The gamma power (in dB) was obtained as follows: p = 20∗log10(Rs/Rb), where Rs = rms value from t0 + 50 ms to t0 + δ ms, and Rb is the rms of the baseline

(time of stimulation: t0), both computed from the gamma-band filtered signal. For plots of bath-applied drug treatment, gamma power was normalized to the control condition at that site. For estimating duration and power of the high-frequency response (Figures S2B and S2E), the same analysis was applied to 3 kHz downsampled traces and high-pass filtered at 500Hz. Computation of spontaneous RGFP966 event duration in Ipc is described in Supplemental Information. Spectral analysis was performed using multitaper spectral estimation with the Chronux toolbox (Mitra and Bokil, 2008). The stimulus-locked MLN8237 part of the response was removed by subtracting the average response across trials from each evoked response, yielding the induced power spectrum. Spectra were computed from t0 + 50 ms to t0 + δ ms, where δ = median duration of the oscillatory episode at each site for each condition. Ratio spectra (R-spectra) were computed

by normalizing induced spectral power at each frequency by power at that frequency during a prestimulation baseline (t0-δ to t0-25 ms). Peak frequencies correspond to the maximum relative power in the trial-averaged R-spectrum in the frequency range of 10–100 Hz. To estimate the gamma oscillation frequency in the drug application experiments, we measured the peak of the raw power spectrum because we were interested only in changes of peak frequency relative to control (Figures 3B, 3D, S2B, and S2D). For sharp electrode recordings in the Ipc, we analyzed subthreshold potentials by low-pass filtering at 200 Hz and the multitaper approach for continuous signals. To compute the spectrum of the bursts, recordings were filtered between 0.5–3.5

kHz, and the spike-times extracted and analyzed with a multitaper spectral estimation algorithm for point processes (Chronux toolbox). Median power, duration, and frequencies were compared across conditions with nonparametric statistics. We used the Friedman test (a nonparametric version of the repeated-measures Interleukin-11 receptor ANOVA) when comparing metrics across conditions applied to the same slice (control, drug wash-in and wash-out). All other comparisons were performed with the Mann-Whitney U test. All p-values were Bonferroni-corrected for multiple comparisons where appropriate. Individual sites (n) represented separate slices, not multiple sites in a given slice. Median values were obtained from 10–40 stimulus repetitions, except for transient drug applications, for which parameters were estimated based on 2–3 repetitions. This work was supported by Stanford Dean’s Postdoctoral Fellowship (C.A.G.), NEI F32 EY018787-01 (C.A.G.), NINDS NS34774 (J.R.H.), and NEI EY019179-31 (E.I.K.).