Spike sorting

In the standard fixed-threshold method, negative-going spikes were detected using an amplitude threshold (2.8–3.2 s.d. of noise floor), with a shadow period of 0.66 ms after each threshold-crossing. We chose a 0.66 ms shadow period because this was the smallest value that prevented single spike events as being detected more than once. Detected spikes were then clipped (1.5-ms waveforms) for spike sorting. Isolatable single-units were labeled through manual inspection and had to satisfy the following criteria: <0.5% refractory period violations (defined as inter-spike interval <1.5 ms) and <30% estimated missed spikes (based on Gaussian fit of detected spike amplitudes relative to the voltage amplitude detection threshold). The mean spike waveform of each single-unit was then used as a template for NTM spike detection. As in the standard method, a shadow period was enforced after each voltage segment detected as a spike (0.66 ms). Clustering was then redone as in the standard method. Details of the NTM method are given in the Results section.

All of the spike sorting used the open-access software UltraMegaSort2000²⁷ (ums2k), implemented in Matlab. In the standard method spike detection, alignment, clustering and manual curation were all done with the functions provided in the ums2k package. For NTM, spike detection was done with custom software in Matlab (the NTM algorithm is described in detail in the results section) and spike alignment, clustering and manual curation were all done with the ums2k software package. Note that NTM can be implemented with any clustering algorithm.