in Figure 4A. To obtain an overall quantification of this effect, we calculated the average cross-frequency self-coherence (Masimore et al., 2004). This method computes the correlation of power over time (taken from multitaper spectrograms computed with the Chronux toolbox, Mitra and Bokil, 2008) between pairs of frequencies within a session. The resulting plot indicated a small overall anticorrelation between the gamma-50 and gamma-80 frequency bands (Figure 4C, white arrow; in line with similar observations obtained from anesthetized rats, Sharott et al., 2009). Because multitaper spectral density estimation methods tradeoff narrowband and broadband bias, potentially underestimating the true correlation (Masimore et al., 2004), we repeated this analysis using the short-time Fourier transform method (MATLAB spectrogram function, 1-s Hamming window with 0.5 s overlap). Similar results were obtained by this method; the multitaper-derived plot is shown here for consistency. Because recordings were referenced against an electrode positioned in or near the overlying corpus callosum, it is theoretically possible that these LFP patterns resulted from this electrode. However, when we repeated the above analysis for the data from the reference electrode itself, we observed none of the above structure, implying that these patterns do not originate in the corpus callosum.
