SUPPLEME TARY FIGURE 1 a b c

Transcription

1 Coherent gamma oscillations couple the amygdala and striatum during learning. Popescu, Popa, Pare SUPPLEME TARY FIGURE 1 a b c LG LP LD R LA BL OT BM HF V HF VP RE VP CL PC d PU e 2 mm R f CP HF 2 mm GP PU GP LG LP HA CM LA BL LA BL BM HF CP 2 mm 2 mm Supplementary Fig. 1 Histological verification of recording sites. Coronal sections stained with cresyl violet. Arrows and arrowheads point to electrolytic lesions performed at the end of the experiments to mark the last recording sites in the BLA (a, d), primary and secondary auditory cortices (b, e), rostral and caudal intralaminar nuclei (c, f), as well as putamen (d, e). Scale bar in e also valid for panels a and d. Abbreviations: BL, basolateral amygdaloid nucleus; BM, basomedial amygdaloid nucleus; CL, central lateral thalamic nucleus; CM, central medial thalamic nucleus; CP, cerebral peduncle; GP, globus pallidus; HA, habenula; HF, hippocampal formation; LA, lateral nucleus of the amygdala; LG, lateral geniculate nucleus of the thalamus; LP, lateroposterior thalamic nucleus; OT, optic tract; PC, paracentral thalamic nucleus; PU, putamen; R, rhinal sulcus; RE, reticular thalamic nucleus; V, ventricle; VP, ventroposterior thalamic nucleus.

2 Coherent gamma oscillations couple the amygdala and striatum during learning. Popescu, Popa, Pare SUPPLEME TARY FIGURE 2 a 8 %.4 CTX Proportion of Cells.2 Spike Duration (ms) R = b Proportion of Cells % Spike Duration (ms) R = 43 STR c Proportion of Cells % Spike Duration (ms).9.6 BLA 2 R = d Proportion of Cells % Spike Duration (ms) R =.57 THL e Spike Duration 1 µv 1 ms Supplementary Fig. 2 Firing rates and spike durations in the cortical, striatal, BLA, and thalamic neurons included in the present study. Following spike sorting, we measured the baseline firing rate and average spike duration of each recorded unit. (a-d) Frequency distribution of firing rates in cortical (a, n = 159), striatal (b, n = 139), BLA (c, n = 152), and thalamic (d, n = 55) neurons. Insets in a-d plot spike duration (y-axis) as a function of firing rate (x-axis). As shown in e, spike duration was defined as the interval between the onset of the negative component of the spike to the peak of the subsequent positivity. Frequency distributions of firing rates revealed evidence of heterogeneity in our samples. Therefore, we ignored neurons whose firing rates fell outside the peak of the distributions.

3 Coherent gamma oscillations couple the amygdala and striatum during learning. Popescu, Popa, Pare SUPPLEME TARY FIGURE 3 a.5 THL BLA Coherence Frequency (Hz) b.5 CTX BLA.4 Coherence Frequency (Hz) Supplementary Fig. 3 Frequency dependence of the coherence between LFPs simultaneously recorded in the BLA and thalamus (a) or BLA and cortex (b). Plots of coherence (y-axis; average ± SEM) vs. frequency (Hz). Dashed lines indicate SEM.

4 Coherent gamma oscillations couple the amygdala and striatum during learning. Popescu, Popa, Pare SUPPLEME TARY FIGURE 4 Gamma Coherence BLA STR CTX STR THL STR BLA THL BLA CTX Distance between recording sites (mm) Supplementary Fig. 4 Impact of distance between recording sites on gamma coherence. Gamma coherence (y-axis; average ± SEM) plotted as a function of distance between recording sites (x-axis) for various combinations of recorded structures, as indicated in the legend shown in the upper right. BLA-striatal gamma coherence (solid red circles) is higher than seen with all other combinations of recorded structures, even when only considering pairs of recording sites separated by the same distance. SUPPLEME TARY FIGURE 5 a b Amplitude CTX/BLA BLA - CTX Offset (degrees) Amplitude THL/BLA 2 1 Normalized Counts BLA - THL Offset (degrees) Supplementary Fig. 5 Normalized frequency distributions of phase lags (x-axis) between BLA vs. cortical (a) or thalamic (b) gamma as a function of normalized gamma amplitude (left y-axis). The lines overlaid on these graphs represent the average gamma cycles seen at the corresponding sites (BLA, dashed lines; cortical and thalamic, solid lines).

5 Coherent gamma oscillations couple the amygdala and striatum during learning. Popescu, Popa, Pare SUPPLEME TARY FIGURE 6 Phase Offset (degrees) STR-BLA STR-CTX STR-THL ± SD (degrees) 45 BLA-CTX BLA-THL Distance between recording sites (mm) Supplementary Fig. 6 Impact of distance between recording sites on phase offset between gamma oscillations. Angle of arrows indicates average phase relation (see polar plot for legend). Origin of arrows indicates distance between recording sites (refer to x-axis). Length of arrows is inversely proportional to standard deviation of phase offset (the shorter the arrows, the higher the variability). Here, a perfectly fixed cycle-to-cycle phase relationship would have been ascribed a value of 1. The length of the arrows was computed using conventional vector averaging techniques (refer to red in polar plot for correspondence) SUPPLEME TARY FIGURE 7 Frequency (Hz) Frequency (Hz) muscimol (4 nm) Time (min) Relative Power (x 1 ) Supplementary Fig. 7 Intra-BLA muscimol infusions reduce striatal gamma power in individual cats. Striatal LFP power (color-coded) in different frequencies (y-axis) plotted as a function of time (x-axis) in experiments where muscimol was slowly infused in the BLA over a period of 25 min. Two examples of this test, performed in different cats, are shown.

6 Coherent gamma oscillations couple the amygdala and striatum during learning. Popescu, Popa, Pare SUPPLEME TARY FIGURE CTX 25 Degrees THL 25 Degrees Supplementary Fig. 8 Frequency distributions of firing peak times for thalamic (left) and cortical (right) neurons in relation to gamma activity picked up by the same electrode as that used for the unit recordings. SUPPLEME TARY FIGURE 9 Frequency (Hz) CTX - BLA THL - BLA Z-Score (SD) Z-Score (SD) < >2.5 < > Proportion of cells Correlation (Percentile) Correlation (Percentile) Frequency (Hz) Proportion of cells Supplementary Fig. 9 Same analysis as in figure 3i-k but for cortical and BLA (left) or thalamic and BLA (right) neurons. Color-coded frequency distributions of correlation indices of central 1 ms of crosscorrelograms (x-axis) plotted as a function the frequency of BLA LFPs (y-axis) used to select spikes included in the crosscorrelograms. The bottom x-axis expresses the data in percentiles. The correspondence in Z-scores can be found in the top x-axis.

7 Coherent gamma oscillations couple the amygdala and striatum during learning. Popescu, Popa, Pare SUPPLEME TARY FIGURE 1 a BLA-STR 4 Hz Relative Coherence (T1 T2) T1 = CS + T2 = CS r1 r2 r3 Recording Session (Day) b CTX-STR 4 Hz Relative Coherence (T1 T2) T1 = CS + T2 = CS r1 r2 r3 Recording Session (Day) c d THL-STR 4 Hz CTX-STR 7 Hz Relative Coherence (T1 T2) Relative Coherence (T1 T2) T1 = CS + T2 = CS r1 r2 r3 Recording Session (Day) T1 = CS + T2 = CS r1 r2 r3 Recording Session (Day) e THL-STR 5 Hz Relative Coherence (T1 T2) T1 = CS + T2 = CS r1 r2 r3 Recording Session (Day) Supplementary Fig. 1 Learning-related changes in correlated gamma are seen only between BLA and striatum, not between striatal and cortical or striatal and thalamic recording sites. For all panels, the y axis plots the difference in coherence elicited by the two tones (y-axis) as a function of recording sessions (x-axis). Panels a-c examines fluctuations in gamma coherence between striatal and BLA (a), striatal and cortical (b), or striatal and thalamic (c) recording sites. (d-e) Learning-related fluctuations in coherence of 7 Hz (d) or 5 Hz (e) LFP activity in cortical and striatal (d) or thalamic and striatal (e) recording sites, respectively. In all cases dashed gray lines indicate SEM.

8 SUPPLEME TARY METHODS Surgery Procedures were approved by the Institutional Animal Care and Use Committee of Rutgers State University, in compliance with the Guide for the Care and Use of Laboratory Animals (Department of Health and Human Services). Six adult male cats were pre-anesthetized with a mixture of ketamine and xylazine (15 and 2 mg/kg, i.m.) and artificially ventilated with a mixture of ambient air, oxygen, and isoflurane. Atropine (.5 mg/kg, i.m.) was administered to prevent secretions. The end-tidal CO 2 concentration was maintained at 3.7 ±.2 %, and the body temperature at o C using a heating pad. Bupivacaine (s.c.) was administered in the region to be incised 15 min prior to the first incision. In sterile conditions, an incision was performed on the midline of the scalp and the skull muscles were retracted. Then, a reference screw was inserted in the skull overlying the cerebellum and silver-ball electrodes were inserted in the supraorbital cavity to monitor eye movements. In addition, four screws were cemented to the skull to later fix the cat's head without pain or pressure. Finally, after trepanation and opening of the dura mater, an array of high-impedance tungsten microelectrodes (1-12 MΩ; Frederic Haer Co., Bowdoin, ME) was stereotaxically lowered to the regions of interest (see below). Finally, the animals were administered penicillin (2, UI/kg, IM) and an analgesic (Ketophen, 2 mg/kg, s.c., daily for 3 days). Recording sessions began eight days after the surgery. Recording sites and construction of microelectrode array

9 The microelectrode array included six electrodes aimed to the basal amygdala nuclei, eight electrodes aimed to the ventral striatum, eight electrodes aimed to primary or associative auditory cortical areas, and five electrodes aimed to rostral or posterior thalamic intralaminar nuclei. To construct the microelectrode array, a computer controlled milling machine was used to drill small holes in a Teflon block, at stereotaxically defined relative positions. Then, microelectrodes were inserted in the holes, adjusting the length of each microelectrode such that recordings could be obtained simultaneously from the various recording sites. After cementing the electrodes, the Teflon block was inserted in a tightly fitting Delrin sleeve, which was cemented to the skull. During the recording sessions, the electrodes could be lowered as a group by means of a micrometric screw. Recordings During the experiments, neuronal activity was sampled at 1 µm intervals. To insure mechanical stability, the microelectrodes were moved only once a day, 3 min prior to beginning data acquisition. The signals picked up by the electrodes ( Hz-2 khz) were observed on an oscilloscope, digitized, and stored on a hard disk. Spike sorting was performed off-line. Muscimol injections To assess the contribution of BLA activity to striatal gamma, we compared the effects of saline vs. muscimol infusions in the BLA on striatal gamma power. To this end, under isoflurane anesthesia and sterile conditions, two cats were implanted bilaterally

10 with stainless steel guide cannulas aimed at the rostro-caudal center of the BLA, under stereotaxic guidance. The cannulas were positioned at the dorsal limit of the BLA, allowing full dorsoventral access for drug infusions. In these cats, we also placed tungsten microelectrodes in the striatum, as described above. After a one-week recovery from the surgery, the animals were gradually adapted to head restraint. During this period, they had restricted access to food, as for the subjects participating in the learning task (see below) and were only fed while in the recording room. Once adapted to head restraint, recording sessions began with a 15 min baseline recording period, after which a total volume of 1 µl/hemisphere of saline or muscimol (4 µm in saline) was infused in the BLA at a rate of.8µl/min. To this end, a microsyringe with a 25-gauge needle was lowered through the guide cannula and the solution was pressure-injected at ten equidistant sites (.2 mm spacing) centered on the inner 2 mm of the BLA. The procedure was repeated for the contralateral side, and the recording continued for an additional 3 min. In both cats, two to three saline or muscimol infusions were performed on alternating days. Behavior Four cats that had restricted access to food (kept at 9% of their initial bodyweight) were trained on a stimulus-response task where the termination of one of two tones (CS+) coincided with the presentation of a liquid food reward (Gerber s pureed baby food Sweet potatoes and turkey ; 2 ml/trial). The food was available for only 1 s, and the animals quickly learned to lick during this interval, consuming the food in more than 9% of CS+ presentations (even during the first training session). The CS+ and CS

11 tones lasted 3 sec and were presented in a random order with 2-4 s inter-tone intervals. The identity of the CS+ and CS (3 or 12 khz) was varied systematically across cats and had no effect on learning progression. Each daily training session, around 6 CS+ and 6 CS trials were performed. Licking behavior was detected when the cats tongues interrupted an infrared beam. The animals were only fed during the recording sessions. As a result, they were aroused and remained awake at all times (as assessed by EEG recordings). The cats weight was monitored daily to maintain it within 1% of the initial value. After five consecutive training sessions, the CS-reward contingencies were reversed. That is, the initial CS+ became CS, and vice-versa. Three such reversal sessions (rd1-rd3 in Figs. 4-5 and supplementary figure 1) were recorded. Learning was assessed by monitoring the proportion of CS+ and CS presentations during which anticipatory licking occurred. In addition, using the 3 sec windows preceding tone onsets, we computed the proportion of trials with spontaneous licking for comparison with tone-evoked behavior. Histology At the end of the experiments, recording sites were marked with electrolytic lesions (.5 ma, 5-1 s). The animals were then given an overdose of sodium pentobarbital (5 mg/kg, i.v.) and perfused-fixed. The brains were later sectioned on a vibrating microtome (at 1 µm) and stained with cresyl violet to verify the position of recording electrodes. Microelectrode tracks were reconstructed by combining micrometer readings with the histology.

12 Data Analysis Spike sorting. Data was analyzed offline with custom software written in Matlab 7.1 (The MathWorks Inc., Natick, MA). Spike-sorting was performed on digitally filtered data (high pass filter > 15 Hz), using principal component analysis and a supervised k- means clustering algorithm. LFP analyses. To analyze LFP interactions in specific frequency bands, the raw data was filtered with a bandpass filter (width of 1 Hz), centered on the required frequency (e.g. results presented for 4 Hz correspond to data filtered in a Hz band). For comparison purposes, all 2D and 3D histograms were normalized to the total number of events used, emphasizing their relative distribution rather than the absolute numbers. Assessing the modulation of unit activity by gamma. To rule out the possibility that the gamma activity seen in the LFPs was volume conducted from a nearby structure, we computed peri-event histograms (PEHs) of unit activity around the peaks of positive gamma cycles the average plus 2.5 SD of the overall signal. In this case, the unit activity and LFPs were picked up by the same microelectrode. To eliminate the possibility that digital filtering at the gamma frequency introduced artefactual gamma activity, prior to filtering, the LFPs were down-sampled (from 1 point every.5 ms to 2 ms). This approach (as opposed to band-pass filtering the raw signal) abolishes most fast transients (like spikes) and greatly reduces the likelihood that spikes introduce artefactual gamma as a result of digital filtering. To determine whether the gamma modulations of unit activity seen in the PEHs were statistically significant, we computed a rhythmicity

13 index (RI) for each cell. The RI was obtained by averaging the difference in spike counts between the three center peaks and troughs of the PEHs and dividing the result by the average of the entire histogram to normalize for variations in firing rates. Statistical significance of the RI was tested by recomputing the peri-event histograms after shuffling the spike times, repeating this process 1 times. The actual RI was considered significant if it was higher than 95% of the randomly-generated RIs. Assessing gamma-related changes in unit coupling. The goal of these analyses is to test whether the occurrence of high amplitude gamma increased unit coupling in the BLA and striatum, as compared to when all spikes were crosscorrelated. To this end, for all BLA-striatal cell couples, we computed two crosscorrelograms of unit activity. The first included all spikes the cells generated. These correlograms were typically flat. The second kind of crosscorrelogram focused on periods when there was high amplitude gamma in the striatum. To compute these, we first searched striatal LFPs for periods of high amplitude gamma. Then, we crosscorrelated the unit activity taking place during these periods by ignoring all striatal spikes that did not coincide with high amplitude gamma (but including all BLA spikes). By comparing the two types of crosscorrelograms, one can assess whether unit coupling is altered during periods of high amplitude striatal gamma. We repeated this analysis for all cell couples. Each cell couple was assigned a correlation index (by comparing the sum of the central ± 5ms bins of the correlograms to that of the correlograms generated after shuffling of the BLA spike trains). The correlation indexes were then rank ordered from the lowest to the highest and expressed as percentiles. To test whether the gamma-related enhancement in BLA-striatal unit coupling where specific to gamma activity, the same analysis was repeated for all

14 frequency bands (in bins of 5 Hz). Color-coded frequency distributions of the correlation indices were then plotted for all frequencies. Such distributions were typically characterized by a strong band at around.5 z-scores, corresponding to the median z- score. The median z-scores were not zero because the frequency distributions were skewed to the right. The cell couples with central peaks falling at the high end (right) of the distributions are those with significant correlations. Learning-related fluctuations in gamma coherence. To study learning-related fluctuations in BLA-striatal coherence, power in the particular frequency band under consideration was calculated in one-second windows (sliding in 1 ms steps) around the onset of the two tones for the two recording sites. Coherence was estimated by computing the product of the powers for the two recoding sites for each one-second time window. Three factors must be taken into account when interpreting the results of these analyses. The first two relate to the physiology of the network under study. The third is methodological in nature. One factor is that from CS onset to the arrival of auditory impulses in the amygdala and striatum, is a short conduction delay, on the order of 2 ms in cats. A second factor is that before the increase in gamma coherence can be detected, BLA and striatal neurons must synchronize and this process necessarily requires a few gamma cycles. Third, reliable coherence measurements require that the epochs analyzed have a minimal duration to avoid spurious variability. By trial and error, we determined that a window duration of 1 ms (sliding in steps of 1 ms) was the minimal duration to allow reliable coherence measurements while preserving an adequate temporal resolution. This means that the first data point during the CS includes the 5 ms immediately before the CS and the 5 ms after CS onset. This temporal smearing tends

15 to reduce the amplitude of the initial CS-evoked changes in gamma coherence and cause a delay in its onset. Statistical analyses consisted of repeated measures ANOVAs followed by Bonferonni-corrected t-tests. All values are reported as average ± SEM.