Aging and Alzheimer’s disease have dissociable effects on local and regional medial temporal lobe connectivity

Abstract Functional disruption of the medial temporal lobe-dependent networks is thought to underlie episodic memory deficits in aging and Alzheimer’s disease. Previous studies revealed that the anterior medial temporal lobe is more vulnerable to pathological and neurodegenerative processes in Alzheimer’s disease. In contrast, cognitive and structural imaging literature indicates posterior, as opposed to anterior, medial temporal lobe vulnerability in normal aging. However, the extent to which Alzheimer’s and aging-related pathological processes relate to functional disruption of the medial temporal lobe-dependent brain networks is poorly understood. To address this knowledge gap, we examined functional connectivity alterations in the medial temporal lobe and its immediate functional neighbourhood—the Anterior-Temporal and Posterior-Medial brain networks—in normal agers, individuals with preclinical Alzheimer’s disease and patients with Mild Cognitive Impairment or mild dementia due to Alzheimer’s disease. In the Anterior-Temporal network and in the perirhinal cortex, in particular, we observed an inverted ‘U-shaped’ relationship between functional connectivity and Alzheimer’s stage. According to our results, the preclinical phase of Alzheimer’s disease is characterized by increased functional connectivity between the perirhinal cortex and other regions of the medial temporal lobe, as well as between the anterior medial temporal lobe and its one-hop neighbours in the Anterior-Temporal system. This effect is no longer present in symptomatic Alzheimer’s disease. Instead, patients with symptomatic Alzheimer’s disease displayed reduced hippocampal connectivity within the medial temporal lobe as well as hypoconnectivity within the Posterior-Medial system. For normal aging, our results led to three main conclusions: (i) intra-network connectivity of both the Anterior-Temporal and Posterior-Medial networks declines with age; (ii) the anterior and posterior segments of the medial temporal lobe become increasingly decoupled from each other with advancing age; and (iii) the posterior subregions of the medial temporal lobe, especially the parahippocampal cortex, are more vulnerable to age-associated loss of function than their anterior counterparts. Together, the current results highlight evolving medial temporal lobe dysfunction in Alzheimer’s disease and indicate different neurobiological mechanisms of the medial temporal lobe network disruption in aging versus Alzheimer’s disease.


ICA-Based Denoising Protocol
A single rater (SH) reviewed all ICA-AROMA classifications and corrected classification inaccuracies.Spatial maps, time courses, and power spectra of every component from every subject were inspected.Scanner noise components were identified by two criteria: (1) majority of spatial activation outside the gray matter, and (2) distinct power spectrum pattern, dominated by high-frequency spikes -generally above 0.11 Hz -with little to no power represented by lower frequencies (i.e., < 0.10 Hz).Cardiovascular and respiratory noise sources were identified based on the guidelines detailed in Griffanti et al. (2017).Head-movementrelated components were largely based on the ICA-AROMA classifications (Pruim et al., 2015) and were generally characterized by slow drifts or sharp signal spikes.Only unambiguous noise components were marked for removal.SH has extensive experience with manual ICA-based denoising and attains approximately 90% intra-rater classification consistency after a 2-week delay (Hrybouski et al., 2021).The dominant head motion artifacts (e.g., global signal drifts with spatial maps localized exclusively to the skull) were removed using the 'aggressive' denoising option in fsl_regfilt, while all other artifacts were removed using the 'soft' denoising option in fsl_regfilt (Beckmann & Smith, 2004;Griffanti et al., 2014).Schaeffer ROIs were excluded because of substantial (>15%) spatial overlap with our anterior or posterior tau-based MTL seeds.To ensure that we did not miss any major cortical regions with FC to the MTL, we also performed seed-to-voxel network identification.Here, one-sample permutation tests (5,000 permutations) for positive connectivity to the bilateral anterior and posterior tau-based MTL ROIs were performed on Fisher-transformed subject-level voxelwise seed-to-voxel connectivity maps (Conn 20.b;Whitfield-Gabrieli & Nieto-Castanon, 2012).The Threshold-Free Cluster Enhancement (TFCE) method with the FDR (q < .05)correction for multiple hypothesis testing was used in these voxelwise tests (Benjamini & Hochberg, 1995;Smith & Nichols, 2009).We considered a given Schaeffer ROI as a part of the broader MTLassociated functional system if it was functionally connected to at least one of the MTL ROIs in the ROI-to-ROI network identification or if more than 40% of that ROI's voxels corresponded to a statistically significant cluster in the voxelwise network identification method.In total, we identified 221 Schaeffer ROIs with positive functional connectivity to the MTL.Together with 4 seed regions, these 221 Schaeffer ROIs (225 ROIs in total) were used in all subsequent analyses of the MTL network function (Fig. 2a in the main text).

ROI-based
Estimation of Network Architecture Using Graphical SCAD First, bivariate correlation matrices were computed for all participants in a group.For normal agers, those correlation matrices were then Fisher-transformed and averaged such that each decade of human lifespan had equal weight on the final connectivity structure.For other groups, simple averaging across all participants was performed after applying Fisher transformations.The resulting average Z-connectivity matrices were converted into the correlational connectivity matrix and used as covariance sources in SCAD-based network estimation.Graphical SCAD optimization relies on two tuning parameters: α and ρ.To minimize the Bayes risk, Fan and Li (2001) recommend α = 3.7.The second tuning parameter, ρ, was selected from a set of ρ = {e -8.0 , e -7.8 , e -7.6 , … , e 0 } by minimizing the Bayesian Information Criterion (Fan et al., 2009;Zhu & Cribben, 2018).Custom MATLAB scripts employing the QUIC optimizer were used to solve the graphical SCAD problem (Hsieh et al., 2014).
Representation of the Extended MTL Network Extra-MTL ROIs with positive functional connectivity (FC) to the MTL were selected from the 400-region 17-Network Schaefer et al. (2018) parcellation.Cortical regions with connectivity to the MTL were identified in normal agers only [i.e., CU young, middle-aged, and Ab− older participants].To test for the presence of FC to the MTL, we performed 4 sets (one per each MTL ROI: left anterior, right anterior, left posterior, right posterior) of one-sample positive-sided ttests [FDR-corrected, q < 0.05] on Fisher-transformed subject-level Pearson correlation coefficients, representing that segment's FC to each of the 393 non-MTL Schaeffer ROIs.Seven

Table 2 .
MTL voxelwise temporal Signal-to-Noise Ratio (tSNR) for raw and preprocessed fMRI datasets, separated by group.

Table 3 .
Framewise displacement (FD) for raw and filtered realignment parameters, separated by group.