Zebrafish sleep displays distinct sub-states

Abstract

Sleep is an essential and evolutionarily-conserved behavior. While mammals and several other species have been shown to exhibit well-defined sleep sub-states, some of which have been ascribed specific functions, it remains unclear to what extent such differentiation exists across the animal kingdom. Here we show, using long-term behavioral data combined with Hidden Markov Modeling, that larval zebrafish display distinct deep and light sleep sub-states. Although both states occur primarily at night, fish respond differently to sleep deprivation and arousing stimuli depending on which sleep sub-state they are in. Moreover, the proportions of deep and light sleep are selectively altered by genetic and pharmacological manipulations of melatonin, serotonin, and norepinephrine signaling, offering new insights into how these neuromodulators shape sleep architecture. These results support zebrafish as a tractable model for dissecting the regulation and function of sleep sub-states. More broadly, they demonstrate that structured, multi-state sleep is a conserved feature of vertebrate behavior.

Figure 1:. Monitoring larval zebrafish activity over 48-h day-night cycles.

(a) Overview of experimental assay. (b) Heat-map of locomotor activity starting at 5 dpf (n = 21 fish). (c) Locomotor activity for one representative fish. (d) Mean activity for all fish, with sem shown in grey. (e) Distribution of locomotor activity for all fish (mean ± sem). In (b-d) the white and black horizontal bars depict the lights-on and lights-off periods on both the days, respectively.

Figure 2:. An HMM with 4 states best fits the data for WT fish.

(a) The distribution of optimal number of states. Inset: Mean BIC as a function of number of HMM states. (b) The four means (λs; units s/min) characterize the Poisson fits to fish locomotor activity for each state. (c) The most likely state sequence inferred from the data for a single example fish, based on parameters of a four state HMM fit. (d) State sequences for 21 fish over 48 hours at 5–6 dpf. (e) Variation of mean posterior probabilities of each state as a function of time, shown as a moving window average. (f) Both mean S1 and mean S2 bout durations were longer at night than during the day (paired t-test, *p<0.05, ***p<0.001, with Benjamini-Hochberg multiple comparisons correction). (g) The transition probabilities averaged across all fishes individual 4-state HMM fit. (h) Sleep amounts obtained based on the conventional definition of sleep (1 minute of no locomotor activity, black) compared to the most-likely state sequence from 4 HMM fit. (i) Comparison of HMM-defined S2 sleep with conventionally-defined sleep during day and night. Filled and empty circles represent night and day sleep amounts, respectively (one dot per fish). In (c), (d), (e) and (h), the white and black horizontal bars depict the lights on and lights off periods, respectively. r and p-value are for Pearson correlation for the fit to the line y=x.

Figure 3:. Fish in S1 vs S2 states show different responses to arousing stimuli and sleep deprivation.

(a-c) Sleep deprivation experiment. (a) Fish were subjected to sleep deprivation (SD) on night 2, when lights were kept on for the first six hours at night (shown within dashed white lines), and then turned off for the last four hours of the night. (b) State sequence for an example fish over 48 hours (RS, rebound sleep). (c) Comparison of S2 and S1 during night 1 and night 2. (d-f) Arousal experiment. (d) An example 60 minutes of data with stimuli every 5 minutes (arrows) shown as a color map, with mean and standard error of mean locomotor activity over fish shown in upper plot. (e) A state sequence for an example fish over the night. In (d) and (e), the black horizontal bars indicate that lights were off. (f) Fish showed a higher proportion of responses (baseline-corrected) to stimuli when in the S1 state than S2 state, and when in the W1 state compared to the S1 state (t-tests with Benjamini-Hochberg multiple comparisons correction).

Figure 4:. Melatonin-deficient aanat2 mutants show loss of S2 and gain of S1.

(a-d) aanat2^−/− fish showed higher activity and lower sleep at night than aanat2^+/− fish. (e,h) Both genotypes were best fit by 4 HMM states. (f,i) aanat2^−/− fish had several significant differences in fitted HMM parameters (p < 0.05 indicated in red; bootstrap sampling with permutation tests; units of λs: s/min). (g,j) aanat2^+/− and aanat2^−/− fish showed strikingly differing patterns of sleep during the night. (k) At night aanat2^−/− fish had significantly less S2 but significantly more S1. (l) During the day there were no significant differences in genotypes between the times spent in each state. Asterisks indicate significant differences (t-tests, with Benjamini-Hochberg multiple comparisons correction; ***p<0.001, **p<0.01, *p<0.05).

Figure 5:. Circadian regulation of sleep at night is due to melatonin-dependent S2.

(a-d) Overall activity was reduced under constant darkness, but aanat2^−/− fish showed higher activity and lower sleep, and lacked circadian oscillations of sleep, at night compared to aanat2^+/− fish. (e,h) Both genotypes had minimum BIC at 3 HMM states. (f,i) aanat2^−/− fish showed significant changes in fitted HMM parameters (p < 0.05 in red; bootstrap sampling with permutation tests; units of λs: s/min). (g,j) aanat2^+/− fish maintained circadian variation in S2 and W states, but aanat2^−/− lost circadian variation in all states.

Figure 6:. tph2 mutants show less deep sleep.

(a-d) tph2^−/− fish showed higher activity and lower sleep than tph2^+/− fish. (e,h) Both genotypes were best fit by 4 HMM states. (f,i) tph2^−/− fish had several significant differences (p < 0.05 indicated in red; bootstrap sampling with permutation tests) in fitted HMM parameters (λ units: s/min). (g,j) tph2^+/− and tph2^−/− fish showed similar patterns of sleep states during the day but differing patterns at night. (k) At night tph2^−/− fish had significantly less S2 but significantly more S1. (l) During the day there were no significant differences in genotypes between the times spent in each state. Asterisks indicate significant differences (t-tests, with Benjamini-Hochberg multiple comparisons correction; ***p<0.001, **p<0.01, *p<0.05).

Figure 7:. Quipazine treatment induces sleep through gain in deep sleep.

(a-d) Quipazine treated fish showed lower activity and more sleep than DMSO controls. (e,h) Both genotypes were best fit by 4 HMM states. (f,i) Quipazine fish had several significant differences (p < 0.05 indicated in red; bootstrap sampling with permutation tests) in fitted HMM parameters (λ units: s/min). (g,j) Quipazine fish showed changes in sleep states during the night. (k) At night quipazine fish had significantly higher S2 but significantly lower S1. (l) During the day there were no significant differences in groups between the times spent in each state. Asterisks indicate significant differences (t-tests, with Benjamini-Hochberg multiple comparisons correction; ***p<0.001).

Figure 8:. dbh mutants show more deep sleep and less wake occupancy.

(a-d) dbh^−/− fish showed lower overall activity and higher sleep than dbh^+/− fish. (e,h) Both genotypes were best fit by 4 HMM states. (f,i) dbh^−/− fish had several significant differences (p < 0.05 indicated in red, bootstrap sampling with permutation tests) in fitted HMM parameters (λ units: s/min). (g,j) dbh^+/− and dbh^−/− fish showed altered patterns of sleep during night and day and wake states during the day. (k) At night dbh^−/− fish had significantly high S2 but significantly less S1. (l) During the day there were significant increases in S2 and S1, and a decrease in W2, in dbh^−/− fish. Asterisks indicate significant differences (t-tests with Benjamini-Hochberg multiple comparisons correction; ***p<0.001, **p<0.01, *p<0.05)

Figure 9:. Prazosin treatment induces sleep through gain in deep sleep and loss of wake.

(a-d) Prazosin-treated fish showed lower activity and higher sleep than DMSO controls. (e,h) Both groups were best fit by 4 HMM states. (f,i) Prazosin-treated fish had several significant differences (p < 0.05 indicated in red, bootstrap sampling with permutation tests) in fitted HMM parameters (λ units: s/min). (g,j) Prazosin fish showed an increase in S2 during the night and day, and reduced W2 during the day. (k) At night prazosin fish showed a significant increase in S2 and a significant decrease in S1. (l) During the day there were significant differences in groups between the times spent in each of S2, S1 and W2. Asterisks indicate significant differences (t-tests with Benjamini-Hochberg multiple comparisons correction; ***p<0.001, **p<0.01, *p<0.05).