Shared and specific dynamics of brain activity and connectivity in amnestic and nonamnestic mild cognitive impairment

Abstract Aims The present study aimed to compare temporal variability in the spontaneous fluctuations of activity and connectivity between amnestic MCI (aMCI) and nonamnestic MCI (naMCI), which enhances the understanding of their different pathophysiologies and provides targets for individualized intervention. Methods Sixty‐five naMCI and 48 aMCI subjects and 75 healthy controls were recruited. A sliding window analysis was used to evaluate the dynamic amplitude of low‐frequency fluctuations (dALFF), dynamic regional homogeneity (dReHo), and dynamic functional connectivity (dFC). The caudal/rostral hippocampus was selected as the seeds for calculating dFC. Results Both aMCI and naMCI exhibited abnormal dALFF, dReHo, and hippocampal dFC compared with healthy controls. Compared with individuals with naMCI, those with aMCI exhibited (1) higher dALFF variability in the right putamen, left Rolandic operculum, and right middle cingulum, (2) lower dReHo variability in the right superior parietal lobule, and (3) lower dFC variability between the hippocampus and other regions (left superior occipital gyrus, middle frontal gyrus, inferior cerebellum, precuneus, and right superior frontal gyrus). Additionally, variability in dALFF, dReHo, and hippocampal dFC exhibited different associations with cognitive scores in aMCI and naMCI patients, respectively. Finally, dReHo variability in the right superior parietal lobule and dFC variability between the right caudal hippocampus and left inferior cerebellum exhibited partially mediated effects on the different memory scores between people with aMCI and naMCI. Conclusion The aMCI and naMCI patients exhibited shared and specific patterns of dynamic brain activity and connectivity. The dReHo of the superior parietal lobule and dFC of the hippocampus‐cerebellum contributed to the memory heterogeneity of MCI subtypes. Analyzing the temporal variability in the spontaneous fluctuations of brain activity and connectivity provided a new perspective for exploring the different pathophysiological mechanisms in MCI subtypes.


| INTRODUC TI ON
Mild cognitive impairment (MCI) is considered a transitional stage between normal aging and dementia 1 and can be divided into amnestic mild cognitive impairment (aMCI) and nonamnestic mild cognitive impairment (naMCI). MCI subtypes are not only theoretical but also underpinned by different pathophysiologies and disease trajectories; 2 aMCI is more likely to develop into Alzheimer's disease, 3,4 and naMCI is more related to other kinds of dementia, such as vascular dementia or dementia with Lewy bodies. 5 Additionally, MCI subtypes differ in aspects of susceptible genes, cardiovascular risk factors, progression courses, 6,7 and patterns of brain abnormalities. 8 Therefore, a deeper understanding of the differences in MCI subtypes will not only contribute to the prediction of dementia type but also provide more therapeutic strategies for preventing the development of dementia.
The different patterns of brain abnormalities between aMCI and naMCI patients have been repeatedly revealed by magnetic resonance imaging (MRI) research. For structural MRI, studies have demonstrated that there are significant differences in the morphology and integrity of gray matter 8,9,10 and white matter between aMCI and naMCI patients. 11,12 Additionally, functional MRI studies suggested that aMCI and naMCI patients exhibited differences in activity and connectivity: (1) aMCI patients exhibited a decreased amplitude of low-frequency fluctuations (ALFF) in the superior temporal gyrus, insula, precentral gyrus, lingual gyrus, and superior frontal gyrus compared with naMCI groups and controls; 13,14 (2) aMCI patients but not naMCI patients exhibited decreased regional homogeneity (ReHo) in the anterior cingulate gyrus compared with controls; 14 (3) compared with controls, aMCI patients and naMCI patients exhibited a different pattern of functional connectivity (FC) between the hippocampus and posterior cingulate cortex 15 and FC within the default mode network; 16 and (4) aMCI patients and naMCI patients exhibited different patterns of activation in temporalparietal regions during memory recognition compared with controls. 17 Moreover, aMCI and naMCI patients exhibited opposite associations between Theory of mind performance and FC between the bilateral temporal pole and the left lateral temporal cortex. 18 All the mentioned studies mainly focus on the static aspect of functional abnormalities, which assume that brain activity and connectivity are static over a whole resting-state functional MRI scan. However, evidence from both task-based fMRI studies and animal electrophysiology demonstrates that functional activity and connectivity may exhibit dynamic changes within time scales of seconds to minutes. 19 Additionally, spontaneous fluctuations in brain activity and connectivity have long been recorded in electrophysiological recordings of single cells, local fields, and surface electroencephalograms. 20 Therefore, important information can be missed when using average functional activity connectivity as the analytical method. Compared with stationary analyses, dynamic analyses facilitate the observation of details that are averaged out in stationary analyses and may offer greater insight into the fundamental mechanisms of activity and connectivity. Additionally, dynamic analyses enable the capture of spontaneously reoccurring patterns of activity and connectivity, which is essential for understanding the temporal variability in the intrinsic organization of the brain. 21 By using the dynamic sliding window method throughout the scanning procedure, the dynamic characteristics of brain function, such as dynamic ALFF (dALFF), dynamic ReHo (dReHo), and dynamic FC (dFC), can be captured effectively. 22,23 Several researchers have successfully applied dynamic analyses to neuropsychiatric diseases, such as AD, 24 Parkinson's disease, 25 bipolar disease, depression, 26,27 and schizophrenia, 28 which provide a novel understanding of their pathophysiologies.
For MCI individuals, studies suggested that they exhibited different patterns of dALFF compared with healthy controls in the working memory state, 29 and the dALFF in the left calcarine cortex was higher in MCI patients than in AD patients. 30 Additionally, a combination of dFC improved the diagnostic performance of MCI from healthy controls. 31,32 This evidence suggests that dynamic analyses enable the capture of spontaneously reoccurring patterns of activity and connectivity in patients with MCI and AD, which provide better knowledge of the pathophysiology of MCI. Nevertheless, the different patterns of dynamic brain function between aMCI and naMCI patients have not yet been investigated. Exploring the different dynamic characteristics of brain function between aMCI and naMCI patients may not only enhance the understanding of the different mechanisms between MCI subtypes but also provide more potential targets for their neuromodulation and prevent them from developing dementia.
Therefore, a sliding window analysis was performed in the present study to compute dALFF, dReHo, and dFC to characterize the temporal variability in the spontaneous fluctuations of activity and connectivity in MCI subtypes in comparison with a cognitively healthy group. We hypothesized that aMCI and naMCI patients would show both shared and specific patterns of abnormal dynamic Laboratory for Innovation Platform Plan, the Science and Technology Program of Guangzhou, China, the Science and Technology Plan Project of Guangdong Province, Grant/Award Number: 2019B030316001 brain activity and connectivity provided a new perspective for exploring the different pathophysiological mechanisms in MCI subtypes.

K E Y W O R D S
Alzheimer's disease, dynamic networks, functional connectivity, mild cognitive impairment, MRI, neuroimaging brain activity and connectivity and that the difference in dynamic characteristics would be associated with their different patterns of cognitive impairment.

| Image processing
Resting-state fMRI data preprocessing was carried out using the Data Processing Assistant for Resting-State 5.0 (DPARSF 5.0). The first ten volumes were removed to preserve steady-state data only.
The remaining images were corrected for timing differences and head motion. Subjects who had images with more than 2 mm of translational movement or more than 2 degrees of rotational movement were excluded from further analysis. The individual structural image (T1-weighted images) was coregistered to the mean functional image after motion correction. The transformed structural images were segmented into gray matter, white matter, and cerebrospinal fluid. Nuisance signals, such as six head motion parameters, global signal, CSF signal, and WM signal were regressed out from each time series. Following this, the motion-corrected functional images were spatially normalized into the Montreal Neurological Institute space and resampled to 3 × 3 × 3 mm 3 using the normalization parameters estimated during unified segmentation. To reduce the effect of low-frequency drifts and high-frequency noise, a bandpass filter (0.01 Hz < f < 0.1 Hz) was applied for the analysis of dFC and dReho.

ReHo, and dynamic FC
The temporal variability in the spontaneous fluctuations of activity was assessed by dynamic ALFF (dALFF) and dynamic ReHo (dReHo). The Hamming window was used to slide the whole-brain BOLD signals. A sliding window size of 100 TR and a window step of 1 TR were selected to evaluate the whole-brain dALFF variability. By using the 100-TR sliding window analyses, the 230 time points were segmented into 131 windows for each participant. In each window length, for a given voxel, the time series was first converted to the frequency domain using a fast Fourier transform.
The square root of the power spectrum was computed and then averaged across a predefined frequency interval (0.01-0.1 Hz).
The average square root was considered to be the ALFF at the given voxel. 43 Then, the standard deviation of the ALFF values (dALFF variability) across all 131 windows was calculated to quantitatively depict the temporal dynamic characteristics of ALFF.
Subsequently, we applied z standardization within the gray matter mask, and the dALFF variability maps were smoothed with a 6 mm full width at half maximum (FWHM) Gaussian kernel. ReHo reflects the degree of local regional neural activity coherence.
Briefly, it was calculated as Kendall's coefficient of concordance (or Kendall's W) of the time course of a given voxel with those of its nearest neighbors (26 voxels). A sliding window size of 50 TR and a window step of 1 TR were applied to calculate the dReho variability of each voxel (181 windows), 26,27 and the other processing was the same as for the dALFF.
The temporal variability in the spontaneous fluctuations of connectivity was assessed by the dynamic FC (dFC). A previous study suggested that functional convergence of the caudal-rostral hippocampus may be a sensitive biomarker of disease severity along the AD spectrum. 44 Therefore, the present study selected the bilateral caudal hippocampus and bilateral rostral hippocampus as the seeds for calculating the dFC variability according to the Brainnetome Atlas (Brainnetome Atlas Viewer, vision 1.0, http://atlas.brain netome. org/). 45 For each sliding window, correlation maps were produced by computing the temporal correlation coefficient between the truncated time series of the seeds and all the other voxels. Consequently, 181 sliding window correlation maps were obtained for each participant. The obtained correlation maps were then converted to z value maps using Fisher's r-to-z transformation to improve the normality of the correlation distribution. Subsequently, we calculated the standard deviation of the z value at each voxel to assess dFC variability.
Finally, we applied z standardization within the gray matter mask, and the dFC variability maps were smoothed with a 6 mm FWHM Gaussian kernel. 26,27

| Statistical analyses
Demographic and clinical data were analyzed by using SPSS, version 25.0 (SPSS). The differences between the aMCI group, naMCI group, and HC group were analyzed using Analysis of Covariance (ancova), and control variables included age, sex, and years of education. The least significant difference (LSD) test was used for post hoc analyses. A chi-squared test was used to compare the sex differences among the three groups. To examine the differences in the variability of dALFF and dReHo among the three groups, ancova was carried out to compare the group differences based on the standard deviation in the z value at each voxel within the gray matter mask, with age, sex, years of education and mean frame-wise displacement (FD) values as control variables. The multiple comparisons of dALFF and dReHo were corrected by using Gaussian random field (GRF) theory (voxel p < 0.001, cluster p < 0.05, cluster size >10).
The one-sample t test was performed to investigate the withingroup dFC variability distribution of each hippocampal seed in patients in the aMCI group, naMCI group, and HC group. The significance level was set at a p < 0.05 (uncorrected). To further examine the difference in dFC variability patterns among the three groups, ancova was performed on the standard deviation in the z value at each voxel within the union mask of one-sample t test results of the three groups. Age, sex, years of education, and mean FD values were included as nuisance covariates in the comparisons. The multiple comparisons were corrected by using Gaussian random field (GRF) theory (voxel p < 0.001, cluster p < 0.05, cluster size >10).

| Demographic and cognitive information
There was one subject with naMCI, 3 subjects with aMCI, and 1 HC who were excluded because they had images with more than 2 mm of translational movement or more than 2 degrees of rotational movement. The demographic and cognitive information of the HC, naMCI, and aMCI groups is listed in Table 1. No significant difference was found in age and sex distribution among the three groups (p > 0.05), and the aMCI group exhibited fewer years of education than the HC and naMCI groups (p < 0.05). For the comparison of cognitive scores, significant differences were found in all assessments among the three groups (p < 0.05). In the post hoc comparisons, both the naMCI and aMCI groups exhibited worse performance in all cognitive scores, and the aMCI group exhibited lower scores in three AVLT aspects than the naMCI group (p < 0.05). No significant difference was found in the other assessments between the aMCI and naMCI groups (p > 0.05).

| Comparison of dALFF variability
Among the HC, naMCI, and aMCI groups, there were significant differences in dALFF variability in the left superior cerebellum, right putamen, right superior temporal gyrus, left Rolandic operculum and right middle cingulum (Table 2, Figure 1A). In the post hoc comparisons, (1) both the naMCI and aMCI groups exhibited higher dALFF variability in the left superior cerebellum and right superior temporal gyrus; (2) the aMCI group exhibited higher dALFF variability in the right putamen, left Rolandic operculum, and right middle cingulum than the HC and naMCI groups; (3) compared with the HC group, the naMCI group exhibited lower dALFF variability, and the aMCI group exhibited higher dALFF variability in the right putamen (p < 0.05) (Figure 2A).

| Comparison of dReHo variability
Among the HC, naMCI, and aMCI groups, there were significant differences in dReHo variability in the left inferior frontal gyrus, left precuneus, and right superior parietal lobule (Table 2, Figure 1B).
In the post hoc comparisons, (1) the naMCI group exhibited higher dReHo variability in the left inferior frontal gyrus than the naMCI and HC groups; (2) both the naMCI and aMCI groups exhibited lower dReHo variability in the left precuneus than the HC group; and (3) the aMCI group exhibited lower dReHo variability in the right superior parietal lobule than the naMCI and HC groups (p < 0.05) ( Figure 2B).

| Comparison of dFC variability
The results of a one-sample t test of hippocampal dFC in HC, naMCI and aMCI were shown in Figure S1 (2) the naMCI group exhibited higher dFC variability in the left rostral hippocampus and left middle frontal gyrus than the aMCI and HC groups and lower dFC variability in the right caudal hippocampus and left inferior cerebellum than the HC group (p < 0.05) ( Figure 2C).

| DISCUSS ION
The present study is the first to compare the temporal variability in intrinsic brain function between aMCI, naMCI, and HC groups The present study suggested that aMCI and naMCI subjects exhibited shared and specific dynamics of brain activity and connectivity. Specifically, their shared pattern included higher dALFF variability in the right superior temporal gyrus and left superior cerebellum, and decreased dReHo in the left precuneus, which were more related to abnormal activity but not connectivity. Abnormalities in the temporal gyrus, precuneus, and cerebellum in AD spectrum diseases have been repeatedly reported in previous studies. 16,46 The superior temporal gyrus plays a necessary role in spoken word recognition because it is related to auditory association and multisensory integration, 48  F I G U R E 2 Post hoc comparison of dALFF variability, dReHo variability, and dFC variability among HC, aMCI, and naMCI groups. *p < 0.05, **p < 0.01, ***p < 0.001. dALFF variability, dynamic amplitude of low-frequency fluctuation dReHo variability, dynamic regional homogeneity, dFC variability, dynamic functional connectivity. aMCI, amnestic mild cognitive impairment; naMCI, nonamnestic mild cognitive impairment; HC, healthy controls.
The present study suggested that the inflexible connectivity with the nearest neighboring regions of the superior parietal lobule may be a specific characteristic of aMCI, and it contributes to the difference in memory heterogeneity between aMCI and naMCI patients. These results are a powerful supplement to the information on the relationship between the superior parietal lobule and AD spectrum diseases, and indicate that the superior parietal lobule may be a potential target for neuromodulation in aMCI patients.
Apart from the dReHo of the superior parietal lobule, the dFC of the hippocampus was also a partial mediator of the memory heterogeneity between aMCI and naMCI patients, and their absence of abnormal dALFF and dReHo suggested that the dynamic brain dysfunction of the hippocampus was more related to connection but not activity.
The hippocampus plays an important role in the cognitive map, and it is widely connected to other brain regions and involved in various complex memory processing tasks. 53 Moreover, the hippocampus is affected early by AD pathology, and its extent of abnormalities reflects the progression of AD development. 54 Abnormal hippocampal FC in AD spectrum diseases has been reported in many studies, 15  prefrontal cortex. 57 Previous studies provided the range of the appropriate window length as 10-75 TR, step = 1 TR, and a moderate sliding window F I G U R E 4 Mediated effect of dynamic brain function on the different cognitive scores between aMCI and naMCI groups. (A) The dFC variability between right caudal hippocampus and left inferior cerebellum partially mediated to the difference in delay recall memory score between aMCI and naMCI groups. (B) The dReHo variability right superior parietal lobule partially mediated the difference in delay recall memory score between aMCI and naMCI groups. (C) The dReHo variability right superior parietal lobule partially mediated the difference in short-term memory score between aMCI and naMCI groups. (D) The dReHo variability right superior parietal lobule partially mediated the difference in recognition score between aMCI and naMCI groups.
length may maximize the statistical power, because it may be an optimal balance between capturing rapidly shifting dynamic relationships (with shorter windows) and achieving reliable estimates of the correlations between regions (with longer windows). 2,64 Additionally, a sliding window size of 50 TR and a window step of 1 TR has been repeatedly used in previous studies, and they were able to capture the dynamics. 65 There are limitations in the present study. First, the present conclusions were based on cross-sectional analyses, and longitudinal studies are needed to further explore the associations between dynamic brain function and dementia progression in aMCI and naMCI individuals. Additionally, combining the use of CSF biomarkers and PET-CT could clarify the relationship between temporal variability in intrinsic brain function and neurodegeneration. Second, 50-TR window lengths were selected to measure dFC variability and dReHo variability, and 100-TR window lengths were selected for dALFF variability analyses in the present study, but it remains unclear whether they are the best choice; this should be further explored by future studies with other window lengths. Third, the relatively imbalanced sample of aMCI and naMCI individuals may have influenced the statistical power, and the present results should be interpreted with caution. Fourth, the present study used the caudal and rostral hippocampus as the seeds for dFC variability analyses, and future studies including more seeds could provide a better picture of the pattern of dynamic connectivity in MCI individuals. Finally, the present subjects were all elderly people, and some of them had general health problems (such as hypertension, diabetes, and coronary heart disease) and were taking various relevant drugs, which may have exhibited potential confounding effects on brain function.
In summary, aMCI and naMCI patients exhibited shared and specific patterns of abnormal dynamic brain activity and connectivity.
The connectivity of the hippocampus-cerebellum and hippocampusfrontal lobe and the activity of the superior parietal lobule contributed to the memory heterogeneity of MCI subtypes. By describing dynamic changes in intrinsic brain activity and connectivity, the present study offers a novel approach for differentiating the pathophysiological mechanisms of MCI subtypes and provides potential targets for individualized intervention.

AUTH O R CO NTR I B UTI O N S
BC acquired the data, analyzed and interpreted the data, and drafted the manuscript. XZ and LH designed and conceptualized the study,

CO N FLI C T O F I NTE R E S T
The authors have no actual or potential conflicts of interest to declare.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used to support the findings of this study are available from the corresponding author upon request.