2.6. Investigating effects of age

We investigated age-related effects on task potency based on the underlying idea that task connectivity modulations that are in common between tasks reflect underlying common mechanisms. Accordingly, we investigated age-related effects on the potency of single edges as well as on an average potency across subgroups of edges. For both analyses, we used least square fitting to investigate the linear change with age, thereby maximising the detection of maturational processes while minimizing the complexity of the model. We applied correction for multiple comparisons across the tested subgroups of edges by implementing FDR correction (q < 0.05).

The subgroups of edges we used were, for each task, 1) edges modulated by this task only, 2) edges modulated by this task and one of the two other tasks, 3) edges modulated by all three tasks. See the Venn-diagram in Fig. 3 for an overview of potential edge subgroups. We propose that similar changes with age will be observed across tasks in connections that they co-modulate. For example, if task potency in one task increases with age for an edge modulated by more than one task, we would expect to observe a similar increase with age in all other tasks modulating this edge.

Effect of age on edges modulated by each task. Each graph illustrates the effect of age for corresponding edges indicated in the Venn-diagram. A: edges modulated during WM; B: edges modulated during STOP; C: edges modulated during REWARD; D: edges modulated by WM and STOP; E: edges modulated by all three tasks. All displayed effects, except for C, reached statistical significance at p < 0.05 after FDR correction. The age effect is calculated by linear regression of age against the average potency over the specified subset of edges. The average potency decreases significantly with age for edges selected in STOP and WM, in all tasks for edges shared by WM and STOP, and in STOP for edges shared by all three tasks.

The average potency across edges within each of the edge subgroups specified above reflects an average underlying mechanism, but potentially obscures effects that play at the single edge level. To gain insight into age effects at the level of single edges we compared the slope of the linear relationship between age and potency for each edge within the task-modulation fingerprint of two tasks. Specifically, we plot the slope of each edge in one task against the corresponding slope of that edge in the other task. We then fit an ellipsoid on the resulting scatter plot using least square fitting to quantify the relationship between the two displayed tasks. If the ellipsoid stretched around the x = y diagonal axis, it indicates a strong relationship between the two parameters, which in our case translates into the observation that connectivity modulation would mature similarly in both tasks. We conducted this analysis independently in edges shared by the two tasks or selected in only one task. To quantify the strength of the relationship we calculated the width/height ratio of the ellipsoid fit. The closer this ratio is to 1, the rounder the ellipsoid, and the weaker the relationship between the two tasks.

Finally, at the single edge level, we tested for second order changes with age, i.e. we tested whether the speed of the maturational changes varied as a function of age. We assumed that age effects would be stronger in younger than in older participants. To this end, we modelled a linear change over a short age window of 1 year including 7 participants from this window. When more than 7 participants were available within an age window we randomly selected 7. We moved this window across our entire population, each time removing the youngest subject of the window and considering a 1 year age span starting from the age of the subject immediately following in age. We extracted the absolute beta value of the linear regression for each window as a marker for the speed of change with age of the task-potency.