Dynamics of gray matter loss in Alzheimer's disease

Abstract

We detected and mapped a dynamically spreading wave of gray matter loss in the brains of patients with Alzheimer's disease (AD). The loss pattern was visualized in four dimensions as it spread over time from temporal and limbic cortices into frontal and occipital brain regions, sparing sensorimotor cortices. The shifting deficits were asymmetric (left hemisphere > right hemisphere) and correlated with progressively declining cognitive status (p < 0.0006). Novel brain mapping methods allowed us to visualize dynamic patterns of atrophy in 52 high-resolution magnetic resonance image scans of 12 patients with AD (age 68.4 +/- 1.9 years) and 14 elderly matched controls (age 71.4 +/- 0.9 years) scanned longitudinally (two scans; interscan interval 2.1 +/- 0.4 years). A cortical pattern matching technique encoded changes in brain shape and tissue distribution across subjects and time. Cortical atrophy occurred in a well defined sequence as the disease progressed, mirroring the sequence of neurofibrillary tangle accumulation observed in cross sections at autopsy. Advancing deficits were visualized as dynamic maps that change over time. Frontal regions, spared early in the disease, showed pervasive deficits later (>15% loss). The maps distinguished different phases of AD and differentiated AD from normal aging. Local gray matter loss rates (5.3 +/- 2.3% per year in AD v 0.9 +/- 0.9% per year in controls) were faster in the left hemisphere (p < 0.029) than the right. Transient barriers to disease progression appeared at limbic/frontal boundaries. This degenerative sequence, observed in vivo as it developed, provides the first quantitative, dynamic visualization of cortical atrophic rates in normal elderly populations and in those with dementia.

Fig. 1.

. Creating 3D average cortical models and maps in populations of elderly people and those with AD: cortical flattening. Using a cortical flattening process (a–f) and sulcal matching techniques (Fig. 2), an average model of the cortex (Fig. 2f) can be built for a group of subjects. The goal of this process is to allow data (such as gray matter volumes) to be averaged from corresponding regions of cortex across subjects, reinforcing features that occur consistently. Briefly, the individual MRI scan (a,gray) is processed to split it up into gray matter (green), white matter (red), and cerebrospinal fluid (blue). A 3D cortical surface model (a) is extracted from the scan, and the following sulci are traced as 3D curves directly on this surface model.b, c, Superior and inferior frontal (SFS, IFS), precentral and postcentral (preCENT, poCENT), central (CENT), intraparietal (IP), superior temporal (STS), Sylvian fissures (SF), paracentral (paCENT), cingulate (CING) and paracingulate (paCING), subparietal (subP), callosal (CC), superior and inferior rostral (SRS, IRS), parieto-occipital (PAOC), and anterior and posterior calcarine (CALCa/CALCp) sulci. Because the surface is made up of discrete triangular tiles (d), a process of geometrical flattening can be applied to lay out the cortical regions, and the sulcal curves that delimit them, as features in 2D (e). Information on where these cortical points originally came from in 3D can still be saved in this 2D image format. Using a color-coding system, cortical point 3D locations (x,y, z) are given unique colors (with intensities of red, green, andblue proportional to x, y, and z, respectively), and these colors are plotted into the flat map. These color images represent the cortical shape and are used in Figure 2 to compute information on cortical pattern differences across subjects and to make an “average shape” cortex for a group of subjects.

Fig. 2.

. Creating 3D average cortical models and maps in elderly populations and those with AD: sulcal matching. The idea behind sulcal matching is to average cortical data from corresponding regions across subjects, accommodating sulcal pattern differences across subjects using an elastic warping process. Briefly, the sulcal curves from all the subjects in the study are flattened, and their shapes are averaged across subjects to make an average set of sulcal curves (a). The sulcal pattern of each individual, as seen in the individual's flattened cortical map (a), differs a little from this average set of curves. A 2D elastic deformation can be applied to an individual's flat map, which drives its features into exact correspondence with the average set of sulcal curves (b). This same deformation can be applied to the color-coded image (c) that stores 3D cortical positions from that individual (see Fig. 1 for an explanation). Images such asc or d can be averaged, pixel by pixel, across all subjects in a group and then decoded to produce a 3D shape. If this is performed before sulcal matching on images such asc, a smooth cortex results (e). It is intriguing that if it is performed on warped color images such asd, a crisp average cortex results, which reinforces group features in their mean anatomic locations (f). This process can create average cortical models for a group of subjects, but it can also transfer cortical data (such as gray matter density information) from many subjects onto a common cortical surface for comparison. In doing so, it accommodates complex differences in cortical patterning across subjects.

Fig. 3.

. Image processing steps applied to individual scans in this study. This flow chart illustrates the key steps used to process the MRI brain scans in this study. They are illustrated here on example brain MRI data sets from a healthy control subject (left) and from a patient with AD (right). First, the MRI images (stage 1) have extracerebral tissues deleted from the scans, and the individual pixels are classified as gray matter, white matter, or CSF (shown ingreen, red, and blue; stage 2). After flattening a 3D geometric model of the cortex (stage 3), features such as the central sulcus (light blue curve) and cingulate sulcus (green curve) may be reidentified. An elastic warp is applied (stage 4), thereby moving these features and entire gyral regions (pink) into the same reference position in flat space. After aligning sulcal patterns from all individual subjects, group comparisons can be made at each 2D pixel (yellow cross-hairs) that effectively compare gray matter measures across corresponding cortical regions. In this study, the cortical measure that is compared across groups and over time is the amount of gray matter (stage 2) lying within 15 mm of each cortical point. The results of these statistical tests can then be plotted back onto an average 3D cortical model made for the group (Fig. 1), and the findings can be visualized as a color-coded map (Figs. 4-8).

Fig. 4.

. Average gray matter loss rates in healthy aging and AD. The maps show the average local rates of loss for gray matter, in groups of controls (top,a–d) and patients with AD (bottom, e–h). Loss rates are <1% per year in controls. They are significantly higher in AD and strongest in frontal and temporal regions (g, h) at this stage of AD (as the MMSE score falls from 18 to 13).

Fig. 5.

Mapping links between cognitive performance and changing brain structure. These maps show the significance of the linkage between gray matter reductions and cognition, as measured by MMSE score. Variations in temporal, parietal, and ultimately frontal (e) tissue are linked with cognitive status. Less gray matter is strongly correlated with worse cognitive performance, in all regions with prominent deficits. Linkages are detected most strongly in the left hemisphere medial temporoparietal zones (d). As expected, no linkages are found with sensorimotor gray matter variation (b), which was not in significant deficit in late AD.

Fig. 6.

Mapping early and late deficits in AD. Deficits occurring during the development of AD are detected by comparing average profiles of gray matter between patients and controls at their first scan (mean MMSE = 18; top) and their follow-up scan 1.5 years later (mean MMSE = 13;bottom). The average percentage loss in patients is shown in the right four panels, and the significance of this loss is shown in the left four panels. Although severe temporal lobe loss (T) and parietal loss have already occurred at baseline (top) and subsequently continue (Fig. 4), the frontal deficits characteristic of late AD are not found until significant cognitive decline has occurred (bottom). A process of fast attrition occurs over the 1.5 years after the baseline scan. Note the relative sparing of sensorimotor cortices at both disease stages (S/M). Regionally significant effects are codedred and assessed by permutation, which corrects for multiple comparisons.

Fig. 7.

. Gray matter asymmetry in healthy controls. The asymmetric loss of gray matter in AD (left hemisphere faster than right) compounds an existing asymmetry in the distribution of gray matter, observed here in healthy controls (a,b). Here, control subjects show an average 15–20% less gray matter in sensorimotor regions (a), which is highly significant (c). These maps are computed by comparing the average gray matter maps on the leftand the right and taking their ratio (L/R). Comparisons are made after adjusting for both cortical pattern differences across subjects and gyral pattern asymmetries, using cortical pattern matching (Fig.2).

Fig. 8.

. Asymmetric deficits progressing across the medial wall in AD. Medial wall deficits in AD in the right and left hemispheres (top, a,b). Colors show the average percentage loss of gray matter relative to the control average. Profound loss engulfs the left medial wall (>15%; b,d). On the right, however, the deficits in temporoparietal and entorhinal territory (a) spread forward into the cingulate 1.5 years later (c), after a five point drop in average MMSE score. Note the prominent division between limbic and frontal zones, with different degrees of impairment (c). The corpus callosum is indicated in white; maps of gray matter change are not defined here, because it is a white matter commissure.