5 EGTA, 10 Na2-phosphocreatine, 4 Mg-ATP, and 0.4 Na2-GTP (pH 7.3, 292 mOsm). The electrode resistance ranged from 7 to 12 MΩ. Electrodes were introduced through a craniotomy, usually 2 mm wide and 6 mm long, at Horsley-Clarke posterior 1–7 mm and near the midline. The electrodes were placed 500–700 μm apart at the cortical surface, angled at 25° relative to one
another so that their tips approached each other as they were driven into the brain. Warm agar solution (3% in saline) was applied to cortical surface to reduce brain movement. Cell pairs were included in the analysis only if the resting Vm of each cell was stable and was more hyperpolarized than −45 mV for long enough (15–60 min) so that we could record data from multiple sets of stimulus presentation. Vm was recorded using an Axoclamp 2A amplifier in bridge mode, anti-alias filtered and sampled at 20 kHz. To reduce capacitive coupling between the two selleck chemicals electrodes, a grounded metal plate was inserted between them. In some experiments (Figure S5), one recording from a pair
was left in juxtacellular mode. For each pair, nonoverlapping blocks (1 s in length) of the spontaneous data were prepared for cross-correlation and spectral analysis through a few steps: (1) spike removal by interpolating the beginning and the end of spikes (Bruno and Sakmann, 2006), (2) subtraction of the Dasatinib DC component so that each block had zero mean, (3) resampling the data from 20 kHz to 4096 Hz, (4) removal of line noise (60 Hz and its harmonics) using Chronux routines (http://chronux.org), and (5) smoothing by Savitzky-Golay mafosfamide filtering (Matlab sgolayfilt function). For visually evoked data, we used only the first 1 or 2 s of the responses (0.25–2.25 s after stimulus onset or 0.25–1.25 s if the stimulation duration was less than 2.25 s).
In addition to the steps listed above for spontaneous data, for each stimulus condition, we also subtracted the stimulus-averaged Vm response in order to remove stimulus-locked component. This step was not critical for complex cells, since by definition they show little temporal modulation at the stimulus frequency (or higher harmonics). After the above preparation, cross-correlation of Vm1Vm1 and Vm2Vm2 for each block of data was calculated as follows (Matlab xcorr function): R12(τ)=∑t=1N−τVm1(t+τ)Vm2(t)∑t=1NVm12(t)∑t=1NVm22(t),τ≥0;R12(τ)=R21(−τ),τ<0where N is the total number of data points (4096 for 1 s block) and τ is the time lag. Cross-correlations of all blocks were then averaged for each stimulus condition. The peak of the cross-correlation was taken as the maximum within 10 ms of zero time lag; the full width of the correlation was measured at half height. Since subtraction of mean response eliminated most stimulus-locked components, the cross-correlation for shift-predictor data (shifted by one trial) was flat (not shown), with no significant peaks near zero time lag.