We report that amyloid deposition is associated with reduced cognitive performance among clinically normal individuals (CDR = 0), and that the relationship is systematically weaker in subjects with higher CR. Clinically normal individuals with higher CR have less susceptibility to amyloid-related impairment than those with lower CR. We confirmed the interaction with CR across a range of cognitive performance including patients with mild AD,35, 36
but found that the subset of subjects above the threshold for amyloid positive seem to derive less benefit from CR, as the main effect of declining performance with increasing amyloid overwhelmed the weaker, moderating effect of CR. Our findings are consistent with epidemiological data suggesting that higher education and occupational attainment is associated with decreased risk for AD,32
with postmortem data relating AD pathology to cognitive function during life in clinically normal individuals,30
and with CR modification for amyloid plaque-related cognitive function, but not for tangle-related function.28
It is not known whether deposited amyloid, soluble forms of β-amyloid, or other associated pathologies are directly responsible for impairment or the effect of CR. However, our findings strengthen existing evidence that brain amyloid burden measured with PET correlates with level of cognitive impairment.7, 16, 20, 64
Similarly, although high CR permits individuals to tolerate encroaching pathology, it is not known whether this comes about because of higher synaptic or neural capacity or greater efficiency in cognitive strategies or network engagement that may have neuroprotective effects related to CR. Conversely, those with lower CR may have been exposed to developmental circumstances that prevent the achievement of higher CR, thus producing increased vulnerability to amyloid's neurotoxic effects. It is also possible but unknown whether the transition from soluble to deposited amyloid itself provides a form of reserve by sequestering putatively more toxic forms of β-amyloid.65, 66
What does seem clear, however, is that attempts to relate amyloid PET to NP performance or to treatment-related changes in NP performance should be interpreted against the background of each subject's level of CR.
The relationship of NP performance to amyloid burden was much less obvious in subjects with high levels of CR when the NP canonical variate was used. Because we suspected that ceiling effects could obscure the relationship, we considered the possibility that more challenging test instruments could improve our ability to detect interaction effects at higher levels of CR. We found in an independent test session that the use of the MCT, an episodic memory test with no evidence of ceiling effects in our sample, permitted us to discover not only the interaction of amyloid with CR, but also the main effect of reduced performance with higher levels of amyloid.
High amyloid burden in the precuneus was related in the full sample of subjects to poorer performance across multiple domains of NP function, including tests of working memory (Digit Span Backward), episodic memory (FRsrt and FCsrt), semantic processing (CAT), naming (BNT), and visuospatial perception (VFDT). In contrast, tests that measured speed of processing (eg, Trails A) and letter fluency (ie, FAS) were not associated with PiB retention in the precuneus but, similar to the findings of Bennett et al in analyses of postmortem data,29
did have a significant interaction with CR (see ). CR could have a stronger effect on particular cognitive tests but not others, because some tests may permit individuals to use a broader range of alternate cognitive strategies that are more accessible to those with higher CR.67, 68
For example, some tasks, such as Trails A and FAS, require more primitive sequencing and phonemic search strategies that may still be accessible to those with higher CR, allowing for the successful completion of the task even in the context of extensive amyloid deposition. The clinical literature has widely reported that letter fluency tasks, such as FAS, are much better preserved in patients with AD.45, 69, 70
Our examination of specific ROIs indicated that PiB retention in the frontal region, which has been reported to bear a substantial amyloid burden,7, 10, 13, 19
was related to NP performance, but there was no interaction with CR. This finding suggests that, compared with precuneus, frontal amyloid deposition may be linked to impairment and disease severity, but the protective effect of CR is not as readily observable. This finding differs from Kemppainen et al,37
who found that highly educated AD patients showed significant increases in PiB uptake in the frontal region compared with less educated AD patients, perhaps because their sample included only impaired subjects. Our finding that CR did not attenuate amyloid-related performance in frontal regions may relate to reports that frontal PiB retention was not as strongly related to atrophy or gray matter loss20, 58
or to fluorodeoxyglucose hypometabolism.7, 8, 15
Lower strata of CR were underrepresented in our sample of NCS (mean IQ = 121.9), and all NCS in the lowest IQ quartile had lower levels of PiB retention (see C). One possible explanation for this is consistent with our hypothesis that individuals in the lowest stratum of CR would have greater levels of impairment as PiB retention increases than those in higher strata. Therefore, lower CR subjects with higher levels of amyloid would be less likely to be classified as clinically normal. To evaluate this possibility, we are currently recruiting subjects specifically targeting the lower strata of CR. If our observations are extrapolated to these individuals, however, they would be the most clinically impaired and least able to participate.
Although we chose to use the AMNART IQ as a proxy of reserve, we realize it has limitations. For example, AMNART IQ measures only 1 aspect of CR, namely verbal ability level, but there are other factors that may contribute to CR, such as early educational experiences, late-life cognitive activities, life-style factors, occupation, and socioeconomic status.28, 71
We also recognize that the AMNART IQ may not be an accurate measure of premorbid ability for all individuals, particularly those with reading difficulties or non–English-speaking participants. Although all subjects in our sample were English speaking, and no one had a history of learning disabilities, this limits the use of AMNART IQ for all populations. Finally, education has been used successfully in several other PiB studies as a proxy of reserve,22, 36
but we found that in our older population, a bias occurred where women did not have the same educational advantages as men. In the past, we found a similar bias directly comparing education and AMNART IQ in older individuals.72
When education was used as an interaction variable with precuneus PiB, we also found a significant main and interaction effect with CR, as others have reported.36, 37
However, when the interaction of rAMNART IQ with CR was added to the model, the interaction with education became nonsignificant. Possibly, AMNART IQ may be closer to the underlying operative of the CR concept, and education may be sharing variance as a confounded correlate. In the future, it will be important to explore further the many dimensions of CR to more closely approximate this concept.
Another limitation of this report is the lack of apolipoprotein E (APOE) genotyping in all subjects. Reiman et al73
have recently reported that fibrillar beta-amyloid is significantly associated with APOE epsilon 4 carrier status in cognitively normal older individuals. It is possible that our findings in the normal control sample may have been attributable to APOE variants. In the entire sample of 83 subjects, 43 had APOE genotyping. We reran the above canonical analysis and included APOE genotyping with 1 or no copies of the APOE epsilon 4 allele as a dummy-coded predictor variable. We found that APOE status was not a significant predictor of the NP canonical variate, whether it was included as an additional predictor in the canonical analysis or merely introduced as an additional covariate in the follow-up regressions. Among the subjects with genotyping, we found a trend (p
= 0.10) in precuneus PiB deposition, with E4 carriers having slightly more amyloid (DVR = 1.35 ± 0.3) than non-E4 carriers (DVR = 1.21 ± 0.2). However, the lack of a significant finding is most likely related to small sample size. Future work will be required to explore the association of genetic factors including APOE genotype to determine if APOE makes an independent contribution to our observed findings of CR modification.