The characteristic patterns of dynamic network reconfiguration underlying subjective cognitive decline

Abstract

Background

Although dynamic network reconfiguration is altered in Alzheimer's disease (AD), its pattern in subjective cognitive decline (SCD), an early preclinical stage, remains unclear.

Objective

This study aims to identify the characteristics of dynamic network reconfiguration in SCD individuals compared to healthy controls (HCs) and AD patients.

Methods

Time-varying multilayer network models were built for AD (n = 111), SCD (n = 115), and HC (n = 111) groups using the GenLouvain algorithm. We compared dynamic reconfiguration features across groups and examined how abnormal reconfiguration affects the relationship between amyloid-β (Aβ) pathology and cognition.

Results

SCD individuals showed higher global recruitment and lower global integration than AD patients, and lower integration than HCs. At the nodal level, SCD was marked by increased recruitment and reduced integration in regions mainly within the default mode network (DMN) and somatomotor network (SMN). In SCD, recruitment in the left anterior insula and ventral frontal cortex linked Aβ pathology to memory performance. In AD, integration in the left superior frontal gyrus mediated the effects of Aβ on memory and verbal fluency. The combination of differences in nodal recruitment and integration across groups helped distinguish between healthy, SCD, and AD individuals.

Conclusions

SCD individuals show distinct patterns of dynamic network reconfiguration, high recruitment, and low integration in DMN and SMN, offering new insights into early AD pathophysiology.

Keywords

Alzheimer's disease amyloid-β pathology dynamic network reconfiguration rs-fMRI subjective cognitive decline

Introduction

Alzheimer's disease (AD) refers to an age-related neurodegenerative disease, which is the most common cause of dementia. Dementia caused by AD accounts for 60–80% of dementia patients, and seriously endangers the life and health of the elderly.^1,2 Subjective cognitive decline (SCD) is characterized by an individual's personal experience of a decline in cognitive abilities or memory, in the absence of measurable impairments through objective clinical assessment.¹ Research has revealed that SCD has similar physiological changes to those with AD, indicating that SCD is an early preclinical stage of AD.³ Due to the unsatisfactory treatment outcomes of therapeutic methods for late-stage AD patients, investigating the early brain alterations of SCD individuals is crucial for the development of early diagnostic tools and efficient disease-modifying therapies.

Resting-state functional magnetic resonance imaging (rs-fMRI) has become a widely adopted tool for mapping the brain's intricate functional architecture, significantly advancing our understanding of its functional dynamics.^4,5 Recent progress in rs-fMRI research has revealed disruptions in functional connectivity (FC) and topological structure across various brain networks in AD, such as the default mode network (DMN), salience network, and executive control network, with these abnormalities potentially playing a role in the advancement of the disease.^6–11 Relative to static FC, network dynamics studies provide a unique perspective for observing dynamic activities and understanding the intrinsic organization of AD.^12–15 As an ultra-early clinical stage of AD, previous studies have revealed that SCD individuals had decreased FC across regions of the putative posterior memory system and retrosplenial-precuneus, and cingulo-opercular network.^16,17 SCD individuals also have stronger DMN FC and limbic FC, and weaker dorsal attention network (DAN) FC than healthy controls (HCs).^18,19 Additionally, studies of brain network dynamics have revealed abnormal dynamic spontaneous neural activity and altered dynamic FC in individuals with SCD.^20,21 Therefore, an altered static or dynamic network could be the neural basis underlying SCD and serve as an objective imaging marker to identify pre-clinically at risk for AD.

Theoretical work in dynamic network neuroscience has linked community structure to network dynamics and proposed the concept of “dynamic network reconfiguration” to assess the time-varying characteristics of brain networks.²² This approach enables us to study more deeply the integrity of the dynamic mechanisms for inter-regional communication of functional networks. Recruitment and integration are two statistical metrics quantifying changes in the community structure over time based on probability, reflecting the dynamic interaction within and between different functional systems. A recent study of time-varying multilayer networks has revealed that dynamic multilayer functional measures were strongly associated with cognition as well as amyloid-β (Aβ) and tau pathology.¹⁵ Compared with the rs-fMRI, the dynamic brain network, by analyzing the temporal reconstruction characteristics of neuronal activities, may have the ability to reveal the dynamic compensation and reconfiguration mechanisms of the brain network in individuals with SCD, compensating for the limitations of rs-fMRI, and providing a unique perspective for understanding the transformation trajectory from SCD to dementia. However, changing patterns of dynamic network reconfigurations and their effects on the relationship between Aβ pathology and cognition in SCD individuals remain to be elucidated.

Hence, based on rs-fMRI, we constructed a time-varying multilayer network in healthy, SCD, and AD individuals. We divided the brain network into communities through a multilayer community detection algorithm and calculated recruitment and integration coefficients at different levels. We aimed to explore (a) the characteristic alterations of dynamic network reconfiguration in SCD individuals compared with HCs and AD patients; (b) the mediation effects of these alteration on the relationship between plasma Aβ pathology and cognition in SCD and AD individuals; (c) obtain a combined index of the dynamic recruitment and integration capabilities of brain network nodes, in an attempt to distinguish between healthy, SCD and AD individuals.

Methods

Participants

The study cohort (n = 337) comprised 111 individuals diagnosed with AD, 115 individuals with SCD, and 111 HCs, who were recruited from two medical centers and one community-based setting. Before participation, each participant provided written informed consent. All participants underwent a series of comprehensive neuropsychological tests, 3.0-T whole brain MRI scanning, measurement of plasma biomarkers of Aβ pathology (SIMOA/qPCR), and detailed clinical evaluations. Education-adjusted norms of the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) were applied to identify dementia and mild cognitive impairment (MCI), respectively.^23–25 The detailed cut-off scores of MMSE and MoCA for different education years were described in our previous study.²⁶ Dementia and MCI due to probable AD were diagnosed according to the recommendations of the National Institute on Aging-Alzheimer's Association.^27,28 Patients with dementia or MCI due to probable AD were classified into the AD group. According to the previous proposed criteria proposed, cognitively normal individuals who reported memory concerns within the past five years and expressed worries about declining memory were classified into the SCD group.³ Those without any cognitive complaints or related anxieties were included as HCs.

The exclusion criteria applied to all participants were: (1) being under 40 years of age; (2) a prior history of stroke; (3) the presence of central nervous system disorders that may lead to cognitive impairment, such as epilepsy, Parkinson's disease, central nervous system infections, subarachnoid hemorrhage, or multiple sclerosis; (4) psychiatric disorder including schizophrenia, major depression disorder; (5) severe systemic diseases such as heart failure, kidney dysfunction, cancer, shock; (6) history of drugs or alcohol abuse; (7) intolerance of MRI examination or inability to complete neuropsychological testing.

Neuropsychological examinations

All subjects underwent a battery of standardized neuropsychological tests. The MMSE and MoCA were employed to evaluate global cognition. Following the methodology established in our prior studies, the current study utilized composite scores to represent the performance in different cognitive domains.^26,29 The original test scores were transformed into normalized Z-scores. Then, the individual score of each cognitive domain was evaluated by averaging the Z-score from the corresponding sub-items. Briefly, visuospatial processing function (VPF) was assessed using the Visual Reproduction–Copy subtest and the Clock Drawing Test. Memory performance was evaluated based on scores from the Auditory Verbal Learning Test-Long Delayed Recall and the Wechsler Memory Scale–Visual Reproduction Delayed Recall. Information processing speed (IPS) was calculated from the Trail Making Test–Part B (TMT-B), Stroop Color and Word Test–Part A (SCWT-A), and SCWT-B. Executive function (EF) was derived from performance on TMT-B and SCWT-C. Higher raw scores on EF and IPS reflect worse performance, to ensure consistent interpretation, the composite scores of IPS and EF were converted to a minus, then obtained Z scores, to facilitate understanding.

Measurement of plasma Aβ_42/40 ratio

Blood samples were collected using ethylene diamine tetraacetic acid (EDTA) anticoagulant tubes and centrifuged at 3000 rpm for 10 min at 4°C. After centrifugation, the resulting plasma was aliquoted into 1.5 mL Eppendorf tubes and stored at −80°C until further analysis. Levels of plasma Aβ₄₀ and Aβ₄₂ were measured using an ultra-sensitive single-molecule array (Simoa) technology on the automated Simoa HD-X analyzer (Quanterix, MA, USA), following the manufacturer's protocol. The multiplex Neurology 3-Plex A assay kits (Catalog No. 101995) were obtained from Quanterix and used as directed. All laboratory personnel conducting the assays were blinded to participants’ clinical information. Plasma Aβ pathology was assessed based on the Aβ_42/40 ratio.

MRI imaging acquisition and preprocessing

rs-fMRI and 3D-T1 weighted imaging were acquired using three 3.0T magnetic resonance scanners (ingeniaCX, ingenia3.0T, Philips Medical Systems, Netherlands) in Nanjing Drum Tower Hospital. To mitigate potential multisite variability arising from the use of different scanners, the ComBat algorithm was applied to adjust for site-specific effects on dynamic network reconfiguration metrics during subsequent data analysis.³⁰ Detailed imaging parameters: high-resolution T1-weighted turbo gradient echo sequence: echo time (TE) = 4.6 ms, repetition time (TR) = 9.8 ms, field of view (FOV) = 250 × 250 mm², matrix sizes = 256 × 256, thickness = 1.0 mm, flip angle (FA) = 8°, number of slices = 192; rs-fMRI: TE = 30 ms, TR = 2000 ms, FOV = 192 × 192 mm², matrix size = 64 × 64, thickness = 4.0 mm, FA = 90°, number of slices = 35, each functional image contained 240 volumes.

Functional images were processed using the Data Processing & Analysis for Brain Imaging toolbox (DPARSF, Version 3.2, www.restfmri.net) in conjunction with SPM12 (www.fil.ion.ucl.ac.uk/spm). The initial 20 volumes were discarded to allow for signal stabilization; subsequently, slice-timing correction and realignment were performed on the remaining image volumes. Then, realignment was performed to correct the movement between time points. Head motion parameters were calculated by evaluating the translation in each direction and the angular rotation on each axis for each volume. Participants with head movement exceeding 3 mm or a rotation angle over 3 degrees were excluded. Following this, individual T1-weighted images were segmented into gray matter, white matter, and cerebrospinal fluid, and then co-registered to the mean functional image using a 6-degree-of-freedom affine transformation without resampling. To minimize artifacts related to head motion, linear trends, along with the Friston 24-parameter model for head motion correction, as well as signals from white matter and cerebrospinal fluid, were removed from the functional data by regressing them out as nuisance covariates.³¹ Of note, we chose not to perform global signal regression due to the attendant controversy.³² Subsequently, the Diffeomorphic Anatomical Registration Using Exponentiated Lie Algebra (DARTEL) approach was employed to create spatial transformation maps from each individual's native anatomical space to the MNI standard space. Following this, temporal band-pass filtering was applied within the frequency range of 0.01 to 0.1 Hz. Lastly, all functional images were spatially smoothed using a 6 mm FWHM Gaussian kernel.

Brain nodes were defined based on Dosenbach's 160 atlas.³³ For each node, a spherical region with a 5 mm radius was created, centered on the corresponding atlas coordinates. We removed cerebellar regions in the template, as it is to date unclear that AD manifests in these brain areas, leaving 142 brain areas. The detailed information about the 142 nodes is provided in Supplemental Table 1. The signal of each node was derived by averaging the preprocessed blood oxygen level-dependent (BOLD) signals of all voxels within the sphere. According to the Yeo atlas, the human brain can be parcellated into seven canonical resting-state networks. Among these, the somatomotor network (SMN), ventral attention network (VAN), visual network (VIS), frontoparietal network (FPN), DAN, and DMN were used in the present study. None of the original 160 Dosenbach regions of interest (ROIs) were located within the limbic network. The present study defined subcortical ROIs as the “subcortical network” instead of “limbic network” as one of the seven networks in our study.

Multilayer network construction

Dynamic FC was calculated based on Sliding-window analysis by using the Dynamic BC toolbox (V2.0, www.restfmri.net/forum/DynamicBC).³⁴ The sliding-window length and the step size were set as 50 TRs (100 s) and 2 TRs (4 s), respectively. The intra-layer network was established using Pearson's correlation, whereas for the interlayer connections, we only considered the connection of the same nodes between adjacent time slices to express the time dependence between windows.³⁵ The block adjacency matrix was applied to represent a multilayer network and can be calculated as:

[\begin{matrix} A_{1} & H_{12} & \dots & H_{1 f_{2}} \\ H_{21} & A_{2} & \dots & H_{2 f_{2}} \\ ⋮ & ⋮ & ⋱ & ⋮ \\ H_{f_{1} 1} & H_{f_{1} 2} & \dots & A_{f} \end{matrix}]

where A_α represents the intra-layer network adjacency matrix of layer α and 1≤ α ≤ f. H_f1f2 is the interlayer connection matrix between layerf₁ and layerf_2, multilayer community detection.

The brain community was calculated by a generalized Louvain method (https://github.com/GenLouvain/GenLouvain) for community detection implemented in MATLAB. Specifically, the communities were partitioned into a multilayer network by maximizing the multilayer modularity quality function Q, defined as:

Q = \frac{1}{2 u} \sum_{ijlr} {(A_{ijl} - γ_{l} P_{ijl}) δ_{l r} + δ_{i j} ω_{j l r}} δ (g_{i l}, g_{j r})

where μ is the total edge weight of the network. δ (g_il, g_jr) is 0 if nodes do not belong to the same community and 1 otherwise.³⁶ A_ijl refers to the correlation matrix in layer l, and the element P_ijl is the matrix of prospective correlation under the Newman-Girvan null model, given by P_ijl =

\frac{k_{i l} k_{j l}}{2 m_{l}}

. If l = r, the δ_lr = 1 and equals 0 otherwise. Similarly, if and only if i = j, δ_ij = 1. The topological resolution parameter gamma (γ) is associated with the edge weight of the intra-layer network, while the edge weight of the inter-layer network is determined by the temporal coupling parameter omega (ω). In this work, as in prior studies, γ and ω were set as default values (i.e., γ = ω = 1) because there is no accepted standard procedure to determine these parameters.^35,36 Owing to the algorithm's stochastic character and the modularity landscape's near degeneracy, the multilayer community detection algorithm was carried out 50 times, and the final estimated values were determined by averaging these metrics over 50 runs, as done in previous studies.³⁷

Module allegiance

Module allegiance was used to encapsulate the consistency with which functionally-specified regions of interest are allocated to communities across time.³⁸ Each element in the N × N square matrix (N is the number of brain nodes) displays the percentage of times that a node shares a community with another node when it is being scanned.

Recruitment and integration

Recruitment and integration were derived using the module allegiance matrix. The primary functions for computing the recruitment and integration coefficients are available in the Network Community Toolbox (http://commdetect.weebly.com/). Rather than splitting each network layer on its own, they offered an exhaustive analysis of the alterations to the community structure in a multilayered network.

The likelihood that a region, during community detection, shares the same community as the nodes in its own network is referred to as its recruitment. Recruitment for a node i in functional network S can be characterized as:

R_{i}^{S} = \frac{1}{n_{S}} \sum_{j \in S} P_{i j}

where n_s is the number of nodes in network S.

The possibility that a given region, during community detection, shares a community with nodes from other functional networks is referred to as its integration. Integration for a node i in functional network S can be defined as:

I_{i}^{S} = \frac{1}{N - n_{S}} \sum_{j \in S} P_{i j}

During community reconfiguration, nodes within various functional networks may initially belong to the same community at a given time point and subsequently form associations with nodes in different functional networks at later stages. This behavior reflects a dynamic process of interaction or communication across brain networks.. Recruitment and integration can measure this phenomenon. Communicating with regions in the same functional system throughout scans indicates a region with high recruitment. A region with high integration indicated that it is simpler to find in a community dominated by other functional network nodes across scans.

Statistical analysis

Basic characteristics, the plasma Aβ biomarker, and cognitive data

All continuous numerical variables were displayed as the mean ± standard deviation (SD). Each categorical variable was displayed as an integer (percentage). One-way analysis of variance (ANOVA) and post-hoc pairwise tests (Bonferroni-corrected) were used to compare the group differences of age, education years, plasma Aβ₄₂ and Aβ₄₀, and plasma Aβ_42/40 ratio among the three groups. The group differences of vascular risk factors and sex among the three groups were compared by using Pearson's χ² test. In addition, one-way analysis of covariance (ANCOVA) and post-hoc pairwise tests (Bonferroni-corrected) were used to compare cognitive data, after adjusting for age, sex, and years of education. The software SPSS 22.0 (IBM Corp., Armonk, NY) was utilized for all of these statistical procedures. It was deemed statistically significant when p < 0.05.

Recruitment and integration at whole-brain, subnetwork, and nodal levels

To evaluate group differences in recruitment and integration among the three groups, one-way ANCOVA was conducted, with age, sex, and years of education as covariates. We utilized Bonferroni correction to regulate multiple comparisons of subnetwork-level metrics (p < 0.05/7; there were 7 comparisons for subnetwork-level metrics) in order to prevent false positive rates caused by multiple comparisons. FDR correction (q = 0.05) was used to do the multiple comparison correction at the nodal recruitment and integration. For functional metrics that remained statistically significant following multiple comparison correction, post-hoc pairwise comparisons were performed to assess differences between any two groups, with additional adjustment for multiple testing using the Bonferroni correction.

Correlation and mediation analysis

Partial correlation analyses were further conducted to look into the correlations between disrupted reconfiguration metrics and biomarkers of plasma Aβ pathology in subjects with SCD and AD after adjustment for age, sex, and education years. For these disrupted reconfiguration metrics related to biomarkers of plasma Aβ pathology, mediation analyses were performed to explore the effect of them on the relationship between Aβ pathology and cognitive function. The Process toolkit proposed by Andrew F. Hayes (http://www.processmacro.org) was used for the mediation test in this study, using bootstrapping (k = 5000 samples), the bias-corrected 95% confidence interval (CI) for the mediating effects was computed. The mediating effect was deemed statistically significant if the 95% CI did not include the value 0. Mediation analyses were carried out in PROCESS for the SPSS 22.0 framework.

Classification analysis

The classification accuracy based on dynamic reconfiguration was evaluated by calculating the area under the curve (AUC) using the auc () function from the R pROC package. The corresponding sensitivity and specificity were determined by Youden's index using the optimal.cutpoints() function and the Youden method from the R OptimalCutpoints package. The recruitment and integration that differed significantly among the three groups (FDR-corrected, q = 0.05) were extracted as classification features. The classification analysis was performed in R Studio software based on R-4.3.3.

Robustness analysis

To validate the robustness of our findings, we performed repeated measures ANCOVA to explore whether the main effect of experimental group remained unchanged on the whole-brain level recruitment and integration in different sliding-window sizes (i.e., 40 TRs, 50 TRs, and 80 TRs) and resolution parameters (i.e., γ = ω = 1, γ = 1 and ω = 0.4, γ = 1 and ω = 0.7, γ = 0.95 and ω = 1, γ = 1.05 and ω = 1).

Results

Basic characteristics, the plasma aβ biomarker, and cognitive data

As displayed in Table 1. There were no significant differences in sex or vascular risk factors among the three groups (all p > 0.05). Between the SCD and HC groups, there were no significant differences in age, sex, education years, vascular risk factors, plasma Aβ_42/40 ratio, and cognitive performance (all p > 0.05). Compared with the HC and SCD groups, the AD group showed older age, lower plasma Aβ_42/40 ratio, and poor cognitive function, including MMSE, MoCA, memory, EF, IPS, language, and VPF (all p < 0.05, Bonferroni-corrected).

Table 1.

Demographic characteristics of the study cohort.

Items	HC	SCD	AD	F/x²	p
Items	(n = 111)	(n = 115)	(n = 111)	F/x²	p
Age (years)	67.78 ± 7.15	65.36 ± 9.47	72.12 ± 8.10^ab	32.33	0.001*
Sex, male (% male)	48 (43.2)	52 (45.2)	43 (38.7)	1.02	0.602
Education (years)	10.92 ± 5.19	12.34 ± 4.05	10.40 ± 4.47^b	5.437	0.005*
Hypertension, n (%)	57 (56.4)	44 (45.8)	38 (43.7)	4.84	0.089
Diabetes, n (%)	19 (18.8)	22 (22.9)	32.8 (23.0)	0.661	0.718
Smoking, n (%)	19 (18.8)	20 (22.2)	13 (16.73)	0.989	0.610
Hyperlipemia, n (%)	31 (30.7)	27 (37.0)	15 (20.5)	4.73	0.094
The plasma Aβ₄₂ (pg/ml)	7.15 ± 6.72	6.72 ± 2.00	5.50 ± 1.46^ab	23.69	<0.001*
The plasma Aβ₄₀ (pg/ml)	104.38 ± 20.78	101.44 ± 25.31	105.10 ± 22.93	0.80	0.451
The plasma Aβ_40/42 ratio	0.069 ± 0.015	0.067 ± 0.015	0.052 ± 0.007^ab	54.101	<0.001*
MMSE	28.64 ± 0.345	28.29 ± 0.350	22.76 ± 0.358 ^ab	82.73	<0.001*
MoCA	25.87 ± 0.327	25.67 ± 0.331	17.46 ± 0.344^ab	191.97	<0.001*
Memory	0.43 ± 0.069	0.30 ± 0.066	−0.78 ± 0.077^ab	76.25	<0.001*
EF	0.28 ± 0.067	0.21 ± 0.067	−0.72 ± 0.095^ab	39.67	<0.001*
IPS	0.31 ± 0.073	0.27 ± 0.074	−0.70 ± 0.092^ab	41.05	<0.001*
VPF	0.36 ± 0.070	0.24 ± 0.073	−0.67 ± 0.083^ab	47.39	<0.001*
Language	0.27 ± 0.067	0.26 ± 0.069	−0.57 ± 0.081^ab	37.15	<0.001*

Values were presented as the mean ± standard deviation (SD) or number (percentage). One-way ANOVA and post-hoc pairwise tests (Bonferroni-corrected) were performed in the analyses of age, education years, and the plasma biomarkers of amyloid-β pathology. One-way ANCOVA and post-hoc pairwise tests (Bonferroni-corrected) were used in the comparison of cognitive function, with age, sex, and education years as covariates. Pearson's χ² test was applied in the comparison of sex and vascular risk factors. *p < 0.05, significant difference among the three groups. ^ap < 0.05, significant difference from the HC group. ^bp < 0.05, significant difference from the SCD group. Aβ: amyloid-β; AD: Alzheimer's disease; EF: executive function; HC: healthy control; IPS: information processing speed; MMSE: Mini-Mental State Examination; MoCA: Montreal Cognitive Assessment; SCD: subjective cognitive decline; VPF: visuospatial function.

Group differences of recruitment and integration at the whole-brain level

Whole-brain recruitment and integration values were calculated by averaging the respective metrics across all brain regions for each participant. As shown in Figure 1A and 1B, there was a significant difference in recruitment (p = 0.001) and integration (p = 0.004) among the three groups. Post-hoc pairwise tests revealed that global integration of SCD patients (p = 0.041, Bonferroni-corrected) and global recruitment of AD patients (p = 0.006, Bonferroni-corrected) were lower than those of HCs. SCD patients exhibited higher global recruitment (p = 0.001, Bonferroni-corrected) and lower global integration (p = 0.005, Bonferroni-corrected) than AD patients.

Figure 1.

Group differences in recruitment and integration at the whole-brain and subnetwork level among the three groups. One-way ANCOVA, along with post-hoc pairwise comparisons, was used for group comparisons, adjusting for age, sex, and years of education as covariates. The effect size was reported as Partial Eta Squared derived from the One-way ANCOVA analysis. (A, B) Group differences of recruitment (A) and integration (B) at the whole-brain level among the three groups. (C, D) Group differences of recruitment (C) and integration (D) at the subnetwork level among the three groups. *p < 0.05. **p < 0.01. ***p < 0.001. AD: Alzheimer's disease; DAN: dorsal attention network; DMN: default mode network; ES: effect size; FPN: frontoparietal network; HC: healthy control; SCD: subjective cognitive decline; SMN: somatomotor network; SUB: subcortical network; VAN: ventral attention network; VIS: visual network.

Group differences in recruitment and integration at the subnetwork level

We further compared the recruitment and integration at the subnetwork level among the three groups. For a given subnetwork, the recruitment and integration coefficients are computed as the mean of the respective node-level coefficients across all regions within that subnetwork. As shown in Figure 1C, we first identified significant differences in recruitment within DMN, SMN, and VAN after Bonferroni correction (P < 0.05/7). A significant difference in integration within DMN and SMN among the three groups was also observed after Bonferroni correction (p < 0.05/7, Figure 1D). Post-hoc pairwise tests revealed that the SCD group displayed higher recruitment within DMN (p = 0.003, Bonferroni-corrected) and SMN (p < 0.001, Bonferroni-corrected), as well as lower integration with DMN (p < 0.001, Bonferroni-corrected) and SMN (p = 0.008, Bonferroni-corrected) than those of the AD group. As compared with the HC group, we found the SCD group had decreased integration within DMN (p = 0.042, Bonferroni-corrected) and SMN (p = 0.012, Bonferroni-corrected). Additionally, we found the AD group exhibited decreased recruitment within SMN (p = 0.028, Bonferroni-corrected) and VAN (p < 0.001, Bonferroni-corrected) compared with the HC group.

Group differences in recruitment and integration at the node level

The recruitment and integration at nodal level were also explored among the three groups. We first observed significant differences in nodal recruitment in 39 brain nodes, which were mainly distributed in DMN, SMN, and VAN (the details in Supplemental Table 2), as well as nodal integration in 10 brain nodes, most distributed in DMN (the details in Supplemental Table 3), after FDR correction (q = 0.05) among the three groups. Post-hoc pairwise tests showed that nodal recruitment in AD patients was widely decreased in SMN (bilateral middle insula, bilateral parietal lobes, right frontal lobe, and right superior parietal lobule), and VAN (right ventral prefrontal cortex, left ventral frontal cortex (vFC.L), bilateral anterior insula (AI), right dorsal anterior cingulate cortex (dACC.R), left basal ganglia (BG.L), medial frontal cortex, and pre-supplementary motor area) when compared with HCs (all p < 0.05, Bonferroni-corrected, top of Figure 2B). We also found widely decreased nodal recruitment in DMN (ventral medial prefrontal cortex (vmPFC), left inferior temporal gyrus, bilateral post cingulate cortex (PCC), left PCUN, bilateral angular gyrus) and SMN (bilateral middle insula, bilateral parietal lobes, bilateral temporal, right superior parietal lobule, right frontal lobe, bilateral precentral gyrus) in AD patients as compared to SCD patients (all p < 0.05, Bonferroni-corrected, top of Figure 2C). Additionally, we found nodal integration of left PCC in AD patients was increased compared with HCs (all p < 0.05, Bonferroni-corrected, bottom of Figure 2B), while nodal integration in DMN (vmPFC, left superior frontal gyrus (SFG.L), left PCC, and intraparietal sulcus (IPS), bilateral PCUN) was increased as compared to SCD patients (all p < 0.05, Bonferroni-corrected, bottom of Figure 2C). To reveal dynamic network reconfiguration specifically distinguished SCD patients from HCs, we executed a post-hoc comparison between the SCD and HC groups and found that SCD patients displayed higher nodal recruitment in left PCC and lower nodal recruitment in medial prefrontal cortex than HCs (all p < 0.05, Bonferroni-corrected, top of Figure 2A). We also found SCD patients showed lower integration in left SFG and IPS, right PCUN, and left temporal relative to HCs (all p < 0.05, Bonferroni-corrected, bottom of Figure 2A).

Figure 2.

Post-hoc analyses in group differences of recruitment and integration at nodal level. The size of the significant nodes reflected the effect size of between-group differences. For all nodes and their corresponding abbreviations please refer to Supplemental Table 1. (A) post-hoc analyses of group difference in nodal recruitment (top) and integration (bottom) between the HC and SCD groups. (B) post-hoc analyses of group difference in nodal recruitment (top) and integration (bottom) between the HC and AD groups. (C) post-hoc analyses of group difference in nodal recruitment (top) and integration (bottom) between the SCD and AD groups. AD: Alzheimer's disease; HC: healthy control; SCD: subjective cognitive decline.

The effect on the relationship between Aβ pathology and cognitive function

To investigate the effect of altered dynamic network reconfiguration on the relationship between Aβ pathology and cognitive function, we first performed partial correlation analysis to investigate the correlations of disrupted nodal recruitment and integration with the plasma Aβ_42/40 ratio in the SCD and AD groups, respectively. As shown in Figure 3, we found nodal recruitment of AI.L (r = -0.227, p = 0.016, Figure 3A), BG.L (r = -0.214, p = 0.024, Figure 3B), and vFC.L (r = -0.222, p = 0.019, Figure 3C) were negatively associated with plasma Aβ_42/40 ratio in the SCD group. Nodal recruitment of dACC.R (r = 0.217, p = 0.026, Figure 3D) and BG.L (r = 0.239, p = 0.014, Figure 3E), as well as nodal integration of SFG.L (r = 0.235, p = 0.015, Figure 3F), were positively associated with plasma Aβ_42/40 ratio in the AD group.

Figure 3.

Correlations of altered nodal recruitment and integration with the plasma Aβ pathology in the SCD (A-C) and AD (D, E) groups. The scatter diagrams were visualized in https://hiplot.cn/basic/line-regression. AI.L: left anterior insula; BG.L: left basal ganglia; dACC.R: right dorsal anterior cingulate cortex; SFG.L: superior frontal gyrus; vFC.L: left ventral prefrontal cortex.

For these disrupted nodal recruitment and integration that are related to plasma Aβ_42/40 ratio, we further explore whether they could fully or partially bridge the connection between plasma biomarker of Aβ pathology and cognitive performance. Firstly, we found nodal recruitment of left AI and vFC co-mediated the relationship between plasma Aβ_42/40 ratio and memory (total indirect effect = 2.47 and 95%CI [0.12, 7.15]; Figure 4A) in SCD subjects. Secondly, we found the indirect effects of plasma Aβ_42/40 ratio on MoCA (indirect effect = −31.84, 95%CI [-87.16, −2.90]; Figure 4B), memory (indirect effect = −5.79, 95%CI [-14.70, −0.78], Figure 4C), and VPF (indirect effect = −5.47, 95%CI [-13.50, −0.98], Figure 4D) were mediated through nodal integration of SFG.L in AD subjects. Aside from significant mediation of all the above, we did not find any other significant mediations. These findings together indicated that the reconfiguration of specific brain regions involved in the relationship between Aβ pathology and cognition in SCD and AD individuals was distinct, which may point to diverse intervention modalities.

Figure 4.

Mediation model of the plasma Aβ pathology on cognition through nodal recruitment or integration. (A) Recruitment of AI.L and vFC.L mediated the relationship between the plasma Aβ pathology and memory in the SCD group. (B-D) Integration of SFG.L mediated the relationship of the plasma Aβ pathology with MoCA (B), memory (C), and VPF (D) in the AD group. *indicated a significant pathway. AI.L: left anterior insula; MoCA: Montreal Cognitive Assessment; SFG.L: left superior frontal gyrus; vFC.L: left ventral frontal cortex; VPF: visuospatial processing function.

Classification ability

As illustrated in Figure 5, ROC curve analysis demonstrated that the combined discrepant nodal recruitment and integration effectively differentiated AD patients (AUC = 0.855, 95% CI [0.806, 0.904]; sensitivity = 77.48%; specificity = 81.98%) and SCD individuals (AUC = 0.790, 95% CI [0.732, 0.847]; sensitivity = 74.78%; specificity = 70.27%) from healthy controls (HCs). Furthermore, the model exhibited an AUC of 0.851 (95% CI [0.802, 0.899]) in distinguishing AD from SCD, with a sensitivity and specificity of 78.38% and 77.39%, respectively.

Figure 5.

Receiver operating characteristic (ROC) curves showing the classification performance. All discrepant nodal recruitment and integration patterns were analyzed to distinguish between HC, SCD, and AD individuals. AD: Alzheimer's disease; HC: healthy control; SCD: subjective cognitive decline.

Robustness results

As illustrated in Figure 6, the analysis revealed a significant main effect of “window size” on both recruitment (F = 15.25, p < 0.001, effect size = 0.044) and integration (F = 8.61, p = 0.001, effect size = 0.025). No significant interaction was observed between “window size” and the experimental group for either recruitment (F = 0.32, p = 0.831, effect size = 0.001) or integration (F = 2.18, p = 0.083, effect size = 0.013). Importantly, the main effect of the experimental group remained statistically significant in both recruitment (F = 6.77, p = 0.001, effect size = 0.039) and integration (F = 6.40, p = 0.002, effect size = 0.037). Furthermore, when testing four additional combinations of resolution and interlayer coupling parameters, a significant main effect of “resolution parameters” was observed for both recruitment (F = 21.10, p < 0.001, effect size = 0.060) and integration (F = 22.89, p < 0.001, effect size = 0.065). However, no significant interaction between “resolution parameters” and the experimental group was found for recruitment (F = 0.83, p = 0.542, effect size = 0.005) or integration (F = 0.38, p = 0.842, effect size = 0.002). Crucially, the experimental group continued to show a significant main effect in both recruitment (F = 8.16, p < 0.001, effect size = 0.047) and integration (F = 5.44, p = 0.005, effect size = 0.032). Collectively, these findings indicate that although variations in sliding-window sizes and resolution parameter settings influenced dynamic properties to some extent, the primary effect of the experimental group (i.e., disease-related differences) reported in this study remained robust.

Figure 6.

Robustness analysis of different sliding-window sizes and resolution parameters through repeated measures ANCOVA. Age, sex, and education years as covariates. (A, B) The effect of sliding-window sizes and experimental groups on the whole-brain level recruitment (A) and integration (B). (C, D) The effect of resolution parameters and experimental groups on the whole-brain level recruitment (C) and integration (D). (E) Statistical results of the main and interaction effects in the repeated measures ANCOVA. AD: Alzheimer's disease; ES: effect size; HC: healthy control; SCD: subjective cognitive decline; TR: repetition time. *p < 0.05. **p < 0.01. ***p < 0.001.

Discussion

The primary objective of this research was to identify the characteristic patterns of dynamic network reconfiguration in SCD individuals and their impact on the connection between Aβ pathology and cognitive function. Our results have highlighted the following: (a) relative to healthy and AD individuals, SCD individuals were characterized by higher recruitment and lower integration, especially in DMN and SMN. (b) nodal recruitment of AI.L and vFC.L co-mediated the relationship between plasma Aβ pathology and memory in SCD individuals, but the effect of Aβ pathology on memory and VPF was mediated by nodal integration of SFG.L in AD individuals; (c) the combination of all discrepant nodal recruitment and integration among the three groups can distinguish healthy, SCD, and AD individuals. This comprehensive study revealed that, in contrast to HC and AD individuals, SCD individuals demonstrate distinctive patterns of dynamic network reconfiguration with high recruitment and low integration. Moreover, the reconfiguration of specific brain regions mediating the intricate relationship between Aβ pathology and cognition in SCD patients differed from that observed in AD patients. This finding offered a novel insight into the pathophysiological mechanisms and suggested potential treatment strategies during the early preclinical stage of AD.

Previous studies examining dynamic networks in both single- and multi-layer configurations have reported widespread disruptions in dynamic functional connectivity, network topological organization, and dynamic reconfiguration among AD patients.^{12,13,15,39,40} Consistent with these findings, our study demonstrated that dynamic network reconfiguration—particularly the recruitment and integration within the DMN, SMN, and VAN—was significantly impaired in individuals with AD. We further investigated the potential contribution of these disruptions to the underlying pathological mechanisms of AD and found that the associations between Aβ pathology and global cognition, memory, and VPF were mediated by the nodal integration of the left SFG in AD patients. The SFG has been implicated in working memory processes.⁴¹ A prior study showed that low-frequency direct cortical stimulation of the left SFG could enhance working memory performance.⁴² Integration is defined as the probability that a brain region belongs to the same community as nodes from other functional networks.³⁸ A higher integration coefficient typically indicates that a brain region is more likely to be found within communities dominated by nodes from other networks. Together with the mediation results of our study, we speculated that an eccentrically increased integration of left SFG might be caused by Aβ pathology and contribute to the progression of AD.

Importantly, the present study provides the first evidence of a distinct pattern of dynamic network reconfiguration at the ultra-early clinical stage of AD, specifically in individuals with SCD. In multilayer network analysis, resting-state research has characterized healthy brain networks as “stable loners”, reflecting a relatively high recruitment coefficient and comparatively low integration coefficient.⁴³ Our findings confirm that both individuals with SCD and healthy controls exhibit high recruitment and low integration. Notably, DMN and SMN in SCD individuals demonstrated reduced dynamic integration with other systems, yet exhibited relatively high levels of dynamic recruitment, suggesting an elevated degree of functional independence for these networks. The SMN is primarily responsible for receiving immediate environmental inputs, processing, and transmitting raw sensory information, whereas the DMN plays a central role in self-referential thinking, scene construction, memory formation, and higher-order information processing.^44,45 In light of this, we can conclude that the increased independence of the DMN and SMN may serve as a compensatory mechanism during the early preclinical stage of AD, enabling SCD patients to maintain normal cognition. However, as AD progresses, this network independence weakens and leads to cognitive impairment.

Furthermore, although individuals with SCD exhibit normal objective cognitive performance, their self-reported memory complaints can significantly affect their quality of life. Our findings indicate that the recruitment of the left anterior insula (AI) and ventral frontal cortex (vFC) plays a critical role in mediating the relationship between Aβ pathology and memory function in SCD individuals. Aβ pathology refers to the accumulation of abnormal protein fragments in the brain, which is a hallmark of AD. Memory impairment is a common symptom of AD. The left AI is involved in various cognitive functions, including emotional processing and interoception, which is the ability to sense the body's internal state.^46,47 The vFC, on the other hand, is known for its role in decision-making and social behavior.⁴⁸ Our findings suggest that the recruitment of these two brain regions may account for the impact of Aβ pathology on memory in individuals with SCD. By identifying specific neural pathways involved in memory dysfunction, our study provides valuable insights into the underlying mechanisms of the early stages of AD. Moreover, classification analyses from this study demonstrate that integrating all distinct nodal recruitment and integration features can effectively differentiate between healthy controls, individuals with SCD, and those with AD. This highlights the potential of dynamic network reconfiguration as a biomarker for identifying individuals at both the preclinical and clinical stages of AD.

Several limitations of the present study warrant discussion. First, the current study employed a cross-sectional design, and the identification of specific brain regions was based solely on rs-fMRI; therefore, the network indicators merely reflect differences in network dynamics and statistical associations rather than causal relationships. Future investigations should incorporate both rs-fMRI and task-based fMRI, together with longitudinal designs, to comprehensively elucidate the neural pathways and underlying mechanisms. Second, although plasma Aβ pathology detected by Simoa technology has demonstrated high diagnostic accuracy and strong correlation with Aβ-PET, future studies employing Aβ-PET are necessary to confirm our key findings.^49,50 Third, due to the relatively small sample size of MCI individuals, we combined MCI and dementia patients into a single AD group to ensure statistical power, which limits disease-stage stratification; we plan to recruit additional participants and conduct longitudinal follow-up assessments to validate our findings. Fourth, the classification analysis based on combined nodal recruitment and integration metrics was exploratory in nature. Fifth, cerebellar regions were excluded based on the Dosenbach atlas, as it remains unclear whether AD manifests in these brain areas. Finally, sleep-related measures were not included in the current study, despite their potential relevance to cognitive decline.

Conclusion

In conclusion, individuals with SCD exhibited a pattern of increased nodal recruitment and reduced nodal integration, particularly within DMN and SMN. The recruitment of AI.L and vFC.L served as a critical mediator linking Aβ pathology to memory dysfunction in SCD individuals, which was distinct from the mediating role of nodal integration in the SFG.L observed in AD patients, where it influenced both memory and VPF. Through this study, our findings may provide some new perspectives on the brain network compensation mechanisms in SCD individuals. The functional network changes in the SFG.L and other cortical areas may serve as early intervention targets for future neuroregulatory treatments.

Supplemental Material

sj-docx-1-alr-10.1177_25424823261451212 - Supplemental material for The characteristic patterns of dynamic network reconfiguration underlying subjective cognitive decline

Supplemental material, sj-docx-1-alr-10.1177_25424823261451212 for The characteristic patterns of dynamic network reconfiguration underlying subjective cognitive decline by Zhihong Ke, Dan Yang, Zhixin Zhou, Wenqian Gao, Shuang Fang, Zheqi Hu, Chengbing Gong, Yue Cheng, Hui Zhao and Yun Xu in Journal of Alzheimer's Disease Reports

Footnotes

Acknowledgements

We gratefully acknowledge all the participants involved in this work for their dedication.

ORCID iDs

Zhihong Ke

Dan Yang

Zhixin Zhou

Wenqian Gao

Zheqi Hu

Chengbing Gong

Yue Cheng

Yun Xu

Ethical considerations

This study was approved by the Ethics Committee of The Affiliated Drum Tower Hospital of Nanjing University Medical School (approval number: 2022-472-01). All patient data were anonymized before analysis to protect privacy. The study was conducted in compliance with the principles of the Declaration of Helsinki and relevant data protection regulations.

Consent to participate

Informed consent was obtained from all individual participants included in the study

Consent for publication

Written informed consent for publication was obtained from all participants or their legally authorized representatives. For participants with cognitive impairment, capacity to consent was assessed by the study physician; when capacity was lacking, consent was obtained from their representative, and participant assent was sought whenever possible. Participants and/or representatives were informed that anonymized neuroimaging and clinical data may be published and shared publicly. This manuscript contains no identifiable personal information.

Author contribution(s)

Zhihong Ke: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Resources; Software; Validation; Visualization; Writing – original draft.

Dan Yang: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Resources; Software; Visualization; Writing – original draft.

Zhixin Zhou: Data curation; Formal analysis; Investigation; Resources; Software; Validation.

Wenqian Gao: Formal analysis; Investigation; Resources; Software; Validation.

Shuang Fang: Data curation; Investigation; Resources; Software; Validation.

Zheqi Hu: Data curation; Investigation; Resources; Software; Validation.

Chengbing Gong: Data curation; Investigation; Resources; Software; Validation.

Yue Cheng: Data curation; Investigation; Resources; Supervision; Validation.

Hui Zhao: Conceptualization; Data curation; Methodology; Project administration; Resources; Supervision; Validation; Writing – review & editing.

Yun Xu: Conceptualization; Formal analysis; Methodology; Project administration; Writing – original draft; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the STI2030-Major Projects (2022ZD0211800), the National Natural Science Foundation of China (U25A2064, 82130036), Jiangsu Province Key Medical Discipline (ZDXK202216), the Key Research and Development Program of Jiangsu Province of China (BE2020620), and the Postgraduate Research & Practice Innovation Program of Jiangsu Province (SJCX23_0670).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

Data described in the article will be available upon request by bona fide researchers for specified scientific purposes by contacting the corresponding authors.

Supplemental material

Supplemental material for this article is available online.

References

Scheltens

De Strooper

Kivipelto

, et al. Alzheimer's disease. Lancet 2021; 397: 1577–1590.

Jia

Chu

, et al. Prevalence, risk factors, and management of dementia and mild cognitive impairment in adults aged 60 years or older in China: a cross-sectional study. Lancet Public Health 2020; 5: E661–E671.

Jessen

Amariglio

van Boxtel

, et al. A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer's disease. Alzheimers Dement 2014; 10: 844–852.

Kashyap

Kong

Bhattacharjee

, et al. Individual-specific fMRI-subspaces improve functional connectivity prediction of behavior. Neuroimage 2019; 189: 804–812.

Vidaurre

Smith

Woolrich

. Brain network dynamics are hierarchically organized in time. Proc Natl Acad Sci U S A 2017; 114: 12827–12832.

Amaefule

Dyrba

Wolfsgruber

, et al. Association between composite scores of domain-specific cognitive functions and regional patterns of atrophy and functional connectivity in the Alzheimer's disease spectrum. Neuroimage Clin 2021; 29: 102533.

de Vos

Koini

Schouten

, et al. A comprehensive analysis of resting state fMRI measures to classify individual patients with Alzheimer's disease. Neuroimage 2018; 167: 62–72.

Pievani

de Haan

, et al. Functional network disruption in the degenerative dementias. Lancet Neurol 2011; 10: 829–843.

Han

Gan

, et al. Biomarkers derived from alterations in overlapping community structure of resting-state brain functional networks for detecting Alzheimer's disease. Neuroscience 2022; 484: 38–52.

10.

Leng

Hinz

Gentleman

, et al. Neuroinflammation is independently associated with brain network dysfunction in Alzheimer's disease. Mol Psychiatry 2023; 28: 1303–1311.

11.

Tumati

Marsman

De Deyn

, et al. Functional network topology associated with apathy in Alzheimer's disease. J Affect Disord 2020; 266: 473–481.

12.

Jing

Chen

Wei

, et al. Altered large-scale dynamic connectivity patterns in Alzheimer's disease and mild cognitive impairment patients: a machine learning study. Hum Brain Mapp 2023; 44: 3467–3480.

13.

Sendi

MSE

Zendehrouh

, et al. Disrupted dynamic functional network connectivity among cognitive control networks in the progression of Alzheimer's disease. Brain Connect 2023; 13: 334–343.

14.

Lin

Huang

, et al. Abnormal dynamic functional connectivity in Alzheimer's disease. CNS Neurosci Ther 2020; 26: 962–971.

15.

Canal-Garcia

Veréb

Mijalkov

, et al. Dynamic multilayer functional connectivity detects preclinical and clinical Alzheimer’s disease. Cereb Cortex 2024; 34: bhad542.

16.

Viviano

Hayes

Pruitt

, et al. Aberrant memory system connectivity and working memory performance in subjective cognitive decline. Neuroimage 2019; 185: 556–564.

17.

Ruiz-Rizzo

Viviano

Daugherty

, et al. Subjective cognitive decline predicts lower cingulo-opercular network functional connectivity in individuals with lower neurite density in the forceps minor. Neuroimage 2022; 263: 119662.

18.

Lee

Park

Kim

. Altered functional connectivity of the default mode and dorsal attention network in subjective cognitive decline. J Psychiatr Res 2023; 159: 165–171.

19.

Jiang

Daamen

, et al. Altered limbic functional connectivity in individuals with subjective cognitive decline: converging and diverging findings across Chinese and German cohorts. Alzheimers Dement 2023; 19: 4922–4934.

20.

Chen

Zhang

, et al. Alterations in dynamic functional connectivity in individuals with subjective cognitive decline. Front Aging Neurosci 2021; 13: 646017.

21.

Liang

Yuan

Wei

, et al. Recurrent and concurrent patterns of regional BOLD dynamics and functional connectivity dynamics in cognitive decline. Alzheimers Res Ther 2021; 13: 28.

22.

Rubinov

Sporns

van Leeuwen

, et al. Symbiotic relationship between brain structure and dynamics. BMC Neurosci 2009; 10: 55.

23.

Katzman

Zhang

Wang

, et al. A Chinese version of the Mini-mental state examination; impact of illiteracy in a Shanghai dementia survey. J Clin Epidemiol 1988; 41: 971–978.

24.

Gorelick

Scuteri

Black

, et al. Vascular contributions to cognitive impairment and dementia: a statement for healthcare professionals from the American Heart Association/American Stroke Association. Stroke 2011; 42: 2672–2713.

25.

, et al. Montreal Cognitive assessment in detecting cognitive impairment in Chinese elderly individuals: a population-based study. J Geriatr Psychiatry Neurol 2011; 24: 184–190.

26.

Yang

, et al. Subsystem mechanisms of default mode network underlying white matter hyperintensity-related cognitive impairment. Hum Brain Mapp 2023; 44: 2365–2379.

27.

Albert

DeKosky

Dickson

, et al. The diagnosis of mild cognitive impairment due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 2011; 7: 270–279.

28.

McKhann

Knopman

Chertkow

, et al. The diagnosis of dementia due to Alzheimer's disease: recommendations from the national institute on aging-Alzheimer's association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 2011; 7: 263–269.

29.

Yang

Tan

Zhou

, et al. Connectome gradient dysfunction contributes to white matter hyperintensity-related cognitive decline. CNS Neurosci Ther 2024; 30: e14843.

30.

Xia

Sun

, et al. Reproducibility of functional brain alterations in major depressive disorder: evidence from a multisite resting-state functional MRI study with 1,434 individuals. Neuroimage 2019; 189: 700–714.

31.

Friston

Williams

Howard

, et al. Movement-related effects in fMRI time-series. Magn Reson Med 1996; 35: 346–355.

32.

Murphy

Fox

. Towards a consensus regarding global signal regression for resting state functional connectivity MRI. Neuroimage 2017; 154: 169–173.

33.

Dosenbach

Nardos

Cohen

, et al. Prediction of individual brain maturity using fMRI. Science 2010; 329: 1358–1361.

34.

Liao

Zhang

Mantini

, et al. Dynamical intrinsic functional architecture of the brain during absence seizures. Brain Struct Funct 2014; 219: 2001–2015.

35.

Telesford

Lynall

Vettel

, et al. Detection of functional brain network reconfiguration during task-driven cognitive states. Neuroimage 2016; 142: 198–210.

36.

Bassett

Wymbs

Rombach

, et al. Task-based core-periphery organization of human brain dynamics. PLoS Comput Biol 2013; 9: e1003171.

37.

Cui

Ding

Wei

, et al. Analysis of dynamic network reconfiguration in adults with attention-deficit/hyperactivity disorder based multilayer network. Cereb Cortex 2021; 31: 4945–4957.

38.

Bassett

Chaitanya

, et al. Disrupted dynamic network reconfiguration of the language system in temporal lobe epilepsy. Brain 2018; 141: 1375–1389.

39.

Chen

, et al. Structural, static, and dynamic functional MRI predictors for conversion from mild cognitive impairment to Alzheimer's disease: inter-cohort validation of Shanghai memory study and ADNI. Hum Brain Mapp 2024; 45: e26529.

40.

Qiao

Wang

Liu

, et al. Brain disorder prediction with dynamic multivariate spatio-temporal features: application to Alzheimer's disease and autism spectrum disorder. Front Aging Neurosci 2022; 14: 912895.

41.

du Boisgueheneuc

Levy

Volle

, et al. Functions of the left superior frontal gyrus in humans: a lesion study. Brain 2006; 129: 3315–3328.

42.

Alagapan

Lustenberger

Hadar

, et al. Low-frequency direct cortical stimulation of left superior frontal gyrus enhances working memory performance. Neuroimage 2019; 184: 697–706.

43.

Mattar

Cole

Thompson-Schill

, et al. A functional cartography of cognitive systems. PLoS Comput Biol 2015; 11: e1004533.

44.

Alexander

DeLong

Strick

. Parallel organization of functionally segregated circuits linking basal ganglia and cortex. Ann Rev Neurosci 1986; 9: 357–381.

45.

Qin

Northoff

. How is our self related to midline regions and the default-mode network? Neuroimage 2011; 57: 1221–1233.

46.

Hopfinger

Slotnick

. Attentional control and executive function. Cogn Neurosci 2020; 11: 1–4.

47.

Namkung

Kim

Sawa

. The insula: an underestimated brain area in clinical neuroscience, psychiatry, and neurology. Trends Neurosci 2017; 40: 200–207.

48.

Hiser

Koenigs

. The multifaceted role of the ventromedial prefrontal cortex in emotion, decision making, social cognition, and psychopathology. Biol Psychiatry 2018; 83: 638–647.

49.

Janelidze

Teunissen

Zetterberg

, et al. Head-to-head comparison of 8 plasma amyloid-β 42/40 assays in Alzheimer disease. JAMA Neurol 2021; 78: 1375–1382.

50.

Schindler

Bollinger

, et al. Validation of plasma amyloid-β 42/40 for detecting Alzheimer disease amyloid plaques. Neurology 2022; 98: e688–e699.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.04 MB

0.00 MB

The characteristic patterns of dynamic network reconfiguration underlying subjective cognitive decline

Abstract

Background

Objective

Methods

Results

Conclusions

Keywords

Introduction

Methods

Participants

Neuropsychological examinations

Measurement of plasma Aβ42/40 ratio

MRI imaging acquisition and preprocessing

Multilayer network construction

Module allegiance

Recruitment and integration

Statistical analysis

Basic characteristics, the plasma Aβ biomarker, and cognitive data

Recruitment and integration at whole-brain, subnetwork, and nodal levels

Correlation and mediation analysis

Classification analysis

Robustness analysis

Results

Basic characteristics, the plasma aβ biomarker, and cognitive data

Group differences of recruitment and integration at the whole-brain level

Group differences in recruitment and integration at the subnetwork level

Group differences in recruitment and integration at the node level

The effect on the relationship between Aβ pathology and cognitive function

Classification ability

Robustness results

Discussion

Conclusion

Supplemental Material

sj-docx-1-alr-10.1177_25424823261451212 - Supplemental material for The characteristic patterns of dynamic network reconfiguration underlying subjective cognitive decline

Footnotes

Acknowledgements

ORCID iDs

Ethical considerations

Consent to participate

Consent for publication

Author contribution(s)

Funding

Declaration of conflicting interests

Data availability statement

Supplemental material

References

Supplementary Material

Measurement of plasma Aβ_42/40 ratio