Deciphering the functional structure and dynamics of neural networks

How fixed versus flexible are brain networks? While it has long been known that synaptic connections can be modified by learning, it was assumed that such plasticity takes place within relatively fixed network structures. However, recent studies of organisms ranging from invertebrates to vertebrates are discovering that neural networks can be surprisingly labile on a moment-to-moment timescale. It is likely that deciphering the mechanisms underlying this flexibility may be one of the keys to understanding processes ranging from motor control to cognition.;The biggest hindrance to our ability to study this important topic is technical in nature. We first need to address such questions at the level of a neuronal population. Here fast-voltage sensitive dyes were used to simultaneously record > 10% of the 1600 neurons contained in the pedal ganglion locomotion network of the marine mollusk Aplysia californica. The pedal ganglion is an attractive target for mapping a flexible neural program to the dynamics and structure of its underlying distributed network as it wholly contains the central pattern generator, as well as the motor and neuromodulatory neurons. This mixture of systems means that population imaging of the pedal ganglion is representative of the analytical challenges that will become increasingly common for large-scale recordings of complex neural systems. Indeed, recent efforts such as the BRAIN Initiative have emphasized the urgency with which we need new analytical tools for mining large-scale data sets, pointing out that such tools are crucial to understand the brain.;In order to understand the locomotion network, it was necessary to first identify the components of the system. This thesis presents a new analytical technique for unsupervised community-detection based consensus clustering, which was used to rapidly and reproducibly return the ensembles of spiking neurons present in individual recordings of the motor program. These ensembles or "modules" were then classified and mapped across preparations to create an atlas of the locomotion network. This modular deconstruction allowed the system to be broken down to fundamental types of ensembles, and revealed that the network is both physically and functionally modular, thereby supporting the recently proposed idea that detecting neural modularity will be highly effective in reducing the dimensional complexity of the brain. Harnessing the knowledge that the system is modular, it was then possible to show that the motor program is built from a small number of dynamical building blocks. Dynamical portraits reveal that the network evolves over the execution of a single motor program, slowly settling into a stable state. Underlying this evolution is a moment-to-moment functional reconfiguration within and between ensembles. These phenomena would not have been discovered without the methods developed and presented here. All such methods have been submitted as a free software package for use by the neuroscience community.;In addition to providing analytical tools, this work paves the way to meet the next challenge of determining the mechanism that drives the network evolution and its underlying moment-to-moment functional reconfiguration.