Skip to main content
. 2020 Oct 29;9:e62071. doi: 10.7554/eLife.62071

Figure 2. Mice rapidly and repeatedly transition between wake and sleep during head-fixation.

(A) Hypnogram showing the arousal states for a single mouse over six days. The hypnogram has a resolution of 5 s. White denotes breaks in data acquisition for saving of data. Note the rapid and frequent transitions between rfc-Awake, rfc-NREM and rfc-REM. (B-I) n = 14 mice. (B) Average percentage of the time spent in each arousal state. (C) Ternary plot showing each individual animal’s percentage in each arousal state. (D) Average probability of an animal being classified in a given arousal state as a function of time since the start of the session. Mice are progressively more likely to sleep and to be in REM sleep the longer they have been head-fixed E, Average probability of the animal being awake as a function of the duration of the period without movement. Mice are more likely to be asleep the longer they go without moving their whiskers or body. (F) Probability distribution of the mean EMG power during individual arousal states (5 s resolution) taken from all animals. (G) Probability distribution of variance in the whisker angle during individual arousal states (5 s resolution) taken from all animals. (H) Probability distribution of the mean heart rate for each arousal state. (I) Mean heart rate during different arousal states. Circles represent individual mice and diamonds represent population averages ± 1 standard deviation. *p<0.05, **p<0.01, ***p<0.001 GLME.

Figure 2.

Figure 2—figure supplement 1. Behavioral measurements demarcate transitions between arousal states.

Figure 2—figure supplement 1.

(A) Power in rfc-Awake electromyograph (EMG) recordings (1.3 ± 1.3 from the nuchal muscles were substantially smaller during rfc-NREM sleep (0.25 ± 2), GLME, p<3.2 × 10−12) and during rfc-REM sleep (0.05 ± 1.6, GLME, p<1.7 × 10−21) in comparison to the rfc-Awake state. (B) Variance in the whisker angle during rfc-NREM (2.4 ± 1.5 deg2, GLME, p<1.3 × 10−12) was significantly less than that of the awake state (25.7 ± 9 deg2). Whisker angle variance during rfc-REM (18.6 ± 7.3 deg2, GLME, p<0.003), though statistically different, was much more similar to the awake state due to mice sporadically moving their whiskers during rfc-REM sleep, analogous to rapid-eye movement seen in humans. (C) Heart rate during the rfc-Awake state was 7.5 ± 0.7 Hz. During rfc-NREM sleep, the heart rate dropped to 6.1 ± 0.6 Hz (GLME, p<1.8 × 10−13), and was elevated slightly during rfc-REM to 7.1 ± 0.5 Hz (GLME, p<0.004) (n = 14 mice). *p<0.05, **p<0.01, ***p<0.001 GLME.
Figure 2—figure supplement 2. Random forest model validation.

Figure 2—figure supplement 2.

All data from the first and last day of imaging from each animal was manually scored as rfc-Awake, rfc-NREM, or rfc-REM. Alternating 15 min periods of data from these two days were divided into two discrete sets: one for model training, the other is held back for model validation beyond the out-of-bag error obtained from the training data set. A confusion matrix containing each IOS animal’s (n = 14) random forest model predictions of the held back, second data set compared to its manual scores are presented in (A). The total model accuracy across all 14 animals was 91.3%, with the most accurate predictions coming from the most prevalent classification class (rfc-Awake), followed by rfc-NREM and then rfc-REM.