Dynamic Causal Modeling of the Relationship between Cognition and Theta–alpha Oscillations in Adults with Down Syndrome

Figures

• 

  Figure 1.. Assessing construct validity of the parametric empirical Bayesian (PEB) approach. (A) A baseline model parameterization is combined with a between-subject random effects regressor by allowing modulation of specific subsets of parameter (here, intrinsic self-inhibitory connections) by the weight given in the between-subject effect regressor. (B) This produces 36 different DCM models, that differ in their intrinsic connectivity in a way that is affected by the between-subject regressor. (C) These models are used to generate synthetic cross-spectral densities (of which the first three eigenmodes are shown here) that vary systematically with the regressor. (D) These data are used as input for DCM model inversion. This approach compares subsets of models where between-subject effects are expressed only in a subset of parameters. (E) Using the free-energy measure of model likelihood for each of these models, we can then compare how good an explanation these models offer for the data, and successfully identify the model with changes in intrinsic self-inhibition as the winning model.

• 

  Figure 2.. A. Mean scalp map of power distribution for theta (5 Hz, top) and low-alpha (8 Hz, bottom) activity. The frequencies shown here are exemplary and chosen for their complementary distribution across the scalp (all statistical inference is made across the whole theta–alpha spectrum). B. Linear relationship between peak alpha amplitude at frontal and occipital regions (eight channel average frontal; seven channel average occipital) and raw KBIT-2 score. C. Mean regional power spectra at frontal and occipital regions (regional channel average). D. T-statistic scalp map showing voxels with a significant relationship between raw KBIT-2 score and power in theta–alpha (4–13 Hz) range; voxels were located over left occipital (a) and right temporal (b) scalp regions.

• 

  Figure 3.. A. Schematic illustrating the canonical microcircuit (CMC) neural mass model, which comprises four distinct neural-masses mapped onto specific cortical layers. The model includes between-population (local, intrinsic) excitatory (black) and inhibitory (red) connections and within-population intrinsic self-inhibition; as well as between-source (long-range, extrinsic) excitatory connections. B. Diagram showing the bilateral network model, including nodes within the primary visual cortex (V1), superior parietal lobule (SPL), and middle frontal gyrus (MFG), also indicating extrinsic forward, backward, and lateral connections; as well as intrinsic self-inhibition. C. Individual subject model fits showing the top three eigenmodes decomposed from whole-scalp data alongside the model fits for each of the 36 participants.

• 

  Figure 4.. A. Diagram showing the seven network-level candidate models, which differed in terms of which forward (F), backward (B), and intrinsic self-inhibitory (i) connections were affected by between-subject differences in cognitive ability and age. Red box denotes the (intrinsic self-inhibition) model with the greatest model evidence B. (Log) model evidence (negative free-energy) for each of the seven candidate models. Model 0i (model 4; intrinsic self-inhibition alone) shows the highest negative free-energy, with a negative free-energy difference to the next highest model of 9.1 (top diagram) and a posterior probability of ~1 (bottom diagram).

• 

  Figure 5.. A. Bar chart showing the linear (between-subject) effect of raw KBIT-2 score on intrinsic self-inhibition for each node in the network of the winning model (allowing for changes in intrinsic self-inhibition alone; shown bottom right). Bayesian confidence intervals indicate that the effects estimated with most certainty are located at right V1. B. Scatter graph showing inverse correlation between raw KBIT-2 score and mean V1 (left V1 and right V1 average) intrinsic inhibition of the winning model (i.e., the model with the highest model evidence in model space).