An ethogram of facial behaviour in domestic horses: evolutionary perspectives on form and function

Collection of video footage

Subjects were 36 adult horses (12 mares, 24 geldings) of mixed breeds and aged 5–19 years (11.3 ± 3.8 yrs; mean ± SD), housed as Sparsholt College Equine Centre, Winchester, UK. This sample size is in line with that of previous facial expression work using network analysis techniques (Rincon et al., 2023). Recordings were made between September and December 2020, between 10:00 and 16:00, on days of routine turnout (horses placed out in a large pasture, with freedom to move around). Management of the animals did not alter for the study. During college term-time they were stabled individually between Monday morning and Friday evening, and turned out over weekends (Fri–Mon). During college holidays, they were turned out full time. Data collection took place during these turn-out periods. Date were collected over 24 days, and the total observation time was 72 h. The number of hours of recording on a given day ranged between two and five hours. Horses were turned out in two established same-sex groups; mares and geldings. The field into which each group was turned out changed periodically throughout the study. Group sizes and composition varied between recording sessions, due to changes in horse availability. The group (mares or geldings) recorded on a given day was pseudo-randomly selected, with consideration given to the number of horses within each group. Because of the temporal variation in group sizes and individuals present over the data collection period, we were not able to pre-determine the observation schedule. Initially groups were selected randomly, using a random number generator, and then adjustments were made during later randomisation in order to balance the number of observations of each individual as much as possible.

Two researchers conducted 30 min focal samples, and subject order was pseudo-randomly selected; observations were balanced across the season and time of day as much as possible, given horse availability varied throughout the study. During a focal observation, the researcher stood between approximately 10–20 m from the focal individual (the exact distance varied depending on the location of other individuals and environmental features, and any movement of animals during recording), and kept a Panasonic HC-VXF1 camcorder, mounted on a tripod, with pre-record function enabled, trained on the focal animal. The camera was used in HD mode in order to generate high quality footage for FACS analysis, and the zoom function enabled the researcher to keep the camera trained specifically on the focal animal and their close neighbours. Upon a social interaction between the focal horse and one or more conspecifics, the experimenter began recording. A social interaction was defined as when horses moved to within one horse length of each other and/or when an animal made a noticeable signal towards another individual. During recording, the researcher aimed to keep the face of the focal animal in shot. Recording continued until the interaction between the animals was complete. This was defined as either when the horses withdrew (moved to > one horse length from each other) or there was a period of social inactivity (horses made no noticeable signals towards one another) of > 3s. In cases of prolonged interaction, it was sometimes necessary to stop and restart recording mid-interaction to prevent video files becoming too large. This was done on an ad hoc basis (typically if an interaction lasted > 60s), and recording was re-started immediately. A GoPro Hero 5 (wide-angled; GoPro, San Mateo, CA, USA), attached to the tripod arms, provided secondary coverage of behaviour for coding behavioural context. Opportunistic recording of interactions between non-focal animals were conducted during periods of focal animal inactivity.

Behavioural context analysis

Videos were initially viewed using Windows Media Player (Microsoft^®, Bellingham, WA, USA). The behavioural context of each interaction was determined using a predefined social ethogram (Table 1). This ethogram represents an amalgamation of three widely used equine social ethograms (Christensen et al., 2002; McDonnell & Haviland, 1995; McDonnell & Poulin, 2002), so as to cover the whole range of potential social interactions, encompassing affiliative (non-play), play, agonistic and attentional contexts. Affiliative behaviours are friendly, peaceful interactions between individuals (del Toro & Nekaris, 2019). Play involves activities which generate a sense of pleasure and elements of surprise, which appear to have no immediate function (McDonnell & Poulin, 2002). Agonistic behavior encompasses aggression, threat, appeasement, and avoidance behavior between conspecifics (McDonnell & Haviland, 1995). Attentional contexts were defined here as the purposeful orientation of two or more sense organs towards a stimulus to garner information. For each interaction, we noted: (1) the identity of the subjects involved, (2) the behaviour of each subject according to our established ethogram, (3) the time (s) of any facial behaviour produced by any of the individuals performing a social behaviour. If the face of an individual was not visible (either out of sight or unclear) for most or all of an interaction (approx. > 50%), their data were discarded as facial behaviour analysis would not be possible. Where multiple behaviours occurred in quick succession, e.g., during a play bout, each of these was coded separately. In situations where behaviours from multiple contexts were present, these were coded as that of the behaviour most associated with the face. For example if rear and nip occurred together, this was coded as nip. To ensure consistency, 10% (n = 81) of behaviours were coded a second time by the same observer (>2 years later). Intra-observer test-retest reliability was calculated using Cohen’s kappa and indicated substantial agreement between coding at the two time points, κ = .74, p < .001. Raw data are available at https://osf.io/zmsvx/.

Analysis of facial behaviour using EquiFACS

The above process identified 1,326 facial behaviours potentially suitable for coding using EquiFACS; however, some behavioural contexts for some horses were over-represented. To prevent the analyses being biased by these contexts and individuals, a maximum of five facial behaviours for an individual horse for a given behavioural context were selected. These behavioural contexts were balanced for date, time of day, and conspecific identity wherever possible, with the aim to avoid clustering of examples within a relatively short time frame or with the same social partner. Facial behaviours were randomly selected where there were multiple examples from an individual within one focal sample or on one day. For example, if, for horse one, 12 occasions of the behaviour ‘displace’ had been identified, five instances of these would be pseudo-randomly selected for FACS analysis. This threshold of five was chosen to reduce over-representation whilst still allowing for within-context individual variation in facial behaviour to be captured. Following this process, 805 facial behaviours remained for EquiFACS coding.

A certified EquiFACS coder (KL) conducted video coding using open source software BORIS (Friard & Gamba, 2016). For each facial behaviour, the peak of the facial behaviour during each behavioural context was identified, by selecting the point where the greatest number AU/ADs were activated and appeared to be at their greatest intensity. In cases where the facial behaviour was held static for any length of time, the middle of this period was selected as the peak. Analyses were conducted over a 1s period surrounding the peak (0.5s either side). The peak of the facial behavior always fell within the performance of the behavioural context, however in some cases the upper and/or lower bounds of the 1s time bracket would fall outside of the behavioural context, particularly when behaviours were short-lived. Coding over a 1s window allowed for the dynamic nature of facial behaviour to be captured, which using a still image does not. We appreciate that not coding for the entirety of a behaviour may have resulted in some AU/ADs being missed, however FACS coding is extremely time intensive and, as such, 1s represents a compromise between precision and time efficiency. Videos were viewed at both full speed and at 0.1x speed. AUs and ADs were coded as having occurred (1) or not having occurred (0) during the 1s period. If an AU/AD was not visible at any point during the clip—due to head orientation, poor focus, or blurring due to the speed of movement—and its use could not be confirmed when it was visible or from resulting facial movements, it was marked as out of sight. Videos were compared with images of the focal horses’ neutral face, wherever this was available, to allow for more accurate coding. Neutral images, i.e., when horses were relaxed and demonstrating no discernible facial behaviour, were obtained during data collection by opportunistically recording horses during periods of inactivity. AUs and ADs contained within the EquiFACS manual (Wathan et al., 2015), as well as the supplementary behaviour and head movement ADs described in the supplementary material, were initially coded (for a full list see Table 2). AD50 (vocalisation) was later excluded as it was not usually possible to identify the specific animal a vocalisation originated from. Both AD133 (blow) and AD38 (nostril dilator) were coded as AD38 as they were often indistinguishable. To ensure accurate coding, the coding was completed by a certified EquiFACS coder and verified by an independent EquiFACS certified coder who coded 5% (n = 60) of a larger bank of videos (n = 1181), of which the data used here are a subset (33 of these 60 were videos used in the present study). Inter-coder reliability was determined using the following equation, as recommended for use in human FACS (Ekman & Friesen, 1978): ${\displaystyle \frac{2\left(\#AUandADsagreedbybothcoders\right)}{\left(\#AUandADscodedbycoder1\right)+\left(\#AUandADscodedbycoder2\right)}.}$

This calculation gives an agreement for each facial behaviour between 0 and 1 (0 = no agreement, and 1 = absolute agreement). Agreement was 0.98, which is well above the criteria required to become a certified FACS coder (0.70).

Statistical analysis

Network analysis techniques, designed for the analysis of FACS data (NetFACS), were used to analyse the data in R v.4.2.3 (R Core Team, 2023) using the package NetFACS v. 0.5.0.9001 (Mielke et al., 2021). Any AU/ADs used less than five times across the whole dataset (n = 8, Table 2) were removed, as rare actions create a high degree of uncertainty in the bootstrapping process when using NetFACS (Mielke et al., 2021). Data were initially categorised into four broad interaction contexts; affiliative (non-play), play, agonistic, and attentional. Network analyses were performed, using the multi.facs function, on each interaction context to establish the AU/ADs which occurred significantly (P < 0.01) more often than observed across all other contexts. AU/ADs were considered ‘frequently used’ if they occurred in ≥ 40% of facial behaviours within a given context. AU/ADs that occurred in < 10% of the data were omitted from results, as no meaningful interpretation was possible. Bipartite networks, that show the paired combination of AU/ADS that are co-activated significantly more than chance, were plotted for each interaction context using the netfacs_network function and modified using the ggraph v.2.1.0 package (Pedersen, 2022). In co-activation networks, connections between AU/ADs were significant if they occurred at a higher probability in the selected behavioural context than would be expected of that combination across all other contexts. Thus it is possible for an AU/AD combination to be significant whilst the individual AU/ADs that make it up are not, as the baselines for calculating significance differ.

The specificity with which AU/AD was associated with an interaction context was calculated using the specificity function. Due to an imbalanced number of observations across interaction contexts, contexts with fewer observations were randomly upsampled prior to the specificity calculation, to correct for any bias in the specificity results from an imbalanced dataset (Rincon et al., 2023). Specificity is the conditional probability of an interaction context given that an AU/AD is observed. It ranges from 0 (when an AU/AD is never observed in a context) to 1 (when an AU/AD is only observed in one context). Low specificity values indicate that an AU/AD was used flexibly across multiple contexts, whereas high values indicate that an AU/AD was used primarily in a single context. The probability of occurrence for each AU/AD in each interaction context were extracted from interaction context networks. The probability of occurrence is the conditional probability that a particular AU/AD will be observed in a given context, and ranges from 0 (when an AU/AD never occurs in a context) to 1 (when an AU/AD always occurs in a context). Low values indicate that an AU/AD is rarely used in a particular context, whereas high values indicate that an AU/AD is present in almost all instances of that context. Bipartite networks, showing how single AU/ADs are connected to the four interaction contexts, were plotted for context specificity and probability of occurrence, weighted by AU/AD specificity and probability of occurrence respectively.

Network analyses were also performed on each behavioural context to establish the AU/ADs which occurred significantly (P < 0.01) more often in that context than by chance. Bipartite networks were plotted, as described above. The significant AU/ADs and the network plots, in combination with observation of video footage, were utilised to produce written descriptions of facial behaviours typical of each behavioural context, in order to create our facial behaviour ethogram. Images illustrating these typical behaviours were also identified from video footage. R code is available at https://osf.io/zmsvx/.

Ethics statement

This study was observational and carried out in accordance with the recommendations in the Association for the Study of Animal Behaviour and the Animal Behaviour Society guidelines for the use of animals in research (ASAB Ethical Committee/ABS Animal Care Committee, 2023). The study was approved by the University of Portsmouth Animal Welfare and Ethical Review Body (Application No. 1219C).

An extension to EquiFACS: AUH21

During analyses, it was noted by researchers that there were changes occurring in the appearance of the equine face during certain interactions (most notably in agonistic and play contexts) that could not be coded using any AU/AD currently available in EquiFACS, yet could play an important role in equine facial behaviour. We thus propose the addition of AUH21 to EquiFACS (Table 3). Personal communication with mammalian facial anatomy and animal FACS expert Professor Anne Burrows, established that the anatomical origin of the observed appearance change was the action of the platysma muscle. In horses, the platysma is a large superficial muscle of the neck and lower face which primarily functions to shiver off flies. Importantly here, the contraction of this muscle causes tension in the lower face in equines, making the underlying structures appear more prominent. Once defined, all facial behaviour videos were reassessed, using the same method, for the inclusion of AUH21. A subset of videos (30%, n = 238) were second coded for AUH21 by an independent certified EquiFACS coder. The agreement across the two coders, calculated using the same formula as described in the methods, was 0.64. This is below the agreement needed to become a certified FACS coder (0.70), however both coders noted that AUH21 is more difficult to code than the other AU/ADs as it is not based on a specific movement of part of the face, rather it is coded when structures appear to become more visible. Visibility can be affected by lighting conditions and the animal’s coat (see Table 3 for more information), in a way that discrete movements are not, resulting in an AU which is more difficult to identify than the majority, if not all, of those in the original EquiFACS.

Table 3:

Summary of action unit AUH21 in EquiFACS compared to Human FACS.

Action unit	Muscle/s	In human FACS	Appearance changes	Considerations when coding
AUH21 Facial tightener	Platysma (a large superficial muscle of the neck and lower face)	Same muscle as AU21, however visual appearance is different due to anatomical differences.	There are no facial movements associated with the AU. Instead, AUH21 is coded if the muscles and structures on the side of the face, between the cheek and muzzle (see region highlighted), become more prominent/visible, or if there is noticeable tightness or tension in this region. These structures may also become more prominent during chewing due to the action of the masseter muscle; if AD81 is present, do not also code AUH21.	AUH21 can be more challenging to identify if the subject: -Has a longer coat and/or is not clipped -Has their mouth open -Is of a stockier breed/is a coldblood type -Has a very dark coat, or the coat is dappled/spotted Lighting should also be taken into consideration when coding AUH21, as this can make the structures of the face appear more or less prominent. Problems with the teeth and/or damage to the inside of the cheek may also cause this region to become more prominent/swollen, and should be considered when coding.
Neutral face			AUH21 activated
			Note the more prominent muscles (zygomaticus and depressor labii inferioris) and tightness of skin across the marked region.
			Note the more prominent muscles (zygomaticus and levator labii superioris) due to the tautness of the overlying skin.
			Note the prominent muscles (zygomaticus and depressor labii inferioris), despite the increased length of coat, and the tightness of the skin the marked region.

DOI: 10.7717/peerj.19309/table-3

Overview of interaction contexts

We generated a bank of 805 facial behaviours recorded during conspecific interactions, between 36 domestic horses, in two established social groups. These were categorised into four discrete interaction contexts; affiliative (non-play) (n = 178), play (n = 357), agonistic (n = 157), and attentional (n = 113).

The context specificity for single AU/ADs in each interaction context were plotted as a bipartite network, showing how single AU/ADs are connected to the four interaction contexts (Fig. 1). Although there were some AU/ADs that were highly specific to one context, the majority of AU/ADs were used flexibly across multiple different contexts. For clarity, edges are only shown in Fig. 1 where an AU/AD was used in > 10% of occurrences. In reality, almost all AU/ADs were observed in all four contexts at least once. Attentional and play contexts have a number of AU/ADs specifically associated with them. For example, the ears being adducted (EAD102L+R) is highly specific to attentional contexts. In contrast, there are no AU/ADs highly specific to agonistic and affiliative interacts.

The probability of occurrence for single AU/ADs in each interaction context were also plotted as a bipartite network, showing how single AU/ADs are connected to the four interaction contexts (Fig. 2). For each context there were a number of AU/ADs that were highly likely to occur in that context. For example, we can see that lower lip depressor (AU16), lips parted (AU25), ears rotated backwards (EAD104L+R), head down (AD54) and nose forward (AD57) all have high probabilities of occurring during play. There are also many AU/ADs that are relatively rarely used in any context, such as upper lid raiser (AU5) and jaw drop (AU26).

Affiliative (non-play) interactions

When looking at affiliation overall, one head movement was frequently (i.e., used in ≥40% of interactions) used in affiliative (non-play) contexts; nose forward (AD57) (Fig. 3). The two less frequently used (used in <40% of interactions) face and head movements significantly associated with affiliative (non-play) contexts were blink (AU145) and half blink (AU47). Chewing (AD81) was identified in 13.5% of behaviours made in this context. Our initial ethogram included the behavioural contexts follow, contact (friendly), and groom initiation within the interaction context category of affiliative (non-play), however we did not observe enough grooming initiations (n = 1) for meaningful analyses. An ethogram of the facial behaviour during the remaining contexts is presented in Table 4.

Table 4:

Facial behaviour ethogram of horses during affiliative (non-play) horse-horse interactions.

Behavioural Context	Single AU/ADs significantly (p < 0.01) associated with context (probability of their occurrence)	Examples	AU/AD network, showing co-activation of pairs of AU/ADs. Only connections with a probability of co-activation of >0.30, and that occurred in at least 10% of observations (or at least one observation where n < 10) for that context are shown.	Description
Contact (friendly) n=132 nhorses=23	Ears: None Eyes: AU47 (0.22) AU145 (0.39) Lower face: AU10 (0.23) AU18 (0.24) AU25 (0.52) Head position: AD52 (0.43) AD53 (0.62) AD55 (0.26) AD57 (0.76) Gross behaviour: None			Facial behaviour during friendly contact is not highly specific. Instead there are a number of different movements which may be present, but the combinations in which these appear are variable. The nose is typically pushed forward. The head is often also raised, and/or may be turned right. In around half of cases of contact the lips are parted slightly. This is sometimes accompanied by raising of the upper lip to reveal a slight view of the upper teeth, or the dropping of the lower lip. There may also be some puckering of the upper lip, which extends to make contact with the other individual. Ears may be in any position and often act independently of one another. Other movements we might see are an increase in blinking and/or half blinking.
Follow n=45 nhorses=21	Ears: EAD103L (0.54) EAD103R (0.46) EAD104L (0.60) EAD104R (0.63) Eyes: None Lower face: AD38 (0.43) Head position: None Gross behaviour: AD81 (0.31)			When following another individual, the ears are often rotated backwards and/or flattened downwards, towards the neck. The eyes remain fairly neutral, with no notable actions being utilised. The nostrils will be flared in around 40% of instances. Chewing is regularly observed during bouts of following on pasture, as this behaviour often occurs during grazing.

DOI: 10.7717/peerj.19309/table-4

Agonistic interactions

Six different facial and head movements were frequently used in agonistic contexts overall; left and right ear flattener (EAD103L + R), left and right ear rotator (EAD104L + R), nostril dilator (AD38), and head down (AD54) (Fig. 3). The three less frequently used face and head movements significantly associated with agonistic contexts were inner brow raiser (AU101), nostril lift (AUH13), and facial tightener (AUH21). Chewing (AD81) was identified in 33.8% of behaviours made in this context.

Our initial ethogram included the behavioural contexts displace, head threat, bite threat, kick threat, bite, kick, strike, chase, and push within the interaction context ‘agonistic’. However, we did not have sufficient data to meaningfully analyse the contexts bite threat (n = 3), bite (n = 2), kick (n = 1), and push (n = 1). The facial movements associated with each of the contexts for which there were enough data are presented in Table 5.

Table 5:

Facial behaviour ethogram of horses during agonistic horse-horse interactions.

Behavioural context	Single AU/ADs significantly (p < 0.01) associated with context (probability of their occurrence)	Examples	AU/AD network, showing co-activation of pairs of AU/ADs. Only connections with a probability of co-activation of >0.30, and that occurred in at least 10% of observations (or at least one observation where n<10) for that context are shown.	Description
Displace n=52 nhorses=23	Ears: EAD103L (0.55) EAD103R (0.50) EAD104L (0.74) EAD104R (0.70) Eyes: AU101 (0.30) Lower face: None Head position: AD54 (0.58) Gross behaviour: AD81 (0.54)			Compared with the other agonistic signals, facial behaviour occurring during displacements uses relatively few individual actions. The ears are usually rotated backwards and/or flattened downwards, towards the neck. When the ears are flattened we may also see flaring of the nostrils. It is typical for the head to be lowered. The brow above the inner corner of the eye may be raised. Chewing is common during displace behaviour when horses are at pasture, as it often occurs during grazing. There are some instances of displace where none of these actions are seen, and instead the ears are forward.
Head threat n=52 nhorses=24	Ears: EAD103L (1.0) EAD103R (0.98) EAD104L (0.62) EAD104R (0.63) Eyes: AU101 (0.40) Lower face: AUH13 (0.35) AD38 (0.46) Head position: AD54 (0.58) AD57 (0.64) Gross behaviour: AD81 (0.37)			Facial behaviour during head threats is similar to that of displace, however more AU/ADs are activated. The ears are always rotated backwards and are typically also flattened downwards, towards the neck. The head is often lowered, as in a displace, however the nose is usually also pushed forward, and the head may turn to the left or the right, towards the threat receiver. The brow above the inner corner of the eye may be raised. In the lower face, the nostrils may be flared. Chewing occurs in around a third of head threats when horses are at pasture, as it often occurs during grazing.
Kick threat n=16 nhorses=13	Ears: EAD103L (0.83) EAD103R (0.90) EAD104L (0.83) EAD104R (0.90) Eyes: AU101 (0.36) Lower face: AUH13 (0.36) AUH21 (0.39) Head position: AD51 (0.38) AD53 (0.63) AD54 (0.44) Gross behaviour: None			During a kick threat the ears are almost always rotated backwards and flattened downwards, towards the neck. Sometimes one or both ears will also flick forwards during the behaviour. The brow above the inner corner of the eye may be raised, the top edges of the nostrils may be lifted in the direction of the eye, and there may be tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. The head may be raised or lowered.
Strike n=15 nhorses=8	Ears: EAD103L (0.54) EAD103R (0.67) EAD104L (0.85) EAD104R (0.85) Eyes: AU5 (0.39) AU101 (0.39) AD1 (0.33) Lower face: AU18 (0.43) AUH21 (0.71) AD38 (0.64) Head position: AD51 (0.47) AD52 (0.53) AD53 (0.93) AD55 (0.23) AD58 (0.67) Gross behaviour: None			During strikes the ears are typically rotated backwards and flattened downwards, towards the neck. The head is almost always raised, and typically turns to one side or the other during the behaviour. The nose is usually pulled back, towards the chest, but may on some occasions be pushed forwards instead. In the lower face we often see flared nostrils and tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. In around 40% of cases we also see puckering of the upper lip. The whites of the eyes become more visible in a third of strikes. We may also see raising of the inner brow or raising of the upper eyelid across its length, widening the eye.
Chase n=15 nhorses=7	Ears: EAD103L (0.73) EAD103R (0.75) EAD104L (0.82) EAD104R (0.67) Eyes: AU101 (0.50) Lower face: AUH13 (0.33) AU17 (0.33) AUH21 (0.50) AD38 (0.47) Head position: AD53 (0.60) Gross behaviour: None			During chasing the ears are typically rotated backwards and flattened downwards, towards the neck. The inner brow is raised in around half of occurrences. The head is typically raised, and in some instances the nose is pushed forwards. In the lower face the nostrils may be flared, or the top edges of the nostrils may be lifted in the direction of the eye. In 50% of instances there will be tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. During chasing there is also sometimes puckering of the lip, or the area under the chin may lift upwards, making the chin look more defined.

DOI: 10.7717/peerj.19309/table-5

Attentional interactions/states

Overall, facial movements frequently used in attentional contexts were ears forward (EAD101L + R) and ears adducted (EAD102L + R) (Fig. 3). Less frequently used face and head movements significantly associated with attentional contexts were half blink (AU47), head down (AD54), and nose back (AD58). For attentional interaction contexts, the initial ethogram consisted of the behavioural contexts alert and olfactory investigation. There was sufficient data to meaningfully analyse both of these, and the facial movements associated with each can be seen in Table 6.

Table 6:

Facial behaviour ethogram of horses during attentional contexts.

Behavioural Context	Single AU/ADs significantly (p < 0.01) associated with context (probability of their occurrence)	Examples	AU/AD network, showing co-activation of pairs of AU/ADs. Only connections with a probability of co- activation of >0.30, and that occurred in at least 10% of observations (or at least one observation where n<10) for that context are shown.	Description
Alert n=38 nhorses=16	Ears: EAD101L (1.00) EAD101R (1.00) EAD102L (0.86) EAD102R (0.86) Eyes: AU145 (0.32) Lower face: None Head position: AD53 (0.87) Gross behaviour: None			An alert horse will have a raised head, and their ears are turned forwards and brought inwards towards each other (adducted). In around a third of alert occurrences there is also an increase in blinking. Flaring of the nostrils may sometimes be seen.
Olfactory investigation n=75 nhorses=28	Ears: EAD101L (0.94) EAD101R (0.94) EAD102L (0.46) EAD102R (0.45) Eyes: AU47 (0.34) Lower face: None Head position: AD54 (0.48) AD58 (0.27) Gross behaviour: None			During olfactory investigation the ears are turned forwards, and may also be brought inwards towards each other (adducted). The head is down in around 50% of cases, but can also be in other positions. In around a third of occurrences there is also an increase in half blinking.

DOI: 10.7717/peerj.19309/table-6

Play interactions

The facial and head movements frequently used in play contexts overall were lower lip depressor (AU16), chin raiser (AU17), lips part (AU25), mouth stretch (AU27), eye white increase (AD1), ear flattener (EAD103L + R), ear rotator (EAD104L + R), head turn right (AD52), head up (AD53), and nose forward (AD57) (Fig. 3). Less frequently used face and head movements significantly associated with play contexts were inner brow raiser (AU101), upper lid raiser (AU5), upper lip raiser (AU10), nostril lift (AUH13), lip pucker (AU18), upper lip curl (AU122), lip pressor (AU24), jaw drop (AU26), facial tightener (AUH21), head turn left (AD51), head tilt left (AD55), head tilt right (AD56) and nose back (AD58).

For play interaction contexts, the initial ethogram consisted of the behavioural contexts frolic, leap, run, chase (play), buck, prance, nip air, nip, bite (play), reach, grasp, neck wrestle, push (play), stamp, rear, kick threat (play), kick (play), balk (play), evasive jump, evasive spin, and head snatch. However, we did not have sufficient data to meaningfully analyse the contexts frolic (n = 0), leap (n = 0), chase (play) (n = 1), buck (n = 2), prance (n = 0), neck wrestle (n = 2), balk (play) (n = 0), and evasive spin (n = 4). The facial movements associated with each of the contexts for which there were enough data are presented in Table 7: run, nip air, nip, bite (play), reach, grasp, push (play), stamp, rear, kick threat (play), kick (play), evasive jump, head snatch.

Table 7:

Facial behaviour ethogram of horses during play contexts.

Behavioural Context	Single AU/ADs significantly (p < 0.01) associated with context (probability of their occurrence)	Examples	AU/AD network, showing co-activation of pairs of AU/ADs. Only connections with a probability of co- activation of >0.30, and that occurred in at least 10% of observations (or at least one observation where n<10) for that context are shown.	Description
Run n=8 nhorses=7	Ears: EAD101L (0.63) EAD104L (0.63) EAD104R (0.86) Eyes: None Lower face: AD38 (0.88) Head position: AD53 (0.63) Gross behaviour: None			When running the ears may be turned forwards or rotated backwards and flattened towards the head, and often they will alternate between the different ear positions. The nostrils are flared and the head is typically raised.
Nip air n=81 nhorses=17	Ears: EAD103L (0.48) EAD103R (0.47) EAD104L (0.63) EAD104R (0.64) Eyes: AD1 (0.50) Lower face: AU10 (0.30) AU16 (1.00) AU17 (0.68) AU18 (0.28) AUH21 (0.40) AU24 (0.18) AU25 (1.00) AU26 (0.43) AU27 (0.59) AU122 (0.56) Head position: AD51 (0.40) AD52 (0.51) AD53 (0.83) AD55 (0.44) AD56 (0.42) AD57 (0.90) Gross behaviour: None			In nip air the ear position is variable; they can be rotated backwards, flattened towards the head, or turned forwards. In around half of occurrences there will be an increase in the amount of eye white visible. In the lower face there may be numerous potential actions involved. Importantly, the lips are always parted and the lower lip is pulled down. We also typically see the area under the chin lifting upwards, making the chin look more defined, as well as the mouth being stretched wide open, and curling upwards of the upper lip. Other actions that may occur are raising of the upper lip revealing the upper teeth, puckering of the top lip, pressing together of the lips (after the nip has been completed), dropping of the lower jaw (as opposed to full stretching open of the mouth), and tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. The head is quite mobile during this behaviour. It may be turned or twisted to the left or right, but is usually raised, with the nose pushed forwards.
Nip n=59 nhorses=18	Ears: EAD104L (0.57) Eyes: AU101 (0.37) AU145 (0.35) Lower face: AU10 (0.48) AUH13 (0.26) AU16 (0.98) AU17 (0.30) AU18 (0.20) AUH21 (0.32) AU24 (0.19) AU25 (1.00) AU26 (0.20) AU27 (0.80) AU122 (0.38) AD38 (0.42) Head position: AD52 (0.44) AD55 (0.49) AD56 (0.34) AD57 (0.80) Gross behaviour: None			During nip air the ear position is variable; they can be rotated backwards, flattened towards the head, or turned forwards. In a third of instances we will observe raising of the inner brow. Similarly, there is increased blinking in around a third of occurrences. As with nip air, there may be numerous potential actions involved in the lower face during nip. Importantly, the lips are always parted and the lower lip is pulled down. We also usually see the mouth stretched wide open, with jaw dropping seen instead of this more intense stretching in only a fifth of cases. Other actions that may occur are raising of the upper lip revealing the upper teeth, puckering of the top lip, pressing together of the lips (after the nip has been completed), the area under the chin lifting upwards (making the chin look more defined), curling upwards of the upper lip, and tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. Flaring of the nostrils, or lifting of the top edges of the nostrils in the direction of the eye, may be seen. The head is usually tilted to one side (either left or right). The nose is typically pushed forwards. On occasion we may also see the head being turned right, or the head being raised.
Bite (play) n=34 nhorses=9	Ears: None Eyes: AU47 (0.26) AU101 (0.42) AD1 (0.29) Lower face: AU10 (0.58) AU16 (0.90) AU17 (0.34) AUH21 (0.38) AU25 (1.00) AU27 (0.91) AU122 (0.39) Head position: AD55 (0.38) AD56 (0.41) AD57 (0.82) Gross behaviour: None			During nip air the ear position is variable; they can be rotated backwards, flattened towards the head, or turned forwards. During play biting the mouth stretches wide open, parting the lips, and the lower lip is pulled down so that the lower teeth are visible. The upper lip is often raised to reveal the upper teeth too. Other lower facial actions that may be seen are raising of the chin to make it more pronounced, curling upwards of the upper lip, and tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. The nose is usually pushed forwards, and is usually tilted to one side (left or right). In some instances, there may be raising of the inner brow, an increase in the amount of visible eye white, or an increase in half blinking.
Reach n=26 nhorses=12	Ears: EAD103R (0.44) Eyes: AU47 (0.25) AD1 (0.40) Lower face: AU16 (0.69) AU17 (0.39) AU18 (0.40) AUH21 (0.46) AU25 (0.78) AU122 (0.36) Head position: AD51 (0.31) AD52 (0.58) AD53 (0.69) AD55 (0.39) AD56 (0.31) AD57 (0.96) Gross behaviour: None			When a horse is performing reach, the position of the ears is quite variable and the ears may move independently of each other. The nose is pushed forwards and the head is usually raised. The head is usually turned and/or twisted (both directions are possible). The lips are typically parted, with the lower lip dropped to reveal the lower teeth. This may be accompanied by raising of the chin, which creates a distinct appearance where the lip appears extended and square (see top left and bottom right plates). We may see the upper lip curling upwards or, alternatively, it may be puckered. In around half of occurrences there will be tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. In some instances, we see an increase in visible eye white or increased half blinking.
Grasp n=40 nhorses=13	Ears: None Eyes: AU145 (0.39) AD1 (0.31) Lower face: AU10 (0.24) AUH13 (0.25) AU24 (0.26) AU25 (0.66) Head position: AD55 (0.45) AD56 (0.25) AD57 (0.75) Gross behaviour: None			Grasps can last a significant length of time. As such, facial behaviour is more static than in the other play behaviours sampled, which are short lived and dynamic. The ears may be in any position, but are more likely to be turned forwards and/or flattened backwards toward the neck, and may work independently. The nose will typically be pushed forwards, and the head is often rotated; this is more often to the left, but the right is also possible. The lips are usually parted around the object (skin or rug) being grasped. Alternatively, the lips can be pressed together to create a seal around the grasped object. Other actions which may be observed are the lifting of the top edges of the nostrils in the direction of the eye, an increase in visible eye white, or increased blinking.
Push (play) n=16 nhorses=8	Ears: EAD101L (0.55) EAD101R (0.67) EAD103L (0.60) EAD103R (0.63) EAD104L (0.73) EAD104R (0.73) Eyes: AU145 (0.46) Lower face: AU17 (0.33) AUH21 (0.44) AU24 (0.40) AD38 (0.44) Head position: AD51 (0.33) AD52 (0.53) AD54 (0.47) AD58 (0.40) Gross behaviour: None			In push (play) the ears may be turned forward, rotated backwards, or flattened downwards towards the neck. They may move during the behaviour, so that multiple of these ear positions are seen in succession. There is an increase in blinking in around a half of instances of push (play). In the lower face, we may see the lips pressing together, raising of the chin (making it appear more pronounced), flaring of the nostrils, or tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible. The head may be up or down, and is often turned to one side. The nose is pulled back towards the chest in around 40% of instances.
Stamp n=10 nhorses=5	Ears: EAD101L (0.88) EAD101R (0.80) Eyes: None Lower face: AD38 (0.50) Head position: AD53 (0.70) Gross behaviour: None			During stamping the ears are almost always pointed forwards at some point during the behaviour, although can also be rotated and flattened backwards towards the neck. The head is typically raised, and the nostrils are flared in around a half of instances of stamp. When nostrils are flared, increased blinking may sometimes be observed. Other actions which may be seen on occasion are the raising of the inner brow and the pressing together of the lips.
Rear n=12 nhorses=7	Ears: EAD103L (0.83) EAD103R (0.82) EAD104L (0.75) EAD104R (0.90) Eyes: AD1 (0.91) Lower face: AU16 (0.91) AU17 (0.64) AU18 (0.46) AU25 (0.92) AU27 (0.50) AD38 (0.46) Head position: AD51 (0.68) AD52 (0.58) AD53 (0.83) AD57 (0.83) AD58 (0.42) Gross behaviour: None			When rearing during play the ears are typically rotated backwards and flattened downwards towards the neck at some point during the behaviour. Sometimes the ears may also rotate forwards during the behaviour. Rearing is also characterised by an increase in the amount of visible eye white, the lips being parted (with the mouth being fully stretched open in 50% of occurrences), and the lower lip being moved downwards to reveal the lower teeth. There may also be raising of the chin, making it appear more distinct. This creates a distinct appearance, where the lip appears extended and square, when it occurs in conjunction with the lowering of the lower lip (see bottom right plate). Puckering of the upper lip, and flaring of the nostrils, are also fairly common. The head is usually up, and the nose will typically be pushed forwards at some point during rearing, although it may also be brought back towards the chest within the same sequence. The head is likely to be turned to one side (left, right, or both in succession).
Kick threat (play) n=16 nhorses=6	Ears: EAD103L (0.68) EAD103R (0.64) EAD104L (0.81) EAD104R (0.92) Eyes: AU5 (0.43) AU47 (0.40) AU101 (0.39) AD1 (0.33) Lower face: AUH13 (0.44) AU24 (0.44) AD38 (0.50) Head position: AD53 (0.69) AD54 (0.44) AD58 (0.31) Gross behaviour: None			Kick threats during play are typically defined by the ears being rotated backwards and flattened downwards towards the neck. The head is usually up, but may alternatively be down, and in some cases the head is lowered during the kick action, then raised immediately afterwards. The nose is brought in towards the chest in around a third of instances. With regards to the eyes, there may be raising of the upper eyelid (making the eye appear larger in size), increased blinking or half blinking, raising of the inner brow, or an increase in the amount of visible eye white. In the lower face we may see pressing together of the lips, flaring of the nostrils, or lifting of the top edges of the nostrils, in the direction of the eye.
Kick (play) n=6 nhorses=4	Ears: EAD104L (0.83) EAD104R (1.00) Eyes: None Lower face: AD38 (1.00) Head position: None Gross behaviour: None			During kick (play) the ears are typically rotated backwards, and flattened downwards towards the neck. The nostrils always flare at some point during the behaviour, and the nose is typically pulled back towards the chest. Other actions we may see include the pressing together of the lips, the head being raised, an increase in visible eye white, raising of the upper eyelid (increasing the size of the eye), raising of the chin (making the chin look more defined), or pulling back at the corners of the lips.
Evasive jump n=24 nhorses=10	Ears: EAD101L (0.57) EAD101R (0.67) EAD103L (0.65) EAD103R (0.53) EAD104L (0.78) EAD104R (0.79) Eyes: AU5 (0.37) AU101 (0.39) AD1 (0.91) Lower face: AU10 (0.36) AUH13 (0.38) AU16 (0.56) AU17 (0.62) AU18 (0.50) AUH21 (0.47) AU24 (0.36) AU25 (0.79) AU27 (0.35) AD38 (0.50) Head position: AD51 (0.34) AD52 (0.63) AD53 (0.88) AD55 (0.33) AD57 (0.71) Gross behaviour: None			During an evasive jump the ear position is variable; they can be rotated backwards, flattened towards the head, or turned forwards. There is almost always an increase in visible eye white, a raised head, and parted lips (in a third of instances the mouth is stretched wide open) at some point during the behaviour. The nose is typically pushed forwards and we often see the lower lip pulled downwards (to reveal a portion of the lower teeth), or raising of the chin (making the chin more defined). Sometimes these latter two occur together, creating a distinctive square, elongated shape with the lower lips (see bottom left plate). The head is typically turned to one side; more often this is to the right, but can also be to the left. Sometimes the head may tilt to the left. Regarding the eyes, we may observe raising of the inner brow, or raising of the upper eyelid, increasing the size of the eye. Lower face actions that may be seen during evasive jumps are raising and/or puckering of the upper lip, the pressing together of the lips, flaring of the nostrils, raising of the upper edge of the nostrils towards the eye, or tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible.
Head snatch n=15 nhorses=8	Ears: EAD101L (0.87) EAD101R (0.91) EAD102L (0.50) EAD104R (0.55) Eyes: AD1 (0.47) Lower face: AUH21 (0.40) AD38 (0.43) Head position: AD51 (0.53) AD52 (0.67) AD53 (0.88) AD55 (0.40) AD58 (0.47) Gross behaviour: None			The ears are almost always turned forwards at some point during head snatching. This may be accompanied by the bringing together of the ears centrally. Sometimes one or both ears may be rotated backwards and/or flattened down and back, towards the neck. The head is usually raised, and is often turned to one side (left or right, with right being more common). In around half of occurrences the nose is pulled backwards towards the chest, and there is sometimes tilting of the head to the left. There is an increase in visible eye white in roughly half of instances (particularly when ears are flattened), and we may see flaring of the nostrils or tightening across the side of the face between the eye, cheek and muzzle, making the underlying structures appear more visible.

DOI: 10.7717/peerj.19309/table-7