SuperAging functional connectomics from resting-state functional MRI

Abstract Understanding the relationship between functional connectivity (FC) of higher-order neurocognitive networks and age-related cognitive decline is a complex and evolving field of research. Decreases in FC have been associated with cognitive decline in persons with Alzheimer’s disease and related dementias (ADRD). However, the contributions of FC have been less straightforward in typical cognitive aging. Some investigations suggest relatively robust FC within neurocognitive networks differentiates unusually successful cognitive aging from average aging, while others do not. Methodologic limitations in data processing and varying definitions of ‘successful aging’ may have contributed to the inconsistent results to date. The current study seeks to address previous limitations by optimized MRI methods to examine FC in the well-established SuperAging phenotype, defined by age and cognitive performance as individuals 80 and older with episodic memory performance equal to or better than 50-to-60-year-olds. Within- and between-network FC of large-scale neurocognitive networks were compared between 24 SuperAgers and 16 cognitively average older-aged control (OACs) with stable cognitive profiles using resting-state functional MRI (rs-fMRI) from a single visit. Group classification was determined based on measures of episodic memory, executive functioning, verbal fluency and picture naming. Inclusion criteria required stable cognitive status across two visits. First, we investigated the FC within and between seven resting-state networks from a common atlas parcellation. A separate index of network segregation was also compared between groups. Second, we investigated the FC between six subcomponents of the default mode network (DMN), the neurocognitive network commonly associated with memory performance and disrupted in persons with ADRD. For each analysis, FCs were compared across groups using two-sample independent t-tests and corrected for multiple comparisons. There were no significant between-group differences in demographic characteristics including age, sex and education. At the group-level, within-network FC, between-network FC, and segregation measurements of seven large-scale networks, including subcomponents of the DMN, were not a primary differentiator between cognitively average aging and SuperAging phenotypes. Thus, FC within or between large-scale networks does not appear to be a primary driver of the exceptional memory performance observed in SuperAgers. These results have relevance for differentiating the role of FC changes associated with cognitive aging from those associated with ADRD.


Introduction
The global population is rapidly aging and the proportion of adults aged 85 or older is growing faster than younger generations. 1 These adults are at the highest risk for both age-related memory decline 2 and the onset of amnestic dementia due to Alzheimer's disease (AD). 3However, significant memory decline is not inevitable.A growing literature on 'successful aging' 4,5 has aimed to describe and investigate older adults with unusually high physical, [6][7][8] social, 7,9 or cognitive functioning. 6,8Understanding mechanisms of successful cognitive aging promises to inform the development of interventions to prevent or slow cognitive decline.
One well-established phenotype of successful cognitive aging is SuperAging, defined by age and cognitive performance as individuals 80 years and older with episodic memory performance equal to or better than 50− to 60-year-olds. 10The multicenter SuperAging Research Initiative (R01AG045571, R01AG067781, and U19AG073153) was launched to understand what factors underlie this memory-specific phenotype and, by comparison, to inform our understanding of normal cognitive decline and AD.Initial results show that SuperAgers share unique neuropsychological, 11,12 psychosocial, 13 genetic, 10,14 and biologic [15][16][17][18][19] features.Post-mortem studies have identified neurobiological features of the SuperAger phenotype, such as relatively large neurons in the entorhinal cortex, 15 a lower AD neuropathologic burden, and a greater density of von Economo neurons in the anterior cingulate cortex. 16Structural MRI studies have found brain features of the SuperAging phenotype that appear more like middle-aged-controls than older-aged-controls (OACs), such as relatively slow atrophy rates 20 and thick cingulate cortices. 18To date, studies of the SuperAger phenotype have focused on brain morphometry.Less is known about the brain functional connectivity (FC) that supports their extraordinary memory abilities.
The brain is thought to be subdivided into distinct brain networks composed of highly interconnected neural regions that communicate to manifest complex behaviours and cognitive abilities. 21Resting-state functional MRI (rs-fMRI) has emerged as a proxy method for exploring the FC of these distributed neurocognitive networks 22 by capturing intrinsic temporal correlations in neural activity 23 indirectly measured from blood-oxygen-level-dependent (BOLD) signal.Exploratory rs-fMRI studies have identified resting-state networks that correspond to major networks previously established by neuroanatomical experiments 21,24 and others with unclear neural connections. 25Some resting-state networks are thought to be involved in unimodal sensory processing, such as the visual 26 and somatomotor networks. 23Other resting-state networks are thought to modulate indistinct higher-order cognitive functions, such as the frontoparietal, 27 ventral attention, [28][29][30] dorsal attention, 28 and default mode networks (DMN). 311][42] The significance of these age-related changes is unclear.Some have suggested that a gradual shift away from 'youthful' functional patterns indicates progressive dysfunction and signifies or precipitates cognitive decline. 41,42One rationale for this hypothesis is that the topography of resting-state networks overlaps with regional activation during specific cognitive tasks.4][45] The DMN is also commonly thought to include the hippocampus, 31 the brain region most commonly associated with episodic memory. 46In some cases, within-network FC has also been related to composites of neurocognitive performances, such as persons with relatively strong within-network FC in the DMN performing better on episodic memory tasks. 34,39,41Nearly all higher-order restingstate networks display a similar pattern. 32As such, the overall degradation of within-network FC may be more closely related to the severe cognitive decline associated with ADRD.
The DMN has become a major focus of cognitive aging research because of its high detectability, topographical overlap with memory-related regions, and enigmatic association with AD dementia.The DMN is one of the most commonly studied resting-state networks. 47It demonstrates unique functional patterns in persons with mild neurocognitive impairments 48 and amnestic dementia due to AD. 49 The pathogenesis of AD amyloidopathy appears to selectively accumulate within regions of the DMN 50,51 early in the disease course.Given these findings, some have argued that preserved DMN integrity in old age supports optimal memory abilities.However, DMN functional anomalies emerge prior to amnestic symptoms of AD. 49,[52][53][54][55] In addition, decreases in DMN cannot distinguish persons with amnestic AD from persons with non-amnestic variants of AD. [56][57][58] Furthermore, functional changes in the DMN are also implicated in non-AD diagnoses including depression, autism spectrum disorder and schizophrenia 44 where memory impairments are not core features.As such, the relationship between the DMN and memory decline in AD is complex, with evidence suggesting the network may be vulnerable to change in multiple disease-related states rather than being specific for AD.
Given the age and AD-related findings, it has recently been proposed that relatively robust FC within neurocognitive networks may support successful cognitive aging.Research differs in the terminology and classification of successful cognitive aging, 59,60 specifically the age range of cohorts, which limits the generalizability of findings.Nonetheless, one recent study of adults aged 60 and over with exceptional episodic memory abilities found that these individuals had relatively strong FC within the DMN and ventral attention neurocognitive networks compared to similar aged cognitive controls. 34They also reported a positive association between FC and performance on a task of episodic memory.However, another recent study using the same methods in a separate but equally sized cohort from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database was unable to replicate their findings. 61Two recent studies have used machine learning to differentiate rs-fMRI signal from successful cognitive agers and controls in participants 60 and older 62 and 80 and older. 63However, results are difficult to interpret because neither study has longitudinal data to ensure that participants were free from emergent neurodegenerative disease.
In summary, the role of resting-state networks in memory and aging is not well defined and the SuperAger phenotype provides a unique opportunity to understand its role in memory preservation beyond the eighth decade.The few studies that have investigated rs-fMRI in successful cognitive aging have had mixed results and no study has longitudinally monitored progressive cognitive decline.This study includes carefully characterized groups of cognitively stable SuperAgers and OACs over two or more visits to investigate the baseline functional integrity of seven canonically defined resting-state networks and subregions of the DMN. and non-memory domains.Tests included the Rey Auditory Verbal Learning Test (RAVLT) for episodic memory, 30-item Boston Naming Test (BNT-30) 65 for picture naming, Trail Making Test Part B (TMT-B) 66 for executive functioning and Category Fluency Test 67 for verbal fluency.Participant neuropsychological performance is summarized in Table 1.

Group criteria
In accordance with the criteria operationally defined for SuperAgers in the study by Harrison et al., 2012, SuperAgers performed above the average range for their peer age normative group on the RAVLT delayed recall (raw score, ≥ 9), at least as good as normative scores for adults in their 50s and 60s, 68 and at least in the average range compared to same-aged peers on the three other tests loading on non-memory cognitive domains (BNT, TMT-B and Category Fluency Test).OAC scored within one standard deviation of average compared to same-aged peers on the memory measure (RAVLT delayed recall score between 3 and 7) and at least in the average range on other cognitive tests.Group classification criteria are summarized in Table 2. Participants maintained group status at consecutive study visits occurring approximately two years apart (mean = 1.7 years; range = 1.16-2.29 years).

Inclusion criteria
Participants with T 1 -weighted (T1w) scans and rs-fMRI scans were considered for this study.Of these participants, we identified those with stable neuropsychological profiles over two visits and excluded those whose group status changed over that interval of time (e.g.participants who developed mild cognitive impairment).MRIs were collected during both visits.Data from the baseline MRI were used in our analysis where possible; data from the second MRI was used only if the baseline scan was unavailable or unusable.We excluded participants with scans containing artefacts in both scans (e.g.magnetic susceptibility, motion, aliasing).In total, 40 participants [mean age (SD): 84.5 (3.2)] were identified for inclusion in the present study.Participant demographics are summarized in Table 1 and a flow chart detailing cohort selection is shown in Fig. 1.

Imaging protocol
MRI scans were acquired for all participants with a standard 12-channel birdcage head coil on a Siemens 3T

Structural imaging
T1w images underwent volumetric segmentations and surface reconstruction by FreeSurfer (v7.2).Trained technicians visually inspected and made iterative edits to optimize FreeSurfer processing.Volumetric segmentations and surface reconstructions were used for registration during rs-fMRI processing.

fMRI quality assessment
Motion artefacts in rs-fMRI are concerning for confounds in older populations. 69,70Quality checking measures were used to censor time series motion artefacts by accounting for deviation in frame-to-frame motion and signal.Respectively, framewise displacement (FD) is a six-dimensional metric of instantaneous head motion calculated from frame-to-frame and DVARS is the relative change in signal from frame-to-frame.We used the eXtensible Connectivity Pipeline 71,72 71,72 were inspected to ensure suitable completion of preprocessing steps.Quality checking metrics are summarized in Table 1.

fMRI processing
Minimal functional MRI preprocessing was performed using fMRIPrep 73 with custom methodologies.fMRIPrep preprocessing included slice-time correction, motion correction using affine registration to the middle time-point, co-registration to the T1w image and resampling into standard space using a single interpolation step.Following minimal MRI preprocessing, XCP 71,72 was used to post-process rs-fMRI.XCP post-processing includes the removal of initial rs-fMRI volumes, outlier detection and filtering, de-spiking and interpolation and a bandpass filter (0.01-0.1 Hz) to reduce low-frequency drift and high-frequency noise in the signal.Global signal, the first principal components from cerebral spinal fluid and white matter, and frame-to-frame motion in six degrees of freedom calculated during motion correction were regressed out to reduce physiologic noise. 74he use of global signal regression is controversial and may remove real neural signal. 75Therefore, all analyses were replicated using rs-fMRI without global signal regression.][79][80] To calculate within-network FC of the DMN, executive control, limbic, ventral attention, dorsal attention, somatomotor and visual networks, Connectome Workbench was used to extract residual signal from all parcels of a seven network (see Fig. 2A), 400 region-of-interest (ROI) parcellation (Schaefer et al., 2018), of a common group-level resting-state atlas (Yeo et al., 2011), resulting in 400 time-series. 25,81ubsequently, pair-wise FC (i.e.Pearson's r converted to Fisher's z) between all ROI time-series were computed to create connectivity matrices (400 × 400).Within-network FC was defined as the average z-value for all pairs within a given  81 To calculate the DMN subcomponent FC, ROI-ROI correlation coefficients were calculated from six bilateral DMN canonical regions: the IPL, PCC, MPC, PHC, MTG and hippocampus.Correlations were calculated between each possible pair, for a total of 30 correlations (15 unique correlations) per person.To ensure that ROIs were completely confined to their predefined regions, cortical areas of interest were defined as the central polygon for each DMN region from the seventeen network parcellations subdivided into 600 parcels 81 (Supplemental Fig. 1).The hippocampal ROI was taken from the Human Connectome Project (HCP) subcortical CIFTI atlas. 78All ROIs were chosen to include homologous regions on both hemispheres.ROI placement is shown in Fig. 3A and cortical ROI identification values (corresponding to the atlas CIFTI metadata) are provided in Supplemental Table 1.Code to generate the custom CIFTI parcellation of cortical ROIs used in the DMN analysis is provided in the Supplemental Material.

Statistical analysis
Independent sample two-tailed t-tests and x 2 tests were used to examine group differences in demographic factors, MRI quality metrics and neuropsychological measures.One ambidextrous SuperAging participant was excluded from the handedness x 2 analysis because no OACs were ambidextrous.The Pearson correlation was calculated between the time-series for each possible pair of ROI.For the 400-parcel atlas, this generated 79 800 unique coefficients which were averaged within-network to create seven average within-network coefficients for every participant.Average between-network connectivity was also computed for each participant (Supplemental Fig. 2).Average within-network FC was defined for each participant as the average of all coefficients between two ROIs of the same network (2618 coefficients per participant).For each participant, average between-network FC was defined as the average FC of all ROIs belonging to separate networks (77 182 coefficients per participant).Finally, system segregation was defined for each participant as the difference between within-network FC and between-network FC divided by within-network FC.
For subcomponents of the DMN, we generated 15 unique coefficients for every participant.Pearson's coefficients were converted to Fisher's z-transformed values for all analyses.After ensuring normality within groups, 82 each withinnetwork coefficient and DMN subcomponent z-value was compared across groups using two-sample independent t-tests with α's adjusted for multiple comparisons using the Benjamini-Hochberg false discovery rate (FDR) at q = 0.05. 83All statistical analyses were performed within RStudio (version 2023.06.1 + 524).

Data inclusion
At the time of our analysis, the SuperAging Research database included 92 participants with T1w scans and rs-fMRI scans from a Siemens 3T MAGNETOM TIM Trio scanner.Of those, 28 were excluded from our analysis due to unstable neuropsychological profiles across two research visits (n = 21) or unclear neuropsychological group profile (n = 7).An additional 24 were excluded due to artefacts in T1w data that made FreeSurfer segmentation fail (n = 6), contained rs-fMRI motion that surpassed FD thresholds (n = 17), or had insufficient useable imaging data (n = 1).Twenty-four SuperAgers and 16 OACs (n = 40) had MRI data and longitudinal neuropsychological profiles that met inclusion criteria for analysis in the present study (Fig. 1).

Demographic and neuropsychological profiles
There were no significant between-group differences in demographic characteristics (all P-values > 0.05) including age, sex, handedness, education or time between research visits.Performance on neuropsychological measures was significantly different between SuperAgers and OACs for episodic memory as measured by the RAVLT delay (t = −11.85;P < 0.01), generative fluencies as measured by category fluency test (t = −2.49;P = 0.02) and confrontation naming as measured by the BNT-30 (t = −2.36;P = 0.03).The between-group difference in RAVLT delay performance is expected due to predefined group classification criteria.There were no significant between-group differences in the overall premorbid functional abilities as measured by the WTAR estimated Full-Scale Intelligence Quotient [FSIQ (t = −1.15;P = 0.26)] or executive functioning as measured by TMT-B (t = 1.38;P = 0.18).Statistical tests for group differences in demographic and neuropsychological profiles are summarized in Table 1.

Functional connectivity analysis
After motion scrubbing, there were no significant between-group differences in mean FD or scan length.Group summary values are included in Table 1.
Average within-network FC of seven resting-state networks were compared between SuperAgers and OACs.There were no significant differences between SuperAgers and OACs in within-network FC of the resting-state networks, including the DMN, executive control, limbic, ventral attention, dorsal attention, somatomotor and visual networks (P-values > FDR adjusted α; Fig. 2B).Similarly, there were no significant group differences in betweennetwork FC (Supplemental Fig. 3).Additionally, broader measures of FC compared between SuperAgers and OACs, including average within-network FC from all networks, average between-network FC from all networks, and system segregation, did not differ significantly between groups (P-values > 0.05; Fig. 4).
FC between subregions of the DMN, including the IPL, PCC, MPC, PHC, MTG and hippocampus, were compared between SuperAgers and OACs.There were no significant between-group differences in FC of the four DMN ROIs after adjusting for FDR (P-values > FDR adjusted α; Fig. 3B).Notably, even before adjusting for FDR, we were unable to find significant between-group differences.Replication of all analyses without global signal regression similarly did not reveal significant group differences.Figures contain results from analyses with global signal regression.

Discussion
The present study compared FC within seven canonical resting-state networks and between major regions of the DMN in SuperAgers and OACs.Results showed no significant group differences in FC between groups for any networks or regions of the DMN.The relationship between the large-scale resting-state networks and the spectrum of cognitive aging is complex and group-based average measurements of FC do not appear to explain the exceptional memory performance observed in SuperAgers.Potential contributors to these results, including the discrepancy between our findings and those that have found relatively strong within-network FC of successful cognitive agers, 34 are likely multifactorial.
One possible contributor to the difference between our results and those of previous studies is our longitudinal inclusion criteria to reduce the risk of undiagnosed neurodegenerative processes among our participants.Of the 92 participants from the SuperAging Research Program with MRI considered for our study, 21 exhibited unstable neurocognitive profiles and were excluded.Participants with unstable neurocognitive profiles are not commonly accounted for in alternate successful cognitive aging rs-fMRI studies 59 in part because the studies only have access to cognitive data from a single time-point.Participants with declining neurocognitive profiles could inadvertently drive differences in within-or between-network FC due to underlying ADRD.As such, group differences observed in prior studies may have been driven by ADRD.This highlights the importance of careful consideration of participant profiles in future studies aiming to elucidate the role of FC in cognitive aging.

Figure 2
Figure 2 Within-network resting-state functional connectivity does not differ between groups.(A) 400-parcel cortical parcellation of seven large-scale resting-state networks from Schaefer et al. (2018) used for within-network analysis.(B) Average within-network functional connectivity across resting-state networks did not differ between SuperAgers and older-aged controls; OAC = older-aged controls.Wilcoxon signed-rank test; P > 0.05 for all comparisons.

Figure 3
Figure 3 Functional connectivity between subcomponents of the DMN do not differ between SuperAgers and controls.(A) Regions-of-interest (ROIs) used for functional connectivity of the default mode network (DMN); Left: Five central polygons from DMN clusters of the 600-parcel cortical parcellation from Schaefer et al., 2018; Right: Hippocampus (Hipp) seed from Glasser et al., 2013.(B) Two-sample independent t-tests found no significant between-group difference in functional connectivity; IPL = inferior parietal lobe; PCC = posterior cingulate cortex; MPC = medial prefrontal cortex; PHC = parahippocampal cortex; MTG = middle temporal gyrus.Wilcoxon signed-rank test; P > 0.05 for all comparisons.

Table 1 Participant demographic and cognitive characteristics
One SuperAger was ambidextrous and not included in the x 2 analysis.b One SuperAger met all SuperAging criteria with the exception of the BNT-30, but scored within expectation in a subsequent visit, 6 months later. a