Differentially disrupted functional connectivity of the subregions of the amygdala in Alzheimer’s disease

2.1 Participants

Thirty-four patients with AD and forty-one normal controls (NCs) participated in the study after giving written informed consent. This study was approved by the Medical Research Ethics Committee of Xuanwu Hospital. The data from five subjects (two AD patients and three NCs) were excluded because of a failure in the image processing (see Image preprocessing). Clinical and demographic information for the remaining 70 participants (32 AD patients and 38 NCs) are shown in Table 1. Previously, a subset of this dataset (16 AD patients and 22 NCs) was used to study the changes in regional brain activity in AD [21]. The AD patients were recruited from individuals who had consulted a memory clinic at Xuanwu Hospital with memory complaints. All AD subjects underwent a complete physical and neurological examination, standard laboratory tests and neuropsychological assessments, which included mini-mental state examination (MMSE), World Health Organization–University of California–Los Angeles Auditory Verbal Learning Test (WHO-UCLA-AVLT), the Extended Scale for Dementia (ESD), Montreal Cognitive Assessment (MoCA), Clock Drawing Task (CDT), Activity of Daily Living Scale (ADL), Functional Activities Questionary (FAQ), Hamilton Depression Scale (HAMD) and Hachinski Ischemic Score (HIS). The diagnosis of AD fulfilled the new research criteria for possible or probable AD [22, 23]. The patients were assessed with the Clinical Dementia Rating (CDR) score [24] as being in the early stages of AD (18 patients with CDR = 1 and 14 patients with CDR = 0.5). The NCs were recruited from the local community by advertisements. The inclusion criteria for NCs were as follows: (1) no neurological or psychiatric disorders, such as stroke, depression or epilepsy; (2) no neurological deficiencies, such as visual or hearing loss; (3) no abnormal findings, such as infarction or focal lesions in conventional brain magnetic resonance imaging (MRI); (4) no cognitive complaints; (5) MMSE score of 28 or higher and (6) a CDR score of 0.

2.2 Data acquisition

MRI data acquisition was performed on a SIEMENS Trio 3-Tesla scanner (Siemens; Erlangen, Germany). Foam padding and headphones were used to limit head motion and reduce scanner noise. Functional images were collected axially using an echo-planar imaging (EPI) sequence with the following parameters: Repetition time (TR)/echo time (TE)/flip angle (FA) = 2000 ms/40 ms/90°, field of view = 24×24 cm², resolution = 64×64 matrix, slices = 28, thickness = 4 mm, gap = 1 mm, voxel size = 3.75×3.75×4 mm³, and bandwidth = 2232 Hz/pixel. During the 478 s scan, subjects were instructed to hold still, keep their eyes closed and not to think of anything in particular. A simple questionnaire after the scan confirmed that none of subjects had fallen asleep. For registration purposes, high-resolution anatomical images were acquired using a 3D magnetization-prepared rapid gradient echo (MPRAGE) T1-weighted sequence with the following parameters: TR/TE/inversion time (TI)/FA = 1900 ms/2.2 ms/900 ms/9°, resolution = 256×256 matrix, slices = 176, thickness = 1 mm, voxel size = 1×1×1 mm³.

2.3 Data preprocessing

Data preprocessing were performed using the Statistical Parametric Mapping (SPM, www.fil.ion.ucl.ac.uk/spm) and Data Processing Assistant for Resting-State fMRI (DPARSF, www.restfmri.net/forum/DPARSF) [25] toolkits. Briefly, preprocessing included removal of the first 10 volumes, slice timing correction and head motion correction. To spatially normalize the fMRI data, the T1-weighted images were used to register the functional data to their corresponding anatomical image, and the resulting aligned T1 dataset was transformed into Montreal Neurological Institute (MNI) space [26]. To improve the coregistration of the fMRI data, a custom T1 template was built by averaging the normalized anatomical images across all subjects. Finally, the normalized functional images were created by applying the transformation of the T1 images to the customized T1 template. Notably, such a custom template-based registration procedure could reduce the inaccuracy of the spatial normalization of functional volumes due to GM atrophy in AD patients and healthy controls. Functional images were resampled to 3 mm isotropic voxels and spatially smoothed with a 4 mm full width-half maximum (FWHM) Gaussian kernel. Linear detrending and temporal band-pass filtering (0.01 –0.1 Hz) were applied to reduce the effect of low-frequency drifts and high-frequency physiological noise. Finally, several nuisance variables, including six head motion parameters, global mean signal, cerebrospinal fluid signal and white matter signal were removed by multiple linear regression analysis. During image preprocessing, no subject was excluded due to excessive head motion (defined by a translation greater than 3 mm or rotation greater than 3°), but five subjects (two AD and three NCs) were excluded due to failures in image normalization.

2.4 Definition of amygdala subregions

We defined three amygdala subregions in each hemisphere using cytoarchitectonically defined probabilistic maps from the JuBrain Cytoarchitectonic Atlas [27] as implemented in the SPM Anatomy Toolbox [28]: (1) LB includes lateral nucleus, the basolateral nucleus, medial nucleus and deputy layer of the basement; (2) CM is composed of the central and medial nuclei; (3) SF includes the area before almonds, transition zone of the amygdala piriform cortex, the amygdala-hippocampus area, the ventral and posterior cortical nucleus (Fig. 1).

2.5 Voxel based morphometry (VBM) analysis

A VBM analysis of structural images was performed to control for the possible confounding effect of atrophy on the functional results. Gray matter intensity maps were obtained by the unified segmentation algorithm as described in the Data Preprocessing section. After spatially smoothed with a Gaussian kernel of 10 mm FWHM, a two-sample t-test was performed on the smoothed gray matter intensity maps to examine the regional gray matter atrophy in the AD patients as compared to the NCs. The statistical threshold was set at P <  0.001 and cluster size >  100 mm³, which corresponded to a corrected P <  0.05 (using the AlphaSim program with parameters: FWHM = 10 mm, within the group gray matter mask).

2.6 RSFCs analysis of the amygdala subregions

For each subject, the voxels of each amygdala subregion were extracted and averaged to obtain the seed point reference time series. A correlation map was produced by computing the correlation coefficients between the reference time series and the time series from all the other brain voxels. Correlation coefficients were then converted to z values using Fisher’s r-to-z transform to improve the normality. For each subject, we obtained 6 z-score maps that represented the intrinsic RSFC patterns of the 6 amygdala subregions.

2.7 Statistical analysis

To examine the within-group RSFC patterns of each amygdala subregion for the AD and NC groups, we performed one-sample t tests on individual z-score maps for each amygdala subregion. The statistical significance threshold was set to P <  0.01 with a cluster size of 50 voxels based on Monte Carlo simulations [29] using the REST AlphaSim utility (www.restfmri.net) [30], which corresponded to a corrected P <  0.05.

To assess the between-group differences of the whole-brain RSFCs of each amygdala subregion, we used general linear model (GLM) analysis (dependent variable: RSFCs; independent variable: group) with age, gender, education and volume as covariates. The significance threshold was set to P <  0.05 with a cluster size of 15 voxels, which corresponds to a corrected P <  0.05. The cluster size for each subregion was determined by Monte Carlo simulations, with the restriction that the significant clusters must belong to significant within-group connectivity maps for one or both groups.

To investigate the relationship between RSFC strength and cognitive behaviors, we performed a GLM analysis (dependent variable: RSFC; independent variable: Cognitive behavior scores) within the regions showing group differences, with age, gender, education and volume treated as covariates. The statistical significance level was set at p <  0.05.

3.1 Demographic and neuropsychological tests

Demographic characteristics are shown in Table 1. No significant differences in gender,age and education level were observed between the AD and NC groups (both Ps >  0.01). All the subjects were right handed. However, the AD group exhibited significantly lower MMSE, AVLT, ESD, MoCA, CDT, ADL and FAQ scores than that of the NC group (Ps <  0.0001).

3.2 VBM analysis

Compared with the controls, the AD patients showed a broad area of significant gray matter loss in the temporal lobe, parietal lobe, occipital lobe as well as subcortical regions. The detailed regions includes the uncus, parahippocampal gyrus, insula, superior and middle temporal gyrus, angular gyrus, supramarginal gyrus, superior occipital gyrus and caudate (p <  0.05, Table 2 and Fig. 2)

3.3 Within-group RSFCs of the amygdala subregions

The subregions of the amygdala include the left and right centromedial (CM), the left and right laterobasal (LB), the left and right superficial (SF). Each subregion showed positive connectivity with the several regions. By visual inspection of the within group RSFC of amygdala subregions (Fig. 3), there was similar pattern between the AD and NC group. In NC within group map, the left CM of amygdala showed positive connectivity with the lentiform nucleus, parahippocampus, superior and middle temporal gyrus and thalamus. The right CM of amygdala showed positive connectivity with the lentiform nucleus, parahippocampus, superior temporal gyrus (STG) and insula. The left LB of amygdala showed significant connectivity with the parahippocampus, superior and middle temporal gyrus, middle occipital gyrus, medial prefrontal cortex and fusiform gyrus. The right LB of amygdala showed connection with the parahippocampus, superior and middle temporal gyrus, middle occipital gyrus, medial prefrontal cortex, uncusand thalamus. The left SF of amygdala showed positive connectivity with the parahippocampus, lentiform nucleus, superior and middle temporal gyrus, insula, middle occipital gyrus, precentral and postcentral gyrus. The right SF of amygdala showed positive connectivity with the parahippocampus, lentiform nucleus, superior and middle temporal gyrus, insula, middle occipital gyrus, pre and postcentral gyrus and inferior frontal gyrus.

3.4 Between-group differences in the RSFCs of the amygdala subregions

When comparing amygdala subregional connectivity between AD patients and controls, several regions showed significantly decreased connectivity to the left LB of amygdala in the AD group. These regions were the left uncus, right insula andprecentral gyrus. In addition, the left superior temporal gyrus and right claustrum showed decreased connectivity to the right LB of amygdala in AD group comparing to NC group. There were no other differences of the amygdala subregional connectivity between the AD and NC group. (Details see Table 3 and Figs. 4, 5)

3.5 Relationship between the RSFCs of the amygdala subregions and cognitive behavioral variables

In the AD and NC group, we didn’t find the significant correlations between cognitive behavior scores and the functional connections of the amygdala subregions, including the CM, LB and SF regions.

4.1 Major findings

By applying RSFC analysis to the resting state-fMRI data acquired from the NCs and AD patients, we observed three distinctive patterns of functional connectivity for the subregions of the amygdala: 1) the CM of amygdala mainly exhibited functional connectivity with the lentiform nucleus, parahippocampus and lateral temporal gyrus; 2) the LB of amygdala functionally was connected with the parahippocampus, lateral temporal gyrus, middle occipital gyrus and medial prefrontal cortex; and 3) the SF of amygdala was involved in the parahippocampus, lentiform nucleus, lateral temporal gyrus, insula, middle occipital gyrus, precentral and postcentral gyrus. Importantly, the AD patients showed disrupted pattern of RSFCs in the LB of amygdala, which predicted disconnection to the left uncus, right insula, right precentral gyrus, the left superior temporal gyrus and right claustrum.

4.2 The functional connectivity of amygdala subdivisions

The functional connectivity of individual amygdala subdivisions showed regions of overlap and regions uniquely related to each subdivision. We noticed that almost all the subregions of amygdala were connected with the parahippocampus and lateral temporal gyrus. Numerous previous studies have confirmed the robust pathway between amygdala and temporal lobe regions in monkeys [31, 32]. Positive RSFC between amygdala and medial temporal gyrus has also been revealed in healthy adults [19]. Task fMRI in healthy adults demonstrated that amygdala-medial temporal gyrus connectivity increases during retrieval of emotional memories, suggesting that the integration of amygdala and temporal gyrus played important role in the modulation of memory [33].

Besides the two common regions,the lentiform nucleus was the significant cluster linked with the the CM of amygdala. In the previous fMRI study, the CM nuclei showed significant functional connectivity with the striatum including caudate, putamen, globus pallidus, which were similar in function, connectivity, and chemistry (neurotransmitter and peptide distribution) to the CM of the amygdala [34]. According to the previous study, the region of CM played an important role in motor responding, reward processing, increasing attention and so on [19].

Besides the parahippocampus and lateral temporal gyrus, the LB of amygdala was uniquely connected to the medial prefrontal cortex and middle occipital gyrus, which is similar with the previous study. A number of recent studies have identified robust reciprocal connections between the frontal cortices and the amygdala in rhesus monkeys [35, 36]. The amygdala–frontal interactions has been shown to be the primary neural substrate of emotion processing and regulation [37]. The middle occipital gyrus also showed predominantly functional connection with the LB subregion, which was, consistent with the previous study [38].

The SF of amygdala showed positive association with the parahippocampus, lentiform nucleus, lateral temporal gyrus, insula, middle occipital gyrus, precentral and postcentral gyrus. Comparing with the CM and LB subregions, the SF subregion showed coupling with a dorsal region including the insula, pre and postcentral gyri, which were involved in olfactory function and affective processes [19, 39].These connected regions were characterized by several previous studies [38, 40].

4.3 Disrupted amygdala subregional functional connectivity in AD

When comparing between the AD and NC subjects, the decreased connectivities were found between the left LB of amygdala and the left uncus, right insula, right precentral gyrus. In addition, disconnection was also found between the right LB subregion and the left superior temporal gyrus, as well as the right claustrum. In the previous study, these regions were consistently found to present disconnection in Alzheimer disease. The left uncus is the important part of hippocampus. In the past several years, most studies have focused on the hippocampus and related medial temporal lobe structures regarding their crucial roles in memory processes [41, 42]. Hippocampus is pathologically involved very early in patients with AD and several studies have demonstrated markedly reduced functional connectivity in hippocampus-related networks in early AD [7, 8].

The insula is also a most vulnerable region in the AD patients. Previous structural MRI studies have demonstrated AD-related gray matter loss in the insula [43]. As a crucial hub of the human brain network, the insula is anatomically connected to a wide range of cortical, limbic and paralimbic structures. It is functionally implicated in higher-order cognition, emotion, autonomic and sensory processes [44]. Previous study have proved the insular network disruption in the early stage of AD [45]. The decreased connectivity between the LB of amygdale and the insula was consistent with the previous study, indicating the important role of the insula on the cognitive impairment in AD patients.

The precentral gyrus showed disconnection with the LB subregion in AD. Task-related fMRI studies have reported decreased activation in the sensorimoter area and premotor cortex in AD during performance of motor-related tasks [46, 47]. Based on a large cohort of 510 human subjects, a recent R-fMRI study found that the sensorimoter network was preferentially affected in AD patients [48].

The left superior temporal gyrus, as part of lateral temporal regions, was part of the default mode network (DMN), which is comprised of posterior cingulate cortex /precuneus, medial prefrontal cortex, inferior parietal lobe, medial and lateral temporal cortex and so on [49, 50]. Connectivity of the DMN plays an important role in activity of human cognition and memory processing. Disruption of DMN has been consistently reported in AD patients by various studies [51, 52]. Such network might be affected by the amyloid deposition, atrophy as well as functional disruption in parietotemporal regions [53].

The claustrum, together with the nucleus caudatus, nucleus lentiformis and amygdale constitutes basal ganglia. These subcortical brain structures also play important roles in memory and learning processes. Although the subcortical region has been less investigated, their structural and functional abnormalities have been consistently reported in AD studies [54, 55].

The brains of AD patients exhibit neuropathological changes, which include senile plaques, neurofibrillary tangles, neuronal loss, and glial reaction. As we all know, extracellular amyloid β-peptide (Aβ) oligomer deposition is an important early event in the pathogenesis of AD. Recent studies indicated myelin can be directly damaged by oligomerized Aβ and postulated that myeline breakdown was the primary process leading to AD [56, 57]. It could be that this process contributed to the white matter damage, resulting in the cortical functional disconnection. Therefore, in our study, we speculated that the functional disconnection of the LB of amygdala in AD can be explained as the myelin breakdown among the above specific brain regions, which was caused by the neuropathological changes of AD.

In the current study, we also examined the gray matter atrophy in the AD patients using VBM analysis. We found there were diffuse gray matter losses in several regions in the AD patients. After controlling for gray matter volumes, there was significant functional connectivity disruption in the AD patients comparing to the control group. This revealed that the functional connectivity alteration was independent of the gray matter atrophy in the AD patients.

It is need to mention that we didn’t found the correlation between RSFC strength and cognitive behaviors. It should be paid attention that we performed a GLM analysis and limited in the regions showing group differences, with age, gender, education and volume as covariates, which is relative strict analysis method. In addition, the sample size of the study is not very large. In the future, further study on the relationship between amygdala connectivity and neuropsychological scores is needed.

Some limitations should be addressed: First, in this study, although we kept the scanner room dim during the scanning and instructed the subjects not to think of anything, the subjects may still think something subconsciously that we could not control. Second, we only collected the resting-fMRI data from AD patients. Recent studies have paid greater attention to individuals at a high risk for AD, including amnestic mild cognitive impairment and ApoE4 carriers. Investigating these populations would provide predictive insight into the pathophysiology of AD and valuable biomarkers for early clinical diagnosis and intervention. Third, a longitudinal design will be necessary to elucidate dynamic RSFC patterns of amygdala subregions. In the future, we will trace these subjects using different time points and explore the amygdala connectivity changes and its influence on cognitive function in AD.