Motivation Noninvasive Scoring of Mouse Sleep and Behavior Farid Yaghouby 1 , Kevin Donohue 2, Bruce O’Hara 3 , Sridhar Sunderam 1* (1) Center for Biomedical Engineering; (2) Electrical and Computer Engineering; (3) Department of Biology Experimental Setup: The Piezo System Conclusions Measures of broadband activity and breathing regularity derived from a piezoelectric sensor signal show potential for noninvasive automated scoring of sleep stage and awake behavior. The percent time spent in each states and typical bout duration mirror values reported in literature based on EEG analysis. The next step is to correlate the states identified with simultaneous EEG scoring of states of vigilance. Successful noninvasive scoring of sleep and behavior will open up new avenues for high throughput analysis of sleep phenotypes and alleviate the need for EEG measurements. References 1. Donohue KD, Medonza DC, Crane ER, O’Hara BF 2008. Assessment of non-invasive high-throughput classifier for behaviors associated with sleep and wake in mice. Biomed Eng Online. 11;7-14. 2. Flores AE, Flores JE, Deshpande H, Picazo JA, Xie XS, Franken P, Heller HC, Grahn DA, O’Hara BF 2007. Pattern recognition of sleep in rodents using piezoelectric signals generated by gross body movements. IEEE Trans Biomed Eng. 54(2):225-33. 3. Friedman L, Haines A, Klann K, Gallaugher L, Salibra L, Han F, Strohl KP 2004. Ventilatory behavior during sleep among A/J and C57BL/6J mouse strains. J Appl Physiol. 97(5):1787-95. 4. Diniz Behn CG, Klerman EB, Mochizuki T, Lin Shih-Chieh, Scammell TE. Abnormal sleep/wake dynamics in orexin knockout mice. Sleep 2010; 33(3):297-306. 5. Zhang S, Zeitzer JM, Sakurai T, Nishino S, Mignot E. Sleep/wake fragmentation disrupts metabolism in a mouse model of epilepsy. J Physiol. 2007; 581.2:649-63. 6. McShane BB, Galante RJ, Jensen ST, Naidoo N, Pack AI, Wyner A 2010. Characterization of the bout durations of sleep and wakefulness. J Neurosci Methods. 193:321-333. Figure 1. Quad cage piezo system [1]. (Left) Cage walls on sensors on base; (Right) Sensor pads on cage floor connected to amplifiers. This noninvasive activity monitoring system was developed by Bruce O’Hara and colleagues for phenotyping sleep in mice. A piezoelectric sensor is inserted at the base of the animal’s cage. This pressure-sensitive device generates a signal that reflects gross and fine movements that typify behavior. Piezo Signal Features Track Changes in Sleep-Wake State Analysis of Bout Durations Identification of genes that contribute to normal and abnormal sleep and wake behaviors would improve our understanding of the mechanisms and functions of these states and suggest new pathways or approaches for the treatment of related disorders. Genetic approaches such as mutagenesis, quantitative trait locus analysis and knockout mice can help identify genes that determine patterns of sleep and circadian rhythms—but screening the relevant phenotypes requires expensive and labor-intensive animal experimentation with EEG/EMG analysis, not just simple actigraphy (i.e., activity monitoring). This severely limits the scale of behavioral experimentation. Our goal is to develop a noninvasive method for scoring stages of sleep and behavior in mice. Figure 4. Teager energy (green) of piezo signal tracks sleep-wake changes in muscle tone (green). Mean Teager energy (TE) computed in 4s epochs is compared to EMG amplitude in a continuous 24h recording. TE provides a broadband measure of the instantaneous power in the piezo signal and appears closely correlated with trends in the EMG. This indicates that the peizo signal can be used to noninvasively determine sleep-wake state (black trace). Figure 5. Comparison of breath variability (red) with EEG theta (4-8 Hz) band power (green). The breathing interval was estimated from the peaks in the piezo signal, and the variance of the inter-breath interval in a 4s epoch used to measure breath variability (BV). Surges in theta power during REM are mirrored to some extent by BV, suggesting that this measure could discriminate between REM and non- REM stages of sleep. However, BV also appears to track brief arousals and awake behavior, which could be discriminated by the Teager energy measure shown in Figure 4. **We are grateful to Paul Franken and Geraldine Mang of the University of Lausanne for providing us with these simultaneous piezo-EEG recordings and visual scores of sleep-wake state in CFW mice.** Figure 2. Peizo signal and respiration [2]. The “piezo” signal can discriminate sleep from wake states with over 90% accuracy. Simultaneous measurement of respiratory effort using an impedance pneumogram shows that changes in breathing can be detected when the mouse is relatively inactive. Figure 3. Can breathing variability be used to detect transitions between REM and NREM sleep? It is well known that breathing is highly regular during non-REM sleep but becomes variable in REM sleep [3]. Pressure changes on the piezo sensor associated with respiratory activity may have signatures characteristic of different stages of sleep as well. Here we see that a regular pattern of breathing in NREM sleep is disturbed by transitions to REM and WAKE states. NREM REM QW AW AW QW REM NREM Figure 6. Mean trends in piezo signal features during NREM-REM (Left) and REM-WAKE (Right) transitions in a CFW mouse. Teager Energy (Top) increases during NREM- REM transitions, which are usually followed by REM-Wake transitions in which there is a further increase in value. Breathing Variability (Bottom), which is independent of signal amplitude, displays similar tendencies. REM- WAKE transitions usually result in a brief arousal, which is why the feature values quickly decrease to NREM levels thereafter. Dashed lines indicate 95% confidence limits estimated from about 50-60 samples in a 24 h period. Figure 7. Distribution of bout duration (in seconds) of different states determined through piezo signal analysis from 4 days of continuous recording in a C57BL/6J mouse. Bout durations for NREM, NREM, QW, and AW were compared using Kruskal-Wallis ANOVA. All four states were significantly different (p<0.05) in the light period (Left) with NREM > QW > AW > REM. In the dark period (Right), however, only REM was significantly different (p<0.05) from the other states and shorter in bout duration. Bout durations of all states compared well with values cited in the literature based on EEG analysis in C57BL/6J mice [6]. Figure 8. Distribution of time spent in NREM, REM, and WAKE states in CFW mice (n=10). A comparison of % time estimated using the piezo classifier with EEG scoring shows that sleep- wake ratios can be estimated with high accuracy. However, NREM sleep is underestimated and REM sleep is overestimated. This may be attributed to the nonspecificity of irregular breathing to REM sleep and possible differences in NREM between light and slow wave sleep, which were not scored. -50 0 50 100 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Time (s) Breath variability NREM-REM Transition -50 0 50 100 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Time (s) Breath Variability REM-WAKE Transition -50 0 50 100 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 Time (s) Teager Energy NREM-REM Transition -50 0 50 100 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 Time (s) Teager Energy REM-WAKE Transition 0 5 10 15 20 Teager energy (green) of piezo signal tracks sleep-wake changes in muscle tone (blue) Time (h) 11.2 11.3 11.4 11.5 11.6 11.7 11.8 11.9 12 Time (h) WAKE NREM REM WAKE NREM REM NREM REM REM WAKE -10 0 10 20 30 40 50 Time (s) Samples of piezo signal changes time-locked to NREM-REM transitions (at Time = 0 s). Wake NREM REM 0 10 20 30 40 50 Percent time spent in each state: comparison of piezo classifier with EEG. Percent time EEG Piezo Wake NREM REM 20 30 40 50 60 70 80 90 100 True Positive (%) Classification using piezo signal features: Performance evaluation (n = 10 mice). Wake NREM REM 20 30 40 50 60 70 80 90 100 True Negative (%)