Disentangling developmental effects of play aspects in rat rough-and-tumble play

Animal play encompasses a variety of aspects, with kinematic and social aspects being particularly prevalent in mammalian play behaviour. While the developmental effects of play have been increasingly documented in recent decades, understanding the specific contributions of different play aspects remains crucial to understand the function and evolutionary benefit of animal play. In our study, developing male rats were exposed to rough-and-tumble play selectively reduced in either the kinematic or the social aspect. We then assessed the developmental effects of reduced play on their appraisal of standardized human–rat play (‘tickling’) by examining their emission of 50 kHz ultrasonic vocalizations (USVs). Using a deep learning framework, we efficiently classified five subtypes of these USVs across six behavioural states. Our results revealed that rats lacking the kinematic aspect in play emitted fewer USVs during tactile contacts by human and generally produced fewer USVs of positive valence compared with control rats. Rats lacking the social aspect did not differ from the control and the kinematically reduced group. These results indicate aspects of play have different developmental effects, underscoring the need for researchers to further disentangle how each aspect affects animals.


Introduction
Animal play is ubiquitous among mammals, captivating scientists' efforts to understand its functions and evolutionary advantages [1][2][3][4].Numerous observational and experimental studies have associated enriched play experience with improved survival, locomotor function and social competence [5][6][7][8][9].However, animal play is complex, encompassing multiple behavioural aspects.For example, in rat rough-and-tumble (RT) play, there is a kinematic aspect where rats engage in pouncing, rotating and running, representing physical manoeuvres in play [10][11][12].There is also a social aspect, requiring rats to communicate and take turns with playmates to maintain play [13].The kinematic and social aspects may contribute differently to the intricate developmental effects of RT play [14][15][16].Disentangling effects from each aspect can further our understanding of why animal play is so widespread and varied.
RT play has been investigated extensively in rats, providing valuable insights into animal play behaviour.Rats engage in RT play from 15 days old and their play frequency peaks when they are around 35 days old, after which the play rates decrease gradually until adulthood [17].In RT play, rats switch roles between 'attacker' and 'defender.'The attacker aims to 'bite' the nape of the defender, and the defender employs various defensive manoeuvres.These manoeuvres often pose the defender in a supine position, and the attacker can pin the defender with paws and mouth before switching their roles [11].We previously developed a paradigm to deprive rats of the kinematic aspect of play using a height-restricted cage [14] while the F344 paradigm [13], using the socially inactive Fischer-344 strain, has been well established to reduce the social aspect of RT play.The most prominent effect of RT play is on social interactions.Rats with reduced play experience were found to escalate social encounters and provoke aggression [18][19][20][21].However, it is not clear whether this compromised social competence is caused by reduced experience with social, kinematic or combined aspects, and how both aspects will affect perception and responses to novel social interactions.
50 kHz USVs are a useful tool to assess social interactions in rats.These USVs are predominantly emitted during positive social interactions such as RT play or sexual behaviour [22,23].This makes 50 kHz USVs a reliable indicator of the positive affective state of the animals [24].Comparing USV production between rats with different play experience can help us understand the developmental effects of play aspects.50 kHz USVs are often further classified into subtypes by inspecting their spectrograms (figure 1).The 'Flat' USV, appearing as a continuous line on a spectrogram, and the 'Short' USV, characterized as a dot-like shape, are easy to identify.Other 50 kHz USVs, referred to as frequency-modulated (FM) USVs, are suggested to reflect the positive affective state in rats [24][25][26][27][28][29], and have been further classified into subtypes [30,31].However, to detect and classify a large quantity of 50 kHz USVs manually is challenging.Reported low repeatability [32] and slow speed of manual classification especially hinder our ability to expand the usage of USV subtypes.Automation [33] and deep-learning models like DeepSqueak [34] have greatly increased the efficiency to detect and extract USVs from longitudinal recordings, but there has yet to be a model that can consistently classify USV subtypes in large datasets based on a smaller subset of accurately labelled USVs.Such models would make the use of 50 kHz USVs a very efficient tool for research on rat social behaviour and affective development.
The emission of 50 kHz USVs can be effectively stimulated through a standardized procedure that mimics RT play, often referred to as 'tickling' [24,[35][36][37][38].This 'tickling' procedure mimics the kinematic sequence of RT play with dorsal contact, pinning and ventral contact [39].When being tickled by a human, rats experience a dose-controlled form of play making it possible to measure their appraisal of the tactile stimuli typical for RT play, and thus providing an excellent method to assess the effects of kinematic and social aspects of play.While tickling incorporates key components of RT play, it is not an exact replication of it [37,38].Nevertheless, in comparison with the rat-rat RT play, it has the advantage of not introducing confounds like individual differences in play partner's motivation [40][41][42][43].
We previously found that rats experiencing kinematically and socially reduced play still appraise tickling positively by predominantly emitting 50 kHz USVs [14].Now, we further investigate whether play aspects specifically affect rats' appraisal of being tactilely stimulated by a human.We hypothesized that rats experiencing kinematically reduced play will emit fewer positive-valenced USVs, especially when being contacted by human, compared with the control and the socially reduced group.

Material and methods
Our experimental procedure has been detailed as the subgroup Batch C in a previous study [14].Eighteen male Long-Evans (LE) rats (21 ± 1 days old, 74 ± 6 g, from the breeding stock at the Institute of Physiology of the Academy of Sciences of the Czech Republic) and six male F344 rats (21-28 days old, 41 ± 3 g, from breeder VELAZ, s.r.o.) were recruited in this experiment.Rats were housed in rat boxes (Ferplast Duna Multy Box, 70 × 46 × 31 cm, with paper bedding and cardboard pieces) separated in the middle by a wire mesh.Pairs in the same rat box had visual, auditory, olfactory and also limited tactile contact, but no physical interactions.Rats had ad libitum access to a standard laboratory rat diet and tap water.The stock room (290 × 590 × 280 cm) was air-conditioned (22-24°C) with an 14 : 10 L : D light cycle (lights on between 08.00 and 22.00 h).Animal welfare was checked each morning by experienced staff.Subjects were donated to Zoopark Zájezd zoo after the experiment.
Briefly, six LE rats were randomly assigned to one of three groups: control, kinematically reduced and socially reduced.Rats in all treatments were allowed to play with a playmate from the same cage for 1 h per work day from 21 to 49 days old in a play cage (42 × 26 × 19.5 cm with paper bedding).Control rats played with their LE playmate in an unaltered play cage.Kinematically reduced rats also played with a LE playmate, but their play cage was height restricted by a metal wire mesh ceiling (height at week of age 4/5/6/7: 3/3.5/4/4.5 cm) that resulted in no pinning or being pinned during play [14] but rats could still touch, contact and crawl around each other.Rats in the socially reduced group played in an unaltered play cage but were paired with a F344 rat, resulting in giving more pinning than being pinned [14].Play sessions started at 09.55 each day.After 5 min of looped music stimulus, all rats were relocated from their home cages to play cages.Experimenters then left the room and returned 1 h later to transport rats back to their home cages.
We started a standardized 2 min human-rat play procedure ('tickling') before (08.00-09.30)rats' daily play sessions when rats reached 35 days old and this procedure, imitating the sequence of dorsal contact, supine pinning and ventral contact in rat RT play, continued for 10 work days.Before the start of tickling sessions, a plastic tickling box (42 × 26 × 19.5 cm) was placed on a table at 100-300 cm distance from rats' home boxes.A microphone (UltraSoundGate 116H, single-channel acquisition and an CM16/CMPA condenser) was placed nearby to record ultrasonic sounds.Two cameras (Milesight) were placed in front and on the right side of the tickling box to record rats' behaviour.Our tickling procedure mostly followed [39]: the experimenter hovered one hand in the middle of the tickling box for 15 s, then in the next 15 s, the experimenter repeatedly tickled the nape three-four times with two fingers, flipped the rat onto its back, tickled its belly for three-four times and allowed the rat to straighten itself up onto its legs.This procedure was in general repeated four times per 15 s; however, it ranged from two to five times.The experimenter alternated between 15 s of hand hovering and 15 s of tickling for 2 min before ending the session and returning the rat to its home cage.The order of rats being tickled was fixed but rats from different treatments were randomly distributed along the tickling order.Two female experimenters performed all tickling sessions to reduce potential impacts of human olfactory cues on rats [44].
To extract USVs from recordings, we first used DeepSqueak [34] and detected 4616 USVs in a subset of our audio recordings.A trained experimenter (M.R.) visually examined the spectrogram (35-80 kHz bandpass, Hanning window, 1024 window 2 royalsocietypublishing.org/journal/rsbl Biol.Lett.20: 20240037 length, 90% overlap; if the detected USV was shorter than 40 ms, its spectrogram was padded to 40 ms with the USV centred in the middle) of all 4616 USVs and labelled each single USV into one of eight subtypes according to our criteria (figure 1, criteria see electronic supplementary material, table S1).The intraobserver repeatability was tested by relabelling 307 USVs, 3 days after the initial labelling.We then merged frequently and reciprocally mislabelled subtypes into one as these subtypes cannot be reliably distinguished by the experienced experimenter, and visually indistinguishable groups could decrease the training efficiency and model accuracy in the following steps.Thus, we ended up with five subtypes.We uploaded all spectrograms of labelled USVs as 'jpg' images to roboflow [45] to train a 'Roboflow 2.0 Classification' model.The 4616 images were split into 3231 : 923 : 462 for training, validation and testing set.All images were resized to 640 × 640 pixels, and three squares (combined size: 10% of the image) were randomly cut-off from images in the training set.Additionally, black pixels were randomly added (up to 5% of the image) on the image as noise for the training set.These augmentations in the training set could increase the training speed and the accuracy of the model.Our model achieved 84.5% accuracy on the testing set.We then applied this model to predict the subtype of 21 590 USVs detected by a customized R script [14] in all audio recordings.To further validate the model, we randomly selected 1000 USVs with predicted subtypes and compared the model prediction with human performance (blind to the prediction).In these 1000 USVs, we reached an agreement of 79.0% with model prediction.Combined with the manually labelled 4616 USVs, we have in total 26 206 USVs in five subtypes.
Rats' behavioural states (dorsal contact, ventral contact, explore cage, explore hand, escape and immobile, ethogram see electronic supplementary material, table S2) during all tickling sessions were observed and documented in BORIS [46] by M.R. Timestamps of all behavioural states were noted and we used these timestamps to calculate the total duration of each behavioural state per tickling session.We then assigned the number of each USV subtype to each behavioural state per session.If rats did not emit any USV in a behavioural state in a session, we assigned 0 to all USV subtypes.
To investigate how treatments and behavioural states affected the emission of USV subtypes, we created a full mixed linear model (glmmTMB function in package 'glmmTMB' in R 4.3.2[47]) with the total number of USVs (log-transformed) as the normally distributed response variable.The three-way interaction of treatments (control/kinematically reduced/socially reduced), USV subtypes (Trill Plus/Trill/Complex/Short/Flat) and behavioural states (ventral contact/dorsal contact/explore cage/explore hand/escape/immobile) was the fixed effect.The total duration of the behavioural state, the order of tickling and tickling days (1-10) were included as covariates too.Rat IDs and session IDs were included as two random intercepts.We then selected the most parsimonious model within 2ΔAICc of the top model as the best model derived from this full model (dredge function in package 'MuMIN') and post hoc tested any significant interaction terms.All used dataset and scripts can be found in the Dryad Digital Repository [48].

Results
Rats from all groups emitted copious 50 kHz USVs during all tickling sessions (see The best model included all two-way interactions but not the three-way interaction between treatments, USV subtypes and behavioural states (see table 2. Model analysis results, estimates and CIs on predictor levels see electronic supplementary material, table S3).The total duration of behavioural states is also included in the best model, meaning the longer the duration of the behavioural state was, the more USVs were emitted.USV emission was affected by rat individuals (repeatability = 0.214 [0.106, 0.334], 'rpt' function from R package 'rptR' using the best model) and by sessions (repeatability = 0.101 [0.074, 0.133]).
Our post hoc tests (figure 2a, details see electronic supplementary material, table S4) revealed that rats experiencing kinematically reduced play emitted fewer Trill Plus, Trill and Complex USVs than controls.In addition, they also in general emitted fewer USVs during ventral and dorsal contacts (figure 2b, for details, see electronic supplementary material, table S5).Rats experiencing socially reduced play were not different from either control or kinematically reduced group for all behavioural states and USV subtypes.In conditions where the control group produced significantly more USVs than the kinematically reduced group, the mean value of the socially reduced group was between the mean values of the other two groups.
The frequency of USV subtypes differed among behavioural states (details see electronic supplementary material, figure S1, table S6).In general, rats emitted Trill Plus and Trill most frequently and on a comparable level.However, in ventral contact,  Notes: Rat IDs and session IDs are included as random intercepts; the marginal and conditional R 2 of this model are 0.72 and 0.80.
Trill Plus was more often emitted than Trill.These two subtypes were always more frequent than Short and Flat.Complex was the third most frequently emitted subtype, and only in escape, Trill Plus and Trill were not more often emitted than Complex.Short and Flat were the least frequent USV subtypes.

Discussion
We found rats experiencing reduced kinematic aspects in play, compared with control rats, emitted fewer Trill Plus, Trill and Complex USVs.They also produced fewer USVs in general during dorsal and ventral contacts compared with control rats.Trill Plus, Trill and Complex are FM USVs associated with positive valence [23] and ludic functions [49][50][51][52][53]. Consequently, this result suggests that although rats lacking the kinematic aspect in play still appraise being tickled positively, the degree of their positive response is weaker than that of rats experiencing full RT play.The diminished appraisal likely due to the absence of the kinematic aspect of play.This indicates that aspects of play have at least short-term developmental effects in rats.Interestingly, the socially reduced group did not appraise human contacts significantly different from either group and was at an intermediate level.This could indicate lacking the social aspect in play (reciprocity, etc.), is less impactful than lacking the kinematic aspect for appraising being tickled positively.The lack of impact of the social play aspect deprivation on appraisal in the current study, seemingly contrasted with previously observed effects of play deprivation on conspecific social encounters [18][19][20][21].This could indicate that appraising social encounters and conducting social encounters are two distinct capacities that are differently affected by play experience.An alternative explanation is that rats perceive human and conspecific play differently.Further research focussed on how social and kinematic play aspect deprivations selectively affect rat-rat RT play appraisal is required to support either hypothesis.A comparison of the current results with our previous study [14] that did not find effects of the same treatment on the total number of 50 kHz USVs indicates that the nuanced play aspects could affect certain USV subtypes instead of general USV production.This suggests that follow-up studies exercise caution when combining all subtypes of 50 kHz USVs in their analyses.
The lack of a kinematic aspect of play could affect cognitive and/or affective development.Rats with kinematically reduced play received and gave no pinning and biting [14], resulting in limited tactile contacts.Tactile stimulation has been shown to reduce the differences between artificially and maternally raised rats [54] in stress coping [55,56], social learning [57,58] and response to rewarding cues [59].Although there lacks direct evidence to link aforementioned cognitive domains to appraising tickling in an unfamiliar social encounter, it is plausible to hypothesize that reduced tactile stimulation in kinematically reduced rats could have broader implications on their development, potentially influencing how they perceive and respond to novel social interactions such as tickling.Alternatively, the reduced appraisal may have arisen from the unfamiliarity of this sensory stimulation.Ventral and dorsal contacts in tickling stimulate a similar sensation as being pinned in RT play, which these rats were deprived of, and the lack of a similar experience could decrease their level of positive response.In this case, kinematically reduced rats should gradually increase their appraisal with increasing positive experience of being tickled or pinned.They might also emit a comparable level of 50 kHz USVs to controls under other positive situations, for example anticipating a reward.Studies should further investigate whether non-tickling tactile contacts like petting [51] could mitigate the effects of kinematically reduced play.In order to control play aspects, rats in all groups were housed for 23 h a day in conditions that prevented direct behavioural interactions.This could have systemically affected their cognitive development as the 1 h play opportunity can only partially compensate for the lack of home cage play [60][61][62].Thus, the observed effect could arise from a combination of a general play restriction applied to all groups and the specifically reduced kinematic experience.
Trill Plus, Trill and Complex were produced more often than Short and Flat.This result is in accordance with most previous studies [32,50] investigating the functions of USV subtypes in RT plays.Our result suggests tickling indeed elicits similar vocal responses as RT play does in rats.We found that rats also emitted more Trill Plus, Trill and Complex when they were exploring the cage or the hand of the experimenter than when they were trying to escape or immobilized.This further confirms that Trill Plus, Trill and Complex are positively valenced.We trained a deep-learning model to classify USV subtypes using annotated spectrograms.This model significantly enhanced the efficiency and capacity for USV analysis, facilitating a more thorough examination of our data compared with traditional methods.The model, not relying on human perception, may reduce human bias that has been reported before in a similar task [32].The model achieved an accuracy of 84.5% on USVs in the validation set, surpassing the repeatability observed in the previous study with human observers.Although our task was somewhat simplified by having fewer subtypes, the model, trained with only 4616 USVs, can substantially improve its performance with more finely labelled USVs from more individuals.As it is openly accessible, this model encourages collaboration among interested research groups.
We developed a machine learning model to efficiently classify USV subtypes and found developmental effects of the kinematic aspect of play in developing rats.Given that the effects of play are often observed in adults as well, further research should test whether these effects of play aspects are transient or persist into later developmental stages.It is also imperative to examine whether this aspect only affects appraisal or if a wider range of behaviours are also affected.Investigating how persistent and extensive aspects of play contribute to the development of animal behaviour will help us understand the functions and evolutionary advantages of animal play.New techniques like machine learning method can enable us to efficiently process data from behavioural observations.Data accessibility.The original data of this paper are presented in the electronic supplementary material file 'Liu et al_Disentangling developmental effects of play aspects in rat rough-and-tumble play_raw data.csv'[63].The R scripts for statistical analysis, visualization and USV extraction can be found in electronic supplementary material file 'Liu et al_Disentangling developmental effects of play aspects in rat rough-and-tumble play_Supplementary script_analysis.R' and 'Liu et al_Disentangling developmental effects of play aspects in rat rough-and-tumble play_USV Spectrogram_cutting.R'.The description of our dateset and all electronic supplementary materials can also be found in the Dryad repository [48].
The USV classification model can be accessed at [64].
Declaration of AI use.Machine learning technology has been used to train the USV classification model, and we have described this in the Material and methods section.

Figure 1 .Figure 2 .
Figure 1.Examples of USV Subtypes.The x-axis shows the time in seconds and the y-axis shows frequency in kHz.The blue to red colour gradient represents increasing energy.Each panel shows a single USV and its annotation.Trill Jump, Trill Flat and Steps were combined into Trill Plus after intraobserver repeatability test.Complex and Composite were combined into the Complex subtype in the same way.

Ethics.
The experiments conducted in this study were approved by the Ministry of Agriculture of the Czech Republic under the licences MSMT-4073/2021-2 and were carried out under the supervision of the Animal Welfare Body of Czech University of Life Sciences Prague in accordance with national and European legislation.This research adhered to the ASAB/ABS Guidelines for the ethical treatment of nonhuman animals in behavioural research and the ARRIVE Guidelines.

Table 1 .
Descriptive statistics of emitted USV numbers per session.

Table 2 .
Model analysis of the best model explaining how treatment and behavioural state affect the production of USV subtypes.Significant effects (p < 0.05) are in bold.