Individual Correspondence of Amyloid-β and Intrinsic Connectivity in the Posterior Default Mode Network Across Stages of Alzheimer’s Disease

INTRODUCTION

In Alzheimer’s disease (AD), there is an intricate relationship between amyloid-β (Aβ) pathology and the brain’s ongoing activity. In particular, the deposition of plaques has been associated with the default mode network (DMN) which is characterized by synchronous infra-slow ongoing activity (i.e., intrinsic functional connectivity [iFC]) [1–6]. This association has been derived due to the striking spatial overlap between Aβ plaque accumulation and the extent of the DMN. The spatial correspondence, together with consistent DMN alterations in AD for iFC [7–9], metabolism [1, 2], and structure [10], have led to the proposal of accumulating Aβ pathology along DMN nodes, i.e., along functionally connected rather than spatially neighboring areas [6, 11–17]. However, the observed spatial overlap hints at two types of relationship: on the one hand, Aβ pathology disrupts local iFC [6, 18–23]; while on the other hand, it exhibits a more “positive” relationship— wherever iFC is high, Aβ tends to be high— suggesting a distribution of Aβ following iFC [6, 24].

Crucially, in a previous work of our group, Myers et al. [6] offered a novel approach to clarify the intricate relationship between Aβ and iFC using a within-subject spatial correlation approach. Looking at several intrinsic networks in a sample of individuals with prodromal AD, they revealed two distinct effects of Aβ on intrinsic connectivity. At the global network level, they found a positive correspondence, where Aβ aggregates in areas of high intrinsic connectivity. While at the local network level, they reported a negative correspondence, where plaques load was negatively associated with connectivity. They found significant results across heteromodal networks, with the strongest effect in the posterior DMN (pDMN). However, this work only examined the prodromal stage of AD, and further clarification is needed about the temporal progression of the within-network effects of Aβ pathology across the stages of the disease. While this previous work provides both the theoretical and methodological backbone of the current study, we aimed at extending those findings by examining the global and local spatial correspondence between Aβ load and iFC within the pDMN in individual patients across stages of AD and healthy older adults.

In the present study, we used a cross-sectional design involving multimodal imaging data of a rich data sample, including healthy Aβ negative individuals, as well as preclinical, prodromal, and clinically manifest stages of AD. Within-subject voxel-wise correlations, were used to derive measures of global and local spatial correspondence between Aβ load and iFC. Based on previous findings from Myers et al. [6], we hypothesized a positive global and a negative local spatial correspondence starting with Aβ accumulation [24, 25]. Both processes were expected to reach a plateau at prodromal stages.

MATERIALS AND METHODS

Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu). The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early AD.

Study participants gave written informed consent. The study was approved by local ethic committees in awareness of the Code of Ethics of the World Medical Association (Declaration of Helsinki). A fuller description of ADNI and up-to-date information is available at http://www.adni-info.org (http://www.loni.usc.edu/ADNI/) [26–28].

Resting-state fMRI data preprocessing

Based on previous findings showing the pDMN to be the primary network affected by both Aβ deposition and iFC reduction [5, 24], we explicitly focused our analyses on the pDMN. Preprocessed data were decomposed into 35 spatially independent components reflecting intrinsic brain networks via a group-level independent component analysis (ICA) framework [33], using the Group ICA of fMRI Toolbox (GIFT, http://www.icatb.sourceforge.net). To automatically select the pDMN, we applied multiple spatial regression analyses on the 35 independent components using publicly available network templates from a multicenter study [34]. Then, for each subject an individual independent component reflecting the pDMN was derived via back-reconstruction as implemented in GIFT. The component was composed of the network’s time course and spatial map of z-scores. Voxels with high iFC within the network have high z-scores, whereas voxels that are not part of the network have z-scores near 0. Individuallevels of pDMN iFC were derived from voxels with z-values ≥1 and with a gray matter probability ≥0.5. A z-threshold of 1 was set in order to omit negative z-values reflecting voxels negatively correlated to pDMN activity but to ensure that the whole extent of the pDMN was included in the analyses (see Supplementary Figure 1A, showing a one-sample t-test over all subject, of the thresholded pDMN used in further analyses). Group comparisons of averaged pDMN iFC z-values were performed within one ANOVA model (specified later on) and related post-hoc t-tests (p < 0.05, Supplementary Figure 1).

OPEN IN VIEWER

Fig.1

Global correspondence between Aβ load and iFC for the pDMN across the AD spectrum. A) The global correspondence between Aβ load and iFC - r_GLOBAL(Aβ, iFC): in the pDMN is positive and increased across the AD spectrum, showing that Aβ aggregates in areas of high intrinsic connectivity, and reaching a plateau in eMCI+, i.e., comparable levels of r_GLOBAL(Aβ, iFC) are found across the patient groups (ANOVA and related post-hoc t-tests, p < 0.05). B) r_GLOBAL(Aβ, iFC) correlates positively with mean iFC z-values within the pDMN (Pearson’s correlation coefficient, p < 0.05). C) r_GLOBAL(Aβ, iFC) correlates positively with mean AV-45 uptake within the pDMN (Pearson’s correlation coefficient, p < 0.05). AD, Alzheimer’s disease; AD+, Aβ positive AD; a.u., arbitrary units; pDMN, posterior default mode network; eMCI+, Aβ positive early mild cognitive impairment; lMCI+, Aβ positive late mild cognitive impairment; CN-, Aβ negative cognitively normal subjects; CN+, Aβ positive cognitively normal subjects; SUVR, standardized uptake value ratio.

Subjects’ AV-45-PET scans were rigidly coregistered to their corresponding structural MRI image and then normalized on the MNI template using the previously acquired normalization parameters. This procedure resulted in structural MRI, resting-state fMRI, and PET data with identical dimensions and in the same space. For each subject, all voxel values were converted to SUVRs by scaling to the cerebellum as reference. Additionally, images were smoothed using an 8×8×8 mm³ Gaussian kernel. In order to derive individual measures of Aβ load, individual AV-45-SUVRs were averaged across those voxels belonging to the individual pDMN mask (i.e., voxels with z-value ≥1, and gray matter probability ≥0.5). The individual pDMN masks were derived by back-reconstructing the pDMN derived from the group-level independent component analysis on resting-state fMRI data, as described in the previous section. In order to assess group differences in Aβ load, individual averaged pDMN AV-45 SUVRs were entered into the aforementioned ANOVA model and post-hoc t-tests were performed (p < 0.05, Supplementary Figure 1).

RESULTS

Positive spatial correlation between patterns of AV-45 uptake and iFC z-scores across stages of AD and healthy older adults - r_GLOBAL(Aβ, iFC)

We examined the global correspondence between Aβ load and iFC in the pDMN across groups via r_GLOBAL(Aβ, iFC). We found a main effect of group on the global spatial correlation between AV-45 uptake and iFC z-values within the pDMN (Fig. 1A, F = 4.60, p < 0.05). More specifically, r_GLOBAL(Aβ, iFC) was positive in Aβ positive subjects and reached a plateau already in eMCI+, i.e., comparable levels of r_GLOBAL(Aβ, iFC) were found between eMCI+, lMCI+, and AD+. Findings for r_GLOBAL(Aβ, iFC) were robust against atrophy patterns in the whole pDMN (Supplementary Figure 2A, B). In the Aβ positive groups, we further found a positive correlation between r_GLOBAL(Aβ, iFC) and mean iFC z-values (Fig. 1B, R = 0.57 p < 0.05). Further, we found a positive correlation between r_GLOBAL(Aβ, iFC) and mean AV-45 uptake within the pDMN (Fig. 1C, R = 0.28 p < 0.05).

OPEN IN VIEWER

Fig.2

The negative local correspondence between Aβ load and iFC within the pDMN is enhanced across the AD spectrum. A) Voxel-wise analyses comparing the local correspondence between Aβ load and pDMN iFC – (Aβ, iFC) – across groups. After accounting for regional variability in global correspondence via decorrelation, local spatial correspondence between Aβ and iFC was calculated by spatial correlations within local spheres via a searchlight approach, indicating that Aβ pathology reduces connectivity in the pDMN as a function of local plaque concentration. Aa) One-sample t-test (p < 0.05 FWE corrected) among all subject showing t-maps of r_LOCAL(Aβ, iFC) peaking within posterior connectivity centers of the pDMN, i.e., the posterior cingulate and precuneus. Ab) ANOVA analysis (p < 0.05 FWE corrected) showing a main effect of group for r_LOCAL(Aβ, iFC) in the same posterior connectivity centers described before. Ac-e) Post-hoc t-tests (p < 0.05 FWE corrected) reveal enhanced negative levels of r_LOCAL(Aβ, iFC) in patients compared to CN- within the posterior cingulate and precuneus. No group differences were detected across patient groups or when compared to CN+. B) Results using the first eigenvariate of r_LOCAL(Aβ, iFC) derived from the group effect cluster. r_LOCAL(Aβ, iFC) is negative, peaks in eMCI+, with a plateau-like course for clinical stages, i.e., comparable levels of r_LOCAL(Aβ, iFC) are found across the patient groups (ANOVA and related post-hoc t-tests, p < 0.05). C) Mean r_LOCAL(Aβ, iFC) and averaged levels of iFC z-values within connectivity centers of the pDMN show a significant negative correlation (Pearson’s correlation coefficient, p < 0.05). D) Mean r_LOCAL(Aβ, iFC) and averaged levels of AV-45 uptake within connectivity centers of the pDMN show no significant correlation (Pearson’s correlation coefficient, p < 0.05). AD, Alzheimer’s disease; AD+, Aβ positive AD; a.u., arbitrary units; pDMN, posterior default mode network; eMCI+, Aβ positive early mild cognitive impairment; lMCI+, Aβ positive late mild cognitive impairment; CN-, Aβ negative cognitively normal subjects; CN+, Aβ positive cognitively normal subjects.

DISCUSSION

OPEN IN VIEWER

Fig.3

Proposed model for the pDMN trajectory of Aβ, iFC, and both local and global correspondence between Aβ and iFC along stages of AD. For visualization purposes, one-sample t-test of the thresholded pDMN maps used in the study in bright green. In the transition phase from healthy to preclinical cognitively normal conditions, where first signs of Aβ pathology are detected (depicted in yellow in the graph and in the gradient bar below the graph), levels of Aβ load (Aβ, line in light green) begin to rise while iFC of the pDMN (iFC, line in orange) gradually decreases. In parallel to increased levels of Aβ load, the global correspondence between Aβ load and iFC (i.e., r_GLOBAL(Aβ,iFC) within the pDMN, line in dark green) begins also to rise, suggesting increased Aβ accumulation within regions of the network with higher iFC (iFC associated Aβ accumulation, gradient bar in green). Analogously, the local correspondence between Aβ load and iFC within the pDMN (i.e., r_LOCAL(Aβ, iFC), line in dark red) begins to decrease in preclinical stages of AD, indicating detrimental effects of Aβ pathology on iFC of pDMN connectivity centers (detrimental effects of Aβ on iFC, gradient bar in red). Global and local correspondence between Aβ load and iFC reach a plateau in eMCI+, in parallel with appearance of first cognitive symptoms (as dashed blue line in the graph and blue gradient bar). In a nutshell, in the pDMN Aβ accumulates in areas of higher iFC in individual patients; and simultaneously, when certain Aβ levels are reached, Aβ pathology has detrimental effects on local iFC; both mechanisms emerge already in the preclinical stages of AD when first Aβ accumulation is present, peak and stabilize at early prodromal stages with the appearance of first cognitive symptoms. Aβ, amyloid-β; AD, Alzheimer’s disease; AD+, Aβ positive Alzheimer’s disease; pDMN, posterior default mode network; eMCI+, Aβ positive early mild cognitive impairment; lMCI+, Aβ positive late mild cognitive impairment; CN-, Aβ negative cognitively normal subjects; CN+, Aβ positive cognitively normal subjects.

Across the AD spectrum, we report a positive within-subject global correspondence between Aβ distribution and iFC of the pDMN (Fig. 1). This outcome was anticipated and is in line with the well-known spatial overlap between the pDMN and Aβ deposition [1]. Critically, such global correspondence starts in preclinical stages CN+ and reached a definitive plateau in early symptomatic stages of MCI (i.e., eMCI+), demonstrating a non-linear trajectory of correlated iFC and Aβ distribution. To clarify, a within-subject positive correlation between voxel-wise iFC and voxel-wise AV45-uptake, i.e., r_GLOBAL(Aβ, iFC), indicates that voxels in the network core (i.e., voxels with high iFC) were burdened with higher Aβ deposition than the network periphery (voxels with low iFC). No significant differences were observed in the global correspondence of Aβ load and iFC between the different stages of the disease (eMCI+, lMCI+, AD+). This data suggests that the deposition of Aβ in areas of high iFC within the pDMN already starts in the earliest clinical stages of AD while reaching a plateau as the disease progresses. This finding is consistent with previous studies showing extensive Aβ deposition within regions of the pDMN at early disease stages, which slows down as the disease advances and saturated levels of Aβ are reached [36–38]. Critically, increased r_GLOBAL(Aβ, iFC) within the pDMN correlated positively with levels of global iFC and levels of global Aβ within the whole pDMN in Aβ positive subjects. This result suggests a mechanistic process, in which available levels of iFC within the pDMN may determine Aβ deposition in areas of high iFC, and finally global levels of Aβ load. We examined additionally the within-subject local correspondence between Aβ load and iFC of the pDMN across groups (Fig. 2), after orthogonalization of Aβ load and iFC at the global level, in order to account for the global correspondence as implemented in Myers et al. [6]. We found a significant negative local correlation between Aβ load and iFC particularly in patients’ precuneus and posterior cingulate, which are primarily affected by Aβ load [2, 24]. To further elucidate, a locally delimited negative correlation between voxel-wise iFC and voxel-wise AV45-uptake within connectivity centers of the pDMN, i.e., r_LOCAL(Aβ, iFC), suggests that Aβ pathology adversely affects intrinsic connectivity within connectivity centers of the pDMN as a function of local Aβ-plaque concentration. Critically, the local negative correspondence between Aβ load and iFC was enhanced along the AD spectrum, starting already at preclinical stages (CN+) and reaching a peak in eMCI+. This finding suggests that the negative effects of Aβ load on iFC are located particularly within centers of the pDMN, start in preclinical stages of CN+, and peak with appearance of first cognitivesymptoms. Extending the work of Myers and colleagues using the same methods, but restricted to a sample with early MCI [6], we found a negative association between mean r_LOCAL(Aβ, iFC) and mean levels of iFC within connectivity centers of the pDMN across subjects of the AD spectrum. These findings indicate that negative Aβ effects on iFC are the stronger, the higher individual levels of iFC. In contrast, no significant correlation was found between mean r_LOCAL(Aβ, iFC) and mean levels of network Aβ load, suggesting threshold effects. These findings suggest that detrimental effects of Aβ load on iFC are stable from eMCI+ on, and determined rather by the strength of individual iFC levels than from Aβ levels.

Our results provide insights about the inter-relatedness of Aβ, iFC, and their global and local correspondence within the pDMN across stages of AD. We have summarized our findings as a trajectory model for the pDMN in Fig. 3, proposing a temporal order of changes in Aβ and r_GLOBAL(Aβ, iFC), as well as iFC and r_LOCAL(Aβ, iFC) across the AD spectrum. In particular, we propose an interaction model of distinct global and local processes between Aβ and iFC, which start at the preclinical stage. In the preclinical stage, where the first signs of Aβ pathology are detectable in the absence of symptoms, Aβ accumulation increases, reaching a plateau in early clinical disease stages. In parallel to the rise of Aβ accumulation, the global correspondence between Aβ and iFC, r_GLOBAL(Aβ, iFC), begins also to rise, suggesting that iFC of the pDMN determines the accumulation levels of Aβ within the network, reaching a plateau in early disease stages.

Network-framed Aβ accumulation is accompanied by local negative effects of Aβ pathology on iFC in network centers, where both iFC and Aβ accumulation have their peaks [39]. Such detrimental local processes take place already in preclinical stages and reach a peak in eMCI+ with the appearance of first cognitive symptoms. The detrimental effect of Aβ on iFC is accompanied by progressive iFC reduction of the pDMN [7–9]. The detrimental Aβ effect is constant when certain levels of Aβ are reached, but it is stronger when individual iFC levels are higher (Fig. 2C). With further iFC reductions along disease progression, the detrimental effect of Aβ on iFC is slightly weakened. Because of progressive iFC decreases across the disease trajectory, we suggest that other factors such as “nodal stress”, in which hub regions undergo activity-related degradation, neuroinflammatory and vascular processes, or increasing tau-based neurodegeneration might become more relevant for iFC decreases, and as a consequence network dysfunction, than constant Aβ effects[4, 41].

In summary, our model is in line with previous hypotheses stating that propagating Aβ accumulation follows a network-based pattern [6, 17]. However, it proposes that when accounting for the global effect of iFC on Aβ pathology accumulation, simultaneous local detrimental effects of Aβ load on iFC start in centers of higher iFC and pathology levels. In other words, wherever Aβ accumulation exceeds a specific threshold, accumulated pathology exerts a stable long-lasting negative influence on local iFC. In our model, both Aβ pathology accumulation and its detrimental effects on iFC start already at presymptomatic stages of AD, reaching a peak and stabilizing with the emergence of first cognitive symptoms at eMCI+. This starting point highlights the need for an effective characterization and further stratification of preclinical stages, where therapeutic interventions could have the most promising effect.

Several limitations need to be considered. First, though our model implicates trajectory of changes, our data is cross-sectional. Longitudinal studies are necessary to test the course of correspondence between Aβ load and iFC, with a special focus on non-linear relationships between iFC and Aβ along disease progression, which we did not account for in the current study. Second, based on previous work of our group finding the strongest effect in the pDMN, our study strictly focuses on this network, neglecting that other DMN subsystems [4] and heteromodal networks [5] are affected by Aβ pathology and iFC changes across stages of AD. A previous study from our group reported similar effects of global and local correspondence between Aβ and iFC in heteromodal networks of patients with MCI, though the effects were of lesser magnitude [6]. We hypothesize that similar trajectories may take place in distinct heteromodal networks but with a delayed onset in respect to the pDMN, following a cross-network gradient [6]. However, recent literature has revealed patterns of non-linearly progressing changes of iFC in AD, most notably within some subsystems of the DMN [5]. Moreover, increases in iFC have been reported both at the intrinsic brain network level, e.g., within the Salience Network [42], and at the regional level, e.g., increased iFC locally confined within the MTL [43, 44]. In order to assess general conclusions on the relationship between Aβ and iFC across stages of AD, our findings need to be extended to other intrinsic brain networks, and need to be completed by the use of complementary iFC measures, such as inter-network functional connectivity and degree centrality. Finally, our study is focused on the correspondence between Aβ pathology and iFC, ignoring that tau-based pathology and neurodegeneration might critically modulate such correspondence [4, 41]. Future studies might include measures reflecting tau-based neurodegeneration in order to draw a more detailed picture.