Serial correlation in neural spike trains: experimental evidence, point process modelling and single neuron variability

Abstract

Event Abstract Back to Event Serial correlation in neural spike trains: experimental evidence, point process modelling and single neuron variability Martin P. Nawrot1, 2*, Clemens Boucsei3, 4, Sonja Grün2, 5 and Farzad Farkhooi1, 2 1 Freie Universität Berlin, Germany 2 BCCN Berlin, Germany 3 Albert-Ludwigs-University Freiburg , Germany 4 BCCN Freiburg, Germany 5 RIKEN Brain Science Institute, Japan The activity of spiking neurons is frequently described by renewal point process models that assume the statistical independence and identical distribution of the intervals between action potentials. However, the assumption of independent intervals must be questioned for many different types of neurons. We review experimental studies that reported the feature of a negative serial correlation of neighbouring intervals. This has been observed in peripheral sensory neurons, and more recently in central neurons, notably in the mammalian cortex [1], and in the insect mushroom body [2]. To model serial interval correlations of arbitrary lags we suggest a family of autoregressive point processes. Its marginal interval distribution is described by the generalized gamma model which includes as special cases the log-normal and the gamma distribution, both have been widely used to characterize regular spiking neurons. We argue that the feature of a negative serial correlation is common to the large class of spike frequency adapting neurons [3] and that it might have been largely overlooked in extracellular single unit recordings, possibly due to spike sorting errors. In numeric simulations of our model we show that a realistic number of between 5% and 15% errors (e.g. [4]) can readily extinguish the significance of the serial correlation. We investigated how serial correlation affects the variance of the neural spike count. The experimentally confirmed negative correlation strongly reduces single neuron variability in our point process model [2] as well as in computational models [3]. This effect is strongest for large observation intervals and, by theoretic argument [5], diminishes for short intervals. This prediction is confirmed in the spontaneous activity of cortical neurons where the Fano factor is decreased by up to 30% [1]. Finally, we examined the effect of serial interval correlation in individual neurons on the significance of coincident spikes in pairs of neurons under stationary rate conditions. We report that a negative serial correlation in the experimentally reported range always reduces the number of chance coincidences [6]. Taken together, we reach the following conclusion: The negative serial correlation - and more generally the SFA mechanism - facilitates the reliable transmission of a rate code as well as the detectability of spike coincidences in postsynaptic neurons. We will test this hypothesis in further experimental and model studies.