Play in juvenile greater rheas: different modes and their evolutionary and socio-cognitive implications

ABSTRACT Even if there is evidence of play from all vertebrate classes suggesting origins in deep time, descriptions of the evolution of play are surprisingly patchy. To bridge this gap, one must study play comparatively and include taxa from key phylogenetic positions. This study is the first systematic description of play in greater rheas, and thereby the first such report on any palaeognath bird. Palaeognaths represent a major subgroup of modern-day birds that have retained many ancestral features from their direct ancestors, the non-avian dinosaurs, making them an ideal window into the behaviors of the earliest birds. We recorded play behaviors of a group of captive rheas, with a focus on the modes and ontogenetic development of their play. Juveniles predominantly engaged in contagious locomotor play, adding a social component to the majority of their play bouts. Interactive social play such as wrestling appeared only around the age of 10.5 weeks and was generally rarer. Based on our findings we hypothesize that early birds, and likely also paravian dinosaurs, played in a similar fashion with a noticeable component of sociality. These hypotheses need to be expanded through more studies on different species of palaeognath birds.


Introduction
Play has captured the fascination of researchers in both biological and social sciences for more than a century. This phenomenon has, however, proven to be surprisingly elusive from an evolutionary viewpoint. While play appears to have deep evolutionary roots, exhibited by mammals (e.g. Byers, 1999;Himmler et al., 2016;Lewis, 2000), birds (e.g. O'Hara & Auersperg, 2017), reptiles (e.g. Dinets, 2015), and fishes (e.g. Burghardt, 2015), it remains unexplained why playa seemingly unproductive behavioral stateis favored by natural selection. Many hypotheses on the adaptive value of play have been forwarded, ranging from the improvement of sensorimotor control, to supporting cognitive development, and the acquisition of social skills. But most hypotheses have various shortcomings and none of them can fully explain the evolution of play (Burghardt, 2005).
One established method to better understand the evolution of particular traits is to compare species that occupy key phylogenetic positions. Such comparisons often provide insights into how a feature has changed in evolutionary time through its expression in various lineages. Further knowledge is gained when the comparative results are correlated with e.g. socio-ecology, brain anatomy, and various other factors. So far, play research has mainly focused on humans, non-human primates, and mammals in general, and has disregarded other species occupying evolutionary key positions (Burghardt, 2005).
The evasiveness of the phenomenon of play is not only apparent from the evolutionary perspective, but also when it comes to defining it. Sometimes, behaviors are intuitively labeled as play that in fact represent serious behaviors, and sometimes it is the other way around such as in leapfrogging fish (e.g. Gudger, 1944) and reptiles tossing around objects (e.g. Burghardt et al., 2002). To tackle this problem, and to avoid a single simple definition which risks to become too narrow, Burghardt (2001) identified five criteria, all of which must be fulfilled for a behavior to be categorized as play.
For a behavior to be play, (1) it must be incomplete in its function in the present context and include elements that do not contribute to current survival, (2) it must be voluntary, rewarding, pleasurable and done for its own sake, (3) it must differ from functional expressions by being incomplete, exaggerated, awkward, or modified, (4) it must be repeated, but not stereotypically, during ontogeny, and (5) it must be initiated only when not under physical or mental stress.
Applying these criteria, one often discriminates between three categories of play: locomotor play, object play, and social play. While this classification might simplify some aspects of play , it allows for a first systematic description of play in a species. Locomotor play includes all forms of play that involve often exaggerated, locomotor movements, such as running, leaping, or prancing. Object play describes manipulations of non-novel objects such as mouthing, pawing, pushing, or pecking. Finally, social play describes all play that is directed towards another individual. Examples of social play are chasing, play fighting, and nipping (Burghardt, 2005). In this study, we adhered to Burghardt's criteria and analyzed the modes of play according to the above.
Even though play can be a solitary activity it is often associated with sociality in group living species. Social play contexts often represent a safe space where juveniles can practice social norms under more tolerant conditions. It is not unusual that animals display play signals to mark that the actions are not serious (Bekoff, 1972;Byosiere et al., 2016;Palagi et al., 2015;Palagi et al., 2016). Thus, it is not farfetched to hypothesize that play might have parts of its adaptive value in the social lives of group living animals. This hypothesis is supported by neurobiological findings suggesting that species engaging more in social compared to non-social play have enlarged brain areas associated with play in primates (Graham, 2011), and higher relative brain masses in birds (Kaplan, 2020). Thus, play is an interesting phenomenon when studying the evolution of social cognition.
In this study, we aimed at filling parts of the gap in the comparative literature on play by providing the first systematic description of play behaviors in a palaeognath bird, the greater rhea (Rhea americana). Palaeognaths comprise an essential taxon when it comes to understanding avian evolution. Palaeognathae is one of two subgroups of birds and retains many ancestral features that are absent in the other order, the Neognathae. In other words, palaeognaths share more features with the earliest birds on earth and their immediate forebearers: the non-avian dinosaurs (Ksepka et al., 2020;Varricchio et al., 2008;Varricchio & Jackson, 2016). Despite this, their play behaviors have never been studied, though some reports point towards the presence of play in these birds (Bohl, 1970;Franz Sauer, 1969;Hallager, 2010;Timothy, 2019). In the present study, we investigate the presence of all three play categories in greater rheas and analyze mechanisms of play contagion. Furthermore, we describe developmental trajectories of different play categories. Lastly, we use our findings to hypothesize about the evolution of play in early birds.

Subjects, location, and observational period
In this study, we observed a captive group of greater rheas consisting of two adult males, two adult females and seven parent-raised juveniles (sex unknown) at Ystad Zoo, Sweden. The animals were observed using continuous recording with a video camera to ensure the capture of all play behaviors. Observational sessions took place between 9 am and 4 pm and lasted for 3-4 h per session. The observations were made in the rheas' summer enclosure, a large pasture shared with llamas and capybaras between mid-August and the end of September 2020. Our focus was on the juveniles, but adult behaviors were also recorded to compare frequency of play and play contagion between the two age classes. Individual recognition of the juveniles was not possible. The juveniles hatched around the 9th of July 2020 (day of first sighting) and were thus about 6 weeks old at the study onset. On the last three days of observation, one chick was limping and did not engage in play.

Video coding
In total, 42 h of video material was recorded out of which the subjects were observable for 38.36 h. An ethogram of all observed play behaviors was created and used in the video coding (for descriptions, see Results and Supplementary Material). 233 instances of play were recorded. All play occurred spontaneously without any interventions from the experimenter. In addition to the play behaviors, we also coded the number of individuals involved, the direction of their movements (congruent or incongruent to observed movement direction) and contagion of play. A play bout was considered contagious when at least one other individual started to play during or within 3 s after another individual's play bout. We further noted whether the elicited play category in contagious bouts was congruent with the demonstrated category.
The video material was coded in the software Solomon Coder (Version: beta 19.08.02). Half of the videos were coded by CZ and the other half by TRJ. Interrater reliability was determined by cross-coding 10% of the other's video material respectively. Agreement was excellent for all coded categories (ICC = 0.963).

Statistical analyses
Statistical analyses were executed using generalized linear mixed models (GLMMs) with the glmer function of the lme4 package (Version 1.1-26) in RStudio (Version 1.4.1717;RStudio Team, 2020). Count data on occurrences of play behaviors were analyzed using Fisher's exact test. Six GLMMS were fitted with duration of play bout (model1), latency of contagion (model2), number of individuals joining (model3), direction of movement (model4), contagion (model5), and congruency of demonstrated and elicited play category (model6) as response variable, respectively. Models 1-3 were run with a Gamma distribution, while models 4-6 were fitted with a binomial distribution. We added category of play and initiator (adult or juvenile) as fixed factors for all models. For model4 we included number of individuals as fixed factor. We added observation session as a random factor for all models to control for daily differences in the birds' behaviors. We reduced the full models using the Akaike Information Criterion (AIC), to identify the models explaining most variance. We used likelihood ratio tests on the final models to determine the effects of the remaining factors.

Systematic description of play behaviors
Two categories of playlocomotor and social playwere observed. Four different types of locomotor play were identified. The most common was play running, which is running without any obvious goal direction, i.e., not ending at a location with food or parent, or without any biologically relevant cue for locomotion such as fleeing, or following a parent. This type of running was often performed in circles. Moreover, play running was frequently accompanied by neck swinging, a snake-like movement of the neck, and wing display, wing flapping while running. Play running was seen in both adults and juveniles. Additionally, the juvenile rheas engaged in leaping, where they jump straight up, often while throwing their necks from side to side. Play running was moreover highly contagious and thus contained a social component. However, due to the lack of active interactions, this type of play was categorized as locomotor play.
Social play was only observed in juveniles. Only play bouts including an active interaction were categorized as social play, including interactive variants of play running.
Chasing was a pursuit that ended with reaching another individual during play running in both individuals. During play runs, they were also bumping, meaning that they were running into each other in the process of play running. These two behaviors could be accidental, however, both coders independently interpreted them as intentional interactions. Furthermore, the subjects were observed pecking one another, mainly in the neck area. This did not appear to be an aggressive interaction as the other individual did not try to avoid being pecked. Pecking was only half of the time reciprocated. The most interactive form of social play observed was wrestling. In this behavior, two individuals were lying next to each other on the ground with their necks intertwined, mutually pecking each other in the neck area and pushing against each other as if trying to roll the other one on its side (see pictures of described behaviors in Figure 1).
We only recorded one instance that might be suggestive of object playan adult repeatedly pecking a detached feather. However, this observation did not fulfill Burghardt's fourth criterion of play: the behavior did not occur repetitively. For that reason, we cannot conclude the presence of object play in greater rheas from our observations. We therefore excluded this category from further analyses.

Descriptive statistics of play behaviors
Juveniles initiated play significantly more often than the adults did (adults: 19 times, juveniles: 214 times; Fisher's exact test: p < 0.001). Nevertheless, more individuals joined play bouts initiated by an adult compared to a juvenile (GLMM, χ² = 6.275, df = 1, p = 0.012).
The average play bout lasted 9.92 s. The predominant category of play was locomotor play (89.7%, 208 instances), followed by social play (10.3%, 24 instances; for more descriptive statistics of play categories see Table 1 and Figure 2).
As described above, contagious play running was rated as locomotor play due to a lack of interaction. Locomotor play could thus either occur solitarily (43% of locomotor play bouts, 90 instances) or was contagious (57% of locomotor play bouts, 118 instances). In contagious bouts of locomotor play, on average 3.76 individuals joined the initiator. Social play was limited to two participants, except for two incidents where a third individual joined. Thereby, significantly more individuals joined a locomotor compared to a social play bout (GLMM, χ² = 12.82, df = 1, p = 0.00034). Social play bouts lasted  significantly longer than locomotor play bouts (GLMM, χ² = 18.158, df = 1, p < 0.001, see Figure 3(A)).

Contagious play
56.7% of all observed play bouts were contagious, i.e., they induced play in another individual (132 instances). Other individuals joined on average 2.87 s after the play initiation. No significant difference in the latency of contagion was found for play categories  Contagious play bouts lasted significantly longer than non-contagious ones (GLMM, χ² = 38.297, df = 1, p < 0.001, see Figure 3(B)). No significant effects of play category (GLMM, χ² = 0.039, df = 1, p = 0.844) or initiator (juvenile or adult; GLMM, χ² = 0.142, df = 1, p = 0.706) on contagiousness could be identified, meaning that locomotor and social play were equally contagious. The most frequently elicited category of play was locomotor play (90.9%, 120 instances), in the rest of contagious play bouts, social play was elicited (9.1%, 12 instances). Locomotor play was moreover significantly more likely to elicit the same category of play than social play (GLMM, χ² = 9.978, df = 1, p = 0.0016, see Figure 4). While 97% of contagious locomotor play elicited locomotor play in conspecifics, only 71% of contagious social play evoked the same category in others. In the other instances, observing siblings play socially induced locomotor play in individuals not directly involved in the social play.

Ontogeny of play
Lastly, an ontogenetic effect on play categories was found. A significant connection could be identified between play category and the subjects' age (Fisher's exact test, p = 0.00046). While locomotor play was observed in every session, social play first occurred one month after the onset of the study period at around 10.5 weeks of age (see Figure 5).

Discussion
This study includes the first systematic description of play behaviors in greater rheas and is thereby to our knowledge the first such description of play in any palaeognath bird. These birds exhibit two categories of play: locomotor and social play.
Locomotor play was by far the predominant category of play. Within this category, juvenile rheas engaged in play running while flapping their wings in a wing display, swinging their necks, and leaping. These behaviors are in line with the first report on rhea locomotion that described a static posture of the neck in non-social running, but accompanying neck and wing movements in social contexts (Raikow, 1968). In most neognath birds, locomotor play revolves around flight, such as soaring in Montagu harriers (Circus pygargus; Pandolfi, 1996), or play flight in juvenile common ravens (Corvus corax; Heinrich & Smolker, 1998). Due to the loss of volant flight in rheas (along with several other palaeognath bird species; Harshman et al., 2008), this obviously does not lie within their repertoire. It is therefore unsurprising that their locomotor play is mainly based on variations of running, similar to many mammalian species such as horses and deer (Carter et al., 2019;McDonnell & Poulin, 2002).
On the first glance, play in juvenile greater rheas appeared to be predominantly solitary in the sense that it was not directed towards another individual. Even though other individuals joined in the majority of locomotor play bouts, direct social interactions during play running were rare and limited to some possibly accidental bumps through running into each other. In human developmental psychology, this type of play is called parallel play, i.e., several children engaging in solitary play in vicinity of each other without interacting (Parten, 1932). Parallel play is commonly categorized as a form of solitary play due to the lack of interaction, even though it takes place in a social setting. In children, this type of play is usually regarded as a developmental stage towards social play (Bakeman & Brownlee, 1980).
Playing parallelly moreover includes an element of contagion where observing another individual play facilitates play in the observer. This effect was clearly seen in our study with almost 60% of play bouts being contagious. At least two possible mechanisms can cause this effect: behavioral synchronization, i.e., the release of speciesspecific motor patterns triggered by the observation of the latter in conspecifics or emotional contagion, i.e., matching emotional states through the spread of a playful mood (De Waal, 2008;Niedenthal, 2007). The latter is regarded as a building block of empathy (Preston & De Waal, 2002). Osvath and Sima (2014) suggested that play can be used to disentangle the two phenomena. A match between the category of play that is observed by an individual and the category of play that might thereafter be evoked in this individual would point towards behavioral synchronization, while a mismatch would be an argument for emotional contagion. According to this framework, our findings could not exclude mere behavioral synchronization, as demonstrated and elicited play categories were highly congruent, i.e., observing locomotor play mainly elicited locomotor play in other individuals. Nevertheless, by adopting a different, but related, approach in the analyses, juvenile rheas appear to not merely synchronize their behavior. In a response to e.g. danger through predation, animals exhibit the tendency to collectively move in the same direction (Couzin, 2009). Contrarily, when analyzing the direction of running during contagious play bouts in greater rheas, we found that in more than one third of those, the birds moved in individually different directions. This indicates that rather than behavioral synchronization, play running behavior could have been guided by the spread of a playful mood. Alternatively, locomotor play might serve as training for anti-predatory responses. Thus, running in different directions might be a form of protean behavior, i.e., an irregular behavior that prevents prediction by predators (Humphries & Driver, 1970). Moreover, they might practice flight responses through self-handicapping, i.e., instead of moving in the same direction they are trying out individual and less efficient movement patterns (Spinka et al., 2001). More studies specifically targeted towards mechanisms of synchronization are needed to understand the species' behavior and the underlying potential for emotional contagion.
One instance suggestive of object play in an adult was observed. However, this does not necessarily mean that juvenile greater rheas do not engage in object play. It is possible that locomotion is simply easier to observe in a large enclosure with high vegetation, than interactions with objects on the ground. The absence of observations in juveniles might also be caused by a lack of conspicuous objects in the enclosure. In pilot experiments on play behaviors in greater rheas, young adults played intensely when provided with familiar, but unusual, objects (in this case a glove, personal observations of MO). It might thus be possible that object play emerges at a later developmental stage, which could also explain why the only instance was recorded in an adult. It is common for different play categories to develop at different times (for example, see below). Alternatively, greater rheas might not engage in object play at all. The described instance of object play might represent a case of explorative behavior. More studies on greater rheas will be needed to explore whether object play is part of the species' play repertoire and when it develops.
Play fighting is interesting with respect to its role in the development of socio-cognitive skills. It requires social tolerance, self-handicapping for the sake of the game, turn-taking, and role-reversals (Burghardt, 2005). It is thus simultaneously a form of restrained competition and cooperation (Pellis & Pellis, 1998). Awareness of all these social signals and of one's own actions takes time to develop. Play has an ontogenetic trajectory with more complex play behaviors, such as social play, occurring later in the development. The development of social play in rats has been described as an 'inversed U-shape' (Panksepp, 1981). The first occurrences have been recorded between day 13 and 17 followed by increasing frequencies up to approximately day 30-40 before decreasing towards sexual maturity (Müller-Schwarze, 1966;Pellis & Pellis, 1990;Thor & Holloway, 1984). First occurrences of running and jumping in rats have however already been recorded on days 8-12 (Baenninger, 1967). Meerkats (Suricata suricatta) first exhibit object play in week 4 before engaging in social play such as wrestling with litter mates in week 5 and then fine-tune their social play skills through learning play-soliciting signals during weeks 9-10, which finally leads to a peak in social play with other juveniles and yearlings at 11-14 weeks (Doolan & Macdonald, 1999). In cheetah cubs (Acinomyx jubatus), the frequency of locomotor play peaks before the frequency of social play. The authors argued that this pattern serves to optimize anti-predatory responses during the most vulnerable developmental periods of the juveniles (Caro, 1995). In the present study, we observed the same developmental pattern within the juvenile greater rheas. They predominantly exhibited locomotor play, perhaps to practice flight responses to predators. Object play was not observed in the juveniles during the study. Social play developed later in the observational period.
However, an alternative strategy has been described for spotted hyena cubs (Crocuta crocuta) where social play develops in week 2, while locomotor play only occurred in week 3 and object play in week 4. Interestingly, the cubs are very aggressive in their natal den and the emergence of social play coincides with their move to a communal den. The authors therefore conclude that the development of social play facilitates sociality in the clan (Drea et al., 1996). A similar development has been found in howler monkeys (Alouatta palliata; Carpenter, 1934). Ravens already engage in object and social object play (i.e. co-manipulation of objects) in the nest, while locomotor play only develops after fledging . Similarly, Australian magpies begin to play with objects in the nest, though their social play only develops after fledging (Pellis, 1981b).
These accounts of different play categories emerging at different developmental stages give the impression of play functioning as practice at appropriate times in the species' life history. However, as noted above, the functions of play remain unclear. When specifically testing for such training effects, some studies show that social play in juveniles can indeed influence adult social behavior (e.g. Blumstein et al., 2013;Nunes, 2014;Perret, 2021). However, many attempts to prove training effects have failed (e.g. Caro, 1980;Sharpe, 2005). Alternatively, empirical evidence suggests that juvenile social play improves social competence through shaping executive functions governed by the prefrontal cortex (e.g. Baarendse et al., 2013;Bell et al., 2010;Burleson et al., 2016;Schneider et al., 2016;Stark & Pellis, 2020). Thus, play in the juvenile period subsequently improves socio-cognitive skills and emotion regulation (Pellis et al., 2014).
While these studies grant insights into some functions of play, they do not shed light on its evolutionary roots. To achieve this goal, one can turn to reptiles, which belong to the same main lineage as birds (Sauropsida). Studies on play in reptiles are scarce, though the current state of knowledge is that their predominant mode of play is object play, as described in Komodo dragons (Varanus komodoensis; Burghardt et al., 2002), Nile softshelled turtles (Trionyx triunguis; Burghardt et al., 1996) and sea turtles (Caretta caretta and Chelonia mydas; Mann & Mellgren, 1996). Komodo dragons have moreover been reported to engage in tug-of-war social play with zookeepers (Burghardt et al., 2002). Other descriptions of social play in reptiles include behaviors whose functions are not fully understood and thereby cannot certainly be defined as play, such as wrestling in African chameleons (Chamaeleo africanus; Burghardt, 1982), head-bobbing in fence lizards (Sceloporus undulatus; Roggenbuck & Jenssen, 1986) and precocial courtship behaviors in emydid turtles (Kramer & Burghardt, 1998). Locomotor play seems to be rare and only some anecdotal reports on a wood turtle (Clemmys insculpta) repeatedly sliding down a board into water exist (Burghardt, 2005). Despite the limited literature on reptilian play, it appears that their play differs considerably from our observations on palaeognaths. It therefore appears that play has evolved independently several times in different clades.
Within their own archosaurian lineage however, inferences can be drawn from palaeognath play to non-avian dinosaur play. Palaeognaths retain many features of non-avian dinosaurs. Ksepka and colleagues (2020) discovered that relative brain sizes of theropod dinosaurs and early birds, including palaeognaths, are similar. Moreover, fossil evidence from animals that died on their nests and the associated clutch sizes suggests that the parental care of troodontids and oviraptorosaurs (both maniraptoran dinosaurs) are akin to the reproductive strategies of palaeognath birds with males incubating eggs from several females and taking care of the young after hatching (Varricchio et al., 2008;Varricchio & Jackson, 2016).
Due to evidence for play from both extant archosaurian lineages-crocodylians (Dinets, 2015) and birds (e.g. Diamond & Bond, 2003), one can unproblematically assume that non-avian dinosaurs played too. When trying to infer play behaviors more specifically, the above-mentioned shared features of palaeognaths and non-avian dinosaurs indicate that fundamental behaviors present in these birds might have also been present in their extinct relatives, at least in the clade containing birds and their very closest relatives, like the dromaeosaurids ('raptors')the paravians. The similar parental care systems indicate comparable social ecologies for juveniles of paravian dinosaurs. This moreover impliesin contrast to other reptilesa noticeable component of sociality in their play. This hypothesis is supported by our finding that the studied palaeognaths showed social aspects in their play already from the study onset at about 6 weeks old through extensive play contagion. They moreover exhibited interactive social play at 10.5 weeks old which is relatively early in their development. As a reference, these birds reach sexual maturity at 20-24 months (Sales, 2006). Their full potential of social play might therefore not even have completely unfolded at the end of the observational period. For these reasons, probable social play in paravian dinosaurs can be inferred from both phylogenetically and ecologically close extant species.
To yield more solid hypotheses about early birds and non-avian dinosaurs, more studies on different palaeognath species are needed to infer whether play observed in rheas is representative for this clade. Especially the study of species with different social systems (e.g. solitary species like the cassowary), and ecologies (e.g. flighted tinamous) is crucial to obtain an overview of different play conditions within Palaeognathae. Additionally, data from basal neognaths such as red junglefowl would allow for more conclusions on the ancestral state of play in birds. Finally, the discovery of species not engaging in play would shed more light on phylogenetic distribution of play and thus the pattern of its evolution.