Abstract
Recently, two architectures have been studied that explain some response properties of layer 4 simple cells in V1: one has anti-phase feedforward inhibition (Troyer et al, 1998 J Neurosci 18:5908–27), the other has isotropic lateral inhibition (McLaughlin et al, 2000 PNAS 97:8087–92). Differences in connectivity between excitatory and inhibitory neurons in these models suggest that their response dynamics should differ for oriented stimuli that shift suddenly in phase or orientation. Also, spike train cross-correlograms (CCGs) should differ in the models. I implemented both types of architectures and configured them to have realistic tuning curves for drifting sinusoidal gratings. Both have three populations of neurons: LGN cells and cortical excitatory and inhibitory cells. Each cell was modeled as a conductance-based integrate-and-fire unit. Realistic temporal dynamics were used for AMPA, NMDA, and GABA_A synaptic conductances. The response dynamics of model units were tested with oriented sinusoidal stimuli that changed abruptly between the preferred state and one of opposite phase, orthogonal orientation, or zero contrast. Model output was compared to data obtained earlier in the macaque (Bair et al, 2002 J Neurosci 22:3189–205). The anti-phase inhibition model was more consistent with the macaque: a stimulus transition from opposite to preferred phase produced a substantial additional delay of response onset. The models were also compared in terms of CCGs for pairs of neurons. The striking difference in CCGs for the models suggests experiments in V1 to determine which architecture is more realistic. However, in both models, details of temporal dynamics and CCGs depended strongly on parameters. The models have been developed within a general framework that accepts arbitrary visual stimuli and returns spikes and intracellular voltages. An online interface to the modeling framework is available at www.imodel.org to allow public testing of the models.
Supported by the Royal Society USA Fellowship