Cognitive Reserve Modifies Age-Related Alterations in CSF Biomarkers of Alzheimer's Disease

Abstract

Importance

Although advancing age is the strongest risk factor for the development of symptomatic Alzheimer's disease (AD), recent studies have shown that there are individual differences in susceptibility to age-related alterations in the biomarkers of AD pathophysiology.

Objective

In this study, we investigated whether cognitive reserve modifies the adverse influence of age on key cerebrospinal fluid (CSF) biomarkers of AD.

Design, Setting, and Participants

Cross-sectional cohort of 268 individuals (211 cognitively normal and 57 cognitively impaired) from the Wisconsin Registry for Alzheimer's Prevention and the Wisconsin Alzheimer's Disease Research Center participated in this study. They underwent lumbar puncture for collection of CSF samples, from which amyloid-β 42 (Aβ42), total tau (t-tau), and phosphorylated tau (p-tau) were immunoassayed. Additionally, we computed t-tau/Aβ42 and p-tau/Aβ42 ratios. Cognitive reserve was indexed by years of education, with ≥16 years taken to confer high reserve. Covariate-adjusted regression analyses were used to test whether the effect of age on CSF biomarkers was modified by cognitive reserve.

Main outcome measures

CSF levels of Aβ42, t-tau, p-tau, t-tau/Aβ42, and p-tau/Aβ42.

Results

There were significant age*cognitive reserve interactions for CSF t-tau (p=.019), p-tau (p=.009), t-tau/Aβ42 (p=.021), and p-tau/Aβ42 (p=.004). Specifically, with advancing age, individuals with high cognitive reserve exhibited attenuated adverse alterations in these CSF biomarkers compared with individuals with low cognitive reserve. This attenuation of age effects by cognitive reserve tended to be more pronounced in the cognitively-impaired group compared with the cognitively-normal group. Lastly, there was modest evidence of a dose response relationship such that the effect of age on the biomarkers was progressively attenuated given additional years of schooling.

Conclusions and Relevance

In a sample comprised of both cognitively normal and cognitively impaired individuals, higher cognitive reserve was associated with a diminution of age-related alterations in CSF biomarkers of AD. This suggests one pathway through which cognitive reserve might favorably alter lifetime risk for symptomatic AD.

INTRODUCTION

Evidence from autopsy,¹ epidemiological,² and cohort studies³ show that advancing age is the strongest risk factor for both the accumulation of Alzheimer's disease (AD)-related pathophysiological abnormalities and the ultimate development of symptomatic AD. Even so, there is significant interindividual heterogeneity in the age-specific prevalence of AD-related pathologies such that some individuals appear relatively spared the development of AD pathognomonic brain lesions even into old age.¹ The concept of cognitive reserve (CR) has been postulated to account not only for the mismatch between AD lesions and cognitive impairment but also for the overall decreased diathesis for these pathological changes in some individuals.^{1, 4} Essentially, the CR hypothesis posits that the intellectual enrichment that accrues from various life exposures—such as high educational attainment and engagement in cognitively-stimulating activities—lessens the adverse effect of brain pathology on cognitive function, and might also attenuate the accumulation of such pathologies.⁴

Although AD brain abnormalities, specifically neuritic plaques and neurofibrillary tangles, have traditionally been quantified histologically at autopsy, recent advances in the field have made possible the in vivo measurement of biomarkers believed to reflect these underlying pathologies.⁵ These biomarkers include cerebrospinal fluid (CSF) amyloid-β 42 (Aβ42), total tau (t-tau), and phosphorylated tau (p-tau) which presumptively tag cerebral amyloid-β plaques, neuronal injury, and neurofibrillary tangles respectively.⁵ These biomarkers have been associated with prospective cognitive decline and risk of progressing to AD in cognitively normal individuals^{6, 7} and in those with mild cognitive impairment (MCI);⁸ as well as with progressive cognitive deterioration and mortality in persons with probable AD dementia.⁹

In this study, we investigated whether educational attainment—the most-widely used proxy for CR—modifies age-related alterations in these CSF biomarkers of AD. Specifically, we hypothesized that the known adverse age-dependent changes in these biomarkers will be attenuated among individuals with high educational attainment, i.e., those with high CR.

MATERIALS AND METHODS

Participants

Two hundred and sixty-eight enrollees in the Wisconsin Registry for Alzheimer's Prevention (WRAP) and the Wisconsin Alzheimer's Disease Research Center (WADRC) participated in this study. The sample comprised 211 cognitively-normal late-middle-aged adults enrolled in either the WRAP or the WADRC, and 57 cognitively-impaired adults enrolled in the WADRC. The cognitively-impaired group included 16 persons with amnestic MCI and 41 persons with mild AD. All participants were diagnostically characterized in standardized, multidisciplinary, consensus conferences: diagnoses of MCI and AD were made using applicable clinical criteria^{10, 11} whereas cognitive normalcy was adjudicated based on intact performance on a comprehensive battery of neuropsychological tests, absence of functional impairment, and absence of neurological/psychiatric conditions that might impair cognition.³ Women comprised 62.3% of the sample and the mean age was 62.62±8.64 years. CR was indexed by years of education. Individuals with less than 16 years of education were considered as having Low CR (n=88) whereas those with ≥16 years of education were considered as having High CR (n=180).^{12, 13} The University of Wisconsin Institutional Review Board approved all study procedures and each participant provided signed informed consent before participation.

CSF Assessment

Lumbar puncture for collection of CSF samples was performed in the morning after a 12-hour fast with a Sprotte 24- or 25-gauge spinal needle at L3/4 or L4/5 using gentle extraction into polypropylene syringes. Each sample consisted of 22 mL of CSF, which was then combined, gently mixed, and centrifuged at 2000g for 10 minutes. Supernatants were frozen in 0.5 mL aliquots in polypropylene tubes and stored at −80°C. The samples were immunoassayed for Aβ42, t-tau, and p-tau (at threonine 181) using INNOTEST enzyme-linked immunosorbent assays (Fujirebio, Gent, Belgium) by board-certified laboratory technicians who were blind to clinical data and used protocols accredited by the Swedish Board for Accreditation and Conformity Assessment as previously described.¹⁴ Using these data, we additionally computed t-tau/Aβ42 and p-tau/Aβ42 ratios.

Statistical Analyses

To investigate whether CR modifies the adverse influence of age on CSF biomarkers, we fitted a series of simultaneous entry linear regression analyses—one for each CSF biomarker—that included terms for age, sex, apolipoprotein E4 (APOE4, 0 vs. ≥1 allele) genotype, cognitive status (i.e., cognitively normal vs. cognitively impaired), CR, and an age*CR interaction. This age*CR interaction term was the effect of primary interest in all models. Where significant, it would indicate a differential effect of age on CSF biomarkers as a function of CR (Low vs. High). Fitting our regression models via simultaneous entry ensured that the age*CR effect, where observed, was independent of—rather than a proxy for—the influence of other covariates (e.g., cognitive status or APOE4) on the CSF biomarkers. Correlational analyses were first used to assess the relationship between age and the CSF biomarkers. All analyses were conducted using IBM SPSS, version 21.0. Only findings with p ≤ .05 (2-tailed) were considered to be significant.

RESULTS

Participant Characteristics

Table 1 details the characteristics of the participants, for the pooled sample and when stratified by cognitive status and CR. Within the pooled sample, the average age was 62.62±8.64 years (range, 45–93), average years of education was 15.94± 2.44 (range, 8–21), 62.3% were women, and 45.9% were APOE4 positive. Among the cognitively normal, those with Low CR had fewer years of education (by design), lower memory test scores, more women, and more persons with a parental family history of dementia compared with the High CR group. Among the cognitively impaired, the Low CR group had less education (by design), worse MMSE, Clinical Dementia Rating sum of boxes, and memory test scores; and higher p-tau levels. These differences were assessed at the pre-defined alpha level of .05 (2-tailed).

Table 1.

Characteristics of study participants

Characteristic	Total sample, n=268	Cognitively Normal, n=211		Cognitively Impaired, n=57
		Low CR, n=64	High CR, n=147	Low CR, n=24	High CR, n=33
Age, y, mean (SD)	62.62 (8.64) [range, 45–93]	58.91 (5.98) [range, 45–73]	59.95 (5.80) [range, 47–72]	74.22 (8.84) [range, 58–93]	73.29 (8.05) [range, 58–91]
Education, y, mean (SD)	15.94 (2.44) [range, 8–21]	13.25 (1.08)† [range, 12–15]	17.33 (1.35) [range, 16–21]	12.46 (1.67)§ [range, 8–15]	17.45 (1.50) [range, 16–20]
MMSE, mean (SD)	28.14 (3.11)	29.22 (1.02)	29.46 (.72)	22.13 (4.25)§	24.70 (3.64)
Women, n (%)	167 (62.3)	51 (79.7)†	97 (66.0)	10 (41.7)	9 (27.3)
APOE4 positive, n (%)	123 (45.9)	26 (40.6)	62 (42.2)	15 (62.5)	20 (60.6)
Family history of dementia, n (%)	197 (73.5)	58 (92.1)†	110 (74.8)	13 (54.2)	16 (51.6)
CDR Global=0/0.5/1.0/2.0, n	203/47/17/1	61/3	139/8	1/13/9/1	2/23/8/0
CDR Sum of Boxes, mean (SD)	.78 (1.64)	.05 (.18)	.06 (.21)	4.17 (1.89)§	2.88 (1.75)
RAVLT Total Learning, mean (SD)	45.18 (13.96)	49.05 (8.25)†	51.69 (8.99)	21.29 (10.14)§	27.27 (6.99)
RAVLT Long Delay, mean (SD)	8.33 (4.69)	10.19 (2.75)	10.40 (3.03)	1.21 (2.38)	1.12 (1.58)
Aβ42, ng/L, mean (SD)	672.24 (237.36)	734.32 (204.75)	733.60 (207.31)	440.21 (211.45)	442.22 (200.55)
t-tau, ng/L, mean (SD)	389.48 (260.11)	295.26 (99.17)	304.79 (134.31)	802.62 (413.76)	648.55 (300.50)
p-tau, ng/L, mean (SD)	47.66 (22.36)	42.14 (12.98)	41.15 (14.13)	81.50 (35.83)§	62.44 (25.38)
t-tau/Aβ42, mean (SD)	.74 (.80)	.42 (.21)	.44 (.26)	2.06 (1.15)	1.74 (1.02)
p-tau/Aβ42, mean (SD)	.09 (.07)	.06 (.03)	.06 (.03)	.21 (.11)	.17 (.09)

Cognitively impaired=having a diagnosis of mild cognitive impairment or probable Alzheimer's dementia; CR=cognitive reserve (Low=less than 16 years of education, High=16 or more years of education); MMSE=Mini-Mental State Exam; APOE4=apolipoprotein E4 allele; CDR=Clinical Dementia Rating Scale; RAVLT=Rey Auditory Verbal Learning Test; Aβ42=amyloid-β 42; t-tau=total tau; p-tau=phosphorylated tau.

^†indicates p <.05 for comparison between Low CR and High CR groups among cognitively normal participants

^§indicates p < .05 for comparison between Low CR and High CR groups among cognitively impaired participants

Association between Age and CSF Biomarkers

Pearson correlations between age and CSF biomarkers within the pooled sample were as follows: Aβ42 (r=−.30, p<.001), t-tau (r=.52, p<.001), p-tau (r=.47, p<.001), t-tau/Aβ42 (r=.48, p<.001), and p-tau/Aβ42 (r=.46, p<.001). In the cognitively-normal group, the correlations were Aβ42 (r=.01, p=.884), t-tau (r=.21, p=.002), p-tau (r=.23, p=.001), t-tau/Aβ42 (r=.19, p=.007), and p-tau/Aβ42 (r=.18, p=.008). In the cognitively-impaired group, they were Aβ42 (r=.24, p=.070), t-tau (r=.13, p=.347), p-tau (r=.14, p=.310), t-tau/Aβ42 (r=−.17, p=.220), and p-tau/Aβ42 (r=−.14, p=.290). Overall, these correlations indicate that age is linked with interindividual variation in these biomarkers, as noted in prior reports.^{1, 3, 15} Although unexpected, the positive correlation between age and Aβ42 within the cognitively-impaired group would appear consistent with accumulating evidence that the well-documented AD-related reduction in Aβ42 levels might be preceded by an initial phase of elevated levels.^16–20

Cognitive Reserve and Age-Related Alterations in CSF Biomarkers

There were significant age*CR interactions for CSF t-tau, p-tau, t-tau/Aβ42, and p-tau/Aβ42 but not for Aβ42 (Table 2). To display these graphically, we followed standard procedure for generating plots for interactions between a continuous (i.e., age) and a categorical (i.e., CR) variable, which entail solving the regression equation at specific “anchor points” on the continuous variable.²¹ In our case, we solved the equation for age=50 years (“Younger”) and age=80 years (“Older”). These solutions, shown in Figure 1A–E, revealed that adverse change in CSF biomarkers due to Older age was more pronounced in Low CR individuals than in High CR individuals. Although the age*CR interaction was not significant for Aβ42, we elected to plot the Aβ42 results for completeness' sake. Overall, percent reduction in the effect of Older age on CSF biomarkers within the High CR group vis-à-vis the Low CR group ranged from 72% (for p-tau) to 180% (for t-tau/Aβ42).

Table 2.

Cognitive reserve modifies the association between age and CSF biomarkers of AD

Biomarker	Age*CR1		Age(Low CR)2		Age(High CR)3
	β (SE)	p	β (SE)	p	β (SE)	p
Aβ42	2.25 (2.84)	.428	---	---	---	---
t-tau	−6.72 (2.84)	.019	271.96 (75.76)	<.001	70.34 (65.96)	.287
p-tau	−.71 (.27)	.009	29.40 (7.15)	<.001	8.19 (6.22)	.189
t-tau/Aβ42	−.02 (.01)	.021	.30 (.21)	.154	−.24 (.18)	.175
p-tau/Aβ42	−.002 (.001)	.004	.04 (.02)	.030	−.02 (.02)	.227

CR=cognitive reserve; β = regression estimate; SE = standard error; Aβ42 = amyloid-β 42; t-tau = total tau; p-tau = phosphorylated tau

¹The regression estimates and associated p values are for the age*cognitive reserve interactive term in each biomarker's model. This term assesses whether cognitive reserve modifies the effect of age on the examined biomarker.

²The regression estimates and associated p values are for the simple main effect of age on each biomarker within the Low CR group.

³The regression estimates and associated p values are for the simple main effect of age on each biomarker within the High CR group.

Variables included in the model were age, sex, apolipoprotein E4 genotype, cognitive status (i.e., cognitively normal vs. cognitively impaired), cognitive reserve, and an age*cognitive reserve interaction; with the age*cognitive reserve interaction term being the effect of primary interest. The analyses were run in the pooled sample of cognitively normal and cognitively impaired persons. Because cognitive status is adjusted for in the models, the age*cognitive reserve effect is deemed to be over and above the influence of cognitive status on the CSF biomarkers.

Figure 1.

Higher cognitive reserve is associated with a diminution of age-related alterations in CSF biomarkers of AD

CSF=cerebrospinal fluid, AD=Alzheimer's disease, CR=cognitive reserve [Low=less than 16 years of education; High=16 or more years of education], t-tau=total tau, p-tau= phosphorylated tau, Aβ42=amyloid-β 42.

Panels display adjusted means and standard errors from analyses that, within the pooled sample of cognitively normal and cognitively impaired persons, modeled each CSF biomarker as a function of age, sex, apolipoprotein E4 genotype, cognitive status (i.e., cognitively normal vs. cognitively impaired), CR, and an age*CR interaction. The age*CR interaction term was the effect of primary interest in all models. Because cognitive status was adjusted for in the models, the age*CR effect is deemed to be over and above the influence of cognitive status on the CSF biomarkers.

Although age was included in the analyses as a continuous variable, for the purposes of graphing the study findings we chose two anchor points to represent Younger Age (age=50 years, black bars) and Older Age (age=80 years, red bars).

Supplemental Analyses

We conducted two supplemental analyses to further investigate our initial age*CR findings. The first was to determine whether CR modified age effects in a dose-response fashion. For this purpose, we redefined CR as either Low (≤12 years of education, n=39), or Medium (13–15 years of education, n=49), or High (≥16 years of education, n=180). When our models were re-fit using this new CR variable, we observed near-significant age*CR effects for p-tau (B(SE)= −.28(.16), p= .083), and p-tau/Aβ42 (B(SE)= −.001(.001), p= .053); trending effects for t-tau (B(SE)= −2.26(1.70), p= .185) and t-tau/Aβ42 (B(SE)= −.01(.01), p= .173); and a non-significant effect for Aβ42 (B(SE)=.91(1.68), p=.590). Simple main effect analyses revealed evidence for a dose-response relationship wherein the adverse effect of age on the biomarkers was progressively diminished at higher CR levels (Figure 2A–E). Of note, when education was treated as a continuous variable, a significant age*CR interaction was only seen for p-tau/Aβ42 (p=.028) with trends for p-tau (p=.117) and t-tau/Aβ42 (p=.168). And, when Low vs. High CR was defined as ≤ 12 vs. >12 years of education, there were no significant age*CR interactions (p's ≥.468).

Figure 2.

Dose-response in the modification of age effect on CSF biomarkers of AD by cognitive reserve

CSF=cerebrospinal fluid, AD=Alzheimer's disease, CR=cognitive reserve [Low=8 to 12 years of education; Medium=13 to 15 years of education; High=16 or more years of education], t-tau=total tau, p-tau= phosphorylated tau, Aβ42=amyloid-β 42.

Panels display adjusted means and standard errors from analyses that, within the pooled sample of cognitively normal and cognitively impaired persons, modeled each CSF biomarker as a function of age, sex, apolipoprotein E4 genotype, cognitive status (i.e., cognitively normal vs. cognitively impaired), CR, and an age*CR interaction. The age*CR interaction term was the effect of primary interest in all models. Because cognitive status was adjusted for in the models, the age*CR effect is deemed to be over and above the influence of cognitive status on the CSF biomarkers.

Although age was included in the analyses as a continuous variable, for the purposes of graphing the study findings we chose two anchor points to represent Younger Age (age=50 years, black bars) and Older Age (age=80 years, red bars).

The second supplemental analysis was performed to ascertain whether the modification of age-related changes in CSF biomarkers by CR was similarly instantiated in the cognitively-normal and cognitively-impaired groups. To do this, we refit the original regression models while including an age*CR*cognitive status interaction term. A statistically significant age*CR*cognitive status term would indicate that the age*CR interaction was differentially instantiated in each group whereas a nonsignificant term would indicate equivalence of instantiation. For this 3-way interaction term to be properly parameterized, it was necessary to also include CR*cognitive status and age*cognitive status interaction terms in the regression models so as to completely elucidate all 2-way interactions (i.e., age*CR, CR*cognitive status, and age*cognitive status) that are nested within the 3-way interaction (i.e., age*CR*cognitive status).

We observed significant age*CR*cognitive status interactions for p-tau (B(SE)= −1.63(.75), p= .030), t-tau/Aβ42 (B(SE)= −.05(.02), p= .014), and p-tau/Aβ42 (B(SE)= −.01(.002), p=.007). The negative sign of the regression estimates in these findings indicates the attenuation of age-related changes in CSF biomarkers among those with High CR was, comparatively speaking, more pronounced in the cognitively-impaired group relative to the cognitively-normal group. That is, High CR exerted a relatively greater abatement of age-related changes in CSF biomarkers in those who are cognitively impaired compared with the cognitively normal. However, when the 3-way age*CR*cognitive status interactions were formally decomposed into within-group age*CR interactions, they failed to attain statistical significance in either group, with the possible exception of p-tau/Aβ42 which neared significance in the cognitively-impaired group (B(SE)= −.55(.31), p=.081) but not in the cognitively-normal group (B(SE)= −.38(.26), p=.135). Nonetheless, all within-group age*CR interactions were in the same direction, which provided reassurance of their consistency. The failure of the within-group age*CR interactions to reach significance was due to reduced power and/or comparatively restricted range in the distribution of age, education, and the CSF biomarkers within each group.

DISCUSSION

In this study, we investigated whether CR modifies age-related alterations in CSF biomarkers of AD in a cohort of cognitively-normal and cognitive-impaired individuals. We found that, with advancing age, individuals with high CR (i.e., ≥16 years of education) exhibited less adverse alterations in CSF t-tau, p-tau, t-tau/Aβ42, and p-tau/Aβ42 compared with individuals with low CR. Follow-up analyses revealed that this modification of age-biomarker associations by CR was comparatively more pronounced among the cognitively impaired relative to the cognitively normal. To our knowledge, this study is the first to adopt this novel approach—i.e., the attenuation of age's effect on AD biomarkers—to the examination of CR. The present observations are consistent with a recent report from our group showing that increased physical activity ameliorates the impact of age on key AD imaging biomarkers such as amyloid burden, glucose metabolism, and hippocampal volume.³

In their landmark study,¹ Braak and Braak investigated the age-dependent evolution of extracellular amyloid deposits and intraneuronal neurofibrillary changes via assessment of a large set of nonselected autopsy cases using a staging heuristic. They observed that there was appreciable interindividual heterogeneity in the incidence of these hallmark AD brain lesions, such that some individuals showed much less pathology than would be expected for their age, suggesting that neither amyloid deposits nor neurofibrillary tangles necessarily accompany old age.¹ By showing that high CR is related to reduced age-related changes in CSF biomarkers of AD and, by proxy, to less AD-related brain pathology, our present findings suggest an important role for CR as a modifier of the evolution of AD pathophysiological changes.

Studies of CR have traditionally framed the construct in terms of the potential for certain life experience to alter the association between brain pathology and clinical outcome. For example, in a sample composed of cognitively normal and demented persons, Roe et al.¹² investigated whether, given elevated cerebral fibrillar amyloid, individuals who had attained greater years of schooling—their proxy for CR—exhibited preserved cognitive function compared with individuals with fewer years of schooling. They found that, across all cognitive outcomes examined, the negative effect of increased fibrillar amyloid on cognition was diminished among those with higher educational attainment. In a similarly designed study, Rentz et al.²² examined whether CR, as measured by the American National Adult Reading Test,²³ modifies associations between amyloid deposition and impaired cognition in a pooled sample of demented and nondemented individuals. They found evidence of such effect modification. Specifically, greater amyloid deposition was only associated with worse cognition at lower CR levels. At higher CR levels, this association was nonexistent. In a large, multi-ethnic study, Yaffe and colleagues²⁴ similarly found that the relationship between plasma Aβ42/40 and 9-year decline in Modified MMSE scores was pronounced in those with low educational attainment but attenuated in those with higher educational attainment. There is also evidence that higher educational attainment likewise diminishes the effect of cerebral atrophy and ischemic damage on cognitive decline.^{25, 26}

Other studies^{13, 27–31} have investigated the beneficial effect of CR from a different perspective. Rather than model the potential for CR to modify the association between pathology and cognition, they instead present evidence that, for a given level of cognitive function, individuals with higher CR harbor greater brain pathology and are, therefore, further along the disease continuum than those with lower CR. For example, Kemppainen et al.¹³ found that, when matched for degree of cognitive impairment and other relevant covariates, mild AD patients with higher educational attainment had greater fibrillar amyloid and lower glucose metabolic rate compared with those who had fewer years of schooling. In a prospective study, Rolstad and colleagues³¹ found that MCI patients with higher education had lower Aβ42 at baseline, and experienced greater drop in Aβ42 longitudinally, compared to those with less education. In contrast with these foregoing approaches to investigating CR, we present novel evidence that greater CR might attenuate the prototypical age-dependent dynamics of AD biomarkers in CSF. Differences in approach notwithstanding, the reports from these studies and ours provide convergent support for the CR hypothesis, and underscore the putative role that certain life exposures might play in forestalling or retarding the AD cascade.

Even so, it bears noting that our observation that CR modifies age-related changes in CSF biomarkers was made for t-tau and p-tau but not for Aβ42, the canonical biomarker of AD pathophysiology.³² It is not entirely clear why this is so. Perhaps, it suggests that CR exerts a stronger influence on neurodegeneration compared with amyloidosis. Consistent with this hypothesis, other studies have found that whereas CR interacted with t-tau and p-tau to predict incident cognitive impairment in preclinical AD, such an interaction was not observed for Aβ42.^{33, 34} One potential caveat to this interpretation is that, in our study, although CR did not modify the association between age and Aβ42, it modified the association between age and both t-tau/Aβ42 and p-tau/Aβ42; and, compared with isolated amyloidosis, coexistent abnormality in t-tau or p-tau and Aβ42 is generally considered to be more prototypic of an AD syndrome across the disease spectrum.^5–9

We observed a significant age*CR*cognitive status interaction, indicating that CR's modification of the relationship between age and the CSF biomarkers is more pronounced in the cognitively-impaired group compared with the cognitively-normal group. This is an interesting finding and, although its full meaning might be open to question, appears consistent with reports in related areas of inquiry suggesting that the protective effect of various life exposures vis-à-vis AD risk is more marked among individuals with greater vulnerability. For example, recent data from a large population-based study revealed that the decreased risk of dementia as a function of higher educational attainment was more striking among those who were APOE4 positive.³⁵ Similarly, engagement in physical activity during midlife has been shown to be differentially associated with a reduced risk of dementia among APOE4 carriers.³⁶

An important limitation of our study is its cross-sectional nature. Although we have estimated how CR might modify age-related alterations in CSF biomarkers using statistical approaches, prospective designs would be better suited for an in-depth examination of this scientific question. We also acknowledge that, although education is the most commonly employed proxy for CR, education is arguably a multifactorial variable with wide ranging interrelationships with a variety of factors. For example, lower education is associated with an array of cardiovascular morbidity including obesity,³⁷ heart disease,³⁸ and stroke,^{38, 39} each of which is associated with increased risk of dementia in general and AD in particular.⁴⁰ Therefore, it is possible that education may reduce the risk of AD through mechanisms that are not directly related to CR. Relatedly, education perforce tracks closely with other indices of socioeconomic status, such as income and occupational attainment, that have been shown to be related to risk of AD.⁴ However, it has been shown that the association between education and risk of AD persists even after adjusting for socioeconomic indicators, medical comorbidities, intelligence, and various lifestyle factors.^{41, 42}

In summary, this study demonstrates that high CR blunts the characteristic age-related dynamic changes in CSF biomarkers of AD, suggesting a pathway through which CR might favorably alter lifetime risk for symptomatic AD. This finding takes on considerable import when placed against the backdrop of ongoing global efforts to thwart a looming AD epidemic that is driven by the rapid expansion in the elderly segment of the world's population.⁴³ Well-designed longitudinal studies will be the logical next step for more rigorously testing these exciting hypotheses.