Fig. 4: Neuronal variability and information encoding across states and the visual hierarchy.

A Raster plots (~10 s) showing the response of 25 units, each from V1 and AM, during two trials in which the mouse was in different states. Each row represents the activity of the same single neuron across the two trials. S_H: High-frequency state (green), S_I: Intermediate state (blue), and S_L: Low-frequency state (pink). B State and area-specific population activity, z-scored and averaged across all mice (\({p}_{{S}_{H},{S}_{I}}\) = 1.4e-05, \({p}_{{S}_{H},{S}_{L}}\) = 3.0e-07, \({p}_{{S}_{I},{S}_{L}}\) = 0.90, one-way ANOVA, n = 25 mice). Error bars represent s.e.m. C Average pairwise correlation between averaged neuronal population activity in different visual areas as a function of oscillation states (\({p}_{{S}_{H},{S}_{I}}\)  = 1.5, \({p}_{{S}_{H},{S}_{L}}\) = 0.002, \({p}_{{S}_{I},{S}_{L}}\) = 0.002, one-way ANOVA, n = 25). Error bars represent s.e.m. D Population shared variance. Top: Separation of shared and independent variance using factor analysis (FA). FA partitions the spike count covariance matrix into shared and independent components. Bottom: Percentage of shared variance plotted against the anatomical hierarchy scores of the visual areas in each oscillation state, averaged across all units (One-way ANOVA: \({p}_{{S}_{H},{S}_{I}}\) = 9.6e-7, \({p}_{{S}_{H},{S}_{L}}\) = 1.3e-146, \({p}_{{S}_{I},{S}_{L}}\) = 1.2e-95; Two-way ANOVA, states: F   431.2, p = 1.5e − 189, areas: F = 78.8, p = 3.3e − 82, states  × area: F = 3.3, p = 2.6e − 4, n = 7609 units). E Neuronal variability across time, quantified using the coefficient of variation (CV). Top-left: Simulated distributions of inter-spike-intervals (ISI) for regular and Poisson-like firing. For a very regular spike train, a narrow peak in the ISI histogram corresponds to CV ≈ 0, whereas Poisson-like variability in the spike trains leads to an exponentially distributed ISI histogram with CV = 1. Top-right: Distribution of ISIs in each oscillation state over a 2.5 sec range. Bottom: CV along the visual hierarchy (quantified as anatomical hierarchy scores) and across oscillation states, averaged across all units (One-way ANOVA: \({p}_{{S}_{H},{S}_{I}}\) = 4.9e-23, \({p}_{{S}_{H},{S}_{L}}\) = 3.9-03, \({p}_{{S}_{I},{S}_{L}}\) = 2.8e-11; Two-way ANOVA, states: F = 42.5, p = 3.6e − 19, areas: F = 88.1, p = 4.5e − 92, states × area: F = 4.8, p = 4.9e − 7, n = 7609 units). F Neuronal variability across trials, quantified using Fano factor (FF). Top-left: Evaluation of FF as an average of the FF ratio over non-overlapping windows of 150 ms with at least ten trials in each state. Top-right: Mean spike count versus variance over all times in each state for an example cell in V1. Bottom: FF along the visual hierarchy and across brain states, averaged across all units (One-way ANOVA: \({p}_{{S}_{H},{S}_{I}}\) = 2.8e-4, \({p}_{{S}_{H},{S}_{L}}\) = 2.4e-33, \({p}_{{S}_{I},{S}_{L}}\) = 3.5e-39; Two-way ANOVA, states: F = 107.7, p = 7.5e − 47, areas: F = 7.1, p = 9.9e − 6, states  × area: F = 0.6, p = 0.8, n = 5017 units). Pearson correlation with hierarchy scores excluding RL, S_H: r_p−RL = − 0.94, p_p−RL = 0.02; S_I: r_p−RL = −0.43, p_p−RL = 0.5; S_L: r_p−RL = − 0.46, p_p−RL = 0.43. G Information encoding along the visual hierarchy across all oscillation states, quantified using mutual information (MI). Top: For each trial, MI was evaluated between the population spike count matrix and a matrix of flattened movie frames at time points corresponding to each state using a matrix-based entropy estimator. Bottom: MI across the visual hierarchy and oscillation states averaged across all mice (Pairwise T-test: \({p}_{{S}_{H},{S}_{I}}\) = 0.01, \({p}_{{S}_{H},{S}_{L}}\) = 7.3e-10, \({p}_{{S}_{I},{S}_{L}}\) = 9.3e-04; Two-way ANOVA, states: F = 3.1, p = 0.04, areas: F = 2.7, p = 0.03, states  × area: F = 0.02, p = 0.99, n = 25). Pearson correlation with hierarchy scores excluding RL, S_H: r_p−RL = − 0.9, p_p−RL = 0.03; S_I: r_p−RL = − 0.86, p_p−RL = 0.06; S_L: r_p−RL = − 0.81, p_p−RL = 0.09. Error bars in D–G represent 95% confidence intervals. All statistical tests were adjusted for multiple comparisons using the Bonferroni correction (***: p < 0.0001, **: p < 0.001, *: p < 0.05). Source data are provided as a Source Data file.