Most neurons and sensory cells respond to input signals with a spike train. They encode the input information into the rate of the spike discharge and/or into timings of individual spikes. In the case of spike codes the regularity of spike discharge becomes a major issue. The main sources of the discharge irregularity are the input signal stochasticity, the intrinsic neuronal noise and the intensity of the input signal. Additive noise has relatively little effect on the spike discharge if the input stimulus is superthreshold. For subthreshold stimuli, however, the neuron might be still able to fire spikes because of the noise that randomly pushes the membrane voltage above threshold. The resulting irregularity of spike discharge in this fluctuation driven regime is high and the interspike interval distribution broad. The distinction between sub- and superthreshold regime has therefore important consequences for the firing and coding behavior of neurons in the presence of noise (Gerstner and Kistler, 2002).

We present a method for establishing whether an output of a neuron is signal or noise driven. The method is based on a measurement of the spike latency in response to a test stimulus, and the time interval between the stimulus onset and the last spike preceding it. We assume that the cell is active in the absence of the test stimulus and that the latency of the response decreases the closer the membrane potential is to the firing threshold. In the superthreshold regime the average membrane potential increases throughout the whole interval between two spikes. The average latency therefore monotonically decreases with increasing distance of the test stimulus from the "prestimulus" spike. In the subthreshold regime, however, the average membrane potential settles at a plateau under the threshold. At some distance away from the "prestimulus" spike the latency therefore does not decrease anymore but reaches a constant value. This hypothesis was tested and confirmed with a computational simulation of the experiment, using standard "leaky integrate-and-fire" and Hodgkin-Huxley type model neurons. The method was then applied to the firebug (Pyrrhocoris apterus) filiform sensilla. These sensilla exhibit an ongoing spike discharge in the absence of a stimulus. By using the method we showed that the resting activity of type T₁ sensilla is signal driven, whereas that of type T₂ is noise driven.

Gerstner W. and Kistler W.M. (2002). Spiking neuron models. Cambridge University Press, UK.