Functional Brain Network Subtypes Are Associated with Alzheimer's Disease Biomarkers in an Aging, High-Risk, Cognitively Normal Cohort: the Prevent-Alzheimer's Disease Study

Alzheimer's dementia (AD) is the end-stage of a chronic illness characterized by neurodegenerative changes long before onset of symptoms. Identification of pre-symptomatic AD biomarkers is hampered by heterogeneity in the aging population and by the incomplete understanding of the cascade of pathophysiological events over the course of disease progression. In 170 members of the PREVENT-AD cohort of aging cognitively intact persons (63.9 ± 5.6 yrs; 126 females) with a first-degree family history of AD, we sought to identify variants in functional brain network organization and to evaluate their association with known AD biomarkers. Resting-state fMRI was used to identify 6 networks putatively affected in AD (Figure 1). A clustering analysis then isolated 5 subtypes of subjects for each network, which shared a combination of localized decrease or increase in network stability when compared to the population average (Figure 2). We finally tested if the similarity between an individual network and a particular subtype would be associated with several AD biomarkers: APOE-ε4 status (n = 161); hippocampal volume (n = 81); CSF Aβ1-42, total tau and apoE protein (n = 59). (a) Twelve functional brain networks were identified at the group level on an independent data set (http://dx.doi.org/10.6084/m9.figshare.1285615). (b) Six of those networks were selected a priori from the literature as being potentially associated with AD: the attentional (att), posterior cingulate cortex (pcc), medial prefrontal cortex (mpfc), motor (mot), language (lang) and limbic (limb) networks. Individual stability maps (not shown) give the probability that a given voxel has maximal correlation with the average signal of the associated network, as compared to all other 11 networks, across many sliding time windows of 1 mn extracted from the 10 mns of resting-state time series. The group average of all 170 individual maps are represented here, with stability values larger than 0.3. Identification of distinct subtypes of resting-state networks, (a) Illustration with the medial prefrontal cortex (mpfc) network as prior, (b) The grand mean stability map is shown at a threshold of 0.1. (c) The mean stability maps are shown for each subtype of the mpfc network, (d) The difference stability maps show how the subtypes maps depart from the grand mean map of the mpfc network, (e) and (f) Such subtypes of networks are associated with subgroups of subjects defined by a clustering procedure applied on a similarity matrix, (g) A test-retest analysis indicated good with in-session test-retest reliability of the similarity of individual networks with subtype averages (intra-class correlation, ICC, in the range 0.5-0.7). The similarity of individual networks with subtype averages was found to have good (0.5-0.7) test-retest reliability (Figure 2). Several associations were found between specific network subtypes and markers of AD risk and pre-symptomatic disease progression (Figure 3). Among these, 2 associations were detected at a conservative q<0.05 (FDR correction over networks and subtypes) and another 6 at an exploratory q<0.2. APOE status was associated with the motor network, with ε4 carriers showing decreased stability in motor areas. Presence of ε4 was also associated with an attentional network with decreased stability in fronto-parietal but increased stability in medial prefrontal areas. Lower hippocampal volume was associated with a posterior cingulate network less stable in the posterior cingulate and more stable in the dorsolateral prefrontal cortex. Finally, increased CSF Aβ1-42 was associated with a posterior cingulate network showing increased stability in superior parietal regions. (a) Three known AD biomarkers (APOE-E4 status. CSF Ab1-42 level and right hippocampal volume) were associated with distinct subtypes of either the motor (mot), attentional (att) or posterior cingulate (pcc) networks. Difference stability maps reveal how subtypes differ from the grand mean maps, (b) Two associations were robust after FDR correction across networks and subtypes (q<0.05). Six associations were identified for further exploration at q<0.2. A similar analysis on an independent retest run in the same subjects showed trends (p<0.1) for 5 out of the 8 associations. Subtypes of functional brain network organization can be identified reliably and may prove to be informative biomarkers in pre-symptomatic AD.