Supporting InformationBorjigin et al. 10.1073/pnas.1308285110SI Materials and MethodsAnalysis of Power Spectrum. The original sampling frequency of1,000 Hz was down-sampled to 500 Hz to reduce computationtime. A notch lter was used to remove the 60-Hz artifact and itspossible superharmonics. In Fig. 2, the spectrograms using short-time Fourier transform were calculated based on discrete Fouriertransform with 2-s segment size and 1 s overlapping for eachfrequency bin (0.5250 Hz with 0.5 Hz bin size; spectrogram.m inMatLab Signal Processing Toolbox; MathWorks Inc.). Eachsegment is windowed with a Hamming window. The absolutepower was expressed in log scale (Fig. 2A, Top and Middle). Therelative power (Fig. 2B) was calculated at each EEG epoch bydividing the absolute power of each frequency bin with the totalpower over all frequency bands (0.5250 Hz). Thus, for eachEEG epoch, the sum of relative power over all frequency bandsshould be 100%. The mean and SD of absolute and relativepowers was calculated (Fig. 2B, Bottom) for eight frequencybands: delta (05 Hz), theta (510 Hz), alpha (1015 Hz), beta(1525 Hz), low gamma (1; 2555 Hz), medium gamma (2; 80115 Hz), high gamma (or 3; 125145 Hz), and ultra gamma (4;165250 Hz). The grouping of the gamma bands into the fourfrequency ranges was designed to avoid the possible artifactcontaminations from 60 Hz and its superharmonic. For the meanspectral power, the 10-min-long EEG epochs in the middle ofthe whole EEG data for waking and anesthesia states, and 20-s-long epochs corresponding to CAS3 state were taken from allrats (n = 9). To test the statistical signicance of the change ofspectral powers across three states, repeated-measures one-wayANOVA and Bonferronis multicomparison were carried out forthe eight frequency bands. The adjusted P values by Bonferronicorrection are denoted in Fig. 2 (*P < 0.05, **P < 0.01, ***P