WM task, power spectrum, phase code and gain modulation.

(A) The animal performed an MGS task in which visual probes appeared during the 1-s fixation period and the 1-s delay period. The probe (white circle) was a brief (200 ms) small (∼1 dva) visual stimulus presented in a 7 × 7 grid of possible locations (open white circles, shown here for illustration only and not present on the screen). In each trial, four probes were presented in succession during both the fixation and memory periods, with an inter-probe interval of 200 ms. This 7 × 7 grid of probes was positioned to overlap with the RF of the recorded neuron based on the preliminary RF mapping. The location of the remembered target could vary with respect to the RF of recorded neurons. (B) The receptive field of a sample MT neuron. Plot shows firing rate in response to probes appearing at various location. (C) Power change (relative to pre-stimulus baseline) over time and frequencies for the sample recording channel as the neuron shown in B. (D) Scatter plot of mean power change in αβ band across 256 recorded channels for memory IN and OUT condition. The diagonal histogram shows the distribution of differences in power. (E) Across 131 MT neurons, the mutual information between the 49 different probe stimuli and the phases of ongoing αβ oscillation at the time of spikes showed a significant increase during memory IN compared with memory OUT. The scatterplot shows the MI for each MT neuron during memory OUT (x axis) and during memory IN (y axis). The diagonal histogram shows the distribution of differences in MI. (F) Visually evoked firing rates increased during the memory IN period. The scatterplot shows visually evoked spiking activity for the optimal probe of each neuron in the memory IN versus memory OUT. The diagonal histogram shows the distribution of changes in the visually evoked spiking activity.

Analysis of Peak Frequencies, Decomposition Method, and Component Frequencies.

(A) The original power spectral density of a single trial (black solid line), fitted power spectral density (red solid line), and aperiodic component of the power spectral density (dashed blue line). (B) The top panel shows a histogram of the number of peak frequencies in trials across all recording sessions, while the bottom panel displays the detected frequencies. (C) Schematic Diagram of a Five-Level MODWT Method. This schematic illustrates the hierarchical decomposition process of a time series signal using MODWT, depicting five successive levels of transformation. Each level shows the corresponding wavelet and scaling coefficients, highlighting how the signal is progressively decomposed into finer frequency components and its approximation. (D) Decomposition of a single trial using MODWT. The first subplot (red line) shows the original signal, and the remaining subplots display the decomposed components, with the peak frequency of each component indicated in the title of each subplot. (E) Scatter plot of the frequency of each of the six components for memory IN and OUT conditions across 256 recording channels.

Power of components, frequency-dependent gain modulation, and regression model for the sample neuron.

(A) Power differences between IN and OUT conditions during the memory period (2400–3000 ms) for the sample neuron shown in Figure 1B. Each subplot represents one component (delta, theta, alpha, beta, low gamma, and high gamma), illustrating the power of the signal across frequency and time. The plotted values reflect the subtraction of power in the memory IN condition from the memory OUT condition. (B) Averaged power across frequencies for different components during the memory period (2400–3000 ms) for the same sample neuron. The plots depict the power differences of the decomposed components over time, calculated as the power in the memory IN condition subtracted from the memory OUT condition. Solid lines on the bottom indicate significant changes in power during that time bin for the corresponding frequency. In B-D, shading shows the standard error of the mean across trials. (C) Firing rate as a function of ranked probes based on responses during the memory IN condition for the sample recording channel shown in Figure 1B. Each subplot corresponds to one component and illustrates firing rate changes across different frequency groups. Here and in D, frequencies are categorized into high-frequency (red line), middle-frequency (black line), and low-frequency (blue line) groups. (D) Linear regression models fitted to the data presented in Figure 3C. The fitted models illustrate the relationship between firing rate changes and probe optimality for different frequency groups across each decomposed component.

Power of components and frequency-dependent gain modulation across the population.

(A) Power differences between IN and OUT conditions during the memory period (2400–3000 ms) across 256 recording channels. Each subplot represents one component (delta, theta, alpha, beta, low gamma, and high gamma), illustrating the power of the signal across frequency and time. The plotted values reflect the subtraction of power in the memory IN condition from the memory OUT condition. (B) Averaged power across frequencies for different components during the memory period (2400–3000 ms) across 256 recording channels. The plots depict the power differences of the decomposed components over time, calculated as the power in the memory IN condition subtracted from the memory OUT condition. Time bins with significant differences from baseline (0-50ms) are indicated by solid lines of the corresponding color below. In B-F, shading shows the standard error of the mean across trials. (C) Averaged firing rate as a function of ranked probes based on responses across 131 neurons during the memory IN condition. Each subplot represents one component and depicts the averaged firing rate changes across different frequency groups. In C-F, frequencies are categorized into high-frequency (red line), middle-frequency (black line), and low-frequency (blue line) groups. (D) Plots same as C, for the memory OUT condition. (E) Averaged linear regression models fitted to the firing rate data of 131 neurons during the memory IN period. The models depict the relationship between firing rate changes and probe optimality for different frequency groups (high, middle, and low) across each decomposed component. (F) Plots same as E, for the memory OUT condition.

Regression model and statistical analysis.

(A) Averaged t-statistics of the linear regression models (n = 131) for the memory IN condition, highlighting the contribution of each factor to firing rate changes across components. These factors include frequency (top panel), probe optimality (middle panel), and the interaction between frequency and probe optimality (bottom panel). The averaged t-statistics illustrate the significance of these factors in modulating firing rates across different frequency bands (delta, theta, alpha, beta, low gamma, and high gamma). (B) Same as A for the memory OUT condition. (C) The number of neurons for which the contributions of each factor—frequency (top panel), probe optimality (middle panel), and the interaction between both factors (bottom panel)—were statistically significant (p < 0.05) in modulating firing rate changes during the memory IN condition. (D) Same as C for the memory OUT condition. (E) Scatter plot showing the slopes of the fitted linear regression models to firing rate changes of 131 neurons for each frequency group and component, for the IN vs. OUT conditions. Red dots represent high frequency, black dots represent middle frequency, and blue dots represent low frequency.