Dependence of FC on power and age
The dependence of FC on age and change in power was further explored with a robust linear regression model applied over all the healthy subjects (refer Methods section) (Figure 6). Alpha, slow gamma, and fast gamma FC was individually regressed with subject age and the change in band power as independent variables, separately for each seed electrode group. Figure 6a shows the scatter plot of alpha, slow gamma, and fast gamma FC versus subject age, while Figure 6b shows the same versus change in power, for the left electrode group (similar results were obtained for other groups; data not shown). The coefficient in the fitted model and its significance (p-value) are also shown in the plots. The p-value relates to the t-statistic of the two-sided hypothesis test\((H:b=0)\), given the other terms in the model. Consistent with previous results, alpha FC depended only on the age factor (\(b_{\text{age}}\) = -0.002; p-value = 0.002) and not on the change in power (\(b_{\text{power}}\) =0.004; p-value = 0.380). This can also be observed in the median alpha FC values binned in 5-year intervals as shown in black traces in Figure 6a, which decreased with age (left plot).
In contrast, regression analysis for slow gamma revealed additional dependencies not observed earlier. In particular, slow gamma FC was found to depend on both, age (\(b_{\text{age}}\) = -0.002; p-value = 0.0003) and change in power (\(b_{\text{power}}\) =-0.026; p-value = 0.0001). Unexpectedly, FC was negatively correlated with change in power (potential reasons are discussed below). The dependence of slow-gamma FC on age can also be observed in the median FC values binned over 5-year intervals (black trace in Figure 6a). This dependence was weaker in our previous analysis because we had partitioned the subjects into the two age groups, middle-aged and elderly. Indeed, when we performed the same regression analysis but after replacing the age with age category (0 for middle aged and 1 for elderly) and excluding the power dependence, the resulting coefficient barely reached significance in the slow gamma band (\(b_{\text{age}}\) = -0.022; p-value = 0.03) even though it remained significant for the alpha band (\(b_{\text{age}}\) = -0.024; p-value = 0.0028). Results for fast gamma FC were largely consistent with previous analyses, showing no dependence on the age factor (\(b_{\text{age}}\) = -0.001; p-value = 0.0816). In addition, there was a weak dependence on the change in power (\(b_{\text{power}}\) =-0.013; p-value = 0.0185), similar to the dependence observed for slow gamma.
The MCI subjects shown in red showed the expected results: these points were scattered across either side of the black traces for alpha and fast-gamma (left and right plots in Figure 6a), but most of the points lay below the black trace for slow-gamma (middle plot).