Trial-by-Trial Variability and Firing Irregularity in Cortical Neurons

We investigate the variability of cortical spiking activity on different time scales. The objectives of this project are the characterization of dynamic changes in variability, the identification of the sources for variability and the quantification of their respective contribution, as well as the determination of functional implications and the resulting limitations for theories of neural coding. To achieve this goal we follow several complementary experimental and theoretical approaches which are shortly outlined below along with further references to recent results.

Disambiguating Different Types of Variability

Conceptionally we discriminate two types of variability. The first is the trial-by-trial variability of neuronal responses as observed during identical repetitions of the same experiment. These changes in response signify variability on a time scale that is long in comparison to the length of the individual trial. Second, on a much shorter time scale, we observe a irregularity of spiking, i.e. the variability in the inter-spike-intervals of individual spike trains.

Stochastic Point Process Theory

The theory of stochastic point processes [Isham/Cox] provides us with a mathematical framework for studying the spiking statistics of neurons. On the one hand it allows us to define measures of stochasticity which, to a large extent, can be treated analytically. Based on experimentally observed parameters such as the firing rate of a neuron we can make predictions of and calibrate for these measures, based on specific point process models. Of particular interest for us is the so-called class of renewal processes. On the other hand we employ point process models as a means to perform numeric simulations of the activity of single neurons and of independent or correlated populations of neurons. This is used (i) for the calibration of new methods of spike train analysis [1] and (ii) for the generation of synaptic input distributions to either control the input to real neurons in in vitro preparations [Nawrot et al.] or the input to model neurons in an in virtu environment [Kuhn et al.].

Development and Calibration of Analysis Methods

Within this project we are interested in the development and improvement of two types of analysis tools. On the one hand, we need means to reliably estimate different types of variability in a time resolved manner and in the presence of rate-dynamics. On the other hand, if we don not simply consider trial-by-trial variability as noise we desire methods that allow for anlysees on the basis of single trials [1,Rickert et al.]. For a calibration of our methods we employ stochastic point processes.

The Influence of Ongoing Activity on Variability In Vivo

In collaboration with the laboratory of Prof. A. Riehle at the CNRS, Marseille, we investigate trial-by-trial variability and firing irregularity in multiple single-unit recordings from the primary motor cortex of behaving monkeys. We found that trial-by-trial variability as estimated by the Fano factor changes systematically during task performance. Furthermore, variability on a long time scales is always and considerably larger than expected from variability on shorter time scales. Taking together, our findings suggest that trial-by-trial variability in vivo is to a large extent caused by ongoing activity, i.e. slow changes in network activity due to processes which are not controlled in our experiment [3,6,8]. Furthermore, using the same data set we study movement prediction in the single trial from the activity of single neurons and neuronal populations while taking into account the individual trial-by-trial variability.

Generation of Background Activity in the Acute Slice Preparation

By means of dynamic photostimulation we are able to provide a single observed neuron with synaptic inputs from presynaptic neurons at many different sites. This creates a climate of ongoing network activity with permanent synaptic bombardement of neurons embedded in the network of the acute slice. This opens new ways to study the influence of permanent network input on the variability of neuronal oputput.

Martin Nawrot