Click to Enlarge

Working memory is defined as a form of memory that is maintained by the active firing of neurons. A central underlying question is the biophysical mechanisms that maintain firing. Although this process is likely to involve reverberating circuits in which neurons stimulate each other at recurrent synapses, there is likely to be further complexity to this process. Pharmacological studies show that NMDA antagonists can affect working memory, but the reason for involvement of NMDA channels has been unclear.

Our recent work with neural network simulations indicates that the voltage-dependence of the NMDA channel could be very important for maintaining the activity of novel memories. A second important issue is the role of intrinsic conductances: if activity stimulated an intrinsic conductance that led to subsequent depolarization, this would maintain the firing of the cell. We believe that cells contain such conductances and that these are important for the maintenance of working memory.

Neural networks in some parts of the brain can also represent scalar quanitities with proportional, persistent changes in neuronal firing rates. For instance, integrator networks in the brain show persistent firing that reflects the sum of previous excitatory and inhibitory inputs from external sources. We have recently developed a new class of models for such networks in collaboration with Alexei Koulakov at the University of Utah and Sridhar Raghavachari and Adam Kepecs in my lab. This class of models relies on self-boostrapping network composed of bistable neurons. Such a ``staircase'' model of integrator networks is extremely robust to both parameter variations as well external noise. The proposed mechanism may be relevant for other forms of parametric working memory, decision making and motor control.

A final important question is the role of oscillations in working memory. A great deal of theoretical work has been done to explore the idea that dual oscillations in the theta and gamma range allow a single network to actively store multiple items: the cells representing a given memory become active during a given gamma subcycle of a theta cycle; since there are about 7 subcycles within a theta cycle, about 7 memories can be actively held in the network consistent with psychological findings. This ordered pattern of memories repeats every theta cycle.

Recently we have begun an experimental program aimed at testing these ideas. This work, which is being done in collaboration with Michael Kahana of the Psychology Dept. and Joe Madsen, of Children's Hospital, takes advantage of the fact that patients with intractable epilepsy are implanted with electrode arrays on the brain surface in order to localize their epilepsy. These patients are often able and willing to participate in psychophysical experiments. We have now given simple working memory tests to these patients and used the signals from the electrode array to ask whether there is any connection between working memory and oscillations. Our results show sites at which large-amplitude theta oscillations are turned on at the beginning of the working memory task and turned off at the end of this task. This correlation is consistent with an integral role of these oscillations in working memory.