
==== Front
CNS Neurosci Ther
CNS Neurosci Ther
10.1111/(ISSN)1755-5949
CNS
CNS Neuroscience & Therapeutics
1755-5930
1755-5949
John Wiley and Sons Inc. Hoboken

10.1111/cns.70029
CNS70029
CNSNT-2022-957.R2
Original Article
Original Article
Dynamic functional network connectivity and its association with lipid metabolism in Alzheimer's disease
Zang et al.
Zang Feifei 1
Liu Xinyi 1
Fan Dandan 1
He Cancan 1 2 3
Zhang Zhijun https://orcid.org/0000-0001-5480-0888
1 2 3
Xie Chunming https://orcid.org/0000-0002-5978-9689
1 2 3 chmxie@163.com

for the Alzheimer's Disease Neuroimaging Initiative
for the Alzheimer's Disease Metabolomics Consortium
1 Department of Neurology, Affiliated ZhongDa Hospital, School of Medicine Southeast University Nanjing Jiangsu China
2 Institute of Neuropsychiatry Affiliated ZhongDa Hospital, Southeast University Nanjing Jiangsu China
3 The Key Laboratory of Developmental Genes and Human Disease Southeast University Nanjing Jiangsu China
* Correspondence
Chunming Xie, Department of Neurology, Affiliated ZhongDa Hospital, School of Medicine, Southeast University, No. 87 DingJiaQiao Road, Nanjing, Jiangsu 210009, China.
Email: chmxie@163.com

20 9 2024
9 2024
30 9 10.1111/cns.v30.9 e7002911 8 2024
07 11 2022
21 8 2024
© 2024 The Author(s). CNS Neuroscience & Therapeutics published by John Wiley & Sons Ltd.
https://creativecommons.org/licenses/by/4.0/ This is an open access article under the terms of the http://creativecommons.org/licenses/by/4.0/ License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Abstract

Aims

The study aims to examine the changing trajectory characteristics of dynamic functional network connectivity (dFNC) and its correlation with lipid metabolism‐related factors across the Alzheimer's disease (AD) spectrum populations.

Methods

Data from 242 AD spectrum subjects, including biological, neuroimaging, and general cognition, were obtained from the Alzheimer's Disease Neuroimaging Initiative for this cross‐sectional study. The study utilized a sliding‐window approach to assess whole‐brain dFNC, investigating group differences and associations with biological and cognitive factors. Abnormal dFNC was used in the classification of AD spectrum populations by support vector machine. Mediation analysis was performed to explore the relationships between lipid‐related indicators, dFNC, cerebrospinal fluid (CSF) biomarkers, and cognitive performance.

Results

Significant group difference concerning were observed in relation to APOE‐ε4 status, CSF biomarkers, and cognitive scores. Two reoccurring connectivity states were identified: state‐1 characterized by frequent but weak connections, and state‐II characterized by less frequent but strong connections. Pre‐AD subjects exhibited a preference for spending more time in state‐I, whereas AD patients tended remain in state‐II for longer periods. Group difference in dFNC was primarily found between AD and non‐AD participants within each state. The dFNC of state‐I yielded strong power to distinguish AD from other groups compared with state‐II. APOE‐ε4+, high polygenic score, and high serum lipid group were strongly associated with network disruption between association cortex system and sensory cortex system that characterized elevation of cognitive function, which may suggest a compensatory mechanism of dFNC in state‐I, whereas differential connections of state‐II mediated the relationships between APOE‐ε4 genotype and CSF biomarkers, and cognitive indicators.

Conclusion

The dysfunction of dFNC temporal–spatial patterns and increased cognition in individuals with APOE‐ε4, high polygenic score, and higher serum lipid levels shed light on the lipid‐related mechanisms of dynamic network reorganization in AD.

Alzheimer's disease is recognized to be associated with dysregulation of lipid metabolism and abnormal tempo‐spatial patterns of dynamic functional network connectivity. The dynamic connectivity represented a better power in differentiating Alzheimer's disease patients from non‐Alzheimer's disease subjects than static connectivity. Differential connectivity of strong‐connected state between APOE‐ε4 carriers and non‐carriers mediated the relationship between Apolipoprotein E‐ε4 genotype and cerebrospinal fluid and cognitive indicators.

Alzheimer's disease
APOE
cerebrospinal fluid biomarkers
dynamic functional network connectivity
lipid metabolism
National Natural Science Foundation of China 10.13039/501100001809 81871069 82071204 Postgraduate Research & Practice Innovation Program of Jiangsu ProvinceKYCX19_0115 Key Project for Research and Development Program of Jiangsu ProvinceBE2018741 Nanjing International Joint Research and Development Project201715013 National Key Projects for Research and Development Program of China2016YFC1305800 2016YFC1305802 Alzheimer's Disease Metabolomics ConsortiumR01AG046171 RF1AG051550 source-schema-version-number2.0
cover-dateSeptember 2024
details-of-publishers-convertorConverter:WILEY_ML3GV2_TO_JATSPMC version:6.4.8 mode:remove_FC converted:20.09.2024
Zang F , Liu X , Fan D , He C , Zhang Z , Xie C , . Dynamic functional network connectivity and its association with lipid metabolism in Alzheimer's disease. CNS Neurosci Ther. 2024;30 :e70029. doi:10.1111/cns.70029

Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp‐content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.

Data used in preparation of this article were generated by the Alzheimer's Disease Metabolomics Consortium (ADMC). As such, the investigators within the ADMC provided data but did not participate in analysis or writing of this report. A complete listing of ADMC investigators can be found at: https://sites.duke.edu/adnimetab/team/.
==== Body
pmc1 INTRODUCTION

Resting‐state functional magnetic resonance imaging (rsfMRI) is used to measure temporal correlations of blood oxygenation level‐dependent signals. It provides a powerful tool to investigate the structurally segregated and functionally specialized brain regions at the large‐scale networks level. 1 Disruption of internetwork functional connectivity (FC) may lead to functional disconnection and pathological spread across interlinked networks in Alzheimer's disease (AD). 2 , 3 However, conventional FC analysis cannot detect the non‐stationary nature of rsfMRI signals within a few seconds. Currently, dynamic functional network connectivity (dFNC) analysis based on a time‐varying sliding‐window method is used to summarize reoccurring patterns of connectivity states. Therefore, it is suitable for measuring high‐level network flexibility, which is necessary for network reorganization. Mild AD is associated with an abnormal tempo‐spatial pattern of dFNC, implying substantial disruptions across the multiple sensory networks and reorganized patterns in default mode network (DMN) and executive control network (ECN). 4 Importantly, AD patients are associated with more occurrences and longer time taken in the weakly‐connected state, and fewer occurrences and shorter time taken in the strongly‐connected state compared with the healthy subjects. 5 , 6 Nevertheless, the time‐varying characteristics of dFNC with the disease continuum of AD and related biological factors are unclear.

Lipid metabolism has been a focus in AD research, with previous study 7 showing that astrocytes use 3‐Hydroxy‐3‐Methylglutaryl‐CoA Reductase (HMGCR) to produce cholesterol from acetyl coenzyme A. Cholesterol, phospholipids, Apolipoprotein E (APOE), and Clusterin (CLU) combine to form high‐density lipoprotein complex, which is secreted out of astrocytes by ATP‐binding cassette transporter A1/7 (ABCA1, ABCA7). This complex binds to low‐density lipoprotein receptor (LDLR), low‐density lipoprotein receptor‐related protein 1 (LRP1), and sortilin‐related receptor 1 (SORL1), facilitating its uptake by neurons and subsequent internalization into endosomes through the actions of phosphatidylinositol‐binding clathrin assembly protein (PICALM) and bridging integrator 1 (BIN1). Following fusion with lysosomes, free cholesterol is released within neurons. Excess cholesterol is esterified for storage in the endoplasmic reticulum, whereas surplus cholesterol is enzymatically converted to 24S‐hydroxycholesterol, which can traverse the blood–brain barrier and enter circulation. This process represents a key aspect of lipid metabolism within the brain. The genes encoding these proteins, which are named after the proteins themselves, play a role in lipid metabolism processes and can be classified as lipid metabolic pathway‐related genes. Prior research has indicated that genes related to lipid metabolism are linked to brain structure and function, as well as poorer cognitive performance in AD. 8 , 9 , 10 , 11 Recent studies have focused on exploring the connections between genetic risk factors and the connectivity within and between large‐scale brain networks. Compared with APOE‐ε2 carriers, APOE‐ε4 carriers reduced static functional network connectivity (sFNC) within visual network (VIN), and spent more time in the state of dFNC with lower connectivity within VIN. 12 Nevertheless, the associations between the accumulative polygenetic effects of the lipid metabolic pathway and the dFNC of large‐scale brain networks in AD spectrum (ADS) individuals are not yet fully understood.

Moreover, abnormal lipid levels have been observed in the blood stream of individuals with AD. 13 , 14 The hypolipidemic agent gemfibrozil could activate autophagy, attenuate AD‐like pathology, and reverse memory deficit. 15 Blood lipid levels may provide insights into the lipid metabolism status of individuals. Additionally, blood lipids can serve as valuable biomarkers for susceptibility, monitoring, and prediction of AD. 16 Previous studies have indicated that serum lipid profiles are associated with brain structure and function as well as cognitive performance in middle‐aged and elderly populations. 17 , 18 Recently, researchers have directed their attention towards investigating the associations between serum lipids and FC within and between brain networks. Internetwork connectivity between salience network (SAN) and ventral sensorimotor network (SMN) could mediate the relationship between serum triglyceride and working memory in young adulthood. 19 Also, the intra‐network connectivity in the right inferior parietal of the dorsal attention network (DAN) could mediate the relationship between serum total cholesterol levels and working memory in young healthy adults. 20 However, these studies have primarily focused on sFNC and select components of the lipid profile. It remains unclear whether serum lipid profiles are linked to dFNC in the ADS individuals.

Furthermore, functional and structural brain connectivity might act as intermediate biomarkers linking molecular pathology to clinical phenotypes. This promising shift in approach may aid the identification of early markers for neurodegenerative diseases, shed light on the molecular neurobiology of network connectivity disruption, and clarify the pathophysiological mechanism of diseases. 21 A hypothesis was proposed that lipid‐related genes and serum lipid profiles may disrupt brain networks, potentially leading to abnormalities in cerebrospinal fluid (CSF) biomarkers and cognitive decline in the ADS individuals.

This study examined the evolving trajectory characteristics of dFNC and its correlations with lipid‐related factors and general cognition among ADS individuals. Initially, the study analyzed the dFNC pattern at the level of large‐scale brain networks. Subsequently, a support vector machine (SVM) method was utilized to classify AD with abnormal dFNC as objective diagnostic biomarkers. Furthermore, the study investigated the connections of APOE‐ε4 status, the polygenes of lipid pathway, and serum lipids with dFNC. Lastly, the study explored the mediating effect of dFNC on the associations between lipid‐related indicators, biological markers, and cognitive phenotypes.

2 METHODS

2.1 Participants

Cross‐sectional data were obtained from Alzheimer's Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu) from June 24, 2010 to September 12, 2018. The diagnostic criterion and inclusion flowchart are shown in Supplementary Methods and Figure S1. This study included 181 ADS patients and 61 cognitive normal (CN) subjects with genetic, lipidomic, CSF, rsfMRI, and cognitive data available. All participants and collateral informants provided written informed consent at each site of ADNI before participation. The ADNI project was approved by all participating sites' institutional review board. This study was approved by the Research Ethics Committee of the affiliated Zhongda Hospital, Southeast University. The ADS patients were classified into subjective cognitive decline (SCD, n = 50), early amnestic mild cognitive impairment (EMCI, n = 75), late mild cognitive impairment (LMCI, n = 35), and AD groups (n = 21). The demographic information of age, sex, and education, and the general cognitive scores of Mini‐Mental State Examination (MMSE) and Alzheimer's Disease Assessment Scale‐13 item cognitive subscale (ADAS) were obtained from ADNI.

2.2 Genes, CSF and lipid biomarkers

This study used 11 well‐established AD risk genetic variants, including CLU rs11136000, 22 LDLR rs5930, 23 LRP1 rs1799986, 24 PICALM rs3851179, 22 SORL1 rs2070045, 25 Cholesterol ester transfer protein (CETP) rs5882, 26 ABCA1 rs2230808, 27 BIN1 rs744373, 28 APOE rs429358 and rs7412, 29 ABCA7 rs3764650, 28 and HMGCR rs3761740. 30 Although these genes did not belong to a specific single pathway, their eponymous proteins were involved in the biological processes of lipid metabolism. 31 Those variants in each gene with minor allele frequency more than 5% were included in the analysis (Table S1). This study used the effect size (odds ratio) of genetic variants multiplying their corresponding allele dosages to construct the weighted polygenic scores (PGS). The PGS excluding APOE (PGSexAPOE)was also calculated due to the strong APOE‐ε4 risk effect. 32

Besides, the concentrations of core CSF biomarkers, including Amyloid‐β 1 to 42 peptide (Aβ), tau phosphorylated at threonine 181 (pTau), and total tau (Tau), of each participant at baseline were obtained from the ADNI database. Serum lipids were obtained from the Alzheimer's Disease Metabolomics Consortium, which adds rich metabolomics datasets into the ADNI database. Each participant had 228 serum lipids.

2.3 Statistical analyses

2.3.1 Demographic and clinical data analysis

The Shapiro–Wilk test was first used to assess the normality of continuous variables. The Levene's test was also examined to assess the homogeneity of variance. One‐way analysis of variance (ANOVA) was applied to compare group differences of age, education, gray matter volume (GMV), framewise displacement (FD), PGS, Aβ, Tau, pTau, MMSE, and ADAS scores, separately. But if the data were not normally distributed, or the variance was unequal, non‐parametric Kruskal–Wallis test should be analyzed instead. Chi‐square test was adopted to compare group differences of categorical variables, including gender and APOE‐ε4 status. The significance threshold was set at p < 0.05. Post‐hoc analyses were analyzed with least significance difference (LSD) correction (p < 0.05). Statistical analyses were performed under SPSS 25.0 software (SPSS, Inc., Chicago, IL, USA).

2.3.2 MRI processing

The MRI acquisitions and processing are shown in Table S2 and Supplementary Methods. Briefly, meaningful independent components (ICs) were selected from rsfMRI data to composite eight functional networks: DMN, SAN, ECN, DAN, SMN, VIN, auditory network (AUN) and cerebellum network (CBN) (Figure S2 and Table S3). Then, a sliding window approach and k‐means clustering algorithm was used to obtain optimal centroid states. The number of optimal centroid states is two.

Group differences of dFNC were evaluated through five‐level one‐way ANOVA, and post‐hoc analysis via Bonferroni correction (p < 0.05). Additionally, due to temporal properties were skewness distributed, we tested group differences using Kruskal Wallis test and post‐hoc Bonferroni correction (p < 0.05). Group differences of topological metrics (global and local efficiency) variance were tested with five‐level one‐way ANOVA and post‐hoc LSD correction (p < 0.05). We then performed Spearman's correlations analyses between topological metrics, temporal properties and CSF, cognitive biomarkers. Moreover, we extracted the common differential connections of sFNC and dFNC at each state, and calculated the group‐level mean value of them, which were then used to classify subjects of one from the other groups and subjects of one from another group (Supplementary Methods).

2.3.3 Difference and correlations in lipid‐related factors subgroups

We selected 20 lipids that correlated to temporal properties with statistical significance (p < 0.00043, Bonferroni correction) to construct composite lipid score (Supplementary Methods and Table S4). We divided all subjects into separate two groups according to the median of lipid score, APOE‐ε4 allele status, and median of PGS, and compared dFNC and temporal properties difference in these pairwise subgroups (Mann–Whitney U‐test, p < 0.05). We also calculated group difference of dFNC per state via two‐sample t‐test (p < 0.05). Then, we examined the Spearman correlations of temporal properties and dFNC with CSF biomarkers and cognitive performance in these separate two subgroups.

2.3.4 Mediation analysis

We applied the simple linear mediation model (Model 4) under v3.5.3 PROCESS macro (https://www.processmacro.org) in SPSS to determine the mediation effect of dFNC on the relationships between serum lipids or lipid‐related genes and CSF or cognitive markers across all subjects (Supplementary Methods). 33

3 RESULTS

3.1 Clinical characteristics

As shown in Table 1, there was no difference in sex and education among groups. A significant decreasing trend in age was found as disease progressed, with the main difference appeared between CN and EMCI, LMCI, AD group. The gray matter volume of CN group was lower than SCD and EMCI group, but higher than that of LMCI and AD group, the former of which might be due to the older age with severe brain atrophy, whereas the latter was associated with severe neurodegenerative effect. There was no discrepancy in respect to PGS, but all groups except SCD group showed a relative increasing trend along the disease process. Significant difference concerning APOE‐ε4 status, CSF biomarkers and cognitive scores were found, with all groups except SCD group exhibiting certain trajectory changes as disease progressed. However, there was no difference in any of items between CN and SCD group.

TABLE 1 Description of demographic and clinical data across the AD spectrum populations.

Items	CN	SCD	EMCI	LMCI	AD	p value	
(n = 61)	(n = 50)	(n = 75)	(n = 35)	(n = 21)	
Age (y)	76.62 ± 0.77 a , b , c	74.54 ± 0.83	72.67 ± 0.77	72.17 ± 1.22	71.76 ± 1.68	0.001	
Gender (M/F)	31/30	20/30	41/34	21/14	12/9	0.375	
Education (y)	16.38 ± 0.32	16.54 ± 0.36	16.11 ± 0.32	16.54 ± 0.40	15.33 ± 0.58	0.395	
GMV (ml)	582.84 ± 6.55 c	594.97 ± 7.05 f	590.42 ± 6.57 h	571.48 ± 10.41 i	526.94 ± 14.39	<0.001	
FD (mm or°)	0.30 ± 0.02 c	0.32 ± 0.03 f	0.33 ± 0.02 h	0.35 ± 0.03 i	0.48 ± 0.07	0.012	
APOE‐ε4 status (+/−)	18/43	12/38	31/44	16/19	15/6	0.002	
PGS	7.93 ± 0.25	8.56 ± 0.33	8.54 ± 2.11	8.79 ± 0.35	9.07 ± 0.49	0.171	
Aβ (pg/mL)	193.95 ± 5.66 c	206.70 ± 6.96 e , f	191.23 ± 6.53 h	174.69 ± 9.10 i	140.40 ± 9.51	<0.001	
Tau (pg/mL)	64.93 ± 3.89 c	66.06 ± 4.26 f	79.97 ± 6.34 h	80.64 ± 7.94 i	129.29 ± 13.40	<0.001	
pTau (pg/mL)	31.29 ± 2.00 b , c	36.68 ± 3.20 f	35.35 ± 2.56 h	45.01 ± 5.24 i	55.23 ± 5.70	<0.001	
MMSE	28.61 ± 0.20 b , c	29.14 ± 0.15 d , e , f	27.93 ± 0.30 h	27.31 ± 0.52 i	22.67 ± 0.54	<0.001	
ADAS	13.30 ± 0.94 b , c	10.92 ± 0.64 d , e , f	15.41 ± 0.93 g , h	18.77 ± 1.33 i	35.81 ± 1.96	<0.001	
Note: Data of continuous variables were presented as mean ± standard deviation, and data of categorical variables as numbers. For categorical variables, p values were obtained via χ 2 test; for continuous variables, p values were obtained by one‐way ANOVA, but if the data were not normally distributed, or the variance was unequal, non‐parametric Kruskal‐Wallis test should be analyzed instead. Post‐hoc analyses were conducted with least significance difference (LSD) correction (p < 0.05).

Abbreviations: AD, Alzheimer's disease; ADAS, Alzheimer's Disease Assessment Scale‐13 item cognitive subscale; Aβ, Amyloid‐β 1 to 42 peptide; CN, cognitive normal; EMCI, early amnestic mild cognitive impairment; FD, framewise displacement; GMV, gray matter volume; LMCI, late mild cognitive impairment; M/F, male/female; MMSE, Mini‐Mental State Examination; PGS, polygenic scores; pTau, tau phosphorylated at threonine 181; SCD, subjective cognitive decline; Tau, total tau; y, years.

a Group difference between CN and EMCI subjects.

b Group difference between CN and LMCI subjects.

c Group difference between CN and AD subjects.

d Group difference between SCD and EMCI subjects.

e Group difference between SCD and LMCI subjects.

f Group difference between SCD and AD subjects.

g Group difference between EMCI and LMCI subjects.

h Group difference between EMCI and AD subjects.

i Group difference between LMCI and AD subjects.

3.2 Dynamic connectivity analysis

This study identified two reoccurring connectivity states in the entire data: state‐I, more frequent with weaker connectivity, and state‐II, less frequent, with stronger connectivity (Figure 1A). Both states showed fluctuating dFNC disruption within the CBN, whereas the dFNC gradually increased in other within‐network connectivity (WNC) as the disease progressed. All between‐network connectivity (BNC) gradually became stronger along the ADS patients, reaching maximum at the AD stage, matching group‐level dynamic patterns (Figure 1B,C). The connectivity trajectories significantly decreased at the SCD stage in WNC and BNC except for cerebellar WNC. The two states had significant group differences in WNC and BNC (Figure 1D). For instance, WNC in the two states had different DMN, DAN, VIN, and CBN, whereas BNC in the two states had differences between association cortex systems (including DMN, ECN, SAN, and DAN) and sensory cortex systems (including SMN, VIN, AUN and CBN). Post‐hoc pairwise contrasts revealed that the difference was mainly due to the differences between the AD and non‐AD groups for each state (Figure 1E).

FIGURE 1 Dynamic functional network connectivity analysis across all subjects. (A) Cluster centroids for each state across all subjects with the number of occurrences and percentage of occurrences were listed above the autocorrelation matrix (State I: Occurrences 15,625, frequency 59%; State II: Occurrences 10,995, frequency 41%). (B) Group‐specific mean cluster centroids for each state in CN and subgroups of AD spectrum subjects, respectively. The color bar indicates z values of dFNC. (C) Dynamic trajectory of within‐ and between‐network connectivity for each state over the disease progress. (D) P value matrix of group‐level differences of cluster centroids for each state among five groups. (E) p value matrix of post‐hoc pairwise group differences of cluster centroids after Bonferroni correction (p < 0.05). The dark pink in color bar represented more significant p values. (F) Percentage of total time spent in each state across all subjects. (G) Significant differences of temporal properties of dFNC including FW, mean DT, and NT among groups. (H) Variance of global and local efficiency across the dynamic network connectivity matrices are presented with violin plots in each group. The red solid line represents the median of variance, and the red dotted line represents the interquartile position. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. (I) and (J) Robust relationships among temporal properties of dFNC, core biomarkers of cerebrospinal fluid, and global cognitive performance across all subjects. (K) ROC curves from SVM classifiers to distinguish AD from non‐AD subjects of all other groups. Dynamic connectivity of state I (red line) and state II (green line) have better power in differentiating AD from non‐AD subjects than sFNC (blue line). (L) SVM model to classify subjects of one versus another group. FW1, Fractional windows of State I; FW2, Fractional windows of State II; DT1, Dwell time of State I; DT2, Dwell time of State II; ROC, receiver operating characteristic.

For dynamic temporal properties, the preclinical AD stage patients had more frequency in state‐I and less frequency in state‐II, whereas the AD patients showed the opposite trend. The AD and non‐AD subjects in each state had slight group‐level differences in fractional windows (FW) and dwell time (DT). Besides, there was no group difference in number of transitions (NT), but there was a gradually increasing trend with the disease process (Figure 1F,G). The variance of global efficiency had a decreasing trend with the disease course. The ADS patients had a lower variance than the CN subjects. In contrast, the variance of local efficiency was not different and showed no significant tendency among the groups (Figure 1H). However, there were no relevance between any network efficiency and CSF, cognitive biomarkers (p > 0.05). Furthermore, the FW and DT in state‐I were negatively correlated with Tau and pTau levels and positively correlated with MMSE scores (Figure 1I). However, the FW and DT in state‐II showed the opposite results (Figure 1J).

The SVM analysis indicated that the averaged common differential connectivity of state‐I and state‐II had better predictive power in differentiating AD from non‐AD, CN, and SCD subjects than sFNC. Although the differential connectivity of state‐I had a better ability to distinguish AD from EMCI and LMCI than state‐II, it could not recognize the preclinical stages of AD (Figure 1K,L). This study did not find any classification ability of serum lipids or lipid score (Table S5).

3.3 Associations between serum lipids and dFNC features

The temporal properties and serum lipids were positively correlated in state‐I, and negatively correlated in state‐II (Figure 2A). The count of observations changed for each state since not all subjects had dynamic windows assigned to each state. State‐I had more subjects than state‐II. In state‐I, the lipidscore_high group had more subjects than the lipidscore_low group, whereas in state‐II, the lipidscore_low group had more subjects than the lipidscore_high group (Figure 2B). The state‐I showed more differential connections between lipidscore_high group and lipidscore_low group than state‐II (state‐I: 63 connections; state‐II: 6 connections), and lipidscore_high > low group of state‐I had fewer connections than lipidscore_high < low group (29/34) in state‐I. Particularly, most altered connections were within VIN, and between association cortex systems and sensory cortex systems (Figure 2C–E). In addition, the differential connections were significantly correlated with CSF biomarkers and cognitive performance in the lipid score subgroups for each state. Similarly, state‐I had more correlated differential connections than state‐II, especially for DMN‐VIN, SAN‐AUN, ECN‐VIN, DAN‐CBN, and VIN‐VIN connections in lipidscore_high group. Particularly, more differential connections were correlated with MMSE and ADAS scores in the lipidscore_high group of state‐I, while there were scarcely few correlations in lipidscore_low group of any state or lipidscore_high group of state‐II (Figure 2F). The FW and DT in state‐I were significantly greater in lipidscore_high group than in lipidscore_low group, while those of state‐II were greater in lipidscore_low group than in lipidscore_high group (Figure 2G). The temporal properties were more significantly related to pTau levels in lipidscore_low group, than Tau levels and MMSE scores in lipidscore_high group, with the same correlation directions in the entire data (Figure 2H).

FIGURE 2 Network features of serum lipid profiles in relation to dynamic network connectivity and its temporal properties. (A) Heat map illustrated rank correlations of dFNC temporal properties with serum lipid components in each state. The color bar indicates the range of Spearman correlation coefficients. (B) Group specific cluster centroids for each state in subgroups divided via the median of lipid composite score, and the number of subjects (N) with at least one window in each state was shown above the correlation matrixes. (C) Group‐level difference (lipidscore_high minus lipidscore_low) in dFNC for each state (p < 0.05), with values plotted as the log of the p‐value with the sign of the associated t‐statistic: −sign(t)log10(p‐value). (D) Differential functional connectivity (FC) in each state, where lipidscore_high group had a stronger or weaker FC pattern compared with lipidscore_low group. (E) Numerical representation of significant differences in FC between the lipid score subgroups for each state with the bar charts. (F) Relationships of differential dynamic connections in each state correlating with cerebrospinal fluid core biomarkers and cognitive performance in separate lipid score subgroups. The color bar represents the range of Spearman correlation coefficients. (G) Difference of temporal properties in dFNC between low and high lipid composite scores groups (lipidscore_low vs lipidscore_high) in each state across all subjects. (H) Significant correlations of temporal properties in dFNC with pTau levels in lipidscore_low group but not in lipidscore_high group, while significant correlations of temporal properties in dFNC with total Tau levels and MMSE scores in lipidscore_high group but not in lipidscore_low group. For the abbreviations of lipid components, refer to Table S4.

3.4 Associations between lipid‐related genes and dFNC features

Most subjects in the APOE‐ε4 subgroups were in state‐I, and a few were in state‐II. The APOE ε4− group had more subjects than the APOE ε4+ group in each state (Figure 3A). The state‐II exposed more differential connections between APOE ε4+ group and APOE ε4− group than state‐I (state I/II: 29/70 connections), and APOE ε4+ > ε4− group of state‐II had 29 connections associated with CBN, VIN, DAN, and SAN, whereas APOE ε4+ < ε4− group of state‐II had 41 connections associated with DMN, SAN, ECN, DAN, SMN and VIN (Figure 3B–D). Besides, the differential dFNC related to both CSF biomarkers and cognition in state‐II of APOE ε4+ group were mainly associated with DMN‐DMN, SAN‐DAN, SAN‐VIN, SAN‐CBN, and VIN‐CBN connections (Figure 3E). In state‐I of APOE‐ε4 carriers, temporal properties were positively associated with Aβ levels, whereas in state‐I of APOE‐ε4 non‐carriers, temporal properties were negatively associated with Tau and pTau levels. Both in the APOE‐ε4 carriers and non‐carriers, the temporal properties were positively associated with MMSE, especially in the APOE‐ε4 carriers, which were negatively associated with ADAS in state‐I (Figure 3F).

FIGURE 3 Network features of APOE and polygenes involved in lipid pathways on dynamic network connectivity and its temporal properties. (A) Cluster centroids for subgroups according to where APOE‐ε4 allele exists or not, and the number of subjects (N) with at least one window in each state was shown above the correlation matrixes. (B) Group difference (APOE ε4+ minus APOE ε4−) in dFNC for each state (p < 0.05). The values were plotted as the log of the p‐value with the sign of the associated t‐statistic: −sign(t)log10(p‐value). (C) Functional connectivity (FC) in each state, where APOE ε4+ had a stronger or weaker FC pattern in comparison with APOE ε4− group. (D) Numerical representation of significant differences in FC between the APOE‐ε4 subgroups for each state with the bar charts. (E) Relationships of differential dynamic connections between APOEε4− and APOEε4+ groups with cerebrospinal fluid core biomarkers and cognitive performance in separate APOE‐ε4 subgroups. (F) Relationships between temporal properties and cerebrospinal fluid core biomarkers and cognitive performance in separate APOE‐ε4 subgroup. (G) Cluster centroids for subgroups divided by the median of lipid pathway‐based polygenic score (PGS). (H) Group difference (PGS_high minus PGS_low) in dFNC for each state (p < 0.05). (I) FC in each state, where PGS_high group had a stronger or weaker FC pattern comparing with PGS_low group. (J) Numerical representation of significant differences in FC between the PGS subgroups for each state with the bar charts. (K) Relationships of differential dynamic connections between PGS_low and PGS_high groups with cerebrospinal fluid core biomarkers and cognitive performance in separate PGS subgroups. (L) Relationships between temporal properties and cerebrospinal fluid core biomarkers and cognitive performance in separate PGS subgroup.

Also, most subjects in PGS subgroups were in state‐I, and a few were in state‐II. However, the number of subjects in each subgroup per state was slightly different (Figure 3G). The discrepancy in dFNC count between PGS_high and low groups in the two states (32/35) was minimal (Figure 3H–J). Only DAN‐CBN and DAN‐SMN connections of state‐II in PGS_high group displayed associations with both CSF biomarkers and cognition (Figure 3K). The temporal properties of PGS_low group were selectively positively related to Aβ and negatively related to pTau levels in state‐I. Both PGS_high and low subgroups showed positive and negative correlations between temporal properties and MMSE in state‐I and state‐II, respectively. Besides, the NT was negatively associated with MMSE in the PGS_high group (Figure 3L). The sFNC feature and its associations with serum lipids or lipid‐related genes were shown in Supplementary Results and Figures S3 and S4. The associations between PGSexAPOE and dFNC features are presented in Supplementary Results and Figure S5.

3.5 Mediation analysis

The relationships between APOE‐ε4 genotype and CSF, cognitive indicators were mediated by the differential state‐II connections between APOE‐ε4 carriers and non‐carriers. The state‐II connection in SAN‐VIN negatively mediated the relations between APOE and Aβ levels. For instance, the APOE‐ε4 carriers with higher state‐II SAN‐VIN connectivity predicted lower Aβ levels (Figure 4A). Similarly, the APOE‐ε4 carriers with higher state‐II DAN‐CBN connectivity predicted higher Tau and pTau levels, with higher state‐II SAN‐CBN connectivity predicted reduced MMSE scores, and with higher state‐II DAN‐VIN connectivity predicted elevated ADAS scores (Figure 4B). This study did not find any indirect effect of differential connections on the relations between lipid score or PGS subgroups on the CSF or cognitive indicators.

FIGURE 4 Differential connections of APOE ε4 subgroups mediated effects of APOE on cerebrospinal and cognitive phenotypes. (A) The differential dFNC in state II mediated the effects of APOE genotype on Aβ, Tau, and pTau levels across all subjects. (B) The differential state II dFNC mediated the relationships between APOE genotype and MMSE, ADAS scores across all subjects. Positive standardized regression coefficients (β) represented positive connections, while negative β indicated negative connections. *p < 0.001 or the 95% bootstrap confidence interval did not straddle zero. ACC, Anterior Cingulate Cortex; LFEF, Left Frontal Eye Field; LIns, Left Insula; LSPL, Left Superior Parietal Lobule; RFEF, Right Frontal Eye Field; RIns, Right Insula; RSPL, Right Superior Parietal Lobule.

4 DISCUSSION

This study explored the dynamics of large‐scale brain networks in ADS populations and highlighted their connections with lipid‐related factors. The primary findings of the study were as follows. First, the dFNC could be clustered into two stable and reoccurring connectivity states with distinct characteristics. Subjects in pre‐AD stages tended to spend more time in the weak‐connected state‐I with lower frequency, while AD patients tended to spend more time in the strong‐connected state‐II with higher frequency. Furthermore, the average strength of differential dFNC in state‐I demonstrated significant power in distinguishing AD from other groups compared with state‐II. Second, the lipidscore_high group exhibited prolonged time spent in state‐I, decreased Tau levels, and raised general cognitive performance. Similarly, the APOE ε4+ group showed increased time spent in state‐I, elevated Aβ levels, and enhanced general cognitive function. Also, the PGS_high group presented prolonged duration in state‐I and elevated cognitive function. Even after removing APOE, the PGSexAPOE_high group continued to show reduced Tau levels and improved general cognition. Third, the lipid‐related differential architectures of dFNC were primarily observed in the large‐scale inter‐networks between association cortex systems and sensory cortex systems of state‐I. In contrast, the differential connections during state‐II were found to play a role in mediating the associations between APOE genotype, and CSF biomarkers and cognitive performance across all participants. These findings indicate the importance of emphasizing the dynamic temporal patterns of large‐scale brain networks and their lipid‐related biological foundations in the ADS populations.

To our knowledge, a few studies have demonstrated the FC dynamics in AD with more concerns on comparing the dynamic features of AD with other types of dementia or healthy controls. In general, AD subjects preferred to have more occurrences and longer dwell time in a sparsely‐connected state, but show fewer occurrences and shorter dwell time in a tensely‐connected state. 5 , 6 , 34 Other recent studies have tried to inspect static brain networks in relations to several serum lipids. The total cholesterol and low‐density lipoprotein cholesterol interacted with gender could influenced the intra‐network connectivity of DAN, lateral VIN, and anterior DMN in young healthy adults. 20 Our previous work has disclosed that a few serum lipids, such as free cholesterol, esterified cholesterol, and sphingomyelins were associated with FC of DMN in the brain regions of postcentral gyrus, supplementary motor area, posterior cingulate cortex, and middle occipital gyrus. 35 Also, 26 serum lipoproteins were associated with WNC and BNC of 10 resting‐state networks. 36 Nevertheless, they never considered linking biological phenotype at different levels of lipid metabolism with large‐scale dynamic brain networks in separate states during the progressing stages of AD.

The transit dynamic patterns of states may reveal the integration and segregation of different functional networks that were characterized by coordinated activity among brain regions at every time point or within predefined time windows of the resting‐state MRI scan, which did not manifest in static analyses. 37 The resting brain activity switches among different states to generate a moment‐to‐moment fluctuation of the spatial patterns of resting‐state brain networks. 38 This study showed that distinctive states have different features. Subjects spent more time in weakly‐connected state‐I, where new connections were more easily established to promote higher network flexibility. 37 However, less frequent state‐II with strong connectivity might compensate for the loss of intertwined brain function under network destruction. The enhanced WNC may facilitate regional information processing, and increased BNC may promote remote information transmission. 37 In contrast to prior studies, 5 , 6 , 34 it was found that individuals with AD exhibited a reversal of the brain's preference for certain state, spending a longer duration with higher frequency in the strong‐connected state‐II compared with pre‐AD subjects. And contrary to the expected cascading network failure in the ADS patients, 39 the connectivity strength of AD reached its peak in both states. These findings suggest that state‐II may be indicative of a connectivity pattern associated with decompensation and limited compensatory capacity, whereas state‐I represents a connectivity pattern with normal compensability. 40 There was a significant group‐level network difference between association cortex systems and sensory cortex systems, similar to previous findings. 5 , 34 Sensory systems integrate visual and auditory signals and respond to the environment, while association cortex networks cope with attention, executive control, and other advanced cognition. Therefore, network disruption between the two systems might characterize biological and behavioral phenotyping in AD. Also, individuals with greater network variability had better behavioral tests, 41 while those with lower temporal variability had reduced network flexibility and poor network communication. Integration across networks (measured as global efficiency) were related to better working memory performance, whereas segregation (measured as local efficiency) related to poorer motor execution function. 42 Consistent with previous studies, 6 , 34 global efficiency was lower in AD than in healthy subjects. Regional information processing characterized by local efficiency was not damaged in AD. However, there were no relevance between any network efficiency and CSF, cognitive biomarkers. Moreover, the temporal properties and molecular, clinical phenotyping had an anti‐correlation pattern in the two states. Frequent occurrences and staying longer in state‐I symbolized alleviated tau pathology of CSF and improved general cognition, whereas frequently occurred and spending more time in state‐II embodied aggregated tau pathology and declined cognition. These findings indicate that FC of state‐I may serve to offset the effects of tau pathology and mitigate cognitive decline, whereas state‐II does not exhibit this compensatory capability. The SVM analysis also showed that dFNC had better predictive power than sFNC in distinguishing AD and non‐AD groups, especially for the dFNC in state‐I, yielded strong power to classify AD from other groups than in state‐II, suggesting the essentiality of extracting distinct dynamic states from the conventional sFNC.

With regard to serum lipid profile, aggregating various lipid components to derive a lipid score can provide insight into lipid metabolism in the bloodstream, thereby facilitating the examination of the relationships between blood lipids, FC, and disease progression within ADS populations. The state‐I showed more differential connections between lipid score subgroups than state‐II, and most of these differential connections of state‐I were correlated with general cognitive scores in the lipidscore_high group. The decreased FC might promote cognitive decline (positive correlation), whereas the increased FC might indicate compensatory processes for the deterioration of cognition (negative correlation). 40 Furthermore, the lipidscore_high group represented more time spent in state‐I and decreased Tau levels as well as raised general cognitive function, which might also indicate compensatory mechanism of dFNC in state‐I. As previous literatures described, high serum high‐density lipoprotein‐cholesterol was associated with preserved memory function in an elderly memory‐clinic population, 18 whereas higher serum triglyceride level was predictive of lower 3‐back accuracy in young adults. 19 Discrepancies were observed in the associations between serum lipids and cognition, highlighting the necessity for additional research to validate age‐specific and lipid species‐specific cognitive alterations in diverse disease populations. However, recent study showed that 27‐hydroxycholesterol contributes to phosphorylation of tau protein by impairing autophagy in mice model. 43 The potential role of elevated lipids in promoting tau pathology remains inconclusive and warrants further investigation.

Furthermore, a greater number of distinct connections were observed between individuals carrying APOE‐ε4 allele and those who did not in state‐II than in state‐I. This discrepancy may be attributed to the heightened susceptibility of ε4 allele to AD, potentially resulting in state‐II exhibiting more pronounced decompensated differential connections. Additionally, APOE‐ε4 carriers spent a longer duration in state‐I than in state‐II. The APOE ε4+ group exhibited increased time spent in state‐I, higher Aβ levels, and enhanced overall cognitive function. The observed enhancement of general cognition in the APOE ε4+ group potentially indicates compensatory mechanisms involving the dFNC of state‐I. Conversely, FC of state‐I cannot offset the effects of Aβ pathology with the presence of ε4 allele. Previous literature has documented that overexpression of APOE‐ε4 reduces APOE lipidation and promotes Aβ accumulation in the murine brain. 44 In comparison with APOE‐ε3 transgenic mice, APOE‐ε4 carriers exhibit lower levels of cholesterol and phosphatidylethanolamine, alongside higher levels of monounsaturated fatty acids in brain tissue. These lipid alterations are also correlated with increased Aβ levels in primary neuronal cultures. 45

This study did not reveal any disparities in dFNC count between high and low PGS group in two states. Following the removal of APOE, the discrepancy of dFNC between PGS subgroups revealed heightened within‐network connectivity in state‐I, suggesting that APOE may guide the direction of polygenes on disrupting between‐network connectivity of state‐I. To date, APOE continues to be recognized as a significant genetic risk factor for late‐onset AD, contributing to the disruption of large‐scale dynamic between‐network connectivity, hindering information exchange between networks, and ultimately leading to abnormalities of CSF biomarkers and cognitive decline. Furthermore, APOE is closely associated with other lipid‐related genes. 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 For instance, Aβ binds to endothelial LRP1 as a free peptide or bound to APOE2 and APOE3, is cleared across the BBB to circulation by receptor‐mediated transcytosis that is regulated by PICALM. 46 The function of ABCA1 membrane recycling on cholesterol efflux was diminished in astrocytes of APOE ε4/ε4 carriers. 48 Consequently, APOE plays a significant role in the lipid metabolic pathway and carries substantial weight in PGS. Additionally, temporal properties‐phenotyping relationships have been observed in the PGS_low group, potentially due to the higher PGS directing attention towards mitigating subtle effects. Also, the PGS_high group exhibited a prolonged duration in state‐I and enhanced cognitive function. Even after controlling for the presence of APOE, the PGSexAPOE_high group continued to stay longer in state‐I with reduced Tau levels and improved general cognition. It is worth noting that the negative correlation between NT and MMSE in the PGS_high group, suggests that polygenes may interrupt the communication between distinct states. As a result, elevated NT may disrupt cognitive representation of information processing.

Of particular importance, the unique connections in state‐II indirectly mediated the relationship between APOE genotype and CSF biomarkers, as well as cognitive performance. State‐II closely resembled the sFNC pattern, suggesting network decompensation of dFNC for cognition, whereas state‐I, possibly indicative of a compensatory process, did not demonstrate this mediation effect. The long‐range connections of the SAN, DAN interacting with the VIN, and the CBN are likely associated with top‐down attentional control. Visual attention, initiated in DAN, particularly in the frontal eye fields, coordinates neural resources through SAN, 56 , 57 facilitating attentional switching among various brain networks.

However, it is important to note that this study has several limitations. This study acknowledges the restraints of the dataset from the ADNI and suggests that future research should utilize diverse datasets to mitigate potential biases. Additionally, the study notes that the limited number of lipid‐related genes obtained from the ADNI database and recommends enriching this pathway in future research. Furthermore, the choices of genes for PGS or lipids for lipid‐score, or the cutoffs for each might influence interpretation of the data, or why lipid scores and PGS might lead to changes in connectivity still need to be careful about phrasing that implies causality. Lastly, there is a need for further investigation into the evolution of dynamic connectivity states over time, particularly in relation to the effects of pharmaceutical and non‐pharmaceutical treatments.

5 CONCLUSION

This study explored the dynamic large‐scale brain networks changes and their associations with lipid‐related factors traversing from normal cognition to AD. Our study revealed that individuals with pre‐AD exhibited a preference for prolonged periods in the weak‐connected state‐I, whereas those with AD tended to remain in the strong‐connected state‐II for extended periods. Additionally, the dFNC of state‐I demonstrated a higher discriminatory power in distinguishing AD from other groups compared with state‐II. Furthermore, the presence of the APOE‐ε4 allele, a high polygenic score, and elevated serum lipid levels were significantly associated with disruptions in the network connectivity between the association cortex system and sensory cortex system, which were indicative of enhanced cognitive function and potentially suggestive of a compensatory mechanism in the dFNC of state‐I. Therefore, this study provides insights into the dynamic neural reorganization, cortical plasticity, and treatment effect underlying AD pathophysiology.

AUTHOR CONTRIBUTIONS

FFZ and CMX designed the research. FFZ, XYL, DDF, and CCH analyzed the data. CMX and ZJZ supervised the research. FFZ and CMX drafted the figures and wrote this manuscript. After discussing the results, all authors reviewed and approved the final manuscript.

FUNDING INFORMATION

This study was granted funding by the National Key Projects for Research and Development Program of China (2016YFC1305800, 2016YFC1305802, CMX), the National Natural Science Foundation of China (82071204, 81871069, CMX), the Key Project for Research and Development Program of Jiangsu Province (BE2018741, CMX), the Nanjing International Joint Research and Development Project (201715013, CMX), and the Postgraduate Research & Practice Innovation Program of Jiangsu Province (KYCX19_0115, FFZ).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Supporting information

Data S1.

ACKNOWLEDGMENTS

Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health grant U01 AG024904), DOD ADNI (Department of Defense award number W81XWH‐12‐2‐0012), and the Alzheimer's Disease Metabolomics Consortium (National Institute on Aging R01AG046171, and 1RF1AG051550). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer's Association; Alzheimer's Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol‐Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann‐La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in ADNI database at http://adni.loni.usc.edu. The raw data required to reproduce these findings cannot be shared at this time due to the large size MRI data, but can be available from the corresponding author upon reasonable request. The corresponding author had final responsibility for the decision to submit for publication.
==== Refs
REFERENCES

1 Xie C , Goveas J , Wu Z , et al. Neural basis of the association between depressive symptoms and memory deficits in nondemented subjects: resting‐state fMRI study. Hum Brain Mapp. 2012;33 (6 ):1352‐1363.21618660
2 Song J , Qin W , Liu Y , et al. Aberrant functional organization within and between resting‐state networks in AD. PLoS OnePLoS One. 2013;8 (5 ):e63727.
3 Brier MR , Thomas JB , Snyder AZ , et al. Loss of intranetwork and internetwork resting state functional connections with Alzheimer's disease progression. J Neurosci. 2012;32 (26 ):8890‐8899.22745490
4 Sendi MSE , Zendehrouh E , Miller RL , et al. Alzheimer's disease projection from normal to mild dementia reflected in functional network connectivity: a longitudinal study. Front Neural Circuit. 2020;14 :593263.
5 Fu Z , Caprihan A , Chen J , et al. Altered static and dynamic functional network connectivity in Alzheimer's disease and subcortical ischemic vascular disease: shared and specific brain connectivity abnormalities. Hum Brain Mapp. 2019;40 (11 ):3203‐3221.30950567
6 Schumacher J , Peraza LR , Firbank M , et al. Dynamic functional connectivity changes in dementia with Lewy bodies and Alzheimer's disease. NeuroImage Clin. 2019;22 :101812.30991620
7 Picard C , Julien C , Frappier J , et al. Alterations in cholesterol metabolism‐related genes in sporadic Alzheimer's disease. Neurobiol Aging. 2018;66 :180.e1‐180.e9.
8 Kerchner GA , Berdnik D , Shen JC , et al. APOE ε4 worsens hippocampal CA1 apical neuropil atrophy and episodic memory. Neurology. 2014;82 (8 ):691‐697.24453080
9 Zhu Y , Gong L , He C , Wang Q , Ren Q , Xie C . Default mode network connectivity moderates the relationship between the APOE genotype and cognition and individualizes identification across the Alzheimer's disease spectrum. J Alzheimer's Dis. 2019;70 (3 ):843‐860.31282419
10 Bai F , Shi Y , Yuan Y , Xie C , Zhang Z . Immunity factor contributes to altered brain functional networks in individuals at risk for Alzheimer's disease: neuroimaging‐genetic evidence. Brain Behav Immun. 2016;56 :84‐95.26899953
11 Chang YT , Hsu SW , Huang SH , et al. ABCA7 polymorphisms correlate with memory impairment and default mode network in patients with APOEε4‐associated Alzheimer's disease. Alzheimer's Res Ther. 2019;11 (1 ):103.31831047
12 Sendi MSE , Zendehrouh E , Ellis CA , et al. The link between static and dynamic brain functional network connectivity and genetic risk of Alzheimer's disease. NeuroImage Clin. 2023;37 :103363.36871405
13 Zarrouk A , Riedinger JM , Ahmed SH , et al. Fatty acid profiles in demented patients: identification of hexacosanoic acid (C26:0) as a blood lipid biomarker of dementia. J Alzheimer's Dis. 2015;44 (4 ):1349‐1359.25428249
14 Llano DA , Devanarayan V . Serum phosphatidylethanolamine and Lysophosphatidylethanolamine levels differentiate Alzheimer's disease from controls and predict progression from mild cognitive impairment. J Alzheimer's Dis. 2021;80 (1 ):311‐319.33523012
15 Luo R , Su LY , Li G , et al. Activation of PPARA‐mediated autophagy reduces Alzheimer disease‐like pathology and cognitive decline in a murine model. Autophagy. 2020;16 (1 ):52‐69.30898012
16 Wong MW , Braidy N , Poljak A , Pickford R , Thambisetty M , Sachdev PS . Dysregulation of lipids in Alzheimer's disease and their role as potential biomarkers. Alzheimers Dement. 2017;13 (7 ):810‐827.28242299
17 Power MC , Rawlings A , Sharrett AR , et al. Association of midlife lipids with 20‐year cognitive change: a cohort study. Alzheimers Dement. 2018;14 (2 ):167‐177.28916238
18 Kinno R , Mori Y , Kubota S , et al. High serum high‐density lipoprotein‐cholesterol is associated with memory function and gyrification of insular and frontal opercular cortex in an elderly memory‐clinic population. NeuroImage Clin. 2019;22 :101746.30856540
19 Wang C , Cai H , Sun X , et al. Large‐scale internetwork functional connectivity mediates the relationship between serum triglyceride and working memory in young adulthood. Neural Plast. 2020;2020 :8894868.33204252
20 Zhang S , Li X , Sun Z , Chen Y , Yu Y . Impact of sex and serum lipids interaction on working memory: a large‐scale brain networks study. Brain Behav. 2023;13 (7 ):e3054.37165747
21 Pievani M , Filippini N , van den Heuvel MP , Cappa SF , Frisoni GB . Brain connectivity in neurodegenerative diseases—from phenotype to proteinopathy. Nat Rev Neurol. 2014;10 (11 ):620‐633.25287597
22 Harold D , Abraham R , Hollingworth P , et al. Genome‐wide association study identifies variants at CLU and PICALM associated with Alzheimer's disease. Nat Genet. 2009;41 (10 ):1088‐1093.19734902
23 Olgiati P , Politis AM , Papadimitriou GN , De Ronchi D , Serretti A . Genetics of late‐onset Alzheimer's disease: update from the alzgene database and analysis of shared pathways. Int J Alzheimer's Dis. 2011;2011 :832379.22191060
24 Wang Y , Liu S , Wang J , et al. Association between LRP1 C766T polymorphism and Alzheimer's disease susceptibility: a meta‐analysis. Sci Rep. 2017;7 (1 ):8435.28814781
25 Rogaeva E , Meng Y , Lee JH , et al. The neuronal sortilin‐related receptor SORL1 is genetically associated with Alzheimer disease. Nat Genet. 2007;39 (2 ):168‐177.17220890
26 Chen JJ , Li YM , Zou WY , Fu JL . Relationships between CETP genetic polymorphisms and Alzheimer's disease risk: a meta‐analysis. DNA Cell Biol. 2014;33 (11 ):807‐815.25105518
27 Wollmer MA . Cholesterol‐related genes in Alzheimer's disease. Biochim Biophys Acta. 2010;1801 (8 ):762‐773.20580938
28 Hollingworth P , Harold D , Sims R , et al. Common variants at ABCA7, MS4A6A/MS4A4E, EPHA1, CD33 and CD2AP are associated with Alzheimer's disease. Nat Genet. 2011;43 (5 ):429‐435.21460840
29 Xiao E , Chen Q , Goldman AL , et al. Late‐onset Alzheimer's disease polygenic risk profile score predicts hippocampal function. Biol Psychiatry Cogn Neurosci Neuroimaging. 2017;2 (8 ):673‐679.29560901
30 Porcellini E , Calabrese E , Guerini F , et al. The hydroxy‐methyl‐glutaryl CoA reductase promoter polymorphism is associated with Alzheimer's risk and cognitive deterioration. Neurosci Lett. 2007;416 (1 ):66‐70.17284348
31 Sun JH , Yu JT , Tan L . The role of cholesterol metabolism in Alzheimer's disease. Mol Neurobiol. 2015;51 (3 ):947‐965.24838626
32 Ahmad S , Bannister C , van der Lee SJ , et al. Disentangling the biological pathways involved in early features of Alzheimer's disease in the Rotterdam study. Alzheimers Dement. 2018;14 (7 ):848‐857.29494809
33 Hayes AF , Rockwood NJ . Regression‐based statistical mediation and moderation analysis in clinical research: observations, recommendations, and implementation. Behav Res Ther. 2017;98 :39‐57.27865431
34 Gu Y , Lin Y , Huang L , et al. Abnormal dynamic functional connectivity in Alzheimer's disease. CNS Neurosci Ther. 2020;26 (9 ):962‐971.32378335
35 Zang F , Zhu Y , Liu X , et al. Polygenic effects of the lipid metabolic pathway accelerated pathological changes and disrupted default mode network trajectory across the Alzheimer's disease Spectrum. J Clin Psychiatry. 2021;82 (6 ):20m13739.
36 Wang Q , Zang F , He C , Zhang Z , Xie C . Dyslipidemia induced large‐scale network connectivity abnormality facilitates cognitive decline in the Alzheimer's disease. J Transl Med. 2022;20 (1 ):567.36474263
37 Bonkhoff AK , Espinoza FA , Gazula H , et al. Acute ischaemic stroke alters the brain's preference for distinct dynamic connectivity states. Brain. 2020;143 (5 ):1525‐1540.32357220
38 Matsui T , Yamashita KI . Static and dynamic functional connectivity alterations in Alzheimer's disease and neuropsychiatric diseases. Brain Connect. 2023;13 (5 ):307‐314.35994384
39 Jones DT , Knopman DS , Gunter JL , et al. Cascading network failure across the Alzheimer's disease spectrum. Brain. 2016;139 (Pt 2 ):547‐562.26586695
40 Qi Z , Wu X , Wang Z , et al. Impairment and compensation coexist in amnestic MCI default mode network. Neuroimage. 2010;50 (1 ):48‐55.20006713
41 Jia H , Hu X , Deshpande G . Behavioral relevance of the dynamics of the functional brain connectome. Brain Connect. 2014;4 (9 ):741‐759.25163490
42 Cohen JR , D'Esposito M . The segregation and integration of distinct brain networks and their relationship to cognition. J Nneurosci. 2016;36 (48 ):12083‐12094.
43 Wang T , Zhang X , Wang Y , et al. High cholesterol and 27‐hydroxycholesterol contribute to phosphorylation of tau protein by impairing autophagy causing learning and memory impairment in C57BL/6J mice. J Nutr Biochem. 2022;106 :109016.35447319
44 Hu J , Liu CC , Chen XF , Zhang YW , Xu H , Bu G . Opposing effects of viral mediated brain expression of apolipoprotein E2 (apoE2) and apoE4 on apoE lipidation and Aβ metabolism in apoE4‐targeted replacement mice. Mol Neurodegener. 2015;10 :6.25871773
45 Lazar AN , Hanbouch L , Boussicaut L , et al. Lipid Dys‐homeostasis contributes to APOE4‐associated AD pathology. Cells. 2022;11 (22 ):3616.36429044
46 Nelson AR , Sagare AP , Zlokovic BV . Role of clusterin in the brain vascular clearance of amyloid‐β. Proc Natl Acad Sci USA. 2017;114 (33 ):8681‐8682.28765369
47 Tachibana M , Holm ML , Liu CC , et al. APOE4‐mediated amyloid‐β pathology depends on its neuronal receptor LRP1. J Clin Invest. 2019;129 (3 ):1272‐1277.30741718
48 Rawat V , Wang S , Sima J , et al. ApoE4 alters ABCA1 membrane trafficking in astrocytes. J Nneurosci. 2019;39 (48 ):9611‐9622.
49 Moulton MJ , Barish S , Ralhan I , et al. Neuronal ROS‐induced glial lipid droplet formation is altered by loss of Alzheimer's disease‐associated genes. Proc Natl Acad Sci USA. 2021;118 (52 ):e2112095118.34949639
50 Guimas Almeida C , Sadat Mirfakhar F , Perdigão C , Burrinha T . Impact of late‐onset Alzheimer's genetic risk factors on beta‐amyloid endocytic production. Cell Mol Life Sci. 2018;75 (14 ):2577‐2589.29704008
51 Johnson LA , Olsen RH , Merkens LS , et al. Apolipoprotein E‐low density lipoprotein receptor interaction affects spatial memory retention and brain ApoE levels in an isoform‐dependent manner. Neurobiol Dis. 2014;64 :150‐162.24412220
52 Jackson RJ , Rose J , Tulloch J , Henstridge C , Smith C , Spires‐Jones TL . Clusterin accumulates in synapses in Alzheimer's disease and is increased in apolipoprotein E4 carriers. Brain Commun. 2019;1 (1 ):fcz003.31853523
53 Zollo A , Allen Z , Rasmussen HF , et al. Sortilin‐related receptor expression in human neural stem cells derived from Alzheimer's disease patients carrying the APOE epsilon 4 allele. Neural Plast. 2017;2017 :1892612.28634550
54 Lee SI , Jeong W , Lim H , et al. APOE4‐carrying human astrocytes oversupply cholesterol to promote neuronal lipid raft expansion and Aβ generation. Stem Cell Rep. 2021;16 (9 ):2128‐2137.
55 Rodríguez E , Mateo I , Infante J , Llorca J , Berciano J , Combarros O . Cholesteryl ester transfer protein (CETP) polymorphism modifies the Alzheimer's disease risk associated with APOE epsilon4 allele. J Neurol. 2006;253 (2 ):181‐185.16096813
56 Veniero D , Gross J , Morand S , Duecker F , Sack AT , Thut G . Top‐down control of visual cortex by the frontal eye fields through oscillatory realignment. Nat Commun. 2021;12 (1 ):1757.33741947
57 Uddin LQ . Salience processing and insular cortical function and dysfunction. Nat Rev Neurosci. 2015;16 (1 ):55‐61.25406711
