| The schematic of a model for theta sequence development

Fast gamma rhythms coordinate a subgroup of place cells, as their spikes are dominantly phase-locked to fast gamma rhythms (FG-cells). The development process of theta sequences is disrupted by excluding these FG-cells. A possible model would be that development of sweep-ahead structure of theta sequences is dependent on the slow gamma modulation, i.e., slow gamma phase precession, whereby spatial information could be highly compressed within theta cycles. FG-cells are those cells exhibiting slow gamma phase precession within theta cycles since from the early stage of sequence development. However, the NFG-cells only excited slow gamma phase locking during late stage, which may unlikely contribute to the sequence development.

| A subset of hippocampal place cells was modulated by fast gamma rhythms during active exploration.

(A) An example of simultaneously recorded LFPs (raw LFP; theta: 4-12Hz; slow gamma: 25-45Hz; fast gamma: 65-100 Hz) and spikes of place cells. The troughs of theta were represented by dark gray vertical lines. Slow and fast gamma episodes were marked by dashed boxes and gray blocks. (B) Firing characteristics of two examples of FG-cells and NFG-cells, including turning curve, fast and slow gamma phase distribution during detected episodes. Spikes were represented by red rasters on top of the gamma traces. Mean vector length of each cell was shown above the gamma phase distribution. (C) The normalized turning curves of place cells sorted by the center of mass of their main place field. The proportion of 4 types of cells identified according to different modulation by fast and slow gamma was shown on the right. (D) Mean vector length of fast gamma phase distribution of FG-cells and NFG-cells across successive running laps (n=12 sessions). Data are presented as mean ± SEM. ***p<0.001.

| Theta sequences development was disrupted without FG-cells.

(A) Top, An example of color-coded decoded probability with time during a single lap. The running trajectory of the animal was indicated as white dashed line. Bottom, 4– 12 Hz bandpass filtered theta rhythm, with cycles divided on trough indicated as red bars. (B) Examples of detected theta sequences decoded by 3 different decoders, all spikes from all cells (top, i.e. raw sequences), all spikes excluding those from FG-cells (middle, i.e. exFG-sequences) and all spikes excluding those from NFG-cells (bottom, i.e. exNFG-sequences). The white triangle indicates the animal’s current position. The white boxes indicate the center of sequences for calculating weighted correlation. (C) Weighted correlations of sequences among three types of decoders (n=12 session). (D) Averaged decoded probabilities for each lap, at a range of the animal’s current position +-63 cm and the mid-time point of theta sequence +-170 ms, of a single recording session. (E) Weighted correlation of raw sequences was significantly increased with running laps (n=12 sessions). (F) The running speeds were maintained across laps. (G) The sweep-ahead structure of the exFG-sequences was disrupted compared to that of exNFG-sequences (n=12 sessions). (H) The effect of excluding either FG-cells or NFG-cells at early (left) and late (right) development of theta sequences (n=24 laps). Data are presented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.001.

| Temporospatially compressed structure of theta sequences required a relatively large proportion of FG-cells.

(A) Examples of detected theta sequences in early development. Each column is a pair of theta sequences decoded by 2 different decoders, i.e. exFG-decoder and exNFG-decoder, in a same theta cycle. The left 2 columns of sequences were detected from laps with a relatively small proportion (∼10%) of FG-cells. The right 2 columns of sequences were detected from laps with a relatively large proportion (∼35%) of FG-cells. Weighted correlation (W.Corr) of each sequence was shown. (B) Same as (A), but for late development of theta sequences. (C) The difference of weighted correlation between exFGsequences and exNFGsequences as a function of the proportion of excluded cells, in early development. Each scatter represents data from each recording session (n=12 sessions). The dashed line is the linear regression line. (D) Same as (C), but for late development of theta sequences.

| Comparison of FG-cells and NFG-cells firing characteristics.

(A) Spatial tuning curves of representative FG-and NFG-cells during exploration on the circular track. Each column of cells exhibited similar place field positions on the track. (B) Distribution of the place field centers of mass (COM) of FG-cells and NFG-cells as a function of position on the track (n=113 cells for FG-cells and downsampled n=113 cells for NFG-cells). (C) Cumulative distribution of FG-cells and NFG-cells’ place field COM. (D) Mean firing rate, (E) Excluded spike counts in decoding, (F) Place field size, and (G) Spatial information of FG-cells and NFG-cells (n=12 sessions). Data are presented as mean ± SEM. *p<0.05

| Slow gamma phase precession during early and late development of theta sequences.

(A) Probability distributions of slow gamma phases of spikes across successive slow gamma cycles within theta sequences. Gamma cycles within theta cycles were ordered, centered at cycle 0 (i.e., gamma cycle with maximal spiking). Slow gamma phases of spikes shifted systematically across successive slow gamma cycles. (B) same as(A) but during early development of theta sequences. Spikes’ slow gamma phase did not significantly backward-shifted across first 2 slow gamma cycles. (C) Same as (A), but during late development of theta sequences. The white dots denote the peak probability of the histogram.

| FG-cells constantly exhibited slow gamma phase precession across laps.

(A) Examples of slow gamma phase precession within theta cycles on single cell level. The top row shows posterior probability for theta sequences. The middle row shows the LFPs waveform in the same theta cycle of the theta sequence, with each slow gamma cycle divided by black lines. The bottom row shows the slow gamma phases of spikes from a representative cell activated in this theta cycle. The spikes occurred at earlier slow gamma phase in the late slow gamma cycles. (B) The averaged slow gamma phase shift across N active cycles were significantly negative (N=2, 3 or ≥4). The red lines indicate the median and the black lines indicate the 25% and 75% quantile of group data (n=12sessions for N=2 and N=3, n=9sessions for N≥4). (C) Histogram of mean slow gamma phase shift of theta cycles across laps. The histograms of real data are shown in blue with blue triangle indicating median. The histograms of mock data are shown in grey with black lines indicating the 95th confident interval of the their median. (D) The same as (C) but for histogram of mean slow gamma phase shift of place cells across laps. (E) Histogram of mean slow gamma phase shift of FG-cells (red) or NFG-cells (yellow) across laps. The red and yellow triangles are the median of phase shift in FG-cells and NFG-cells respectively. *p<0.05, **p<0.01, ***p<0.001.

| The sweep-ahead structure of FG-cell sequences was positively correlated with the intensity of slow gamma phase precession during both early and late sequence development.

(A) Scatter plot of averaged slow gamma phase shift as a function of weighted correlation of sequences during all laps. Left panel shows real data (blue) and right panel shows mock data (gray). (B) Probability distributions of sequences falling in 4 quadrants (Q1-Q4) with slow gamma phase shift (<0 or >0) and weighted correlation (<0 or >0). (C) The count of sequences falling in Q1 (blue triangle) was significantly higher than 95% quantile of the shuffled data (black line). The gray histogram indicates the relative probability of mock data falling in Q1 by 1000 times shuffling. (D) Scatter plot of averaged slow gamma phase shift as a function of weighted correlation of FG-cell sequences during early development (top) and late development (bottom). (E) Probability distributions of FG-cell sequences falling in 4 quadrants during early development (top) and late development (bottom). (F) The count of FG-cell sequences falling in Q1 (red triangle) was significantly higher than 95% quantile of the shuffled data (black line) during both early development (top) and late development (bottom) of sequences. (G-H) Same as (D-E), but for NFG-cell sequences. (I) The count of NFG-cell sequences falling in Q1 (red triangle) was significantly higher than 95% quantile of the shuffled data (black line) only during late (bottom) but not early (top) development of sequences. *p<0.05, ***p<0.001.