At the end of training, starlings were presented with random nondifferentially reinforced (with secondary reinforcer only) probe stimuli consisting of each of the eight training motifs in isolation (i.e., not paired) to obtain behavioral confirmation that all four task-relevant motifs were recognized ( Figures 1E and 1F). Probe stimuli were randomly interleaved on 8%–20% of all trials during these probe sessions. Starlings were anesthetized with urethane (20% by volume, 7–8 ml/kg) and head fixed to a stereotactic apparatus inside a sound-attenuating chamber. A small craniotomy was made dorsal to CLM, and multichannel silicon electrode arrays (177 μm2 electrode surface area, 50 μm spacing, 1 × 16 and 1 × 32 electrode

layout; NeuroNexus technologies) were inserted into CLM. For some Rucaparib subjects, only the 1 × 32 array was used (Figure S2M). Motif stimuli were presented free field from a speaker 30 cm from the bird at sound pressure levels matched to those during behavioral training (mean, 65 dB; peak, 96 dB). Electrode arrays were advanced while presenting the 12 motif stimuli until two or more auditory single units were isolated. Once single units were isolated, all 12 single motifs and the set of training

motif pairs were presented pseudorandomly in blocks while the extracellular electrical activity was amplified (5,000 × gain; AM Systems), filtered (high pass, 300 Hz; low pass, 3–5 kHz), sampled (20 kHz), and and saved digitally learn more for offline analysis (Spike2; Cambridge Electronic Design). Putative action potentials in the recorded voltage traces were identified by amplitude and sorted into single units with principal components analysis on waveform shape

using Spike2 software (Cambridge Electronic Design). Only large amplitude spike waveforms that formed a clear cluster in principal component space and that had very few refractory period violations were considered to be single units. In our sample, 99.3% (133/134) of all (Wide Spiking+Narrow Spiking) neurons had no refractory violations (interspike intervals of less than 1 ms) and one neuron had a single violation, which accounted for less than 0.005% of all measured ISIs for that neuron. Since presentation of task-relevant, task-irrelevant, and novel motifs was temporally interleaved, none of the effects reported here can be due to changes in neuron isolation or changes in anesthetic state. Because the recording sites on each multichannel array were only 50 μm apart, stereotrode sorts were used to further improve spike-sorting quality. All but one of the WS neuron pairs analyzed here were recorded from different electrode channels on the multichannel arrays. Omitting the one pair recorded from the same channel does not alter the main results. Only neurons that were driven by at least one motif were used in subsequent analyses. All further analysis was performed using custom-written MATLAB (MathWorks) software.