Abstracts

Rationale:
The underlying changes to effective connectivity networks that are critical for transitioning a healthy brain into one that has a propensity to generate seizures are poorly understood. Specifically, the role of hubs, highly connected neurons that are critical for orchestrating synchronization, in the interictal state is incompletely defined as hubs are a sparse cell population and most single cell imaging studies capture activity from only a small window. However, recent technological advances enable the acquisition of whole brain neural activity at single cell resolution in larval zebrafish. We therefore performed effective network inference on whole-brain cellular resolution functional calcium imaging data and discovered that network stability is causally linked to the local network architecture surrounding hub neurons. Collectively, characterizing the organizing principles of seizure networks built from single cells may offer insight towards developing strategic therapies.
Method:
Whole brain cellular-resolution calcium imaging of larval zebrafish pentylenetetrazol (PTZ) model of acute seizures was performed using light-sheet microscopy. Hubs were identified from effective networks that were inferred using a 1:1 chaos-generating recurrent neural network constrained by a structural wiring atlas. Higher-order motifs are crucial for determining network function. Therefore, we applied the novel Motif-based Approximate Personalized PageRank (MAPPR) local clustering algorithm to characterize the higher-order architecture surrounding hub neurons. Cluster integration was quantified using motif conductance, which is proportional to the number of motif instances that project out of the cluster. Similar analysis was performed on 2p calcium images of dentate gyrus granule cells from kainic acid mouse model of chronic temporal lobe epilepsy (TLE).
Results:
Effective networks constrained by zebrafish structural atlas trained successfully and displayed robust partitions resembling major macroscale brain regions. Identified hubs were spatially localized to diencephalon and projected to telencephalon and mesencephalon. Computational simulations reveal that perturbation of individual hubs in preseizure networks had significantly more influence over network dynamics compared to perturbation of individual hubs in baseline networks. Using MAPPR, we discovered that hubs in preseizure brain were surrounded by a dense plexus of unstable feedforward motifs. Hubs and their respective local feedforward clusters, termed higher-order hubs, were significantly more integrated with the rest of the network in preseizure state than in baseline. Disconnecting a small fraction of higher-order hubs with highly integrated local feedforward clusters reduced oscillatory power and rendered the network more resilient to perturbation. Identical results were found in dentate gyrus granule cells of chronically epileptic mice compared to control.
Conclusion:
Network instability in the preseizure brain is attributed to enhanced integration of higher-order hubs and their local feedforward clusters. The causal relationship we identified may provide a path towards a novel network-level biomarker. Furthermore, our findings provide the explicit experimental prediction that perturbation of outgoing hubs in preseizure brains but not control brains exerts strong influence over network dynamics. Lastly, by showing that findings are true for both acute zebrafish and chronically epileptic mouse models, we provide insight towards a general organizing principle of preseizure networks.
Funding:
:Stanford Interdisciplinary Graduate Fellowship w/ Wu Tsai Neurosciences Institute (DH); US NIH Grant NS094668 (IS)