Neural similarity across task load relates to cognitive reserve and brain maintenance measures on the Letter Sternberg task: a longitudinal study

Abstract

The aging process is characterized by change across several measures that index cognitive status and brain integrity. In the present study, 54 cognitively-healthy younger and older adults, were analyzed, longitudinally, on a verbal working memory task to investigate the effect of brain maintenance (i.e., cortical thickness) and cognitive reserve (i.e., NART IQ as proxy) factors on a derived measure of neural efficiency. Participants were scanned using fMRI while presented with the Letter Sternberg task, a verbal working memory task consisting of encoding, maintenance and retrieval phases, where cognitive load is manipulated by varying the number of presented items (i.e., between one and six letters). Via correlation analysis, we looked at region-level and whole-brain relationships between load levels within each phase and then computed a global task measure, what we term phase specificity, to analyze how similar neural responses were across load levels within each phase compared to between each phase. We found that longitudinal change in phase specificity was positively related to longitudinal change in cortical thickness, at both the whole-brain and regional level. Additionally, baseline NART IQ was positively related to longitudinal change in phase specificity over time. Furthermore, we found a longitudinal effect of sex on change in phase specificity, such that females displayed higher phase specificity over time. Cross-sectional findings aligned with longitudinal findings, with the notable exception of behavioral performance being positively linked to phase specificity cross-sectionally at baseline. Taken together, our findings suggest that phase specificity positively relates to brain maintenance and reserve factors and should be better investigated as a measure of neural efficiency.

Keywords: Brain maintenance; Cognitive reserve; Longitudinal; Within-subjects fMRI; Working memory.

Figure 1.. Schematic of LS task presentation.

In the stimulus presentation phase (3 seconds), participants were presented with a string of letters of either 1, 3, or 6 letters in uppercase. They were asked to retain this sequence in memory during the retention phase (7 seconds). Finally, during the probe phase, they were presented with a single letter in lowercase and asked to indicate whether that letter was present during stimulus presentation. Responses were made by button press. Trials were interleaved with a 3-second intertrial interval (ITI).

Figure 2.. Behavioral performance plots of participant-level LISAS per task load and mean accuracy across loads, divided by time point.

In the upper panel, from left to right, the linear integrated speed-accuracy score (LISAS), which provides a corrected measurement of RT in milliseconds, in addition to mean accuracy (i.e., mACC) across task loads, is plotted for baseline, follow-up, and the difference between them, respectively. Within each graph, LISAS is plotted for load 1 (L1), load 3 (L3), and load 6 (L6), with values depicted on the left y-axis, and the mean accuracy across loads, with values being depicted on the right y-axis. Green lines represent younger participants while magenta lines represent older participants. For mean accuracy, thick horizontal bars represent the mean of each group and the color-corresponding boxes represent their standard error of the mean (SEM).

Figure 3.. Correlation matrices at each time point for all task pairings.

Fisher’s Z correlation matrices of all 36 task pairings, where the correlation is computed across all voxels and averaged within subjects. Y-labels denote the load level (L1= Load 1, L3= Load 3, L6 = Load 6) and X-labels denote the corresponding phase. Black square boxes indicate within-phase correlations between task loads. Centered values within each square indicate the average within-phase correlation not considering a task’s perfect correlation with itself. Left plot denotes the correlations at baseline (BL), the middle plot at follow-up (FU), and the right plot, the difference between the two (follow-up – baseline).

Figure 4.. Results from the region-level regression analyses.

In the first panel are the results from the baseline regression, in the second panel for 5-year follow-up, and in the third panel, the difference between the two time points. Variables that significantly predicted phase specificity (PS) at p < 0.005 are color-coded at the bottom of the figure. N*CT denotes the interaction term of NART * Cortical Thickness. NOTE: All regions displayed bear positive relationship between PS and the respective factor, except LISAS, which displays a negative relationship.

Figure 5.. Plot of interaction effect in the left precuneus between CT and NART IQ at baseline (BL).

Phase specificity (PS) is plotted on the y-axis, cortical thickness (CT) on the x-axis, and data is median split to show the moderation effect of NART on the relationship between PS and CT. CT has been mean-centered. Colored error ribbons reflect the 95% confidence band.