Cosyne 2007 Workshops
February 26-27, 2007
The Canyons, Utah
Francesco Savelli
A minimal model of Hebbian plasticity for the formation of place fields from grid-cell inputs
Francesco Savelli and James J. Knierim
The discovery of grid cells in the medial entorhinal cortex (MEC) permits the input-output characterization of a hippocampal subsystem in much greater detail than was previously possible. Anatomical considerations suggest that grid cells and head direction cells represent the major entorhinal input to the proximal-septal area of the CA1 subregion, while lesion studies have indicated that the EC-CA1 component alone (deprived of its CA3 inputs) is still able to produce CA1 place fields. A striking feature of such a system is the transformation of the multi-peaked spatial firing pattern of grid cells into the discharge activity of each place cell, which tends to occur mostly in a single location of the recording enclosure. Understanding how this is accomplished is bound to constrain our theories of hippocampal information processing. Our goal in this study was to investigate how place fields may be obtained from grid-cell inputs by computational modeling. We used integrate-and-fire units, representing place cells, driven by grid-cell spike trains simulated from real animals' trajectories tracked during unconstrained foraging. The scale of such grid-like inputs approximately spanned the range observed in dorso-caudal MEC, and every simulated place cell received up to 100 random inputs. Simply thresholding the input activity with no plasticity failed to produce single-peak place-fields, even when place cells were brought to fire only a few spikes over 15 minutes. By contrast, we found that rapid, postsynaptically gated, Hebbian plasticity permitted the formation of stable, single fields that accounted for most of the cell activity. Other synaptic-rule variants considerably underperformed these results, which points to a critical role of the presynaptic level of activity in determining the direction of plasticity. In further support of the critical role of presynaptic activity, the more complex Oja rule produced a surprisingly similar activity profile for a cell with the same set of grid inputs. Interestingly, the model predicts that the specific trajectory of the animal during initial exploration of a novel environment influences where a cell will form its field. For instance, in accordance with experimental observations, place fields produced by our simulations were preferentially localized along the border of the recording enclosure, where the rat is behaviorally biased to limit its initial exploration of a novel environment. Our model supports the existence of heterosynaptic depression at perforant-path synapses, as observed in vivo, or the hypothesis that synaptic plasticity at the same synapses is modulated by CA1 recurrent inhibition, as suggested by some in vitro experiments. Supported by a Cross-Disciplinary Fellowship from the Human Frontier Science Program and by NIH grant P01 NS038310.
