands. Given the scope of the present study, we cannot be entirely certain as to what biological mechanisms may underlie this higher NF in the visual network, however future studies should explore localized relationships between pathological burden and NF in order to get a better sense of the role of pathology in explaining these findings.

The current study has several limitations. The time resolution of fMRI is quite large (TR=3). Neural flexibility may differ when we have high time-resolution fMRI data so replication studies using high resolution fMRI data needs to be done in the future. Further, in healthy aging, task-based neural flexibility, and not resting state neural flexibility, was related to participant age and cognitive function, however the present analyses were limited to resting state data. However, we believe that restricting the present analyses to resting state data is in line with existing studies.^1–7 Additionally, given the challenges of administering a cognitive task to individuals at risk for or diagnosed with dementia, and the paucity of correct trials available for analysis when performance on the task is at or near floor, a similar analysis of task-based data is not likely to be feasible in this population. One additional large limitation of the ADNI dataset is the fact that this is an overwhelmingly non-Hispanic White sample. Thus, it is unclear whether results of the present study will generalize to other racial/ethnic groups. Finally, a relatively small proportion of individuals converted to dementia over the course of the ADNI study, suggesting that we have to be relatively cautious in touting the diagnostic utility of our predictors of conversion to AD. The current study had relatively low rate of AD conversion (8.6%) since large proportion of the non-demented participants were CN and early MCI. Thus, the survival analysis findings did not pass multiple comparison corrections. Future studies should address these limitations by exploring these patterns and predictors in other, more racially/ethnically diverse samples of individuals along the AD spectrum (CN/MCI/AD).

Finally, we used Power Atlas to organize node-level estimates of neural flexibility into network-level neural flexibility in this work. However, network composition could be altered in AD, and using atlases derived from healthy adults may not capture AD-specific reorganization of the human functional network. To the best of our knowledge, no AD-specific functional network parcellation has been validated and established. That being said, given that we are fundamentally interested in biomarkers of conversion from healthy aging to AD, we made the decision to optimize the network parcellation scheme for our healthy individuals in order to contextualize changes in AD with respect to “optimal” network assignment in healthy older adults. In the future, though, it is of interest to revalidate the findings in AD-specific functional networks.

In summary, this study represents a critical step in exploring the effects of AD and MCI on brain function and connectivity, and in identifying potential biomarkers of risk for AD conversion. Critically, it also establishes a set of measures that can easily be applied to new samples of individuals at risk for AD to explore whether these predictors can aid in identifying those who may convert to AD in the near future.

Data availability statement

The data that support the findings of this study are available from the corresponding author, SL, upon reasonable request.