University of Pennsylvania
The unifying theme for our lab is the study of neuronal dynamics. Historically, our lab has been interested in neuronal mechanisms of anesthesia. Anesthetics have been indispensable for medicine for over 100 years, but the mechanisms through which these drugs act in order to produce a state of unconsciousness remain mysterious. Questions concerning neuronal mechanisms of anesthesia are very closely related to perhaps the most profound question in neuroscience: What is the relationship between brain activity and conscious subjective experience? Our lab is specifically interested in two related questions. The first question is: What are the properties of neuronal activity that distinguish the conscious from the unconscious state. We proposed a novel theory that in order to be able to respond to a stimulus — a sine qua non of consciousness — stability of brain dynamics must be tightly controlled. The brain is never silent, even in the state of deep anesthesia the brain continues to generate complex spontaneous patterns of activity. A stimulus, therefore, acts as a perturbation to this ongoing dynamics. If the dynamics of the brain were too stable, any perturbation will quickly dampen. Conversely, if the dynamics were too unstable, then any perturbation would grow without bound. From this simple intuition it follows that, in order to be able to consciously perceive a stimulus, the dynamics must be precisely balanced between the stable and the unstable regime.
A related, but distinct question is how is the brain able to recover consciousness after its activity has been profoundly disrupted by anesthetics? We have demonstrated that en route to recovery from deep anesthesia, the brain traversed through several discrete transiently stabilized activity patterns. The transitions among these patterns are highly structured such that certain activity patterns form hubs arrival into which are a pre-requisite for further recovery. While, the stabilization of activity patterns greatly shrinks the “activity space”, the structured transitions among them form a dynamical scaffold that gives rise to boot sequence which ultimately allows recovery of consciousness to occur on a physiological time scale.
A more recent approach in our lab is to model neuronal dynamics outside of the context of anesthesia. A technological revolution has enabled massively parallel recordings of neuronal activity using high density electrodes and imaging techniques. Yet, the theoretical approaches needed to extract meaningful information from such complex datasets is critically lacking. Motivated by statistical mechanics approaches we developed a modeling approach for extracting dynamical structures from multivariate, nonlinear, and noisy recordings of neuronal activity. We validated the predictions of this modeling approach using published datasets of whole brain imaging in nematode C. elegans. Using this approach, we can predict when the nematode will switch direction of locomotion. Remarkably, the predictions of the model of macroscopic dynamics are valid despite consistent individual differences in the patterns of activation of individual identified neurons during execution of the same locomotor behavior in genetically identical worms. This suggests that one cannot necessarily expect biophysically realistic models of the nervous systems to be generalizable across individuals. Rather, this observation points to the fact that evolutionary constraints operate at the level of macroscopic dynamics which assure that an adequate behavioral is implemented. We are exploring the consequences of this hypothesis by training artificial neuronal networks on a variety of behavioral tasks and are studying the relationship between network dynamics and task demands.
A pizza lunch will be served.