Recent experimental work has identified a novel form of synaptic plasticity (synaptic scaling) in which changes in the level of ongoing cortical activity scale the strength of all of a neuron's excitatory and inhibitory inputs up or down in the direction needed to stabilize network activity. The direction of change in synaptic strengths depends on the identities of both the presynaptic and the postsynaptic neurons, so that excitatory to excitatory (E to E), excitatory to inhibitory (E to I), and I to E connections are all regulated independently, and in the correct direction to stabilize firing rates. Here we propose a collaboration between the Turrigiano lab (experimental) and the Wang lab (computational) to study the computational consequences of these scaling rules for the function of highly recurrent cortical networks. The Wang lab has found that robust models of persistent activity states in prefrontal cortex (such as those thought to underlie working memory) require that synaptic strengths be """"""""tuned"""""""" very finely, and synaptic scaling of excitatory inputs offers an effective and biologically plausible means of performing such synaptic fine-tuning. In this proposal we will work back and forth between experimental and theoretical studies to determine more generally whether the plasticity rules we have uncovered are suitable and sufficient to allow the fine-tuning and optimization of activity states in highly recurrent cortical networks. We will ask whether the scaling rules are implemented differently for different classes of neuron (pyramidal vs. interneurons) and how they interact with classic Hebbian-type learning rules to stabilize and optimize network activity. Computational studies of synaptic scaling are still in their infancy, and many assumptions about how this plasticity will function and interact with classical forms of plasticity such as LTP and LTD are entirely untested. By working back and forth between experiment and theory we hope to begin to illuminate this important subject, and expect these studies to open up a wealth of new ideas about how recurrent networks are set up and stabilized during deveiopment and learning. These studies will shed light on the mechanisms by which perturbations in network activity, such as those produced by chronic use of addictive substances, cause fundamental alterations in cortical circuitry and computational capabilities.
Showing the most recent 10 out of 12 publications
