Available via license: CC BY 4.0
Content may be subject to copyright.
RESEARCH ARTICLE
Are Eyes a Mirror of the Soul? What Eye
Wrinkles Reveal about a Horse’s Emotional
State
Sara Hintze
1,2
*, Samantha Smith
3
, Antonia Patt
4
, Iris Bachmann
2
, Hanno Wu
¨rbel
1
1Division of Animal Welfare, Vetsuisse Faculty, University of Bern, Bern, Switzerland, 2Agroscope, Swiss
National Stud Farm, Avenches, Switzerland, 3Royal (Dick) School of Veterinary Studies, University of
Edinburgh, Edinburgh, United Kingdom, 4Department of Animal and Avian Sciences, University of
Maryland, College Park, Maryland, United States of America
*sara.hintze@vetsuisse.unibe.ch
Abstract
Finding valid indicators of emotional states is one of the biggest challenges in animal wel-
fare science. Here, we investigated in horses whether variation in the expression of eye
wrinkles caused by contraction of the inner eyebrow raiser reflects emotional valence. By
confronting horses with positive and negative conditions, we aimed to induce positive and
negative emotional states, hypothesising that positive emotions would reduce whereas
negative emotions would increase eye wrinkle expression. Sixteen horses were individually
exposed in a balanced order to two positive (grooming, food anticipation) and two negative
conditions (food competition, waving a plastic bag). Each condition lasted for 60 seconds
and was preceded by a 60 second control phase. Throughout both phases, pictures of the
eyes were taken, and for each horse four pictures per condition and phase were randomly
selected. Pictures were scored in random order and by two experimenters blind to condition
and phase for six outcome measures: qualitative impression, eyelid shape, markedness of
the wrinkles, presence of eye white, number of wrinkles, and the angle between the line
through the eyeball and the highest wrinkle. The angle decreased during grooming and
increased during food competition compared to control phases, whereas the two phases
did not differ during food anticipation and the plastic bag condition. No effects on the other
outcome measures were detected. Taken together, we have defined a set of measures to
assess eye wrinkle expression reliably, of which one measure was affected by the condi-
tions the horses were exposed to. Variation in eye wrinkle expression might provide valu-
able information on horse welfare but further validation of specific measures across
different conditions is needed.
Introduction
Animal welfare comprises both physical health and emotional well-being. Whereas physical
health can be assessed directly, the subjective experience of emotions must be inferred [1].
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 1 / 15
a11111
OPEN ACCESS
Citation: Hintze S, Smith S, Patt A, Bachmann I,
Wu¨rbel H (2016) Are Eyes a Mirror of the Soul?
What Eye Wrinkles Reveal about a Horse’s
Emotional State. PLoS ONE 11(10): e0164017.
doi:10.1371/journal.pone.0164017
Editor: Glenn F. Browning, The University of
Melbourne, AUSTRALIA
Received: February 18, 2016
Accepted: September 19, 2016
Published: October 12, 2016
Copyright: ©2016 Hintze et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: All data files are
available from the figshare database (accession
number(s) 10.6084/m9.figshare.2361208).
Funding: SH was funded by Agroscope.
Competing Interests: The authors have declared
that no competing interests exist.
Ideally, indicators of emotional state are non-invasive and can be assessed easily and reliably in
various contexts.
Charles Darwin already observed that ‘man and animals express the same state of mind by
the same movements’ ([2], p.352), highlighting facial expressions as potential indicators of ani-
mal emotions. However, while studied extensively in humans (e.g. [3,4]) and non-human pri-
mates (e.g. [5]), facial expressions have been largely neglected in other species until more
recently (e.g. [6,7]).
In animals, facial expressions have mainly been investigated using Facial Action Coding
Systems (FACS) and Grimace Scales (GS). FACS was originally developed in humans [3] and
relies on the systematic characterisation of facial expression based on individual Action Units
which are caused by contraction or relaxation of one or more muscles. It has been proposed as
a tool to assess pain and distress in neonates and infants who cannot express their feelings ver-
bally [8], as well as for the diagnosis of depression [9]. Based on the approach in humans,
FACS have now been developed for non-human animals, including various primates (e.g.
ChimpFACS [5], OrangFACS [10]), dogs (DogFACS [11]), and horses (EquiFACS [12]). GS
for pain assessment have been developed and validated by comparing facial expressions in
painful and painless conditions in mice (MouseGS [7]), rats (RatGS [13]), rabbits (RabbitGS
[6]), cats [14], sheep (Sheep Pain Facial Expression Scale (SPFES) [15]), and horses (HorseGS
[16], Equine Pain Face [17]).
In horses, two independent studies were performed in which pain was induced by either
surgery [16] or ischemic pain and neurogenic inflammation [17], from which the HorseGS
[16] and the Equine Pain Face [17] were developed. More recently, a FACS for horses (Equi-
FACS) has been established which describes all possible facial movements [12]; however, it has
not been applied to specific emotional contexts (e.g. pain) so far.
One particular aspect of facial expression in horses is the formation of wrinkles above the
eyeball. Eye wrinkles are caused by contraction of the inner eyebrow raiser (levator anguli oculi
medialis muscle and corrugator supercilii muscle) and have been identified as an Action Unit
(AU 101) in EquiFACS [12]. Among horse owners, eye wrinkles are also known as ‘worry wrin-
kles’, suggesting that they may be associated with a negative emotional state. Furthermore, eye
wrinkles were more pronounced in the pain treatment as assessed by the Equine Pain Face
[17], and in the Horse GS a ‘tension above the eye area’ was described in horses that were in
pain ([16], p.4). Thus, present evidence indicates that pain may be associated with eye wrinkles.
Whether the expression of eye wrinklesmay serve more generally as a sign of negative emo-
tional states, has not been studied so far. Supporting evidence stems from studies in humans,
where it was found that the same Action Unit associated with eye wrinkles in horses was associ-
ated with fear and sadness in people [18]. Based on this, we aimed at examining the link
between emotional state and eye wrinkle expression in horses.
For this purpose, we exposed horses to distinctively valenced conditions. Since neither Equi-
FACS nor the Horse GS have looked at eye wrinkle expression in more detail than describing the
presence or absence of the wrinkles, we established a range of outcome measures that can be
scored reliably to assess variation in eye wrinkle expression. We hypothesised that positive emo-
tions would reduce the expression of eye wrinkles whereas negative emotions would increase it.
Materials and Methods
Animals and housing
This study included 16 horses (15 stallions, 1 mare) of three breeds (Franches-Montagnes,
warmblood, trotter) aged between 3 and 20 years (10.4 ± 4.7, Table 1). The sample was ran-
domly selected from the horse population of the stud farm, excluding subjects that were not
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 2 / 15
responding to food competition during pilot observations (see below). All horses were housed
in standard single boxes (3.00 x 3.50 m) at the Swiss National Stud Farm of Agroscope
(Avenches, Switzerland). The stable was divided into six discrete areas. Each area contained
eight horse boxes in two rows separated by an aisle (3 m). All boxes had a window facing into
the adjacent aisle. The 16 experimental horses were distributed across each of the six areas,
with a maximum of four horses from the same area. They were kept on wood shavings or straw
and had visual contact with conspecifics. Horses were fed hay and concentrate three times a
day, and water was provided ad libitum by automatic drinkers. On a daily basis, horses were
exercised (riding, carriage riding), allowed free movement on a sand paddock, or walked in a
horse walker, according to the Swiss Animal Welfare Ordinance [19].
Experimental design
Each horse was exposed to four conditions, two positive (grooming, food anticipation) and two
negative (food competition, waving a plastic bag).
Each condition was divided in two phases lasting 60 seconds each: control phase followed
by treatment phase. Food anticipation, food competition, and waving of a plastic bag did not
involve the presence of an experimenter inside the box, whereas grooming wasperformed by
the experimenter standingnext to the horse inside the box. Consequently, the experimenter
was also present during the control phase of the grooming condition, imitating the movement
of grooming beside the horse without touching it. Throughout the whole duration (= 60 sec-
onds) of both treatment and control phase, pictures of the eyes were taken by two professional
photographers (camera: Canon EOS-1D X, Switzerland; lens: EF 70–200 mm f/.8 L IS II USM,
Switzerland) who were positioned in the aisle. Horses were also exposedto all stimuli but
grooming from the side of the aisle.
A total of ten experimental sessions was conducted during May and June 2014 with five to
eight horses being exposed to one condition per session. Sessions took place during off work-
hours (between11:45 and 13:30) to avoid disruption of thehorses’ attention and emotional
changes caused by management procedures. During all sessions the same two experimenters
and two photographers were present. All 16 horses were exposed to all four conditions, with
Table 1. Sex, age and breed of the 16 horses used in this study.
Sex Year of birth Breed
Stallion 2011 Franches-Montagnes
Stallion 2010 Warmblood
Stallion 2008 Warmblood
Stallion 2006 Franches-Montagnes
Stallion 2005 Franches-Montagnes
Stallion 2005 French Trotter
Stallion 2004 Franches-Montagnes
Stallion 2003 Franches-Montagnes
Stallion 2003 Franches-Montagnes
Stallion 2002 Franches-Montagnes
Stallion 2002 Franches-Montagnes
Stallion 2000 Franches-Montagnes
Stallion 1999 Franches-Montagnes
Stallion 1997 Franches-Montagnes
Stallion 1994 Franches-Montagnes
Mare 2009 Warmblood
doi:10.1371/journal.pone.0164017.t001
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 3 / 15
the order of the conditions being counterbalanced between horses. No more than two condi-
tions per session were presented in the same area of the stable, with the positive conditions
(grooming, food anticipation) always preceding the negative conditions (food competition,
plastic bag) because the positiveconditions affected only the experimental horse, whereas nega-
tive conditions could affect all horses within the same area. In one session, it was unavoidable
to expose two horses within the same area to negative conditions; however, food competition, a
common condition for the horses, preceded the plastic bag condition by nearly 90 minutes,
thereby minimising the risk of contamination.
Before the start of each session,the forelocks of the horses were braided to ensure clear view
of both eyes. Lights were switched on and two additional flood lights were arranged to improve
the illumination of the boxes. Horses were habituated to the floodlights before the start of the
experiment. Habituation was deemed successful when horses did not react to the flood lights
anymore. Windows heading to the aisle were opened to allow the horses to put their heads out
during testing.
Conditions of different emotional valence
Grooming (G). Grooming has been shown to reduce heart rate in horses [20,21] which is
indicative of a more relaxed state and to induce positive behavioural responses, like leaning
into the massage [21]. In the present experiment, horses were groomed continuously for 60
seconds by a familiar experimenter (SH) standing on the left side of the horses and moving the
fingertips of both hands along the withers, the shoulder and the neck. Grooming was preceded
by a 60 second control phase with the experimenter imitating the grooming movements with-
out touching the horse.
Food anticipation (FA). Anticipation is a motivational state between a cue signaling the
arrival of an event and the actual arrival of this event [22,23]. In order to induce a positive
anticipatory state, horses were conditioned to anticipate a special food reward (British Mash
Original, Landmühle, Switzerland, mixed with apples and carrots) when confronted with a
blue and yellow striped bucket for three trials across two days preceding the test day. During
conditioning as well as on the test day, the bucket was shown to the experimental horse, and
after a 60 second waiting period (= anticipation and therefore treatment phase during the
experiment), the horse was given access to the bucket with the food. Horses were trainedindi-
vidually and with a bucket and food different from their usual food to distinguish clearly
between food anticipation and food competition. The 60 seconds preceding presentation of the
bucket were used as control phase.
Food competition (FC). Witnessing other horses being fed their regular rations of con-
centrates while waiting for being fed is stressful for stabled horses [24]. From our pilot observa-
tions we knew that most horses at the stud farm respond strongly to their neighbours being
fed, for example, by shaking the head quickly up and down, flattening the ears, and threatening
the neighbour. We used the common procedure of feeding concentrate to the neighbour horses
to induce frustration in the experimental horse. Moreover, instead of horses being fed between
11:00 and 11:30 by thefarm workers,feeding wasdelayed by about one hour toincrease frus-
tration [25]. Although both control and treatment phase were affected by this delay in feeding,
we expected a larger contrast due to enhanced frustration during the treatment phase. One of
the experimenters entered the aisle with a food trolley, feeding all horses with the experimental
horse being fed last and after a minimum of 60 seconds. The 60 seconds preceding the experi-
menter entering the aisle with the food trolley were usedas control phase.
Plastic bag (PB). Waving a plastic bag has previously been used to induce fear in horses
[26,27]. Similar to thesestudies, a plastic bagfilled with cans and tied to the end of a whip and
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 4 / 15
a green umbrella were used to startle the horses.The objects were waved in front of the
box throughoutthe 60 seconds treatment phase. The 60 secondspreceding thewaving period
were used as control phase. To minimise the risk of habituation to this stimulus, only one
horse per area was exposed to the plastic bag within one session. Moreover, whenever possible,
other experimental horses of the same area were brought to outside paddocks before the start
of the session.
Picture processing
All pictures were screenedand blurry pictures and pictures with obscuredview of the eye wrin-
kles were excluded.Pictures were defined as blurry if eye wrinkleswere not clearly visible or
the beginning or endof wrinkles was not clearly detectable. The view of eye wrinkles was con-
sidered as obscured if either both eyes were visible (frontal picture) or if only parts of the eye-
ball were visible (frontal picture or picture from the back).
From the remaining pictures, four pictures per condition and phase were selected randomly
for each horse using a random number generator in R (version 3.1.3, function ‘sample’, exclud-
ing repetitions), resulting in a total of 512 pictures (16 horses x 4 conditions x 2 phases x 4 pic-
tures). Selected pictures were cropped and rotated with Adobe Lightroom 5.4 to make eye size
and eye position comparable between pictures
Outcome measures
Outcome measures were defined based on pilot observations during which pictures of horses
varying in eye wrinkle expression were compared in view of identifying those parts that dif-
fered between pictures, similar to the selection process of Actions Units in the HorseGS [16].
We defined six measures including four categorical measures: qualitative assessment (‘qualita-
tive’), eyelid shape (‘shape’), markedness of eye wrinkles (‘markedness’), and presence of eye
white (‘eye white’), as well as two continuous measures: numberof wrinkles (‘number’), and
angle (‘angle’). The definitions of the outcome measures and the details of scoring are specified
in Fig 1.
Scoring
All pictures were scored using CorelDRAW Graphics Suite X7 (Corel Corp.). The order of the
pictures was randomised and the coder (SH) was blind to condition and phase. After every 50
th
picture, ten out of these 50 pictures (i.e. 20%) were randomly chosen using the sample function
in R and re-scored to assess intra-observer reliability. Additionally, 10% of the total sample
were scored by a second coder to test for inter-observer reliability. The second coder was
trained with example pictures before coding the sample.
Ethical considerations
This experiment was carriedout in accordance with the ethical policy of the International Soci-
ety for Applied Ethology. It was approved by the Cantonal Veterinary Office according to the
Swiss Animal Welfare legislation (Vaud, Switzerland, Approval Number VD 2804).
Statistical analyses
Intra- and inter-observerreliability. Intra- and inter-observer reliability were analysed
using SPSS (version 22). For categorical variables (‘qualitative’, ‘shape’, ‘markedness’, and ‘eye
white’), Cohen’s kappa was calculated, with a significantp-value and a kappa-value above 0.8
indicating ‘almost perfect agreement’ [28]. Reliability of continuous measures (‘number’,
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 5 / 15
‘angle’) was assessed by intraclass correlation coefficients (ICC). ICCs were calculated using a
two-way mixed design to assess the absolute agreement between the two scores [29,30], with
an ICC greater than 0.75 indicating ‘excellent’ agreement [31]. Lower bounds of the 95% confi-
dence interval (CI) were assessed as a measure of deviation from the ICC.
Assessment of the outcome measures. (Generalised) linear mixed-effectsmodels and
ordered logistic regressions were used for the analysis of the outcome measures to adequately
reflect dependencies in the experimental design (nesting, repeated measures). Data were ana-
lysed in R (version 3.1.3).
All models were built based on the hypothesis that the two positive treatment phases would
decrease eye wrinkle expression whereas the two negative treatment phases would increase it
compared to the respective control phases. Thus, we expected an interaction between
Fig 1. Measures for the assessment of eye wrinkle expression. (a) Qualitative assessment: Categories reflect the first subjective impression of the
expression of eye wrinkles based on the number of wrinkles, their markedness, and the angle they are forming. ‘No wrinkle’: no wrinkle visible. ‘Weak’:
the overall impression of the wrinkles is weak with, e.g., some weakly marked wrinkles forming a narrow angle. ‘Strong’: the overall impression of the
wrinkles is strong with, e.g., several marked wrinkles forming a wide angle above the eyeball. (b) Eyelid shape: ‘Round’: smooth curve without any sign
of the lid being pulled in the dorso-medial direction. ‘Weakly pulled’: curve is continuous but slightly pulled in the dorso-medial direction. The eye looks
more angled. ‘Strongly pulled’: the lateral part of the lid is an almost straight line. (c) Markedness: the depth and width of the wrinkles is assessed. If the
markedness differs between wrinkles, the most prominent wrinkle is assessed. ‘No wrinkle’: no wrinkle visible. ‘Weak’: wrinkles are flat and narrow lines.
‘Strong’: wrinkles are pronounced in depth and width. (d) Eye white: The sclera (eye white, also when brownish due to pigments) is assessed as visible
(‘yes’) or not visible (‘no’). (e) Number: Only wrinkles above the eyelid and those of a minimum length of one third of the eyeball’s diameter are
considered. A deep indent, often seen in older horses, is not considered as a wrinkle (as it is not caused by muscle contraction of the inner eyebrow
raiser). Moreover, wrinkles originating on the eyelid (mostly one or two) are not counted. (f) Angle: The degree of the angle is measured on the
intersection of the extension of a line drawn through the eyeball and the extension of the highest wrinkle. The line through the eyeball extends from the
medial to the lateral corner of the eyeball. If the medial corner is not clearly defined, the line goes through the middle of the tear duct.
doi:10.1371/journal.pone.0164017.g001
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 6 / 15
‘condition’ (factor with four levels: grooming (G), food anticipation (FA), food competition
(FC), plastic bag (PB)) and ‘phase’ (factor with two levels: control, treatment) to have an effect
on the outcome measures. Since the position of the horse within its box was unpredictable dur-
ing data collection, the side of the eye (left eye, right eye) varied among pictures. For each
horse, condition, and phase, we only compared pictures of the same eye with each other. If a
condition for a horse included pictures of both the left and the right eye, we excluded the pic-
tures of the side containing fewer pictures. Further, ‘side’ was included as an additional fixed
effect in all models to control for potential laterality effects (effect between horses).
Condition, phase, side, all two-way interactions (between condition and phase, condition
and side, and phase and side)and the three-way interaction (condition,phase, and side) were
implemented as fixed effects in all models. Random effects for all models were phase nested in
condition nested in horse. All six measures describedas outcome measures were considered as
primary outcome variables (see above and Fig 1). The threshold of significance was adjusted to
a p-value of p <0.0083 using the Bonferroni correction for multiple hypothesis testing in
order to control for type I error (false positive) rate.
In cases of significant interactions between condition and phase, Bonferroni-corrected post-
hoc analyses were conducted (function: glht, package: multcomp).
Model assumptions were verified by graphical inspections of the residuals. We checked for
normal distributionand homogeneity of variancefor the linear models,and for homogeneity
of variance for the generalised model and the ordered logistic regression. Transformation of
the data was not necessary for any of the models.
Categorical measures: The binomial outcome measure ‘eye white’ was analysed using a
generalised linear mixed-effects model (function: glmer, package: lme4 [32]). The multinomial
outcome measures ‘qualitative’, ‘shape’ and ‘markedness’ were scored across three ordered lev-
els each, and were thus analysed using an ordered logistic regression (function: clmm, package:
ordinal [33]).
Continuous measures: For the continuous outcome measures ‘number’ and ‘angle’,
linear mixed-effects models were run (function: lme, package: nlme [34]). For ‘angle’, pictures
were only included in the analysis if a wrinklewas present such that an angle could be mea-
sured. Therefore, the sample for the outcome measure ‘angle’ consisted of all pictures with
‘number’ 1.
Results
Sample size and data structure
Based on the study design,we aimed at a total of 512 pictures (16 horses x 4 conditions x 2
phases x 4 pictures) for the scoring of eye wrinkles. However, the final sample of pictures meet-
ing our inclusion criteria comprised 417 pictures. Some pictures included in the final sample
were not good enough for assessing all outcome measures reliably. Thus, missing values
occurred for the measures ‘shape’ (n = 11), ‘markedness’ (n = 6), ‘number’ (n = 49), and ‘angle’
(n = 31), but not for ‘qualitative’ and ‘eye white’. ‘Angle’ could only be analysed when there was
at least one wrinkle (‘number’ 1), leading to a final sample for the analysis of ‘angle’ of 227
pictures.
Among the final sample, both levels of ‘eye white’ and all three levels of ‘shape’, ‘marked-
ness’, and ‘qualitative’ were represented, and the ‘number’ of wrinkles ranged from 0 to 6 wrin-
kles, whereas ‘angle’ varied between 10.3° and 42.5°.
Pictures were fairly equally distributed across eye side, with 197 pictures of the left eye and
220 pictures of the right eye, and eye side was fairly well balanced across horses and conditions
(S1 Table).
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 7 / 15
Intra-observer reliability
Comparison of the first and the second scoring revealed high intra-observer reliability for all
six outcome measures. For the categorical measures, all p-values were highly significant
(p <0.001) and all kappa-values were above 0.8 (‘qualitative’: κ= 0.856, ‘shape’: κ= 0.823,
‘markedness’: κ= 0.841, ‘eye white’: κ= 0.978). For both continuous measures, ICCs revealed
excellent agreement between the two scores (‘number’: ICC
average
= 0.976, CI lower
bound = 0.963; ‘angle’: ICC
average
= 0.979, CI lower bound = 0.968).
Inter-observer reliability
Comparison of the scorings from the two coders revealed ‘almost perfect’ agreement for
the categorical measures, ‘qualitative’, ‘markedness’ and ‘eye white’ (‘qualitative’: κ= 0.810,
‘markedness’: κ= 0.847, ‘eye white’: κ= 0.839), whereas agreement was moderate for ‘shape’
(κ= 0.497). P-values for all categorical measures were highly significant (p <0.001). The agree-
ment for the two continuous measures was ‘excellent’ (‘number’: ICC
average
= 0.970, CI lower
bound = 0.934; ‘angle’: ICC
average
= 0.947, CI lower bound = 0.902).
Assessment of the outcome measures
‘Angle’ was significantly affected by the interaction between condition and phase (F
3
,
22
=
5.922, p = 0.004, Table 2). Post-hoc comparisons revealed a significant difference between
Table 2. Statistical models and results for the six outcome measures.
Outcome measure Model Interaction Test statistic P-value
1
Qualitative assessment Ordered logistic regression condition x phase x side χ
23
= 3.795 0.28
condition x phase χ
23
= 3.902 0.27
condition x side χ
23
= 2.568 0.46
phase x side χ
21
= 3.095 0.08
Eyelid shape Ordered logistic regression condition x phase x side χ
23
= 2.008 0.57
condition x phase χ
23
= 1.853 0.60
condition x side χ
23
= 3.337 0.34
phase x side χ
21
= 0.521 0.47
Markedness Ordered logistic regression condition x phase x side χ
23
= 0.259 0.97
condition x phase χ
23
= 3.191 0.36
condition x side χ
23
= 1.455 0.69
phase x side χ
21
= 1.923 0.17
Eye white Generalised linear mixed-effects model condition x phase x side χ
23
= 3.448 0.33
condition x phase χ
23
= 8.33 0.04
condition x side χ
23
= 1.818 0.61
phase x side χ
21
= 1.569 0.21
Number Linear mixed-effects model condition x phase x side F
3, 250
= 1.101 0.35
condition x phase F
3, 42
= 0.872 0.46
condition x side F
3,250
= 1.185 0.32
phase x side F
1,250
= 0.004 0.95
Angle Linear mixed-effects model condition x phase x side F
3,146
= 1.287 0.28
condition x phase F
3,22
= 5.922 0.004 *
condition x side F
3,146
= 3.511 0.02
phase x side F
1,146
= 2.189 0.14
1
Critical p-value after Bonferroni correction for multiple hypotheses testing: p = 0.0083.
doi:10.1371/journal.pone.0164017.t002
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 8 / 15
control and treatment phase in the G condition(z = -3.258, p = 0.008) and in the FC condition
(z = 2.803, p = 0.033). As predicted by our hypothesis, ‘angle’ decreased from the control to the
treatment phase during the G condition and increased from the control to the treatment phase
during the FC condition (Fig 2). However, there was no significant change between control
and treatment phase during the FA and the PB conditions. There was also a non-significant
interaction for ‘angle’ between condition and side but independent of phase, indicating the pos-
sibility of laterality effects (F
3,146
= 3.511, p = 0.02, Table 2). The control phases did not signifi-
cantly differ from each other (F
3,15
= 1.732, p = 0.2).
For ‘eye white’ there was a tendency to vary according to our hypothesis, with fewer horses
showing eye white in the positive, and more horses showing eye white in the negative treatment
phases compared to control phases (χ
23
= 8.33, p = 0.04). However, the interaction between
condition and phase failed to reach significance after Bonferroni correction.
‘Qualitative’ assessment, ‘shape’, ‘markedness’, and the ‘number’ of wrinkles were not
affected by any of the fixed effects. Thus, there was neither a significant three-way interaction
(condition x phase x side), nor a significant two-way interaction (condition x phase, condition
x side, or phase x side, Table 2).
Discussion
The aim of the present study was to investigate whether variation in the expression of eye wrin-
kles reflects emotional state in horses. To this end, horses were exposed to distinctively
valenced conditions, with the hypothesis that positive conditions would decrease the expres-
sion of eye wrinkles whereas negative conditions would increase it. Eye wrinkle expression was
assessed using a set of six outcome measures, of which all but ‘shape’ could be assessed reliably.
The angle was narrower in the G and wider in the FC condition compared to control phases,
indicating relaxation of the inner eyebrow raiser during G and contractionof this muscle dur-
ing FC. Moreover, more horses showed eye white in the PB condition than during the G and
FA conditions, whereas in none of these conditions treatment phase significantly differed from
the respective control phase. The four other measures did not vary consistently during the four
test conditions.
Intra- and inter-observer reliability
When tested for intra-observer reliability, all six outcome measures could be assessed highly
reliably as indicated by the strong agreement between the two scores for all six measures.
Almost perfect agreement was also found for the two coders for five of the six measures; only
the agreement for ‘shape’ was weaker. We therefore propose to either exclude this measure in
future studies or to revise its definition to render it reliable.
In order to arrive at such high reliability, the categorical variables had to be defined more
broadly than initially planned, resulting in less fine-grained coding. The disadvantage of this
was that ceiling and floor effects were more likely for measures with only three expression lev-
els (‘qualitative’, ‘shape’, and ‘markedness’), and with only two expression levels, ‘eye white’
could only vary in one direction depending on its expression during the control phase. By con-
trast, the continuous outcome variables allowed a more precise assessment. However, the num-
ber of missing values was higher for continuous than for categorical measures, suggesting a
need for higher picture quality.
Nevertheless, these measures reflect the type and range of measures by which eye wrinkle
expression can be assessed in situ. Given the high intra- and inter-observer reliability, we
believe that they may be a useful tool to assess eye wrinkle expression in future studies.
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 9 / 15
Fig 2. Effect of the four conditions on eye wrinkle expression. The effect of the four conditions (Grooming (G),
Food anticipation (FA), Food competition (FC), Plastic bag (PB)) on the six outcome measures is shown with
respect to phase (Control (C), Treatment (T)). (a, b) ‘angle’, ‘number’: boxplots with median (black line),
interquartile range (box), 1.5 x interquartile range (whiskers). (c) ‘eye white’: bar chart±binomial confidence
interval. (d, e, f) ‘qualitative assessment’, ‘eyelid shape’, ‘markedness’: stacked bar charts with the three colours
representing the three different scores. (d) ‘qualitative assessment’: ‘no wrinkle’ (black), ‘weak’ (dark grey), ‘strong’
(light grey). (e) ‘eyelid shape’: ‘round’ (black), ‘weakly pulled’ (dark grey), ‘strongly pulled’ (light grey). (f)
‘markedness’: ‘no wrinkle’ (black), ‘weak’ (dark grey), ‘strong’ (light grey).
doi:10.1371/journal.pone.0164017.g002
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 10 / 15
Assessment of the outcome measures
Angle. The angle between the horizontal line through the eyeballand the highest wrinkle
was narrower in the G and wider in the FC condition compared to control phases, reflecting
relaxation and contraction of the underlying muscle, respectively. Eye wrinkleexpression is
elicited by the level of contraction of the inner eyebrow raiser, and the angle may therefore be a
good indicator of contraction. Contraction of the inner eyebrow raiser has been considered a
sign of fear and sadness in humans [18], and in horses the eye wrinkles are more pronounced
when the animals are in pain [17]. Interestingly, variation in angle was not only consistent in
one negative (FC) but also in one positive condition (G). With the growinginterest in the
assessment of positive emotional states in animals [22], indicators covering the range from neg-
ative to positive emotions are particularly important, as they may not only provide information
on the absence of negative emotional states but also on the presence of positiveemotional states
[35]. However, since results were consistent only for one condition per valence (positive, nega-
tive), they do not allow for generalisation to all negative and positive conditions. Thus, it is pos-
sible that the angle of eye wrinkles varies depending on discrete emotional states (e.g. those
elicited during G and FC) rather than emotional valence in general. Moreover, as discussed in
more detail below, we do not have independent evidence that each condition did indeed elicit
the expected emotional valencein the horses (in this case consideringthe FA and the PB
condition).
Potential reasons why the other outcome measures were not affected. The other five
outcome measures did not systematically vary depending on the different conditions. This may
simply reflect the fact that these measures were not affected by the horses’ emotional states.
However, other explanations should be considered as well.
Firstly, it should be discussed whether the four different conditions the horses were exposed
to elicited the expectedemotional valence. All conditions were chosen based on literature dem-
onstrating the expected effects on horses’ emotional states. However, since there are as yet no
validated measures of emotional valence for horses (hence the present study), we cannot rule
out the possibility that the four conditions failed to induce the predictedemotional states in the
horses. In particular, the FA condition may have induced frustration while the horses were
waiting for the food to arrive instead of a state of positiveanticipation [36]. However, the
horses were trained to expectthe 60 seconds delay before the arrival of the food reward,which
should have reduced the likelihood of frustration [36].
Secondly, the duration of the conditions may have affected our results. Emotions are short-
term states that are specific to stimuli or events [1]. In contrast, mood states last longer, reflect-
ing the cumulativeexperienceof short-term emotional states, thereby representing past experi-
ence [1]. In the present study, we exposed the horses to specific acute conditions hypothesising
to induce positive and negative emotional states. It is, however, possible that the duration of
the exposure to the four conditions (60 seconds) was not long enough to induce changes in the
expression of our outcome measures but that long-term mood could lead to detectable effects.
In the horse community, eye wrinkles are thought to be associated with both short-term states,
e.g. nervousness and tension [37], as well as long-term states, e.g. depression [38]. As a conse-
quence, future studies should not only investigate the influence of valence and specific condi-
tions but also both the effect of short-term emotional states as well as long-term mood on eye
wrinkle expression.
Another possibility is that either emotional state or the associated eye wrinkle expression
varied so strongly across the 60 second treatment phases that our pictures, which were selected
randomly, did not reflect the induced emotional state. Other possibilities for picture selection
would be toinclude more pictures by sampling at a shortertime interval,or by using
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 11 / 15
independent behavioural indicators of emotion expression as cues to time picture selectionto
peak emotional state. However, if emotional state or eye wrinkle expression varied so strongly
across the 60 second treatment phases that random picture selection did not provide accurate
results, then it might be questioned whether the conditions used here were appropriate to
study emotion expression or whether eye wrinkle expression can be a reliable indicator of
horses’ emotional states.
Thirdly, the categorical variables with only two or three expression levels may have been too
constrained, resulting in floor and ceiling effects which may have masked effects of condition
and phase. Even the number of eye wrinkles (‘number’), despite being treated as a continuous
variable, was rather constrained with a maximum number of six wrinkles and many zeroes,
which may have led to floor effects.
Finally, the substantial variability within the four control phases has to be considered. In
order to assess the effects of the different conditions on eye wrinkle expression independent of
external factors (e.g. time of day, weather condition, and management factors), every treatment
phase was paired with a preceding control phase, and the effects were assessed relative to the
control phase to which measures during the treatment phase were compared. Variation was
relatively high across the four different control phases both between and within horses, sug-
gesting that variation was notonly due toindividual differences but that individualhorses also
varied across control phases. Differences between horses were not surprising since individual
differences in emotional state are to be expected [39]. Moreover, despite habituation of the
horses to the different procedures and the flood lights, the experimentalcontext may have
affected them differently. For example, some horses continued to eat straw and did not seem to
be affected by the procedure, while others were focusing on the photographers. By contrast,
intra-individual differences would have to be explained by changes in emotional state between
the different experimental sessions. However, given that intra-individual variation was rela-
tively large, we cannot rule out the possibility that eye wrinkle expression reflects acute and
transient emotional states, or that it is sensitive to confounding factors.
Eye side and laterality
We controlled for eye side (left eye, right eye) by including it as fixed factor in our statistical
model. Even thoughthe results were not identical for both eyes, eye side didnot significantly
affect the outcome measures, providing little indication for laterality effects. There was only a
non-significanttendency for an effectof eye side on variation in ‘angle’ across the four condi-
tions but independentof phase. In the literature, visual laterality in response to the valence of
an object [40] and in response to the presence of a human [41] have been discussed. Our study
was not designed to investigate laterality effects, however, for further validation of ‘angle’ and
possibly other measures of eye wrinkle expression as indicators of emotions in horses, it should
be examined whether laterality can affect eye wrinkle expression depending on the valence of
the conditions the horses are exposed to.
Conclusion
To our knowledge this was the first study systematically investigating the effect of distinctively
valenced conditions on eye wrinkleexpression in horses. We established a set of six measures
of which five allowed assessing various aspects of eye wrinkle expression reliably. We found
that the angle between a horizontal line through the eyeball and the highest wrinkle caused by
contraction of the underlying inner eyebrow raiser was consistently affected in two out of the
four conditions: G resulted in a narrower angle through muscle relaxation while FC resulted in
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 12 / 15
a wider angle through muscle contraction. The other outcome measures were not affected by
the four conditions.
Further research is needed to investigate which aspects of eye wrinkle expression reflect
emotional valence in general or are more reflective of specific conditions and thus discrete
emotional states. Moreover, it is important to assess how longer lasting conditions affect these
measures in view of using them as indicators of emotional well-beingin horses.
Supporting Information
S1 Table. Distribution of pictures from the left and the right eye across horses (A) and situa-
tions (B).
(DOCX)
Acknowledgments
We would like to thank the two photographers Olivier Bloch and Raphaël Remund who enthu-
siastically and patiently worked with us. We gratefully acknowledge the staff from the stud
farm for their logistic support. Many thanks to Jeremy Bailoo, Thomas Reichlin, and Lorenz
Gygax for their fruitful statistical advice, and to Lisa Schanz for acting as second coder.
Author Contributions
Conceptualization:SH IB HW.
Formal analysis: SH AP HW.
Funding acquisition: IB HW.
Investigation: SH SS.
Methodology: SH SS HW.
Project administration: SH IB HW.
Resources: IB HW.
Supervision: HW.
Writing – original draft: SH.
Writing – review & editing: SH AP HW.
References
1. Mendl M, Burman OHP, Paul ES. An integrative and functional framework for the study of animal emo-
tion and mood. Proc Biol Sci. 2010; 277(1696):2895–904. doi: 10.1098/rspb.2010.0303 PMID:
20685706
2. Darwin C. The expression of the emotions in man and animals. London; 1872.
3. Ekman P, Friesen W V. Measuring facial movement. Environ Psychol nonverbal Behav. 1976; 1:56–
75. doi: 10.1007/bf01115465
4. Morris JS, Frith CD, Perrett DI, Rowland D, Young AW, Calder AJ, et al. A differential neural response
in the human amygdala to fearful and happy facial expressions. Nature. 1996; 383:812–5. doi: 10.
1038/383812a0 PMID: 8893004
5. Vick S-J, Waller BM, Parr LA, Smith Pasqualini MC, Bard KA. A Cross-species Comparison of Facial
Morphology and Movement in Humans and Chimpanzees Using the Facial Action Coding System
(FACS). J Nonverbal Behav; 31(1):1–20. doi: 10.1007/s10919-006-0017-z PMID: 21188285
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 13 / 15
6. Keating SCJ, Thomas A a, Flecknell P a, Leach MC. Evaluation of EMLA cream for preventing pain
during tattooing of rabbits: changes in physiological, behavioural and facial expression responses.
PLoS One; 7(9):e44437. doi: 10.1371/journal.pone.0044437 PMID: 22970216
7. Langford DJ, Bailey AL, Chanda ML, Clarke SE, Drummond TE, Echols S, et al. Coding of facial
expressions of pain in the laboratory mouse. Nat Methods; 7(6):447–9. doi: 10.1038/nmeth.1455
PMID: 20453868
8. Ahola Kohut S, Pillai Riddell R. Does the Neonatal Facial Coding System differentiate between infants
experiencing pain-related and non-pain-related distress? J Pain; 10(2):214–20. doi: 10.1016/j.jpain.
2008.08.010 PMID: 19081306
9. Cohn JF, Kruez TS, Matthews I, Yang Y, Nguyen MH, Padilla MT, et al. Detecting depression from
facial actions and vocal prosody. In: Affective Computing and Intelligent Interaction. 2009. p. 1–7.
10. Caeiro CC, Waller BM, Zimmermann E, Burrows AM, Davila-Ross Marina. OrangFACS: A muscle-
based facial movement coding system for Oragnutans (Pongo spp). Int J Primatol. 2013; 34(115–129).
doi: 10.1007/s10764-012-9652-x
11. Waller BM, Peirce K, Caeiro CC, Scheider L, Burrows AM, McCune S, et al. Paedomorphic facial
expressions give dogs a selective advantage. PLoS One; 8(12):e82686. doi: 10.1371/journal.pone.
0082686 PMID: 24386109
12. Wathan J, Burrows AM, Waller BM, McComb K. EquiFACS: The Equine Facial Action Coding System.
PLoS One; 10(8).
13. Sotocinal SG, Sorge RE, Zaloum A, Tuttle AH, Martin LJ, Wieskopf JS, et al. The Rat Grimace Scale:
a partially automated method for quantifying pain in the laboratory rat via facial expressions. Mol Pain;
7(1):55. doi: 10.1186/1744-8069-7-55 PMID: 21801409
14. Holden E, Calvo G, Collins M, Bell a., Reid J, Scott EM, et al. Evaluation of facial expression in acute
pain in cats. J Small Anim Pract. 2014; 55(12):615–21. doi: 10.1111/jsap.12283 PMID: 25354833
15. McLennan KM, Rebelo CJB, Corke MJ, Holmes M a., Leach MC, Constantino-Casas F. Development
of a facial expression scale using footrot and mastitis as models of pain in sheep. Appl Anim Behav
Sci; 2016; 176:19–26. doi: 10.1016/j.applanim.2016.01.007
16. Dalla Costa E, Minero M, Lebelt D, Stucke D, Canali E, Leach MC. Development of the Horse Grimace
Scale (HGS) as a pain assessment tool in horses undergoing routine castration. PLoS One; 9(3):
e92281. doi: 10.1371/journal.pone.0092281 PMID: 24647606
17. Gleerup KB, Forkman B, Lindegaard C, Andersen PH. An equine pain face. Vet Anaesth Analg; 42
(1):1–12.
18. Arellano D, Perales FJ, Varona J. Mood and Its Mapping onto Facial Expressions. Proc 8th Int Conf
AMDO 2014;31–40.
19. Swiss Federal Council. Animal Welfare Ordinance. 2011 p. 20.
20. Feh C, de Mazières J. Grooming at a preferred site reduces heart rate in horses. Anim Behav; 46
(6):1191–4. doi: 10.1006/anbe.1993.1309
21. McBride SD, Hemmings A, Robinson K. A preliminary study on the effect of massage to reduce stress
in the horse. J Equine Vet Sci. 2004; 24(2):76–81. doi: 10.1016/j.jevs.2004.01.014
22. Boissy A, Manteuffel G, Jensen MB, Moe RO, Spruijt B, Keeling LJ,et al. Assessment of positive emo-
tions in animals to improve their welfare. Physiol Behav; 92(3):375–97. doi: 10.1016/j.physbeh.2007.
02.003 PMID: 17428510
23. Van Den Bos R, Meijer MK, Van Renselaar JP, Van der Harst JE, Spruijt BM. Anticipation is differently
expressed in rats (Rattus norvegicus) and domestic cats (Felis silvestriscatus) in the same Pavlovian
conditioning paradigm. Behav Brain Res. 2003; 141(1):83–9. doi: 10.1016/s0166-4328(02)00318-2
PMID: 12672562
24. Bachmann I, Bernasconi P, Herrmann R, Weishaupt M a., Stauffacher M. Behavioural and physiologi-
cal responses to an acute stressor in crib-biting and control horses. Appl Anim Behav Sci. 2003; 82
(4):297–311. doi: 10.1016/s0168-1591(03)00086-8
25. Ninomiya S, Kusunose R, Sato S, Terada M, Sugawara K. Effects of feeding methods on eating frus-
tration in stabled horses. Anim Sci J. 2004; 75(5):465–9. doi: 10.1111/j.1740-0929.2004.00214.x
26. Christensen JW, Malmkvist J, Nielsen BL, Keeling LJ. Effects of a calm companion on fear reactions in
naive test horses. Equine Vet J. 2008; 40(1):46–50. doi: 10.2746/042516408X245171 PMID:
18083659
27. Von Borstel UUK, Duncan IJH, Lundin MC, Keeling LJ. Fear reactions in trained and untrained horses
from dressage and show-jumping breeding lines. Appl Anim Behav Sci. 2010; 125(3–4):124–31. doi:
10.1016/j.applanim.2010.04.015
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 14 / 15
28. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;
33(1):159–74. doi: 10.2307/2529310 PMID: 843571
29. Shrout PE, Fleiss JL. Intraclass correlations: Uses in assessing rater reliability. Psychol Bull. 1979; 86
(2):420–8. doi: 10.1037//0033-2909.86.2.420 PMID: 18839484
30. Landers R. Computing (ICC) as Intraclass Estimates Correlations of Interrater Reliability in SPSS. The
Winnower. 2015;(2: :e143518.81744 (2015)):1–4.
31. Cicchetti D V. Guidelines, criteria, and rules of thumb for evaluating normed and standardized assess-
ment instruments in psychology. Psychol Assess. 1994; 6(4):284–90. doi: 10.1037//1040-3590.6.4.
284
32. Bates D, Maechler M, Bolker B, Walker S. lme4: Linear mixed-effects models using Eigen and S4.
2014.
33. Christensen RHB. ordinal—Regression models for ordinal data. 2015.
34. Pinheiro J, Bates D, DebRoy S, Sarkar D, Team RC. nlme: Linear and nonlinear mixed effects models.
2015.
35. Sandem AI, Braastad BO, Bu
`e KE. Eye white may indicate emotional state on a frustration-contented-
ness axis in dairy cows. 2002; 79:1–10. doi: 10.1016/s0168-1591(02)00029-1
36. Moe RO, Nordgreen J, Janczak AM, Spruijt BM, Zanella AJ, Bakken M. Trace classical conditioning as
an approach to the study of reward-related behaviour in laying hens: A methodological study. Appl
Anim Behav Sci. 2009; 121:171–8. doi: 10.1016/j.applanim.2009.10.002
37. Wieland K. What are my horse’s eyes telling me? [Internet]. 2013. Available: https://15minutehorsefix.
wordpress.com/2013/01/02/what-are-my-horses-eyes-telling-me/.
38. Crowe J. Body language 101—Part two: the eyes and nose [Internet]. 2009.
39. Boissy A. Fear and fearfulness in animals. Q Rev Biol. 1995; 70(2):165–91. doi: 10.1086/418981
PMID: 7610234
40. De Boyer Des Roches A, Richard-Yris MA, Henry S, Ezzaou??a M, Hausberger M. Laterality and emo-
tions: Visual laterality in the domestic horse (Equus caballus) differs with objects’ emotional value. Phy-
siol Behav. 2008; 94(3):487–90. doi: 10.1016/j.physbeh.2008.03.002 PMID: 18455205
41. Farmer K, Krueger K, Byrne RW. Visual laterality in the domestic horse (Equus caballus) interacting
with humans. Anim Cogn; 13(2):229–38. doi: 10.1007/s10071-009-0260-x PMID: 19618222
Eye Wrinkle Expression in Horses
PLOS ONE | DOI:10.1371/journal.pone.0164017 October 12, 2016 15 / 15
