Data processing and analysis

Signals were first examined visually using AxoScope 10.6 (Molecular Devices) to choose data segments for further analyses. Analyses were computed and visualized using custom-written code in MATLAB (version R2015b, MathWorks), before being processed with CorelDRAW X6 (Corel). Data analyses included the following: zero phase delay digital filtering, evoked potential averaging, power and phase profile and spectral analyses, coherence, current source density, single- and dual-channel spectra, auto- and cross-correlations, spike-triggered averaging (STA), and spike phase preference. Spectral analysis was used to confirm brain state in chosen segments before conducting other analyses.

Autopower, cross-power, coherence, and cross-phase spectra were computed and plotted for both individual and pairs of field signals (details further below). Spectra were estimated from a series of 6-s-long, Hanning-windowed samples with 2 s overlap using Welch’s periodogram method. Power spectrograms (Wolansky et al., 2006; Whitten et al., 2009) were computed using a sliding window procedure, allowing discrete spectra to be calculated at specific time points across the entire time segment. Windows were 30 s in duration, and slid across the entire file in 6 s increments. These discrete spectra were then analyzed as described above. Spectral profiles were also created for activity recorded with the linear multiprobe in the same way, except that each multiprobe channel was compared against a fixed (nCTX or HPC) bipolar reference, and then power values at spectral peak frequency values for both SO and theta states were extracted. The spatial locations of the channels were then estimated based on the power profile for theta (Wolansky et al., 2006), with the phase reversal point being at the interface between stratum pyramidale and stratum radiatum, and the theta maximum being at stratum lacunosum moleculare.

For chemogenetic experiments, 3 min samples of SO activity were extracted based on the analysis of cortical power spectrograms pre- and post-CNO. This duration was selected as a balance between a long-term, stable SO sample, without being compromised by potential nonstationarities. SO time points were all chosen at least 30 min post-CNO injection to ensure adequate time for the ligand or its metabolites to enter the brain (Whissell et al., 2016). A profile of the power at the peak SO frequency for each sample was then created for each channel across the linear multiprobe in the HPC. Using this, we could determine the channel of maximal SO power, which has been previously demonstrated to be the relative location of the SLM (Wolansky et al., 2006). The CSD (described below) profile of the linear multiprobe was then computed. Spectral (particularly power and coherence) estimates were computed as described above, comparing the CSD of the SLM channel to a fixed nCTX or HPC electrode.

The relationship between neocortical and hippocampal field with RE spikes was assessed by STA. The preferred phase of the unit to the field was computed separately by filtering the field within a specific bandwidth (0.5–1.5 Hz for SO; and 3–5 Hz for HPC theta), and then computing the time points of negative to positive zero crossings. Unit activity was binned (bin size, 18°) according to the phase of the field cycle, from 0° to 360° (from one zero crossing to the next). Spike rates, interspike intervals, and autocorrelation histograms (10–100 ms bin size) were computed to analyze spike train dynamics.

STA significance was computed by comparison with the distribution of STAs using a series (n = 100–1000) of randomized (shuffled) spike trains derived from the original data. Spike trains were shuffled using random assignment based on the actual interspike intervals computed for the original spike train. The resulting STA distribution had a variance that was proportional to the amplitude of the original field signal, but that was lower than the original fluctuations for signals with a strong correspondence.

Significance for autocorrelation histograms was computed in a similar way, using the average bin value of individual point processes (single-unit activity) and their fluctuations within a randomized distribution (n = 100–1000). The 99% confidence limits were computed as the average value ± 2.6 SEM. We classified units with systematic and periodic fluctuations beyond this window as being rhythmic. Rayleigh statistics for circular data were used to statistically evaluate phase histograms (Zar, 1999).

Root mean square (rms) envelopes were created for multiunit RE activity by using a 200 ms window slid by 50 ms increments. The resulting envelope was inverted by multiplying by −1 to ensure that it was not antiphase to the ongoing SO. Power, coherence, and phase estimates were computed as described above for field signals. The peak cross-power and coherence frequency was used to estimate the spectral phase angle between the RE rms and field signal. Coherence between the signals was used as a measure of phase preference (radius). To determine a 99% confidence limit for coherence estimates, a series of time-reversed coherence spectra were computed, and the distribution of values across the entire spectrum was assessed.

CSD analysis was conducted on spontaneously collected field samples or evoked potential averages recorded using the linear multiprobe, following the assumptions of Freeman (1975), Rodriguez and Haberly (1989), and Ketchum and Haberly (1993). Briefly, CSD was computed by estimating the second spatial derivative of adjacent multiprobe voltage traces.