The positive impacts of early-life education on cognition, leisure activity, and brain structure in healthy aging

The effects of age×education on cognitive performance

A total of 659 participants were enrolled in the present study from Beijing Ageing Brain Rejuvenation Initiative (BABRI) database (Supplementary Figure 1). The demographic and clinical characteristics of the participants in the large samples are presented in Table 1. No differences in age or APOE ε4 genotype were found between the high education and low education groups. A higher proportion of males had more education than females. Additionally, participants with a high level of education had significantly better cognitive performances compared to those with low education attainment (Supplementary Figure 2). Among the cognitive scores, only TMT-B and SCWT-A exhibited a significant education×age interaction (Table 1), suggesting a more age-related reduction in the executive function of elderly with less education.

The influence of education levels on leisure activities

Habitual leisure activity levels were grouped according to intellectual, physical and social factors. Scores of intellectual, physical and social types of leisure activities were significantly higher for participants with a high level of education, but no significant education×age interaction was found (Table 1).

Group differences in leisure activities between the elderly with higher education levels and those with lower education levels were compared using a scale that represents a wide range of elderly people’s daily life in China (Supplementary Table 1). The highly educated elderly prefer to participate in knowledge-dependent activities (including reading, writing, attending a senior university, using a computer, doing crossword puzzles, magic cube, or solitaire, and doing Chinese traditional martial arts) and the less educated elderly participated in knowledge independent activities (including playing chess, poker, or mahjong, manufacturing by hand, and owning pets).

Mediation analysis

In the mediation analysis, the independent factor was the education level and the dependent variables were cognitive measures that showed significant education×age interactions, such as the TMT-B and SCWT-A. The proposed mediators were the intellectual, physical and social activity scores. The mediation analysis indicated that intellectual activity mediates the effect of education on the TMT-B (Z=-3.40, p<0.0005) and SCWT-A (Z=-3.54, p<0.0005) performance, and social activity mediates the effect of education on the TMT-B (Z=-2.00, p<0.05) significantly and SCWT-A (Z=-2.17, p<0.05) performance (Figure 1). Furthermore, we added all the cognitive tests as outcome variable, and we did find some interesting results. The results show that education affects cognitive performance in almost all fields through mental activities, while education can affect attention function, executive function and memory function through social activities. However, we did not find a significant mediating effect of physical activity (Supplementary Figure 3).

Figure 1. The mediation model. The mediation model illustrating the direct effect of education on intellectual/social/ physical activity, the direct effect of education on TMT-B and SCWT-A performance, the direct effect of intellectual/social/ physical activity on TMT-B and SCWT-A performance and the mediating effect of intellectual/social/ physical activity on the association between educational levels and TMT-B/SCWT-A performance, as indicated by the path coefficients and the p values.

The age×education effects on gray matter volume

The demographic and clinical characteristics of the small samples with available MRI data are presented in Supplementary Table 2. A main effect of educational status was found on the left precentral gyrus, the bilateral medial orbital superior frontal gyrus (ORBsupmed), the bilateral rectus, the bilateral posterior cingulate cortex, and the left amygdala (p<0.05, uncorrected, Figure 2A). After gender and APOE ϵ4 status were regressed out, significant age×education interaction effects were found in several frontal, temporal and occipital regions (p<0.05, uncorrected, Figure 2A), including the ORBsupmed.L, the left anterior cingulate, and the paracingulate gyrus (ACG.L), which persisted after false-discovery rate (FDR) corrections (Figure 2B). There was a trend toward higher age-related GM atrophy among the less educated elderly compared to the more educated elderly in these regions.

Figure 2. The age×education effects on gray matter volume. (A) Maps of the significant age×education interaction on gray matter volume. A significant main effect of education on gray matter was also found (p<0.05, uncorrected). (B) Scatterplots illustrating the age×education interaction for ORBsupmed.L and ACG.L (q<0.05, FDR-corrected). Specifically, the plots were drawn to illustrate how the GM volume changes within the actual range of age. ORBsupmed.L=left medial orbital superior frontal gyrus; ACG.L=left anterior cingulate and paracingulate gyrus.

The age×education effects on white matter integrity

Similarly, we investigated the interaction effects on the fibers. The main effect of education was significant only on the left cingulum (hippocampus part) mean diffusivity MD value. The MD of the forceps major and left superior longitudinal fasciculus (SLF, temporal part) were affected by the age×education interaction, while the FA was affected only in the forceps major, as shown in the scatter diagram (p<0.05, uncorrected, Figure 3). Our post-hoc analysis found that the years of education was negatively correlated with the MD values of the forceps major and the SLF.temporal part (Supplementary Figure 4).

Figure 3. The effect of age×education on white matter integrity. The fibers in red exhibited significant age×education interaction effects (p< 0.05), including the FA and MD of the forceps major, and the MD of the SLF.temporal part. The plots were drawn to illustrate how the white matter changes within the actual range of age. FA=fractional anisotropy; MD=mean diffusivity; SLF=superior longitudinal fasciculus.

Structural equation modeling

The results above demonstrate that the interaction of education and age has effects on both cognitive performance and brain structure. We further tested whether the brain-cognition relationship was also influenced by the age×education interaction. We analyzed the data using structural equation modeling to discuss the relationship between education, WM, GM and cognition, as shown in the following figure. Please notice that the only the MD value of the SLF. Temporal part was included in the model as the WM latent variable. The reason why we only considered SLF.temporal part (MD value) instead of the three white matter indicators was because the poor relationship among the three indicators made it unreasonable to extract a latent variable, so it wouldn’t make sense if the three indicators were considered together in one model. Instead, we modeled on each one separately, and only modeling on SLF.temporal part (MD value) revealed a meaningful model fit (RMSEA=0.0468, NFI=0.934, CFI=0.989, GFI=0.963, AGFI=0.868).

As shown in Figure 4, the first path of each measurement model was latent variable scaling, and the values on each path were marked with the estimated values of standardized parameters of the path. Results indicated that education had a significant main effect on cognition only, but not on GM or WM. The interaction between age and education, on the other hand, had a significant effect on WM, but not on cognition or GM. However, the significant effect from WM to cognition was detected, indicating the interaction between age and education could affect cognition indirectly through SLF.temporal part.

Figure 4. The structural equation modeling of the relation among white matter, grey matter and cognition. COG= Cognitive Tests; GM=gray matter; WM=white matter; ORB_L= left medial orbital superior frontal gyrus; ACG_L=left anterior cingulate and paracingulate gyrus; MD_SLF= the mean diffusivity of the SLF.temporal part; SCWT-A=Stroop Color and Word A Test; TMT-B=Trail Making B Test.

Limitations

Our study has some limitations. Firstly, we only recruited Chinese elders who lived in the city and had a minimum of primary education. It is also important to investigate the effects of education in rural Chinese populations and to determine whether education influences the regional GM and WM differently among those with less than 6 years of schooling. Secondly, the elderly ranged in age from 48 to 87 years in our cognition sample, and from 60 to 81 years in the MRI sample. Further studies with a larger age span are needed to evaluate this interaction. Thirdly, data from prospective population-based autopsy studies of older individuals have suggested that the protective effect of education varied by brain pathological severities or cognitive stages [46]. The present study only focused on cognitively normal elderly, and did not include patients with cognitive impairments or AD. Furthermore, we did not rule out the role of occupation and income, which are the factor that is largely responsible for associations between education and later life cognitive functioning. Finally, our way of categorizing study participants into high/low education classes was only a proxy, it cannot reflect all aspects of latent and more abstract ‘education’.

Participants

Participants in the present study were from the BABRI database, which is an ongoing longitudinal study that plans to collect cognitive behavioral, sociodemographics, lifestyle, potential influencing factor, and clinical data from Beijing’s elderly adults in five time-windows over 10 years, with more than 1600 eligible urban participants. The study was undertaken after obtaining the understanding and written consent of each subject, and the study conformed to the guidelines set out in the 2013 World Medical Association Declaration of Helsinki. The use of human subjects for this study has been approved by the institutional review board (IRB) at the Imaging Center for Brain Research at Beijing Normal University. Written informed consent was obtained from each participant.

A total of 1211 urban elderly living in Beijing, China were included in the BABRI database. Finally, 659 participants were enrolled in the present study (Supplementary Figure 1). These individuals included all right-handed native Han Chinese. Participants qualified for this study if they met the following criteria: 1) MMSE≥24; 2) no history of neurologic, psychiatric, or systemic illnesses known to influence cerebral function, including serious vascular diseases, head trauma, tumor, current depression, alcoholism, and epilepsy; 3) having completed 6 or more years of education; and 4) without a history of using psychoactive medications. People with suspected MCI or AD were excluded from the study. The diagnostic criteria for MCI was stated in Petersen [47]. The protocol was approved by the Institutional Review Board of the State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University. Written consent was obtained from each subject. Education level was assessed at baseline by recording the number of years of schooling. Two education levels were defined: the low education level (≤12 years) and the high education (>12 years) level.

Neuropsychological testing

All participants underwent a battery of neuropsychological tests which was comprised of the following 5 cognition domains with the tests included in parentheses: 1) memory (Auditory Verbal Learning Test (AVLT), Rey-Osterrieth Complex Figure test (ROCF)(recall) and backward Digit Span); 2) attention (Trail Making Test (TMT) A, Symbol Digit Modalities Test (SDMT) and Stroop Color and Word Test (SCWT)-B); 3) visuo-spatial ability (ROCF (copy), Clock-Drawing Test (CDT)); 4) language (Category Verbal Fluency Test (CVFT), Boston Naming Test (BNT)); and 5)executive function (TMT-B and SCWT-C). The specific neuropsychological test procedures have been described previously [27].

Analysis of genotyping

DNA was extracted from the subjects’ blood samples according to standard procedures for the subsequent characterization of the APOE genotype using PCR (Applied Biosystems, Foster City, CA). All participants were genotyped for two SNPs in the APOE gene (rs429358 and rs7412) using previously published methods [48].

The measurement of leisure activity

Habitual leisure activity levels were measured using a personal information questionnaire, which is a subjective measure including 23 questions that ask participants to recall their leisure activity in the past one year. Leisure activities were defined as activities performed for enjoyment; these activities were independent of work and included reading; writing; participating in a senior citizens’ university; playing chess, poker, or mahjong; and handcrafts. The frequency of participation in each of the above activities was defined as frequent if the subject participated several times per week and rare if the subject participated less than once per week. Additionally, each leisure item was grouped according to intellectual, physical and social factors, and three aggregate scores were assigned to each participant for the subsequent analysis. In an attempt to create a valid data set, we excluded incomplete questionnaires.

MRI data acquisition

Among the participants, 78 subjects (41 elderly with low education and 37 elderly with high education) received high-quality MRI scanning, which included a 3D T1-weighted MRI scan and a DTI scan. The MRI data were acquired using a SIEMENS TRIO 3T scanner in the Imaging Center for Brain Research at Beijing Normal University. Participants laid supine with their head fixed snugly by straps and foam pads to minimize head movement. T1-weighted, sagittal 3D magnetization prepared rapid gradient echo (MP-RAGE) sequences were acquired and covered the entire brain [176 sagittal slices, repetition time (TR)=1900 ms, echo time (TE)=3.44 ms, slice thickness=1 mm, flip angle=9°, inversion time=900 ms, field of view (FOV)=256×256 mm², acquisition matrix=256×256]. For each DTI scan, images covering the whole brain were acquired by an echo-planar imaging sequence with the following scan parameters: TR=9500 ms, TE=92 ms, 30 diffusion-weighted directions with a b-value of 1000 s/mm², and a single image with a b-value of 0 s/mm², slice thickness=2 mm, no inter-slice gap, 70 axial slices, acquisition matrix =128×128, FOV=256×256 mm², averages=3.

Diffusion MRI data processing

DTI data were processed with the FDT toolbox in FSL (www.fmrib.ox.ac.uk/fsl). First, the DICOM files of all subjects were converted into NIfTI using the dcm2nii tool embedded in MRIcron. Second, the brain mask was estimated, and the resulting brain mask was used for the subsequent processing steps. Third, the non-brain spaces in the raw images were cut off leading to a reduced image size, reducing the memory cost and speeding up the processing in subsequent steps. Fourth, each image was coregistered to its corresponding b0 image using an affine transformation to correct the eddy-current induced distortions and simple head-motion artifacts. The diffusion gradient directions were adjusted accordingly. Fifth, a voxel-wise calculation of the tensor matrix and the diffusion tensor metrics were yielded for each subject, including the FA and mean diffusivity (MD).

To identify major white matter tracts, we used the digital WM atlas JHUICBM-DTI-81 (http://cmrm.med.jhmi.edu/), a probabilistic atlas generated by mapping the DTI data of 78 subjects to a template image. The atlas contains 20 main WM bundles and has three individual sets of sub-templates with different probability levels in the probability tractography map: 0%, 25% and 50%. In this study, we chose to use the 25% threshold subtemplate, which contains 20 major tracts. The JHU-WM atlas was overlaid on the WM skeleton of each subject in the CBM-DTI-81 space, such that each skeleton voxel could be categorized into one of the major tracts. Then, the FA and MD at the skeleton voxels within each tract were calculated.

T1-weighted MRI data processing

All subjects’ individual structural images (3D T1-weighted anatomical images) were processed using theVBM8 toolbox of SPM (http://dbm.neuro.uni-jena.de/vbm/). The transformed T1-weighted images were then segmented and spatially normalized into GM, WM and cerebrospinal fluid in standard MNI space using a unified segmentation algorithm [49]. Modulation was further applied to the GM images to compensate for the effect of spatial normalization using linear and nonlinear methods.

ROIs were defined by the parcellation of the brain into the 90 regions of the automated anatomical labeling (AAL) template based on T1-weighted images. Each ROI volume for each subject was obtained by averaging the voxel values over the GM maps across all voxels within each AAL region.

Statistical analysis

Independent two-sample t-tests were used to assess between-group differences in age and education. The chi-square test was used to compare proportional differences in the gender, APOE ε4 genotype, and frequency of leisure activities. We then assessed the effect of the age×education interaction on each cognitive test, and the aggregate leisure activity scores (intellectual, physical and social scores). Specifically, a GLM with “age”, “education” and “age×education” as predictor variables was used, wherein gender and APOEε4 genotype were included as covariates.

We further investigated whether the expected education-related cognitive performance was mediated or moderated by leisure activity alterations in the large sample. To this aim, the mediation analyses were performed by the Hayes Process macro [50] (http://www.afhayes.com/). A mediator model typically consists of one independent variable, a mediator variable, and a dependent variable. This macro runs a series of ordinary least-squares regressions with the centered product term. In the present study, education constitutes the independent variable, cognitive measure was the dependent variable and aggregate leisure activity scores were the mediating variable. Thus, we would assume that the association between education and cognitive performance is mediated through the effect of education on leisure activity (i.e., an indirect effect). The significance threshold for the Sobel test was set at p<0.05. Structural equation modeling was employed to examine the association among education, age, cognition, gray matter volume and white matter integrity by Lisrel.