Older NC subjects were recruited from the community via newspaper advertisement. Eligibility requirements for recruitment in this study were no MRI contradictions, living independently in the community, Mini-Mental State Examination (MMSE) ≥ 26, absence of neurological or psychiatric illness, lack of major medical illnesses and medications that affect cognition, and normal performance on cognitive tests. Forty-four subjects underwent PET imaging and fMRI for this study.
Additionally, resting state fMRI data were collected on 17 young subjects to define a template for the DMN (mean age = 23.0 (2.9), 9 females). Young subjects were recruited from the community through online postings. PIB-PET data from 22 AD patients were used for comparison purposes (mean age = 65.9 (10.7), 10 females). AD patients were recruited from the University of California San Francisco (UCSF) Memory and Aging Center. The diagnosis of AD was based on a comprehensive multidisciplinary evaluation that includes a clinical history and physical examination, a caregiver interview and a battery of neuropsychological tests (Kramer et al. 2003
). All AD subjects met National Institute of Neurological Disorders and Stroke criteria for probable AD (McKhann et al. 1984
) and had no significant comorbid medical, neurologic, or psychiatric illnesses.
All NC subjects underwent detailed cognitive testing in multiple domains to ensure normality. Normal cognitive performance was defined by creating composite scores in episodic memory (EM) (long delay free recall portion of the California Verbal Learning Test [CVLT] [Delis et al. 2000
] and Wechsler Memory Scale [WMS-R] visual reproduction [Wechsler 1987b
]), working memory (Wechsler Adult Intelligence Scale [WAIS-R] digit span backwards [Wechsler 1987a
] and listening span total recall [Salthouse et al. 1991
]), and frontal function (Trails B
] and Stroop total correct in 60 s [Zec 1986
]) across a larger cohort of subjects ≥ 60 years old who underwent neuropsychological testing (198 subjects aged ≥ 60 were enrolled at the time of this study, mean age = 73.1 (7.6) and MMSE = 28.7 (1.7)). Subjects were considered ineligible if 1 composite score fell below 2 standard deviations (SDs) from our cohort defined age/gender/education-adjusted means. The individuals in this cohort were high functioning, such that a 2 SD cutoff yielded raw scores that were generally well within the range of age-adjusted normative data. These scores are thus more conservative than normative-derived cutoffs used in diagnosing mild cognitive impairment.
For subjects that had undergone multiple testing sessions, scores closest to the PET scan date were used (mean delay between PET and closest testing session was 3.93 (2.62) months, and the maximum delay was 10.2 months). Given the minimal amount of cognitive decline that may be expected in high PIB subjects in this short time period (Storandt et al. 2009
), as well as the very slow rates of change in PIB uptake over time (Engler et al. 2006
; Jack et al. 2009
), this short delay is unlikely to have any effect on the results.
PIB was synthesized at the Lawrence Berkeley National Laboratory's (LBNL) Biomedical Isotope Facility using a published protocol and described in detail previously (Mathis et al. 2003
; Mormino et al. 2009
). PIB-PET imaging was performed using a Siemens ECAT EXACT HR PET scanner (Siemens Medical Systems) in 3D acquisition mode. Ten to fifteen mCi of PIB was injected into an antecubital vein. Dynamic acquisition frames were obtained as follows: 4 × 15 s, 8 × 30 s, 9 × 60 s, 2 × 180 s, 8 × 300 s, and 3 × 600 s (90 min total). A 10-min transmission scan for attenuation correction were obtained for each PIB scan. Filtered backprojected reconstructions were performed on the transmission and emission data to judge transmission alignment with each frame of emission data. In the case of misalignment, the transmission image was coregistered to that individual emission frame and then forward projected to create an attenuation correction file specific to that head position. PET data were reconstructed using an ordered subset expectation maximization algorithm with weighted attenuation. Images were smoothed with a 4-mm Gaussian kernel with scatter correction.
MRI data on old and young NC subjects were collected at LBNL on a 1.5 T Magnetom Avanto System (Siemens Medical Systems) with a 12 channel head coil run in triple mode. The MRI session includes the following sequences (in order of acquisition): a T2-weigted fluid attenuated inversion recovery scan (FLAIR, axially acquired, time repetition [TR]/time echo [TE] = 9730/100 ms, flip angle = 150°, 0.80 × 0.80-mm2 in-plane resolution, 3.00-mm thickness with no gap), 3 T1-weighted volumetric magnetization prepared rapid gradient echo scans (MP-RAGE, axially acquired, TR/TE/time to inversion [TI] = 2110/3.58/1100 ms, flip angle = 15°, 1.00 × 1.00-mm2 in-plane resolution, 1.00-mm thickness with 50% gap), a T1 structural scan in plane to the resting state scan (axially acquired, TR/TE = 500/10 ms, flip angle = 150°, 0.80 × 0.80-mm2 in-plane resolution, 3.5-mm thickness with 15% gap), and a resting state fMRI scan (acquired axially, TR/TE = 1890/50 ms, flip angle = 90°, 3.0 × 3.0-mm2 in-plane resolution, 3.5-mm3 thickness with 15% gap, 250 TRs total). The FLAIR scan was used to screen for stroke, whereas the in-plane T1 and the MP-RAGE scans were used in subsequent processing steps. There was a mean time delay of 2.7 (6.3) months between PET and MRI scanning.
Resting state fMRI was not available for UCSF AD subjects. MP-RAGE scans for these subjects were used in the analysis of PIB-PET data. For 13 UCSF AD subjects, MP-RAGE scans were collected coronally on a 1.5 T Vision System (Siemens Medical Systems) with a quadrature head coil (TR/TE/TI = 10/7/300 ms, flip angle = 15°, 1.00 × 1.00-mm2 in-plane resolution, 1.40-mm slice thickness with no gap). For the remaining 9 UCSF AD subjects, MP-RAGE scans were collected sagittally on a Bruker MedSpec 4 T system with an 8 channel head coil (TR/TE/TI = 2300/3.37/950 ms, flip angle = 7°, 1.00 × 1.00-mm2 in-plane resolution, 1.00-mm slice thickness with no gap).
PIB data were preprocessed using the SPM2 software package (http://www.fil.ion.ucl.ac.uk/spm
). Realigned PIB frames corresponding to the first 20 min of acquisition were averaged and used to guide coregistration to the subject's structural MRI scan. Distribution volume ratios (DVRs) for PIB images were created using Logan graphical analysis with frames corresponding to 35–90 min postinjection and a gray matter masked cerebellum reference region (Logan et al. 1996
; Price et al. 2005
When applicable, PET images were coregistered to the subject's high-resolution structural scan and transformed to Montreal Neurological Institute (MNI) space using parameters defined from nonlinear alignment between the high-resolution structural scan and the MNI template (see the fMRI Preprocessing section). After normalization to template space, these PET images were smoothed an additional 8 mm, resulting in a total of 8.9 mm smoothing for these images.
Structural MRI Processing
MP-RAGE scans were processed as described previously using FreeSurfer version 4.3 (http://surfer.nmr.mgh.harvard.edu/
) (Mormino et al. 2009
). In brief, MP-RAGE scans were realigned and averaged to create a single high-contrast structural image. Anatomical masks relevant to PET processing were derived in each subject's native space using this analysis stream (Dale et al. 1999
; Fischl et al. 2001
; Fischl et al. 2002
; Segonne et al. 2004
). Specifically, a gray matter only cerebellum mask was used as a reference region for PIB and the mean PIB DVR value from a FreeSurfer derived cortical gray matter mask was extracted for each subject and used as a measure of global PIB uptake (“global PIB”) and as a regressor in the voxelwise FC analysis. These global PIB values were highly correlated with the PIB index value used in our previous publication (R2
= 0.98) (Mormino et al. 2009
fMRI data were processed using FMRIB Software Library (FSL) version 4.1 (http://www.fmrib.ox.ac.uk/fsl
). Images were motion corrected, low-pass (2.8 s) and high-pass filtered (100 s), and smoothed with a 5-mm Gaussian kernel. To define the spatial transformation from fMRI space to MNI template space, a multistep registration procedure was employed. First, the mean fMRI image was linearly registered to the subject's in-plane T1
structural image using 7 degrees of freedom. The in-plane T1
image was then registered to the high-resolution structural scan using 6 degrees of freedom. Finally, the high-resolution structural scan was nonlinearly aligned to the standard MNI 152 brain using FNIRT, and resulting parameters were used to transform fMRI data.
FC analyses were conducted with FSL version 4.1 and adapted from the goodness of fit procedure described by Greicius et al. (2004)
. Supplementary Figure 1
provides a detailed schematic flow chart of the processing stream. A DMN template was defined using resting state fMRI data from the group of 17 young subjects. A seed in the posterior cingulate cortex (PCC; sphere of 8 mm centered at MNI coordinates −12, −50, 32) was transformed from template space into native fMRI space for each young subject. Additionally, nuisance regions were defined on the MNI template and transformed to native space (a white matter 8-mm spherical seed centered at MNI coordinates −24, −16, 36, a lateral ventricle mask drawn on the template, and a whole brain mask derived from segmenting the template brain and combining gray and white matter). Time series were extracted across all these regions and entered into a general linear model, and resulting contrast maps reflecting voxels correlated with the PCC time series (covarying the signal associated with white matter, lateral ventricle, and whole brain) were entered into a higher-level 1-sample t
-test (height and extent joint threshold of P
< 0.05, corrected). Based on evidence that the DMN may be “split” into posterior and anterior components with ICA (Damoiseaux et al. 2006
), we excluded the anterior medial prefrontal cortex (mPFC) portion from this map and used the remaining posterior clusters as our template during the goodness of fit procedure (a “posterior” DMN template). This would ensure that posterior DMN components were selected for subjects with separable DMN components (27/44 subjects showed separable DMN components). A posterior rather than anterior template approach was employed because the posterior regions of the DMN have been implicated in EM processes (Andrews-Hanna et al. 2010
) and show early dysfunction in AD (Minoshima et al. 1997
; Killiany et al. 2002
; Sorg et al. 2007
; Thompson et al. 2007
). Resting state data for each older NC was decomposed at the individual subject level using ICA with FSL's Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC) (Beckmann and Smith 2004
). MELODIC isolates a multitude of components for each subject, and for each component, voxels are assigned a z
-score that reflects the extent to which that voxel's time series is correlated with the time series associated with the specific component. A goodness of fit procedure was applied to spatially normalized z
-maps for each subject to determine which component most closely resembled the DMN template (average z
-score of voxels within the template minus average z
-score of voxels outside the template; z
-values from the removed mPFC area were excluded from this calculation to ensure that high mPFC connectivity did not inflate the average value “outside” the template). The z
-map for the best-fit component for each subject was used in subsequent higher-level analyses.
Correlations between global PIB and demographic variables were completed using the statistical programming language R version 2.8 (http://www.r-project.org/
). Partial correlation coefficients (r
) were reported for relationships between global PIB and EM/MMSE (controlling for demographic predictors). Caret version 5 was used to display voxelwise results (http://brainvis.wustl.edu/wiki/index.php/Caret:About
Spatial overlap between the DMN and the PIB uptake was qualitatively examined by overlaying statistical maps. The DMN FC map was derived from a 1-sample t-test of DMN best-fit components from the elderly NC group, whereas the PIB uptake map was derived from contrasting low PIB NC with high PIB NC DVR images (defined with a median split of global PIB values). The FC map was thresholded at P < 0.001 and k = 100, uncorrected and the PIB map was thresholded at P < 0.0002 and k = 100, uncorrected, binarized, and overlaid upon a 3D rendered brain.
Within the NC group, global PIB was treated as a continuous value and regressed against voxelwise DMN FC using permutation testing with FSL's Randomise (using 5000 permutations, http://www.fmrib.ox.ac.uk/fsl/randomise/
). This analysis was restricted to the voxels that were significant in the DMN 1-sample t
-test of elderly NC subjects (thresholded at P
< 0.001 and k
= 100, uncorrected; this mask is displayed in ). Age, gender, and education were controlled for in this analysis, and results were considered significant at P
< 0.05 and k
= 50, uncorrected. Cluster size, maximum t
-value, and peak MNI coordinates for significant regions are reported.
Figure 1. DMN FC and global PIB uptake overlap. One sample t-test of DMN best-fit components (yellow), 2-sample t-test between high and low PIB subjects (blue), and overlap (red) are displayed. These maps highlight congruence and incongruence between the DMN and (more ...)
To explore focal PIB uptake within regions showing a relationship between FC and global PIB, region of interest (ROI) analyses were completed with R. ROIs were created from the primary FC voxelwise results by casting an 8-mm sphere around the voxel with the highest t-value in a subset of significant clusters (voxels showing the strongest relationship between FC and global PIB). PIB values from these ROIs were extracted from spatially normalized PET images. A repeated measures analysis of variance (ANOVA) was conducted with PIB DVR values as the dependent variable, group (AD and NC) as a between-subjects factor and ROI as a within-subjects factor. Main effects of group and ROI were investigated further with post hoc contrasts [1) AD vs. NC for each ROI and 2) each ROI vs. global PIB controlling for diagnosis]. Within the NC group, mean FC was extracted from these ROIs and regressed against EM. Age, gender, and education were controlled for in each model. Results were considered significant at P < 0.05.