Presented at the Society for Neuroscience convention, Washington, DC, USA, November, 2008.

The effect of up and down network states on pairwise and 3-way correlation in V1

Ryan C. Kelly1, Matthew A. Smith1, Rob E. Kass2 and Tai Sing Lee1

1Center for the Neural Basis of Cognition, Carnegie Mellon University, Pittsburgh, PA, USA
2Department of Statistics, Carnegie Mellon University, Pittsburgh, PA, USA


We have previously shown that the local field potential (LFP) is correlated with up and down states in primary visual cortex (V1). These states correspond to an increase or decrease in firing rates, unrelated to the visual stimulus, across a large population of neurons. Such global fluctuations in responsivity, which occur on a time scale of several hundred milliseconds, lead to high levels of pairwise correlation among neurons. Correlation can also be induced from other sources, such as common input from upstream neurons. Most existing studies of correlated activity in visual cortex have involved pairwise recordings of nearby single units. Such data is not sufficient to disambiguate large scale fluctuations from local effects. To decouple global responsivity fluctuations from other sources of correlation, we recorded from V1 of anesthetized, paralyzed macaque monkeys using the Cyberkinetics "Utah" array, a 100 electrode recording device with similar properties to conventional single microelectrodes. The array consisted of a 10 x 10 grid with 400 micron spacing between adjacent electrodes. We implanted the electrodes 0.6 mm into cortex, resulting in superficial layer recordings. We recorded activity during a blank stimulus and also during the presentation of natural movies. From each electrode, we recorded all responses which exceeded a threshold and sorted these waveforms offline. Simultaneously, we recorded the LFP from each electrode. We analyzed the spiking correlation among groups of neurons in V1 and its relationship to the LFP. We find that fluctuations in the LFP are able to predict a substantial portion of the slow time scale spiking correlation between neuronal pairs. Our spike train data also contains near-synchronous firing across many cells above chance levels according to independence. We have applied statistical tests of pairwise and higher order interactions and shown that pairwise interactions alone supply the bulk of the information about cell firing. However, we have also found 3-way interactions at levels which reach statistical significance. We developed a model of cell firing which computed the expected correlation between pairs of neurons due to the global network state and separated it from other sources of correlated activity. Furthermore, among certain group of cells the apparent interactions could be due entirely to network up and down states.