Dynamics of temporal integration in the lateral geniculate nucleus

Abstract

Before visual information from the retina reaches primary visual cortex, it is dynamically filtered by the lateral geniculate nucleus (LGN) of the thalamus, the first location within the visual hierarchy at which non-retinal structures can significantly influence visual processing. To explore the form and dynamics of geniculate filtering we used data from monosynpatically connected pairs of retinal ganglion cells (RGCs) and LGN relay cells in the cat that, under anesthetized conditions, were stimulated with binary white noise and/or drifting sine-wave gratings to train models of increasing complexity to predict which RGC spikes were relayed to cortex, what we call “relay status”. In addition, we analyze and compare a smaller data set recorded in the awake state to assess how anesthesia might influence our results. Consistent with previous work, we find that the preceding retinal inter-spike interval is the primary determinate of relay status with only modest contributions from longer patterns of retinal spikes. Including the prior activity of the LGN cell further improved model predictions, primarily by indicating epochs of geniculate burst activity in recordings made under anesthesia, and by allowing the model to capture gain control-like behavior within the awake LGN. Using the same modeling framework, we further demonstrate that the form of geniculate filtering changes according to the level of activity within the early visual circuit under certain stimulus conditions. This finding suggests a candidate mechanism by which a stimulus specific form of gain control may operate within the LGN.

Significance

The LGN is a dynamic, tunable filter, transforming information as it flows from the retina to primary visual cortex. In this work we utilize a large data set of monosynaptically connected RGC and LGN cell pairs to model the filtering function performed by individual LGN neurons in the anesthetized or awake state. We demonstrate that, while much of the filtering that the LGN performs can be accounted for by temporal summation, other factors, such as the bursting activity of relay cells, also play a role. Additionally, we show that the time scale of summation is dynamic under certain stimulus and network conditions and that the integration dynamics are largely similar between the anesthetized and awake states.