Precise neuronal firing coexists with large response variability in vivo

Purpose: We investigated the variability and precision of neuronal firing in order to resolve conflicting reports about the size and nature of neuronal noise. Methods: We recorded extracellular discharges from retinal ganglion and lateral geniculate nucleus (LGN) cells in anesthetized and paralyzed cats that were stimulated with drifting gratings on a CRT screen. Two types of analysis were performed: 1) high-resolution PSTHs showed the firing rate following stimulus onset; 2) Fourier coefficients were calculated as measures of the cell's response and response variability. We also created a 'leaky integrator' model with a noisy threshold and compared its behavior with experimental results. Results: We show that neurons can fire temporally precise and reproducible spike trains in response to visual inputs, and that highly precise firing (as measured by PSTHs) can coexist with large response variability (as measured by Fourier analysis) in spike trains generated by the same neurons. The precision of spike times across trials increases with stimulus contrast even while the variability among individual responses is stimulus-independent and large. The data agree with the predictions of our leaky integrator model. Conclusions: Precise firing and high variability can coexist in the same neuron, which can be accurately modeled for this purpose as a noisy, leaky integrator. Demonstrating that a stochastic-threshold device can convey precise temporal information removes the apparent contradiction between two views: 1) that neuronal firing must be approached statistically; 2) that spike timing is what carries information in the nervous system.