Relation of cognitive reserve and task performance to expression of regional covariance networks in an event-related fMRI study of nonverbal memory☆

Abstract

Cognitive reserve (CR) has been established as a mechanism that can explain individual differences in the clinical manifestation of neural changes associated with aging or neurodegenerative diseases. CR may represent individual differences in how tasks are processed (i.e., differences in the component processes), or in the underlying neural circuitry (of the component processes). CR may be a function of innate differences or differential life experiences. To investigate to what extent CR can account for individual differences in brain activation and task performance, we used fMRI to image healthy young individuals while performing a nonverbal memory task. We used IQ estimates as a proxy for CR. During both study and test phase of the task, we identified regional covariance patterns whose change in subject expression across two task conditions correlated with performance and CR. Common brain regions in both activation patterns were suggestive of a brain network previously found to underlie overt and covert shifts of spatial attention. After partialing out the influence of task performance variables, this network still showed an association with the CR, i.e., there were reserve-related physiological differences that presumably would persist were there no subject differences in task performance. This suggests that this network may represent a neural correlate of CR.

Introduction

Several studies have suggested that differential susceptibility to age-related memory changes and dementia is related to variables such as education and IQ Elias et al., 2000, Qiu et al., 2001, Scarmeas et al., 2001, Schmand et al., 1997. These studies provide epidemiological evidence for the presence of cognitive reserve (CR), where subjects with greater reserve may show less severe effects of the aging process. Cognitive reserve may represent individual differences in how tasks are processed (i.e., differences in the component processes), or in the underlying neural circuitry (of the component processes). These differences may result from innate differences in intelligence or they may be acquired through life events such as educational or occupational experience. Different levels of reserve would result in some subjects being less susceptible to the effects of aging or pathology than others Stern, 2002, Stern, et al., 2003.

In this scheme, CR would be equally present in similarly aged individuals who are unaffected by any disease pathology. In contrast to the epidemiologic literature, recent cognitive neuroimaging research dealing with the aging process (Cabeza et al., 1997a,Cabeza et al., 1997b, Reuter-Lorenz, 2002, Stern et al., 2000 highlighted many age-related differences in brain activation, but did not explicitly relate these findings to CR. To our knowledge there is only one study so far (Scarmeas et al., 2003) that could associate age-related differences between young and elderly with CR.

Prompted by the finding of a relation between CR and age-related activation differences between age groups, we asked whether CR might also affect differences in activation within one age group and potentially account for variability that is routinely observed during performance of cognitive neuroimaging tasks. Focusing on one age group should keep the cumulative CR aspect of life experience constant, while allowing an investigation of the innate component of CR that is already present at low age. In particular, young subjects would be expected to show differences in neural activation during task processing that might be related to differences in CR measures such as IQ or education. Rather than focusing on differences in functional connectivity across groups Cabeza et al., 1997a, Cabeza et al., 2002, Grady et al., 1999, we set out to map out commonalties in functional connectivity within a group of health young adults. We predicted that the level of expression of the same activation pattern would be related to the level of CR in each subject.

We used fMRI to examine healthy young adults during performance of a nonverbal serial recognition test. There were two task conditions. The low demand (L) condition required encoding and recognition of single items. The titrated demand (T) condition required the subject to encode a larger list of items. Prior to scanning, this study list size (SLS) was adjusted for each subject, such that recognition accuracy was 75%. This procedure was intended to control for task difficulty, matching difficulty across subjects. Our intention was to explore how individual differences in measured CR are related to changes in expression of a reserve- and task-related brain network from the low to the titrated demand task.

We have previously reported univariate General Linear Model (GLM) analyses (Stern et al., 2003) that demonstrated a correlation between IQ and task-related activation differences between low and titrated demand conditions. In contrast, the regional covariance analysis employed in this paper first aimed to identify a network that underlies task performance in the experiment. We then determined whether individual expression of this network differed as a function of our reserve variables. Therefore, our method offers a more parsimonious account than the different voxels identified on the sole basis of the correlation of their activation with CR variables in the univariate analysis. Specifically, a task-related network whose expression also correlates with CR would provide a more convincing demonstration of having isolated a true neural substrate of CR via manipulation through an experimental design. The type-I error of erroneously detecting such a confluence of factors should be smaller than in univariate analyses where voxels of CR-related activation might bear no relationship to the task performed in the cognitive experiment.

Because univariate and multivariate analyses are truly complementary and focus on different signatures of the BOLD signal, it is difficult to judge whether they give congruent or incongruent results on the basis of statistically significant voxels identified in both analyses. For a scenario of very good experimental control it might be conceivable that a covariance pattern can be isolated whose highest voxel weights also exhibit significant group differences in a univariate analysis. However, the converse seems to be true more often: when several cognitive processes are simultaneously operational, only one of which is successfully manipulated by the experimental design, univariate and multivariate analyses might identify different brain regions. This is understandable given their different premises.