Alzheimer's disease: The influence of age on clinical heterogeneity through the human brain connectome

Abstract

Introduction: One major factor that influences the heterogeneity of Alzheimer's disease (AD) is age: younger AD patients more frequently exhibit atypical forms of AD. We propose that this age-related heterogeneity can be understood better by considering age-related differences in atrophy in the context of large-scale brain networks subserving cognitive functions that contribute to memory.

Methods: We examined data from 75 patients with mild AD dementia from Alzheimer's Disease Neuroimaging Initiative. These individuals were chosen because they have cerebrospinal fluid amyloid and p-tau levels in the range suggesting the presence of AD neuropathology, and because they were either younger than age 65 years early-onset AD (EOAD) or age 80 years or older late-onset AD (LOAD).

Results: In the EOAD group, the most prominent atrophy was present in the posterior cingulate cortex, whereas in the LOAD group, atrophy was most prominent in the medial temporal lobe. Structural covariance analysis showed that the magnitude of atrophy in these epicenters is strongly correlated with a distributed atrophy pattern similar to distinct intrinsic connectivity networks in the healthy brain. An examination of memory performance in EOAD dementia versus LOAD dementia demonstrated relatively more prominent impairment in encoding in the EOAD group than in the LOAD group, with similar performance in memory storage in LOAD and EOAD but greater impairment in semantic memory in LOAD than in EOAD.

Discussion: The observations provide novel insights about age as a major factor contributing to the heterogeneity in the topography of AD-related cortical atrophy.

Keywords: Age; Alzheimer's disease; Cortical thickness; MRI; Memory.

Fig. 1

Cortical atrophy maps of (A) early-onset Alzheimer's disease versus age-matched controls and (B) late-onset Alzheimer's disease versus age-matched controls. Maps depict regions in which the thickness of the cortex in AD patients is at least 0.3 mm thinner than controls, and all are statistically significant (P < .01). Maps are shown on a partially inflated average cortical surface template from 40 individuals (fsaverage). Abbreviation: AD, Alzheimer's disease.

Fig. 2

Epicenters of cortical atrophy. (A) The posterior cingulate cortex/ventral precuneus is the epicenter of cortical atrophy in early-onset Alzheimer's disease versus age-matched controls (P < 1 × 10⁻⁷). (B) The anterior medial temporal (entorhinal and perirhinal) cortex is the epicenter of cortical atrophy in late-onset Alzheimer's disease versus age-matched controls (P < 1 × 10⁻¹¹).

Fig. 3

Structural covariance analyses of each of the two epicenters reveal two largely distinct distributed patterns of atrophy. (A) Regions where cortical atrophy is correlated with PCC atrophy include bilateral PCC, the inferior parietal lobule, posterior lateral temporal cortex, and lateral and medial superior frontal gyrus. (B) Regions where cortical atrophy is correlated with aMTL atrophy include entorhinal and perirhinal cortex, temporal pole, and anterior lateral and ventral temporal cortex, as well as subgenual cingulate cortex. (C) An overlap map comparing the EOAD-epicenter structural covariance map to the LOAD-epicenter structural covariance map clarifies these as two nearly independent maps, with minimal overlap primarily in the caudal lateral temporal cortex, with smaller regions of overlap in the angular gyrus and posterior cingulate and retrosplenial cortex. Abbreviation: PCC, posterior cingulate cortex.

Fig. 4

The overlap of this spatial pattern of atrophy and the localization of large-scale brain networks clearly demonstrate the involvement of multiple networks at the mild clinical stage of AD dementia, with a pattern that differs based on age of onset. Compared with the seven-network parcellation by Yeo et al. (2011) (B), the EOAD-epicenter structural covariance map (A) largely overlaps with DMN and frontoparietal executive control networks, whereas the LOAD-epicenter structural covariance map largely overlaps with the anterior temporal/paralimbic network. Compared with the Yeo et al. (2011) 17-network parcellation (C), the EOAD-epicenter structural covariance map overlaps with dorsal DMN and language networks, whereas the LOAD-epicenter structural covariance map overlaps with the anterior temporal/paralimbic network. Thus, the overlap of AD-related atrophy and large-scale brain networks depends both on the characteristics of AD patients (with age being one major factor) and the scale at which brain networks are examined. Abbreviations: AD, Alzheimer's disease; DMN, default mode network.

Fig. 5

Comparison of cortical thickness of epicenters of cortical atrophy. Bar graphs depict thickness of (A) precuneus which is the epicenter of EOAD and (B) anterior medial temporal cortex which is the epicenter of LOAD, compared to age-matched younger and older control groups of cognitively normal, amyloid-negative individuals. Abbreviation: aMTL, anterior medial temporal lobe.

Fig. 6

The EOAD-epicenter and LOAD-epicenter structural covariance overlap map (B) shows a topography that is very similar to the original AD signature (A), which compared patients with mild AD dementia (n = 73) to age-matched controls (n = 194). The original AD signature included patients with a relatively wide age range, although most were late onset. Major regions include anterior MTL and ventral temporal cortex, temporal pole, lateral temporal cortex, inferior parietal lobule, PCC and precuneus, and medial and lateral superior frontal gyrus. Abbreviations: AD, Alzheimer's disease; MTL, medial temporal lobe; PCC, posterior cingulate cortex.

Fig. 7

Patients with EOAD are more impaired than those with LOAD on Rey Auditory Verbal Learning Test trial-1 encoding and total learning across all trials. The groups are similarly impaired on delayed retention and recognition discriminability, although the LOAD may be very slightly more impaired on these measures. In contrast, patients with LOAD are more impaired than those with EOAD on semantic memory measured using the Boston Naming Test.