top of page


Lateral Inhibition in the Olfactory Bulb


Lateral inhibition is a key feature of circuits in sensory systems. In the olfactory system, the breadth and organization of lateral inhibition across glomeruli, the functional units of odor coding and processing, is not well understood. Does each glomerulus inhibit all or most others, or are they more selective? If glomeruli inhibit just a few others, are their targets of inhibition random, or is there an underlying logic?


In collaboration with the Wachowiak Lab, we constructed a model of interglomerular inhibition, grounded in anatomical data, whose neural activity is driven by previously-published datasets of sensory inputs. We found that the canonical global inhibition model does not produce responses to stimuli that match experimental data and that selective inhibitory connections can be tuned (based on input statistics) to enhance decorrelation of similar stimuli.

D. Zavitz, I. Youngstrom, A. Borisyuk, M. Wachowiak. "Effect of Interglomerular Inhibitory Networks on Olfactory Bulb Odor Representations". Journal of Neuroscience, 40 (31).

Effect of Connectivity Structures on High Dimensional Odor Representations


The mushroom body is a learning and memory center in the drosophila brain that receives input from the olfactory system’s antennal lobe.  The antennal lobe, much like the olfactory bulb, is comprised of functional units called glomeruli. Each glomerulus is associated with a set of projection neurons (PNs) that mediate feedforward excitation to the mushroom body’s 2000 Kenyon cells (KC). Previous theoretical studies have suggested that purely random PN-KC connections produce optimal KC odor representations. However, recent anatomical studies have argued that two distinct connectivity motifs may be present, giving the connectivity structure.


In this study, we collaborated with the Caron Lab to examine what the proposed PN to KC connectivity motifs, bias and grouping, might have on KC representations of odors. We found that the motifs allowed for distinct features in representations of ethologically significant odors and in the overall discriminability (and its inverse, generalizability) of KC odor representations. 

D. Zavitz, E. Amematsro, A. Borisyuk, SJC Caron. "Connectivity patterns shape sensory representation in a cerebellum-like network". bioRxiv.

A Multiscale Model of Connectivity Between Functional Units


Network models are used to analyze the brain at many spatial scales. However, the effect of connectivity properties of one spatial scale on another is not well understood. In this project, we created a network model in which nodes correspond to multicellular functional units (olfactory bulb glomeruli, cortical columns, etc.), and edges result from connections made by each node’s constituent neurons. We postulated wiring rules for individual neurons based on previously published single-cell anatomy statistics. The resulting networks have structural properties unlike some frequently used network models but can be approximated by more traditional models in some parameter regimes.

In Preparation

Role of Network Structure in Refractory Dynamics


The relationship between structural properties of a complex network and dynamics over such a network is of interest in many fields, including systems neuroscience. In this project, we use an idealized, dynamical model of the spread of neuronal network activity with refractoriness (loosely inspired by dynamic responses in the antennal lobe to sensory neuron input) to explore this relationship. Staring with several broad network classes, we systematically altered their connectivity and tracked the resulting number and lengths of the stable periodic sequences of activity (the coding space).  In each network class there exists a narrow range of "optimal" connectivity where the coding space has the largest repertoire of available stable activity patterns. A detailed examination of the underlying structure of "optimally-diverse" networks reveals that most network-structure characteristics differ from one network class to the next. However, in all of them, the size of the largest strongly connected component predicts the coding properties of the network.

In Preparation

bottom of page