Selective changes of resting-state networks
in individuals at risk for Alzheimer’s disease
Christian Sorg*†, Valentin Riedl‡§, Mark Mu ¨hlau‡, Vince D. Calhoun¶?, Tom Eichele**, Leonhard La ¨er††,
Alexander Drzezga‡‡, Hans Fo ¨rstl*, Alexander Kurz*, Claus Zimmer††, and Afra M. Wohlschla ¨ger‡††‡‡
Departments of *Psychiatry,‡Neurology,††Neuroradiology, and‡‡Nuclear Medicine, Klinikum Rechts der Isar, Technische Universita ¨t Mu ¨nchen,
Ismaningerstrasse 22, 81675 Munich, Germany;¶MIND Institute, 1101 Yale Boulevard, Albuquerque, NM 87131;?Department of Electrical
and Computer Engineering, University of New Mexico, Albuquerque, NM 87131; **Cognitive Neuroscience Group, Faculty of Psychology,
University of Bergen, Jonas Lies Vei 91, 5011 Bergen, Norway; and§Munich Center for Neurosciences, Brain and Mind, Ludwig-Maximilians Universita ¨t;
Grosshadernerstrasse 2, 82152 Martinsried, Germany
Edited by Leslie G. Ungerleider, National Institutes of Health, Bethesda, MD, and approved September 25, 2007 (received for review September 17, 2007)
Alzheimer’s disease (AD) is a neurodegenerative disorder that
prominently affects cerebral connectivity. Assessing the functional
connectivity at rest, recent functional MRI (fMRI) studies reported
on the existence of resting-state networks (RSNs). RSNs are char-
acterized by spatially coherent, spontaneous fluctuations in the
blood oxygen level-dependent signal and are made up of regional
patterns commonly involved in functions such as sensory, atten-
tion, or default mode processing. In AD, the default mode network
(DMN) is affected by reduced functional connectivity and atrophy.
In this work, we analyzed functional and structural MRI data from
healthy elderly (n ? 16) and patients with amnestic mild cognitive
impairment (aMCI) (n ? 24), a syndrome of high risk for developing
AD. Two questions were addressed: (i) Are any RSNs altered in
aMCI? (ii) Do changes in functional connectivity relate to possible
structural changes? Independent component analysis of resting-
state fMRI data identified eight spatially consistent RSNs. Only
selected areas of the DMN and the executive attention network
demonstrated reduced network-related activity in the patient
group. Voxel-based morphometry revealed atrophy in both medial
temporal lobes (MTL) of the patients. The functional connectivity
between both hippocampi in the MTLs and the posterior cingulate
of the DMN was present in healthy controls but absent in patients.
We conclude that in individuals at risk for AD, a specific subset of
RSNs is altered, likely representing effects of ongoing early neu-
rodegeneration. We interpret our finding as a proof of principle,
demonstrating that functional brain disorders can be characterized
by functional-disconnectivity profiles of RSNs.
default mode network ? intrinsic brain activity ? mild cognitive impairment
ropsychiatric symptoms (1). AD is neuropathologically defined
by tau pathology and amyloid aggregations (2). Tau pathology
starts in regions of the medial temporal lobe (MTL) and is well
correlated with cell loss and atrophy; amyloid deposition pri-
marily affects distributed neocortical regions but is not especially
prominent in the MTL (2, 3). Atrophy of the MTL is correlated
with the degree of dementia and also the extent of temporopa-
rietal hypometabolism; both results are assumed to reflect
changes in cerebral connectivity, especially between the MTL
and the neocortex (3–5). In non-human primates, prominent
structural connectivity between the MTL and neocortical re-
gions as well as broad neocortical hypometabolism after ablation
of parts of the MTL were demonstrated (6, 7). Evidence for
disrupted structural and functional connectivity (FC) further
suggests that AD includes a disconnection syndrome (5, 8–10).
Mild cognitive impairment (MCI) is a syndrome with cogni-
tive decline greater than expected for an individual’s age and
educational level but not interfering notably with activities of
daily living; prevalence of MCI is ?15% in adults older than 65
years; more than half of patients with MCI progress to dementia
within 5 years; the amnestic subtype of MCI has a high risk of
lzheimer’s disease (AD) is a neurodegenerative disorder
clinically characterized by progressive dementia and neu-
progression to AD constituting a prodromal stage of AD (1, 11,
12). Previous results of task-related functional MRI (fMRI) in
patients with MCI (13, 14) indicate that FC seems to be already
impaired in prodromal stages of AD (15). Reduced white matter
volumes of the MTL in amnestic MCI (aMCI) point at changed
MTL–neocortex connectivity (16). Very recent fMRI studies in
AD reported on FC changes especially during rest (5, 9, 17).
Together, these findings suggest that the functional integration
of brain areas in the cerebral resting state in individuals at risk
for AD is disturbed and that functional changes are related to
The study of intrinsic brain activity may be central for under-
standing the physiology of functional brain disorders (5, 18–20).
Functional brain disorders such as AD, schizophrenia, or autism
are characterized by structural alterations that are subtle or have
an uncertain relationship to clinical symptoms (21). Such struc-
tural lesions might be functionally related to alterations of
intrinsic brain activity that are reflected by changes of connec-
tivity (18, 19). Here, the study of spontaneous coherent fluctu-
ations of the blood oxygen level-dependent (BOLD) signal at
rest by fMRI is of special interest. Synchronized BOLD fluctu-
ations overlap with brain systems that are involved in functions
such as motor, sensory, language, attention, or default mode
processing (22–28). Evidence for the neuronal nature of so-
called resting state networks (RSNs) comes from studies that
employ simultaneous fMRI and electroencephalograms (EEGs)
(29, 30), from the observation of altered connectivity caused by
neurological diseases (5), and from the existence of homologous
RSNs in non-human primates that overlap with neuroanatomi-
cally defined systems (31).
Regions including the posterior cingulate, inferior parietal,
and medial prefrontal cortex, constitute a RSN called default
mode network (DMN) (32, 33). The areas of the DMN show
consistently greater BOLD activity during rest than during any
attention-demanding task, a phenomenon called deactivation;
the same regions are prominently involved in episodic memory
at rest are anticorrelated to the spontaneous fluctuations of a
widely distributed neocortical system that largely overlaps with
attention-related RSNs (5, 22, 27, 28, 32, 34–36). In AD, parietal
Author contributions: C.S. and V.R. contributed equally to this work; C.S., A.D., H.F., A.K.,
and A.M.W. designed research; C.S., L.L., H.F., A.K., C.Z., and A.M.W. performed research;
V.D.C. and T.E. contributed new reagents/analytic tools; C.S., V.R., M.M., and A.M.W.
analyzed data; and C.S., V.R., and A.M.W. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Freely available online through the PNAS open access option.
†To whom correspondence should be addressed. E-mail: email@example.com.
This article contains supporting information online at www.pnas.org/cgi/content/full/
© 2007 by The National Academy of Sciences of the USA
November 20, 2007 ?
vol. 104 ?
regions of the DMN demonstrate altered functional connectivity
at rest within the DMN itself, to the MTL, and to neocortical
areas that are involved in attention processes (5, 9, 17). Ac-
counting for these findings, the hypothesis suggested above of
altered FC at rest in individuals at high risk for AD can be
specified in terms of RSNs with a focus on FC changes in the
DMN and attention-related RSNs. In general, the finding of
RSNs allows for a broadened perspective on functional brain
disorders in the field of neuroimaging: the brain shows rich
intrinsic dynamics in absence of tasks, with external stimuli more
modulating than determining brain activity (19, 20, 27, 31, 37);
the plurality of RSNs reflects this structured intrinsic activity,
and selective changes of RSNs may characterize traits and states
of functional brain disorders such as AD (18, 19, 21, 26).
Most previous studies investigating RSNs have used region-
of-interest (ROI)-based correlation analyses (9, 17, 22–24). The
signal time course of a selected ROI is correlated with remaining
brain areas, resulting in a ROI specific correlation map. More
recently, instead of defining a priori spatial hypotheses, a number
of studies used model-free approaches involving independent
component analysis (ICA) to describe RSNs in rest-fMRI data
(5, 25, 26, 30). ICA allows for the extraction of distinct spatio-
temporal patterns by identifying spatially independent and tem-
porally synchronous brain regions (38).
In the present work, we combined ICA and ROI-based
correlation methods to investigate RSNs in patients with aMCI.
We focused on the following questions: (i) Are any RSNs
changed in aMCI compared with healthy elderly? (ii) Are there
any volumetric changes in the patient group pointing at possible
ongoing neurodegeneration? (iii) How are potential functional
and structural changes related with respect to their spatial
We examined 24 patients and 16 healthy elderly (Table 1).
Participants were instructed to close their eyes and relax during
4 min of fMRI scanning. Rest-fMRI data were analyzed by using
a group ICA approach involving subject-wise concatenation of
the individual fMRI data sets and subsequent back-projection
into subject space (15, 39). Voxel-based morphometry (VBM) of
additional structural MRI data and ROI-based correlation
analyses of the rest-fMRI data were performed to analyze
possible causes of altered functional connectivity.
RSNs in Healthy Elderly and Patients with aMCI. The ICA was
performed employing a group ICA model for fMRI data
(GIFT).§§After the IC estimation on the subject-wise concate-
nated data, the toolbox performs the analysis in three stages:
data reduction through principal component analysis (PCA),
ICA decomposition of aggregate data, and subsequent back-
reconstruction of individual subject maps and time courses (39,
40). Results of the separate analyses for each study group are
depicted in Fig. 1 [see also supporting information (SI) Table 2].
Each group IC image contains a pair of two spatial IC patterns
that are correlated (red) or anticorrelated (blue) with the time
course of the component (data not shown). In the corresponding
glass brain projection, the result of a one-sample t test for the
back-reconstructed individual subject IC patterns across both
groups is shown [P ? 0.05 false discovery rate (FDR) corrected
for multiple comparisons]. Eight IC patterns represented func-
tionally relevant RSNs as described previously (26). Both IC
patterns of Fig. 1e/E and f/F were pairwise anticorrelated within
one IC. The first two IC patterns of Fig. 1 partly coincide with
the visual cortex: a/A covers inferior parts of the occipital cortex
such as the striate area [Brodmann area (BA) 17/18]; b/B shows
lateral and dorsal parts of the occipital gyrus (BA 19). The
transverse (BA 41) and superior temporal gyrus (BA 22/42) of
the auditory cortex are depicted in Fig. 1c/C. Fig. 1d/D repre-
sents a right lateralized frontoparietal network comprising the
supramarginal gyrus (BA 39/40), middle temporal gyrus (BA
21), and middle frontal gyrus (BA 8) on the right side as well as
bilateral frontal areas covering middle and inferior frontal gyrus
attention network (34).
In Fig. 1e/E the upper IC pattern covers medial and lateral
parts of the superior parietal lobe as well as the precuneus (BA
7). This network is constituted by areas well known to be active
during spatial attention (41). The anticorrelated network of Fig.
1e/E (blue) includes a number of areas relevant for sensorimotor
coordination in pre- and postcentral gyri (BA 3/4), the caudal
zone of the cingulate motor cortex (BA 24), and parts of the
superior frontal gyrus (BA 8) (41). The upper IC pattern of Fig.
1f/F includes medial prefrontal areas (BA 9/10/11), anterior (BA
12/32), and posterior cingulate cortex (PCC) (BA 23/31), the
inferior temporal gyrus (BA 21), and bilateral supramarginal
gyrus (BA 39) in the parietal lobe. This pattern corresponds to
the DMN (26, 32, 42). Involvement of the hippocampus (HC), as
seen by Greicius et al. (5), was not observed here. The anticor-
7/40), V5/MT, and inferior temporal gyrus (BA 37), as well as
parts of the inferior frontal gyrus mainly on the right side. These
areas comprise a dorsal/executive attention network (34).
Altered RSNs in Patients with aMCI. The second analysis aimed at
between-group differences in corresponding RSNs. We per-
formed a two-sample t test on each of the eight RSNs contrasting
the individual, back-reconstructed IC patterns of both groups.
The only contrasts revealing any group difference concerned
both IC patterns of Fig. 1f/F and are shown in Fig. 2 (P ? 0.05,
FDR corrected). None of the two-sample t tests on the five
remaining components revealed significant group difference at
P ? 0.001, uncorrected for multiple comparisons (cluster extent
IC pattern demonstrated reduced component time course-
related activity in the patient group (see also SI Table 3): left
PCC (BA 31), right medial prefrontal cortex (BA 10), and two
small clusters in parietal cortex bilaterally (BA 39). Accordingly,
we found reduced activity in the patient group in bilateral
superior parietal lobes (BA 7, 40) and in the right prefrontal
cortex, namely precentral (BA 4) and inferior frontal gyrus (BA
9) of the anticorrelated IC pattern of Fig. 1f/F.
Patients with aMCI Have Reduced Gray Matter Volume in the Medial
Temporal Lobes. To assess possible causes of reduced functional
connectivity, we analyzed our data for structural differences
between both study groups. Neither the analyses of global
volumes of gray matter (GM) and white matter (WM) nor the
on Acoustics, Speech, Signal Processing, Philadelphia, PA, pp. 401–404.
Table 1. Subject demographic information
Education, ?/?12 years
CDR, sum of boxes
CERAD (delayed recall)
68.1 ? 3.8
0 ? 0
29.6 ? 0.5
7.4 ? 1.3
69.3 ? 8.1
2.2 ? 0.9*
27.7 ? 1.1*
4.3 ? 2.1*
Data are presented as mean ? SD. NC, normal control; CDR, Clinical De-
mentia Rating scale; MMSE, Mini-Mental State Exam; CERAD, Consortium to
Establish a Registry for Alzheimer’s Disease.*, P ? 0.05 difference between
Sorg et al.PNAS ?
November 20, 2007 ?
vol. 104 ?
no. 47 ?
voxel-wise analysis of WM revealed significant differences.
However, the voxel-wise analysis of GM revealed GM loss
bilaterally in the MTL including the HC, the thalamus, the
insular cortex, as well as in patches within the inferior parietal
lobe (Fig. 3). These findings comply with previous reports on
subtle abnormalities in structural brain images of patients with
aMCI (16, 43). These areas did not overlap with those regions
found to be altered in RSNs of the patient group.
At-Rest Coactivation of HC and PCC Is Absent in Patients with aMCI.
To examine the relation of neocortical disconnectivity and MTL
atrophy we investigated the functional connectivity between the
HC and the RSNs with ROI correlation analyses. We hypoth-
esized reduced coactivation of HC and PCC of the DMN in the
patient group. The PCC is the central region of the posterior
DMN that is primarily affected by AD-associated alterations
such as hypometabolism or elevated atrophy rate; impaired
PCC–MTL connectivity is assumed to be the main cause for
prominent metabolic PCC changes in AD (3–7, 17). We com-
puted the pairwise correlations between the time courses from
each HC and the PCC cluster which was identified by ICA. The
average correlation between left/right HC and PCC was z ?
0.37/0.30, SE ? 0.08/0.08 in the control group and z ? ?0.03/
?0.09, SE ? 0.10/0.08 in the patient group. We found significant
group differences for each HC (P ? 0.04/0.02, Bonferroni-
anticorrelated (blue) with the time course of the component (data not shown). (a/A–d/D) Each pair of IC patterns is shown within one brain image. (e/E and f/F)
Each upper row represents the correlated IC pattern, each lower row the anticorrelated one. IC patterns are superimposed on a single-subject high-resolution
T1 image. The black to yellow/light blue color scale represents z values, ranging from 1.8 to 8.0. Glass brain projections illustrate results of one-sample t tests
on the individual back-reconstructed subject IC patterns across both groups (P ? 0.05, FDR-corrected). (a/A–d/D) One-sample t test on the anticorrelated
individual subject IC patterns provided no significant results. In axial view the right hemisphere is displayed on the right. NC, normal controls.
RSNs of normal controls (a–f) and patients with aMCI (A–F). Each group IC image contains a pair of two spatial IC patterns that are correlated (red) or
(Left) Maps corresponding to DMNs of Fig. 1f/F, Upper. (Right) Maps corre-
sponding to executive attention networks of Fig. 1f/F, Lower. Color maps
represent significant (P ? 0.05, FDR-corrected) voxels of higher component-
color-coded with black to yellow (from 0 to 5.0) to light blue (from 0 to 6.2).
Maps are superimposed on a single-subject high-resolution T1 image. The
patient group did not show any significant higher activation for DMN and
executive attention network. For all remaining RSNs of Fig. 1, two-sample t
Contrasts of RSNs between normal controls and patients with aMCI.
decreased GM in aMCI compared with controls superimposed on sagittal and
coronal slices transformed in standard MNI space. Corresponding t values are
color-coded with black to yellow (from 0 to 6.3). The distribution of the
Color maps showing significant (P ? 0.05, FDR-corrected) voxels of
www.pnas.org?cgi?doi?10.1073?pnas.0708803104Sorg et al.
corrected) by applying an analyses of covariance, which included
corresponding HC volume as covariate. The correlation between
left and right HC did not differ significantly between healthy
controls (z ? 0.76, SE ? 0.10) and patients (z ? 0.62, SE ? 0.08)
in a two-sample t test. To evaluate the degree of functional
coactivation of the HCs and any RSN, we separately correlated
the time courses of each HC with the time course of each RSN
across all subjects. The strongest and only significant effect was
found between left HC and the DMN (z ? 0.13, SE ? 0.06, P ?
0.003, Bonferroni-corrected, one-sample t test), indicating rel-
evant functional integration of the HC and the DMN during
Discussion and Conclusion
In this work we investigated spatiotemporal patterns of hemo-
dynamic activity during relaxed wakefulness and underlying
brain volumes in healthy elderly and patients with amnestic MCI.
By applying ICA, VBM, and ROI-based correlation analyses, we
characterized eight RSNs that were spatially consistent across
subjects and corresponded with functionally relevant patterns.
Areas of the DMN and the executive attention network showed
diminished functional connectivity in patients, whereas the
respective volumes remained unaffected. Atrophy was found for
both MTLs, including HCs in the patient group. Functional
connectivity between both HCs and left PCC of the DMN was
absent in patients. These results suggest that in aMCI a selected
subset of RSNs is affected by altered functional connectivity,
likely representing effects of early neurodegeneration.
disconnection phenomena associated with prodromal stages of
AD are specific to a subset of RSNs whereas other networks
remain unaffected. By applying ICA in patients with aMCI, we
found left PCC and right medial prefrontal cortex of the DMN
as well as bilateral superior parietal lobes and bilateral inferior
frontal gyri of the executive attention network to be selectively
affected (Fig. 2). Regions of the affected RSNs did not differ
significantly in volumetric aspects between the two groups,
confirming the functional nature of the observed alterations
(Fig. 3). Our finding concerning the DMN is consistent with
previous results demonstrating altered deactivation in regions of
the DMN in MCI (44). The areas constituting the DMN and the
executive attention network also seem to be involved in early
changes of AD: In patients at risk for AD activity of the PCC or
superior parietal cortex is changed during memory or executive
attention-related tasks, respectively (14, 15). In mild AD, these
regions show changed functional connectivity at rest (5, 9) and
overlap with regional patterns of atrophy, glucose hypometab-
olism, and hypoperfusion (for overview, see ref. 3). Also, regions
showing early amyloid deposition overlap with posterior parts of
the default mode and executive attention network (2, 3, 45). In
summary, our result of selective changes of RSNs in individuals
at risk for AD corresponds very well with changes that have
previously only been described in AD.
Changes of the MTL are discussed as a possible factor causing
neocortical disconnectivity in AD (2, 4, 5, 7). The HC and
posterior parts of the DMN display coherent spontaneous
activity at rest (36) and constitute an episodic memory network
that is linked to successful recollection (35). Functional connec-
tivity of the HC with neocortical areas, especially with the PCC,
is reduced in AD during rest (17). The VBM analysis (Fig. 3)
demonstrated reduced GM within the MTL in the patient group.
This finding is in line with previous results for aMCI (16, 43) and
points to MTL neurodegeneration (3) as well as impaired
MTL—neocortex connectivity (16). Our ROI correlation anal-
connectivity with the left PCC of the DMN in patients compared
with controls. Among all RSNs, the DMN was functionally most
strongly related to the HC at rest. This relation was demon-
strated by the only significant correlation between left HC and
DMN across all subjects, which is in line with previous results (5,
36). Taken together, the presented results indicate relevant
functional integration of the HC in the DMN at rest as well as
an impaired HC–parietal memory system in aMCI.
of RSNs in individuals at high risk for AD reflects altered connec-
tivity between the MTL and neocortical areas (4, 5, 7, 9, 15, 44).
Apparently, MTL—neocortex disconnectivity is related to neuro-
degeneration, which is expressed by MTL atrophy (16). In addition
to FC changes in the DMN, our result also points to relevant
is in line with observed attentional deficits in MCI and AD (9, 14,
15). This finding indicates impaired interaction between these two
anticorrelated networks that prominently organize intrinsic brain
for diagnostic purposes; one possible way would be to use ICA-
derived patterns to define ROIs where FC analysis can classify
patients from controls. We are currently in the process of assessing
symptom progression in our patients in a two-year follow-up
examination where FC analysis might provide a more accurate way
than neuropsychological testing and structural imaging alone.
RSNs and Functional Brain Disorders. In this work we identified
spatially consistent RSNs across both study groups that match
previous results (22, 24, 26, 46). Damoiseaux et al. (26) evaluated
the spatial consistency of RSNs and found a set of 10 reliably
many parameters such as ICA model, age, and health condition of
participants, the regional patterns of detected RSNs show large
regional concordance with their findings. Identified RSNs can be
1 a/A, b/B, c/C, and e/E) are associated with sensory or sensori-
motor functions, characterized in several previous studies (23, 24,
46, 47). The remaining networks encompass regions involved in
higher cognitive functions (34, 35, 41, 48, 49). The bilateral poste-
rior parietal network of map e/E including precuneus and intrapa-
rietal sulcus is known to be associated with spatial attention
processes (41). This network is anticorrelated to the RSN of map
e/E representing regions normally involved in sensorimotor inte-
gration; areas of both networks are part of a functional system
participating in goal-directed movement coordination (41). The
spatial pattern of Fig. 1f/F (Upper) covers areas of the DMN; this
network is suggested to support default mode function, such as
maintaining a background level of attention for the detection of
salient events by monitoring internal and external environment
(49). Evidence increases that large parts of the posterior DMN
together with the hippocampal system are associated with autobio-
network of Fig. 1f/F (Lower), involving bilateral superior parietal
cortex, intraparietal sulcus, and inferior frontal gyrus, and the right
lateralized frontoparietal network of Fig. 1d/D overlap with the
regions of the dorsal/executive and ventral/reorienting attention
systems (34). The dorsal attention system is assumed to be involved
in top–down direction of attention, and the ventral system, later-
alized to the right, supports reorienting of attention in response to
salient stimuli (27). Rest fluctuations of the two systems, which are
considered member of a group of areas routinely exhibiting task-
related activation, are anticorrelated to spontaneous BOLD fluc-
tuations of the DMN, demonstrated by ROI-based methods (22,
28). Using ICA, we found the anticorrelated coupling of the DMN
limited to the dorsal attention network. Our result is supported by
the finding that the two attention networks can be distinguished by
their spontaneous rest activity (27); it seems plausible that a system
concerned with attentional shifts in response to salient external
Sorg et al.PNAS ?
November 20, 2007 ?
vol. 104 ?
no. 47 ?
stimuli is less strongly connected to introspectively oriented default
orienting of attention.
In summary, we identified eight distinct, partly anticorrelated,
and spatially consistent RSNs in healthy elderly and patients with
aMCI. The division of the rest-related BOLD signal into separate
RSNs presumably reflects the functional organization of brain
activity into stabilized functional–anatomical systems (18, 19, 31).
Regarding the synchronicity of spontaneous BOLD fluctuations
during rest, pathology-induced changes seem to exist also in pa-
tients with autism, schizophrenia, attention deficit hyperactivity
disorder, or major depression (51–54). Taking into account these
findings and our result of selective changes of RSNs in aMCI, we
suggest that rest-fMRI and especially RSNs constitute very prom-
ising tools for the functional characterization of functional brain
disorders, for intergroup comparisons, and possibly with some
potential for assessing functional connectivity on a single-subject
Materials and Methods
Subject data and ICA of rest-fMRI data are described below.
Detailed information regarding ROI-based correlation analyses
of fMRI data and VBM analysis of structural MRI data can be
found in SI Materials and Methods.
Subjects and Task. Sixteen healthy controls (6 female, ages 63–73
aMCI participated in the study. All subjects provided informed
consent in accordance with the Human Research Committee
Mu ¨nchen. Patients were recruited from the Memory Clinic of the
Department of Psychiatry, controls by word-of-mouth advertising.
examination, informant interview (only for patients), neuropsycho-
logical assessment [CERAD battery, Consortium to Establish a
Registry for AD (55)], structural MRI, and for patients additional
blood tests. Patients met criteria for aMCI which contain reported
and neuropsychologically assessed memory impairments, largely
intact activities of daily living, and excluded dementia (Table 1)
(11). Exclusion criteria for entry into the study were other neuro-
logical, psychiatric, or systemic diseases (e.g., stroke, depression,
alcoholism), or clinically remarkable MRI (e.g., stroke lesions)
which could be related to cognitive impairment. Five controls and
nine patients were treated for hypertension (?-blockers, angioten-
sin-converting enzyme inhibitors, and calcium channel blockers),
and six controls and eight patients were treated for hypercholes-
terolemia (statins). None of the subjects had diabetes mellitus.
None of the subjects or patients received psychotropic medication,
especially cholinesterase inhibitors.
All subjects underwent 4 min of resting-state scan first followed
by an attention and a memory task, which are not discussed here.
For the resting-state scan, subjects were instructed simply to keep
their eyes closed, not to think of anything in particular, and not to
Imaging Methods. Imaging was performed on a 1.5T Siemens
Symphony system. Functional data were collected by using a
gradient echo EPI sequence (TE ? 50 ms, TR ? 3,000 ms, flip
angle ? 90°, FoV ? 200 mm2, matrix ? 64 ? 64, 33 slices, slice
thickness ? 4 mm, and 0.4-mm interslice gap) (where TE is echo
time, TR is repetition time, FoV is field of view, and T1 is
inversion time) for a 4-min period resulting in a total of 80
volumes. The first three functional scans were discarded before
the subsequent analysis. A T1-weighted anatomical dataset was
obtained from each subject by using a magnetization-prepared
rapid acquisition gradient echo sequence (TE ? 3.93 ms, TR ?
1,500 ms, TI ? 760 ms, flip angle ? 5°, FoV ? 256 mm2, matrix ?
256 ? 256, 160 slices, voxel size ? 1 ? 1 ?1 mm3).
Oxford Centre for Functional Magnetic Resonance Imaging of the
Brain Software Library (FMRIB; FSL version 3.2), statistical
parametric mapping (Wellcome Department of Cognitive Neurol-
ogy; SPM5), and in-house software for Matlab (MathWorks).
In a first step, nonbrain structures were removed from the echo
planar imaging volumes. Next we performed a mean-based inten-
sity normalization of all slices within a volume by the same factor
(56). Data were then motion-corrected, spatially normalized into
the stereotactic space of the Montreal Neurological Institute
(MNI), and spatially smoothed with an 8 ? 8 ? 8 mm Gaussian
kernel with SPM5. Before they were entered into the ICA, a
voxel-wise transformation was applied on the time course data
yijk(t), y ˆijk(t) ? [yijk(t) ? ?yijk?]/?ijk(for each voxel: t, time; i, j, k, three
directions in space; ?yijk?, mean; ?ijk standard deviation). This
respect to BOLD amplitudes from the four-dimensional data set
y ˆijk(t). Sensitivity for variance correlation was thereby rendered
independently of variance magnitude.
ICA. We performed the ICA by using group ICA for fMRI
toolbox (GIFT version 1.3b; icatb.sourceforge.net)§§established
for the analysis of fMRI data (15, 39, 54). The toolbox supports
a group ICA approach, which first concatenates the individual
data across time, followed by the computation of the subject-
specific components and time courses. For each of the two study
groups the toolbox performed the analysis in three stages: (i)
data reduction, (ii) application of the ICA algorithm, and (iii)
back-reconstruction for each individual subject (39).
In the first step (i), data from each subject were reduced by
using PCA, whereby computational complexity was reduced
and most of the information content of the data was preserved.
After concatenating the resulting volumes, the number of
independent sources was estimated by the GIFT dimension-
ality estimation tool based on the aggregated data: 28/31 ICs
for the control/patient group (57). The final reduction step
according to the selected number of components was achieved
by PCA again. In the second stage of the analysis (ii) we chose
the Infomax algorithm for running the proper IC analysis and
a GM mask based on all subjects. In the final stage of
back-reconstruction (iii), time courses and spatial maps were
computed for each subject. After back-reconstruction, the
mean spatial maps of each group were transformed to z scores
for display (39). Before any statistics were applied to the
individual subject maps, the initially calculated scaling factor
?ijkwas reintegrated into the data by voxel-wise multiplication.
Each IC contains a pair of two spatial IC patterns that are
correlated or anticorrelated to the time course of the component.
(26, 46). The remaining IC patterns were attributed to two major
forms of artifacts: IC patterns representing tissue border artifacts
near the ventricular system, the skull, and cerebrospinal fluid space
?20 (MNI) in axial slices burdened by major vessel artifacts and
lack of reliability in EPI scans. Individual subject IC patterns
effects analyses in SPM5. Results were thresholded at P ? 0.05,
sample t tests were masked with a within-group mask thresholded
at P ? 0.05, uncorrected.
We thank two reviewers for constructive suggestions. This work was
supported by the German Federal Ministry of Education and Research
Grant 01EV0710 and by Kommision fu ¨r Klinische Forschung, Klinikum
Rechts der Isar, Mu ¨nchen Grant 8765160.
www.pnas.org?cgi?doi?10.1073?pnas.0708803104Sorg et al.
1. Blennow K, de Leon MJ, Zetterberg H (2006) Lancet 368:387–403. Download full-text
2. Braak H, Braak E (1991) Acta Neuropathol 82:239–259.
3. Buckner RL, Snyder AZ, Shannon BJ, LaRossa G, Sachs R, Fotenos AF,
Sheline YI, Klunk WE, Mathis CA, Morris JC, et al. (2005) J Neurosci
4. Meguro K, LeMestric C, Landeau B, Desgranges B, Eustache F, Baron JC
(2001) J Neurol Neurosurg Psychiatry 71:315–321.
5. Greicius MD, Srivastava G, Reiss AL, Menon V (2004) Proc Natl Acad Sci USA
6. Lavenex P, Suzuki WA, Amaral DG (2002) J Comp Neurol 447:394–420.
7. Blaizot X, Meguro K, Millien I, Baron JC, Chavoix C (2002) J Neurosci
8. Van Hoesen GW (1997) J Neuropsychiatry Clin Neurosci 9:331–341.
9. Wang K, Liang M, Wang L, Tian L, Zhang X, Li K, Jiang T (2007) Hum Brain
10. Delbeuck X, Van der Linden M, Collette F (2003) Neuropsychol Rev 13:79–92.
S, Brodaty H, Bennett D, Chertkow H, et al. (2006) Lancet 367:1262–1270.
12. Petersen RC, Smith GE, Waring SC, Ivnik RJ, Tangalos EG, Kokmen E (1999)
Arch Neurol 56:303–308.
13. Dickerson BC, Salat DH, Greve DN, Chua EF, Rand-Giovannetti E, Rentz
DM, Bertram L, Mullin K, Tanzi RE, Blacker D, et al. (2005) Neurology
14. Dannhauser TM, Walker Z, Stevens T, Lee L, Seal M, Shergill SS (2005) Brain
15. Celone KA, Calhoun VD, Dickerson BC, Atri A, Chua EF, Miller SL, DePeau
K, Rentz DM, Selkoe DJ, Blacker D, et al. (2006) J Neurosci 26:10222–10231.
16. Stoub TR, DeToledo-Morrell L, Stebbins GT, Leurgans S, Bennett DA, Shah
RC (2006) Proc Natl Acad Sci USA 103:10041–10045.
17. Wang L, Zang Y, He Y, Liang M, Zhang X, Tian L, Wu T, Jiang T, Li K (2006)
18. Buckner RL, Vincent JL (2007) NeuroImage 37:1091–1096.
19. Fox MD, Raichle ME (2007) Nat Rev Neurosci 8:700–711.
20. Raichle ME, Mintun MA (2006) Annu Rev Neurosci 29:449–476.
21. Matthews PM, Honey GD, Bullmore ET (2006) Nat Rev Neurosci 7:732–744.
22. Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle ME
(2005) Proc Natl Acad Sci USA 102:9673–9678.
23. Hampson M, Peterson BS, Skudlarski P, Gatenby JC, Gore JC (2002) Hum
Brain Mapp 15:247–262.
24. Biswal B, Yetkin FZ, Haughton VM, Hyde JS (1995) Magn Reson Med
25. Beckmann CF, Smith SM (2004) IEEE Trans Med Imaging 23:137–152.
26. Damoiseaux JS, Rombouts SA, Barkhof F, Scheltens P, Stam CJ, Smith SM,
Beckmann CF (2006) Proc Natl Acad Sci USA 103:13848–13853.
27. Fox MD, Corbetta M, Snyder AZ, Vincent JL, Raichle ME (2006) Proc Natl
Acad Sci USA 103:10046–10051.
28. Fransson P (2005) Hum Brain Mapp 26:15–29.
29. Laufs H, Kleinschmidt A, Beyerle A, Eger E, Salek-Haddadi A, Preibisch C,
Krakow K (2003) NeuroImage 19:1463–1476.
30. Mantini D, Perrucci MG, Del Gratta C, Romani GL, Corbetta M (2007) Proc
Natl Acad Sci USA 104:13170–13175.
31. Vincent JL, Patel GH, Fox MD, Snyder AZ, Baker JT, Van Essen DC, Zempel
JM, Snyder LH, Corbetta M, Raichle ME (2007) Nature 447:83–86.
32. Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman
GL (2001) Proc Natl Acad Sci USA 98:676–682.
33. Raichle ME, Snyder AZ (2007) NeuroImage 37:1083–1090.
34. Corbetta M, Shulman GL (2002) Nat Rev Neurosci 3:201–215.
35. Shannon BJ, Buckner RL (2004) J Neurosci 24:10084–10092.
36. Vincent JL, Snyder AZ, Fox MD, Shannon BJ, Andrews JR, Raichle ME,
Buckner RL (2006) J Neurophysiol 96:3517–3531.
37. Buzsaki G, Draguhn A (2004) Science 304:1926–1929.
38. Calhoun VD, Adali T (2006) IEEE Eng Med Biol Mag 25:79–90.
39. Calhoun VD, Adali T, Pearlson GD, Pekar JJ (2001) Hum Brain Mapp
40. Schmithorst VJ, Holland SK (2004) J Magn Reson Imaging 19:365–368.
41. Grefkes C, Fink GR (2005) J Anat 207:3–17.
42. De Luca M, Beckmann CF, De Stefano N, Matthews PM, Smith SM (2006)
43. Whitwell JL, Petersen RC, Negash S, Weigand SD, Kantarci K, Ivnik RJ,
Knopman DS, Boeve BF, Smith GE, Jack CR, Jr (2007) Arch Neurol 64:1130–
44. Rombouts SA, Barkhof F, Goekoop R, Stam CJ, Scheltens P (2005) Hum Brain
45. Klunk WE, Engler H, Nordberg A, Wang Y, Blomqvist G, Holt DP, Bergstrom
M, Savitcheva I, Huang GF, Estrada S, et al. (2004) Ann Neurol 55:306–319.
46. Beckmann CF, Smith SM (2005) NeuroImage 25:294–311.
47. Xiong J, Parsons LM, Gao JH, Fox PT (1999) Hum Brain Mapp 8:151–156.
48. Alescio-Lautier B, Michel BF, Herrera C, Elahmadi A, Chambon C, Touzet C,
Paban V (2007) Neuropsychologia 45:1948–1960.
49. Gusnard DA, Raichle ME, Raichle ME (2001) Nat Rev Neurosci 2:685–694.
50. Addis DR, Wong AT, Schacter DL (2007) Neuropsychologia 45:1363–1377.
51. Cherkassky VL, Kana RK, Keller TA, Just MA (2006) NeuroReport 17:1687–
52. Greicius MD, Flores BH, Menon V, Glover GH, Solvason HB, Kenna H, Reiss
AL, Schatzberg AF (2007) Biol Psychiatry 62:429–437.
53. Liang M, Zhou Y, Jiang T, Liu Z, Tian L, Liu H, Hao Y (2006) NeuroReport
54. Garrity AG, Pearlson GD, McKiernan K, Lloyd D, Kiehl KA, Calhoun VD
(2007) Am J Psychiatry 164:450–457.
55. Morris JC, Heyman A, Mohs RC, Hughes JP, van Belle G, Fillenbaum G,
Mellits ED, Clark C (1989) Neurology 39:1159–1165.
56. Manjaly ZM, Marshall JC, Stephan KE, Gurd JM, Zilles K, Fink GR (2003)
57. Li YO, Adali T, Calhoun VD (2007) Hum Brain Mapp 28:1251–1266.
Sorg et al. PNAS ?
November 20, 2007 ?
vol. 104 ?
no. 47 ?