Hierarchical clustering was used to delineate modules of co-active clusters in each correlation matrix. The correlation coefficient between each cluster pair was converted to the complete Euclidean distance for clustering. The derived dendrograms were trimmed at different tree heights (between 30–100% of the maximal tree height), to compare the total number of modules across hierarchical levels. A reliable difference in modularity should be consistent and robust between groups across different tree-cutting thresholds. We chose to highlight modularity results cut at 60% of the maximum tree height, which is near the elbow of the height by modularity curve. Hierarchical clustering was conducted using the built-in hclust function in the R stats package and was visualized using the heatmap.2 function in gplots (version 3.1.3).
Inspired by work from Kimbrough et al. (Kimbrough et al., 2020 (link), 2021 (link)), looking at inter- and intra-module characteristics of between-region co-activity as defined by c-Fos+ cell counts, we used the same graph theory approach to examine psilocybin-induced changes in regional centrality metrics, including the Z-scored version of within-module degree (WMDz) and the participation coefficient (PC). The WMDz represents the relative importance of a cluster within its own module, which corresponds to intra-module connectivity. The PC measures the extent a cluster correlates with multiple modules, which corresponds to the inter-module connectivity (Guimerà & Nunes Amaral, 2005 (link)). As recommended elsewhere (Kimbrough et al., 2020 (link), 2021 (link)) we thresholded the co-activity correlograms by removing co-activity edges weaker than r = 0.75 and excluded negative co-activity edges following conventions as no consensus was reached regarding how to handle them in graph theory analysis (Hallquist & Hillary, 2018 (link)). Graph theory analysis was conducted using the Brain Connectivity Toolbox (Rubinov & Sporns, 2010 (link)) in Matlab R2020b.