Opportunities for Refinement in Neuroscience: Indicators of Wellness and Post-Operative Pain in Laboratory Macaques

Being able to assess pain in nonhuman primates undergoing biomedical procedures is important for preventing and alleviating pain, and for developing better guidelines to minimize the impacts of research on welfare in line with the 3Rs principle of Refinement. Nonhuman primates are routinely used biomedical models; however, it remains challenging to recognize negative states, including pain, in these animals. This study aimed to identify behavioral and facial changes that could be used as pain or general wellness indicators in the rhesus macaque (Macaca mulatta). Thirty-six macaques scheduled for planned neuroscience procedures were opportunistically monitored at four times: Pre-Operative (PreOp), Post-Operative (PostOp) once the effects of anesthesia had dissipated, Pre-Analgesia (PreAn) on the subsequent morning prior to repeating routine analgesic treatment, and Post-Analgesia (PostAn) following administration of analgesia. Pain states were expected to be absent in PreOp, moderate in PreAn, and mild or absent in PostOp and PostAn when analgesia had been administered. Three potential pain indicators were identified: lip tightening and chewing, which were most likely to occur in PreAn, and running, which was least likely in PreAn. Arboreal behavior indicated general wellness, while half-closed eyes, leaning of the head, or body shaking indicated the opposite. Despite considerable individual variation, behavior and facial expressions could offer important indicators of pain and wellness. They should be routinely quantified and appropriate interventions applied to prevent or alleviate pain and promote positive welfare. This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 International license (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided the original work is appropriately cited.


Introduction
Although primate use is a small proportion of total animal use in bioscience research, these animals are routinely subjected to procedures or conditions that directly and indirectly affect their welfare (e.g.,, Capitanio et al., 1996;Balcombe et al., 2004;Carlsson et al., 2004;Rennie and Buchanan-Smith, 2006a,b,c;Olsson and Westlund, 2007;Wolfensohn and Lloyd, 2013).It is a societal expectation that animals used in bioscience experience good welfare, which is characterized by an absence of unnecessary suffering (Lund et al., 2012;Leaman et al., 2014).Such considerations are also important for scientific validity as poor animal welfare may confound experimental results and affect the translation of scientific findings to human health benefits (Poole, 1997;Würbel, 2001;Tasker, 2012;Everds et al., 2013;Hall et al., 2015; general activity decreases after abdominal surgery, as does standing and vigilance, while locomotion and foraging seem unaffected (Allison et al., 2007), although this may not be a pain-specific response as all animals were provided with pain relief.These indicators provide a useful starting point for pain assessment, but more objective experimental data on NHP pain expression is required and across a wider range of commonly used laboratory species.Ideally, specific indicators should be identified and developed for each protocol that experimental animals are likely to experience (Morton, 2000), and humane endpoints employed and re-evaluated regularly (Hawkins, 2002).
Some pain reactions are likely to be automatic protective responses (Sneddon et al., 2014), but the expression of pain may benefit individuals by recruiting assistance from conspecifics (Langford et al., 2006;de Waal, 2008) although empirical evidence that it serves this function is limited.However, many animal species including primates are thought to "hide" their pain as a survival-enhancing strategy, making pain identification challenging (Plesker and Mayer, 2008;Murdoch et al., 2013;Fenwick et al., 2014;Gaither et al., 2014).Furthermore, distinguishing chronic pain presents an even greater challenge than for acute pain because it can be difficult to obtain a pain-free baseline for comparison (Brearley and Brearley, 2000), and there is the potential that other associated negative affective states (e.g., depression) will trigger or amplify the pain experience and diminish pain recognition and treatment (Bair et al., 2003).Assessment of animal welfare would therefore be improved by the development of sufficiently specific and sensitive indicators of pain and other negative affective states (Wolfensohn and Lloyd, 2013).
Recently, several studies have identified facial changes that indicate pain in a variety of mammalian species including mice (Langford et al., 2010), rats (Sotocinal et al., 2011), horses (Dalla Costa et al., 2014), andsheep (McLennan et al., 2016).These "grimace scales" assess the presence and intensity of pain, and are practical to use because they rely on scoring position changes of only a few facial features, typically the ears, eyes, cheeks and nose (see Descovich et al., 2017 for a review), and exploit our strong attention bias towards animal faces relative to other body regions (Leach et al., 2011).This approach has not been applied to NHP species, although primate facial expressions have been well-studied in the context of social communication (Chevalier-Skolnikoff, 1973;Partan, 2002;Ghazanfar and Logothetis, 2003;Waller and Micheletta, 2013) and facial expressions may provide insight into internal states in animals (Descovich et al., 2017).
Pain-related facial changes are widely evident in humans, present even in premature neonates although they are modified during development (Craig et al., 1993(Craig et al., , 2001;;Johnston et al., 1993).The human pain face is typically characterized by lowered brows, tightening of the eyes, raising of the cheeks, nose wrinkling, upper lip raising and horizontal stretching of the mouth (LeResche, 1982;Craig and Patrick, 1985;Prkachin, 2009) although pain type and individual variation affects expression of some elements (Prkachin, 1992;Prkachin and Solomon, 2009).Human facial expressions are commonly measured using the Facial Action Coding System (FACS), which records observable movements of the underlying facial musculature as "Action Units" (AUs) (Ekman and Friesen, 1978).Macaques share a similar facial muscle 2006b,c), however others, such as the experience of pain, are widely relevant to all sentient species (Bateson, 1991;Carstens and Moberg, 2000;Sneddon et al., 2014;Mellor and Beausoleil, 2015).Pain can be defined as "an unpleasant sensory and emotional experience associated with actual or potential tissue damage, or described in terms of such damage" (Loeser and Treede, 2008, p. 475), and is a particularly pertinent issue in biomedical science (Hawkins, 2002;Stokes et al., 2009).Research in both humans and animals suggests that identification of pain states can be challenging as individuals vary in pain sensitivity, tolerance or expression due to life history, social connectedness, sex, health, genotype, and temperament (Mogil and Kest, 1999;Mogil et al., 2000;DeWall and Baumeister, 2006;Defrin et al., 2017;Lush and Ijichi, 2018).The alleviation of pain should be based on the needs of the individual in terms of frequency or dosage (Roughan and Flecknell, 2002;Pham et al., 2010), and Refinement protocols should consider the potential for minimizing undesirable side effects from analgesia administration (Fleming and Coombs, 1992;Cooper et al., 2009;Schaap et al., 2012).In humans, pain assessment generally incorporates eliciting verbal feedback from the patient about their experience of pain (e.g., Jensen et al., 1986), unless the patient is not able to communicate effectively (e.g., pre-verbal infants: Taddio et al., 2009).However, animal pain levels are typically assessed by an observer, resulting in two sources of variation, within-patient and within-observer (Morton, 2000), and therefore creates additional challenges in pain assessment.
Behavioral observation is a useful, non-invasive technique for the identification of welfare states in animals (Dawkins, 2006b;Fraser, 2009).A normal behavioral baseline for a species or individual can be assumed to reflect wellness (e.g., Lambeth et al., 2013) and deviation may indicate compromised welfare.In this context we use "wellness" to indicate a state of being in good health, and as an antonym of illness.Ideally, specific negative states (e.g., pain, nausea, fear) should be identifiable so that appropriate treatment (e.g., analgesia/anti-emetic/environmental adjustments) can be implemented (Morton and Griffiths, 1985;Carstens and Moberg, 2000;Roughan and Flecknell, 2002;Mellor and Beausoleil, 2015;Sneddon, 2017).Unfortunately, validated behavioral indicators of specific negative states are lacking for many species, and assessment is reliant on subjective criteria (Carstens and Moberg, 2000;Honess and Wolfensohn, 2010;Wolfensohn and Lloyd, 2013).Despite the widespread use of nonhuman primate (NHP) species in translational bioscience, evidence-based guidelines for distinguishing pain and illness from other negative states are lacking.Wolfensohn and Honess (2005, p. 60) and the National Research Council Committee (NRCC) Guidelines (2009, p. 50 and 57) suggest that pain behavior in NHPs is expressed through a "miserable appearance", huddling, "sad" or contorted facial expressions, moaning or grunting, teeth clenching, restlessness, eye rolling and shaking.Other general pain behaviors may be wincing, vocalization, difficulty in movement (Lambeth et al., 2013), hunching or arm crossing (Morton and Griffiths, 1985).Grooming, social interaction, eating and drinking behaviors may also decrease (Wolfensohn and Honess, 2005;NRCC, 2009), however any deviation from an individual's normal behavioral repertoire could indicate pain or distress (Hawkins, 2002;Wolfensohn and Lloyd, 2013).In female olive baboons (Papio anubis), structure to humans (Burrows et al., 2009) and the FACS method has been adapted for rhesus macaques (Macaca mulatta), providing a comparable tool to examine facial movement (MaqFACS; Parr et al., 2010).
One key challenge in the study of negative animal welfare states is that experimental conditions must, by definition, induce or invoke these states.However, opportunistic sampling of animals undergoing planned experimental protocols, or "animal sharing", is consistent with Reduction within the 3Rs framework (e.g., Walker and Srinivas, 2013) and is desirable from an ethical standpoint because it offers benefits without causing additional harms.While this approach limits control over the experimental design and allows the influence of potential confounders, there is an important advantage in that data can be collected in contexts that reflect the actual severity or specificity of existing biomedical protocols.Pain elicited during standard analgesiometric tests is not comparable to post-operative pain as distinct nociceptive pathways are implicated (Roughan and Flecknell, 2002), which suggests that pain responses resulting from the first pain type may not be relevant to detection in the second.NHPs are used as models in neuroscience research where they undergo acute procedures such as cranial implants (Niekrasz and Wardrip, 2012;Azimi et al., 2016).These protocols are likely to cause significant pain and discomfort and it would be ethically unacceptable to undertake similar procedures solely to examine their effects on welfare, or to withhold analgesia in such cases to isolate pain responses.Therefore, opportunistic observation of animals before and after planned experimental protocols allows behavior to be measured under applied conditions.
The aim of this study was to identify general and facial behavioral changes that occur with pain states, and those that may indicate general wellness (or a deviation from wellness) in rhesus macaques, one of the most commonly used primate species in bioscience (Carlsson et al., 2004).Sampling was opportunistic, therefore, in addition to time period, the potential impact of several other variables was also examined, including sex, age, the severity classification of the procedure, and the interval since analgesia administration (to the start of the post-analgesia observation period).Finally, while this study monitors the impact of a single event, an individual's experience of previous procedures (e.g., Lutz et al., 2003) and the presence of any indicators of ill-ness prior to the procedure were also considered as factors that may indicate longer term effects related to cumulative severity (Honess and Wolfensohn, 2010).

Subjects and housing
This study was conducted between 2010 and 2014 and included 36 rhesus macaques (Macaca mulatta) (22 male, 14 female) undergoing planned neuroscience procedures.They were aged between 4 and 13 years (Tab. 1 and Tab.S1 1 ).This study was conducted opportunistically, and the number of animals used was therefore directly dependent on other unrelated studies within the collaborating facility, which determined both the number of animals that could be included and procedures that could be studied.Animals were originally obtained from a national (UK) breeding facility at a mean of 2.3 years prior to the study.Animal housing was in accordance with UK Home Office requirements (2014) with all being group housed with 1-9 other individuals, except for two adult males who were housed singly because they were incompatible with others.The facility underwent a major refurbishment during data collection and the majority of subjects (n = 22) were observed in the new accommodation.Space allowance per macaque was similar, but in the older facilities, animals were housed in 3 separate holding areas, (up to 10-12 animals per area), whereas the new accommodation housed up to 50-60 animals in a single larger area.Enclosures allowed a minimum floor area of 4.40 m 2 for each individual or pair of animals.All enclosures incorporated vertical space using raised platforms, and environmental enrichment, including a wood shavings substrate (Eco-pure, Datesand, Manchester, UK) for bedding and foraging, was available.The macaques were provided with appropriate nutrition and daily forage including commercially prepared food (Mazuri Primates Extended, Banana Chunks and Trio Munch Grains from Special Diets Services, Witham, U.K. LBS Biotechnology, U.K.) supplemented with forage mix and fruit.Water was available ad libitum unless restricted for other studies.The lightdark cycle at this facility was 12 h:12 h and the temperature was maintained at 22°C, with relative humidity at 24%. during PreAn and are significantly different from other periods were considered to have the most potential as pain indicators.
Filming was undertaken in the same location for all periods.Most macaques were filmed in a recovery cage directly adjacent to their home enclosure although a few (n = 3) were filmed in a restricted section of similar size in their home enclosure.Animals were alone in their enclosure during all filming sessions but could maintain vocal contact and limited visual contact with cage-mates or other conspecifics.Two camcorders recorded simultaneously; one captured the entire enclosure and the second was mounted on a motorized pan/tilt tripod head (CamRanger MP-360) and was maneuvered by remote control to zoom in on the macaque's face.Various camera models were used (e.g., Canon Legria HF M52; Sony Handycam) with any analogue footage converted into a digital format using Mediacruise (Canopus Co. Ltd.).
Twenty-five minutes of macaque footage was recorded for each period but the first five minutes were excluded from analysis to allow the camera operator to move out of view to minimize observer effects (Iredale et al., 2010;Peterson et al., 2017).Although the camera operator remained out of sight of the macaque, familiar facility staff were in the general vicinity and intermittently in direct view.Experimenter bias (Kilkenny et al., 2013) was minimized by relabeling videos with blind labels and by coding in a randomized video order, however fully blinded coding was not feasible as the effects of some surgical procedures (e.g., shaved patches of hair) were visible.

Behavioral recording
Behavior was coded according to a catalogue of defined behaviors (Tab.2) with long duration behaviors (e.g., walking) recorded continuously and short, discrete behaviors (e.g., vocalizations) counted on each occurrence.One experienced observer coded all macaque behavior; however, inter-observer reliability (IOR) was checked against a second trained observer to ensure consistency.Both had more than one year's experience in the observation of animal behavior including primates.Observers watched four 20min videos of different macaques and coded the behavior at 30 s intervals.The target percentage of agreement between observers was 80% and IOR test was 73%.Behavior definitions were discussed and clarified prior to coding different videos for which agreement reached 89%.Facial expressions were recorded by two MaqFACS accredited coders (Parr et al., 2010).Seven macaques were excluded from MaqFACS coding as facial movements were not clear enough in at least one of their videos.MaqFACS facial "action units" were recorded continuously from the video with the exception of blinking and vocalizing, which were counted as events (Tab.2).Due to the complexity of measuring facial movements, each face video was coded twice -once for the upper face and once for the lower face -and playback speed was reduced to between 0.1x and 0.2x to capture the activation and deactivation of muscles as they occurred.Due to the location of surgeries and head movements, ear positions/movements were excluded from coding as they could not be reliably differentiated.All coding was completed using Cowlog (Hanninen and Pastell, 2009) with output files processed in Excel (Microsoft ® Excel for Mac 2017 version) prior to analyses.

Ethical considerations
No regulated procedures were performed for the purposes of this study, and no veterinary procedures were delayed, omitted or modified from standard protocols.All procedures were performed under UK Home Office license and peri-operative protocols conformed with their Guidance on the Operation of the Animals (Scientific Procedures) Act, 1986 incorporating European Directive 2010/63/EU.All surgical protocols were approved by the Newcastle University Animal Welfare and Ethical Review Committee; animals received appropriate anesthesia/sedation and analgesia according to the approved experimental protocols for the relevant project, with analgesia drug and dosage determined by an experienced laboratory veterinarian (Tab.S1 1 ).Behavioral data collection methods were approved by the Psychology Ethics Committee, University of Stirling and adhered to the guidelines for the treatment of animals in behavioral research and teaching (ASAB, 2018).

Experimental protocol
The behavior of rhesus macaques undergoing planned surgical neuroscience research procedures was recorded.Procedures were prospectively categorized by the UK Home Office as moderate (n = 27) or mild in severity (n = 9).Moderate procedures included cranial implants (CRI), electromyography wire implants (EMG) and optogenetic surgery (OPT).Mild procedures included implant maintenance and repair (SCP and MAR).All animals received appropriate anesthesia/sedation and analgesia according to the approved experimental protocols for the relevant project, with analgesia drug and dosage determined by an experienced laboratory veterinarian.The influence of anesthesia and analgesia were not included in the statistical analysis due to the complexity and variation in drug protocols used.
Video footage of each macaque was collected during four periods: i) PreOp: on a day preceding the procedure when the animal was expected to experience no pain and to be performing normal behaviors for that individual, ii) PostOp: on the day of the procedure when the animal had recovered sufficiently from the anesthesia to move normally around the cage.All animals in this period received analgesia during the procedure, iii) PreAn: on the morning following the procedure directly before the animal received routine administration of analgesia, and iv) PostAn: on the same morning as the PreAn period, and 60 minutes following a routine administration of analgesia, except for a small subset (n = 5) filmed 30 minutes following a routine administration of analgesia.The PreOp period represented the baseline for each individual, and wellness indicators were defined as behaviors that were different in the pre-operative period when compared to all others.The PreAn period was considered the most informative for isolating pain responses in terms of both likelihood and intensity; PreOp was assumed to be a pain-free baseline, while analgesia was administered for both PostOp and PostAn and pain was assumed to be reduced or absent.Anesthesia and surgery, however well they are conducted, will have a range of non-specific effects potentially including but not limited to nausea, fatigue, disorientation, and metabolic or endocrine effects due to the surgical stress response.Therefore, behaviors that either peak or trough Tab.2: Catalogue of macaque behavior recorded (including behaviors that had no significant effects when analyzed or were too infrequent for analysis)

Locomotion
Type Descriptions

Lip corner pull State
The mouth corners are pulled obliquely upwards and backwards towards the ears (AU12: Zygomatic major ).

Lower lip depress State
The lower lip is pulled downwards in a smooth curve exposing the teeth, stronger actions may reveal the lower gums (AU16: Depressor labii inferioris ).

Chin raise State
The chin is pulled upwards causing the skin to flatten beneath the lower lip (AU17: Mentalis muscle).

True pucker State
Purses the lips medially forward towards each other, narrowing the mouth corners medially and protruding the lips (AU18i: Orbicularis oris incisivii labii superioris and inferioris ).

Outer pucker State
The lips protrude and cinch together at a point distal to the midline, causing them to part and appear inflated (AU18ii: Orbicularis oris, Incisivii labii inferioris and superioris ).
firmation using Akaike Information Criterion (AIC) and likelihood ratio tests with the ANOVA function (R base package.R Core Team, 2017) to ensure removal did not result in a weakened model.This was undertaken until either all fixed effects were significant or the variable to be removed was Period, the main variable of interest.Models including and excluding Period were then compared using AIC and likelihood ratio tests.When models indicated that Period was a significant variable, pairwise contrasts between different periods were undertaken using the pairs function (Lenth, 2016) with Tukey corrections for multiple comparisons.The likelihood of some behaviors could not be analyzed because these were displayed by either almost all or too few animals: walking (all walking and quadrupedal walking), eyebrow raise, true pucker, lip part and jaw drop.

Duration and frequency analysis
Analysis of behavioral durations and frequencies were conducted similarly to the presence/absence model, except with generalized linear mixed models (GLMMs) using the glmmADMB function.The residuals were not normally distributed as the data were over-dispersed; therefore, a negative binomial distribution was specified with a log link.Some behavior or facial movements could not be analyzed because these were either too rare or displayed by too few animals: bipedal running, affiliation, head jerk, pull on the head piece, hand shake, stretch, nose wrinkle with upper lip raise, upper lip raise, chin raise, true pucker, lip tuck.Plots indicated that model fitting was not optimal, and therefore confirmatory non-parametric tests were also undertaken using Friedman's tests with Bonferroni-corrected Wilcox signed-rank post-hoc contrasts.As both methods resulted in similar outcomes and interpretations, only the results from the GLMMs are reported because these allow for more complexity in the model and were slightly more conservative overall.

Statistical analysis
Statistical analysis was undertaken using R (R Core Team, 2017) in RStudio (Version 1.0.153,RStudio, Inc).For all analyses, the aim was to identify how behavior and facial movements differed between experimental periods (PreOp, PostOp, PreAn, and PostAn) in order to detect pain and wellness indicators.As outlined above, PreOp and PreAn were considered to be the most salient periods for identifying wellness and pain indicators, respectively.

Presence-absence analysis
To determine if behavior and facial movements were more likely to occur in particular periods, dependent variables were converted to a 1/0 data structure (0 = absence, 1 = presence).A logistic regression was conducted (glmmADMB function, glm-mADMB package, Skaug et al., 2016) with individual allocated as a random effect to account for the repeated measures design, and a binomial distribution specified with the canonical logit link function.The main fixed effect of interest was "Period".Secondary fixed effects included in the full models were "Sex" (male/female), "Age" (integer from 4 to 13), "Severity" (Mild/ Moderate), "Pre-operative illness" (Yes/No -whether the macaque was unwell at the time of the procedure as some procedures were undertaken for medical reasons), "PreviousOps" (Yes/No -whether the macaque had undergone procedures previously), and "PostAnTime" (30/60 minutes).Although all data collection periods were 20 minutes in length, on some occasions the animal's behavior was not visible or was unclear.To account for deviations in observation time, sampling effort (log transformed time in sight in seconds) was included as an offset in the model.Models were simplified by stepwise selection using Wald Chi tests from the ANOVA function (Car package, Fox and Weisberg, 2011) to identify the least important variables, with con- Lip tighten State Tightening and narrowing of the lips (AU23: Orbicularis oris )

Lips part State
The lips part so some space is observable between them (AU25: Several AUs may cause the lips to part ).

Jaw drop State
The bottom jaw is relaxed and lowered (AU26).

Mouth stretch State
The lower jaw is actively stretched to open the mouth, often occurs during yawning (AU27: Mylohyoid, Depressor angulioris, Levator labii inferioris).

Vocalization Event
Macaque makes any type of vocalization.

Tongue out State
The tongue protrudes in front of the teeth and is visible (AD19).

Chew State
Macaque performs a chewing motion of the mouth but is not eating.

Lip tuck State
Lower lip appears to tuck behind the upper lip.
The probability of occurrence of pain related behaviors was expected to be most evident in the PreAn period.The likelihood of two behaviors, i.e., lip tightening and chewing, peaked in PreAn but these were only significantly different from the PreOp period, with PostOp and PostAn falling more or less between the two (Tab.3).Similarly, the likelihood of running behavior (quadrupedal running and all running) troughed in PreAn but was not significantly lower than in either the PostOp or PostAn periods (Tab.3).

Multivariate analysis
Behaviors were analyzed with a multi-level sparse partial least squares discriminant analysis (sPLSDA) with the sPLSDA function (mixOmics package, Le Cao et al., 2017), which allows for repeated measures on the same individuals.Data were log transformed, and variables were scaled and centered during the analysis.Firstly, a sPLSDA was undertaken on all of the data to determine which behaviors influenced classification of cases to period.Secondly, to determine if the model would allow predictions, the dataset was randomly divided into training (67%) and test sets (33%).The training set was tuned on 20 components, using 10-fold, leave-one-out cross validation, and maximum distance specifications, and this model was then used to predict classification of Period in the test data.

Presence-absence analysis
Monkeys had clearly different behavior in the pre-operative period (PreOp), when they were assumed to be well, than at other times.Arboreal behavior such as climbing, descending, hanging, and the combined "all-arboreal" category were all most likely to occur in PreOp, as was cage manipulation (Tab.3).There was an effect of sex on arboreal behavior with males less likely to be off the ground [probability with 95% confidence intervals: 0.23 (0.01-0.87)] than females [0.74 (0.25-0.96)], (χ 2 (1, n = 36) = 5.95, p = 0.015).Conversely, body shake had the opposite pat-

Behavior
There were no behaviors that clearly peaked or troughed during PreAn relative to all other periods.

Multivariate analysis
sPLSDA analysis of the training dataset classified the periods using three components and was plotted to visualize how behaviors related to each other (Fig. 2a-c).Only component 1 of the three could be sensibly labelled and appeared to indicate an axis of ac-

Duration and frequency analysis
Several behaviors occurred for significantly more time in the PreOp period compared to other periods: This general pattern applied to walking (quadrupedal and all walking), arboreal behaviors (climbing, descending, hanging, and all arboreal), standing (quadrupedal, bi-and all stand), crouching, cage manipulation, and cage monitoring (Tab.4).Arboreal behavior was also influenced by sex and pre-operative illness; females were off the ground more often than males (χ 2 (1, n = 36) = 11.06,p < 0.001), as were those individuals not considered to be ill prior to the procedure (χ 2 (1, n = 36) = 8.3, p = 0.004; Fig. 1a,b).Similarly, those individuals with pre-operative signs of illness spent less time standing following the procedure (χ 2 (1, n = 36) = 7.18, p = 0.007; Fig. 1c).
In contrast, the duration of half-closed eyes and leaning head, and the frequency of body shake was lower during the PreOp period (Tab.4).Half-closed eyes was affected by several other explanatory variables including sex, age, severity, wellness and whether the animal had undergone previous procedures (Tab.4).On average, males half-closed their eyes significantly more [39.1

Tab. 4: Minimum adequate model summaries of statistical analysis using duration (timed in seconds) and frequency (counted) behavioral data of rhesus macaques over a 20-min observation time
Significant differences between experimental Periods (PreOp, PostOp, PreAn, PostAn) in the prevalence of behavior are indicated with letters in italics (different letters = significant differences).Akaike information criterion (AIC) given for models with and without Period as the main variable of interest.3a), and the classification rate was high across all three components (Tab.5a).In the training data, the PreAn period was best identified by component 2 but the classification rate was moderate (61%).PostAn was not correctly classified for any components (Tab.5a) and had large areas of overlap with other periods (Fig. 3).When applied to the test dataset, the mean successful classification rates of the test dataset across all three components were moderate for PreOp (48.1%),PostOp (33.3) and PreAn (25.9%), and unsuccessful for PostAn (0%) (Tab.5b).

Indicators of pain
The primary aim of this study was to identify potential behavioral and facial changes that indicate acute post-operative pain states in rhesus macaques.The period where pain was expected to be at its highest was PreAn, on the morning following the surgical procedure prior to routine administration of analgesia.At this point the effects of anesthesia and peri-surgical analgesia were expected to have dissipated significantly or entirely, as the next dose of pain relief was due to be given.Behavior that changes when pain is likely to be present and returns towards baseline levels after administration of analgesia can be used as evidence of pain (Roughan and Flecknell, 2002;Sotocinal et al., 2011;Wolfensohn and Lloyd, 2013;Sneddon et al., 2014).Three behaviors appeared to be potential pain indicators either in their presence (lip tightening and chewing, which peaked in probability at PreAn) or absence (running, which troughed in probability at PreAn), but these behaviors did not differ in amount or frequency during the period.Although our results suggest that behavior and facial expressions alone are insufficient to assess pain states in NHPs, these may nonetheless make an important contribution to perioperative welfare, for example, in monitoring wellness or medication effects (Flecknell, 2018), and triangulation with physiological measures could enhance our understanding of pain responses (Allison et al., 2007).Not all pain indicators that have been previously proposed (e.g., Morton and Griffiths, 1985;Wolfensohn and Honess, 2005;NRCC, 2009;Lambeth et al., 2013, see introduction) were observed.This is possibly because proposed indicators were constructed from subjective impressions rather than empirical research with experimental controls such as blinding, minimization of observer effects and randomized analysis.Alternatively, it may also be that some indicators are specific to particular types of procedures or pain.

Indicators of wellness
Wellness indicators were considered to be those that were significantly different in the baseline PreOp period compared to all other periods.There were clear changes in behavior from PreOp levels, indicating that the procedures carried out considerably impacted the behavioral repertoire of macaques, with effects remaining for at least 12 hours post-procedure regardless of the administration of analgesia.This was particularly evident in the multivariate analysis as PreOp was clearly different from the other groups and had the most successful classification rate.This was also supported by the univariate analyses where many behaviors either peaked or troughed in PreOp.Several behaviors, primarily those indicative of activity level and alertness, were more likely to be present and/or performed more in the baseline period, and while behaviors of this nature can indicate an absence of pain (Roughan and Flecknell, 2002;Sneddon, 2017) in this case they appear to reflect general wellness because they were insensitive to analgesia administration.These behaviors include arboreal behaviors (such as climbing and hanging), standing, crouching and two cage-related behaviors (cage manipulation and cage monitoring) that may indicate motivation to return to the home cage from the recovery cage.Arboreal behavior and standing were also lower in monkeys that had indicators of illness prior to their procedures, supporting the interpretation that they may be good general indicators of wellness.Rhesus macaques are primarily terrestrial in the wild (Wells and Turnquist, 2001), however in experimental facilities much of their time is spent in elevated positions (Clarence et al., 2006), which is likely to be an anti-threat behavior.The recovery cage lacked an elevated perch and therefore the reduction in arboreality after the PreOp period may indicate a need to reduce energy expenditure, or reflect discomfort in movement (e.g., Allison et al., 2007).However, there was also an interaction with sex; females were more likely to spend time off the ground than males, which may reflect sex differences in threat or stress responses, consistent with previous findings of rhesus macaque reactions to an unfamiliar observer (Iredale et al., 2010).The post-operative reduction in standing and cage monitoring are consistent with previous research on female baboons undergoing abdominal surgery (Allison et al., 2007).Posture may reflect reduced alertness or general reduction in activity, congruent with telemetry measures taken in female baboons (Allison et Lip tightening (AU23, Tab. 2) is not an action typically recruited in the human pain face (Prkachin and Solomon, 2009), however mouth tension, e.g., horizontal mouth stretching (AU20) and upper lip raising (AU10), is and comparable actions are observed in animal pain behavior, such as a "strained mouth" in horses (Dalla Costa et al., 2014).Chewing can also be a pain behavior in both humans and other animals.For example, donkeys chew more during mechanical nociceptive tests in comparison to sham tests (Grint et al., 2017), sheep grind their teeth when experiencing duodenal distension (Kania et al., 2009), and chewing can have an analgesic effect in humans (Weijenberg and Lobbezoo, 2015).In terms of locomotion, only running decreased during PreAn, however difficulty in movement is considered a negative correlate of quality of life in NHPs (Lambeth et al., 2013).In a previous study, female baboons decreased overall activity after abdominal surgery, although no effect on locomotion was found (Allison et al., 2007).Previous procedures (PreviousOps) contributed to some models (Tab.3, 4), but there was not strong evidence of a link with a continued underlying pain state (i.e., no cumulative severity) for putative pain indicators.
However, it is important to note that changes in these behaviors did not map perfectly onto anticipated pain states because the probability of occurrence in PreAn did not differ significantly from either PostOp or PostAn.This may be for several possible reasons.Firstly, it may be that behavior is a relatively poor indicator of pain in NHPs (Allison et al., 2007), or secondly, that the animals in the study are masking their pain (Fenwick et al., 2014;Gaither et al., 2014) due to the presence of care staff and other macaques in the vicinity.A third possibility is that actual pain states do not align with the predicted pain states within periods.The efficacy and effects of anesthesia and analgesia regimes in NHPs undergoing surgery are not yet well understood (Bertrand et al., 2018) and the absence of significant behavioral change following analgesia could indicate that the drug or dosage was not optimal in managing the levels of pain experienced, or at least not for all individuals.However, there was also no effect of procedure severity on these behaviors, which would be expected for reliable pain indicators.
While it is possible that the monkeys were not experiencing pain, this is very unlikely, as similar protocols generate significant levels of pain when conducted on humans (Dunn et al., 2016).The opportunistic experimental design may have resulted in behavioral variation that masked some of the pain-specific responses; e.g., grimace scale studies have typically assessed pain using analgesiometric tests or following a standardized procedure (Langford et al., 2010;Sotocinal et al., 2011;Dalla Costa et al., 2014).Lastly, the prevalence and frequency of behaviors may also be confounded by individual variation in drug response, or in pain reaction or tolerance.Personality in primates is recognized to have a significant effect on behavior (Coleman, 2012) as well as general health and welfare (Robinson et al., 2016(Robinson et al., , 2018)).For example, in female baboons individuals significantly varied in their response to the same standardized surgical procedure (Allison et al., 2007).Studies in dogs, horses and humans have also suggested that pain expression, if not the actual pain experience, is affected by personality (Williams, 2002;Ijichi et al., 2014;Lush and Ijichi, 2018), and this is a key area for future study on pain behavior in NHPs.haviors suggest a general and expected decrease in activity and environmental engagement rather than being specific to pain or feeling unwell, but are highly relevant in applied contexts; it is important to distinguish sedation effects from other negative affective states to avoid inflated pain scores due to similarities in behavioral response (Langford et al., 2010;Miller et al., 2015).

Practicality of potential indicators
Clinical monitoring of animals is largely reliant on cage-side observation, and behavioral or facial patterns identified in this study provide insight into how macaque welfare may be monitored.Generalization of the results through replication and refinement of observable indicators is needed, with consideration given to which indicators are likely to be most effective and most practical.
One challenge in the interpretation of these findings is the influence of secondary, explanatory variables, which are likely to reflect the complexity of the pain experience and variation in response.For example, arboreal behavior and body shake were (positively and negatively, respectively) associated with wellness, while face rubbing was associated with sedation, however all of these behaviors were also influenced by the sex of the animal.Ideal indicators of pain or wellness would be effectively generalized; however, it is possible that if such indicators existed then robust guidelines on identifying pain in macaques would already exist, and as our results suggest, in practice some variation should be expected.
A second practical challenge is the prevalence of behavior evident during monitoring; behaviors which are rare or change too subtly in relation to wellness states are unlikely to be sufficiently robust to use in assessments.For example, sedation effects appear to decrease lowering of the glabella, but given the mean frequency was less than 3 seconds in any period, this would be difficult to practically detect at the cage-side.Similarly, the likelihood of chewing behavior increased with assumed pain state, however even in the baseline period when pain was presumed to be absent, the probability of occurrence within the observation period was high.This limitation could potentially be overcome by developments in automated monitoring; however, this may also be impractical in cage-side contexts.Based on our results, the most promising indicator of wellness appears to be the presence of arboreal behavior when macaques were not in their home enclosure, while the following would warrant close monitoring if occurring within a 20-minute period: two or more body shakes; more than 1 minute of head leaning or 10 seconds of half-closed eyes; and less than 1 minute of standing or cage manipulation.It is recommended that these indicators are included in facility welfare assessments, and imperative that housing offers macaques vertical space as per guidelines for housing research primates (Jennings and Prescott, 2009;NHMRC, 2016) The third practical challenge is the potential impact of observer effects.In this study, the camera operator was not in visual range of the animal, however it was not possible to fully control potential observer effects because intermittent staff activity in the vicinity could evoke or suppress some behavioral responses (Iredale et al., 2010;Peterson et al., 2017).In applied contexts, animal monitoring is likely to be carried out using cage-side observation, however this may be insufficient for clinical assess-al., 2007).Cage monitoring, which is diminished following anesthesia, can be interpreted as indicating alertness (similar to the "checking" behaviour in Allison et al., 2007), and potentially associated with motivation to return to the home enclosure from the temporary cage.The exhibition of arboreal behavior and environmental manipulation seem to be good indicators of general wellbeing in macaques, at least in the context of temporary separation in a holding cage, although it is unclear whether post-operative reductions are attributable to pain or other factors because the frequency of these behaviors remained low across all postoperative periods, even after analgesia.
The converse pattern was identified for three behaviors that were either less likely to occur, or had lower durations, in the baseline pre-operative period; half-closed eyes, body shake and head leaning, which may indicate reduced wellness, again without specificity to pain.Half-closed eyes and head leaning occurred less before a procedure than during all subsequent periods, while body shake was both less frequent and less common.Monkeys that were potentially unwell prior to the procedure halfclosed their eyes more frequently, supporting the interpretation that this behavior is influenced by reduced wellness.In previous research, reduced eye aperture is related to sedation effects (Bertrand et al., 2016), may function as a protective mechanism (Defensor et al., 2012) and is a common hallmark of the pain face in mammals including mice (Langford et al., 2010), horses (Dalla Costa et al., 2014), rats (Sotocinal et al., 2011), sheep (McLennan et al., 2016), and seals (MacRae et al., 2018).Similarly, body shaking may indicate reduced wellness as the likelihood of occurrence was higher for procedures of moderate severity, and this behavior has been linked to anxious states in clinically ill macaques (Gaither et al., 2014).Head leaning occurred for significantly lower durations in PreOp than in the post-operative periods, and could serve a similar function to the behavior of pressing hand to head described in clinically ill rhesus macaques, and may alleviate pain via manual pressure (Gaither et al., 2014), however, there was no influence of procedure severity that could specifically indicate pain.Head leaning shares postural similarities with huddling/hunching which has been suggested as a potential pain or distress behavior (Morton and Griffiths, 1985;Wolfensohn and Honess, 2005) and this supports the interpretation that it may be a potential indicator of malady.In terms of potential signs of cumulative severity, individuals who had undergone previous procedures had an increased likelihood of body shaking and reduced duration of half-closed eyes, however wellness indicators were mostly insensitive to previous procedures.

Indicators of sedation
Although the macaques were only filmed once cage-side observation suggested they had recovered from sedation, some behaviors peaked in the PostOp period when any residual effects of anesthesia would be expected to be most evident.These "sedation-related" behaviors were leaning head, half-closed eyes and face rubbing (the likelihood of occurrence but not the duration of time/frequency), and a non-vigilant state (duration but not likelihood).Behaviors that were most suppressed by sedation effects were standing (likelihood), lowered glabella (duration), focused vigilance (duration), and cage shaking (both frequency and likelihood).These be-

Conclusion
Surgical research protocols have undesirable welfare implications for animals used as experimental models, such as the rhesus macaque.The 3Rs principle of Refinement guides scientists to minimize the effects on such animals; however, this is reliant on accurate assessment of negative affective states including, but not limited to, pain.This project has identified several potential behavioral indicators of pain and general wellness, however practical implementation to applied contexts is likely to have challenges.Although macaques are thought to hide their responses to pain, they were clearly negatively impacted by the procedures, as evidenced by changes in their behavior.It is difficult to disentangle common indicators of pain, malaise and sedation in macaques due to the complexity of interactions with other factors and given the confounds of the opportunistic sampling and applied context of the current study.Directions for future research should aim to extend this work to different experimental interventions and to examine the influence of individual behavioral variation on pain response.The precautionary principle should be applied to pain relief until sensitive and robust measures of pain are identified and validated (Flecknell, 1984;Roughan and Flecknell, 2002;Sneddon et al., 2014).Evaluating improved anesthetic and intraoperative care regimens using "wellness" indicators could lead to significant refinements of research procedures.
smacking of the lips together, the teeth are covered, the tongue may protrude, and actions may be accompanied by vocalization (AD181: Orbicularis oris ).

Fig. 1 :
Fig. 1: Boxplots of behavior prevalence for rhesus macaques experiencing neuroscience procedures in relation to explanatory variables a) & c) Pre-operative illness (ill/not ill); b) sex (female/male); and d) procedure severity (mild/moderate).Large figures depict behavioral prevalence for each 20-min observation period for the explanatory variable.Subset figures depict behavioral prevalence for the explanatory variable only.
to inactivity (+1.0).Using the training data only, the PreOp and PostOp periods were clearly different from the other periods (Fig.

Animal Age Ill (Y/N) Year Time at Procedure UK Home Analgesia Post- First Time Face (years) of proce-facility Office severity analgesia procedure since last coded dure (days) classification interval (Y/N) proce- (Y/N) (min) dure (days)
Time since last procedure: Number of days since the macaque's previous procedure.If this was the macaque's first procedure this is blank.Face coded: Indicates whether facial movements were coded using MaqFACS (Yes/No).