Skip to main content
NCBI home page
Cerebral Cortex (New York, NY) logoLink to Cerebral Cortex (New York, NY)
. 2021 Jul 13;32(2):397–407. doi: 10.1093/cercor/bhab215

Person-Based Similarity Index for Cognition and Its Neural Correlates in Late Adulthood: Implications for Cognitive Reserve

Anna West 1,2, Noah Hamlin 2,2, Sophia Frangou 3,4, Tony W Wilson 5, Gaelle E Doucet 6,✉
PMCID: PMC8754370  PMID: 34255824

Abstract

Healthy aging is typically associated with some level of cognitive decline, but there is substantial variation in such decline among older adults. The mechanisms behind such heterogeneity remain unclear but some have suggested a role for cognitive reserve. In this work, we propose the “person-based similarity index” for cognition (PBSI-Cog) as a proxy for cognitive reserve in older adults, and use the metric to quantify similarity between the cognitive profiles of healthy older and younger participants. In the current study, we computed this metric in 237 healthy older adults (55–88 years) using a reference group of 156 younger adults (18–39 years) taken from the Cambridge Center for Ageing and Neuroscience dataset. Our key findings revealed that PBSI-Cog scores in older adults were: 1) negatively associated with age (rho = −0.25, P = 10−4) and positively associated with higher education (t = 2.4, P = 0.02), 2) largely explained by fluid intelligence and executive function, and 3) predicted more by functional connectivity between lower- and higher-order resting-state networks than brain structural morphometry or education. Particularly, we found that higher segregation between the sensorimotor and executive networks predicted higher PBSI-Cog scores. Our results support the notion that brain network functional organization may underly variability in cognitive reserve in late adulthood.

Keywords: brain functional connectivity, brain structural morphometry, cognitive reserve, late adulthood, person-based similarity index

Introduction

Healthy aging is commonly associated with slight declines and slowing in general cognitive function, even in those who have not been diagnosed with dementia or any other form of neurodegenerative disorder (Deary et al. 2009; Daffner 2010). Yet, there is substantial variation in cognitive performance among older adults (Morse 1993; Ardila 2007; Foss et al. 2009; Costa et al. 2013). For example, some older adults continue to perform at a level similar to younger adults, whereas others will show extensive decline (Harrison et al. 2012; Rogalski et al. 2013; Gefen et al. 2014; Sun et al. 2016). The mechanisms behind differences in healthy cognitive aging remain unclear and are likely multifactorial (Deary et al. 2009; Daffner 2010; Depp et al. 2012).

At the behavioral level, clustering approaches have consistently demonstrated that older adults are cognitively heterogeneous and can be further categorized into smaller, more homogeneous groups based on a range of cognitive performance scales covering general cognition, executive function, and/or memory (Ardila 2007; Foss et al. 2009; Costa et al. 2013). At the biological level, age-related variability in cognitive performance has been linked to changes in the dynamics of information processing and neural response amplitude, which may be the result of age-related decreases in the integrity of neuronal membranes, altered metabolic functions, and cell death (Hipkiss 2006; Terman et al. 2006; Leuner et al. 2007; Tosato et al. 2007; Daffner 2010).

There is extensive evidence of changes in both structural and functional brain organization throughout the lifespan (Betzel et al. 2014; Dima et al. 2021; Doucet et al. 2021; Frangou et al. 2021), yet a link between these changes and variability in cognitive decline remains relatively unclear (Habeck et al. 2017; Nilsson and Lovden 2018). Most studies focusing on the association between brain and cognition in late adulthood have tried to link it with interindividual differences in cognitive reserve (Stern 2002; Nilsson and Lovden 2018). Cognitive reserve has been defined as the ability to optimize or maximize performance through differential recruitment of brain networks and other compensatory mechanisms (Stern 2002; Nyberg et al. 2012). However, one major limitation in this context is the lack of quantitative measures to evaluate it objectively.

Herein, we propose a new metric, the “person-based similarity index” score for cognition (PBSI-Cog), and use it to quantify the similarity between the cognitive profile of an older participant and those of younger healthy participants. Accordingly, high PBSI-Cog values indicate that the cognitive profile of any one older adult is similar to those of younger adults, which may reflect higher cognitive reserve. In contrast, low PBSI-Cog values signify low similarity (i.e., more deviation) in the cognitive profile of the older relative to younger adults (i.e., reflecting lower cognitive reserve). We hypothesized that the PBSI-Cog score would: 1) decrease with older age, as both cognitive decline and cognitive variability increase with age (Ardila 2007; Chan et al. 2018); 2) be predicted by fluid intelligence and executive function variables (Daffner 2010); and 3) be predicted by variation in brain structural morphometry and network functional connectivity. Particularly, we expected network functional connectivity to be a better predictor than brain structural metrics (Li et al. 2020; Pistono et al. 2020), and that the sensorimotor and executive networks would be substantial contributors to our cognitive proxy measure based on previous studies showing an age-related increase in shared variance between cognition, speed and sensory abilities (Li and Lindenberger 2002). Lastly, because education is a strong proxy of cognitive reserve (Stern et al. 1994; Nilsson and Lovden 2018), we also tested our hypotheses against this standard predictor. To test these predictions, we utilized the structural (sMRI), resting-state functional magnetic resonance imaging (rs-fMRI), and cognitive data from healthy individuals participating in the Cambridge Center for Ageing and Neuroscience project (CamCAN; Shafto et al. 2014; Taylor et al. 2017).

Materials and Methods

Sample

The CamCAN study used an epidemiologically-informed recruitment framework and is available upon request (https://www.cam-can.org/; Shafto et al. 2014; Taylor et al. 2017). It includes a total of 652 individuals (330 females, age range: 18–88 years). From this cohort, we selected 2 subsets based on an age cutoff: one composed of older healthy adults (age range: 55–88 years, referred to as “CamCAN55+”), and one composed of younger healthy adults (age range: 18–39 years, referred to as “CamCAN40−”), with this latter defined as the reference group. The age cutoff of 55 years for the older subset was based on the age criterion used in the Alzheimer’s disease neuroimaging initiative (ADNI; Jack et al. 2008; Petersen et al. 2010).

For the CamCAN55+ subset, we selected 326 individuals (49% of females, mean age [standard deviation, SD] = 71.8 [9.3] years). Among this subsample, we excluded 89 participants from further analyses because they: 1) reported neurological or neuropsychiatric disorders (n = 74), 2) did not have both rs-fMRI and structural MRI data available (n = 2), 3) had excessive head motion during the rs-fMRI scan (n = 4, head motion above 2-mm translation or 1° rotation), and/or 4) had at least two-thirds of missing neuropsychological data (n = 9). The final sample of older adults used in the current study was 237 (48.5% of females, mean age [SD] = 70.9 [9.1] years). For the CamCAN40− subset, we selected 169 healthy individuals (52% of females, mean age [SD] = 30.6 [5.8 years]). For this reference subset, only the cognitive assessment was used. Among this subset, we excluded 13 participants because they had at least two-thirds of missing neuropsychological data, resulting in a final sample of 156 young individuals (46.8% of females, mean age [SD] = 30.9 [5.7] years). The CamCAN55+ subgroup was statistically comparable with the reference CamCAN40− subgroup in sex distribution (chi-squared test, P = 0.4; Table 1).

Table 1.

Sociodemographic characteristics of the individuals in each CamCAN subsample

Measure CamCAN40− N = 156 CamCAN55+ N = 237
Age (years) 30.9 [5.7] 70.9 [9.1]
Female sex 73 [46.8%] 115 [48.5%]
Education (degree)
 College degree or higher 121 [77.56%] 59 [24.9%]
 A-level 22 [14.1%] 114 [48.10%]
 GCSE/O-level 13 [8.33%] 36 [15.19%]
 None 0 [0%] 28 [11.81%]
Socioeconomic status (occupation)
 I—professional 39 [25%] 41 [17.30%]
 II—intermediate 58 [37.18%] 102 [43.04%]
 IIIN—skilled non-manual 13 [8.33%] 32 [13.50%]
 IIIM—skilled manual 30[19.23%] 45 [18.99%]
 IV—partly skilled 8[5.13%] 12 [5.06%]
 V—unskilled 0 [0%] 4 [1.69%]
 Not specified 8 [5.13%] 1 [0.64%]
Employment*
 1—Paid or self-employed 136 [87.18%] 83 [35.02%]
 2—Retired 24 [15.38%] 165 [69.62%]
 3—Looking after home/family 5 [3.21%] 39 [16.46%]
 5—Unemployed 3 [1.92%] 1 [0.42%]
 6—Doing unpaid or voluntary work 8 [5.13%] 29 [12.24%]
 7—Student 5 [3.21%] 0 [0%]
Open in a new tab

Notes: Continuous variables are shown as mean (SD); Categorical variables are shown as number (percentage. %); CamCAN = Cambridge Centre for Ageing and Neuroscience; additional cognitive variables are provided in Supplementary Tables S1 and S2.

* Some participants reported > 1 employment, which led to a total sum above 100% across participants.

Selection of the Cognitive Variables

The CamCAN project provides cognitive assessment data collected outside of the MRI scanner (Shafto et al. 2014, 2019; Samu et al. 2017). For the current study, we used 36 variables covering fluid intelligence, language, executive function, visual working memory, and motor function. Details of the tasks and the variables are provided in Supplementary Table S1. Across all tasks, we excluded categorical variables where >90% of the sample endorsed the same outcome (n = 48), that were colinear (r > 0.9, n = 114), or that had >10% of missing values (n = 101) in the 2 subsets combined. For psychometric tests with multiple correlated outcome variables we selected those that are more commonly reported in the literature. For variables with <10% of missing values, we used expectation maximization (EM) imputation implemented in SPSS version 25. The EM algorithm maximizes the log-likelihood of the available data, with missing data marginalized, so that the log-likelihood for the full data (available plus missing) is greater than that for the available data only.

MRI Acquisition

Resting-state fMRI data were acquired while participants rested with their eyes closed on a 3T Siemens TIM Trio scanner with a 32-channel head coil with the following acquisition parameters: time repetition/time echo (TR/TE) = 1970/30 ms, 32 axial slices, flip angle = 78°; field of view (FOV) = 192 mm × 192 mm; voxel-size = 3 mm × 3 mm × 4.44 mm, acquisition time = 8 min 40 s, and number of volumes = 261. The structural MRI data were acquired using a T1-weighted, 3D MPRAGE sequence with the following parameters: TR/TE/time to inversion = 2250/2.99/900 ms, voxel size = 1-mm isotropic, flip angle = 9°, FOV = 256 × 240 × 192 mm3, and duration of acquisition = 4 min 32 s.

Resting-State fMRI Preprocessing

The rs-fMRI data were preprocessed using SPM12 and the DPABI (DPARSF V4.5) Toolbox (Chao-Gan and Yu-Feng 2010; Yan et al. 2016). Preprocessing procedures included: removal of the first 3 volumes, motion correction to the first volume with rigid-body alignment, co-registration between the functional scans and the anatomical T1-weighted scan, spatial normalization of the functional images into Montreal Neurological Institute (MNI) stereotaxic standard space, spatial smoothing within the functional mask with a 6-mm at full-width at half-maximum Gaussian kernel, wavelet despiking (Patel et al. 2014), linear detrending, and regression of motion parameters and their derivatives (24-parameter model; Friston et al. 1996) as well as white matter (WM) and cerebrospinal fluid (CSF) signals. The WM and CSF signals were computed using a component based noise reduction method (CompCor, 5 principal components; Behzadi et al. 2007). Lastly, bandpass filtering was applied at [0.01–0.1] Hz (Cordes et al. 2001).

Extraction of Functional Connectivity Measures

We used Atlas55+, a previously established functional brain atlas specifically designed for late adulthood (Doucet et al. 2021), to partition the functional connectome into the 5 most replicated resting-state networks (RSNs) and their 15 subnetworks: default mode (DMN; 4 subnetworks), salience (SAL; 1 subnetwork), executive control (ECN; 4 subnetworks), sensorimotor (SMN; 4 subnetworks), and visual (VIS; 2 subnetworks; detailed in Supplementary Table S2). In each participant, Fisher Z-transformed Pearson’s correlation coefficients were computed to calculate functional connectivity within- and between- networks. Within-network functional connectivity (WN-FC) was computed as the average correlation of each voxel’s blood oxygen level-dependent (BOLD) signal time series with every other voxel within the network. Between-network functional connectivity (BN-FC) was computed as the correlation between the average time-series of each pair of networks. We extracted these measures for both the 5-network and the 15-subnetwork parcellations.

Structural MRI Preprocessing

Across all participants, parcellation and segmentation of the sMRI datasets was implemented in FreeSurfer 6.0 (http://surfer.nmr.mgh.harvard.edu/). The steps included removal of nonbrain tissue using a hybrid watershed/surface deformation procedure (Segonne et al. 2004), automated Talairach transformation, segmentation of the subcortical WM and deep gray matter volumetric structures (Fischl et al. 2002, 2004), intensity normalization (Sled et al. 1998), tessellation of the boundary between the gray and WM, automated topology correction (Fischl and Dale 2000; Segonne et al. 2007), and surface deformation following intensity gradients to optimally place the gray/WM boundaries and gray/CSF borders at the location where the greatest shift in intensity defines the transition to the other tissue class. All participants’ data passed the quality control protocols developed by the ENIGMA initiative (http://enigma.ini.usc.edu/).

Extraction of Subcortical Volume and Cortical Thickness Measures

Following FreeSurfer segmentation and parcellation based on the Desikan atlas (Desikan et al. 2006), we obtained 68 regional cortical thickness measures, 16 subcortical volume measures, and 3 global measures (right hemisphere cortical thickness, left hemisphere cortical thickness, and intracranial volume [ICV]) for each participant.

Computation of the PBSI-Cog Scores

For each participant, regardless of subgroup, their cognitive profile was defined by concatenating the 36 cognitive variables (detailed in Supplementary Table S1) into a single vector. In order to evaluate the degree of cognitive reserve in the older CamCAN55+ adults, we computed a PBSI score for cognition (PBSI-Cog). This score has been described in several previous works (Doucet et al. 2019, 2020a; Janssen et al. 2020). The PBSI-Cog score, as computed here, provides an estimation of the degree of similarity between the cognitive profiles of each participant in the CamCAN55+ group compared with that of CamCAN40− participants. Computation followed a 3-step procedure. First, for each participant, regardless of subgroup, their cognitive profile was defined by concatenating 36 cognitive variables into a single vector. Second, we calculated the Spearman’s correlation coefficient between the cognitive profile of each CamCAN55+ participant with the profile of each of the CamCAN40− participants (n = 156); for each CamCAN55+ participant, this process generated 156 interindividual correlation coefficients. Third, for each CamCAN55+ participant, these interindividual correlation coefficients were averaged to yield one score referred to as the PBSI-Cog. Higher scores (with a maximum of 1) denote greater similarity between the profile of each CamCAN55+ participant and the profiles of the CamCAN40− participants. The PBSI-Cog score was computed using an in-house script in Matlab2018b.

The contribution of each of the 36 cognitive variables to the PBSI-Cog score was investigated using a leave-one-out approach. Specifically, the PBSI-Cog score of each participant was recalculated after leaving out one cognitive variable at a time. This process yielded 36 recalculated PBSI-Cog scores per CamCAN55+ participant. We then conducted paired t-tests between the original PBSI and each recalculated PBSI score to quantify the contribution of each variable. A larger positive contribution meant that the recalculated PBSI-Cog decreased more, that is, the cognitive variable contributed more to the original PBSI-Cog score. A more negative contribution value meant that excluding the cognitive variable increased the recalculated PBSI-Cog more, that is, the variable had a larger negative impact on the original PBSI-Cog score.

Neuroimaging Correlates of the PBSI-Cog in the CamCAN55+ Subgroup

In the CamCAN55+ subgroup only, we conducted 3 separate stepwise selection regression analyses in order to evaluate the power of different types of neuroimaging variables versus more typical predictors (i.e., age, sex, and education) in predicting the PBSI-Cog score. The 3 models included respectively: 1) structural morphometric measures only (i.e., cortical thickness, subcortical volumes and global measures (i.e., ICV, left and right hemisphere cortical thickness); n = 87), 2) structural morphometric measures (i.e., cortical thickness, subcortical volumes, and global measures), and functional measures of the 5-network parcellation from Atlas55+ (WN-FC and BN-FC; n = 102), 3) structural morphometric measures (i.e., cortical thickness, subcortical volumes and global measures), and functional measures of the 15-subnetwork parcellation from Atlas55+ (WN-FC and BN-FC; n = 192). The 2 parcellations of Atlas55+ were analyzed separately because the 15 subnetworks are direct subdivisions of the 5-network parcellation, leading to high collinearity. We also tested using the subcortical volumes after adjusting for variation in ICV and the results remain the same.

Age, sex, and education level were also added in each model as independent variables. A regression using a stepwise selection is based on a mixture of forward and backward elimination. It operates by successively adding and removing predictor variables in order to select the best grouping of predictor variables that accounts for most of the variance (R2) in the outcome. In other words, the model starts with zero predictors and then adds the strongest predictor, to the model if its b-coefficient is statistically significant (P < 0.05). It then adds the second strongest predictor, and so on. Because some iterations may render previously entered predictors nonsignificant, some may be removed during an iteration. This process continues until none of the excluded predictors contributes significantly to the included predictors. Stepwise regression analyses are prone to R2 value inflations and are sensitive to the presence of collinearity amongst the predictor variables. To address these concerns, we confirmed that none of the regression results showed collinearity based on variance inflation factor (VIF) < 2 and tolerance > 0.2. The regression analysis was considered significant if it reached a P-value < 0.05, after applying a Bonferroni correction (correcting for the 3 models tested). Linear regression analyses were conducted in SPSS version 25.

Subsidiary Analyses

We conducted a series of supplementary analyses to ensure the reproducibility and reliability of the PBSI-Cog scores. First, we recomputed the PBSI-Cog score 100 times for all CamCAN55+ individuals, after randomly selecting a subset of 50% within the CamCAN40− individuals, in order to confirm that the PBSI-Cog scores were not dependent on the reference sample selection. We then compared the recomputed scores with the original scores using Spearman correlation analysis. Second, we computed the effect size (Cohen’s d) for each cognitive variable between the CamCAN55+ and CamCAN40− participants. We then tested the association between the variables’ effect size and their respective contribution to their PBSI-Cog score, conducting Pearson’s correlation analysis. Last, we computed linear regression analyses using fluid intelligence (Cattell total score, see Supplementary Table S1) as the dependent variable, instead of the PBSI-Cog score, in order to test the degree of specificity of the findings related to the PBSI-Cog score. Since the Cattell score is an absolute measure of cognitive ability, and the PBSI-Cog score is a relative index of cognitive performance, we expected that these analyses would provide a different set of significant predictors, particularly involving FC of the VIS, SMN, and DMN (Wen et al. 2020).

Results

Demographic Information and PBSI-Cog Scores in the CamCAN55+ Subgroup

Spearman’s correlation indicated a significant negative association between age and PBSI-Cog scores (Spearman ρ = −0.25, P = 10−4, Fig. 1). No sex differences in PBSI-Cog scores were detected (Mann–Whitney U test, P = 0.98). Individuals with a college degree showed significantly higher PBSI-Cog scores than those without (t = 2.4, P = 0.018). No significant differences in PBSI-Cog scores were detected based on employment or socio-economic status (all P > 0.05).

Figure 1 .

Figure 1

Open in a new tab

Association between age and PBSI-Cog scores in the CamCAN55+ subgroup.

Cognitive Variable Contribution to the PBSI-Cog Scores in the CamCAN55+ Subgroup

In the leave-one out analyses, variables related to executive function, fluid intelligence, and language comprehension were major positive contributors to the PBSI-Cog score, which indicates that the presence of these variables increase the PBSI-Cog score (Fig. 2). In contrast, variables related to language retrieval seemed to have a negative impact on the PBSI-Cog score. However, because all these variables reflected negative outcomes in regard to language retrieval (i.e., higher scores reflected increased errors), these data indicated that better language retrieval was associated with higher PBSI-Cog score.

Figure 2 .

Figure 2

Open in a new tab

Individual contribution of the cognitive variables to the PBSI-Cog score, in the CamCAN55+ subgroup. A positive T-score indicates a positive contribution (i.e., variable helps increase the PBSI score); whereas a negative T-score indicates a negative contribution (i.e., variable helps lower the PBSI score). Cognitive domains are coded by pattern. Details of the variables are in Supplementary Table S1. *P < 0.05.

Neuroimaging Predictors of the PBSI-Cog Scores in the CamCAN55+ Subgroup

Structural Morphometric Measures Only

The best model explained 6% of the variance in PBSI-Cog, with the volume of the left amygdala and the thickness of the right paracentral lobule being the only positive contributors to the model. The thickness of the right inferior temporal gyrus was observed to be a negative significant contributor (Table 2).

Table 2.

Summary of stepwise regression analyses for neuroimaging predictors of PBSI-Cog score in older adults

Variables B (SE) β Adjusted R2 F P-value
Model with structural morphometric measures only 6% 6.188 4.65E−04
 Left amygdala volume 4.290E−05 (1.56E−05) 0.181
 Right paracentral thickness 6.07E−02 (2.19E−02) 1.930
 Right inferior temporal thickness −.053(.025) −0.015
Model with structural morphometric and functional connectivity measures (5-network atlas) 9% 8.559 2.10E−05
 Left amygdala volume 4.696E−05 (.00015) 0.198
 BN-FC SMN–VIS 0.038 (0.011) 0.219
 Education 0.016 (0.007) 0.155
Model with structural morphometry and functional connectivity measures (15-subnetwork atlas) 17% 7.473 4.44E−08
 Left amygdala volume 4.653E−05 (1.48E−05) 0.196
 WN-FC VIS2 −0.120 (0.035) −0.211
 BN-FC ECN2–SMN1 −0.039 (0.019) −0.133
 BN-FC ECN3–SMN3 0.049 (0.017) 0.186
 BN-FC SAL1–SMN2 −0.053 (0.017) −0.205
 Right paracentral thickness 0.045 (0.020) 0.144
 BN-FC ECN2–SAL1 −0.036 (0.016) −0.147
Open in a new tab

Note: Detail of the networks and their subdivisions can be found in Supplementary Table S2.

Structural Morphometric and Functional Connectivity (5-Network Atlas55+) Measures

The best model explained 9% of the variance in PBSI-Cog, with the volume of the left amygdala, and the functional connectivity between the SMN and VIS network being significant positive contributors. To a lower level, education was also a positive contributor to the PBSI-Cog score (Table 2 and Fig. 3).

Figure 3 .

Figure 3

Open in a new tab

Schematic presentation of the results of regression analyses involving the structural and network functional connectivity (FC) measures as predictors of the PBSI-Cog score in CamCAN55+ subgroup. (A) based on the structural morphometric measures only, (B) based on structural and FC from the 5-network atlas, (C) based on structural and FC from the 15-subnetwork atlas. Lines connecting two networks represent their BN-FC. One network represents WN-FC. For each significant model, straight and dashed lines to the PBSI-Cog box show positive and negative predictors, respectively. The details of the regression analyses are shown in Table 2. More details on each subnetwork (C) can be found in Supplementary Table S2. One color is associated with one network and its subnetwork.

Structural Morphometric and Functional Connectivity (15-Subnetwork Atlas55+) Measures

The best model explained 17% of the variance in PBSI-Cog. The neuroimaging variables that made a statistically positive significant contribution to the model were the volume of the left amygdala, the thickness of the right paracentral lobule, and the functional connectivity between the ECN3 and SMN3 subnetworks (Table 2 and Fig. 3). Conversely, the neuroimaging variables that made a statistically negative contribution to the model were the FC within the VIS2 network, the FC between the subdivisions of the SMN, ECN, and SAL networks (i.e., between SMN1 and ECN2, between SMN2 and SAL1, and between the ECN2 and SAL1).

Subsidiary Analyses

Bootstrap analyses confirmed that the PBSI-Cog scores in older individuals were reproducible with minimal variation when recalculated for random subsets from 50% of the CamCAN40− subgroup. The average correlation between the recalculated and original PBSI scores was 0.999 (SD = 5.10−4). Second, the individual contributions of the cognitive variables to the PBSI-Cog score were not related to the degree of cognitive impairment in the older relative to the younger individuals (r = 0.07, P = 0.7; see Supplementary Fig. S1 and Supplementary Table S3). Third, in order to ensure that our results related to the PBSI scores were not driven by the participants’ global level of intelligence, we reconducted the regression analyses using fluid intelligence as the dependent variable. The results show different sets of predictors for each model and are detailed in Supplementary Table S4.

Discussion

The current study aimed to understand the cognitive changes associated with aging by deriving a proxy of preserved cognitive function, and to evaluate the contribution of functional and structural brain measures to this preservation. Importantly, we did not limit our investigation to one cognitive domain. Instead, we evaluated cognitive change across multiple domains and computed a PBSI-Cog score for each older individual using a younger healthy adult group as a reference. We found that the PBSI-Cog score was not dependent on the reference sample used, was negatively associated with age, was higher in individuals with higher education, and that it was predicted by functional connectivity between the networks supporting lower- versus higher-order cognitive functions, rather than by brain structural morphometry.

The PBSI score was originally derived from brain structural and functional variables and was found to be robust and reliable in our previous works (Doucet et al. 2019, 2020a, 2020b). It quantifies the intersubject similarity between a specific set of variables of an individual to that of other individuals. In the current study aiming to investigate the brain correlates of cognitive variability in late adulthood, the PBSI-Cog score quantified the degree of similarity of each older individual’s cognitive profile, when compared with that of younger healthy individuals. As hypothesized, we found that older age was associated with lower PBSI-Cog scores, meaning that the older the individual in the CamCAN55+ group the less similar their cognitive profile was to that of individuals in the CamCAN40− group. We also reported that individuals with a college degree had a higher PBSI-Cog score than those without. This result is in line with previous studies showing the role of education in cognitive reserve (Stern et al. 1994; Nilsson and Lovden 2018) and further suggests that the PBSI-Cog score may be used as a proxy of cognitive reserve.

We found that not all cognitive variables contributed equally to the PBSI-Cog score. In particular, variables related to executive function, fluid intelligence and language capacity showed a strong positive contribution, relative to the other cognitive domains investigated (i.e., leaving these variables out decreased the PBSI-Cog score). In other words, our findings indicate that lower fluid intelligence, executive function and/or language capacities lead to lower PBSI-Cog scores in older adults (i.e., more variability in the older adults’ cognitive profile, compared with those of younger adults). Executive function and fluid intelligence have been typically shown as declining, whereas verbal skills remained relatively preserved with aging (Buckner 2004; Hedden and Gabrieli 2004; Ardila 2007). Ardila (2007) has investigated the degree of dispersion of several cognitive subtests from the WAIS-III (Wechsler 1997) throughout the lifespan, and found that the degree of cognitive heterogeneity during normal aging was also dependent on multiple cognitive domains. Abilities related to executive functions and attention showed higher cognitive heterogeneity, whereas other abilities such as working memory and visuoconstructive abilities were associated with a more homogenous pattern of decline (Ardila 2007). Furthermore, other behavioral studies have consistently reported that broad cognitive constructs, such as intelligence, were among the strongest predictors of cognitive aging (Daffner 2010). Cognitive reserve has also been linked to higher intellectual capacity and language function (Stern 2002; Nyberg et al. 2012). The results of the current study indicate that higher fluid intelligence, superior executive function, and higher language function predicted higher PBSI-Cog scores, which is one would expect if the PBSI-Cog score was a valid index of cognitive reserve.

We further investigated the role of brain morphometry and resting-state functional connectivity measures as predictors of the PBSI-Cog scores in older adults. The parameters of network functional connectivity considered here were better predictors of the PBSI-Cog scores than the brain morphometry metrics or demographic information such as age and education. It is important to note that education is typically considered a robust proxy of cognitive reserve (Nilsson and Lovden 2018) and our findings that functional connectivity measures proved to be more reliable predictors suggest that rs-fMRI approach may have strong potential to further characterize and quantify the biological mechanisms behind cognitive reserve in late adulthood. It is also worth noting that the left amygdala was reported as one of the only significant structural predictors across models. The role of amygdala in cognitive reserve remains unclear (Soldan et al. 2018) although some studies using animal models have suggested that the amygdala was sensitive to adaptive mechanisms to preserve learning in aged rats (Samson et al. 2017).

The association between functional, structural measures, and cognitive ability in late adulthood is in line with previous studies reporting a stronger association of cognitive performance with network functional connectivity than with brain structural morphometry in healthy older adults (Li et al. 2020; Pistono et al. 2020). Specifically, the functional interactions between the SMN and ECN/SAL networks emerged as key predictors of higher PBSI-Cog scores among older adults. These findings can be better understood within the context of age-related trajectories and organizational principles of human brain network architecture (Doucet et al. 2011; Betzel et al. 2014). At the highest level, the brain functional architecture of healthy adults is intrinsically organized into several large-scale neural networks that constantly interact (Doucet et al. 2011); the ECN typically supports higher order cognitive functions, concerned with maintenance and manipulation of goal-directed mental operations (Curtis and D'Esposito 2003; Smith et al. 2009), while the SAL is involved in switching between directed and undirected mental activity (Menon and Uddin 2010), and the DMN is associated with unconstrained mental activity (Buckner et al. 2008), whereas, the SMN and VIS networks cover primary cortices and support lower-order activities driven by external stimulation. Typically, in young adulthood, the networks related to higher-order functions (i.e., ECN and DMN) work in opposition to the lower-order networks (i.e., SMN and VIS), whereas the SAL plays the role of mediator by being alternatively connected to either group of networks to help information exchange (Menon and Uddin 2010; Doucet et al. 2011). However, late adulthood has been typically associated with both higher interactions between brain networks and weaker integration within networks, leading to a reduction in brain network modularity (Meunier et al. 2009; Betzel et al. 2014; Damoiseaux 2017). Our results indicate that older adults with lower functional connectivity between lower- and higher-order networks had cognitive profiles that were more similar to those of younger adults (i.e., higher PBSI-Cog score). We further infer that larger reductions in functional connectivity between these networks may reflect a more youthful pattern, as these networks will be more segregated as seen in younger adults (Doucet et al. 2011; Betzel et al. 2014). Such interpretation is consistent with the concept of brain resilience to aging processes (Nyberg et al. 2012; Nilsson and Lovden 2018). The current study also further extends structural neuroimaging studies that have described super-agers (i.e., older adults with similar cognitive capacity as young adults) as having a “younger” brain signature (Sun et al. 2016), by indicating the existence of the same mechanisms in the brain functional architecture.

In previous studies, cognitive reserve has been linked to a combination of factors, including genetic, past lifestyle, educational and/or environmental conditions (Nyberg et al. 2012; Habeck et al. 2017; Nilsson and Lovden 2018). Herein, we infer that age-related reductions in brain modularity may be critical to the cognitive heterogeneity observed among older adults, with the impact being proportional to the degree of vulnerability in each individual.

In our current study, the SMN, which supports sensory and motor processing, was a key contributor in both the 5- and 15-network model. Previous studies have shown that sensory and cognitive changes in late adulthood are not independent of each other (Li and Lindenberger 2002). Particularly, both cross-sectional and longitudinal studies have reported an age-associated increase in the link between sensory and intellectual abilities, suggesting that cognitive aging may be attributable to the deterioration of common neurological processes, or that age-related declines in sensory function may directly affect higher-order processing (see review by Li and Lindenberger (2002)). When considered together, our observations are in line with this interpretation and further suggest that the growing interdependence between cognitive and sensorimotor domains in later life may be due to functional changes in the SMN and its interactions with higher-order cognitive networks, which may lead to an altered reallocation of resources between these specific networks and thereby cognitive decline.

Lastly, the DMN was not a key predictor of PBSI-Cog scores in late adulthood, despite its major role in the functional architecture of the brain and cognition (Buckner et al. 2008). We believe this may reflect that the PBSI-Cog score did not include variables related to episodic memory, which is typically supported by the DMN (Buckner et al. 2008; Andrews-Hanna et al. 2010). Alternatively, this may reflect that DMN functional connectivity is more associated with absolute (rather than relative) cognitive performance in older adults (Andrews-Hanna et al. 2007; Huo et al. 2018), as was also suggested by our supplementary analyses (see Supplementary Table S3). Examination of other cognitive domains, including episodic memory, and their contribution to the PBSI-Cog score in late adulthood may provide a more definitive answer among these alternatives.

To our knowledge, this is one of the first studies focused on developing a quantitative proxy of cognitive reserve using a person-based approach. The results reinforce the importance of the functional network interaction in healthy cognitive aging. However, it is important to acknowledge the limitations of this study. First, although our results show that higher PBSI-Cog scores are associated with higher education and higher cognitive scores (e.g., fluid intelligence, executive function, and language ability), we cannot be certain on whether it is a proxy of cognitive reserve. It will be important to reproduce these findings and further test this measure in independent datasets and against other typical predictors such as cognitive reserve score and other IQ measures. Longitudinal studies of healthy mid- to late-life individuals would be especially helpful in understanding the relationship between PBSI-Cog scores and cognitive reserve. In fact, the cross-sectional nature of the current study is a limitation, as it does not allow us to account for intersubject differences in other factors, such as life experience, or environmental conditions that may alter the degree of cognitive reserve across older individuals (Ylikoski et al. 1999; Ardila 2007; Foss et al. 2009; Depp et al. 2012; Costa et al. 2013; Chan et al. 2018). However, to our knowledge, there are currently no publicly available datasets that include longitudinal cognitive data from early to late adulthood. Second, because we applied a careful quality control on the cognitive variables selected to compute the PBSI-Cog score, we had to exclude variables related to other cognitive domains, such as episodic memory, because they had numerous missing points. It will be important to extend the PBSI-Cog throughout the lifespan by using larger cognitive datasets, including other cognitive domains and data derived from different neuropsychological tests. We also note that the CamCAN dataset includes other neuroimaging modalities (e.g., diffusion weighted imaging and magnetoencephalography) that could be incorporated, and that other analytic methods such as graph theory were not considered here. We chose to focus on structural MRI and rs-fMRI as these represent the most commonly used neuroimaging approaches in late adulthood and are also easier to implement because of their relative brevity and lack of requirement for active engagement. Lastly, to identify the brain networks, we used a specific functional brain atlas of RSNs which may affect the results. However, Atlas55+ was chosen because it is the only atlas derived from brain networks of older healthy adults (aged range 55–95), and therefore has been specifically constructed to investigate brain networks in late adulthood and reduce any age-related influence on the anatomical definition of the networks (Doucet et al. 2021).

Conclusion

The current study investigated the origins of cognitive aging and particularly sustained cognitive function in healthy late adulthood. The study provides new evidence that the preservation of cognitive function in older adults is largely supported by variability in the functional interactions among higher- versus lower-order networks, rather than brain structural atrophy. We believe this preservation of function is central to cognitive reserve and that the PBSI-Cog may be useful as a quantitative proxy of this important construct. Our findings also provide novel, putative, mechanistic insights into cognitive reserve among older individuals. These neural network phenotypes should be further explored in future studies, with regards to their importance in other populations who exhibit cognitive heterogeneity, such as those with neuropsychiatric disorders (Joyce and Roiser 2007; Depp et al. 2012; Feczko et al. 2019; Vaskinn et al. 2020).

Funding

Funding for this work was provided by the National Institute on Aging (R03AG064001); the National Institute of Mental Health (R01MH116782, RF1MH117032, R01MH118013); the National Institute on Drug Abuse (R01DA047828). The CamCAN data collection was provided by the UK Biotechnology and Biological Sciences Research Council (grant no. BB/H008217/1), together with support from the UK Medical Research Council and University of Cambridge, United Kingdom.

Notes

Conflict of Interest: None declared.

Supplementary Material

Supplementary_Material_Final_R2_bhab215
Click here for additional data file. (139.7KB, docx)

Contributor Information

Anna West, Institute for Human Neuroscience, Boys Town National Research Hospital, Omaha, NE 68010, USA.

Noah Hamlin, Institute for Human Neuroscience, Boys Town National Research Hospital, Omaha, NE 68010, USA.

Sophia Frangou, Department of Psychiatry, Djavad Mowafaghian Centre for Brain Health, University of British Columbia, Vancouver, BC V6T 1Z3, Canada; Department of Psychiatry, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA.

Tony W Wilson, Institute for Human Neuroscience, Boys Town National Research Hospital, Omaha, NE 68010, USA.

Gaelle E Doucet, Institute for Human Neuroscience, Boys Town National Research Hospital, Omaha, NE 68010, USA.

References

  1. Andrews-Hanna  JR, Reidler  JS, Sepulcre  J, Poulin  R, Buckner  RL. 2010. Functional-anatomic fractionation of the brain's default network. Neuron. 65:550–562. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Andrews-Hanna  JR, Snyder  AZ, Vincent  JL, Lustig  C, Head  D, Raichle  ME, Buckner  RL. 2007. Disruption of large-scale brain systems in advanced aging. Neuron. 56:924–935. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Ardila  A. 2007. Normal aging increases cognitive heterogeneity: analysis of dispersion in WAIS-III scores across age. Arch Clin Neuropsychol. 22:1003–1011. [DOI] [PubMed] [Google Scholar]
  4. Behzadi  Y, Restom  K, Liau  J, Liu  TT. 2007. A component based noise correction method (CompCor) for BOLD and perfusion based fMRI. Neuroimage. 37:90–101. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Betzel  RF, Byrge  L, He  Y, Goni  J, Zuo  XN, Sporns  O. 2014. Changes in structural and functional connectivity among resting-state networks across the human lifespan. Neuroimage. 102(Pt 2):345–357. [DOI] [PubMed] [Google Scholar]
  6. Buckner  RL. 2004. Memory and executive function in aging and AD: multiple factors that cause decline and reserve factors that compensate. Neuron. 44:195–208. [DOI] [PubMed] [Google Scholar]
  7. Buckner  RL, Andrews-Hanna  JR, Schacter  DL. 2008. The brain's default network: anatomy, function, and relevance to disease. Ann N Y Acad Sci. 1124:1–38. [DOI] [PubMed] [Google Scholar]
  8. Chan  D, Shafto  M, Kievit  R, Matthews  F, Spink  M, Valenzuela  M, Cam  CAN, Henson  RN. 2018. Lifestyle activities in mid-life contribute to cognitive reserve in late-life, independent of education, occupation, and late-life activities. Neurobiol Aging. 70:180–183. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Chao-Gan  Y, Yu-Feng  Z. 2010. DPARSF: a MATLAB toolbox for "pipeline" data analysis of resting-state fMRI. Front Syst Neurosci. 4:13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Cordes  D, Haughton  VM, Arfanakis  K, Carew  JD, Turski  PA, Moritz  CH, Quigley  MA, Meyerand  ME. 2001. Frequencies contributing to functional connectivity in the cerebral cortex in "resting-state" data. AJNR Am J Neuroradiol. 22:1326–1333. [PMC free article] [PubMed] [Google Scholar]
  11. Costa  PS, Santos  NC, Cunha  P, Palha  JA, Sousa  N. 2013. The use of bayesian latent class cluster models to classify patterns of cognitive performance in healthy ageing. PLoS One. 8:e71940. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Curtis  CE, D'Esposito  M. 2003. Persistent activity in the prefrontal cortex during working memory. Trends Cogn Sci. 7:415–423. [DOI] [PubMed] [Google Scholar]
  13. Daffner  KR. 2010. Promoting successful cognitive aging: a comprehensive review. J Alzheimers Dis. 19:1101–1122. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Damoiseaux  JS. 2017. Effects of aging on functional and structural brain connectivity. Neuroimage. 160:32–40. [DOI] [PubMed] [Google Scholar]
  15. Deary  IJ, Corley  J, Gow  AJ, Harris  SE, Houlihan  LM, Marioni  RE, Penke  L, Rafnsson  SB, Starr  JM. 2009. Age-associated cognitive decline. Br Med Bull. 92:135–152. [DOI] [PubMed] [Google Scholar]
  16. Depp  CA, Harmell  A, Vahia  IV. 2012. Successful cognitive aging. Curr Top Behav Neurosci. 10:35–50. [DOI] [PubMed] [Google Scholar]
  17. Desikan  RS, Segonne  F, Fischl  B, Quinn  BT, Dickerson  BC, Blacker  D, Buckner  RL, Dale  AM, Maguire  RP, Hyman  BT, et al.  2006. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage. 31:968–980. [DOI] [PubMed] [Google Scholar]
  18. Dima  D, Papachristou  E, Modabbernia  A, Doucet  GE, Agartz  I, Aghajani  M, Akudjedu  TN, Albajes-Eizagirre  A, Alnæs  D, Alpert  KI, et al.  2021. Subcortical volume across the lifespan: data from 18,605 healthy individuals aged 3-90 years. Human Bain Mapping. doi: 10.1002/hbm.25320 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Doucet  G, Naveau  M, Petit  L, Delcroix  N, Zago  L, Crivello  F, Jobard  G, Tzourio-Mazoyer  N, Mazoyer  B, Mellet  E, et al.  2011. Brain activity at rest: a multiscale hierarchical functional organization. J Neurophysiol. 105:2753–2763. [DOI] [PubMed] [Google Scholar]
  20. Doucet  GE, Glahn  DC, Frangou  S. 2020a. Person-based similarity in brain structure and functional connectivity in bipolar disorder. J Affect Disord. 276:38–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Doucet  GE, Labache  L, Thompson  PM, Joliot  M, Frangou  S, Alzheimer's Disease Neuroimaging  I. 2021. Atlas55+: brain functional atlas of resting-state networks for late adulthood. Cereb Cortex. 31:1719–1731. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Doucet  GE, Lin  D, Du  Y, Fu  Z, Glahn  DC, Calhoun  VD, Turner  J, Frangou  S. 2020b. Personalized estimates of morphometric similarity in bipolar disorder and schizophrenia. NPJ Schizophr. 6:39. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Doucet  GE, Moser  DA, Rodrigue  A, Bassett  DS, Glahn  DC, Frangou  S. 2019. Person-based brain morphometric similarity is heritable and correlates with biological features. Cereb Cortex. 29:852–862. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Feczko  E, Miranda-Dominguez  O, Marr  M, Graham  AM, Nigg  JT, Fair  DA. 2019. The heterogeneity problem: approaches to identify psychiatric subtypes. Trends Cogn Sci. 23:584–601. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Fischl  B, Dale  AM. 2000. Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proc Natl Acad Sci U S A. 97:11050–11055. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Fischl  B, Salat  DH, Busa  E, Albert  M, Dieterich  M, Haselgrove  C, van der  Kouwe  A, Killiany  R, Kennedy  D, Klaveness  S, et al.  2002. Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain. Neuron. 33:341–355. [DOI] [PubMed] [Google Scholar]
  27. Fischl  B, Salat  DH, van der  Kouwe  AJ, Makris  N, Segonne  F, Quinn  BT, Dale  AM. 2004. Sequence-independent segmentation of magnetic resonance images. Neuroimage. 23(Suppl 1):S69–S84. [DOI] [PubMed] [Google Scholar]
  28. Foss  MP, Formigheri  P, Speciali  JG. 2009. Heterogeneity of cognitive aging in Brazilian normal elderls. Dement Neuropsychol. 3:344–351. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Frangou  S, Modabbernia  A, Doucet  GE, Papachristou  E, Williams  SCR, Agartz  I, Aghajani  M, Akudjedu  TN, Albajes-Eizagirre  A, Alnæs  D, et al.  2021. Cortical thickness across the lifespan: data from 17,075 healthy individuals aged 3-90 years. Human Bain Mapping. doi: 10.1002/hbm.25364 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Friston  KJ, Williams  S, Howard  R, Frackowiak  RS, Turner  R. 1996. Movement-related effects in fMRI time-series. Magn Reson Med. 35:346–355. [DOI] [PubMed] [Google Scholar]
  31. Gefen  T, Shaw  E, Whitney  K, Martersteck  A, Stratton  J, Rademaker  A, Weintraub  S, Mesulam  MM, Rogalski  E. 2014. Longitudinal neuropsychological performance of cognitive SuperAgers. J Am Geriatr Soc. 62:1598–1600. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Habeck  C, Razlighi  Q, Gazes  Y, Barulli  D, Steffener  J, Stern  Y. 2017. Cognitive reserve and brain maintenance: orthogonal concepts in theory and practice. Cereb Cortex. 27:3962–3969. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Harrison  TM, Weintraub  S, Mesulam  MM, Rogalski  E. 2012. Superior memory and higher cortical volumes in unusually successful cognitive aging. J Int Neuropsychol Soc. 18:1081–1085. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Hedden  T, Gabrieli  JD. 2004. Insights into the ageing mind: a view from cognitive neuroscience. Nat Rev Neurosci. 5:87–96. [DOI] [PubMed] [Google Scholar]
  35. Hipkiss  AR. 2006. Accumulation of altered proteins and ageing: causes and effects. Exp Gerontol. 41:464–473. [DOI] [PubMed] [Google Scholar]
  36. Huo  L, Li  R, Wang  P, Zheng  Z, Li  J. 2018. The default mode network supports episodic memory in cognitively unimpaired elderly individuals: different contributions to immediate recall and delayed recall. Front Aging Neurosci. 10:6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Jack  CR Jr., Bernstein  MA, Fox  NC, Thompson  P, Alexander  G, Harvey  D, Borowski  B, Britson  PJ, Jennifer L  Whitwell, Ward  C, et al.  2008. The Alzheimer's disease neuroimaging initiative (ADNI): MRI methods. J Magn Reson Imaging  27:685–691. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Janssen  J, Díaz-Caneja  CM, Alloza  C, Schippers  A, de  Hoyos  L, Santonja  J, Gordaliza  PM, Buimer  EEL, van  Haren  NEM, Cahn  W, et al.  2020. Dissimilarity in sulcal width patterns in the cortex can be used to identify patients with schizophrenia with extreme deficits in cognitive performance. Schizophr Bull. 47(2):552–561. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Joyce  EM, Roiser  JP. 2007. Cognitive heterogeneity in schizophrenia. Curr Opin Psychiatry. 20:268–272. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Leuner  K, Hauptmann  S, Abdel-Kader  R, Scherping  I, Keil  U, Strosznajder  JB, Eckert  A, Muller  WE. 2007. Mitochondrial dysfunction: the first domino in brain aging and Alzheimer's disease?  Antioxid Redox Signal. 9:1659–1675. [DOI] [PubMed] [Google Scholar]
  41. Li  KZ, Lindenberger  U. 2002. Relations between aging sensory/sensorimotor and cognitive functions. Neurosci Biobehav Rev. 26:777–783. [DOI] [PubMed] [Google Scholar]
  42. Li  Q, Dong  C, Liu  T, Chen  X, Perry  A, Jiang  J, Cheng  J, Niu  H, Kochan  NA, Brodaty  H, et al.  2020. Longitudinal changes in whole-brain functional connectivity strength patterns and the relationship with the global cognitive decline in older adults. Front Aging Neurosci. 12:71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Menon  V, Uddin  LQ. 2010. Saliency, switching, attention and control: a network model of insula function. Brain Struct Funct. 214:655–667. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Meunier  D, Achard  S, Morcom  A, Bullmore  E. 2009. Age-related changes in modular organization of human brain functional networks. Neuroimage. 44:715–723. [DOI] [PubMed] [Google Scholar]
  45. Morse  CK. 1993. Does variability increase with age? An archival study of cognitive measures. Psychol Aging. 8:156–164. [DOI] [PubMed] [Google Scholar]
  46. Nilsson  J, Lovden  M. 2018. Naming is not explaining: future directions for the "cognitive reserve" and "brain maintenance" theories. Alzheimers Res Ther. 10:34. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Nyberg  L, Lovden  M, Riklund  K, Lindenberger  U, Backman  L. 2012. Memory aging and brain maintenance. Trends Cogn Sci. 16:292–305. [DOI] [PubMed] [Google Scholar]
  48. Patel  AX, Kundu  P, Rubinov  M, Jones  PS, Vertes  PE, Ersche  KD, Suckling  J, Bullmore  ET. 2014. A wavelet method for modeling and despiking motion artifacts from resting-state fMRI time series. Neuroimage. 95:287–304. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Petersen  RC, Aisen  PS, Beckett  LA, Donohue  MC, Gamst  AC, Harvey  DJ, Jack  CR  Jr, Jagust  WJ, Shaw  LM, Toga  AW, et al.  2010. Alzheimer's disease neuroimaging initiative (ADNI): clinical characterization. Neurology. 74:201–209. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Pistono  A, Guerrier  L, Peran  P, Rafiq  M, Gimeno  M, Bezy  C, Pariente  J, Jucla  M. 2020. Increased functional connectivity supports language performance in healthy aging despite gray matter loss. Neurobiol Aging. 98:52–62. [DOI] [PubMed] [Google Scholar]
  51. Rogalski  EJ, Gefen  T, Shi  J, Samimi  M, Bigio  E, Weintraub  S, Geula  C, Mesulam  MM. 2013. Youthful memory capacity in old brains: anatomic and genetic clues from the Northwestern SuperAging project. J Cogn Neurosci. 25:29–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Samson  RD, Lester  AW, Duarte  L, Venkatesh  A, Barnes  CA. 2017. Emergence of beta-band oscillations in the aged rat amygdala during discrimination learning and decision making tasks. eNeuro. 4(5):ENEURO.0245–17.2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Samu  D, Campbell  KL, Tsvetanov  KA, Shafto  MA, CANc  C, Tyler  LK. 2017. Preserved cognitive functions with age are determined by domain-dependent shifts in network responsivity. Nat Commun. 8:14743. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Segonne  F, Dale  AM, Busa  E, Glessner  M, Salat  D, Hahn  HK, Fischl  B. 2004. A hybrid approach to the skull stripping problem in MRI. Neuroimage. 22:1060–1075. [DOI] [PubMed] [Google Scholar]
  55. Segonne  F, Pacheco  J, Fischl  B. 2007. Geometrically accurate topology-correction of cortical surfaces using nonseparating loops. IEEE Trans Med Imaging. 26:518–529. [DOI] [PubMed] [Google Scholar]
  56. Shafto  MA, Henson  RN, Matthews  FE, Taylor  JR, Emery  T, Erzinclioglu  S, Hanley  C, Rowe  JB, Cusack  R, Calder  AJ, et al.  2019. Cognitive diversity in a healthy aging cohort: cross-domain cognition in the Cam-CAN project. J Aging Health. 32(9):1029–1041. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Shafto  MA, Tyler  LK, Dixon  M, Taylor  JR, Rowe  JB, Cusack  R, Calder  AJ, Marslen-Wilson  WD, Duncan  J, Dalgleish  T, et al.  2014. The Cambridge Centre for Ageing and Neuroscience (Cam-CAN) study protocol: a cross-sectional, lifespan, multidisciplinary examination of healthy cognitive ageing. BMC Neurol. 14:204. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Sled  JG, Zijdenbos  AP, Evans  AC. 1998. A nonparametric method for automatic correction of intensity nonuniformity in MRI data. IEEE Trans Med Imaging. 17:87–97. [DOI] [PubMed] [Google Scholar]
  59. Smith  SM, Fox  PT, Miller  KL, Glahn  DC, Fox  PM, Mackay  CE, Filippini  N, Watkins  KE, Toro  R, Laird  AR, et al.  2009. Correspondence of the brain's functional architecture during activation and rest. Proc Natl Acad Sci U S A. 106:13040–13045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Soldan  A, Pettigrew  C, Albert  M. 2018. Evaluating cognitive reserve through the prism of preclinical Alzheimer dsease. Psychiatr Clin North Am. 41:65–77. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Stern  Y. 2002. What is cognitive reserve? Theory and research application of the reserve concept. J Int Neuropsychol Soc. 8:448–460. [PubMed] [Google Scholar]
  62. Stern  Y, Gurland  B, Tatemichi  TK, Tang  MX, Wilder  D, Mayeux  R. 1994. Influence of education and occupation on the incidence of Alzheimer's disease. JAMA. 271:1004–1010. [PubMed] [Google Scholar]
  63. Sun  FW, Stepanovic  MR, Andreano  J, Barrett  LF, Touroutoglou  A, Dickerson  BC. 2016. Youthful brains in older adults: preserved neuroanatomy in the default mode and salience networks contributes to youthful memory in superaging. J Neurosci. 36:9659–9668. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Taylor  JR, Williams  N, Cusack  R, Auer  T, Shafto  MA, Dixon  M, Tyler  LK, Cam  C, Henson  RN. 2017. The Cambridge Centre for Ageing and Neuroscience (Cam-CAN) data repository: structural and functional MRI, MEG, and cognitive data from a cross-sectional adult lifespan sample. Neuroimage. 144:262–269. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Terman  A, Gustafsson  B, Brunk  UT. 2006. Mitochondrial damage and intralysosomal degradation in cellular aging. Mol Aspects Med. 27:471–482. [DOI] [PubMed] [Google Scholar]
  66. Tosato  M, Zamboni  V, Ferrini  A, Cesari  M. 2007. The aging process and potential interventions to extend life expectancy. Clin Interv Aging. 2:401–412. [PMC free article] [PubMed] [Google Scholar]
  67. Vaskinn  A, Haatveit  B, Melle  I, Andreassen  OA, Ueland  T, Sundet  K. 2020. Cognitive heterogeneity across schizophrenia and bipolar disorder: a cluster analysis of intellectual trajectories. J Int Neuropsychol Soc. 26:860–872. [DOI] [PubMed] [Google Scholar]
  68. Wechsler  D. 1997. WAIS-III: Administration and scoring manual. San Antonio: The Psychological Corporation. [Google Scholar]
  69. Wen  X, He  H, Dong  L, Chen  J, Yang  J, Guo  H, Luo  C, Yao  D. 2020. Alterations of local functional connectivity in lifespan: a resting-state fMRI study. Brain Behav. 10:e01652. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Yan  CG, Wang  XD, Zuo  XN, Zang  YF. 2016. DPABI: Data Processing & Analysis for (resting-state) brain imaging. Neuroinformatics. 14:339–351. [DOI] [PubMed] [Google Scholar]
  71. Ylikoski  R, Ylikoski  A, Keskivaara  P, Tilvis  R, Sulkava  R, Erkinjuntti  T. 1999. Heterogeneity of cognitive profiles in aging: successful aging, normal aging, and individuals at risk for cognitive decline. Eur J Neurol. 6:645–652. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary_Material_Final_R2_bhab215
Click here for additional data file. (139.7KB, docx)

Articles from Cerebral Cortex (New York, NY) are provided here courtesy of Oxford University Press

RESOURCES