Computer separation of unitary spikes from whole-nerve recordings

A practical and efficient off-line computer technique is described for automatically separating unitary waveforms from multiunit whole-nerve spike train data with no a priori knowledge about the number of units or their waveforms. The procedure requires two recording eletrodes which provide 3 measures on each putative unit: (1) peak-to-peak amplitude on the proximal channel, (2) amplitude on the distal channel, and (3) temporal offset (depends on conduction velocity) between proximal and distal spikes. On the basis of these 3 measurements, individual unitary spikes are automatically separated into clusters according to empirically-determined limits of variability. The results of the program are displayed in 3-D plots of the 3 measures on each unitary spike and in plots of superimposed waveforms from each cluster. These plots can be used to interactively correct clustering errors. The procedure is illustrated with a 1-min segment of spike train data recorded in vivo from the siphon nerve of a freely-behaving Aplysia. We routinely obtain about 10 relatively well-isolated units in such segments. By utilizing the average waveforms and conduction velocities for individual clusters, it may eventually be possible to separate unitary spikes from compound waveforms resulting from simultaneous of two or more units.