Figure 1.

Experimental design and methodology. A, The sinusoidal amplitude modulation (AM) stimulus consisted of a continuous broadband noise for which amplitude was either modulated or unmodulated. AM segments fluctuated periodically, at a single rate from 2 to 32 Hz, or irregularly, with individual periods from the same range presented in pseudorandom order. Gerbils performed a detection task in which they could safely drink water from a metal spout during all AM sounds. Unmodulated noise predicted a small electrical shock, which subjects learned to avoid by withdrawing from the spout momentarily. B, Neural activity was recorded in auditory cortex during task engagement, and single units (SUs) were isolated offline. Waveforms of two units are shown on one shank of a 64-channel probe. C, The two units were separated in principle component (PC) space. D, The autocorrelation of spike times for each unit was confirmed to show a clean refractory period. E, The distributions of waveform amplitudes across the session were inspected to ensure that a majority of spiking events were captured and that drift did not degrade unit quality. If all conditions were met, the unit was labeled as a SU and included in further analyses. F, The width of spike waveforms for all SU resulted in a clean bimodal distribution, which was used to separate regular/broad spiking (RS) cells from narrow spiking (NS) cells.