Mean spike frequency as a single measure of neural activity does not represent the full extent of coding. Similarly, description of instantaneous frequency as a Poisson process is not fully adequate. It assumes that adjacent spikes occur independently which is not true, because for example of the absolute refractory period.

The methods proposed here adopt a direct approach to describe spike interval patterning because this constitutes the information that reaches the axon terminals. An information model has been developed in which spike patterning increases the predictability of interspike intervals (ISIs). Log ISI histogram probability distributions were used to evaluate the entropy, an absolute measure of the disorder of the interval distribution. Very regular intervals exhibit minimal disorder and convey only frequency information.

The ISI histogram provides a useful representation of the mean spike frequency variability of coding but neglects the order in which the ISIs occurred. Plotting the duration of adjacent ISI pairs against each other as a scatter reveals statistical relationships between neighbouring intervals. The strength of association is given by the mutual information: the reduction in entropy of one ISI given the duration of its neighbour. To describe the scatter distribution, a probabilistic (Bayesian) approach has been adopted and random (Markov) simulations have been undertaken to detect non-stationarity in spike activity.

Oxytocin cell activity in female Wistar rats were tested under four experimental conditions: supraoptic neurones of virgins in vitro and in vivo, and in lactating rats in the absence of presence of suckling pups. All experiments were undertaken in accordance with the Animal (Scientific Procedures) Act, 1986, and experiments in vivo were conducted under urethane anaesthesia (1.1 g kg^-1 I.P.); all animals were killed humanely. Although there were no significant differences in mean spike frequency between the different groups, the ISI entropy was significantly lower in vitro (P < 0.02, ANOVA with post-hoc Student’s t test) than in each of the in vivo conditions; the impoverished coding was possibly a result of the loss of connections in vitro. The mutual information was significantly higher (P < 0.001) in suckled rats than in the other recordings in vivo, reflecting the increased extent of spike patterning.

Appropriate parameters of neural activity thus revealed significant differences in coding between cells in the different experimental groups. Reduced synaptic input in slices decreased the complexity of coding and spike patterning was increased during reflex milk ejections. The study highlights the limitations of mean spike frequency to represent neural activity.

This work was supported by Merck, Sharp and Dohme, and the James Baird Fund.