Sleep fragmentation disrupts vocal interactions in rats

ABSTRACT Repeated interruption of the sleep cycle, commonly known as sleep fragmentation, is associated with a plethora of health issues, ranging from mood swings and memory loss to severe neurodegenerative disorders. Despite being a significant health problem with consequences on the social lives of individuals, its effect on vocal communication has been poorly studied. Here we show that sleep fragmentation induces a decrease in vocal production of a social rodent, without altering the acoustic characteristics of the vocalisations emitted. We conducted an experimental study using Rattus norvegicus rats, known for their ultrasonic vocal repertoire, in which we frequently woke pairs of individuals during their daily sleeping period. The rats whose sleep was artificially fragmented produced fewer vocalisations during their active periods than control pairs whose sleep was not disturbed. This decrease in vocal activity occurred after only two phases of fragmented sleep and was maintained throughout the 4 weeks of the experimentation. Conversely, sleep fragmentation had no effect on the rats’ vocal repertoire. Our results demonstrate that fragmented sleep impacts vocal interactions and emotional expression in a social mammal, and that this effect is maintained over weeks without recovering.


Introduction
As a widespread phenomenon in aged persons and patients with sleep disorders, sleep fragmentation induces a large spectrum of life issues, including cognitive decline, increased daytime sleepiness and mood disorders (Lee et al. 2022).Although these effects may impact the social abilities of people, the effects of sleep fragmentation on social interactions have been largely neglected.Previous studies have reported impaired performance during the processing of emotional information (Lee et al. 2022), a change in the way people respond to stimuli (Drummond et al. 2006;Anderson and Platten 2011), and alterations of emotion regulation (Tomaso et al. 2021).However, and to the best of our knowledge, no study has already investigated the effect of sleep fragmentation on vocal production, communication and expression of affective state.
In the present work, we investigated the effect of sleep fragmentation on vocal production in the rat Rattus norvegicus.In this species, vocalisations play a critical role during social interactions (Brudzynski 2014).The rat vocal repertoire contains several calls, associated with different behavioural contexts.Rat calls can be classified in two main categories: one category being associated with aversive (anxiety) contexts, and the other category being associated with positive (hedonic) contexts (Brudzynski 2018).These two types of calls differ by their frequency range (around 22 kHz for the aversive vocalisations and around 50 kHz for the positive vocalisations), their duration and how they are modulated in frequency.As a window into the biobehavioral, motivational and emotional states of the emitter, the rather straightforward vocal repertoire of the rat offers a behavioural proxy to assess individual's physiological and emotional states (Simola and Granon 2019).
To test whether sleep fragmentation impacts the vocal production in quantity (number of vocalisations) and/or quality (type of vocalisations), we recorded the calls emitted by pairs of healthy rats over a period of 28 days.The animals were placed in an apparatus designed to create sleep fragmentation in a standardised way (Leenaars et al. 2011).Using automatised detection and analysis of vocalisations, we quantified the number of calls emitted during the daily active phase of the rats, and specified the vocal repertoire by measuring the acoustic characteristics of the vocalisations throughout the experiment.We predicted that sleep-fragmented rats will be less vocal and that their calls will be more aversive than control rats.We also predicted that the effect of sleep fragmentation will increase with the duration of the experiment.

Animals
We worked with 1-year-old Wistar rats (n = 8 males; provided by Janvier Labs, France).Prior to the experiment, the animals were housed during 14 days in the SAINBIOSE Lab animal facilities under a controlled environment (12 h/12 h day/night cycle, temperature 19-21°C, food and water ad libitum).

Experimental design
Pairs of rats (n = 4 pairs) were housed in experimental cages specifically designed to induce fragmented sleep ('CaResS cages', (Leenaars et al. 2011).Each cage consists of one cylindrical Plexiglas box (diameter = 30 cm), divided in two compartments by a transparent glass panel.Each compartment of the cage accommodates one rat.The cage floor can be rotated while the glass panel remains still, which results in the rats being pushed and awaked (rotation speed of the cage floor = 180° in 10 s).This setup was validated to induce sleep fragmentation (Ringgold et al. 2013), as their rats' electroencephalogram revealed a significant decrease in rapid eye movement sleep, which almost reached 0%, on days 1, 3 and 9 compared to preprotocol values, and no significant difference of non-rapid eye movement was recorded compare to pre-protocol.The rats were under a 12 h light − 12 h dark cycle (lights turned off at 09:00 am and turned on at 09:00 pm), with food and water ad libitum.The floor of the cage was covered with beddings, and one piece of wood was provided to each rat as an enrichment.
The rats were placed in their experimental cages 2 weeks prior the beginning of the sleep-fragmentation protocol.During this habituation period, the cage floor remained motionless for all rats.The sleep-fragmentation protocol started at the end of the habituation period and lasted 28 days (Figure 1).
Four rats were submitted to sleep fragmentation ('sleep-fragmented rats') while the four other rats were used as control ('control rats').In the sleep-fragmented situation, we provoked 30 interruptions of sleep per hour during the sleeping period (light phase of the day; sleep interruption induced by a cage floor rotation of 9 s every 120 s, a commonly employed fragmentation rhythm (Li et al. 2014).In the control situation, the cage floor remained motionless during the entire light period.During the dark (active) period, both sleep-fragmented and control rats had floor rotations during 15 min every hour.This protocol was decided to prevent a phase shift (sleep cycle inversion) of the sleepfragmented rats.

Recording of vocal interactions
One recorder (Audiomoth 1.2.0 with firmware version 1.6) was placed on top of each cage, directed towards the animals.The recorders were programmed to record 15 min every hour throughout the entire experiment (total duration: 31 days, resulting in 186 h of recordings for each experimental cage).Recording periods occurred during the 15 first minutes of each hour, corresponding to the 15 min of each hour when the dark cycle of rat was disturbed by the floor rotation.The sampling rate was 250 kHz, enabling to record sounds with frequencies up to 125 kHz.Batteries and SD memory cards were changed twice a week, during cage cleaning which was done during the rats' active period.

Detection and acoustic analysis of rats' vocalisations
We recorded the ultrasound vocal signals produced by the rats from the last 3 days of the habituation period to the end of the experiment.We performed the acoustic analysis using Deepsqueak v3.0, a software dedicated to the automatic detection and analysis of rodents' ultrasound vocalisations (Coffey et al. 2019).Spectrograms were generated with a 0.0056 s window size, with 90% overlap and 0.0056 s NFFT.The vocalisations were detected using both Long Rat and Rat Detector YOLO R1 trained Neural Networks.To improve the automatic detection accuracy of the software, we first manually surveyed all vocalisations produced during one full day, testing for the best detection parameters, and chose to set at 0.25 and 0.55 the tonality and the score thresholds, respectively.
To characterise the magnitude of the vocal interactions within the pairs of rats and their temporal dynamics throughout the experimental protocol, we measured the number of vocalisations emitted per 15 min.To test whether the acoustic characteristics of the vocalisations may change according to the condition (sleep-fragmented or control) and/ or along the experiment, we automatically measured 10 acoustic parameters from each detected vocalisation (Table 1).

UMAP representation and acoustic clustering
We built multi-dimensional graphical representations of the rats' vocal repertoire using the UMAP dimension reduction algorithm ((McInnes et al. 2020); uwot R package, (Melville 2022); R version 4.2.2 (R Core Team 2023)).The dimension reduction has been performed over the 10 acoustic parameters extracted with Deepsqueak (Table 1), to obtain a two-dimensions representation of the vocalisations.The UMAPs defined three acoustic clusters (see Results) which correspond to the acknowledged classification of rats vocalisations (Burgdorf et al. 2008;Wöhr et al. 2008;Brudzynski 2018): -'22-kHz calls': vocalisations weakly modulated in frequency, centred around 20 kHz (20-35 kHz).These calls are usually emitted in aversive contexts.
-'flat 50-kHz calls': vocalisations weakly modulated in frequency, between 35 and 80 kHz.These calls are usually emitted in positive contexts.
-'frequency modulated 50-kHz calls': vocalisations highly modulated in frequency, above 50 kHz.These calls are usually emitted in positive contexts.

Statistics
We analysed the acoustic data using Bayesian multilevel models fitted with the 'brms' R package (Bürkner 2017), and using default priors.As both sleepfragmented and control rats remained almost silent during the inactive (light) phase (Supplementary Figure S1), we only considered the vocal interactions occurring during the active (dark) phase.To test whether sleep fragmentation has an effect on the number of vocalisations produced by the rats, we fitted a first model with the number of vocalisations per 15 min as the response variable (modelled with a lognormal distribution) and the interaction effect between the experimental condition (sleep-fragmentation or control) and the week of experiment (from habituation period to the 4 th week).Due to the limited number of replicates, we did not include random effects in the main models.To target the potential acute effect of sleep fragmentation, we built a complementary model which included the interaction effect between the experimental condition and the day of recording during the first week, from the last day of the habituation period to 7 th day of the experiment.To test whether sleep fragmentation could shape the rats' vocal repertoire, we built a model including a three-way interaction between the conditions (sleep-fragmented or control), the week of the experiment (from habituation period to 4 th week) and the call type (22 kHz, flat 50 kHz, and modulated 50 kHz).We also modelled the variation of three acoustic parameters (duration, principal frequency and frequency modulation) of each call type across weeks of experiment, to investigate potential intra-call types shift during the protocol.Posterior distributions of the models' parameters were summarised by their median and 95% credible interval (CI).Prediction variables of models have a significative effect when the CI of difference between their levels is strictly positive or negative, indicated that this difference is reliable and robust (Kruschke and Liddell 2018).Random effects on animals' identities were applied in the models to investigate the uncertainty in the effect size due to limited number of individuals recorded.Uncertainty is reported as the probability of observing no difference between control and sleep-fragmented rats.

Ethics
The experimental protocol was approved by the Ministry of Higher Education, Research and Innovation of France .
Figure 2(b) reports the detail of the rats' vocal production during the first 3 days following the habituation period.The decrease in the number of vocalisations in the sleep-fragmented rats appeared after two perturbated sleeping periods (Day 1 compared to habituation period: −28 voc/h for sleep fragmented rats, 95% CI [−47, −13], uncertainty 0.61%).This decrease was marginal after the first sleep-fragmented sleeping period (Day 0 compared to habituation period: −13 voc/h for sleep fragmented rats, 95% CI [−36, 8], uncertainty 35.87%).
Although control rats stayed more vocal than sleep-fragmented rats throughout the duration of the experiment, their vocal production nevertheless decreased on week 3, i.e. after 5 weeks in the experimental cage (comparison between control rats' vocal production during week 3 and habituation period: −17 voc/h, 95% CI [−39, −2], uncertainty 1.80%).

Sleep fragmentation does not alter the acoustic characteristics of vocalisations
The 2D acoustic spaces (UMAPs) on Figure 3 illustrate the vocal repertoires of the rats for both control and sleep-fragmented conditions during the habituation period and during the weeks 1 and 4 of the experimental procedure.These vocal repertoires are remarkably consistent across conditions and time, and are organised according to three acoustic clusters.These clusters confirm the use of 22-kHz, flat 50-kHz and modulated 50-kHz vocalisations as a correct proxy of repertoire structure in our rats.
Each one of the three types of vocalisations remain stable across time (no significant change in duration, principal frequency or frequency modulation) for all conditions (Supplementary Table S1).

Discussion
This study demonstrates that sleep fragmentation induces a decrease in vocal production in rats.This effect appeared after only two fragmented sleeps, and remained stable during the 4 weeks of the experiment.Contrary to our prediction, the type of emitted calls did not change over the course of the experiment.The sleep-fragmented rats and the control rats displayed the same vocal repertoire and there was no tendency of sleep-fragmented rats to emit more or a greater proportion of aversive vocalisations.Sleep fragmentation strongly influences vocal production since the number of calls produced during the active period was almost divided by two.This decrease may likely be due to the fatigue induced by the disturbed nights, similar to daytime sleepiness observed in humans in similar circumstances.It is, however, to be noticed that, after 3 weeks of the experimental procedure, control rats also decreased their vocal production.It is known that rats are highly social animals that require inter-individual contacts (Brudzynski 2005).In our experimental protocol, the two rats housed together in the cage remained physically separated.As it has been shown that social contact increases the number of 50-kHz USVs produced by rats (Brudzynski and Pniak 2002), this form of social deprivation may explain the decrease of the vocal interactions in control rats after a while.Moreover, social deprivation may also have an impact on sleep-fragmented rats leading to a higher decrease in vocal production compared to what sleep disruption could already cause on its own.One remarkable result is the fact that the vocal repertoire remained without any strong change in the proportion of emitted call types throughout the experiment, for both control and sleep-fragmented rats.The flat 50-kHz calls have been suggested to serve as a means of social coordination (Burgdorf et al. 2008;Wöhr et al. 2008).Their presence may thus indicate that the two rats present in each cage kept on interacting together throughout the weeks, even in the case of sleep fragmentation.The rather small proportion of frequency-modulated 50 kHz calls (associated with positive emotional states) and the rather high proportion of 22 kHz calls (associated with state of anxiety) in both conditions may suggest that the rats did not like very much the experimental apparatus, either because of its design and functioning or because of the restricted physical interactions.Short-term study noticed that acoustic characteristics of call types remained stable across experimental sessions (Wright et al. 2010).Our findings are in line with this result, showing that sleep fragmentation does not alter call-type structure in rats.
Sleep fragmentation thus impacts vocal production in a quantitative but not qualitative way.These changes may reflect change in the animals' affective mood, likely due to inappropriate sleep for 28 consecutive days.This could be considered as a low-arousal and long-term change, as modifications in acoustic parameters of calls are most likely to occur during 'highly emotional' situations (Brudzynski 2021).
This research underlines the interest of using a bioacoustics approach to monitor variations in social performance of rats under a stress, and their vocal emotional expression.This non-invasive approach looks especially adequate for medium or long-term studies with freely moving animals.Although the measurement of vocal production and the analysis of the acoustic characteristics of vocalisations require some bioacoustics knowledge, the automatic analysis tools available today -as the one we used herefacilitate the work by allowing the analysis of long recordings without manual inspection.By giving proxies of the social abilities as well as of the mental states of the animals throughout an experiment, the bioacoustics approach is a powerful means allowing continuous monitoring of those important traits.This approach could complement cognitive tests performed at the end of an experiment or post-mortem examinations.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Figure 1 .
Figure 1.Sleep fragmentation experiment in rats.Pairs of rats were placed in two adjacent compartments of experimental cages with a movable floor.During their inactive phase (light, 9 pm-9 am), experimental rats were awakened every 2 minutes by a rotation of the cage floor, while control rats were left undisturbed.During the active phase (dark, 9 am-9 pm), the floor rotated for both control and experimental rats during 15 minutes per hour.This protocol lasted 4 weeks.The rats' vocalisations were recorded 15 minutes/hour throughout the experiment.

Figure 3 .
Figure 3. Distribution of vocalisations in a two-dimensional acoustic space.(a) The acoustic space is a UMAP representation constructed from 10 acoustic parameters describing each recorded vocalisation (see main text for details).Each dot represents one vocalisation (for clarity, only 1500 vocalisations taken at random from all vocalisations emitted for a given week and condition are represented).The colour scale indicates the frequency of the vocalisations in kHz.The dashed lines delimit three main clusters corresponding respectively to sounds with a frequency around 20 kHz (purple), flat sounds with a frequency between 30 and 60 kHz (blue), and modulated sounds with a frequency above 50 kHz.The bar charts show the proportion of each cluster as a percentage of the total number of vocalisations recorded per week per condition.(b) the spectrograms represent examples of vocalisations belonging to each of these three clusters (purple −22-kHz call; blue -flat 50-kHz calls; greenfrequency modulated 50-kHz call).The distribution of vocalisations in the acoustic space, as well as the relative proportions of the acoustic clusters, did not differ significantly between weeks and conditions.

Table 1 .
List of automatically extracted acoustic parameters from Deepsqueak software.