(252h) Modularity Analysis of Single-Cell Resolved Neural Activity Recordings: Bridging the Gap between Molecular, Cellular, and Brain-Wide Changes in Biological Neural Networks

Biological neural networks are complex, multi-scale systems that rely on coordination of neural activity across multiple scales. To investigate the molecular and cellular mechanisms underlying this multi-scale coordination, technological advances have increased both the scale and resolution of neural activity datasets. Notably, single-cell resolved whole-brain imaging has been demonstrated in both the nematode roundworm Caenorhabditis elegans and zebrafish (Danio rerio). While whole-brain imaging provides valuable insights into brain-wide dynamics and changes in network structure during stimulus processing and behavior, fully leveraging the single-cell resolution of these whole-brain imaging datasets to link observed brain-wide changes to underlying molecular and cellular mechanisms has remained challenging.

To bridge the gap between brain-wide changes, individual neurons, and pairwise correlations, neurons can be clustered into modules based on similarities in their activities. However, it can be challenging to evaluate how accurately clusters from clustering algorithms describe the true structure of the network. Applying network modularity, a concept popular in network analysis for community detection, can help address this challenge. Network modularity not only enables clustering of network nodes (neurons) into modules (groups of neurons) that interact more strongly with each other than the rest of the network, but also scores how well a given clustering describes the true structure of the network. We apply network modularity analysis to both single-cell resolved whole-brain (~100 neurons) and multi-cell (20-30 neurons) neural activity recordings from C. elegans. These methods allow us to not only measure dynamic changes in brain-wide state through changes in network structure, but also investigate the molecular and cellular mechanisms driving them.

We first analyze network modularity in multi-cell neural activity recordings, where activity is only recorded from neurons expressing GLR-1 (a synaptic receptor subunit for the neurotransmitter glutamate). We find that food odor cues decrease the modularity of functional connectivity across this network of neurons expressing GLR-1. This decrease in modularity during food sensing suggests that food odors synchronize activity across the GLR-1 network, making it more appropriate to consider this network all-to-all connected in this context, rather than as discrete modules. To determine whether excitatory or inhibitory connections, we implement a modification of modularity that accounts for signs of correlations between neurons, not just their strength. We find that correlation strength alone is enough to observe decreases in network modularity during food sensing, but significantly stronger decreases are observed when also accounting for correlation sign.

In whole-brain neural activity recordings, we find fluctuations in network modularity during spontaneous neural activity. These fluctuations are driven more strongly by changes in strength of correlations between neurons, rather than changes in correlation sign. Despite these fluctuations in network modularity during spontaneous activity, we find that whole-brain functional connectivity remains highly modular, relative to a null model of shuffled functional connectivity matrices. Additionally, the whole-brain dataset contains recordings where the activities of key hub interneurons are silenced. When some of these hub interneurons, such as RIM (a C. elegans hub interneuron involved in regulating transitions between forward and reverse locomotion), are silenced, we find that the timescale and magnitude of fluctuations in network modularity changes. This suggests that these key hub interneurons may play a role in stabilizing the modular structure and organization of neural activity during spontaneous, brain-wide dynamics.

By analyzing network modularity in large-scale, single-cell resolved neural activity datasets, we demonstrate the utility of these methods in linking brain-wide changes in network structure to molecular and cellular mechanisms. Developing these methods to more fully leverage the single-cell resolution of whole-brain imaging datasets meets a broader need driven by both the complex, multi-scale nature of biological neural networks and technological advances that have increased the prevalence, scale, and resolution of large-scale neural activity recordings from these systems.