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.

Hrybouski et al. report an inverted ‘U-shaped’ functional connectivity pattern in Alzheimer’s disease, characterized by elevated connectivity levels during the preclinical stage of the disease. This effect was centered around areas of the anterior medial temporal lobe from which tau pathology originates and spreads.

Graphical Abstract

Graphical abstract.

Introduction

Structures of the medial temporal lobe (MTL) have attracted extensive scientific inquiry for over 50 years because of this region’s role not only in episodic memory but also in the processing of emotional content, social cognition, creative thought, short-term memory and object discrimination.^1–10 Both human and animal studies implicate the MTL as the key region involved in memory decline in aging and Alzheimer’s disease.^11–14

Core memory-associated MTL subregions are the hippocampus (HP) and the perirhinal (PRC), entorhinal (ERC) and parahippocampal (PHC) cortices.^6,7,13 Although unresolved, there is considerable evidence that the functional organization of the MTL can be divided along the anterior-posterior axis.^6,7,15 Further, a growing body of clinical literature indicates the differential vulnerability of the anterior versus posterior MTL to tissue atrophy and pathological processes.^16–20 The anterior MTL consists of the anterior HP (primarily head), PRC and ERC, while the posterior MTL consists of the posterior (body + tail) HP and PHC. Because the MTL is one of the most densely connected regions in the brain,²¹ MTL dysfunction is likely to propagate to extra-MTL brain areas with which various MTL subregions interact. Thus, changes in MTL function ought to be considered in the context of their functional communities. Two brain networks have been hypothesized to explain MTL-cortical interactions: the Anterior-Temporal (AT) and Posterior-Medial (PM).^6,22–26 The anterior MTL regions (i.e. PRC, ERC, anterior HP and amygdala) belong to the AT network, which also includes a number of extra-MTL regions such as lateral temporal and orbitofrontal cortices, and the temporal pole.⁶ The PM network consists of the posterior MTL (i.e. posterior HP and PHC, but sometimes also includes medial ERC) along with a number of medial parietal regions, especially the posterior cingulate, precuneus and retrosplenial.^6,27 In Alzheimer’s disease, tau pathology is more pronounced in the AT network, while amyloid plaques are more prevalent in the PM network.^24,28,29

MTL dysfunction is an early feature of Alzheimer’s disease,^{11,12,30–33} corroborated by previous task-based functional neuroimaging studies which reported increased HP activation in patients with Mild Cognitive Impairment (MCI),^34–38 and in individuals at genetic risk for Alzheimer’s disease, including asymptomatic carriers of the apolipoprotein ɛ4 allele.^39–44 MTL hyperactivity has also been observed in cognitively normal older adults with elevated levels of tau in the MTL.⁴⁵ Results from functional connectivity (FC) studies have been less clear, with mixed findings of both increased and decreased MTL connectivity in patients with MCI or dementia-level impairment due to Alzheimer’s disease.^22,46–50 In addition, changes in the MTL function at the systems level in normal aging are largely unexplored, and despite the AT and PM networks’ relevance to declining memory in aging and Alzheimer’s disease, the architectural profiles of their direct functional interactions have not been mapped. Given these knowledge gaps in the field, the primary aim of our current study was to determine whether and how age-associated FC changes within the MTL and its immediate functional neighborhood—the AT and PM network systems—differ from those that are caused by Alzheimer’s disease, especially during the earliest preclinical stage of the disease, when clinical interventions have the greatest potential. Our secondary goal was to construct detailed maps of MTL-associated networks. To do so, we examined granular MTL subregional measures within the MTL while we used MTL regions based on the earliest areas of tau deposition in postmortem studies to examine extra-MTL connectivity change. The latter leveraged a recent ex vivo study by our group²⁰ that revealed the presence of an anterior-posterior gradient within the MTL, with posterior subregions being less vulnerable to tau neurofibrillary tangle (NFT) accumulation. Regions with higher/lower NFT burden scores had a remarkable degree of spatial overlap with the MTL portion of the AT/PM network and were, thus, used as seed regions for the MTL-AT/MTL-PM connectivity analyses.

As suggested by animal models of Alzheimer’s disease that revealed amyloid-induced hyperexcitability within the MTL circuitry, including subclinical seizures, during the early stages of the disease,^51–57 we predicted MTL hyperconnectivity in individuals with preclinical Alzheimer’s disease. Since animal work has also demonstrated the activity-dependent nature of tau accumulation and spread,^52,58 we hypothesized that the aforementioned hyperconnectivity effect would be most prominent in the anterior MTL—and the PRC in particular—because this region includes the transentorhinal cortex, the earliest locus of tau pathology in Alzheimer’s disease.^{12,14,17,21,59} Furthermore, because rising tau is associated with synapse loss, cell death and circuit breakdown,^60–62 we predicted that symptomatic stages of Alzheimer’s disease would be characterized by reduced, not excessive, FC levels. In normal aging, we expected to see a generalized loss of network integrity, consistent with previous observations in other systems,^63–69 but perhaps more prominently in the PM network, given more robust declines in cognitive function ascribed to this network in prior studies of cognitive aging.^6,7,70–74

Materials and methods

Participants

In this cross-sectional study, we analysed data from 179 individuals from the Aging Brain Cohort (ABC) study of the University of Pennsylvania Alzheimer's Disease Research Center (Penn ADRC). These participants undergo annual cognitive evaluations, including psychometric testing as prescribed by the Uniform Data Set 3.0.⁷⁵ Consensus diagnoses are reached by a team that includes neurologists and neuropsychologists using clinical history and cognitive scores. Amyloid status was determined by a visual read of amyloid PET scans (see below for acquisition parameters). An additional group of younger adults (<60 years of age) were recruited beyond the ABC study to capture the full adult age span. Thus, four groups were formed: (i) 36 cognitively unimpaired (CU) participants < 60 years of age were classified as normal young and middle-aged adults [age range 23–59 years; mean Montreal Cognitive Assessment (MoCA)⁷⁶ total score ± SD = 26.1 ± 3.7], (ii) 80 Aβ-negative CU participants > 60 years of age were classified as normal older adults [age range 63–86 years; mean MoCA ± SD = 27.4 ± 2.0], (iii) 23 Aβ-positive CU participants > 60 years of age were categorized as preclinical Alzheimer’s disease [age range = 65–87 years; mean MoCA ± SD = 26.9 ± 1.6], and (iv) 40 Aβ-positive cognitively impaired (CI) participants were categorized as symptomatic Alzheimer’s disease [age range 57–87 years; mean MoCA ± SD = 20.3 ± 4.2]. In all analyses aimed at studying the effects of normal aging on MTL connectivity, the first two groups (i.e. normal young, middle-aged and older adults) were merged into a single normal agers cohort. Additional information about our participants is presented in Table 1. The study was approved by the University of Pennsylvania Institutional Review Board.

Table 1.

Demographic information and medial temporal lobe structural measures for each group in the study

	Young and middle-aged (N = 36)	Aβ− CU (N = 80)	Aβ+ CU (N = 23)	Aβ+ CI (N = 40)	Group differences
Age (years)					Young/middle < others *** Aβ− CU < Aβ+ CU **
Mean ± SD	39.2 ± 11.8	71.1 ± 4.9	75.2 ± 6.0	71.7 ± 6.9
Range (min/max)	23/59	63/86	65/87	57/87
Race (White/Black/Asian/Other)	20/12/3/1	61/17/0/2	21/2/0/0	35/4/0/1	Young/middle different from others *
Sex (males/females)	14/22	29/51	7/16	23/17	None
Education (years) ± SD	16.2 ± 2.6	16.6 ± 2.6	16.9 ± 1.7	16.5 ± 2.6	None
Total MoCA ± SD	26.1 ± 3.7	27.4 ± 2.0	26.9 ± 1.6	20.2 ± 4.3	Aβ+ CI MoCA < others ***
CDR (sum/global)	N/A	0.038/0.031	0.087/0.065	2.963/0.638	Aβ+ CI CDR > Aβ+ CU CDR *** Aβ+ CI CDR > Aβ− CU CDR ***
Composite Aβ SUVR ± SD (Florbetaben n)	N/A	0.950 ± 0.062 (n = 68)	1.280 ± 0.201 (n = 21)	1.438 ± 0.144 (n = 37)	Aβ− CU SUVR < Aβ+ CU SUVR *** Aβ− CU SUVR < Aβ+ CI SUVR *** Aβ+ CU SUVR < Aβ+ CI SUVR ***

Amyloid-PET standard uptake value ratio (SUVR) is shown for the Florbetaben tracer only.

* P < 0.05; ** P < 0.01; *** P < 0.001 CU = cognitively unimpaired; CI = cognitively impaired.

Image acquisition

All structural and functional MRI datasets were acquired using a 3.0 T Siemens MAGNETOM Prisma scanner (Erlangen, Germany) at the Center for Advanced Magnetic Resonance Imaging and Spectroscopy (University of Pennsylvania, Philadelphia, PA). During an eyes-open resting-state functional MRI (rs-fMRI) scan, 420 functional volumes were collected axially using a T2*-sensitive Gradient Echo Planar Imaging pulse sequence sensitive to blood oxygenation level-dependent (BOLD) contrast [repetition time (TR): 720 ms; echo time (TE): 37 ms; flip angle: 52°; field of view: 208 × 208 mm²; voxel size: 2 × 2 × 2 mm³; 72 interleaved slices; phase encoding direction: anterior to posterior; partial k-space: 7/8; multi-band acceleration factor: 8]. A whole-brain T1-weighted 3D Magnetization Prepared Rapid Gradient Echo sequence [TR: 2400 ms; TE: 2.24 ms; inversion time: 1060 ms; flip angle: 8°; field of view: 256 × 240 × 167 mm³; voxel size: 0.8 × 0.8 × 0.8 mm³] was used to acquire anatomical images for MTL subregion segmentation and registration to template space. To estimate B₀ inhomogeneity, gradient echo field maps were also collected [TR: 580 ms; TE1/TE2: 4.12/6.58 ms; flip angle: 45°; field of view: 240 × 240 mm²; voxel size: 3 × 3 × 3 mm³; 60 interleaved slices].

Furthermore, all CU participants who were 60 years of age or older and all patients with symptomatic MCI or early Alzheimer’s disease underwent amyloid PET scanning. Our amyloid PET protocol used an injection of 8.1 mCI ± 20% of ¹⁸F-Florbetaben or 10 mCi ± 20% of ¹⁸F-Florbetapir and a 20 min brain scan (4 frames of 5-min duration) following a standard uptake phase (90 min for Florbetaben and 50 min for Florbetapir). 44% of PET scans were acquired within 6 months of rs-fMRI and 85% within a year of rs-fMRI. Cognitively normal young and middle-aged adults underwent rs-fMRI but not PET imaging because this age group is unlikely to have measurable amyloid pathology.

Image pre-processing

Initial processing of structural and functional images was performed on the Flywheel Imaging Informatics Platform (http://flywheel.io). We used a customized fMRIprep 20.0.7⁷⁷ analysis gear, which consisted of mcflirt realignment,^78,79 susceptibility distortion correction using a custom workflow of SDCFlows,⁸⁰ functional-to-structural boundary-based rigid-body registration,⁸¹ SyN diffeomorphic registration to the 2-mm MNI152 template using Advanced Normalization Tools,^82,83 followed by volumetric smoothing with a 6-mm FWHM Gaussian kernel, and decomposition of the fMRI time series into signal sources using the Probabilistic Spatial Independent Component Analysis (ICA).⁸⁴

Next, manual ICA-based denoising was performed by a single rater (SH) with the aid of the automated movement component classifier, ICA-AROMA.⁸⁵ A short description of the protocol that was used for ICA-based artefact detection and removal is provided in the Supplementary Materials.^67,86,87 Following the ICA-based denoising, the first 10 time points were discarded, and the rest of the fMRI pre-processing was done in the Conn toolbox (v. 20.b)⁸⁸ for MATLAB (release 2020b; The MathWorks Inc., Natick, MA). Rather than employing global signal regression, which can introduce artefactual anti-correlations into the fMRI time series,^89–91 we used 32 CompCorr⁹² regressors—16 from the White Matter and 16 from the cerebrospinal fluid (CSF)—that were extracted from the ICA-de-noised fMRI data using eroded tissue probability maps. Those 32 CompCorr regressors were supplemented with movement regressors based on the Friston 24-Parameter model⁹³ as well as low-frequency sine and cosine waves (6 waveforms with periods of π, 2π and 3π over 410 fMRI volumes) that were included to account for slow BOLD signal drifts. Since realignment parameters in sub-second fMRI acquisitions are contaminated by breathing-associated magnetic field perturbations,^94,95 those parameters were filtered using MATLAB scripts provided by Gratton et al.⁹⁴ prior to being used as nuisance covariates. In the last step of our pre-processing pipeline, the fMRI time series were band-pass filtered (0.008–0.150 Hz) and underwent linear detrending. Descriptions of our fMRIprep structural pipeline as well as our amyloid-PET pre-processing pipeline are provided in the Supplementary Materials.

The MTL subregions were segmented using the automated segmentation of hippocampal subfields-T1 (ASHS-T1) pipeline.⁹⁶ We used ASHS-T1 to generate Brodmann Area 36 (BA36), Brodmann Area 35 (BA35), ERC, PHC, as well as the anterior and posterior HP (aHP and pHP, respectively) segmentations. Because the theoretical framework tends to treat the PRC as a single functional unit,⁶ we merged the BA35 and BA36 ROIs into a single PRC ROI. For visual reference, a single participant’s ASHS-T1 segmentation is shown in Fig. 1A. Each subject’s ASHS-T1 segmentations of the MTL were then registered to MNI space using the same deformation fields that were used to register that subject's fMRI data to the 2-mm MNI152 template.

Figure 1.

MTL subregions and their functional interactions with each other. (A) Example ASHS-T1 segmentation from a single participant; (B) Intra-MTL connectome in normal agers (N = 116). One-sided one-sample t-tests with the FDR correction for multiple comparisons were used to test for the presence of positive functional coupling. The connectogram on the left is based on standard correlational connectivity. The connectogram on the right is based on partial correlations and represents direct connectivity within the MTL. Edge thickness in both connectograms is proportional to the t-static score of each connection. Abbreviations: PRC = perirhinal cortex; ERC = entorhinal cortex; PHC = parahippocampal cortex; aHP = anterior hippocampus; pHP = posterior hippocampus; MTL = medial temporal lobe; FDR = false discovery rate; ROI = region of interest.

Intra-MTL functional connectivity analyses

The most common technique for modelling the brain’s functional architecture is to use the bivariate Pearson correlation coefficient as a proxy for FC between brain regions.^97,98 Despite its high sensitivity, this approach may produce biased connectivity estimates because it does not control for indirect paths.^98,99 Such problems can be ameliorated by replacing bivariate correlation coefficients with partial correlations, controlling for signals from all other ROIs in the network model. However, the improvement in architectural accuracy comes at a price: noisier connectivity estimates and reduced statistical power when studying group effects.⁹⁹ Because each FC metric has a unique set of trade-offs, we examined the effects of age and Alzheimer’s disease pathology on intra-MTL FC using both approaches.

All statistical analyses of intra-MTL FC were performed using the Network-Based Statistics (NBS) toolbox.¹⁰⁰ The following NBS settings were used: (i) uncorrected connection-level t-statistic threshold was set to match two-tailed P < 0.05; (ii) the cluster-mass ‘intensity’ option in the NBS algorithm was turned on; (iii) positive and negative contrasts were run separately, and their combined family-wise error-corrected significance threshold was set at P < 0.05; (iv) 25 000 Freedman and Lane¹⁰¹ permutations were used to construct each cluster-mass distribution under the null hypothesis. Prior to performing age effect analyses and group comparisons of intra-MTL FC, correlation matrices were Fisher Z-transformed, and overall intra-MTL connectivity maps were generated using all participants from the normal agers cohort [positive-sided one-sample connection-level t-tests with the false discovery rate (FDR)¹⁰² correction for multiple comparisons]. These overall connectivity maps were used to select edges of interest (Fig. 1B). Since all MTL ROIs were functionally related to each other in the bivariate correlational method, all pairwise connections between ASHS-T1 ROIs were examined for the effects of age and Alzheimer’s disease pathology (Fig. 1B, left panel). For partial correlations, the above test identified 23 (out of 45) edges representing direct functional interactions within the MTL (Fig. 1B, right panel). In all subsequent intra-MTL FC analyses that used partial correlations, only those 23 edges were studied.

To study the effects of age on intra-MTL FC, we ran a general linear model (GLM) on FC data from the normal agers cohort with sex and four aggregate head movement metrics [mean filtered framewise displacement (FD), maximum filtered FD, mean DVARS, max DVARS; see Power et al.¹⁰³ for metric details] as nuisance covariates. To examine how MTL function differs between normal agers and individuals at different stages of Alzheimer’s disease progression, we performed all pairwise group comparisons of intra-MTL connectivity between normal older adults, preclinical Alzheimer’s disease and symptomatic Alzheimer’s disease. In these group comparisons, age, sex and aggregate head movement metrics were used as covariates. Connection-specific and supplementary statistical comparisons were performed using SPSS v. 28 (IBM Inc., Armonk, NY). Age-, sex- and motion-corrected intra-MTL connectivity matrices for each of the four groups are available in the Supplementary Materials (Supplementary Fig. 1).

Identification of cortical regions with functional connectivity to the MTL

Seminal work by Braak and Braak¹⁷ demonstrated that tau NFT pathology originates at the border of the PRC and ERC in the anterior MTL, and a recent ex vivo study by our group revealed the presence of an anterior-posterior NFT gradient within the MTL with posterior subregions being less vulnerable to NFT accumulation.²⁰ Regions with higher/lower NFT burden scores had a remarkable degree of spatial overlap with the MTL portion of the AT/PM network, and together with the AT network’s vulnerability to tauopathy versus PM network’s vulnerability to amyloidosis, seem to reflect partially dissociable patterns of molecular pathology within the anterior versus posterior MTL and might also relate to dissociable patterns of network disruption in the AT versus PM system.^6,22,24,28 Thus, when modelling MTL interactions with the rest of the cortex, we employed four tau-based MTL ROIs [left/right × anterior/posterior]. Our group recently built a novel atlas of MTL NFT accumulation using fusion of ex vivo MRI and serial histological imaging (Supplementary Fig. 2).²⁰ Left and right tau-based MTL ROIs were defined by thresholding and binarizing this NFT accumulation map (inclusion threshold: WildCat-based metric of NFT burden > 0.2; see Yushkevich et al.²⁰ for threshold details), and the anterior/posterior split was defined using the aHP/pHP boundary of the ASHS-T1 protocol.^7,96,104 Next, we identified cortical regions with positive FC to at least one of the four tau-based MTL ROIs in the normal agers cohort (Fig. 2). In ROI form, this extended MTL network (i.e. all cortical regions displaying statistical association with the MTL) was represented by the anterior and posterior tau-based MTL ROIs (aMTL_tau and pMTL_tau, respectively) and 221 ROIs from the 400-region 17-Network Schaefer et al.¹⁰⁵ parcellation (Fig. 2; Supplementary Methods).

Figure 2.

The extended MTL network. Cortical regions with positive functional connectivity to the anterior or posterior tau-based MTL ROIs were identified and mapped onto ROI-based representation. The resulting set of 225 ROIs was termed the Extended MTL Network. The Extended MTL Network was identified using the normal agers group (N = 116). Abbreviations: MTL = medial temporal lobe; TFCE = threshold-free cluster enhancement; FDR = false discovery rate; ROI = region of interest.

Estimating MTL network architecture

The functional architecture of the extended MTL network was estimated using the graphical SCAD algorithm (Fig. 3).^106–109 Similar to other sparse estimation techniques, graphical SCAD eliminates spurious or indirect connections by penalizing excessive model complexity if there is a lack of evidence in the data to support a more complex graph. Because sparse network estimation based on 225 ROIs and a relatively short fMRI acquisition is unlikely to produce sufficiently stable connectome estimates at the subject level, we performed group-level network estimation instead. Custom MATLAB scripts employing the QUIC optimizer¹¹⁰ were used to solve the graphical SCAD problem (see Supplementary Methods for methodological details of the graphical SCAD estimation).

Figure 3.

AT and PM network estimation. Direct functional connections within the Extended MTL Network (see Fig. 2) from the normal agers group (N = 116) were identified using the graphical SCAD estimation method with BIC-based model selection. Network communities within the Extended MTL Network were identified using consensus Louvain modularity. Abbreviations: AT = anterior-temporal; PM = posterior-medial; MTL = medial temporal lobe; Ant. = anterior; Post. = posterior; Dors. = dorsal; SCAD = smoothly clipped absolute deviation estimator; ROI = region of interest.

To identify the AT and PM network communities within the broader MTL connectome, we used the two-sided Louvain modularity algorithm that incorporates both positive and negative edge weights in its community search.^111,112 Since modularity-based network parcellations can produce inconsistent networks from one iteration to another, we built a consensus parcellation from 5000 separate iterations of the fine-tuned two-sided Louvain community search.^112–114 Module detection was performed on the normal agers’ sparse partial correlation matrix, estimated using the graphical SCAD. In total, we identified six stable functional modules that could explain functional relationships among the 225 MTL-affiliated ROIs (Fig. 3). Based on visual inspection, two of those modules were consistent with the AT and PM networks that have been reported in the literature.^6,22,24,26 The parcellation procedure was performed using functions from the Brain Connectivity Toolbox.¹¹⁵

Statistical analyses of intra-AT, intra-PM and AT-PM connectivity

Age effect analyses and Alzheimer’s disease group comparisons were performed on bivariate connectivity matrices but only on those connections that survived group-level graphical SCAD estimation in at least one test-related group. For age effect analyses, in addition to the overall connectivity map that was estimated above, we also estimated separate SCAD-based maps for younger (normal agers < 60 years) and older (normal agers > 60 years) participants. If a connection was positive in at least one of the three (i.e. younger, older, overall) group-level sparse connectivity matrices, it was examined for age effects. Negative edges and those with conflicting signs in the bivariate versus graphical SCAD models were excluded because of interpretability issues. The final set of intra-AT, intra-PM, and AT-PM connections that were used in age effect analyses is shown in Fig. 4. For Alzheimer’s disease group comparisons, similar criteria were used, except that there were two instead of three group-level connectivity estimates per comparison. For example, two separate group-level graphical SCAD connectivity estimates were computed for the preclinical versus symptomatic Alzheimer’s disease connectivity comparison: the first for the preclinical group and the second for the symptomatic group.

Figure 4.

AT and PM network architecture. The top panel depicts direct intra-AT and intra-PM functional connections in the normal agers group (N = 116). The bottom panel depicts direct AT-PM inter-network connections in the normal agers group (N = 116). The AT-affiliated ROIs and intra-AT connections are in purple; the PM-affiliated ROIs and intra-PM connections are in green. The AT-PM inter-network connections are in grey/black. The anterior and posterior tau-based MTL ROIs are marked with a yellow asterisk. Abbreviations: AT = anterior-temporal; PM = posterior-medial; MTL = medial temporal lobe.

Statistical analyses of the AT and PM intra-module and AT-PM connectivity were performed using the NBS GLM approach¹⁰⁰ with all settings, including covariates, matching those that were used in the intra-MTL analyses described above. Connection-specific follow-up comparisons were performed using SPSS v. 28 (IBM Inc., Armonk, NY). All AT and PM network results were visualized with the BrainNet Viewer.¹¹⁶

Data availability

Anonymized pre-processed data and in-house analysis scripts will be made available upon request for the sole purpose of replicating procedures and results presented in this article.

Results

Effects of age and Alzheimer’s disease progression on functional connectivity between the MTL subregions

The NBS algorithm identified a single cluster of 14 edges that represented declining FC in normal agers (family-wise error P < 0.001; Fig. 5A; Supplementary Fig. 3A). Most of the cluster’s connections were interhemispheric (10 out of 14 edges), and all but one represented either anterior-with-posterior or posterior-with-posterior functional interactions. The PHC was involved in 9 out of 14 connections with a negative correlation to age. We did not observe any positive associations between age and intra-MTL FC.

Figure 5.

The effects of (A) age and (B-E) Alzheimer’s disease progression on intra-MTL functional connectivity. The NBS-based GLM approach was used to assess the statistical significance of each contrast/comparison (N_{Aβ− CU} = 80, N_{Aβ+ CU} = 23, N_{Aβ+ CI} = 40). (B) Matrix-form representation of connectivity differences between Aβ-positive individuals with preclinical Alzheimer’s disease and Aβ-negative age-matched controls (univariate GLMs at the connection level; N_contrast = 103). (E) Left panel: intra-MTL connection cluster with an inverted ‘U-shaped’ functional connectivity pattern in Alzheimer’s disease (GLM-based quadratic trend analysis; N_contrast = 143). Right panel: violin plot depicting this cluster’s mean functional connectivity scores in controls, individuals with preclinical Alzheimer’s disease, and patients with symptomatic Alzheimer’s disease. Data points represent connectivity scores from individual participants. Abbreviations: PRC = perirhinal cortex; ERC = entorhinal cortex; PHC = parahippocampal cortex; aHP = anterior hippocampus; pHP = posterior hippocampus; CU = cognitively unimpaired; CI = cognitively impaired; Aβ− = amyloid-negative; Aβ+ = amyloid-positive; GLM = general linear model; NBS = network-based statistic(s).

Comparing intra-MTL connectivity profiles of the Aβ− CU and Aβ+ CU older adult groups revealed increased functional coupling between the right PRC and other MTL subregions in the preclinical Alzheimer’s disease group (Fig. 5B and C). Although PRC hyperconnectivity reached statistical significance in the right hemisphere only, similar trends were observed in the left hemisphere as well (Fig. 5B). To test whether these alterations in MTL function were linked to structural atrophy in the MTL, we used two-way ANCOVAs [Group and Gender as fixed factors; Age as a covariate] to compare the Aβ− CU versus Aβ+ CU groups’ MTL volumetric measurements (thickness for PRC, ERC, and PHC; ICV-corrected volumes for aHP and pHP; thickness/volumes averaged across hemispheres; Supplementary Table 1). Because we did not detect any differences in MTL structure when comparing normal older adults to individuals with preclinical Alzheimer’s disease (all P-values > 0.10), our results suggest that MTL network dysfunction may precede neuronal loss in Alzheimer’s disease. However, we note that longitudinal data will be needed to conclusively demonstrate this effect.

To investigate the effects of disease progression on the MTL function, we compared intra-MTL connectivity profiles of the Aβ+ CI group to those of the Aβ+ CU group. This comparison revealed reduced PRC- and ERC-associated connectivity in the Aβ+ CI group (Fig. 5D). Relative to normal older adults, individuals in the Aβ+ CI group displayed lower HP connectivity suggesting that functional changes in the symptomatic disease are characterized by functional decoupling of the two HP from each other and from the rest of the MTL (Supplementary Fig. 3B).

The above results suggest that the PRC (and possibly ERC) connectivity follows an inverted ‘U-shaped’ pattern characterized by an initial rise during the asymptomatic stage and a subsequent decline that likely coincides with the onset of clinical symptoms. To test this hypothesis, we performed a quadratic trend analysis on intra-MTL connectivity measures from the Aβ− CU, Aβ+ CU and Aβ+ CI groups. As expected, a single cluster of connections in the anterior MTL, most of which involved the PRC, demonstrated the inverted ‘U-shaped’ effect (i.e. Aβ− CU FC < Aβ+ CU FC > Aβ+ CI FC) in Alzheimer’s disease (Fig. 5E, left panel). Visual inspection of mean connectivity scores within this cluster revealed a single outlier in the Aβ+ CU group (Fig. 5E, right panel); however, its removal does not eliminate the statistical significance of the quadratic contrast in the PRC. Therefore, our results support the notion of increased functional synchronicity in the anterior MTL, specifically during the asymptomatic stage of Alzheimer’s disease (age-, gender- and motion-corrected PRC connectivity means in Fisher’s Z-score units: M_{Aβ− CU} = 0.086, M_{Aβ+ CU} = 0.171, M_{Aβ+ CI} = 0.057; Fig. 5E).

In summary, our intra-MTL connectivity comparisons revealed PRC hyperconnectivity in cognitively normal amyloid-positive individuals and HP hypoconnectivity in patients with symptomatic Alzheimer’s disease. In normal agers, we observed age-associated FC decline that was centred around the PHC connections. Furthermore, these trends did not change if partial correlations were used to model intra-MTL functional interactions (Supplementary Fig. 4) or if ASHS-T1 structural measurements were included as covariates (Supplementary Fig. 5).

The effects of age and Alzheimer’s disease progression on the AT and PM networks

Both network systems displayed declining intra-network FC as a function of age, while no positive relationships to age were detected in either system (Fig. 6A). Relationships to age within the AT and PM systems were represented by single 20-edge and 40-edge clusters, respectively. Only the AT network’s cluster contained an edge with a direct link to one of the tau-based MTL ROIs (Fig. 6A). In analyses of AT-PM inter-network FC, only the aMTL_tau-pMTL_tau edge in the left hemisphere had a statistically significant negative relationship to age (Supplementary Fig. 6), demonstrating that FC between the anterior and posterior MTL segments declines with age. We did not observe any statistically significant positive relationships between age and AT-PM inter-module FC.

Figure 6.

Effects of (A) age and (B-D) Alzheimer’s disease progression on intra-network connectivity of the AT and PM systems. In each comparison, NBS-based GLMs were used to test for the presence of statistical group differences (N_{Aβ− CU} = 80, N_{Aβ+ CU} = 23, N_{Aβ+ CI} = 40). AT nodes and P-values are in purple; PM nodes and P-values are in green. Tau-based MTL ROIs are marked with a yellow asterisk. Abbreviations: AT = anterior-temporal network; PM = posterior-medial network; MTL = medial temporal lobe; CU = cognitively unimpaired; CI = cognitively impaired; Aβ− = amyloid-negative; Aβ+ = amyloid-positive; GLM = general linear model; NBS = network-based statistic(s); ROI = region of interest.

Consistent with the intra-MTL effects detailed above, the preclinical Alzheimer’s disease group displayed AT-specific hyperconnectivity relative to age-matched amyloid-negative controls. This excess connectivity was represented by a single cluster of connections, all but one of which emanated from the aMTL_tau ROIs. As in the intra-MTL results, the AT-specific hyperconnectivity was more prominent in the right hemisphere and was not present in the symptomatic group (Fig. 6B). No FC differences between the Aβ− CU and Aβ+ CU groups were detected in the PM system (Fig. 6B). Furthermore, in congruence with our intra-MTL findings, our analyses of the AT and PM networks indicate that hyperconnectivity within the AT system declines as pathology and symptoms progress beyond the preclinical stage (Fig. 6C). Relative to normal agers, patients with MCI and prodromal Alzheimer’s disease displayed reduced FC within the PM but not the AT system (Fig. 6D). No differences in the AT-PM inter-network FC were found between the Aβ− CU, Aβ+ CU and Aβ+ CI groups.

To further examine the effects of age on those connections that link the aMTL_tau and pMTL_tau ROIs to their respective networks, we first computed average FC scores of all such connections (i.e. one-hop aMTL_tau-AT and pMTL_tau-PM connectivity; Figs. 7 and 8). In normal agers, both the aMTL_tau-AT and pMTL_tau-PM one-hop FC was negatively associated with age (Fig. 7). Next, we evaluated the effect of disease progression on the aMTL_tau-AT and pMTL_tau-PM connectivity (Fig. 8). Disease stage group differences were statistically significant for the aMTL_tau-AT connections (F(2,134) = 10.96, P < 0.0001). Follow-up pairwise comparisons revealed that the aMTL_tau-AT one-hop connectivity was greater in the Aβ+ CU group relative to the Aβ− CU age-matched control group [F(1,134) = 15.99, Holm-Bonferroni-corrected P = 0.0001, M_diff = 0.086 Fisher’s Z-score units; Fig. 8] and relative to the Aβ+ CI patient group [F(1,134) = 20.506; Holm-Bonferroni-corrected P < 0.0001, M_diff = 0.109 Fisher’s Z-score units; Fig. 8]. No statistical differences in the aMTL_tau-AT connectivity were observed between the Aβ− CU and Aβ+ CI groups (uncorrected P = 0.221). Removing the influential outlier in the Aβ+ CU group (Fig. 8) did not alter the statistical inference [Holm-Bonferroni-corrected P = 0.014 for the Aβ+ CU versus Aβ− CU comparison; Holm-Bonferroni-corrected P = 0.002 for the Aβ+ CU versus Aβ+ CI comparison]. For the pMTL_tau-PM analyses, we observed a trend towards lower connectivity in the Aβ+ CI group relative to the Aβ− CU group [F(1,131) = 5.16, Holm-Bonferroni-corrected P = 0.076; Fig. 8].

Figure 7.

Age relationships for aMTL_tau-AT and pMTL_tau-PM functional connectivity. The aMTL_tau-AT/pMTL_tau-PM connectivity metric was computed by averaging connectivity values of those edges that represented direct functional interactions between the aMTL_tau (or pMTL_tau) ROIs and their one-hop neighbours in the AT (or PM) network (top and bottom glass brains, respectively). Lifespan trajectories for normal agers were fitted using multiple regression (N = 116). Thin lines around fitted age trajectories represent 95% confidence intervals. The AT network’s nodes and edges are shown in purple; those of the PM network are shown in green. Tau-based MTL ROIs are marked with a yellow asterisk. Abbreviations: AT = anterior-temporal; PM = posterior-medial; MTL = medial temporal lobe; aMTL_tau = tau-based anterior MTL ROI; pMTL_tau = tau-based posterior MTL ROI; CU Aβ− = cognitively unimpaired amyloid-negative; Ant. = anterior; Post. = posterior; ROI = region of interest.

Figure 8.

Effect of Alzheimer’s disease progression on aMTL_tau-AT and pMTL_tau-PM functional connectivity. The violin plot in the centre depicts mean aMTL_tau-AT and pMTL_tau-PM connectivity scores in Aβ-negative CU older adults (N_{Aβ− CU} = 80), Aβ-positive CU individuals with preclinical Alzheimer’s disease (N_{Aβ+ CU} = 23), and Aβ-positive individuals with symptomatic Alzheimer’s disease (N_{Aβ+ CI} = 40). Data points represent connectivity scores from individual participants. GLMs were used to test for the presence of connectivity differences between the three groups. Statistically significant Holm-Bonferroni-corrected pairwise comparisons are highlighted. The aMTL_tau-AT/pMTL_tau-PM connections from which overall aMTL_tau-AT/pMTL_tau-PM connectivity values were computed are shown to the left/right of the violin plot. Tau-based MTL ROIs are marked with a yellow asterisk. Abbreviations: AT = anterior-temporal; PM = posterior-medial; MTL = medial temporal lobe; aMTL_tau = tau-based anterior MTL ROI; pMTL_tau = tau-based posterior MTL ROI; CU = cognitively unimpaired; CI = cognitively impaired; Aβ− = amyloid-negative; Aβ+ = amyloid-positive; GLM = general linear model; ROI = region of interest.

In summary, intra-network FC of both network systems displayed a progressive decline with age in amyloid-negative cognitively normal adults. The preclinical stage of Alzheimer’s disease, on the other hand, was characterized by increased functional synchronicity within the AT system, driven mainly by the aMTL-AT one-hop connections. In patients with symptomatic Alzheimer’s disease, our results revealed hypoconnectivity in the PM system.

Discussion

We conducted a detailed investigation of MTL FC changes in normal aging and in preclinical and symptomatic Alzheimer’s disease. We paid particular attention to the anterior-posterior axis of the MTL to determine whether the anterior or posterior MTL segments were more vulnerable to age- or Alzheimer’s disease-related network dysfunction. Our most intriguing finding is the inverted ‘U-shaped’ FC pattern in the anterior MTL from normal aging to preclinical Alzheimer’s disease to symptomatic Alzheimer’s disease. The preclinical stage of Alzheimer’s disease was characterized by increased FC centred around areas of the anterior MTL that are prone to early tauopathy. In patients with symptomatic disease, hyperconnectivity was no longer present. Instead, we observed reduced connectivity within the PM system, between the anterior and posterior segments of the HP, and between the left and right hippocampi. In contrast to the hyperconnectivity seen in preclinical Alzheimer’s disease, advanced age was associated with reduced intra-network FC in both the AT and PM systems. Within the MTL, we observed greater vulnerability of the posterior MTL, especially the PHC connections, to age-associated connectivity decline. Together, our results indicate that MTL network dysfunction in Alzheimer’s disease is not simply an accelerated form of normal aging.

Hyperconnectivity in the anterior MTL in preclinical Alzheimer’s disease

In our earlier work, we reported increased FC between the ERC and other regions of the MTL in amnesic MCI.⁴⁶ Later studies by others confirmed those findings.^22,48 However, it was not obvious how early in the course of Alzheimer’s disease this effect first appears, and conflicting results have been published on the nature of the MTL network disruption at more advanced stages.^48,117 Our results demonstrated that the anterior MTL hyperconnectivity is present in individuals with preclinical but not prodromal Alzheimer’s disease. Inverted ‘U-shaped’ FC patterns of this type have been previously identified in the default mode and salience networks.¹¹⁸ To our knowledge, only one other study reported elevated FC levels in the AT network among older individuals with memory complaints but not in patients with dementia due to Alzheimer’s disease.¹¹⁷ However, that study did not use PET or CSF biomarkers to screen for preclinical Alzheimer’s disease, and because of resolution limitations, it had limited anatomical specificity within the MTL.

The physiology underlying network hyperconnectivity in preclinical Alzheimer’s disease is unknown. Epidemiological studies indicate an elevated prevalence of subclinical seizures in both early-onset and late-onset Alzheimer’s disease,^119–121 frequently originating in the temporal cortex and associated with earlier onset of cognitive symptoms.¹²² Even though we did not observe any differences in cognitive capacity between normal agers and CU participants with preclinical Alzheimer’s disease, it is plausible that increased anterior MTL connectivity in this group represents circuit dysfunction as opposed to compensatory processes. For instance, Bakker et al.¹²³ showed that administering a low dose of the anti-epileptic drug levetiracetam to patients with amnesic MCI can result in memory improvement.

To our best knowledge, only a single human imaging study by Berron et al.²² examined the effects of preclinical and symptomatic Alzheimer’s disease on both the AT and PM networks. They reported decreased FC between the medial prefrontal cortex and the ERC/PRC region of the MTL in preclinical Alzheimer’s disease and decreased MTL-PM FC in Aβ+ individuals with amnesic MCI. Thus, it is plausible that functional abnormalities in preclinical Alzheimer’s disease are characterized not only by local and regional anterior MTL hyperconnectivity but also by reduced connectivity between the anterior MTL and more distant regions of the neocortex. Furthermore, according to our current results, the MTL subregions most likely to show NFT accumulation are also the ones to display patterns of excessive connectivity in the early stages of Alzheimer’s disease. Harrison et al.¹²⁴ reached a similar conclusion for intra-HP connectivity, demonstrating a positive relationship between the intra-HP signal coherence (a measure of local connectivity) and tau burden in the ERC among CU older adults.

Animal and in vitro research has revealed that hyperactivity within the MTL enhances both amyloid and tau pathology in the HP and ERC,^{51,56–58,125,126} while the earliest stages of neuronal dysfunction in Alzheimer’s disease mice are characterized by Aβ-induced hyperactivity with hypoactivation appearing in later stages of the disease.¹²⁷ Using mice models of Alzheimer’s disease, Angulo et al.⁵² studied the effects of Aβ and tau on ERC neurons specifically. Consistent with findings from non-ERC MTL subregions, Aβ—in the absence of tau—induced hyperexcitability in the ERC circuits; however, co-expression of both Aβ and tau moderated the excitatory effects of Aβ on the ERC.⁵² This is broadly consistent with our findings of increased connectivity in the anterior MTL only during the preclinical stage of Alzheimer’s disease when tau levels are expected to be low. Alternatively, by symptomatic stages of the disease, most individuals are expected to have a more significant tau burden both within the MTL and beyond.^{12,16,58,128–131} Indeed, a recent human imaging study by Berron et al.¹³² reported decreased FC in the AT system as a function of rising MTL tau.

In this context, our results support the hypothesis that the earliest pathological processes in Alzheimer’s disease result in MTL network dysfunction centred around the anterior MTL subregions, possibly caused by Aβ-induced hyperexcitability. Such elevated activity levels likely accelerate NFT formation and spread, leading to synapse loss and network breakdown over time. Alternatively, the observed hyperconnectivity effect in the anterior MTL might represent a functional reserve that allows for normal cognitive capacity despite Alzheimer’s disease-associated pathology.

Intra-MTL functional connectivity changes in normal aging

We showed that posterior MTL subregions, and especially the PHC, become increasingly decoupled from the rest of the MTL with age. On standardized psychological assessments, older adults attain lower scores on tests of recollection and relational/associative memory, while familiarity-based item recognition is largely unaffected by age.^70–74,133 Declining performance on tests of associative/relational memory indicates posterior, as opposed to anterior, MTL dysfunction in older adults.^6,7,134 Consistent with the evidence from cognitive work, Panitz et al.¹³⁵ reported reduced intra-HP connectivity in older participants but only in the middle-posterior HP segments. Similarly, some structural MRI studies reported greater vulnerability of the posterior MTL subregions to atrophy in normal aging.^18,136–138

In a task-based study of intra-MTL FC, Stark et al.¹³⁹ compared PRC-HP, ERC-HP and PHC-HP connectivity profiles of CU young (<40 years) and CU old (≥70 years) adults. The authors reported reduced aHP-PHC connectivity among the elderly and a smaller but also significant reduction of the PRC-HP FC. However, group differences were not found in the more anterior ERC-HP connections. In the rs-fMRI literature, the effects of age on HP- and PHC-associated FC have been examined by Damoiseaux et al.,¹⁴⁰ who reported negative age relationships for the anterior-posterior and interhemispheric intra-HP connections. By focusing on all memory-related MTL subregions, not just the HP and PHC, we addressed many of the outstanding questions about the effects of normal aging on intra-MTL FC.

Effects of age on functional connectivity properties of the AT and PM networks

This is the first FC study to document age-related alterations in the AT and PM intra- and inter-network connectivity for the entire adult lifespan. Initial imaging evidence for altered network dynamics in old age was demonstrated in task-based fMRI and PET experiments, which showed an over-recruitment of frontal and parietal association cortices in older cohorts relative to younger cohorts across a wide variety of cognitive tasks.^141–151 Studies of spontaneous BOLD signal fluctuations also found a negative relationship between age and network specialization across standard cerebral networks, which often manifests as reduced intra-network and increased inter-network connectivity in older participants.^69,152–159 Such functional desegregation in the AT and PM networks was recently reported by Cassady et al.,¹⁵⁹ and this reduced specialization was associated with more tau in the AT and more Aβ in the PM systems. For normal agers, our current results broadly agree with previous work.^{63,66–69,159–161} However, contrary to prior reports of increased inter-network connectivity in aging individuals,¹⁵⁹ we did not detect any positive relationships between age and AT-PM inter-network connectivity. Instead, our results indicate reduced AT-PM interactions in this group. This lack of agreement with earlier literature might have resulted from our exclusive focus on direct AT-PM interactions.

Limitations and future work

A few caveats merit discussion. First, as is the case with most fMRI studies of the MTL, the anterior MTL subregions in our study were more vulnerable to susceptibility artefacts from the skull base than their posterior counterparts. However, because both raw and pre-processed temporal signal-to-noise ratio profiles for each ASHS-T1 ROI were similar across groups (Supplementary Table 2), it is unlikely that susceptibility effects biased our main findings. Second, because of acquisition constraints, we assumed a common network architecture for all participants in a given statistical comparison. Despite high levels of structural consistency, inter-individual variability in network architecture exists,^162–164 and exploring individual differences in the AT and PM network profiles at different stages of Alzheimer’s disease progression is a valuable avenue for future research.

In the current study, we did not analyse the effects of age and Alzheimer’s disease progression on the amplitude of BOLD signal fluctuations, even though previous studies revealed that BOLD signal amplitude is affected in both aging and Alzheimer’s disease.^67,165 Based on recent evidence from Strain et al.,¹⁶⁵ we do not expect our findings on altered MTL connectivity in preclinical Alzheimer’s disease to be confounded by amplitude effects; however, AT- and PM-associated measures of BOLD signal amplitude might provide additional insight into the pathophysiology of Alzheimer’s disease, and this topic merits further research.

Head motion has been shown to modulate FC in multiple brain networks.^166–168 As in other studies involving older adults,^67,169 head movement in our current study was correlated with age. Our pre-processing pipeline was designed with this issue in mind. Furthermore, among the three older groups [i.e. CU Aβ−, CU Aβ+, CI Aβ+] the amount of head motion was similar (Supplementary Table 3). Consequently, it is unlikely that our main findings were driven by head movement artefacts.

Conceptually, the present study focused only on the anterior-posterior axis of the MTL; however, work by others demonstrated functional specialization and differential Alzheimer’s disease pathology within the medial-to-lateral axis as well.^{10,20,38,139,170–173} Future studies employing high-field high-resolution fMRI acquisitions will be able to study the effects of Alzheimer’s disease progression on both axes simultaneously.

Although we infer opposite effects of amyloid and tau on anterior MTL connectivity, the proposed model needs further validation on participants who underwent both amyloid- and tau-PET scans. Direct imaging of both amyloid and tau is also necessary for bridging our current findings on amyloid-associated AT hyperconnectivity in preclinical Alzheimer’s disease with recent work by others that focused on the relationship between MTL function and tau.^{29,45,132,159} In addition, even though our results suggest that the anterior MTL connectivity in symptomatic Alzheimer’s disease reverts to approximately the level seen in controls, it is not clear from our data whether this is caused by a simple return to the baseline or by a separate pathological process that cancels out the hyperconnectivity effect seen in preclinical Alzheimer’s disease. Lastly, the cross-sectional nature of our design and the associated limitations must be acknowledged.

Conclusion

Our results suggest that MTL-AT connectivity follows an inverse ‘U-shaped’ pattern in Alzheimer’s disease. The earliest phase of Alzheimer’s disease is characterized by elevated MTL-AT connectivity. This effect originates from the MTL regions most susceptible to early tauopathy. As the disease progresses, MTL-AT connectivity declines and hypoconnectivity appears in the PM network. In normal aging, both the AT and PM networks displayed intra-network connectivity decline; however, the deleterious effects of age were more pronounced in the posterior MTL subregions. Thus, MTL functional connectivity is a dynamic process in normal aging and in Alzheimer’s disease.

Supplementary material

Supplementary material is available at Brain Communications online.

Funding

The work in the present study was supported by the Alzheimer’s Association (Grant AARF-21- 48972) and by the National Institutes of Health (Grants P30-AG072979, RF1-AG069474, R01-AG056014, R01-AG055005, R01-AG072796 and R01-AG070592).

Competing interests

D.A.W. has served as a paid consultant to Eli Lilly, GE Healthcare and Qynapse. He serves on a DSMB for Functional Neuromodulation. He receives research support paid to his institution from Biogen. I.N. serves on the Scientific Advisory Board for Eisai and does educational speaking for Biogen. All other authors report no competing interests.