Keywords
Alzheimer Disease - Magnetic Resonance Imaging - Longitudinal Studies - Cognitive Dysfunction - Biomarkers
Palavras-chave
Doença de Alzheimer - Imageamento por Ressonância Magnética - Estudos Longitudinais - Disfunção Cognitiva - Biomarcadores
INTRODUCTION
Alzheimer disease (AD) is a progressive neurodegenerative disorder that affects brain function, leading to cognitive decline and behavioral changes. It is the most common cause of dementia.[1] Currently, the pharmacologic therapies available for AD have been unable to arrest disease progression and consequent neuronal damage.[2] Hence, early identification and tracking of AD are essential in enhancing the patient's quality of life. Researchers are attempting to track AD progression and localize brain damage, in the hopes of improving existing treatment plans and thereby preventing the development of AD and reducing its severity.[3]
Neuroimaging methods, such as structural magnetic resonance imaging (MRI) scans, functional MRI (fMRI) scans, positron emission tomography (PET) scans, and electroencephalograms (EEGs) have their strengths and limitations compared to resting-state fMRI (rs-fMRI) in AD. Structural MRI provides detailed anatomical information, while fMRI captures brain activity; PET scans offer molecular imaging, and EEG measures electrical activity. Comparatively, rs-fMRI could offer insights into both structural and functional connectivity changes over time, potentially providing a more comprehensive view of AD[4] progression.
Resting-state fMRI is a technique that measures brain activity while a person is at rest, not performing any specific tasks. It helps identify spontaneous fluctuations in the brain's neural activity and functional connectivity (FC) patterns, providing insights into the brain's intrinsic organization. Resting-state fMRI is widely used to study brain networks, neurological conditions, and the impact of various interventions on brain function.[5]
Functional connectivity in rs-fMRI refers to the temporal correlation between the spontaneous low-frequency fluctuations in the blood oxygen level-dependent (BOLD) signal across different brain regions. It can be further classified into static functional connectivity (SFC) and dynamic functional connectivity (DFC): SFC indicates the average correlation between brain areas during the rs-fMRI scan, and DFC captures recurring connectivity patterns, which are essential to understand temporal variability in brain organization.
Neuroimaging research has predominantly focused on SFC analysis.[6] Previous investigations on AD-based SFC has revealed a decreased connection in significant regions associated with cognition, including the hippocampus,[7] prefrontal cortex, posterior cingulate cortex (PCC)/precuneus,[8] and the inferior parietal lobule,[9] which constitute the task-negative default mode network.[10] Memory and attention-related regions such as the insula,[11] anterior cingulate cortex (ACC),[12] thalamus,[8] and amygdala[13] have also been reported to exhibit alterations in AD. Furthermore, AD patients exhibited stronger increased connectivity in the occipital and temporal regions.[14] Thus, SFC analysis is a useful way to study potential regional alterations that could lead to AD in people at risk or in the early stages.[15]
The dynamic aspects of brain activity in AD reveal more distinct information such as fluctuations in functional connectivity patterns over time and variability in neural network dynamics.[16] The identification of distinct meta-states, representing recurring connectivity patterns in AD rs-fMRI studies, significantly enhances our understanding of the dynamic brain function during the progression of the disease.[17] The most common approach to assess these meta-states in rs-DFC is by the sliding window approach, whereby the fMRI data are segmented in (overlapping) windows, and functional interconnections between different brain areas are assessed within each window. Analyzing DFC provides insights into the nuanced perspective of the underlying neural mechanisms and potential biomarkers for the early identification of AD.[18]
There are few published studies that have explored DFC in AD; one of them,[19] while demonstrating the nonstationary nature of the brain's modular organization, used AD only for validation. A past study compared the static and dynamic brain functions in AD to those of other subtypes of dementia.[20] Resting-state functional network analysis formed the framework in all these studies. However, there is a lack of comprehensive and in-depth analyses of the region-based DFC in AD relative to healthy control (HC) individuals.
In the present study, we investigated the static and dynamic mean regional FC in both healthy and AD patients at baseline and after 1 year. By combining the SFC and DFC, we aimed to explore rs-fMRI time series at a finer scale. Furthermore, longitudinal dynamic FC analysis for region-specific assessment has not been previously conducted. The present study aims to bridge the gap by integrating the SFC and DFC to demonstrate the longitudinal differences in AD patients throughout 1 year and identify potential region-based biomarkers of AD.
METHODS
Participants data acquisition
The participants selected were scanned using 3.0 Tesla Philips (Amsterdam, Netherlands) scanners under the Alzheimer's Disease Neuroimaging Initiative (ADNI) protocol, with the individuals keeping their eyes open. The image collection was acquired using a single-shot sensitivity encoding (SENSE) gradient and echo planar imaging (EPI) with 48 slices, covering the entire brain in 140 volumes, a repetition time (RT) of 3 seconds, and an echo time (EC) of 30 milliseconds. During these scans, all the participants followed an identical procedure, and the participant's flow is as illustrated in [Figure 1].
Figure 1 Participant flowchart.
Data preprocessing
Standard preprocessing routines were applied to the rs-fMRI data using Data Processing Assistant for Resting-State fMRI (DPARSFA)[21] and Statistical Parameter Mapping (SPM 8)[22] toolbox based on MATLAB 2013a platform. The preprocessing pipeline included the following steps:[23]
-
Discarding the first 10 time points to enable signal stabilization.
-
Implementing slice timing correction to remove slice-dependent delays in the data.
-
Realignment of images and constraining head motion to less than 2 mm or 2° to mitigate motion artifacts.
-
Co-registering the structural and functional images to align them spatially.
-
Using the EPI template provided in SPM8 for spatial normalization.
-
Finally, applying spatial smoothing using a Gaussian kernel with a 6-mm full-width half-maximum (FWHM).
Ethical statement
Data collection and sharing for this project was funded by the ADNI (National Institutes of Health Grant U01 AG024904).
Spatial component extraction
Group independent component analysis (GICA) of the fMRI toolbox (GIFT) was employed in the extraction of spatially constrained components. The spatially constrained independent component analysis (C-ICA)[24] of rs-fMRI was used to address challenges in identifying components of interest and determining the optimal number of components in the ICA analysis. This method incorporates available prior spatial information about the sources into standard blind ICA. Previous studies have shown the benefits of using C-ICA in fMRI analysis.[24] This approach offers several strengths: it is fully automated, adapting component maps and time courses to individual subjects, simplifying cross-validation, and ensuring comparability across studies.
In the present analysis, the desired source included eighty regions of interest which were grouped under six brain regions: frontal, occipital, parietal, and temporal lobes along with basal ganglia and subcortical regions, as mentioned in the Supplementary Material (https://www.arquivosdeneuropsiquiatria.org/wp-content/uploads/2024/04/ANP-2023.0272-Supplementary-Material.docx) [refer Table S1: list of independent components and regions]. The WFU_ Pick Atlas toolbox[25] supported by SPM8 generated eighty region masks based on the automated Anatomical Labelling (AAL) brain standard atlas.[26] Before proceeding further, additional postprocessing such as detrending, 3dDespike, and filtering using fifth-order Butterworth low-pass filters with a cut-off of 0.15 Hz were performed on the obtained time courses.[27]
Functional connectivity investigation
Functional connectivity measures the temporal association of functional activation in various brain areas, and it can be quantified through pairwise Pearson correlation coefficients, covariance, or mutual information between time series. The present study employed both SFC and DFC pipelines ([Figure 2]). The SFC is a conventional approach that measures the correlation of time courses in large spatial components whereas the DFC is a more contemporary form of brain functional analysis. Of late, this windowed approach is being used to operate on smaller segments of time courses.[28]
Figure 2 Schematic representation of data analysis.
Static functional connectivity (SFC)
Static FC evaluates the correlations between the spatial brain map time courses, assuming that the statistical properties of these time courses do not change over time.[29] The MANCOVAN framework, a part of the GIFT toolkit, was used to conduct the SFC analysis by evaluating the changes in eighty spatial IC pairings for FC. Pearson correlation between all pairs of ICs was computed and converted using the Fisher Z-transform. The significant connectivity differences regarding the baseline and 1-year assessments were evaluated through statistical analysis.
Dynamic functional connectivity (DFC)
Dynamic FC during the resting state is considered altered across the scan cycle. Furthermore, it serves as a novel biomarker of complex brain functional structure by acquiring time-dependent connections over short periods.[6] We used the GIFT's temporal DFC toolbox to analyze DFC, calculating it across eighty ICA time processes using the sliding window approach, which employed the convolution method.[30] The covariance matrix estimation in all individuals was performed using L1-regularization, repeated ten times. The resultant functional connectivity matrices were converted to Z-scores to improve normality. Thus, the estimated individual matrices represented the dynamic changes during the complete rs-fMRI scan period and were used for further connectivity state analysis.[31]
Statistical analysis
Demographic and clinical data were analyzed using statistical methods, and the Chi-squared test was used for gender differences. One-way analysis of variance (ANOVA) was used to assess differences in age and Mini-Mental State Examination (MMSE) scores between AD patients at baseline (AD0), HCs at baseline (HC0), AD patients after 1 year (AD1), and HCs after 1 year (HC1). Further, the SFC and DFC between the groups were evaluated using analysis of covariance (ANCOVA) in GIFT, while regressing out age and gender covariates, which provides a way of statistically controlling the effect of covariates. In the present study, values of p lower than 0.001 with a false discovery rate (FDR) corrected for multiple comparisons were considered statistically significant.
RESULTS
Participants
The data for this research was sourced from the Alzheimer's Disease Neuroimaging Initiative 2 (ADNI2) database.[32] It is a consortium of universities and medical centers established to develop standardized imaging techniques and biomarker procedures in individuals with normal, mild cognitive impairment (MCI) and with AD. We selected AD and NC subjects with both rs-fMRI and structural MRI scans over two visits in a span of 1 year. The scans of 35 individuals comprising 15 AD and 20 NC were taken up for analysis. [Table 1] below provides an overview of the clinical and neuropsychological characteristics of the groups.
Table 1
Demographics of the study sample
|
Groups
|
AD0 (n = 15)
|
HC0 (n = 20)
|
AD1 n = 15)
|
HC1 (n = 20)
|
Test
|
p-value
|
|
Sex (M:F)
|
6:9
|
8:12
|
6:9
|
8:12
|
Chi-squared
|
0.72
|
|
Age (years)
|
74.5 ± 7.08
|
72.79 ± 5.24
|
75.57 ± 7.09
|
73.83 ± 5.23
|
ANOVA
|
0.61
|
|
MMSE
|
|
|
19.4 ± 4.37
|
29.15 ± 1.18
|
ANOVA
|
0.0016
|
|
CDR
|
0.5-1
|
0
|
0.5-2
|
0
|
|
|
Abbreviations: AD0, Alzheimer patients at baseline; AD1, Alzheimer patients after 1 year; ANOVA, analysis of variance; CDR, clinical dementia Rate; F, female; HC0, healthy controls at baseline; HC1, healthy controls after 1 year; M, male; MMSE, Mini-Mental State Examination.
Static functional connections
The individual SFC patterns of AD0, AD1, HC0, and HC1 obtained through the one-sample t-test demonstrated decreased connections in the AD group compared to the HC group both at baseline and after 1 year ([Figures 3A,B]). The heat-map indicates the strength of the connections between brain regions where, darker red color indicates a higher positive connection and darker blue color represents higher negative connection. In the case of FC, a significant variation was observed in the right medial-orbital-frontal-gyrus region at baseline, while in the 1-year assessment, the right hippocampus and left caudate appeared to be altered ([Figure 4A]). These regional changes were statistically significant, with p< 0.001 after FDR correction.
Figure 3 (A) Static functional connectivity averaged across subjects at baseline; (B) static functional connectivity averaged across subjects after 1 year.
Figure 4 (A) Static functional connectivity alterations at baseline and after 1 year; (B) static functional connectivity between regions at baseline; (C) static functional connectivity between regions after 1 year.
The ANCOVA test compared SFC at baseline (AD0-HC0) and after 1 year (AD1-HC1) to identify component alterations. Significant functional alterations in the AD group at baseline showed increased connectivity between the left superior-occipital-gyrus region and the precuneus, and also displayed decreased connectivity between the precuneus and left the thalamus, the right middle-occipital-gyrus and left superior-occipital gyrus regions, and the left hippocampus and right postcentral regions ([Figure 4B]).
However, in the follow-up after 1 year, decreased connectivity was observed between the right superior-temporal-pole and right middle-occipital-gyrus regions, and decreased FC, between the left lingual-right medial orbital frontal and left fusiform- left hippocampus. Additionally, an increased FC between the right superior-temporal-pole and right insula was also observed ([Figure 4C]).
Dynamic functional connections
The DFC matrices for baseline and the 1-year follow-up were obtained through the K-means clustering algorithm. We were able to identify three highly-structured FC states that recurred throughout individual scans and across subjects. Significant differences between the groups were observed in states 1 and 3, with greater frequency.
The baseline and 1-year group comparisons were performed using ANCOVA ([Figures 5A,B]). Apart from repetition of the SFC, additional increased FC was observed between the left lingual and left fusiform regions, and decreased FC, between the left lingual-right medial-orbital-frontal and left fusiform- left hippocampus. In the 1-year DFC assessment, we also observed increased connectivity between the right superior-temporal and right middle-frontal regions, as well as between the PCC and the right middle-temporal-pole region.
Figure 5 (A) Altered dynamic functional connectivity at baseline in both groups (AD0-HC0) for states 1 and 3; B) altered dynamic functional connectivity after 1 year in both groups (AD1-HC1) for states 1 and 3.
As a result, in the AD group, alterations within the occipital and subcortical regions, as well as the frontal-subcortical and basal-subcortical regions, were observed in the baseline dynamic analysis. The propagation was directed within the temporal, temporal-frontal, temporal-subcortical, and temporal basal regions over the course of 1 year. The findings were noted to be statistically significant, with p < 0.001after FDR correction.
DISCUSSION
In the current study, we present a region-specific analysis using C-ICA over a 1-year period in the context of AD research, uncovering alterations in FC. By integrating static and dynamic approaches, our investigation focuses on identifying abnormalities within the frontal, occipital, parietal, temporal, and subcortical regions that are susceptible to the onset of AD within 1 year. The present research aims to offer valuable insights into the regional dynamics of FC abnormalities, underlining the importance of considering both static and dynamic conditions in longitudinal studies.
Static functional alterations at baseline
Exploring the static functional alterations at baseline provides key insights into AD, as numerous studies[33] on cognitive decline have reported altered SFC in multiple brain regions associated with AD. Notably, the present study reveals significant baseline alterations (AD0) in the right medial-orbital-frontal gyrus region, a key region within the frontal lobe linked to cognitive functions such as self-evaluation, decision-making, and emotion processing.[34] This aligns with existing evidence indicating correlations between functional abnormalities in medial frontal brain areas and neurodegenerative illnesses, including AD.[35]
A decrease in SFC was observed, specifically between the right middle-occipital and left superior-occipital regions, which are associated with visual processing and perception. Previous research[36] has emphasized the occipital lobe as a crucial area of interest, distinguishing AD patients from those without AD. The functional alterations between the left postcentral gyrus, responsible for processing somatosensory information, and the left hippocampus, responsible for verbal memory, have been reported to be altered due to AD.[14] The present study revealed reduced connectivity between the left postcentral gyrus and the left hippocampus, indicating these as prominent regions in the context of AD.
Neuroimaging studies have revealed a strong structural relationship between the thalamus and precuneus, suggesting a potential pathway for modulation during impaired consciousness.[37] The static baseline analysis of the present study revealed reduced connectivity between the precuneus and the left thalamus, both of which are part of the default mode network. We have also observed increased connectivity between the precuneus and left superior-occipital gyrus, possibly reflecting enhanced integration of visual information with other cognitive processes in AD.
Dynamic functional alterations at baseline
Dynamic functional alterations at baseline (AD0) emphasize the significant involvement of the left fusiform and left lingual regions. The lingual gyrus, part of the occipital lobe that is associated with visual and cognitive processes, showed increased connectivity with the left fusiform gyrus, which is involved in visual and language-related cognitive functions. Notably, a previous study[38] on individuals with both depressive and non-depressive AD has highlighted the potential importance of the lingual gyrus in understanding the neurophysiology of depression in AD.
Furthermore, the left lingual gyrus exhibited decreased DFC with the right medial-orbital frontal gyrus, corroborating the findings from studies on abnormal cortical networks. The lingual gyrus and orbital frontal gyrus are indicated to be the hub regions of the cortical networks, suggesting their involvement in AD.[39]
An important observation of the present study was the presence of negative connectivity between the left hippocampus and the left fusiform. Previous studies[40] have identified DFC changes in the left hippocampus, which is a significant region for early AD detection. Another previous study[41] has reported altered connectivity between the fusiform and para-hippocampal gyri, which provided additional support for the presented findings. These dynamic alterations contribute to a deeper understanding of the evolving connectivity patterns in AD at its baseline stage.
Static functional alterations at the 1-year follow-up
Examining static functional alterations at the 1-year follow-up (AD1) revealed increased engagement of the right insula and right hippocampus in the AD patients. The insula, a critical brain hub with connections to cortical, limbic, and paralimbic regions, influences cognition, emotions, and sensory perception.[11] Previous studies[42] have identified disruptions in the insula among individuals with AD, potentially impacting the progression and manifestation of AD symptoms. Meanwhile, the right hippocampus, which is involved in spatial memory and emotional behavior, shows alterations in functional connectivity that have been documented in the early stages of AD.[43] The findings of the present are consistent with previous research reporting disruptions in hippocampal functional connectivity.
The hippocampus is crucial for memory formation and retrieval, while the caudate nucleus is associated with functions such as learning, memory, and decision-making. the current research has revealed increased connectivity between the right hippocampus and the left caudate regions, suggesting a unique potential alteration in the neural network underlying memory processes in individuals with AD.
Additionally, the insula, which acts as the central hub of the salience network, has shown increased connectivity with the right superior-temporal pole. Both of these regions play a significant role in the neurodegeneration observed in AD, particularly concerning social cognition. Previous research[44] has documented atrophy in the insular regions and the loss of grey matter in superior temporal regions due to AD.
Dynamic functional alterations at the 1-year follow-up
In addition to the results shown by SFC, the DFC analysis identified substantial involvement of the right middle-occipital, PCC, and regions of the temporal lobe. One of the novel findings from the present study is the negative connectivity between the right middle-occipital region and the right superior-temporal pole region, which is associated with visual and auditory processing, in individuals with AD.[45] This observation is more specific than that of the previous study,[46] which provided a more general overview, indicating the involvement of the middle-occipital-gyrus and superior-temporal-gyrus in AD.[46]
Another significant observation of the present study, is the negative connectivity between the PCC,[47] which is associated with the default mode network, and the right middle-temporal pole, which is responsible for processing audio-visual emotion. A previous study[48] on abnormal cortical networks in MCI and AD has indicated that dysfunction in the middle temporal regions and PCC is associated with depressive symptoms in AD.
Another previous study[49] conducted among patients with early- and late-onset AD, using voxel-based morphometry analysis, has highlighted the structural abnormality of the superior-temporal-gyrus and midle-frontal-gyrus regions. More specifically, the authors[49] observed negative connectivity between the right part of the superior-temporal-gyrus and middle-frontal-gyrus regions, both of which are known to be commonly associated with social cognition functions.
The hippocampus palys a crucial role in AD, being involved in functional disconnection with other parts of the brain.[50] In the DFC analysis of the present study, a notable observation was the progression of alterations from the left hippocampus toward the right hippocampus throughout 1 year. This observation illustrates the propagation of memory deterioration over the longitudinal course of AD.[7]
While the current study yielded valuable results and crucial observations, it is important to acknowledge the limitations associated with our findings. The sample size was small, limiting the ability to detect true effects, generalize findings, and potentially leading to less reliable results and an increased risk of false negatives. Moreover, the 1-year follow-up limited the monitoring of regional changes. Extending this period would have enabled us to better track the long chronic course of AD and disease progression. Despite these limitations, our results indicate that combining static and dynamic connectivity analyses could enhance the identification of abnormalities in AD.
Significantly, the present research contributes to identifying the specific regions primarily associated with disease development. This approach could lead to improved understanding and monitoring of AD progression, offering valuable insights for future research and potential clinical application.
In conclusion, the current study represents an innovative approach by integrating both SFC and DFC analyses to observe changes in core brain regions longitudinally. In the baseline analysis of AD, we identified connectivity alterations in key regions associated with consciousness and perceptions, namely the frontal, occipital, and core basal ganglia. Over the subsequent year, these alterations appeared to propagate towards regions in the temporal lobe and caudate, particularly those engaged in memory processes.
Significant cognitive regions affected in the AD group at baseline included the right cortex regions such as medial-orbital-frontal gyrus, middle-occipital gyrus, postcentral gyrus, and left cortex regions, such as the superior-occipital gyrus, thalamus, fusiform gyrus, lingual gyrus, hippocampus, and precuneus. In the AD analysis after 1 year, alterations were found in regions of the temporal lobe and basal ganglia, including the right part of the superior temporal gyrus, middle temporal pole, hippocampus, insula, left caudate, and PCC.
This comprehensive examination using SFC and DFC measures enhances our understanding of region-specific differences in neurological data in AD. The aforementioned regions significantly contribute to the diagnosis and tracking of AD. Moreover, the findings of the present study highlight the potential value of future research involving longitudinal scans with multiple time points over extended periods to validate and further build upon our results.
Bibliographical Record
Kuppe Channappa Usha, Honnenahally Ningappa Suma, Abhishek Appaji. Regional-based static and dynamic alterations in Alzheimer disease: a longitudinal study. Arq Neuropsiquiatr 2024; 82: s00441787761.
DOI: 10.1055/s-0044-1787761
