Bell Labs, Lucent Technologies, Physiology and Neuroscience, New York University Medical Center, Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology
Cellular and circuit mechanisms of persistent neural activity
In the last few decades, neural activity related to short-term memory has been observed in many brain areas. This activity encodes information about a past stimulus, and can persist for up to tens of seconds after the stimulus is gone. Both cellular and circuit mechanisms have been proposed to explain this phenomenon. According to the cellular explanations, persistence is intrinsic to single neurons, resulting from biophysical processes with long time scales. In contrast, the circuit explanations hold that the time scale of persistence is an emergent property of networks of neurons interacting by synaptic feedback loops. We are trying to resolve the debate between cellular and circuit mechanisms by studying persistent neural activity in the goldfish (Carassius auratus) brainstem horizontal velocity-to-eye-position integrator, where neuronal firing rates store eye position memory (Pastor et al., PNAS (1994) 91:807).
One experimental approach we are taking is to identify manipulations that have well-understood effects on single cells and synapses in vitro, apply those manipulations in vivo, and quantify the resulting perturbations in gaze and neural activity. For instance, allosteric modulators can be used to artificially change the effective weight of a subset of synapses in a circuit. Likewise, ion channel blockers can be used to alter the input-output properties of neurons (Wang et al., 1998, Soc. Neurosci. Abst. 602.4) or to partially inactivate a brain area (Aksay et al., 1998, Soc. Neurosci. Abst. 75.4). Our results so far are consistent with a network mechanism in which persistent activity is sustained in part by tuned positive synaptic feedback within the neural integrator. However, the amplitude of the effects we have observed is smaller than predicted by attractor-based network models employing tuned positive feedback, leaving a quantitative discrepancy between experiment and theory that is not yet resolved. (Supported by Lucent Technologies.)