Neocortical origin and progression of gray matter atrophy in nonamnestic Alzheimer's disease

Abstract

Amnestic Alzheimer's disease (AD) is characterized by early atrophy of the hippocampus and medial temporal lobes before spreading to the neocortex. In contrast, nonamnestic Alzheimer's patients have relative sparing of the hippocampus, but the pattern in which the disease spreads is unclear. We examined spreading disease in nonamnestic AD using a novel magnetic resonance imaging-based analysis adapted from pathologic staging studies, applied here to cross-sectional imaging data. We selected 240 T1-weighted scans from 129 patients with pathology confirmed by autopsy or cerebrospinal fluid, and atrophy maps were computed relative to 238 scans from 115 elderly controls. For each phenotype, the frequency of atrophy in 116 brain regions was used to infer the anatomical origin of disease and its progression across 4 phases of atrophy. Results from the amnestic cohort were used to determine appropriate parameter settings for the phase assignment algorithm, based on correspondence to Braak pathology staging. Phase 1 regions, which represent the origin of disease, included the hippocampus for the amnestic group (comprising 33 scans); left lateral temporal lobe for logopenic-variant primary progressive aphasia (88 scans); occipitoparietal cortex for posterior cortical atrophy (51 scans); temporoparietal cortex for corticobasal syndrome (31 scans); and frontotemporal cortex for behavioral/dysexecutive variant AD (37 scans). In nonamnestic patients, atrophy spread to other neocortical areas in later phases, but the hippocampus exhibited only late-phase atrophy in posterior cortical atrophy and corticobasal syndrome. Region-specific phase values were also associated with regional measures of tau, beta amyloid, neuronal loss, and gliosis for the subset of patients (n = 17) with neuropathology findings; this comparison represented a first validation of the phase assignment algorithm. We subsequently assigned a phase to each patient scan based on the similarity of regional atrophy patterns with atrophy predicted for the corresponding phenotype at each phase. Scan-specific phases were correlated with disease duration as well as global and domain-specific cognition, supporting these phase values as global estimates of patients' disease progression. Logistic regression models based on spatial overlap with model-predicted atrophy patterns reliably discriminated nonamnestic phenotypes from each other and from amnestic AD. The frequency-based phase assignment algorithm used in the present study thus represents a promising approach for studying the neocortical origin and spread of disease in nonamnestic AD.

Keywords: Alzheimer's disease; Atrophy; Braak staging; Corticobasal syndrome; Disease progression; Frontalvariant Alzheimer's disease; Hippocampus; Logopenic-variant primary progressive aphasia; Magnetic resonance imaging; Non-amnestic Alzheimer's disease; Posterior cortical atrophy; tau.

Figure 1

Frequency-based MRI phase assignment algorithm. Within each patient group, ROIs were ranked by frequency of atrophy. Phase 1 comprised ROIs with an atrophy frequency of 90–100 percent of the maximum frequency; Phase 2 represented ROIs atrophied in 80–90 percent of the maximum; Phase 3 included ROIs atrophied in 70–80 percent of the maximum; and Phase 4 included ROIs atrophied in 60–70 percent of the maximum. ROIs with less frequent atrophy constituted a hypothetical Phase 5, i.e., they were assumed to be atrophic only in late disease stages. Each scan was subsequently assigned the highest phase value for which it exhibited atrophy in at least 75 percent of the corresponding ROIs. Scans that did not meet criteria for Phase 1 are classified as “Phase 0” and assumed to reflect very mild disease progression.

Figure 2

Scatterplot of ROI phases for the optimal aAD model vs. the stage in which tau neurofibrillary tangle pathology first appears, according to the Braak model. Numbers within each black circle indicate the number of data points overplotted in that circle. The blue line is the best-fit regression line. The optimal aAD model used a Z-score threshold of −1 to determine atrophy and a frequency interval of 10% to distinguish ROI phases. ROI phase was correlated with Braak stage (ϱ=0.35, p<0.001).

Figure 3

Results of the MRI phase assignment algorithm for each phenotype indicate a continuous anatomical progression of atrophy from early- to late-phase regions. Phase 1 (brown): putative anatomical origins of atrophy. Phase 2: red; Phase 3: orange; Phase 4: yellow. Left: cortical surface views of atrophic regions. Right: coronal slices through the hippocampus and precuneus. Hypothetical Phase 5 regions, comprising the remainder of the brain, are left uncolored.

Figure 4

Associations between ROI phase and regional micropathology ratings. Top left: tau neurofibrillary tangle (NFT) burden; top right: β-amyloid burden; bottom left: neuronal loss; and bottom right: gliosis burden. Each data point represents one of 11 brain areas in one of 17 naAD patients. Numbers within each circle indicate the number of observations overplotted in that circle. The dashed black line in each plot is plotted from the slope and intercept terms of the corresponding linear mixed effects model.

Figure 5

Associations between MRI phase and (a) disease duration or (b–i) neuropsychological performance. Each data point represents a single observation associated with a single scan. All phenotypes are plotted together; the x-axis indicates the MRI phase associated with each scan, as a measure of disease progression. Each plot title includes the regression coefficient for the association with MRI phase, the number of observations in each analysis, and the p-value for the MRI phase regression coefficient.

Figure 6

Phenotype discriminability based on disease progression models. Each panel represents a logistic regression model trained to discriminate between two AD phenotypes. Each point represents a single scan from a patient with one of the two clinical phenotypes. The x- and y-axes represent spatial overlap (expressed by Sorensen-Dice coefficients) between the binary atrophy map for a single scan and the model-predicted atrophy map for the appropriate phenotype and MRI phase. Sen: sensitivity, indicating proportion of scans correctly identified for the phenotype on the y-axis; Spec: specificity, indicating scans correctly identified for the phenotype on the x-axis.