Extracellular signals were amplified (×1000) and wide-band pass filtered (1 Hz–5 kHz). Intracellular signals were buffered (×1) and amplified by a DC amplifier (Axoprobe 1A; Axon Instruments). Signals were continuously acquired at 20 kHz on one or two synchronized 64 channel DataMax systems (16 bit resolution; RC Electronics, Santa Barbara, CA). All analyses were conducted off-line. The stimulation signal was directly fed from the stimulator output (STG1008 Stimulus Generator; Multi Channel Systems, Germany) to one of the recording channels. The peak-to-peak amplitude of the stimulator output is referred to as the stimulation intensity. All intensity values are given as peak-to-peak amplitudes. For LFP display and analysis, the volume-conducted stimulus (TES) artifact was removed as follows. The TES-related component of the LFP was estimated using least-square fit between the stimulation signal and the LFP. Time delay parameter and the scaling factor of the fit were adjusted dynamically in 10-second windows, sliding in 7-second steps. By subtracting the best-fit component from the LFP a continuous artifact-free signal was constructed (
SI Fig. 4). Spectrograms of the LFP trace (
Fig. 6A) were computed using windows of 4-second lengths, sliding in 1-second steps.
Unit activity was detected from the high-pass filtered (>0.8 kHz) trace and single units were semi-automatically isolated using Klusta-Kwik© (Harris et al., 2000 (
link)), followed by a manual refinement using a custom-made software (Hazan et al., 2006 (
link)), which utilized spike waveforms, auto-correlograms and cross-correlograms. To ensure that only high quality units were used in the analyses, we set two criteria for spike inclusion. First, the amplitude of the spike should be >60 µV. This value is based on our previous observations that smaller amplitude spikes result in clusters bordered closely by other clusters (at least in the hippocampal CA1 region; Henze et al., 2000 (
link)). Second, the peak-to-peak spike amplitude (p2p) should exceed 3.6 times the temporally local background noise, determined according to the standard deviation of the signal in the immediate vicinity of each spike (between 0.8-0.4 msec before and 0.4–0.8 msec after the extracellular trough), and scaled by the natural logarithm of the number of samples across which the standard deviation was computed. This definition of spike signal/noise, SNR = p2p / SD(noise) / ln(n), is independent of the number of samples used to compute the noise and is asymptotically 1 for band pass-filtered white noise (
SI Fig. 1). Although these strict criteria likely eliminated several true spikes, it reduced the likelihood of obtaining false positive results. A cluster was classified as ‘multiunit’ if the autocorrelogram lacked a clear refractory period. Multiunit clusters recorded from the same electrode (e.g., the shank of the probe) were merged into a single multiunit cluster to avoid potential oversampling by neighboring recording sites. Short latency (< 5 msec) temporal interactions with other isolated neurons were used to identify single neurons as putative excitatory or inhibitory cells (Bartho et al., 2004 (
link); Isomura et al., 2006 (
link); Fujisawa et al., 2008 (
link); Sirota et al., 2008 (
link)). The significance of monosynaptic connections was evaluated according to the global and point-wise significance values obtained by jittering (n=500) the cross-correlograms between single unit pairs. Cross-correlograms with significant peaks < 5 msec of the reference neuron spikes were regarded as monosynaptically connected (Fujisawa et al., 2008 (
link)).
TES stimulation was applied multiple times (minimum of 5 trials) for a given stimulation protocol, that varied in intensity, frequency (0.8 to 1.7 Hz) and duration (15 cycles to 60 cycles; 1 minute long for few acute experiments) across sessions. Each stimulation trial was followed by a stimulation-free period (> 40 sec, except for short duration stimulation trials, for which it was 10 sec). The signal was a sinusoid waveform. In most sessions, stimulation was applied through 3 stimulating electrodes (3-poles), such that the same polarity was applied to both hemispheres (side) versus the opposite polarity applied to the center (frontal) electrode (
Fig. 1Aa, d). This configuration yielded synchronous electric fields in both hemispheres. In a few experiments, bipolar stimulation was used (
Fig. 1Ab–c). Trials using the same stimulation protocol were combined for statistical analysis.
To assess the effect of stimulation on unit firing, each spike was assigned to the instantaneous phase of the reference TES signal obtained by Hilbert transformation. The significance of phase-modulation was evaluated using Kuiper’s test of random deviation from uniformity on circular data (Fisher, 1993 ). This omnibus test was preferred over the more standard Rayleigh test, since our findings (such as bimodal phase preference) precluded having a predefined assumption regarding the modality of the TES-induced phase entrainment and thus the application of conventional (unimodal) phase entrainment measures (e.g., Rayleigh statistic). Since omnibus tests require a sufficiently large sample of data, we applied 2 criteria a) a minimum number of spikes (at least 250 per stimulation protocol) and b) a minimum number of trials (at least 5) for a given stimulation intensity, frequency and configuration. Because both multi- and single units had different number of spikes and both were tested against the same uniformity hypothesis, we examined whether the number of spikes biased our results. A fixed number of spikes (n=600, n=1000) was randomly chosen from a given TES session and tested against uniformity. Such sub-sampling did not show any particular trend substantially different from the case when all spikes were included. Probabilities <0.01 were regarded as significant.
For the color-coded population display, the spike count of each unit was normalized by subtracting the mean spike count of that unit from the spike count of each bin and dividing by the standard deviation (20° phase bins across one period of TES signal). To examine the relationship between spikes and the intrinsic slow oscillation, the peak spectral power of LFP (peakF) within the 0.4–10 Hz frequency range was computed in 3.2-second windows. The original LFP signal was band-pass filtered around peakF, between [0.75×peakF 1.25×peakF] Hz, using 2
nd order, Butterworth Filter (a built-in Matlab function). The phase of the band-pass filtered LFP was then derived by Hilbert transformation. Phase-modulation of units by the slow oscillation was evaluated by Rayleigh circular statistics, since the unimodal phase modulation by the intrinsic slow oscillation could be safely assumed from previous experimental evidence (Steriade et al., 1993 (
link)). For the display of cycle-by-cycle unit firing, peri-event time histograms of spike times around the first/last troughs of the reference TES signal were constructed.
For the analysis of the intracellular signal (<1,250 Hz, low-pass), the signal was band-passed filtered (0.5–10 Hz) to remove the action potentials. The relationship between the filtered intracellular potential (V
i) and the phase of the slow oscillation or TES was assessed using joint probability density. This measure allows the assessment of the correlation between network or TES effects and V
i across (20°) phase bins (Isomura et al., 2006 (
link)). To compensate for the added effect of the TES-induced field on V
i, TES was repeated after the pipette was withdrawn from the neuron. The obtained extracellular average polarization for each phase bin of TES signal was subtracted from the V
i to arrive at a field-corrected joint density (
Fig. 1D).
To evaluate the state-dependence of stimulation, the Chi-square test was performed, comparing the distribution of P-values, the significance of phase modulation (Kuiper’s test), of each unit. Either two or three P-value categories were constructed: not-significant (nS, P>0.05), significant (S, P<0.005
Ozen S., Sirota A., Belluscio M.A., Anastassiou C.A., Stark E., Koch C, & Buzsáki G. (2010). Transcranial Electric Stimulation Entrains Cortical Neuronal Populations in Rats. The Journal of Neuroscience, 30(34), 11476-11485.