Computational modeling of network activity in the cerebellar cortex

Date: 1 April 2016

Venue: UAntwerp, Campus Drie Eiken, T.642 - Universiteitsplein 1 - 2610 Wilrijk (Antwerp)

Time: 4:00 PM - 6:00 PM

PhD candidate: Shyam Kumar Sudhakar

Principal investigator: Erik De Schutter

Short description: PhD defence Shyam Kumar Sudhakar - Faculty of Pharmaceutical, Biomedical and Veterinary Sciences, Department of Biomedical Cciences


Biologically realistic, large-scale modeling of neural networks has emerged as a popular tool to simulate population-level activity across brain regions. It is a powerful technique that should result in better understanding of system-level computations, and how such computations arise from cellular and network mechanisms. Large-scale network modeling could also help to reveal functions of specific components of biological networks and complex input-output transformations of the network. Thalamo-cortical and hippocampal network models have already resulted in better understanding of the underlying systems and have led to novel physiological predictions. Modeling such detailed networks requires enormous amounts of information regarding the components and connections of the underlying biological system.

The cerebellum is an ideal candidate for large-scale modeling for many reasons. First, though known to be involved in a variety of functions ranging from locomotion to cognition, its computations are not understood completely. Next, the cortex of the cerebellum is highly compartmentalized and characterized by uniform cyto-architecture with same types of neurons throughout the folia. Moreover, the membrane biophysical properties of most neurons in the cerebellar network have been extensively studied in vitro. Other information, such as cell numbers and their ratio, density of individual neurons, convergence and divergence of network connectivity, and dendritic and axonal distribution patterns of component neurons, has been characterized through decades of seminal research work.  In this thesis, I use modeling as a tool to study information processing in the cerebellar cortex network and to answer specific questions related to it.

I first modeled a network of the molecular layer of the cerebellar cortex, consisting of basket and stellate neurons, popularly known as interneurons. With this model, I have shown how gap junctions between interneurons spatially link them and broaden the spatial convergence of inhibitory input onto Purkinje neurons. In addition to that, I quantified the convergence between interneurons and Purkinje neurons. I discovered that each Purkinje neuron receives inhibition from an average of 4.5 molecular layer interneurons, which increases to 12 in the presence of gap junctions. In addition to emphasizing the role of electrical coupling, this research also demonstrates how modeling can be used to replicate experimental paradigms and to answer questions that are difficult to address experimentally.

Using a previously developed 2D network model of the granular layer of the cerebellar cortex, I modeled the effect of alcohol on granule neuron IPSPs. The model consists of Golgi and granule neurons with connections between them. Additionally, the network also included gap junctions between Golgi neurons. Experimental work has shown that alcohol increases the excitability of Golgi neurons by blocking Na+/K+-ATPase. In the model, in the absence of mossy fiber input, inhibition of Na+/K+-ATPase by a drug called ‘ouabain’ increased the frequency of granule neuron IPSPs without having any effect on their amplitude.

The granular layer of the cerebellar cortex is characterized by excitatory granule neurons and inhibitory Golgi neurons. The mossy fibers that provide input to the granular layer exhibit a variety of firing patterns. I created a biologically realistic 3D network model of the granular layer of the cerebellar cortex to study its oscillatory properties and network activity patterns in response to physiological mossy fiber input patterns. The network activity pattern is characterized by tunable, synchronous oscillations, oscillation cycle skipping by Golgi neurons, and center-surround activity patterns in the firing of granule neurons. Further I calculated the correlation between neurons in the network along the transverse and sagittal axes for different network configurations and also studied the role of NMDA receptors on network dynamics. This modeling work is significant because it employs physiological mossy fiber input patterns with spatial and temporal structure, unlike previous network models of the cerebellar cortex. In addition, the research explains population level mechanisms that promote sparse activity in granule neurons and the role of individual components (e.g., NMDA receptors, electrical coupling) in network dynamics.

Rate coding and time coding are two important coding strategies by which neurons transfer information to one another. Neurons of the cerebellar nuclei are important as they transmit the final output of cerebellar information processing to various parts of the brain. Whether neurons of cerebellar nuclei use rate coding or time coding to send output to various downstream pathways has been a topic of intense debate. Decades of work suggest primarily rate coding, but a recently discovered time coding mechanism bolsters the involvement of the cerebellum in timing functions. Recent work has shown that their presynaptic partners, the Purkinje neurons, signal multiplexed code with pauses synchronized among neighboring neurons providing a time code and regular spikes acting as a pure rate code. Therefore it is important to analyze the coding strategies of cerebellar nuclei neurons with respect to the temporal structure of presynaptic Purkinje neuron spike trains. In this thesis, I address this issue, and I have focused on the effect of synchronized pauses on the coding strategies of cerebellar nuclei neurons. My results indicate that Purkinje neuron synchrony is mainly represented in the form of rate changes in downstream nuclei neurons. Additionally, pause beginning synchronization produced a unique effect on coding strategies of nuclei neurons, while the effects of pause ending and pause overlapping synchronization could not be distinguished.

I conclude that large-scale network modeling based on accurate description of single neuron models and anatomical patterns of connectivity help to better understand population-level mechanisms and to derive predictions that can be tested experimentally. The modeling techniques that I have employed in my thesis are novel and can be used to understand system-level mechanisms across all brain regions. Based on this research I propose extensive use of computer modeling and large-scale network simulations to understand and characterize neurophysiological mechanisms.