Highlight
Ups and Downs of Visual Cortex
Receptive fields are sharper when measured using down-state spikes.
Achievement/Results
Carnegie Mellon graduate student Ryan Kelly, and his advisor, Professor Tai Sing Lee, have discovered that neurons in visual cortex reveal sharper tuning curves when the current activity of the surrounding network of neurons is taken into account. With funding from NSF’s IGERT program, Kelly, a computer science doctoral student, learned to do neurophysiological recording from the brains of macaque monkeys. Then, working with Lee, statistics professor Rob Kass. and computer science professor Gary Miller, Kelly developed a statistical model of the network of neurons he was recording from. The model was based on the idea that the network’s global activity could be approximated by two states, an up-state and a down-state. In the up state, many neurons are simultaneously active, influenced in part by their neighbors. In the down state, neuron firing is more closely tied to properties of the visual stimulus and the neuron’s unique tuning curve.
Using this model, Kelly was able to show that neurons in primary visual cortex (area V1) exhibit better-defined receptive fields when the fields are measured using down-state rather than up-state spikes, even when the number of spikes is the same.
V1 cells are known for their orientation selectivity: they tend to respond best to edges at a particular “preferred” orientation, which differs from one cell to the next. Kelly and Lee developed a regression model of network activity, measure by the local field potential (LFP), that allowed them to separate out the stimulus component of firing from the component due to the influence of other cells in the network. By using this model to subtract out the network influence, they were able to show that cells’ orientation tuning curves are sharper than the raw spiking response suggests.
Kelly and Lee also examined local models of the network state using local field potentials from just nearby electrodes, or a single electrode, instead of averaging over the entire electrode array. They found that this produced even sharper tuning curves, suggesting that different regions of V1 may be independently switching between up and down states.
Address Goals
This discovery illustrates how the interplay between neuroscience, computer science, and statistics can yield new ways of looking at the brain that change our understanding of visual cortex, Understanding the neural basis of vision has wide ranging applications in psychology, education theory, medicine, and robotics.
This activity also illustrates how the interdisciplinary training fostered by the IGERT program is producing new scientists who are able to bring together tools from multiple areas and pursue novel approaches to fundmental scientific problems. Very few computer scientists or engineers personally record from the brains of behaving animals, but in the Center for the Neural Basis of Cognition’s IGERT program, this is not uncommon.
