Introduction
Amnestic mild cognitive impairment (aMCI) is considered as the middle stage of the Alzheimer's disease (AD) spectrum (Kong et al. Reference Kong, Chen, Tang, Wu, Driesen, Womer, Fan, Ren, Jiang, Cao, Blumberg, Xu and Wang2013) and, when present, significantly increases the subsequent risk of developing AD (Gauthier et al. Reference Gauthier, Reisberg, Zaudig, Petersen, Ritchie, Broich, Belleville, Brodaty, Bennett, Chertkow, Cummings, de Leon, Feldman, Ganguli, Hampel, Scheltens, Tierney, Whitehouse and Winblad2006). Structural imaging studies using volume, shape or cortical thickness measures have demonstrated clear differences between normal subjects and those with mild cognitive impairment (MCI) (for reviews, see Smith, Reference Smith2010, Reference Smith2012). Prominent gray matter (GM) volume reductions in aMCI patients may occur in the medial temporal lobes and thalamus (Bell-McGinty et al. Reference Bell-McGinty, Lopez, Meltzer, Scanlon, Whyte, Dekosky and Becker2005; Whitwell et al. Reference Whitwell, Petersen, Negash, Weigand, Kantarci, Ivnik, Knopman, Boeve, Smith and Jack2007a , Reference Whitwell, Przybelski, Weigand, Knopman, Boeve, Petersen and Jack b ; Karas et al. Reference Karas, Sluimer, Goekoop, van der Flier, Rombouts, Vrenken, Scheltens, Fox and Barkhof2008; Balthazar et al. Reference Balthazar, Yasuda, Pereira, Pedro, Damasceno and Cendes2009; Morbelli et al. Reference Morbelli, Piccardo, Villavecchia, Dessi, Brugnolo, Piccini, Caroli, Frisoni, Rodriguez and Nobili2010; Venneri et al. Reference Venneri, Gorgoglione, Toraci, Nocetti, Panzetti and Nichelli2011; Serra et al. Reference Serra, Giulietti, Cercignani, Spano, Torso, Castelli, Perri, Fadda, Marra, Caltagirone and Bozzali2013), extending into more posterior regions of the temporal lobe, parietal lobe, posterior cingulate and part of the occipital lobe, as well as substantial involvement of the frontal lobes (Bell-McGinty et al. Reference Bell-McGinty, Lopez, Meltzer, Scanlon, Whyte, Dekosky and Becker2005; Whitwell et al. Reference Whitwell, Przybelski, Weigand, Knopman, Boeve, Petersen and Jack2007b ). Interestingly, reductions in hippocampal volume and more widespread cortical atrophy in aMCI patients are similar to AD, and associated with a greater risk of conversion of aMCI to AD (Hamalainen et al. Reference Hamalainen, Tervo, Grau-Olivares, Niskanen, Pennanen, Huuskonen, Kivipelto, Hanninen, Tapiola, Vanhanen, Hallikainen, Helkala, Nissinen, Vanninen and Soininen2007; Ferreira et al. Reference Ferreira, Diniz, Forlenza, Busatto and Zanetti2011). Moreover, intermediate amounts of AD-related neuropathological findings, including amyloid deposition and neurofibrillary tangles, were also found in the mesial temporal and neocortical structures in aMCI patients relative to age-matched normal adults and patients with AD (Bourgeat et al. Reference Bourgeat, Chetelat, Villemagne, Fripp, Raniga, Pike, Acosta, Szoeke, Ourselin, Ames, Ellis, Martins, Masters, Rowe and Salvado2010; Markesbery, Reference Markesbery2010). These findings strongly suggest that the GM atrophy observed in aMCI substantially increases the risk of aMCI progression to AD.
Recently, functional magnetic resonance image (fMRI) task-specific studies have shown abnormal hippocampal activation during episodic memory performance in aMCI (Dickerson et al. Reference Dickerson, Salat, Greve, Chua, Rand-Giovannetti, Rentz, Bertram, Mullin, Tanzi, Blacker, Albert and Sperling2005; Bai et al. Reference Bai, Zhang, Watson, Yu, Shi, Yuan, Zang, Zhu and Qian2009b ; Machulda et al. Reference Machulda, Senjem, Weigand, Smith, Ivnik, Boeve, Knopman, Petersen and Jack2009). Moreover, greater hippocampal activation predicts cognitive decline in persons with aMCI (Miller et al. Reference Miller, Fenstermacher, Bates, Blacker, Sperling and Dickerson2008). These studies have identified task-related brain network activation and substantially advance our understanding of the regions involved in the performance of a given task, performance being dependent on the difficulty of the task, subject's compliance, and so on. To overcome the task-dependent fMRI pitfalls, the resting-state functional connectivity MRI (R-fMRI) approach has been employed to detect the relationship between different regions; R-fMRI measures temporal correlations between the spontaneous blood oxygenation level-dependent (BOLD) signals in different brain regions, and provides new information into how structurally segregated and functionally specialized brain regions are interconnected. Abnormal resting-state functional connectivity has been found in aMCI in several networks, including in frontal–parietal (Rombouts et al. Reference Rombouts, Barkhof, Goekoop, Stam and Scheltens2005; Qi et al. Reference Qi, Wu, Wang, Zhang, Dong, Yao and Li2010), posterior cingulate cortex (PCC)–temporal lobe (Bai et al. Reference Bai, Watson, Yu, Shi, Yuan and Zhang2009a ), hippocampus–PCC (Sorg et al. Reference Sorg, Riedl, Muhlau, Calhoun, Eichele, Laer, Drzezga, Forstl, Kurz, Zimmer and Wohlschlager2007) and insula–prefrontal (Xie et al. Reference Xie, Bai, Yu, Shi, Yuan, Chen, Li, Zhang and Li2012) networks in aMCI subjects. These investigations suggest that cognitive impairment may result from the disruption of functional integrity in the distributed brain networks in aMCI subjects at the systems level (Palop et al. Reference Palop, Chin and Mucke2006). Taken together, volumetric MRI and fMRI studies have separately revealed the potential contribution of GM volumes and distributed networks, which both collect information from multiple domain-specific brain systems, to higher-order cognitive processing in the human brain (Buckner et al. Reference Buckner, Sepulcre, Talukdar, Krienen, Liu, Hedden, Andrews-Hanna, Sperling and Johnson2009).
Recently, multimodal neuroimaging techniques have been engaged in exploring the associations among GM atrophy, intrinsic connectivity in distributed networks coordinated in large-scale brain systems, and cognitive performance. For example, GM atrophy is associated with altered intrinsic connectivity in various types of dementias (Seeley et al. Reference Seeley, Crawford, Zhou, Miller and Greicius2009), heroin addicts (Yuan et al. Reference Yuan, Qin, Dong, Liu, Sun, Liu, Zhang, Wang, Wang, Li, Zhao, von Deneen, Liu, Gold and Tian2010) but not in schizophrenia (Lui et al. Reference Lui, Deng, Huang, Jiang, Ma, Chen, Zhang, Li, Li, Zou, Tang, Zhou, Mechelli, Collier, Sweeney, Li and Gong2009). Moreover, GM atrophy and intrinsic connectivity are partially correlated with behavioral performance in these studies, whereas other fMRI studies in aMCI subjects removed the confounding effects of brain atrophy on the functional connectivity changes (Sorg et al. Reference Sorg, Riedl, Muhlau, Calhoun, Eichele, Laer, Drzezga, Forstl, Kurz, Zimmer and Wohlschlager2007; Bai et al. Reference Bai, Zhang, Watson, Yu, Shi, Yuan, Zang, Zhu and Qian2009b ). Currently, it is still unclear as to whether the changes in brain anatomy, particularly in the atrophic regions, affect the alterations of intrinsic connectivity, and how GM atrophy and intrinsic connectivity affect cognitive performance, in non-demented elderly. Thus, we propose a network-based hypothesis that the reduced GM volumes seen in critical brain structures and atrophy-related intrinsic connectivity can jointly affect cognitive deficits, and GM volumes will significantly influence the altered intrinsic connectivity in aMCI subjects.
In the current study, therefore, we use the R-fMRI approach to test the following network-based hypotheses: in aMCI subjects, (i) reduced GM volumes will be seen in critical brain structures; (ii) GM atrophy-related intrinsic connectivity will predict cognitive deficits; and (iii) reduced GM volumes will be significantly associated with altered intrinsic connectivity.
Method
Participants
A total of 33 aMCI patients and 30 cognitively normal (CN) healthy subjects were recruited to this study. All subjects were enrolled through community health-screening activities and newspaper advertisements (all were Chinese Han and right handed) in Nanjing, China.
Protocol was approved by the Research Ethics Committee of the Affiliated Zhongda Hospital, Southeast University, and written informed consent was obtained from all participants. All procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Three aMCI and four CN subjects were excluded, because of excessive motion during fMRI runs (i.e. exceeding more than 1 mm translational movement or more than 1° rotational movement) or incomplete image scans, yielding a total of 30 aMCI and 26 controls for this analysis.
All subjects underwent comprehensive neuropsychological assessments, including the Auditory Verbal Memory Test-Delayed Recall (AVMT-DR) and the Rey–Osterrieth Complex Figure Test-Immediate Recall (ROCFT-IR), Trail Making Tests A and B, Digit Span Test, Symbol Digit Modalities Test, and the Clock Drawing Test, which covered memory, executive function, perceptual speed, and visual–spatial domains. All aMCI patients met the diagnostic criteria described previously (Petersen et al. Reference Petersen, Smith, Waring, Ivnik, Tangalos and Kokmen1999; Petersen, Reference Petersen2004; Bai et al. Reference Bai, Watson, Yu, Shi, Yuan and Zhang2009a ; Liu et al. Reference Liu, Qin, Li, Fan, Wang, Jiang and Yu2013). Exclusion criteria were: regular use of drugs or medications; history of neurological or psychiatric illness; and contraindications for MRI scanning. Control subjects were required to have a Clinical Dementia Rating of 0, a Mini Mental State Examination (MMSE) score ⩾26, and an AVMT-DR score >4 for those with 8 or more years of education. The inclusion and exclusion assessments were performed by an experienced neuropsychiatrist, who administered a structured interview to subjects and their informants.
MRI protocol
Imaging was performed using a General Electric 1.5 Tesla scanner (USA) with a homogeneous birdcage head coil. High-resolution spoiled gradient-recalled echo (SPGR) three-dimensional axial images were acquired for anatomical reference. The parameters were: repetition time (TR) = 9.9 ms, echo time (TE) = 2.1 ms, flip angle (FA) = 15°, acquisition matrix = 256 × 192, field of view (FOV) = 240 × 240 mm, thickness = 1.0 mm, gap = 0 mm, number of excitations (NEX) = 1.0. Axial resting-state (no cognitive tasks were performed, eyes closed, and ears occluded) functional connectivity fMRI (R-fMRI) datasets were obtained in 7 min and 6 s with a gradient-recalled echo-planar imaging (GRE-EPI) pulse sequence, and in total 142 volumes were acquired. The R-fMRI imaging parameters were: TR = 3000 ms, TE = 40 ms, FA = 90°, acquisition matrix = 64 × 64, FOV = 240 × 240 mm, thickness = 4.0 mm, gap = 0 mm, NEX = 1.0.
Voxel-based morphometry (VBM) analysis
Optimized VBM, combined with Statistical Parametric Mapping (SPM8) software (http://www.fil.ion.ucl.ac.uk/spm) on a Matlab 7.5 (MathWorks, http://www.mathworks.com/) platform were utilized to perform GM volume analysis (Duyn & Moonen, Reference Duyn and Moonen1993). First, the study group-specific template was created by averaging the high-resolution T1-weighted images for all participants after spatial normalization. Second, two-step segmentation of the brain tissues was done using the analysis of likelihood of each voxel of the brain image to be classified as GM, white matter or cerebrospinal fluid (CSF). Third, spatial normalization for the GM volume of all participants was conducted based on study group-specific GM template and the absolute GM volume modulated for each subject. Fourth, all segmented GM images were smoothed with an 8-mm Gaussian kernel and transformed using a logit transformation [logit(n) = 0.5 ln(n/[1 − n])] to make the data more normally distributed. Finally, these data were transformed to Montreal Neurological Institute (MNI) space. To find brain regions with a robust group difference, a two-sample voxelwise t test was performed, and the significance level was set at p < 0.0001 (corrected for multiple comparisons with family-wise error). Several regions of interests (ROIs) reflecting significant clusters of GM atrophy found in aMCI subjects compared with control subjects were selected as seed regions for the functional connectivity analysis and these ROIs volumes were also extracted from all subjects using masks generated from the group comparison.
Functional connectivity analysis
R-fMRI data analysis was carried out using Analysis of Functional NeuroImages (AFNI) software (http://afni.nimh.nih.gov/afni). Briefly, motion correction was performed by volume registration on the R-fMRI data (3dvolreg). Then, detrending was carried out to remove Legendre polynomials (3dDetrend). Possible contamination from the signals in white matter and CSF, and six-motion vectors were regressed out from the whole brain (3dDeconvolve), and the global signal was also regressed out (Chen et al. Reference Chen, Chen, Xie, Ward, Li, Antuono and Li2012). Then, a band-pass filter was applied to keep only low-frequency fluctuations within the frequency range of 0.015 and 0.1 Hz.
The time course of each subject's ROI from the functional EPI images was extracted. The spatial resolutions in the SPGR (1 × 1 × 1 mm) and EPI images (3.75 × 3.75 × 4 mm) were different; therefore, only those voxels in the EPI images that contained at least 50% of the individual ROI voxels masked on the three-dimensional SPGR images were included in the voxel time course analysis (Xu et al. Reference Xu, Xu, Wu, Antuono, Rowe and Li2008; Xie et al. Reference Xie, Li, Shao, Fu, Goveas, Ye, Li, Cohen, Chen, Zhang and Yang2011). Then, the averaged time course of each ROI, as the seed time course, was correlated with the time courses of all brain voxels using Pearson cross-correlation. Next, we applied the Pearson correlation coefficients (r) to a Fisher z transform analysis, which yielded variants of approximately normal distribution [m = 0.5 ln(1+ r)/(1− r)] (Zar, Reference Zar1996). Finally, the data were spatially normalized to the Talairach template image, resampled to 2-mm isotropic voxels, and smoothed with a Gaussian kernel (6 mm full width half maximum), using AFNI software.
Statistical analysis
Behavioral data
The composite scores for cognition were developed in a two-step process (Xie et al. Reference Xie, Bai, Yu, Shi, Yuan, Chen, Li, Zhang and Li2012). First, the composite scores for different cognitive domains were created: z scores for each test for each subject were converted from the raw scores with reference to the means and standard deviations of all the subjects. We then calculated composite scores by averaging the z scores of individual tests: Episodic memory (two tests including AVMT-DR and ROCFT-IR) and non-episodic memory (five tests, executive function including Trail Making Tests A and B, working memory including the Digit Span Test, perceptual speed including the Symbol Digit Modalities Test, and visual–spatial function including the Clock Drawing Test). The global cognitive function score was formed by averaging the z scores of all seven tests. These composite scores were compared between aMCI and CN subjects by t tests. The MMSE was performed for description and diagnostic classification, but not used in the composite cognitive measures. These composite scores increase power by reducing random variability, floor and ceiling effects, and the number of correlated data in analyses (Schneider et al. Reference Schneider, Boyle, Arvanitakis, Bienias and Bennett2007), and are widely used (Ringman et al. Reference Ringman, Diaz-Olavarrieta, Rodriguez, Chavez, Fairbanks, Paz, Varpetian, Maldonado, Macias-Islas, Murrell, Ghetti and Kawas2005; Schneider et al. Reference Schneider, Boyle, Arvanitakis, Bienias and Bennett2007; Wilson et al. Reference Wilson, Leurgans, Foroud, Sweet, Graff-Radford, Mayeux and Bennett2010).
Group-level functional connectivity analysis
Individual connectivity maps of ROIs for each subject were analysed with a random-effects one-sample t test to identify voxels. This showed a significant positive or negative correlation with the seed time series, and the pattern of each ROI network for CN and aMCI groups was obtained, respectively (p < 0.01, corrected with AlphaSim, cluster size >450 mm3). To find brain regions with a significant group difference, a voxelwise two-sample t test for each network was performed at the threshold of p < 0.01 (corrected with AlphaSim, cluster size >450 mm3). Then, the average strength of functional connectivity in each cluster was extracted from each subject using masks generated from the group comparison in the distributed network, and then underwent the following regression analysis using commercially available statistical software (SPSS 16.0, USA).
Joint effects of GM atrophy and altered intrinsic connectivity on behavioral performance
To investigate the effects of GM atrophy and altered intrinsic connectivity on cognitive performance, a multivariate linear regression analysis was employed as below:
$$Beh = {\beta _0} + {\beta _1} \times GMV + {\beta _2} \times Networ{k_{FC}} + {\beta _3} \times AGE + {\beta _4} \times Edu + {\beta _5} \times Gen + {\beta _6} \times Group + {\rm \in} $$
where Beh is the composite scores of global cognition, episodic memory, and non-episodic memory in each subject, GMV is the total GM volumes of eight atrophic regions and Network FC is the averaged strength of functional connectivity in all eight functional networks for individual subject (see description below). β 0 is the intercept of straight line fitting in the model; β 1 and β 2 are the effects of GMV and Network FC on the cognitive performance, respectively; β 3, β 4, β 5, and β 6 are the effects of age, education, gender and group in the above model; ∈ is the error of this model.
To increase the statistical power, the raw GM volumes of each ROI and the functional connectivity strength in the distributed network for each subject were converted to z scores according to the mean and standard deviation for all participants, respectively, and then, the total z scores of GM volumes for all ROIs were summed to yield the composite measures of the GM volumes, and the z scores of functional connectivity strength in all networks were averaged to obtain composite measures of functional connectivity for each subject, separately. These are similar to the behavioral data processing. Then, we performed the above multivariate linear regression model using these variables after controlling for the effects of age, education, gender and group.
Results
Subject characterization
Demographic information and clinical evaluations for all participants are shown in Table 1. No significant differences in age or gender were noted between the aMCI and CN groups. In examining both groups, significant cognitive deficits were found in the aMCI group.
Demographic information and psychometric data on the cognitive tests and composite measures
Data are given as mean (standard deviation) unless otherwise indicated.
aMCI, Amnesic mild cognitive impairment; CN, cognitively normal; n.s., non-significant; MMSE, Mini-Mental State Examination; AVMT-DR, Auditory Verbal Memory Test-Delayed Recall; ROCFT-IR, Rey–Osterrieth Complex Figure Test-Immediate Recall.
a p Value was obtained by χ2 test.
b p Values were obtained by two-sample t tests (two-tailed).
Reduced GM volume in aMCI group
Brain regions with significant GM atrophy included the bilateral precuneus and insula, and the left postcentral gyrus (PostCG), medial frontal gyrus (MeFG), middle frontal gyrus (MFG) and hippocampus, as shown in Fig. 1 (detailed information for these eight regions are described in online Supplementary Table S1).
Group comparison identifying atrophic gray matter (GM) regions. The results display significant GM volume reduction in: 1, the left postcentral gyrus; 2, left hippocampus; 3, left medial frontal gyrus; 4, left middle frontal gyrus; 5, right precuneus; 6, left precuneus; 7, left insula; and 8, right insula on representative sections of the template brain. Blue color indicates decreased GM volumes in the amnestic mild cognitive impairment patients compared with the cognitively normal subjects. The color bar is presented with T values.
Resting-state functional connectivity patterns in CN and aMCI subjects
Resting-state functional connectivity maps of each atrophy-related region in the CN and aMCI groups (p < 0.01, with cluster size >450 mm3) are briefly illustrated in Fig. 2. Generally, each network is composed of a positive network and anticorrelated (negative) network, and the strongest connectivity of each network was adjacent to its seed region. Non-adjacent regions also showed significant connectivity associated with each seed ROI. Detailed information for each network pattern is described in online Supplementary Table S2.
Resting-state functional connectivity maps between each gray matter atrophy-related region of interest and the rest of the brain in cognitively normal (CN) and amnestic mild cognitive impairment (aMCI) subjects (p corrected < 0.01). These maps show the functional connectivity pattern of each network between CN and aMCI subjects. Bright color indicates voxels with significant positive connectivity and blue color indicates voxels with negative (anticorrelation) connectivity with the seed region. The color bar is presented with z scores. LHip, Left hippocampus; LMeFG, left medial frontal gyrus; LMFG, left middle frontal gyrus; LPostCG, left postcentral gyrus; LPcu, left precuneus; LIns, left insula; RPcu, right precuneus; RIns, right insula.
Altered functional connectivity in distinct networks in aMCI group
We examined the group differences of connectivity specificity in each network. As shown in online Supplementary Tables S3 and S4, significantly decreased functional connectivity in eight distributed networks was observed in the aMCI cohort. Briefly, these brain regions were found in: (1) the left hippocampus network, primarily including the bilateral cuneus, right middle occipital gyrus, precuneus and left inferior frontal gyrus (IFG); (2) the left MeFG network, primarily involved in the left lingual gyrus (LG) and anterior temporal pole (aTP) and right precuneus; (3) the left MFG network, mainly encompassing the left PCC, right inferior parietal cortex (IPC), precuneus, IFG and fusiform area; (4) the left PostCG network, including the right superior temporal gyrus (STG) and PCC; (5) the left precuneus network, particularly in the right precuneus and STG, left SFG and LG; (6) the left insula network, mainly located in the left IPC, dorsolateral prefrontal cortex (DLPFC), pre-somatomotor area, MeFG and aTP; (7) the right precuneus network, especially in the left anterior cingulate cortex, IFG and MeFG; and (8) the right insula network, primarily localized in the right middle cingulate cortex, aTP, left PostCG and IFG. We did not find significantly increased functional connectivity of individual networks in the aMCI subjects compared with CN subjects, after correction for multiple comparisons. Representative connectivity changes for the distributed functional networks are shown in Fig. 3.
Representative regions of significantly altered functional connectivity in distinct networks between cognitively normal (CN) and amnestic mild cognitive impairment (aMCI) subjects. The results illustrate decreased functional connectivity strength (m value) in the distinctive network in aMCI subjects, as opposed to that in CN subjects. In each panel, the left brain image shows functional connectivity from the seed region (red color) to the target region (blue color). The right image shows the numerical presentation of the functional connectivity strength in the target region connected to the seed region in the CN group (red color) and the aMCI group (blue color). Values are means, with standard deviations represented by vertical bars. lIns, Left insula; lDLPFC, left dorsolateral prefrontal cortex; lHip, left hippocampus; lIFG, left inferior frontal gyrus; rPcu, right precuneus; lACC, left anterior cingulate cortex; rIns, right insula; lPreCG, left precentral gyrus; rSTG, right superior temporal gyrus; lMFG, left middle frontal gyrus; rIPC, right inferior parietal gyrus; lPcu, left precuneus; lSFG, left superior frontal gyrus; lMeFG, left medial frontal gyrus.
Relationship of behavioral performance, GM atrophy and altered intrinsic connectivity
A multivariate linear regression model was established (R 2 = 0.24, p = 1.79 × 10−3). Further, we also identified that GM volumes were positively correlated with global cognition (R 2 = 0.13, p < 0.0006), episodic memory (R 2 = 0.09, p < 0.01) and with non-episodic memory (R 2 = 0.06, p < 0.07); intrinsic connectivity strength was positively correlated with global cognition (R 2 = 0.26, p < 0.0001), episodic memory (R 2 = 0.18, p < 0.001) and non-episodic memory (R 2 = 0.14, p < 0.004), after controlling for the effects of the age, education, gender and group, as described in Fig. 4. These different R 2 values suggest the distinct contribution of GM atrophy and intrinsic functional connectivity to cognitive impairment in aMCI patients.
Joint effects of gray matter (GM) volumes and intrinsic functional connectivity (FC) strength on cognitive performance across all subjects: cognitively normal (CN) and amnestic mild cognitive impairment (aMCI) subjects. Multivariate linear regression analysis demonstrated that both GM volumes in atrophic regions and FC strength contribute to cognitive performance. (a) GM volumes in atrophic regions are significantly correlated with global cognition (GC) (R 2 = 0.13, p < 0.007) and episodic memory (EM) (R 2 = 0.09, p < 0.01). There is a trend in correlation with non-EM (R 2 = 0.06, p < 0.07). (b) Intrinsic FC strength is significantly correlated with GC (R 2 = 0.26, p < 0.0001), EM (R 2 = 0.18, p < 0.001) and non-EM (R 2 = 0.14, p < 0.004).
Further, partial correlation analysis identified that GM volumes were significantly correlated with intrinsic connectivity strength after controlling for the effects of the above covariance (p < 0.025).
Discussion
In this study, we demonstrated reduced GM volumes and altered functional connectivity of GM atrophy-related networks in the aMCI cohort relative to CN subjects, and, importantly, GM atrophy and these disrupted functional connectivities jointly affected cognitive performance in these non-demented elderly. In addition, we also found that GM volumes potentially influenced functional connectivity strength across all subjects.
We first identified that the GM atrophic regions were mainly located in the frontal–parietal system, bilateral insula and left hippocampus. This is consistent with previous studies reporting that these regions were typically involved in AD-related pathology (Karas et al. Reference Karas, Scheltens, Rombouts, Visser, van Schijndel, Fox and Barkhof2004; Gili et al. Reference Gili, Cercignani, Serra, Perri, Giove, Maraviglia, Caltagirone and Bozzali2011). Importantly, these brain areas are widely recognized to mediate episodic memory encoding, and are densely connected to other neocortical regions (Brambati et al. Reference Brambati, Belleville, Kergoat, Chayer, Gauthier and Joubert2009; Schmidt-Wilcke et al. Reference Schmidt-Wilcke, Poljansky, Hierlmeier, Hausner and Ibach2009). Therefore, regional abnormalities, such as hippocampal atrophy, as frequently seen in aMCI and AD subjects, may lead to a dysfunction in information processing related to episodic and semantic memory (Schmidt-Wilcke et al. Reference Schmidt-Wilcke, Poljansky, Hierlmeier, Hausner and Ibach2009; Joubert et al. Reference Joubert, Brambati, Ansado, Barbeau, Felician, Didic, Lacombe, Goldstein, Chayer and Kergoat2010). Recently, a meta-analysis of VBM studies in the aMCI subjects showed that significant GM volume reduction, especially in the left hippocampus and parahippocampus, serves as a good biomarker for predicting the conversion from aMCI to AD (Ferreira et al. Reference Ferreira, Diniz, Forlenza, Busatto and Zanetti2011). It is also intriguing that widespread cortical areas, as well as the medial temporal lobe, are affected in the early stages of AD. For example, atrophy in the precuneus, extending to more diffuse medial and lateral posterior parietal regions, as well as extensively accelerated atrophy in frontal regions were identified in very mild stages of AD (Buckner et al. Reference Buckner, Snyder, Shannon, LaRossa, Sachs, Fotenos, Sheline, Klunk, Mathis, Morris and Mintun2005). In addition, insula atrophy has often been reported in the previous studies that focused on aMCI subjects and AD patients, suggesting that insula atrophy may be a biomarker to differentiate progressive MCI and stable MCI in the early stages of AD, and to monitor the progression of aMCI conversion to AD (Spulber et al. Reference Spulber, Niskanen, Macdonald, Kivipelto, Padilla, Julkunen, Hallikainen, Vanninen, Wahlund and Soininen2012). The consistency of these observations indicates that GM atrophy in the above regions in aMCI subjects is associated with a high risk of subsequent progression to AD.
Next, significantly decreased functional connectivity of eight GM atrophy-related networks were found in aMCI subjects relative to the CN subjects (online Supplementary Tables S3 and S4). Briefly, in the CN subjects, these eight ROI networks, as expected, presented similar patterns to previous findings (Cavanna, Reference Cavanna2006; Bai et al. Reference Bai, Zhang, Watson, Yu, Shi, Yuan, Zang, Zhu and Qian2009b ; Xie et al. Reference Xie, Bai, Yu, Shi, Yuan, Chen, Li, Zhang and Li2012). However, in the aMCI subjects, the disrupted intrinsic connectivity of each network was primarily located in the frontal–parietal system, temporal and occipital lobes, and subcortical regions. Recently, network analyses of anatomical and functional connectivity in non-human and human primates have indicated that some brain areas, known as hubs, are highly interconnected and functionally specialized (Bullmore & Sporns, Reference Bullmore and Sporns2009). These hubs, often including the DLPFC, MeFG, PCC and pMTG, play an essential role in the integration of diverse information, balancing the signal transfer among multiple, specific neural pathways in order to perform high-order cognitive process (Sporns et al. Reference Sporns, Honey and Kotter2007; Hagmann et al. Reference Hagmann, Cammoun, Gigandet, Meuli, Honey, Wedeen and Sporns2008; Buckner et al. Reference Buckner, Sepulcre, Talukdar, Krienen, Liu, Hedden, Andrews-Hanna, Sperling and Johnson2009). Damage to these hubs can disrupt the resting-state default mode network activity, which as a core network has been suggested to play a pivotal role in cognitive processes, including mind wandering, goal-directed behavior and self-preference (Dosenbach et al. Reference Dosenbach, Fair, Miezin, Cohen, Wenger, Dosenbach, Fox, Snyder, Vincent, Raichle, Schlaggar and Petersen2007; Fox & Raichle, Reference Fox and Raichle2007; Mason et al. Reference Mason, Norton, Van Horn, Wegner, Grafton and Macrae2007). Also, destruction of the precuneus and insula networks perhaps leads to dysfunctional attention and episodic memory impairment (Cavanna, Reference Cavanna2006; Dorfel et al. Reference Dorfel, Werner, Schaefer, von Kummer and Karl2009; Xie et al. Reference Xie, Bai, Yu, Shi, Yuan, Chen, Li, Zhang and Li2012), which present as core features characterizing aMCI patients. Moreover, disruption of these delicately balanced regulations in the functional integrity among various neural networks, but not structural damage, increases vulnerability to AD pathology (Buckner et al. Reference Buckner, Sepulcre, Talukdar, Krienen, Liu, Hedden, Andrews-Hanna, Sperling and Johnson2009). This evidence, at least in part, supports the notion that AD, and possibly aMCI, are disconnection syndromes (Greicius et al. Reference Greicius, Srivastava, Reiss and Menon2004; Delbeuck et al. Reference Delbeuck, Collette and Van der Linden2007). Taken together, these disrupted connectivities in aMCI patients may reflect the decoupling of functional brain networks underlying interoception, memory processing, appropriate emotional and cognitive responses, and probably represent the potential atrophy-related network organization among individuals with aMCI.
It is widely accepted that high-order cognitive processing results from the dynamic interaction of distributed brain areas operating at the large-scale neural-network level instead of isolated brain regions (Pessoa, Reference Pessoa2008; Mesulam, Reference Mesulam2009). Currently, the fact that GM volumes and intrinsic connectivity of specific neural networks are functionally linked to distinctive behavioral modulation indicates that intactness in the cognitive functions requires the integrity of the complex neural networks and stability in the GM volumes in discrete anatomic brain regions at the system level. Understanding the underlying mechanism of this fact is critical for gaining new insight into the neural basis of the conversion of aMCI to AD. First, accumulating evidence has suggested that separate physiological processes are associated with distinct cognitive performances (Andrews-Hanna et al. Reference Andrews-Hanna, Snyder, Vincent, Lustig, Head, Raichle and Buckner2007). Previous studies have reported that GM atrophy or dysfunctional connectivity are associated with the decline in cognition in aMCI patients (Laird et al. Reference Laird, Eickhoff, Li, Robin, Glahn and Fox2009; Wang et al. Reference Wang, LaViolette, O'Keefe, Putcha, Bakkour, Van Dijk, Pihlajamäki, Dickerson and Sperling2010), further indicating that coherent mental representation generated through reorganization of neural networks to improve the ability to perform cognitively demanding tasks are seen in patients with MCI and AD (Agosta et al. Reference Agosta, Rocca, Pagani, Absinta, Magnani, Marcone, Falautano, Comi, Gorno-Tempini and Filippi2010). Second, we also found that overall connectivity strength in aMCI patients was lower than that in CN subjects (see Fig. 4), suggesting the functional decoupling of information in aMCI patients. Interesting, our findings identified a distinct contribution of GM atrophy and intrinsic functional connectivity to cognitive impairment in aMCI patients; that is, the regional GM atrophy was significantly associated with poorer cognitive performance, including global cognition, episodic memory, and even non-episodic memory, and separately accounted for about 13, 9 and 6% of the variance in cognitive performance (Ersche et al. Reference Ersche, Jones, Williams, Turton, Robbins and Bullmore2012). By contrast, the disrupted intrinsic connectivity can account for 26, 18 and 14% of the variance of poorer cognitive performance, as shown in Fig. 4, indicating the greater contribution of intrinsic connectivity over regional GM volumes to cognitive performance.
Additionally, we found that GM atrophy was significantly correlated with intrinsic connectivity, indicating that functional integrity in multiple distributed neural networks is selectively and vulnerably disrupted by structural damage. This will extend our understanding of fundamental cognitive process in the human brain with memory disorders, such as aMCI and AD. In addition, these atrophic regions are also key components of the higher-order cognitive networks; therefore, GM atrophy in these regions may result in the disruption of distributed cognitive networks, and thus, detrimentally affect information processing, especially for episodic memory, in aMCI subjects (Balthazar et al. Reference Balthazar, Yasuda, Cendes and Damasceno2010).
Our study is not without limitations. First, we predefined the atrophic areas as seed regions and proved the behavioral significance of the GM volumes and intrinsic connectivity. This may not reflect the significance of whole-brain GM atrophy and altered functional networks. Future studies are needed to investigate the functional role of the large-scale, distributed, functional networks, including the whole-brain regions in aMCI subjects. Second, anatomical connectivity studies with diffusion tensor imaging (DTI) have highlighted an important contribution of white matter fiber in the functional integrity of structural networks, so tracing white matter fibers based on structural connectivity analysis using the DTI approach to investigate the functional integrity of the neural network should also be a focus for future studies. Third, the combination of R-fMRI and specific cognitive tasks, which can reflect the subtle decline of cognition, can detect the association between cognitive performance and the neural network reorganization underlying the different pathological state. This is essential in determining the putative etiologic roles of functional network abnormalities in subjects with aMCI. Additionally, we focused on addressing the potential contribution of combining GM atrophy and altered functional networks to cognitive function in the current study, and did not consider the individual GM atrophic region's or network's influence. This may limit our understanding of the contribution of each region or network to cognition in the early stage of the AD process.
Conclusions
In summary, this study demonstrated that GM atrophy can significantly affect intrinsic connectivity, and that GM atrophy and altered functional connectivity jointly affect cognitive deficits, indicating that combining quantitative information about GM volumes and the strength of functional connectivity may successfully predict cognitive performance in non-demented elderly.
Supplementary material
For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291714002876
Acknowledgements
C.X. receives research funding support from the National Natural Science Foundation of China (81171323, 91332118); F.B. receives support from the National Natural Science Foundation of China (81201080, 91332104). H.Y. was supported by the National Natural Science Foundation of China (81070877); Y.S. was funded by the National Natural Science Foundation of China (81171021); Y.Y. receives research funding support from the National Natural Science Foundation of China (30970814, 81071101), the Nature Science Foundation of Jiangsu Province (BK2009050) and the Key Program of Medical Development of Nanjing (ZKX07018); D.W., Z.-S.Z. and B.Y. do not report disclosures; Z.-J.Z. receives research support from the National Basic Research Program of China (973 Program) (2007CB512308), the National Hi-Tech Research and Development Program of China (863 Program) (2008AA02Z413), National Science and Technology Major Projects for ‘Major New Drugs Innovation and Development’ (2012ZX09506-001-009) and the National Natural Science Foundation of China (30770779, 30825014, 30971016, 81061120529).
The authors thank Dr Gavin Reynolds for editorial assistance and Mr B. Douglas Ward (Biophysics, Medical College of Wisconsin, WI 53226, USA) for his contribution to the statistical analysis.
Declaration of Interest
All authors have made substantial intellectual contribution to this paper in one or more of the following areas that included design or conceptualization of the study, analysis or interpretation of the data, or drafting or revising the manuscript. C.X. conducted the statistical analysis. All authors gave final approval of the manuscript. No authors of this paper have any possible conflict of interest, financial or otherwise, related directly or indirectly to this work.
