Low and High Gamma Oscillations in Rat Ventral Striatum have Distinct Relationships to Behavior, Reward, and Spiking Activity on a Learned Spatial Decision Task.

Cell classification. Average extracellular waveforms were plotted by peak width, valley width, and firing rate (note log scale). Consistent with previous reports (Berke et al., 2004; Sharott et al., 2009) waveforms separated into a large and a small cluster, thought to contain mostly medium spiny neuron (MSN) and fast-spiking interneuron (FSI) waveforms respectively. Classification based on spike train statistics only (A) identified 55 high-firing neurons (HFNs; Barnes et al. 2005; Schmitzer-Torbert and Redish, 2004b, 2008; PFNs are “phasically firing neurons”). Classification based on waveform shape and firing rate (waveform peak width smaller than 0.15 ms, waveform valley width smaller than 0.35 ms, and a firing rate above 2 Hz) identified 73 putative FSIs (B). The 53 neurons that satisfied both the spike train and waveform-based classification criteria for putative FSIs (C) were included for analysis. Because FSIs are thought to be rare in ventral striatum (Berke et al., 2004; Cowan et al., 1990; Kita et al., 1990), the relatively high (10–15%) percentage of FSIs we observed may stem from the fact that our recording sites were somewhat more dorsal (but clearly in ventral striatum overall, see Figure 3).

Gamma-50 and gamma-80 power are differentially modulated before turnaround points on the central segment of the maze, in the absence of reward. (A) Gamma-50 power was briefly elevated shortly before turnaround (thin vertical line at time 0), while gamma-80 power (grey line) did not show sharp peaks but was higher overall before turnaround compared to after (B). The animals' average speed indicated an onset of movement at the time of turnaround (C).

Putative fast-spiking interneuron (FSI) spike-triggered averages (STAs) indicate interactions with local field potentials. (A,B) Example STA of two FSIs. The STA of the top neuron shows clear gamma modulation; note also the asymmetric slow component following a spike. The bottom neuron shows 12 Hz, but not gamma, modulation. In general, different FSIs exhibited a range of different STA patterns, and a significant proportion were not modulated. (C) Average STA across all FSIs; note gamma oscillation and asymmetric slow deflection. The sharp spike at time 0 likely resulted from the spike waveform interfering with the LFP, a consequence of recording both from the same tetrode.

Putative fast-spiking interneurons (FSIs) showed significant gamma phase modulation. (A) Example LFP traces showing episodes of high coherence at 50 Hz (top) and 80 Hz (bottom). (B) Gamma phase histograms at 50 Hz (top) and 80 Hz (bottom) for an example neuron (same neuron as A, top). (C) Polar plots of the phase angle and magnitude for all FSIs at 50 Hz (top) and 80 Hz (bottom). For both frequencies, the majority of cells showed a significant phase preference (Rayleigh's r, p < 0.001, corrected for multiple comparisons; 32/53 for gamma-50, 36/53 for gamma-80). At both frequencies, the distributions of mean phase angles were significantly non-uniform (see main text for details).

Putative fast-spiking interneurons (FSIs) showed a variety of firing rate tuning relationships to gamma-50 and gamma-80 power. (A,B) Firing rate tuning (top row) and coherence tuning (bottom row) to gamma-50 power (left column) and gamma-80 power (right column) for two example FSIs. Firing rates were normalized to each neuron's minimum firing rate of its raw tuning curve. Power was z-scored within sessions in order to allow comparison of tuning curves on the same scale. The neuron in (A) increased its firing rate with increasing gamma-50 and gamma-80 power, by a factor of 1.5 or more for the latter, while the neuron in (B) decreased its firing rate with increased gamma-50, but increased firing with stronger gamma-80. For neither neuron did coherence change significantly with gamma power. (C) Population scatterplot of tuning to gamma-50 against tuning to gamma-80. Significant linear regression fits (corrected for multiple comparisons) are shown in black; 31 of 53 neurons were tuned to either gamma-50 power, gamma-80 power, or both. While overall there was a significant negative correlation between gamma-50 and gamma-80 power tuning, this relationship depended on the strongly tuned point in the top left corner. With this point removed, the two were uncorrelated (r = 0.0023, p = 0.99). Arrows indicate the neurons in the examples (A,B). (D) While about half of FSIs (26/53) showed an effect of gamma power on coherence, coherence tuning to gamma-50 and tuning to gamma-80 power were uncorrelated.

Distinct putative fast-spiking interneurons (FSIs) cohere with gamma-50 or gamma-80. (A) Average spike-field coherence for all FSIs (red) and the average obtained from randomly interspike-interval-shuffled bootstrap samples (black) shows overall coherence in the wide gamma band that cannot be explained by first- or second-order spiking statistics. (B) Coherence for all FSIs individually, z-scored against the distribution of bootstrap samples, ordered by coherence z-score at 50 Hz (bottom to top). While some neurons cohere at both 50 and 80 Hz, there are clear, distinct subgroups that cohere at one, but not the other. (C) Correlation between (raw) coherence at different frequencies across all FSIs. The observed correlation between 50 and 80 Hz was much less than that expected by chance (bootstrap-shuffled correlation in (D), white arrows).

Behavior on the Multiple-T maze. (A) Diagram of a single Multiple-T configuration (“RRLL”), with the left side rewarded. T1-T4 indicate turns, with T4 the final choice point. Food reward is delivered at the feeder sites (F1, F2) when the rat crosses the active feeder trigger lines. (B) Pseudocolor heat map of average speed over laps on the linearized track. Note low average speed at the feeder sites, where rats pause to eat, and similar speeds on the T4-F1 and F2-S (start) segments. Units are cm/s.

Histology. Over 90% of recording sites were localized to ventral striatum, with the majority in the core of the nucleus accumbens and in ventral caudate-putamen. Symbols indicate final electrode positions; however, over all days that data were taken, electrodes were only lowered by small amounts (mean maximum deviation across all tetrodes, 125 μm, median, 90 μm) such that the locations indicated are representative. While we could not exclude the possibility that two recording sites from the more dorsal striatum were included in analyses (we were unable to conclusively assign these electrode tracks to particular electrodes, although their location in comparison with turning records suggests these were likely either unused references or broken electrodes) these constitute a small fraction of the total data set. As our results were consistent across subjects, it is unlikely this possibility affected our results or conclusions.

Ventral striatal local field potentials contain low and high gamma frequencies. (A) Example unfiltered local field potential trace containing alternating gamma-50 and gamma-80 power (top) and associated spectrogram (bottom) with characteristic structure. For display only, the spectrogram was obtained using a 250-ms time window at 10-ms time step resolution. Units on the spectrogram are in dB, a log-transform of power (10 × log10 (μV2)/Hz). Taken from session R117-2007-06-01, tetrode 7. (B) Average power spectral density (PSD) over all sessions. Note the peak at 50 Hz and a wider “hump” defined by inflection points at 60 and 100 Hz. Error bars (SEM over sessions) were too small to be visible. (C) Average cross-frequency autocorrelation (over all sessions) for different frequencies within each spectrogram. Note the distinct, slightly anticorrelated (white arrow) zones at 45–55 Hz (gamma-50) and 70–85 Hz (gamma-80) frequencies. See main text for details.

Low (45–55 Hz, “gamma-50”) and high (70–85 Hz, “gamma-80”) gamma power are differentially modulated across the track and over time. (A) Representative spatial distribution of gamma-50 power over the track for early laps (1–10, left) and late laps (41–50, right) from a single session. (B) Spatial distribution of gamma-80 power over the track for early laps (1–10, left) and late laps (41–50, right) from the same session (R132-2007-10-17, tetrode 4). (C) Distribution of gamma-50 power over position on the track (horizontal axis) and lap (vertical axis) averaged over correct (rewarded) laps from all sessions. Averages over laps and over space are shown to the top and to the right respectively. Note the abrupt increase in power at the feeder sites, the more gradual return back to baseline, and the development of this pattern over the first few laps. (D) Distribution of gamma-80 power over position on the track (horizontal axis) and lap (vertical axis) with panel layout as in (C). Note the contrast with (C) in both space and time: high gamma power increases gradually up to the reward sites, and is highest during early laps.

Distinct spatial modulation of gamma-50 and gamma-80 power is consistent across subjects. Power values were z-scored across space for each session individually and plotted in black (gamma-50) and grey (gamma-80). Error bars represent SEM over sessions.

The spatial distribution of gamma-50 power, but not gamma-80 power, is affected by the presence or absence of reward receipt at the reward sites. (A) Spatial distribution of gamma-50 power on error laps (red line, no reward received), matched correct laps (blue line, reward received) and all correct laps (grey line). Note the higher power at F2 on correct laps. (B) Spatial distribution of gamma-80 power; panel layout as in (A). While gamma-80 power was increased elsewhere on the track, power at the reward sites was similar between rewarded and non-rewarded laps (but see Figure 9 for a closer look). (C) Distribution of error laps (top); most errors occurred before lap 10, when gamma-50 and gamma-80 power levels had not yet stabilized (Figures 5C,D). To avoid biasing the comparison between error and correct laps, we compared error laps to a set of matched correct laps (the preceding or following correct lap for each error lap, chosen randomly). Different rats contributed a similar number of error laps (bottom).

Gamma-80, but not gamma-50, power was elevated at the final choice point (T4) during early laps. Shown are average z-scored power values for the first quarter of recording sessions (first 10 min, black) and the last quarter (final 10 min, grey) for gamma-50 (left) and gamma-80 (right). Insets show a close up at T4. As shown in Figure 5, overall power levels for both gamma-50 and gamma-80 changed between early and late laps; this analysis reveals the spatial distribution of that power normalized for absolute levels.

Gamma-50 and gamma-80 power are differentially modulated around the time of reward receipt. (A) Peri-event power averages, aligned to arrival times at the first reward site (F1, top) or the second reward site (F2, bottom) for gamma-50 (left column) and gamma-80 (right). As before, error laps (no reward received, red line) were matched with a set of correct laps that occurred at comparable times within sessions (blue line). The average for all correct laps is shown in grey. Food pellet reward usually arrived within 0.5 s of the rat's arrival at reward sites. Over all correct laps, a clear response to reward delivery was apparent in gamma-50 power (peaking at about 0.75 s). During error laps, this response was absent, while a small response to reward delivery at the second reward site could be distinguished, indicating that this effect was not due to a non-specific suppression of gamma-50 power or modulation. Gamma-50 power exhibited a sustained increase following reward receipt. Gamma-80 power also distinguished between rewarded and non-rewarded laps, with a strong reduction in power following reward receipt absent during non-rewarded trials. (B) Average running speed; panel layout as in (A).