Subgenual anterior cingulate cortex controls sadness-induced modulations of cognitive and emotional network hubs

The regulation of cognitive and emotional processes is critical for proper executive functions and social behavior, but its specific mechanisms remain unknown. Here, we addressed this issue by studying with functional magnetic resonance imaging the changes in network topology that underlie competitive interactions between emotional and cognitive networks in healthy participants. Our behavioral paradigm contrasted periods with high emotional and cognitive demands by including a sadness provocation task followed by a spatial working memory task. The sharp contrast between successive tasks was designed to enhance the separability of emotional and cognitive networks and reveal areas that regulate the flow of information between them (hubs). By applying graph analysis methods on functional connectivity between 20 regions of interest in 22 participants we identified two main brain network modules, one dorsal and one ventral, and their hub areas: the left dorsolateral prefrontal cortex (dlPFC) and the left medial frontal pole (mFP). These hub areas did not modulate their mutual functional connectivity following sadness but they did so through an interposed area, the subgenual anterior cingulate cortex (sACC). Our results identify dlPFC and mFP as areas regulating interactions between emotional and cognitive networks, and suggest that their modulation by sadness experience is mediated by sACC.

: Selection of the parameter value γ-=0.45 based on the identification of anatomical communities (left) and on the maximization of the modularity Q difference between WM1 and WM2 (right).
Left: Connectivity matrix organized according to the communities identified by the modularity algorithm in the two task conditions (Resting-WM1 and Sadness-WM2, two columns) and for different values of the parameter γ-(rows, see title for each panel). Notice that for γ-< 0.8 the algorithm identifies subcortical areas as belonging to a separate community within the ventral subnetwork.
Right: Difference in Q value obtained in Resting-WM1 relative to Sadness-WM2, for different values of γ-(x-axis). The maximum occurs for γ-=0.45, indicating an optimal condition to explore modularity differences in relation to our task. A: Connectivity matrix organized according to the communities identified by the modularity algorithm in the two task conditions (Neutral-WM1 and Sadness-WM2, two columns) and for different values of the parameter γ-(rows, see title for each panel). Notice that for γ-< 0.8 the algorithm identifies subcortical areas as belonging to a separate community within the ventral subnetwork.
B: Difference in Q value obtained in Neutral-WM1 relative to Sadness-WM2, for different values of γ-(x-axis). The maximum occurs for γ-=0.45, indicating an optimal condition to explore modularity differences in relation to our task. Thick horizontal line marks a significant difference in modularity Q between Neutral-WM1 and Sadness-WM2 (permutation test, p<0.05. Between 0.42 and 0.5, all threshold increments by 0.01 give p<0.05 except for 0.43, p=0.073, and 0.45, p=0.054). Figure S3: Variance across areas of the degree as a function of the threshold used to binarize the connectivity matrix.

A B
The degree is computed as the number of connections for each node, and connections are validated if they exceed a specific fraction of the maximal absolute connection value (threshold). The variance across areas of the degree shows a maximum as a function of threshold for both task paradigms (left and right panels). The maximal variance occurs for a threshold at 35%. This is our principled threshold choice, where we expect to get the best power in our data to identify task-related differences in degree between the areas. C. Significantly greater activation of cognitive areas during delay WM1 compared to delay WM2 (after sadness).

Figure S5: Cognitive and emotional communities for Neutral-WM1
A. 3D-graphical representation of the networks. The ROIs are located according to real-world coordinates. Mean significant correlations across subjects are plotted (positive correlations in red, negative correlations in blue dashed lines). Shaded brain for schematic purposes.

B.
Matrix of the mean correlations across subjects. The graph analysis identified two main modules, the cognitive and emotional communities separated by the dashed black line. Within the emotional community, two sub-communities were found (separated by the red line), corresponding to emotional areas in the cortex and the limbic system (subcortical areas).

C.
Correlation distributions (collapsing all subjects). The correlations between cognitive and emotional modules were mainly negative (inter-modules, plotted in blue, mean ± SEM = -0.14 ± 0.004). The correlations within the cognitive module (intra-cognitive, plotted in orange, mean ± SEM = 0.073 ± 0.005) and the correlations within the emotional module (intra-emotional, plotted in red, mean ± SEM = 0.037 ± 0.008) were both mainly positive.

Figure S6: Parametric analysis of communities for High-sadness group (left) and Low-sadness group (right) during Sadness-WM2 task.
Connectivity matrix organized according to the communities identified by the modularity algorithm in the two subgroups (High-sadness and Low-sadness, two columns) and for different values of the parameter γ-(rows, see title for each panel). Notice that for γ-> 0.6 the algorithm identifies subcortical areas as belonging to an integrated community within the ventral subnetwork only in the High-sadness group.

Figure S7: Hub identification was consistent across thresholds of the correlation matrix
The classification of the hubs identified was stable across thresholds (from 30% to 45%) during Neutral-WM1 (A) and Sadness-WM2 (B). Regions classified as hubs are plotted as triangles. Error bars mark standard error of the mean.