Successful cognitive aging is associated with thicker anterior cingulate cortex and lower tau deposition compared to typical aging

Abstract INTRODUCTION There is no consensus on either the definition of successful cognitive aging (SA) or the underlying neural mechanisms. METHODS We examined the agreement between new and existing definitions using: (1) a novel measure, the cognitive age gap (SA‐CAG, cognitive‐predicted age minus chronological age), (2) composite scores for episodic memory (SA‐EM), (3) non‐memory cognition (SA‐NM), and (4) the California Verbal Learning Test (SA‐CVLT). RESULTS Fair to moderate strength of agreement was found between the four definitions. Most SA groups showed greater cortical thickness compared to typical aging (TA), especially in the anterior cingulate and midcingulate cortices and medial temporal lobes. Greater hippocampal volume was found in all SA groups except SA‐NM. Lower entorhinal 18F‐Flortaucipir (FTP) uptake was found in all SA groups. DISCUSSION These findings suggest that a feature of SA, regardless of its exact definition, is resistance to tau pathology and preserved cortical integrity, especially in the anterior cingulate and midcingulate cortices. Highlights Different approaches have been used to define successful cognitive aging (SA). Regardless of definition, different SA groups have similar brain features. SA individuals have greater anterior cingulate thickness and hippocampal volume. Lower entorhinal tau deposition, but not amyloid beta is related to SA. A combination of cortical integrity and resistance to tau may be features of SA.


BACKGROUND
Successful cognitive aging (SA) can be conceptualized as one extreme of a continuum with patients with dementia at the other extreme and typical aging (TA) falling somewhere in the middle.Although most of the research in cognitive aging focuses on pathological and age-related cognitive decline, studying SA is also crucial to uncover protective mechanisms.
2][3] Studies have described SA using different terms such as SuperAgers, [4][5][6][7] optimal memory performers, 8 supernormals, [9][10][11][12][13] superior memory performers, 14 or high-performing older adults, 15,16 and have used different criteria, including a minimum age and cognitive measures.8][19] Another approach defines SA as older adults with higher cognitive performance (e.g., composite memory scores above the 80th or 90th percentiles) compared to individuals within the same age range. 8,9Despite the broad methodological and conceptual differences in previous work examining SA, a comparison between SA definitions and related brain features has never been formally conducted.
Existing evidence on the neural substrates of SA is limited but suggests that SA is characterized by greater hippocampal volume and greater thickness in several regions of frontal cortex. 4,8,14,22Less clear is the relationship with Alzheimer's disease (AD)-related pathology.
Previous findings indicate that SA may be resilient to the negative effects of amyloid beta (Aβ), although the role of Aβ may be more detectable by longitudinal cognitive measures. 8,14Neuropathological evidence suggests lower tau neurofibrillary tangles (NFTs) in SuperAgers compared with TA, but no differences in amyloid plaque density. 6,25Despite these findings, evidence about the presence of AD pathology in SA is very limited.
A growing number of studies have used machine learning models to estimate brain age from structural neuroimaging data. 26The same method can be used to obtain age estimates based on individuals' cognitive functioning, whereby a normative model is built using cognitive data to predict chronological age.Measuring cognitive-predicted age is a novel approach that has been introduced only recently. 27is method has the potential to capture patterns of holistic cognitive aging that are not easily detectable by standard neuropsychological measures considered separately.SA can be defined as individuals deviating from normative cognitive trajectories independent of age, rather than being indirectly inferred using thresholds on single or composite measures of cognitive performance.

RESEARCH IN CONTEXT
1. Systematic review: A systematic literature search was carried out using PubMed to identify articles investigating the brain features related to successful cognitive aging (SA).Despite the different approaches that have been used to define SA, a comparison between SA definitions and related brain features has never been formally conducted.

Interpretation:
The results of this work indicate that different SA definitions identify only partially overlapping groups of older adults.Despite this, common brain features were found across definitions.These findings suggest that a feature of SA is resistance to tau pathology and preserved cortical integrity, especially in the anterior cingulate and midcingulate cortices.
3. Future directions: Longitudinal observations are needed to fully comprehend the many phenomena associated with SA.Moreover, it will be important in the future to investigate the relationship between both genetics and modifiable lifestyle factors related to SA and brain pathology.
The present study aimed at using novel and existing approaches to define SA to examine underlying concepts of resistance to the emergence of AD pathology (i.e., SA showing lower levels of pathology than would be predicted by group-level data) and resilience to its negative effects on cognition (i.e., SA showing comparable levels of pathology). 28 ensure that our findings were not driven by specific SA criteria, we considered different SA definitions.reviewed and approved the study.All participants provided written, informed consent for their participation in this study.

Neuropsychological assessment
The BACS protocol comprises a comprehensive neuropsychological battery assessing a variety of cognitive domains, including verbal and visual memory, working memory, processing speed, executive functioning, and attention.In this study the following tests were used:

SA definitions
The magnitude of cognitive decline is heavily influenced by advancing age and, therefore, age is an important criteria to include in the definition of SA. 37 For this reason, in the present study, only participants 70 years of age or older from the BACS test cohort (n = 184) were included to explore differences between SA and TA, in line with previous studies. 9,14,17,21Cross-sectional cognitive data closest to each participant's PiB scan were used to identify cutoffs to classify participants as either SA or TA.For 93% of participants, the time between the PiB scan and the cognitive session was ≤6 months; the time interval was less than a year for all subjects.Tau PET-related analyses involved a subsample of participants (n = 114).The cognitive session closest to the FTP scan was used for tau PET-related analyses.
For 91% of participants, the time between the FTP scan and the cognitive assessment was ≤6 months; for only two participants, the time interval was slightly over a year (1.03 and 1.07 years).For this reason, all analyses were repeated including the time interval between scan date and cognitive session date as covariate of no interest in addition to age, sex, and years of education.
Individuals were identified as SA using four different definitions.
The first definition was based on CAG scores.Consistent with the MRI brain age literature, negative CAG scores reflect younger cognitive age than chronological age.Therefore, participants with CAG scores within the lowest 20th percentile were defined as SA-CAG.The cutoff was placed at the 20th percentile to be consistent with previous approaches defining SA as participants with cognitive performance in the top 20%. 8The second definition was similar to SA definitions used previously based on the performance on memory composite scores in comparison with average performance of individuals with the same age.the CVLT LDFR, as previous definitions. 14,17,22In particular, we defined SA as participants with a performance comparable to individuals 18 to 32 years of age (score of 14 or above; max score = 16). 14,22All individuals in the cohort not meeting any of the four SA criteria were defined as typical agers.In all analyses comparing SA and TA, therefore, each SA group was compared with the same group of typical agers (n = 110).

MRI acquisition and processing
Each and PiB scans were acquired within 6 months for all subjects.The time between the MRI scan and the cognitive session used to define SA/TA groups was ≤6 months for 91% of participants and was less than a year for all participants.[40] Briefly, FreeSurfer reconstructs three-dimensional (3D) pial and white matter surfaces based on the relative intensity differences at the boundaries of each tissue class.Cortical thickness is calculated across ≈150,000 vertices per hemisphere as the average distance of the vectors perpendicular to the triangular faces of the white matter and pial surfaces. 38The FreeSurfer volumetric segmentation was used to calculate hippocampal volumes and total intracranial volume (TIV). 41

PET acquisition and processing
3][44] PiB and FTP were synthesized at the LBNL Biomedical Isotope Facility.
For PiB-PET images, 90 minutes of dynamic emission data frames was acquired after an injection of 15 mCi of PiB tracer.A computerized tomography (CT) scan was obtained pre-injection and used for attenuation correction.PiB-PET images were reconstructed using an ordered subset expectation maximization algorithm with weighted attenuation and smoothed with a 4-mm Gaussian kernel with scatter correction.
Distribution volume ratio (DVR) was generated with Logan graphical analysis on PiB frames over 35 to 90 minutes post-injection and normalized using a cerebellar gray matter reference region. 45,46obal cortical PiB DVR was calculated using FreeSurfer-derived cortical regions of interest (ROIs), 39,47 and Aβ positivity determined using a global PiB DVR threshold of 1.065. 42Centiloid (CL) values were calculated using a conversion equation employed previously in our laboratory 48 and developed for our processing pipeline: CL = (DVR × 142.73) -141.99.
For FTP-PET images, participants were injected with 10 mCi of tracer and scanned from 80 to 100 minutes post-injection, usually on the same day as the PiB-PET scan.CT scans were used for attenuation correction.FTP-PET images were reconstructed using an ordered subset expectation maximization algorithm with scatter correction and smoothed with a 4-mm Gaussian kernel.
To create FTP standardized uptake value ratio (SUVR) images, the mean tracer uptake 80 to 100 minutes post-injection was normalized to the inferior cerebellar gray matter reference region. 49ometric transfer matrix partial volume correction (PVC) on the Desikan-Killiany FreeSurfer-derived ROIs was used for FTP data processing to account for partial volume effects. 50,51The Desikan-Killiany atlas was used to define ROIs of the entorhinal cortex (EC) and inferior temporal (IT) cortex, which were used to explore differences between SA and TA in FTP uptake.These regions were chosen because tau accumulation has been shown to start focally in the EC, and the IT was chosen as an early-stage tau deposition region outside the medial temporal lobe (MTL).
A GLM was used to explore differences between SA and TA in TIVadjusted hippocampal volume, global PiB DVR, and FTP SUVR in the EC and IT ROIs separately for each SA group.Specifically, we performed analyses of covariance (ANCOVA) controlling for age, sex, and years of education.We repeated the analyses including days between cognitive session and MRI/PET scan as an additional covariate of no interest.Partial eta squared (partial η 2 ) was used as a measure of the effect size for between-group differences (small = 0.01, medium = 0.06, large = 0.14).
Significance level was set at p < 0.05.

Cognitive age model accurately predicts age from neuropsychological tests
To estimate cognitive-predicted age we used PLSr, which performs latent space modeling, deriving latent features that have maximum covariance with the response variable.We trained the CA model using an independent data set (BACS training cohort) than the data set used to associate with imaging variables (BACS test cohort).These training and testing data sets were generally well matched (Table 1); however, there was a significantly smaller proportion of female participants and greater rates and duration of follow-up for the testing set.The percentage of variance explained by each component in the trained CA model is described in the Supplementary results, which suggest that only components 1 and 2 are well associated with age (explaining 33% and 13% of variance, respectively).Figure S2 shows the contribution of each predictor to the first two components.
Applying the parameters learned from the trained model to predict age in the independent test cohort (n = 1141 sessions), the model accurately predicted age with a MAE of 4.36, explaining 41% of the variance in chronological age (Figure 1).This value was similar to the variance explained in the training sample (R 2 = 0.49).4][35] Overall, these results validated our modeling approach to predict age using cognitive data.

Four definitions of SA (70+ years old)
Using

SA and TA cohort characteristics
Cohort characteristics for SA and TA are summarized in Table 2.
Successful cognitive aging defined using CAG, EM, NM, or CVLT were also combined in a single group of individuals meeting any SA criterion any SA group and TA were found in age, BMI, history of hypertension, or MMSE and GDS scores.Finally, we were interested in the effect of the APOE ε4 allele, the major genetic risk factor for sporadic AD, and the APOE ε2 allele for its protective effect. 58,59Participants were grouped as APOE ε2 carriers (ε2/ε3, n = 16), APOE ε3 homozygotes (ε3/ε3, n = 116), or APOE ε4 carriers (ε3/ε4, n = 42; ε4/ε4, n = 1).No APOE ε2/ε2 homozygotes were identified, whereas APOE ε2/ε4 heterozygotes (n = 3) were excluded from these analyses.No differences in APOE genotype were found across SA definitions compared with TA.
Neuropsychological tests are summarized in Table S1.

Greater medial prefrontal and temporal thickness, and greater hippocampal volume in successful cognitive aging
Vertex-wise cortical thickness analyses in FreeSurfer revealed regions of greater cortical thickness in SA groups compared to TA (p < 0.001 uncorrected, Figure 3).All results were adjusted for age, sex, and years of education.For the SA-ALL group, the following regions were thicker compared with TA: left anterior midcingulate cortex (aMCC), also known as dorsal anterior cingulate cortex, left posterior midcingulate cortex (pMCC), left middle and inferior temporal gyri, right rostral anterior cingulate cortex (rACC) extending to the medial orbitofrontal cortex (mOFC), and MTL, mostly in the bilateral parahippocampal gyrus and left EC.All other SA groups except SA-CVLT showed thicker left MCC, right rACC/mOFC, and ACC.SA-EM, SA-NM, and SA-CVLT had thicker regions in the MTL.A cluster in the lateral temporal lobe was thicker in SA-CAG and SA-NM compared with TA.It is important to note that there were no regions in the brain where cortex was thicker in TA compared with SA, regardless of definition.When education was removed from the vertex-wise analyses, the findings were consistent.
A series of GLMs were run in FreeSurfer to test the continuous relationship between cognitive measures and cortical thickness in the whole sample (including all SA groups and TA).Sex and years of education were included as covariates of no interest; in addition, age was included in CVLT-related analyses.GLMs revealed a negative association between CAG scores and the left aMCC/pMCC (Figure 4A     Findings were replicated when years of education was removed from the models.

Lower entorhinal tau burden in successful aging, but no differences in global Aβ
We examined the relationship between successful aging and AD biomarkers, including PET measured PiB and FTP to measure Aβ and tau burden, respectively.3][54][55] We also investigated FTP uptake in ACC/MCC ROIs, since previous neuropathological evidence has shown lower NFTs in the rACC and aMCC regions in SuperAgers compared with age-matched controls. 25fferences between SA groups and TA in global PiB DVR and PVC FTP uptake in each ROI were explored using ANCOVA models includ- Sex and years of education were included as covariates of no interest in all analyses, and age was additionally included in CVLT-related analyses.
η 2 = 0.05) (Figure 6).The results remained similar when we included days between cognitive session and PET scan as a covariate of no interest in addition to age, sex, and years of education (see Supplementary results).Moreover, findings were replicated when years of education was removed from the models.Next, we repeated the model includ- Previous studies have defined SA predominantly as older adults with exceptional memory performance, mostly due to the vulnerability of memory abilities to both aging and AD. 7 There may be conceptual and neurobiological distinctions associated with the use of domain-specific SA definitions.Hence, in the present study, we decided to include both memory and non-memory cognition definitions to investigate potential divergent neurobiological substrates.Moreover, substantial data indicate that women have better performance on verbal memory tests, 60 which is also reflected in our findings.For example, SA-CVLT had a significantly higher percentage of women, but no sex differences were found between SA-CAG and TA.Our novel definition based on a cognitive age model aimed at capturing patterns of holistic cognitive aging that may not be easily detectable by single or composite measures of cognitive performance.SA-CAG showed the highest overlap with the other definitions, suggesting that it may be a more comprehensive measure capturing multiple aspects of SA.Moreover, SA-CAG was the only SA group that did not differ from TA in years of education, suggesting that this definition may be less dependent on educational attainment, while it appeared to be a feature of the other definitions.For this reason, we repeated the analyses both with and without education in our models, and the results remained consistent, indicating that education, overall, had no significant impact on our outcome measures.
To address our second aim, we explored differences between TA and each SA group in brain features.We found regions of greater cortical thickness in the aMCC/pMCC, rACC/mOFC in SA using most definitions, compared with TA, confirming previous results in successful memory aging. 4,14,22The observation that the SA-CVLT group showed thicker cortex limited to the MTL may reflect the episodic memory-predominant definition of this group.The consistency of thicker aMCC/pMCC cortex across SA definitions was striking.Furthermore, when we explored the continuous relationship between cognitive scores and cortical thickness, we found a negative association between continuous CAG and the aMCC/pMCC thickness (i.e., greater thickness for younger predicted cognitive age).This is consistent with our interpretation that the CAG captures cognitive performance well across multiple definitions of SA and may be more sensitive to superior cognitive performance rather than pathological cognitive impairment.The strong relationship between continuous CAG and thicker aMCC/pMCC suggests the CAG is a promising continuous measure of normal cognitive aging that can be used as an alternative approach to the dichotomization into two separate groups (i.e., TA vs SA).
Thickness in the ACC/MCC and its relationship to SA is particularly interesting due to its unique neurobiology.Evidence from studies of SuperAging demonstrate greater cortical thickness of this brain region in SA, even with different specific comparator groups. 4,25This region is also one of several, including the orbitofrontal cortex and frontal regions of the insula, that contain relatively high densities of Von Economo neurons (VENs).These unusual spindle-shaped neurons have unclear functional significance; are seen in humans, great apes, and cetaceans; and are selectively vulnerable to neurodegeneration in frontotemporal dementia in humans. 61,62Brain regions with a high density of VENs have been implicated in a variety of neuropsychiatric disorders involving emotional-social functions, and also comprise the main hubs of the salience network. 63,64Postmortem studies of SuperAgers have shown that these individuals have substantially higher numbers of VENs in the aMCC compared to age-matched cognitively normal individuals as well as younger controls. 65Lower levels of tau pathology in rACC and aMCC have been shown previously in SuperAgers compared to both age-matched controls and cases of mild cognitive impairment (MCI). 25In our study, however, we found no differences in FTP uptake in ACC/MCC regions, which could be due to lower levels of tau burden in both SA and TA groups in our sample in these regions.
Taken together, the structural MRI findings in this cohort, along with other histological studies, suggest that the MCC is an important region for optimal brain aging outcomes.Greater cortical thickness in this brain region is consistent across studies of SuperAgers 4,25 and other definitions of SA, 14,22 and metabolic preservation in this region in cognitively normal older people also behaves as a signature of brain resilience. 668][69] It is not clear whether these brain regions provide support to successful aging outcomes through a dynamic or adaptive process such as hypertrophy or even neurogenesis, or whether they reflect lifelong advantages related to genetic or early life environmental effects.Regardless, together these findings provide strong evidence for the importance of the ACC and MCC in maintaining superior cognition at an older age, regardless of SA definition and methodological differences across studies.
We also found evidence that preserved integrity of the MTL was a feature of SA.Greater hippocampal volume has been related previously to SA. 8,14,22 This may be interpreted as greater brain reserve, so that interindividual neurobiological differences may allow SA to overcome the effects of brain aging. 70However, further evidence from longitudinal studies is needed to clarify the role of individual brain morphological changes over time in SA.The conjoint findings of volume preservation in MTL/hippocampus and reduced deposition of pathological tau raise the possibility that these findings are related.2][73] There is also strong evidence of relationships between tau accumulation and regional MTL atrophy in cognitively normal older people. 74,75Based on this evidence, it is possible that the finding of larger hippocampal volumes and thicker MTL cortex in the SA sample could reflect the reduced tau pathology, lifelong or early life-reserve factors, a dynamic response to pathology, or a combination of these factors.
Lower tau in our SA participants was seen despite comparable levels of Aβ.The mean DVR value measured with PiB-PET in the TA group was 21 CL, whereas the mean values in the SA groups ranged from 15 to 19 CL.It is difficult to know whether these small differences are related to differences in EC tau burden.SA individuals may be resistant to EC tau deposition, consistent with previous neuropathological reports on SuperAging, 6 and this may underlie, at least in part, their exceptional cognitive performance.Preliminary results from a small group of individuals from the Alzheimer's Disease Neuroimaging Initiative (ADNI) showed lower tau deposition in temporal and medial parietal lobe in SA compared with TA. 76 It is important to note that our most robust findings on tau differences reflect tau deposition in the EC and not IT.
These brain regions are different because EC tau is common in people without brain Aβ, whereas tau spread to IT usually reflects higher levels of Aβ. 44,[77][78][79][80][81] These findings, along with the relatively small differences in brain Aβ, are consistent with a resistance of age-related tau deposition in our SA participants rather than a resistance to AD pathology.
Whether Aβ pathology plays a role in driving tau deposition in our SA group will require longitudinal observation.
Various limitations should be considered when interpreting the results of the present study.First, the BACS cohort is a fairly homogeneous and highly educated sample that is not fully representative of the diversity of typical cognitive aging.Second, there were some differences between the training and test cohorts for calculation of CAG, (i.e., presence of neuroimaging scans, number of follow-up sessions, proportion of women), although all individuals underwent the same comprehensive screening process and received identical cognitive evaluations.Despite differences between the cohorts, when we applied the parameters learned from the trained model to predict age in the test sample, the model explained 41% of the variance in chronological age, which was very similar to the 49% variance explained in the training sample.This confirms the validity of our modeling approach by demonstrating that the model is robust to differences in training and test samples.Another limitation is that our methods reflect numerous choices of thresholds that are likely to affect our results.This includes the age threshold, the threshold for composite score definitions of SA or CVLT performance, and thresholds for statistical significance in a variety of analyses.Finally, this is a cross-sectional investigation, and it is increasingly apparent that cognitive aging requires longitudinal observation to fully comprehend the many associated phenomena. 17r findings support the hypothesis that a combination of structural integrity and resistance to tau pathology may underlie SA regardless of its exact definition and promote effective cognitive functioning at an older age.It will be important in the future to investigate the relationship between both genetics and modifiable lifestyle factors related to SA and brain pathology.A better understanding of the neural features related to SA may lead to the identification of targets for new interventions aiming at promoting healthy aging.This becomes even more relevant in the light of increasing evidence suggesting the importance of interventions promoting brain health in helping mitigate cognitive decline. 82 participant within the test cohort underwent a high-resolution T1-weighted magnetization-prepared rapid gradient echo (MPRAGE) scan acquired on a 1.5 T Siemens Magnetom Avanto scanner at LBNL.The following acquisition parameters were applied: repetition time (TR) = 2110 ms, echo time (TE) = 3.58 ms, flip angle = 15 • , 1 mm slice thickness, and 1 × 1 mm 2 in-plane resolution.For each participant, we selected the MRI scan closest to the PiB scan.MRI a subset of BACS individuals aged 70 and older from the test cohort (n = 184), we explored the number of participants identified as exceptional cognitive performers, herein called SA, by different new and previously accepted definitions.Participants were identified as SA based on (1) ≤20th percentile as CAG (SA-CAG, n = 37); (2) ≥80th percentile of age-adjusted EM composite (SA-EM, n = 37); (3) ≥80th percentile of age-adjusted NM cognition composite cutoff score (SA-NM, n = 37); and (4) performance comparable to young adults on the CVLT LDFR (score of 14 or above; max score = 16) (SA-CVLT, n = 31).All individuals in the cohort not meeting any of the four SA criteria were defined as TA (n = 110).Cohen's kappa statistics were used to explore the strength of agreement between definitions.The number of overlapping participants defined as SA by different definitions and Cohen's κ coefficients, displayed in Figure2, showed only a fair to moderate strength of agreement between definitions, with only 6 of 74 SA participants defined as SA by all definitions.The strength of agreement given

Note:F I G U R E 2
Values represent either mean (SD) or n (%).Differences between groups were investigated using Welch two-sample t-test for continuous variables and chi-square test for categorical variables.Abbreviations: Y, yes. a Missing data for two participants in the training cohort.b Missing data for 44 participants in the training cohort.c Including Hispanic or Latino and White (n = 4), Hispanic or Latino and Asian (n = 1), Hispanic or Latino and Black (n = 1).d Including Asian and White (n = 2).e Including Black and White (n = 2).f Including Hispanic or Latino and Asian (n = 1), Hispanic or Latino and White (n = 6), Hispanic or Latino and Native Hawaiian or other Pacific Islander (n = 2).F I G U R E 1 Cognitive-predicted age from the partial least squared (PLS) regression model.Scatterplots showing chronological age by cognitive-predicted age before age-bias correction in the training (left) and test (right) cohorts.The dashed lines are lines of identity (x = y) where cognitive-predicted age = chronological age.R 2 values refers to the total variance explained, and r-values are the Pearson correlation coefficients of cognitive-predicted age with chronological age.by the Cohen's κ coefficients was interpreted as poor (<0), slight (0-0.20),fair (0.21-0.40), moderate (0.41-0.60), substantial (0.61-0.80), and almost perfect (0.81-1). 57The CAG-based definition presented the best overlap with other definitions, with the lowest number of participants defined as SA-CAG only (n = 4), as opposed to other definitions (only SA-EM = 10; only SA-NM = 7; only SA-CVLT = 11).When we repeated the analyses with the same number of participants within each SA group to match the SA-CVLT number of participants (n = 31), the results were similar (Cohen's κ coefficients: SA-CAG and SA-EM, κ = 0.42; SA-CAG and SA-NM, κ = 0.53; SA-CAG and SA-CVLT, κ = 0.22; SA-EM and SA-NM, κ = 0.30; SA-EM and SA-CVLT, κ = 0.26; and SA-NM and SA-CVLT, κ = 0.30).Number of overlapping participants included in the successful cognitive aging (SA) groups using different definitions (left), and Cohen's κ values indicating the strength of agreement between definitions (right).The strength of agreement given by the Cohen's κ coefficients was interpreted as poor (<0), slight (0-0.20),fair (0.21-0.40), moderate (0.41-0.60), substantial (0.61-0.80), and almost perfect (0.81-1).57

4 DISCUSSION
ing global PIB DVR as a covariate of no interest: EC FTP uptake ∼ SA Group + Age + Sex + years of education + PiB DVR.There was a significant effect of SA group on EC FTP uptake when controlling for PiB DVR for SA-ALL (F(1, 107) = 6.71, p = 0.01, partial η 2 = 0.06), SA-EM (F(1, 91) = 6.14, p = 0.02, partial η 2 = 0.06), and SA-NM (F(1, 80) = 6.43, p = 0.01, partial η 2 = 0.07), but not SA-CAG (F(1, 82) = 2.66, p = 0.11, partial η 2 = 0.03) and SA-CVLT (F(1, 87) = 3.47, p = 0.07, partial η 2 = 0.04).Differences between SA and TA in IT FTP uptake were significant only in SA-ALL (F(1, 109) = 4.20, p = 0.04, partial η 2 = 0.04), with no significant effects when SA was defined as SA-CAG (F(1, 84) = 3.49, p = 0.07, partial η 2 = 0.04), SA-EM (F(1, 93) = 3.31, p = 0.07, partial η 2 = 0.03), SA-NM (F(1, 82) = 2.45, p = 0.12, partial η 2 = 0.03), and SA-CVLT (F(1, 89) = 1.93, p = 0.17, partial η 2 = 0.02).The results remained unaltered when we included days between cognitive session and PET scan as covariate of no interest in addition to age, sex, and years of education (see Supplementary results).When years of education was removed from the model, we found a significant effect of SA group on IT FTP when all SA were grouped together (SA-ALL: F(1, 110) = 4.98, p = 0.03, partial η 2 = 0.04), and SA-EM (F(1, 94) = 4.29, p = 0.04, partial η 2 = 0.04), but no effect when SA was defined as SA-CAG, SA-NM, and SA-CVLT.Finally, no significant differences were found between SA groups and TA in rACC, aMCC, and pMCC FTP uptake.Despite imperfect overlap between the four SA groups, our findings suggest common brain features across definitions.Except for SA-CVLT, all SA groups presented greater cortical thickness in the aMCC/pMCC, rACC/mOFC compared to TA. SA also had regions of thicker cortex in the MTL (SA-EM, SA-NM, and SA-CVLT) and lateral temporal regions (SA-CAG and SA-NM).In addition, SA-CAG, SA-EM, and SA-NM had greater hippocampal volume and all SA groups had lower EC tau burden compared to TA.Overall, these findings suggest that a feature of SA, regardless of its exact definition, may be resistance to tau pathology, whereas greater thickness in aMCC, rACC/mOFC, and MTL may be interpreted as greater brain reserve.The first goal of this study was to develop a new measure to define SA using a CA prediction model.Our findings suggest that we can reliably predict age from neuropsychological tests and use these age estimates to calculate biologically meaningful CAG scores.We also defined SA based on EM and NM cognition composite scores as well as performance on the CVLT LDFR.Our findings revealed that different definitions of SA identified only partially overlapping groups of older adults, highlighting the heterogeneity of the successful aging concept and its definition.Moderate strength of agreement was found between SA-CAG and SA-NM/EM definitions, but only fair agreement was shown between the others.The imperfect overlap of SA groups highlights the importance of considering different SA definitions and approaches when interpreting results from studies on SA.

F I G U R E 5 F I G U R E 6
Compared to TA greater TIV-adjusted hippocampal volume was found in SA-ALL, SA-CAG, SA-EM, and SA-CVLT controlling for age, sex, years of education.CAG, cognitive age gap; CVLT, California Verbal Learning Test; EM, episodic memory; NM, non-memory cognition; SA, successful cognitive aging; TA, typical aging; TIV, total intracranial volume.*p < 0.05, **p < 0.01.Compared to TA, lower entorhinal FTP PVC SUVR was found in all SA groups controlling for age, sex, years of education.CAG, cognitive age gap; CVLT, California Verbal Learning Test; EM, episodic memory; FTP, flortaucipir; NM, non-memory cognition; PVC, partial volume correction; SA, successful cognitive aging; SUVR, standardized uptake value ratio; TA, typical aging.*p < 0.05, **p < 0.01.
Cognitive age model cohort characteristics.
TA B L E 1

TA SA-CVLT vs. TA
Values represent either mean (SD) or n (%).Differences between typical aging (TA) and each successful cognitive aging (SA) group were assessed using Welch two-sample t-test (Cohen's d as effect size) for continuous variables and chi-square test (Cramér's V as effect size) for categorical variables.
Abbreviations: BMI, body mass index; CAG, cognitive age gap; CVLT, California Verbal Learning Test; EM, episodic memory; GDS, Geriatric Depression Scale; MMSE, Mini-mental State Examination; NM, nonmemory cognition; PiB DVR, Pittsburgh compound B distribution volume ratio; SA, successful cognitive aging; TA, typical aging; Y, yes. a Missing data for five TA individuals.b Missing data for three TA individuals.c Missing data for five TA and one SA-CAG/SA-ALL (participants were grouped as APOE