Memory testing

One hundred fourteen of the participants had complete data on a Memory Recall and Recognition test. This test is a variant of the Rey Auditory Verbal Learning and Memory task (Rey, 1964; Geffen et al., 1990), commonly used to provide measures of auditory-verbal learning, recall, and recognition. Briefly, the participant is presented with a sequence of 12 words binaurally via headphones, 1 s at a time. The participant is instructed to say back as many of the words as they can remember from the list, in any order. This procedure is then repeated three more times, with the same instructions. The total number of words recalled during these four trials is recorded. The participant is then presented with a second list of words (which are neither semantically nor phonetically related to the first list) and asked to say back as many of these as can be remembered (Distractor Trial). Immediately following this trial, the participant is asked to recall as many words as can be remembered from the first list (Immediate Recall Trial). Approximately 25 min later, after completing a number of other tests in the battery, the participant is again asked to recall as many words as possible from the first list (Delayed Recall Trial). The subject is then presented one at a time with a series of 24 words on the computer screen (Recognition Trial). Half of these words are the words from the first list; the remaining words are new words. The words are in fixed, pseudorandom order. Following each word, the participant is required to touch a “Yes” or “No” button on the touchscreen according to whether or not the word was in the first list.

MRI scan acquisition

MRI was conducted on a 1.5 Tesla Siemens Sonata system at Westmead Hospital, Sydney and a 1.5 Tesla Siemens Sonata at Perrett Imaging, Flinders University, Australia. The MRI protocol included a 3-D T1-weighted image acquired in the sagittal plane using an MPRAGE sequence (TR = 9.7 ms; TE = 4 ms; echo train 7; flip angle = 12°; TI = 200 ms; NEX = 1). A total of 180 1 mm slices (without gap) were acquired with a 256 × 256 matrix with an in plane resolution of 1 mm × 1 mm. Approximately 1% of the MRI data were excluded from the overall database due to MRI technical acquisition errors/artifact.

MRI analysis

Volumetric segmentation was performed on the 3D T1-weighted structural images with the FreeSurfer image analysis suite (version 4.3) (http://surfer.nmr.mgh.harvard.edu/), which has been shown to be comparable in accuracy to manual labeling (Fischl et al., 2002). The technical details of these procedures are described in detail elsewhere (Dale et al., 1999; Fischl et al., 1999; Fisher and Defries, 2002; Grieve et al., 2011). Volumetric measurements of the hippocampus and the amygdala were generated with FreeSurfer's automatic quantification of subcortical structures, which assigns a neuroanatomical label to each voxel in an MRI volume based on probabilistic information estimated from a manually labeled training set (see Figure ​Figure1).1). The classification technique employs a non-linear registration procedure that is robust to anatomical variability, as described in detail elsewhere (Fischl et al., 2002). Anatomically labeled structures were manually inspected for accuracy, and corrected if necessary. Total brain volumes, which were used as a covariate in the statistical analyses, were also generated automatically with the FreeSurfer processing stream.

Open in a separate window

Figure 1

Hippocampus and amygdala regions of interest. The bilateral hippocampi are depicted in yellow. The bilateral amygdalae are depicted in blue.

Main effects—hippocampus

Initial bivariate correlations showed that age was significantly negatively associated with hippocampal size across the entire age range (r = −0.374; p < 2.1 × 10⁻¹¹). However, the data were better fit by a quadratic function for age (r = −0.402; p < 4.3 × 10⁻¹²), reflecting the fact that hippocampal volume tends to increase until approximately age 30, at which point hippocampal volume tends to decline (Grieve et al., 2011). Age² was therefore included as a regressor in all models below.

Additionally, across the entire age range, educational attainment was positively and linearly associated with hippocampal size (R = 0.236; p < 7.9 × 10⁻⁵). When adjusting for age, sex, and total brain volume, however, the main effect of educational attainment did not account for significant unique variance in total hippocampal volume (Beta = 0.078; p < 0.076), left hippocampal volume (Beta = 0.086; p < 0.053), or right hippocampal volume (Beta = 0.065; p < 0.15).

Main effects—amygdala

Similar to the findings in the hippocampus, bivariate correlations revealed a significant negative association between amygdala volume and age (r = −0.239; p < 2.9 × 10⁻⁵), with the data best fit by a quadratic function for age (r = −0.254; p < 2.04 × 10⁻⁵) (Grieve et al., 2011).

A significant linear association between amygdala volume and educational attainment was also observed (r = 0.154; p < 0.01). Again, however, the main effect of educational attainment on total amygdala volume was no longer significant when adjusting for age, sex, and total brain volume (Beta = 0.029; p = 0.515). Nor was this effect significant when the left (Beta = 0.073; p < 0.094) or right (Beta = −0.016; p < 0.744) amygdala were considered separately.

Age by education interactions

Of interest next was the extent to which the effect of age on regional volume varied by an individual's educational attainment. To assess this hypothesis, models above were extended to include terms for an age × education interaction. We first assessed the hippocampus. A significant age × education interaction was present (R² change = 0.021, Beta = 0.498 p < 0.001), suggesting that the effect of age on hippocampal volume is not constant across all levels of educational attainment. This finding is portrayed in Figure ​Figure2,2, where it can be seen that the age-related decline in hippocampal volume is moderated by education. In this figure, the data are divided according to median split of individuals by age and education. There is no difference in hippocampal size by education among individuals younger than 35 years of age [t₍₁₃₆₎ = −0.636; p = 0.53]. However, among individuals older than 35 years, there is a highly significant difference in hippocampal size, with more highly educated individuals having larger hippocampi [t₍₁₃₅₎ = −5.6; p < 0.0001].

Open in a separate window

Figure 2

Higher educational attainment buffers age-related decrease in hippocampal volume. A significant education × age interaction is present for hippocampal volume (Beta = 0.498 p < 0.001), such that age-related decline in hippocampal volume is smaller among individuals with higher educational attainment. For clarity of presentation, educational attainment and age are represented graphically by median split. There was no difference in hippocampal volume according to education in the younger half of the subject pool. However, among older participants, more highly educated individuals had significantly larger hippocampi (^***p < 0.0001). Error bars represent ±1 standard error.

When considered separately, both left and right hippocampi showed a similar interaction (left hippocampus: Beta = 0.508, p < 0.001; right hippocampus: Beta = 0.462, p < 0.002).

Figure ​Figure33 shows a “dose-dependent” effect of education: The observed age-related volumetric decreases in hippocampal volume are greater among less educated individuals relative to more highly educated individuals. Less difference in hippocampal volume is seen with age among the highest educated individuals.

Open in a separate window

Figure 3

A dose-dependent moderating effect of education on age-related decrease in hippocampal volume. A “dose-dependent” effect of education is observed. Age-related decline in hippocampal volume is most pronounced among the least educated individuals. In contrast, age-related reduction in hippocampal volume is mitigated among individuals with higher educational attainment. Hippocampal volume is portrayed as the unstandardized residual, after adjusting for sex and total brain volume. The figure excludes the 13 data points that represented “influential points” as identified using the standardized difference in beta. Note that all analyses were performed using continuous variables for age, years of educational attainment, and hippocampal volume. For clarity of presentation, educational attainment is represented graphically in bins, with high school or less portrayed with circles/solid line; some college portrayed with diamonds/dashed-dotted line; and college or beyond portrayed with squares/dotted line.

Of note, there was a wide span of educational attainment across the entire age range, with the educational spectrum represented across each age range, from young adults to elderly individuals (see Table ​Table1).1). Nonetheless, there was a significant correlation between age and education (r = −0.257; p < 0.0001), such that older participants were on average less educated than younger participants. For this reason, we also examined the full model excluding “influential points” in this correlation. Influential points were identified using the standardized difference in beta, defined as the difference in the standardized regression coefficient as a result of deleting a particular case, divided by the standard error of this difference. Any value greater than 2/√n was excluded (Pedhazur, 1997b). Among the remaining 267 participants, the age × education interaction remained significant (Beta = 0.454; p < 0.001), suggesting that the moderating effect of education on age-related decreases in hippocampal volume is an accurate description of the sample, and not a spurious finding driven by the influence of a small number of subjects with extreme values. These 267 participants are represented in Figure ​Figure33.

No significant education × age interaction was observed in total amygdala volume, or left or right hemispheres considered separately, regardless of whether influential points were included or excluded, suggesting that the effect of age on amygdala volume is independent of educational attainment.

Implications

The idea that, among older individuals, higher levels of education confer benefits in brain structure or function that buffer against age-related changes or brain pathology is often referred to as “brain” or “cognitive” reserve (Stern, 2002, 2006). For instance, some studies have found that Alzheimer's Disease (AD) is less prevalent among individuals with higher levels of education (Kidron et al., 1997), and it has been suggested that this may be due to a protective effect whereby individuals with more education may not manifest clinical symptoms of disease until higher levels of neural pathology are reached (Scarmeas et al., 2006; Julkunen et al., 2010; Seo et al., 2011; Liu et al., 2012).

The present study provides some support for this theory. A “dose-dependent” effect of education was found, such that age-related decreases in hippocampal volume were steepest among the least educated individuals. In contrast, age-related decrease in hippocampal volume was less pronounced among individuals with higher educational attainment. It is thus possible that, among this sample of healthy individuals, higher levels of education were able to counteract typical age-related changes in hippocampal structure. The functional significance of these differential rates of decline may reflect greater retention of the cognitive functions, including declarative memory performance, which rely on this structure (Grieve et al., 2011).

Possible mechanisms

Many components of SES tend to vary with educational attainment and may themselves be important factors in understanding mechanistic pathways. Other objective socioeconomic indicators include measures of income and occupation, which were not available in the dataset studied here. Although education, occupation and income tend to be highly correlated (McLoyd, 1998), they may nonetheless act independently in predicting different outcomes of interest (Duncan and Magnuson, 2012; Noble et al., 2012). For instance, previous work has suggested that family income levels and parental educational attainment may not act in concert in predicting hippocampal and amygdala volumes in children (Hanson et al., 2011; Noble et al., 2012). Further, subjective social status, or an individual's perception of where he or she stands in the social hierarchy, may predict regional brain structure and/or function over and above objective socioeconomic measures (Gianaros et al., 2007a, 2008).

Socioeconomic variables—including differences in educational attainment—likely represent relatively distal forces operating on cognitive and neural outcomes. However, SES has been associated with many factors that have been shown to exert more proximal effects on regional brain structure. One well described pathway includes differences in exposure to stress or allostatic load (McEwen, 1998, 1999, 2000, 2001a,b; Evans, 2004; Evans and Schamberg, 2009; Gianaros and Manuck, 2010; McEwen and Gianaros, 2010).

In both animals and humans, the experience of stress has been linked to differences in the structure of the hippocampus (Rao et al., 2010; Tottenham and Sheridan, 2010). The animal literature has shown cascading negative effects of early life stress on hippocampal development at the cellular, anatomic and functional levels (Sapolsky, 1996; McEwen, 1999; Brunson et al., 2005; Gianaros and Manuck, 2010; McEwen and Gianaros, 2010; Tottenham and Sheridan, 2010). Conversely, adult offspring of female rodents that exhibit high licking and grooming of pups—a model of “parental nurturance”—show increased hippocampal synaptic density and plasticity (Liu et al., 2000; Champagne et al., 2008), and improved performance in hippocampal-dependent forms of memory (Liu et al., 2000).

In studies of human adults, self-reported stress exposure, stress-related mental illness, and the prior experience of child abuse have all been associated with decreased hippocampal volume (Sheline et al., 1996; Bremner et al., 1997; Geuze et al., 2005; Kitayama et al., 2005; Gianaros et al., 2007b; Tottenham and Sheridan, 2010; Dannlowski et al., 2012). [Findings concerning the association between stress exposure and hippocampal size in children, however, have been less consistent (Woon and Hedges, 2008; De Bellis et al., 2010; Rao et al., 2010)].

In the socioeconomically diverse sample of adults studied here, one possibility is that differences in the experience of stress may mediate the association between educational attainment and hippocampal volume. Future research will be necessary to directly test this putative mediating pathway.

Interestingly, and contrary to initial predictions, we only found a moderating effect of educational attainment on regional volume in the hippocampus, and not in the amygdala. Research in animals has demonstrated that the experience of stress profoundly affects development of the amygdala (Makino et al., 1994, 1999; Bonaz and Rivest, 1998; Wood et al., 2003; Moriceau, 2004), and human neuroimaging studies also suggest that the amygdala may be structurally altered by stress (Tottenham et al., 2010). Thus, if the reported effects of education could be completely explained by differences in exposure to stress, one would have expected to see effects in the amygdala as well. It is therefore likely that some other mechanistic factors might explain the effects of educational attainment on hippocampal volume described here. Of course, many factors, including but not limited to prenatal factors, nutrition, exercise, parenting, and the quality and quantity of the schooling experience, are associated with an individual's educational attainment, and may serve as more proximal mechanisms underlying the results reported here. Indeed, it is likely that multiple such experiential factors contribute to the moderating effects of educational attainment on hippocampal volume.

Finally, while differences in experience have direct effects on the hippocampus, it should also be noted that interactions between age, educational attainment and hippocampal volume may in part be explained by differences in genes and/or gene-environment interactions. Future research will be necessary to disentangle the various causal pathways.

Effects of education across the lifespan

In a recent study of 60 socioeconomically diverse children and adolescents, we found that parental education levels predicted children's amygdala volumes, but not hippocampal volumes (Noble et al., 2012). (In contrast, parental income levels predicted hippocampal but not amygdala volumes.) This differs from the findings reported here, in which educational attainment moderated hippocampal, but not amygdala volume. Several notable differences may help to reconcile these findings.

The two studies included a non-overlapping age range of participants: the previous study examined children between the ages of 5 and 17, whereas the present study includes individuals between the ages of 17 and 87. Human neuroimaging studies have suggested that the effects of adversity on amygdala structure may vary based on the age of the population studied. In adulthood, the experience of highly stressful events has been associated with smaller amygdala size (Driessen et al., 2000; Schmahl et al., 2003; Tottenham and Sheridan, 2010), whereas in children, adverse caregiving circumstances have been associated with atypically large amygdala volume (Mehta et al., 2009; Tottenham et al., 2010). Thus, it is similarly possible that any effects of education on amygdala structure may be different in childhood versus adulthood, even considering the broad range of adult ages studied here.

A second possible explanation is that the previous study in children examined parental education levels, whereas the current study examined the participants' education levels. (Unfortunately parental educational attainment was not available in the current dataset.) It is therefore possible that differences in parental education may reflect differences in childhood experience or parenting that are not necessarily captured by one's own ultimate educational attainment. For instance, lower parent education has been associated with lower levels of parental nurturance (Brooks-Gunn and Markman, 2005), which may have particular importance for amygdala structure (Tottenham et al., 2010). Additionally, it is possible that the association between adult educational attainment and hippocampal volume may be mediated by the material resources individuals were exposed to as children (e.g., parental SES). Indeed, differential effects on hippocampal volumes have been described as a function of childhood SES versus adult SES. For example, several reports have found that higher childhood SES is associated with larger hippocampi (Hanson et al., 2011; Jednoróg et al., 2012; Noble et al., 2012). Intriguingly, Staff et al. (2012) recently reported that childhood SES, as operationalized by adult recall of childhood home conditions and paternal occupation, predicted adult hippocampal volume, whereas adult SES did not. This suggests that the high degree of neural plasticity in childhood may underlie effects observed decades later. Ideally, future work in adults will incorporate measures of both parental education and participant education to better sort out these disparate effects.

A third possibility is that the previous study, with only 60 participants, had reduced power with which to detect an effect of education on hippocampal volume, whereas the greater sample size in the present study allowed such an effect to be revealed here. This explanation alone would not, however, explain why we failed to find a link between educational attainment and amygdala volume in the present study.

We acknowledge the data and support provided by BRAINnet; www.BRAINnet.net, under the governance of the BRAINnet Foundation. BRAINnet is the scientific network that coordinates access to the Brain Resource International Database for independent scientific purposes. We also thank the individuals who gave their time to participate in the database. Funding for this work was partially supported by the John M. Driscoll, MD Scholars Program awarded to Kimberly G. Noble, and by NIH grants AG029949 and AG034189, and a grant from the Alzheimer's Association, awarded to Adam M. Brickman.