Fig. 2: Properties of internal oscillation states identified from local field potentials in awake behaving mice.

A Top: Local field potential (LFP) power modulations in V1 recorded from mice passively viewing a naturalistic movie. Bottom: Time course of running speed and pupil area during the same time period. B Schematic to identify oscillation states using LFPs. Discrete states are defined based on frequency-specific transients of LFPs from six visual areas. Hidden Markov model (HMM) uses Hilbert transforms in the theta (3–8 Hz), beta (10–30 Hz), lower gamma (30–50 Hz), and higher gamma (50–80 Hz) frequency ranges. C Top: Model comparison among HMMs over a range of latent states using three-fold cross-validation. The cross-validated log-likelihood (LL) estimate, normalized by the top eigenvalue of the state definition matrix, is reported for each mouse (hollow circles) along with across-subject averages (solid circles, n = 25 mice, error bars represent s.e.m). For each mouse, the optimal number of states was identified as the point where the normalized LL was maximized. Final model selection was based on the majority rule across all mice.  Bottom: Evaluation of state similarity (λ₁) as the top eigenvalue of the state definition matrix. D Top: State posterior probabilities identified by the HMM. Bottom: LFPs from randomly selected channels from V1, displayed alongside their respective latent states over the same duration. E LFP power distribution in the three-state model. Shaded lines represent the state-specific z-scored power distributions in individual mice, and the solid black line represent the average across all mice (N = 25 mice). In state-1, or the high-frequency state, LFPs are dominated by high-frequency gamma oscillations. State 3, or the low-frequency state, has characteristic slow oscillations in the theta band. F Histogram of state dwell times in each trial across all states and all mice. G Average probability of observing 3-step or 2-step (inset) transition sequences to different states. Transition probabilities were calculated from observed sequences averaged across all mice (n = 25, error bars represent s.e.m). Source data are provided as a Source Data file.