Clinical effect of white matter network disruption related to amyloid and small vessel disease

Participants.

We prospectively recruited 251 participants with cognitive impairment who underwent Pittsburgh compound B (PiB)-PET and structural brain MRI, from July 2007 to July 2011. All participants were clinically diagnosed at Samsung Medical Center. Patients with amnestic mild cognitive impairment (aMCI, n = 45) and those with subcortical vascular mild cognitive impairment (svMCI, n = 67) met the Petersen criteria for mild cognitive impairment with modifications.¹³ Patients with probable AD dementia (n = 69) met criteria proposed by the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association.¹⁴ Patients with SVaD (n = 70) met the diagnostic criteria for vascular dementia as determined by the DSM-IV, and also met the imaging criteria for SVaD proposed by Erkinjuntti et al.¹⁵ Patients with svMCI and SVaD had severe WMH on fluid-attenuated inversion recovery (FLAIR) images, which was defined as periventricular WMH ≥10 mm and deep WMH ≥25 mm, as modified from the Fazekas ischemia criteria.¹⁶ Patients with aMCI and AD were classified as having minimal (periventricular WMH <5 mm and deep WMH <5 mm) or moderate WMH (between minimal and severe WMH). The WMH meets the definition of WMH of presumed vascular origin, recently proposed by Wardlaw et al.¹⁷

Patients with territorial infarctions, WMH due to radiation injury, leukodystrophy, multiple sclerosis, or vasculitis were excluded. All patients underwent a clinical interview and neurologic examination. In addition, complete blood count, blood chemistry, vitamin B₁₂/folate level, syphilis serology, thyroid function, and APOE genotype were tested through blood sample.

Standard protocol approvals, registrations, and patient consents.

This study was approved by the institutional review board of Samsung Medical Center. We obtained written consent from each patient.

PiB-PET acquisition and data analysis.

All patients completed an [¹¹C]PiB-PET scan at Samsung Medical Center or Asan Medical Center. All participants completed the same type of PET scan with a Discovery STe PET/CT scanner (GE Medical Systems, Milwaukee, WI). When measuring PiB retention, cerebellar gray matter was used as the reference region. We defined the PiB retention ratio as a continuous variable representing amyloid burden. The detailed radiochemistry profiles, scanning protocol, and PiB-PET data analysis are described in appendix e-1 on the Neurology® Web site at Neurology.org and in a previous study.⁵

MRI techniques.

We acquired standardized T2, 3-dimensional T1 turbo field echo, 3-dimensional FLAIR, and DTIs from all participants at Samsung Medical Center using the same 3.0T MRI scanner (Philips 3.0T Achieva; Philips Healthcare, Andover, MA). Detailed imaging parameters are described in appendix e-2. In whole-brain DTI-MRI examination, sets of axial diffusion-weighted single-shot echo-planar images were collected with the following parameters: 128 × 128 acquisition matrix; 1.72 × 1.72 × 2 mm³ voxel size; 70 axial slices; 22 × 22 cm² field of view; echo time 60 milliseconds, repetition time 7,696 milliseconds; flip angle 90°; slice gap 0 mm; b-factor of 600 s/mm². Diffusion-weighted images were acquired from 45 different directions using the baseline image without weighting (0, 0, 0). All axial sections were acquired parallel to the anterior commissure–posterior commissure line.

Measurement of WMH volume and lacunes.

We measured volume of WMH (in milliliters) on FLAIR images using an automated method.¹⁸ Detailed methods are described in appendix e-3.

A lacune was defined as a lesion ≥3 mm and ≤15 mm in diameter with low signal on T1-weighted images, high signal on T2-weighted images, and perilesional halo on FLAIR images. This meets the definition of lacune of presumed vascular origin, recently proposed by Wardlaw et al.¹⁷ Two neurologists manually counted the number of lacunes, with a κ value of 0.78.

Network node definition.

We used the automated anatomical labeling (AAL) atlas¹⁹ to parcellate the entire cerebral cortex into 78 areas (39 regions in each hemisphere) and to define the nodes of a brain graph. Individual T1-weighted images were nonlinearly registered to the ICBM152 T1 template in Montreal Neurological Institute (MNI) space.²⁰ The AAL atlas was transformed from MNI space to T1 native space using the inverse transformation with a nearest-neighbor interpolation method.

Network edge definition.

We corrected distortions in DTIs caused by eddy currents and simple head motions by using the diffusion toolbox of the FSL (FMRIB's Software Library) package (www.fmrib.ox.ac.uk/fsl/fdt). Diffusion tensor models were estimated, and fractional anisotropy and mean diffusivity were calculated at each voxel. Whole-brain WM fiber tracts in native diffusion space for each participant were reconstructed using fiber assignment by a continuous tracking algorithm²¹ that is embedded in the Diffusion Toolkit (trackvis.org).²² Tracking was terminated when the angle between 2 consecutive orientation vectors was greater than 45°, or when both ends of the fibers extended outside of the WM mask that was generated by the tissue segmentation process.^23,24 A fiber cutoff filter was applied so that fibers shorter than 20 mm and longer than 200 mm were filtered.

Using the affine registration tool from the FSL package (www.fmrib.ox.ac.uk/fsl/flirt), we coregistered T1-weighted images to the b0 images. Reconstructed whole-brain fiber tracts were inversely transformed into T1 space and fiber tracts and AAL-based parcellated regions were located in the same space. We assume that 2 nodes (regions) were structurally connected by an edge when at least the end points of 3 fiber tracts were located in these 2 regions. We selected a threshold number of fiber tracts to reduce the risk of false-positive connections due to noise or limitations of deterministic tractography.^24,25 In this study, a mean fractional anisotropy value along all the fibers connecting a pair of regions was used to weight the edge. Finally, weighted WM networks represented by symmetric 78 × 78 matrices were constructed for each individual.

Network analysis.

Graph theoretical analyses were performed on weighted connectivity networks using the Brain Connectivity Toolbox (www.brain-connectivity-toolbox.net).²⁶ We used nodal efficiency as a nodal topological characteristic. Nodal efficiency is defined using the inverse of the weighted shortest path length between a given node and all other nodes in the network.²⁶ This metric quantifies the importance of nodes for communication within the network. Nodal efficiency was measured for each node, and averaged values of the nodal efficiency (mean nodal efficiency) in the frontal and temporoparietal regions predefined in the AAL atlas were used for global analysis.

Of 251 participants, we excluded 6 patients for whom WMH volume measurement failed because of segmentation errors, and 13 for whom the quality of diffusion image (low signal-to-noise ratio) was not sufficient to reconstruct reliable fiber tracts. Thus, network analysis was performed in 232 participants.

Cortical thickness data analysis.

T1-weighted images were processed using the standard MNI anatomical pipeline. Further image processing for cortical thickness measurements is described in appendix e-4. In this way, we obtained the mean thickness of the frontal lobe and temporoparietal lobe.

Neuropsychological tests.

Participants underwent neuropsychological tests using a standardized neuropsychological battery.²⁷ Based on the neuropsychological results, we calculated memory and frontal-executive subdomain scores.²⁸ Memory-domain score (memory score) was calculated by summing scores in orientation, verbal memory, and visual memory tests. The memory score ranged from 0 to 150. Frontal-executive-domain score (executive score) was calculated by summing scores in a category word generation task, a phonemic word generation task, and the Stroop color reading test. The executive score ranged from 0 to 55.

Because of missing data in the memory score (n = 10) and executive score (n = 21), path analysis for the memory score was performed with data from 222 patients, and path analysis for the executive score was performed with data from 211 patients.

Statistical analysis.

To evaluate the association between neuroimaging markers (PiB retention ratio, WMH volume, lacune number) and nodal efficiency, multiple linear regression analysis was performed for each node. We entered age, sex, education, PiB retention ratio, WMH volume, and lacune number to find the independent effects of each imaging marker. False discovery rate correction was performed to correct for multiple comparisons in 78 node regions.

To evaluate whether alteration of the WM network mediates the effects of SVD markers and amyloid burden on cortical thickness and cognition, path analyses were performed after controlling for age, sex, and education. Path analysis was used to simultaneously consider the direct, indirect, and total effects of predictors on outcomes through mediators. Because our intention was to focus on frontal-executive and memory function, we selected nodes or cortical regions that were related to frontal-executive function (bilateral frontal lobe) when evaluating frontal-executive function, and selected nodes or cortical regions that were related to memory (bilateral temporal and parietal lobe) when evaluating memory function (appendix e-5). Path analysis using executive score as the outcome was performed using mean frontal nodal efficiency and mean frontal thickness as mediators (path analysis A). Path analysis using memory score as the outcome was performed using mean temporoparietal nodal efficiency and mean temporoparietal thickness as mediators (path analysis B). The pathobiological model of complex relationships between various factors is shown in figure e-1. We also performed subgroup analyses in patients with aMCI and AD (appendix e-6, table e-1, figure e-2). Amos version 18.0 software (SPSS, Chicago, IL) was used for all path analyses using maximum likelihood estimation.