Dysfunctional interactions between the default mode network and the dorsal attention network in subtypes of amnestic mild cognitive impairment

Altered anticorrelation strengths between the DMN and DAN

The anticorrelation between the DMN and DAN during rest is a core feature of the human brain's intrinsic architecture [16, 17]. This pattern remains stable when individuals perform cognitive tasks that demand various cognitive domains, including memory, cognitive control and visuospatial planning, and subjects with a stronger anticorrelation achieve better performance on cognitive measures [25, 27]. Our results of inter-network functional connectivity suggested that aMCI-s and aMCI-m groups showed different magnitudes of enhanced anticorrelation between the DMN and DAN relative to other groups. Based on the aforementioned information, these results were indicative of a neural mechanism of functional compensation that helps to improve cognitive deficits and maintain behavior at relatively normal levels.

Specifically, the aMCI-s group revealed increased anticorrelation strengths between the left PCC seed and the left IPS, as well as between the left FEF seed and the right ventral PCC, compared to the AD group. The aMCI-m group showed enhanced anticorrelation strengths between the left IPS seed and the right dorsal PCC compared to the AD group. These areas are central hubs of the DMN and the DAN. PCC sends dense projections to the medial temporal lobe, hippocampal formation, and parahippocampal cortex, which are important areas for memory [41]. Furthermore, it is particularly vulnerable to early deposition of amyloid beta-protein, one of the hallmark pathologies of AD [42]. IPS and FEF contribute to mediation of goal-directed process and orienting to external stimuli [20]. Prior studies demonstrated that attention deficits co-occurred with memory deficits in early stage AD [43, 44]. Consistent with the results of functional connectivity within the DMN and DAN, aMCI-s and aMCI-m groups used more functional resources of the DMN and DAN to compensate for cognitive function to perform tasks relatively normally. Our correlation analysis further demonstrated that better performance on various cognitive measures was associated with stronger negative functional connectivity between the DMN and DAN among the four groups. It is worth noting that both the dorsal and ventral PCC are important areas for memory [23, 43]. The results from a previous study showed that the ventral PCC had a closer relationship with ‘MCI/AD’ and can be seen as a more important central hub of the DMN [45]. This result may explain why the aMCI-s group showed more robust changes in functional connectivity in the ventral PCC, which could be utilized as a potential biomarker to detect early-stage AD.

Our present results also revealed that the aMCI-s group had a stronger anticorrelation between the left PCC seed and the left inferior precentral sulcus than NC and aMCI-m groups. Based on previous studies, the inferior frontal eye field and the inferior frontal junction area have been suggested to be located at the inferior precentral sulcus and are thought play critical roles in visual-biased attention, which may subsequently influence short-term memory [46, 47]. Considering the evidence mentioned above, enhanced anticorrelation strengths between the left PCC and the left inferior precentral sulcus in aMCI-s may rescue impaired memory. This result would also be useful to distinguish different subtypes of aMCI. Moreover, the aMCI-m group had significantly increased negative functional connectivity between the left mPFC seed and the superior occipital gyrus compared to NC. Our findings suggested that aMCI-m and AD groups showed hyperconnectivity in areas outside the typical networks. More specifically, we found that the aMCI-m group showed greater functional connectivity to the occipital gyrus within the DAN. Therefore, dysfunctional interactions between the DMN and DAN are not isolated and have a close relationship with functional changes within the DMN and DAN during the progression from MCI to AD. Hyperconnectivity in areas outside the typical DMN and DAN could subsequently disturb functional interactions between the DMN and DAN.

Overall, the aMCI-s group had the most significant increase in the DMN-DAN functional anticorrelation, and then, the increase in the anticorrelation for the aMCI-m group was lower in magnitude; comparatively, the anticorrelation strengths of the AD group were near-zero values. These findings suggested a gradient of DMN-DAN anticorrelation attenuation during the progression of the disease that was consistent with the behavioral assessments, revealing a significant trend of gradual impairment across different groups. A previous study also reported that the aMCI group showed an enhanced negative correlation between the salience network (SAL) and the DMN, which may compensate for and help sustain the anticorrelation between the DMN and DAN in early-stage AD [38]. Numerous studies have shown an attenuated anticorrelation between the DMN and DAN during the normal aging process both during task performance and at rest [31, 38, 39]. Reduced anticorrelation between the DMN and DAN in older adulthood is thought to be related to the poor modulation of attentional processes in response to shifting cognitive demands and inefficiency in processing cognitive resources [48–50]. In this way, the progression from MCI to AD may accelerate this process. In the early stage of pathological processes, different subtypes of aMCI patients still have the ability to allocate extra cognitive resources to compensate for different levels of cognitive deficits. With the progression of the disease, dysfunctional interactions between the DMN and DAN reach a significant level in AD patients, overcoming possible compensatory mechanisms, thus leading to gradual cognitive decline.

Relationship between the DMN-DAN anticorrelation and functional connectivity within the DMN and DAN

Consistent with prior studies, our results replicated previous reports and showed reduced functional connectivity within the DMN and DAN in AD and MCI patients during the resting state [11, 35, 36]. Compared with NC, aMCI-m and AD groups showed reduced functional connectivity strength in the right precuneus within the DMN. The precuneus and the PCC comprise the core of the DMN and display the highest resting metabolic rates [51, 52]. These regions are involved in multiple cognitive processes, including, but not limited to, episodic memory, visuospatial processing, self-referential processing and consciousness [53]. We also observed that the AD group showed reduced functional connectivity in the right retrosplenial cortex within the DMN relative to NC. This area forms part of the PCC and has emerged as a crucial transit region between the hippocampus and cingulate cortex, having been implicated in a range of cognitive functions including episodic memory, spatial navigation and imagination [54–56]. Importantly, previous studies have indicated that atrophy of the retrosplenial cortex is present in the earliest clinical stage of AD and is a site often affected in AD [55, 56].

The aMCI-m and aMCI-s groups revealed decreased functional connectivity in the left angular gyrus compared to NC. A previous study demonstrated that the left angular gyrus is critical for episodic memory [57]. Our results showed that the aMCI-s group only had decreased functional connectivity in the left angular gyrus compared with NC, which is consistent with the definition of isolated memory impairment. Furthermore, we also found that the AD group showed reduced functional connectivity in the right retrosplenial cortex and the right PCC within the DMN compared to the aMCI-m group and had decreased functional connectivity in the right postcentral gyrus within the DAN compared to the aMCI-s group. The aMCI-m group showed decreased functional connectivity in the right supramarginal gyrus extending to the right postcentral gyrus compared to the aMCI-s group. Considering the aforementioned findings, our results supported a gradient of functional deficits. Observed hypoconnectivity in the aMCI-s, aMCI-m and AD groups relative to NC may reflect different levels of impairment among a variety of cognitive domains during the progression from MCI to AD. Preserved functional connectivity within the DMN and DAN may represent internal potential to maintain a reciprocal pattern between the DMN and DAN.

Hyperconnectivity was also observed within the DMN. Our results showed that aMCI-s and aMCI-m groups had increased functional connectivity in the left precuneus relative to NC. These findings can be explained by the compensation theory, which suggests that individuals with aMCI-s and aMCI-m may still have the ability to invoke the resources of the DMN to compensate for declining functional integrity. We also found that aMCI-m and AD groups, compared to NC, showed expanded connectivity to the middle occipital gyrus, which is outside the typical default network. In accordance with a previous study, these findings supported the concept of network expansion as a neural mechanism of functional compensation [58]. Additionally, aMCI-m and AD groups showed stronger and more extended connectivity to the lingual gyrus and the right visual cortex, respectively, than NC. These results demonstrated that individuals with aMCI-m and AD may need to recruit more cognitive resources from outside the typical default network to compensate for the declining integrity of the DMN. The altered functional connectivity within the DMN and DAN, including the hypoconnectivity in restricted regions in both networks and hyperconnectivity in areas outside the typical networks, may subsequently disturb functional interactions between the DMN and DAN.

Some limitations of this study need to be further discussed. First, age and educational level were not well matched across the four groups. However, the average age of all participants approached 65 years old. There were no significant differences among the aMCI-s, aMCI-m and AD groups and only a slight difference in age between the aMCI-s and NC groups. Therefore, age likely had little influence on our results. We also included educational level and age as covariates to avoid possible confounding effects on any observed differences in the functional connectivity of the brain among groups in our analysis. Second, although the sample size in the current study was relatively small, our results replicated the results from previous reports, and strict correlation maps between the DMN and DAN were employed to examine differences between aMCI, NC and AD groups. Future studies should increase the sample size to reveal more stable and objective differences between groups. Finally, although the present study revealed a gradient of deficits increasing from the NC group to the AD group, more cross-validation results and longitudinal studies are needed to better understand the progression of AD pathology.

Participants

Our study sample included 20 AD participants (11 females), 22 aMCI-m participants (10 females), 29 aMCI-s participants (14 females) and 23 normal controls (NCs) (13 females). All AD and aMCI participants were recruited from the Department of Neurology at Xuanwu Hospital, Capital Medical University. The sex-matched NCs were recruited from the local community by advertisements. All participants received financial compensation for their participation. The experimental procedure was approved by the Research Ethics Committee of Xuanwu Hospital. Written informed consent was obtained from all participants or their relatives after the study had been fully explained.

All AD and aMCI participants underwent a detailed physical and neurological examination. They were screened by two experienced neurologists according to the standard criteria [59, 60]. The AD participants were assessed based on the Diagnostic and Statistical Manual of Mental Disorders-V (DSM-V) criteria for Alzheimer's Dementia, and the National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria for AD [61–63]. Participants with aMCI were diagnosed and classified according to Petersen's clinical diagnostic criteria [59] and the National Institute on Aging-Alzheimer's Association criteria for MCI due to AD [60]. The inclusion criteria for aMCI were as follows: 1) memory complaint, preferably corroborated by an informant; 2) memory impairment for age; 3) essentially normal performance on general cognition and preserved activities of daily living; 4) Clinical Dementia Rating (CDR) score of 0.5 [64]; and 5) the absence of dementia. Exclusion criteria for all participants included the following: 1) cognitive impairment caused by head trauma or cranial surgery; 2) current or lifetime history of neurological or psychiatric disorders that could cause cognitive impairment, such as stroke, depression, or epilepsy; 3) neurological deficiencies, such as visual or hearing loss; and 4) contraindications to MRI. Demographic and behavioral data were analyzed with SPSS 20 using one-way ANOVA and chi-square tests. All demographic and neuropsychological assessment results are shown in Table 1.

Neuropsychological assessment

All participants underwent an extensive battery of neuropsychological assessments that included the CDR [64], the Mini-Mental State Examination (MMSE) [65] and the Montreal cognitive assessment (MoCA) [66] as screening measures. Four cognitive domains were assessed: 1) memory function was measured with the Auditory Verbal Learning Test (short delay free recall) [67]; 2) naming skill was evaluated with the Boston Naming Test (BNT) [68]; 3) executive function was measured with the Trail Making Test (TMT) [69]; and 4) visuospatial ability was assessed with the clock drawing test (CDT; 3-point) [70]. The neuropsychological data were collected on the day before the MRI session.

MRI data acquisition

Imaging data were acquired using a SIEMENS 3-Tesla scanner (Siemens Medical Solutions, Erlangen, Germany). Participants were instructed to keep their head still and their eyes closed and to refrain from thinking about anything in particular. Foam padding and headphones were used to limit head movement and minimize scanner noise. The high-resolution T1-weighted anatomical images were acquired using a multiecho magnetization prepared rapid gradient echo (MPRAGE) sequence. The acquisition parameters were as follows: repetition time = 1900 ms, echo time = 2.2 ms, inversion time = 900 ms, flip angle = 9°, acquisition matrix = 224 × 256 × 176, voxel size = 1 × 1 × 1 mm³. BOLD functional imaging was obtained in an axial orientation using a T2*-weighted, multislice gradient echo planar imaging (EPI) sequence with the following parameters: repetition time = 2000 ms, echo time = 40 ms, flip angle = 90°, acquisition matrix = 64 × 64, field of view = 256 mm × 256 mm, slice thickness = 4 mm, gap = 1 mm, 239 volumes, 28 slices, voxel size = 3.75 × 3.75 × 4 mm³.

Image preprocessing

Preprocessing and analysis of rs-fMRI images were conducted with Statistical Parametric Mapping (SPM 8, University College London, London, UK; http://www.fil.ion.ucl.ac.uk/spm/software/spm8) and the CONN-fMRI functional connectivity toolbox v18a (http://www.nitrc.org/projects/conn [71]). Prior to preprocessing, the first 10 volumes of the functional images were discarded to allow for signal stabilization and the participants’ adaptation to the noisy environment. The remaining volumes were slice-time corrected (to the middle slice) and realigned to the first volume to reduce the confounding effects of head motion (using a 6 parameter rigid body transformation). Eight participants whose head motion exceeded 2 mm in translation or 2° in rotation were excluded from further analysis (4 AD participants, 1 aMCI-m participant, 2 aMCI-s participants and 1 healthy participant were excluded for this reason). We also calculated motion parameters (estimated by a realignment algorithm) for each group, and the difference between groups was statistically analyzed. For spatial normalization to the MNI space, the individual T1-weighted structural image was coregistered to the mean functional image. The T1 image was bias-corrected and segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using template (ICBM) tissue probability maps. Parameters obtained from this step were subsequently applied to the functional data (resampled to 3 mm isotropic voxels) during normalization to the MNI space. Then, the data were spatially smoothed by a Gaussian filter with a full width half maximum (FWHM) of 6 mm.

Using the CONN toolbox, spurious sources of noise, such as heart rate and respiration signals, were estimated and regressed out using the anatomical component base noise reduction strategy (aCompCor) [72]. The aCompCor method efficiently removes principal components from WM and CSF regions. Specifically, the anatomical image for each participant was segmented into white matter, gray matter, and CSF. The white matter and CSF masks were eroded by one voxel, which resulted in substantially smaller masks than in the original segmentations. Then, the eroded white matter and CSF masks were used as noise regions of interest to extract signals from the unsmoothed functional volumes to avoid the additional risk of contaminating white matter and CSF signals with gray matter signals. Global signal regression was not used because it can artificially introduce anticorrelations into the BOLD signal [73]. The motion parameters (3 translations and 3 rotations), along with their first-order and second-order temporal derivatives, were also regressed out. The BOLD time series were simultaneously filtered with the recommended bandpass filter (0.009–0.08 Hz) during this step [74], followed by removal of the linear trend.

Functional connectivity analysis

After data preprocessing, we first performed group spatial independent component analysis (ICA) to decompose the data into functional networks using the CONN toolbox and then to identify networks of interest [75]. Data reduction was performed by principal component analysis (PCA) to decrease computational complexity. Then, the resulting volumes were temporally concatenated, and PCA was performed again. After data reduction, we performed ICA decomposition using the Infomax algorithm to identify 20 independent components (ICs), which were suggested to provide a reliable identification of the DMN and DAN [76, 77], estimated by minimum description length criteria (MDL) [78]. ICs and time courses were back-reconstructed for each participant. After visually inspecting all ICs, we also calculated the spatial correlation values between ICs and the templates from the CONN network’s default networks file to identify the DMN and DAN networks. The functional connectivity of the DMN and DAN networks was converted to Z-values using Fisher’s r-to-z transform before further correlation analyses were performed. Next, one-sample t-tests were separately performed on the correlation maps of the DMN and DAN to obtain statistical functional connectivity maps of each network. The significance level was set at a voxel level of P < 0.05, corrected for multiple comparisons using familywise error (FWE). The results were saved as the masks for subsequent inter-network analysis.

Seed-based resting state connectivity analysis was employed to explore DMN-DAN interactions. Two regions of interest (ROIs) in the DMN and 2 ROIs in the DAN that had been created by a previous study that parcellated the brain based on 1000 participants were defined for analysis [79]. The mean time series of the DMN ROIs were extracted from the posterior cingulate cortex (PCC; MNI coordinates: -7, -52, 26) and the medial prefrontal cortex (mPFC; MNI coordinates: -7, 49, 18) using spherical seeds with a 6 mm radius. The mean time series of the DAN ROIs were extracted from the intraparietal sulcus (IPS; MNI coordinates: -34, -38, 44) and the frontal eye field (FEF; MNI coordinates: -22, -8, 54) using spherical seeds with a 6 mm radius. Then, seed-based resting state connectivity maps were generated for each participant by calculating the correlation coefficients between the mean signal of each ROI and the time series at each voxel of the whole brain. Individual correlation maps were converted to Z-value maps using Fisher’s r-to-z transformation to improve the normality of the correlation coefficients. A one-sample t-test was performed separately for HC, AD, aMCI-m and aMCI-s participants on the correlation maps to identify the brain areas that were negatively correlated with each seed with a voxel threshold of P < 0.05, corrected for multiple comparisons using FWE. The resulting statistical maps for each group were subsequently combined within each ROI to generate specific masks for inter-network analysis. A voxelwise one-way ANOVA was performed to compare the intra-network functional connectivity difference between the DMN and DAN networks of the four groups. The inter-network functional connectivity represented the ROIs of the DMN or DAN connectivity maps, masked with correlation maps of the DAN or DMN, and the specific mask that negatively correlated with the corresponding DMN node or DAN node, as defined above.

Statistical analyses

Statistical analysis was conducted for group differences using one-way analysis of variance (ANOVA) with group as a factor (4 groups: HC, aMCI-s, aMCI-m and AD) after regressing out the influence of age, education years and global correlation (GCOR). The main effect map of the group was first generated to create a sample-specific mask with a voxel threshold of the FWE correction (P < 0.05). Post hoc comparisons were used to identify meaningful intra-network and inter-network functional connectivity differences among the four groups within the corresponding sample-specific mask. The statistical significance was AlphaSim corrected at a voxel level of P < 0.05, which was obtained by clusters with a minimum size of 10 voxels and an individual voxel height threshold of P < .005, according to a Monte Carlo simulation (1000 simulations, FWHM = 4 mm) within the mask using REST software (http://restfmri.net/forum/REST_V1.8 [80]). To investigate the relationship between network interaction and cognitive ability, we also explored Pearson’s correlations between the behavioral data and the inter-network functional connectivity with significant between-group differences among aMCI-s, aMCI-m, AD and NC groups. A value of P < 0.05 was considered statistically significant.