Relationship among indicators of pain and stress in response to piglet surgical castration_ an exploratory analysis

Multi-dimensional approaches are suggested as advantageous for evaluation of pain and/or stress at piglet castration, but the interpretation of data from such approaches is limited by a lack of understanding of relationships among different types of indicators and associated parameters. This paper used an exploratory approach to evaluate the interrelationships between different physiological, vocal, and behavioral indicators obtained during and after surgical castration of 580 piglets aged 3-4 days. The data, obtained from two experimental studies, were examined by analyses of correlations and by principal component analyses (PCA). Principal components were analyzed in mixed effects models. Each type of indicator (vocalization, resistance movement, saliva cortisol concentration, behavioral response in a social motivation test, behavior in the home pen, reaction to human) mainly contributed to separate principal components in the PCA and showed relatively low correlation coefficients between each other, indicating a variation in the response explained by the indicator types. Even within a type of indicator, specific parameters were, in several cases, found to explain different aspects of the piglet response. Overall, the results point to the importance of careful consideration of the differences existing among indicator types and highlight the need for further methodological development in that domain. © 2023 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Surgical castration in piglets is a welfare concern, due to the pain involved in the procedure and the following healing process (e.g., Taylor and Weary, 2000;Hay et al., 2003).Different pain mitigating strategies are currently implemented, including the use of different combinations of analgesia, local anesthesia, and/or general anesthesia (Bonneau and Weiler, 2019).The potential of these mitigative measures to improve welfare has been studied in the last decades, but despite several reports on the effects of these, discussions remain regarding their efficacy and sufficiency (Rault et al., 2011;Bonneau and Weiler, 2019).This lack of clear conclusions may partly be explained by a diversity in the selected indicators of pain and stress, as well as methodological differences.While this variation allows the study of different aspects of the potential effects of the procedure and lives up to recommendations for a multi-dimensional assessment (Leliveld et al., 2016;Baysinger et al., 2021), it may limit the possibility for valid comparisons of results, including, for example, meta-analyses and/or other analytical assessments across studies (Dzikamunhenga et al., 2014;Sheil and Polkinghorne, 2020).Thus, the formulation of a consensus for best practice for pain relief may be hindered (Bateson, 1991).Moreover, making conclusions on the effect on welfare of a procedure as evaluated by use of a multi-dimensional approach can be challenging.In most studies, conclusions seem to be formulated based on patterns observed across indicators, using a quantitative approach.In other terms, the more indicators that seem to point in a certain direction, the more this effect seems plausible.Other approaches include ranking of the indicators based on the perceived relevance for the assessment of pain and stress (O'Connor et al., 2014).Yet, with limited knowledge on the relative importance of the different indicators and with ongoing debates over the interpretation of certain indicators, conclusions should be drawn with care.In addition, the consequences of certain pain mitigating measures, for example, potential effects of a drug on thermoregulation and proprioception, are not always fully recognized (Coutant et al., 2022c), and may lead to wrongful interpretation of, for example, shivering or activity levels as signs of pain.This lack of consensus on the significance and interpretation of multiple indicators increases the risk of misinterpretation of the results (Sheil and Polkinghorne, 2020) and formulation of misled conclusions.
The present paper took advantage of a large dataset obtained in studies of piglet castration using different pain mitigating treatments and followed a novel approach to explore the potential relationship between different types of indicators used to evaluate the welfare effects of castration in piglets.The data used in the paper were extracted from two large-scale experiments examining effects of procaine-based local anesthetic injection on piglets' acute responses and post-procedural behaviors.Differences between the experimental treatments have been reported elsewhere (Coutant et al., 2022a(Coutant et al., , 2022b(Coutant et al., , 2022c) ) and will not be further discussed.Focus will instead be given to the response of the piglets to castration.In total, 40 parameters from six main indicator types were included and analyzed, with the aim to bring further insights into their interpretation in terms of stress and pain, and thus their potential use in a multi-dimensional approach examining welfare consequences of the procedures.

Materials and methods
The two studies underlying this paper have already been described in Coutant et al. (2022aCoutant et al. ( , 2022bCoutant et al. ( , 2022c)).In this section, a short description of the materials and methods is provided.For further details, we refer to the aforementioned papers.

Ethical and other permits
The experimental studies underlying this paper were performed in compliance with the EU Directive 2010/63/EU for animal experiments, the Ministry of Food, Agriculture and Fisheries, and The Danish Veterinary and Food Administration under act 474 of May 15, 2014, and executive order 2028 of December 14, 2020.The two experiments were approved as clinical trials by the Danish Medical Agency

Animals
Data were collected in July-October 2020 (Study 1) and June-September 2021 (Study 2).Both studies were conducted in the same Danish conventional herd producing crossbred (Landrace and Yorkshire × Duroc) piglets housed with the sow in farrowing pens (3.1 × 2.8 m).Piglets included were 3-4 days of age on the day of data collection (with day zero defined as the day of birth of the last piglet of a litter born alive), weighing between 0.94 and 2.68 kg.All experimental piglets were clinically healthy, free of overt anatomical malformations and signs of lameness, and only sows with a rectal temperature below 39 °C were included.Within litters, either six (Study 1) or five (Study 2) male piglets living up to the inclusion criteria were randomly attributed to a treatment number, following an allocation plan ensuring a balanced distribution of experimental treatments across litters, experimental days, and experimental weeks.
Experimental piglets could be cross-fostered in the first days of life, up to 24 hours prior to experimentation.Non-experimental littermates could be cross-fostered up to the morning prior to castration.Litters with experimental piglets were not ear tagged, tail docked, or teeth clipped before castration, and were administered a Non-Steroidal Anti-Inflammatory Drug (NSAID) (intramuscular injection of 0.3 mL Melovem, 5 mg/mL, Dopharma, the Netherlands) as analgesic within 24 hours after castration, after completion of the data collection.

Study design
On the day of castration, experimental piglets were weighed.Saliva samples were taken approximately 35-40 minutes before bringing the piglets to a testing area, a calm room outside the farrowing room.Piglets were injected with the local anesthetic, castrated, or sham-handled one by one, respecting a randomized testing order, and following a predefined schedule ensuring that the experimental intervals between procedures were respected.During all procedures, piglets were fixated while lying on their back, in a commercially available castration bench (Unitron A/S, Kolding, Denmark) modified to increase the possibility for piglets to vocalize and move their front legs, as explained in Coutant et al. (2022a).Inbetween procedures, piglets were returned to a heated box with some of their littermates.A detailed description of the procedures of intra-funicular and intra-testicular injections of local anesthetic, castration, and sham-handling is given in Coutant et al. (2022a).
Immediately after castration or last sham handling, piglets were individually subjected to a 3-minute social motivation test performed in the corridor outside the farrowing room (Study 2 only).After the test, each piglet was brought back to the home pen, where continuous behavioral observations were performed for a duration of 10 minutes upon placement in the pen (Study 2 only).On average 17 minutes after castration or last sham handling, a second saliva sample was taken in the farrowing unit.Later in the afternoon, approximately 6 hours after castration, piglets were individually subjected to a human-animal relationship (HAR) test followed by a handling test, and a last saliva sample was taken for cortisol determination (Study 1 only; Figure 1).

Study 2
A total of 290 piglets were assigned to one of five treatments (Table 2): surgical castration without local anesthesia (CC), intratesticular injection of 0.5 mL of local anesthetic per testis and subsequent castration after 5 minutes (IT05), intra-testicular injection of 0.3 mL of local anesthetic per testes and subsequent castration after 5 minutes (IT05 V03 ), sham handling with no tissue damage inflicted with one (sham castration; SH00) or two stays in the castration bench with a 5 minutes interval (sham anesthesia and sham castration; SH05).For details on the calculations justifying the sample size for Study 1 and 2, refer to Coutant et al. (2022b).

Vocalizations
For each piglet, intra-procedural vocal responses were recorded during castration, using a microphone (Sennheiser E614, Sennheiser, Wennebostel, Germany) fixed 30 cm ahead of the piglet's snout, at the level of the head of the piglet.The microphone was connected to an amplifier (Audiobox USB 96, PreSonus, Louisiana, USA) connected to a computer, from which recordings were manually started and stopped upon the placing and the removal of the piglet in the castration bench.All vocal files were analyzed by an observer, blinded to the experimental treatments, using Raven Pro 1.6 bioacoustics analysis software (Cornell Lab of Ornithology, Ithaca, New York, USA), and call characteristics were recorded and analyzed for each procedure [Table 3; see methodology in Coutant et al. (2022a)].

Resistance movements
Four distinct categories of front leg movements were recorded during each procedure using a camera (GoPro HERO7 Black, GoPro, San Mateo, California, USA; 60 frames per second, FPS) placed on a stand 30 cm to the right of the castration bench, approximately 50 cm above the bench.Video footage were analyzed by two observers, blinded to the experimental treatments, at low speed (5 FPS) using the Behavioral Observation Research Interactive Software (BORIS; Friard and Gamba, 2016), and the occurrence of each category of behavior (flexion, extension, kick, and blow; Table 4) was counted for each front leg in the interval between closing and opening of the castration bench, following a method developed by Coutant et al. (2022a).In addition, duration of blocking, corresponding to a leg being unable to move due to physical blocking, was also recorded.Resistance movement parameters were transformed from count to average count per second, taking into consideration the duration of the procedure and the duration of legs being blocked.The piglets were brought to the castration room with some of their littermates, anaesthetized and castrated one by one, then individually subjected to a social motivation test prior to being returned to their home pen.In Study 1, piglets were individually subjected to saliva sampling 17 minutes after castration, then placed back in the pen again and left undisturbed until being brought back to the corridor outside of the farrowing room for individual human-animal testing and saliva sampling 6 hours after castration.The piglets were then brought back to the home pen and left undisturbed.In Study 2, after the social motivation test, the behavior of the piglets was observed continuously for 10 minutes upon return in the home pen.After saliva sampling 17 minutes after castration, the piglets were left undisturbed in the home pen.PT, procedure type: CA, castration after injection of local anesthesia, CC, castration without anesthesia, SH, sham handling.IT05, castration 5 minutes after intra-testicular administration of 0.5 mL of drug per testis; IT05 V03 , castration 5 minutes after intra-testicular administration of 0.3 mL of drug per testis; SH1, sham handling with one stay in the bench; SH2, sham handling with two stays in the bench.
In addition, the four categories of movements were summed into a total count of movements per second.

Saliva cortisol concentrations
For baseline, one saliva sample was collected per piglet, immediately after the piglet was picked up from the home pen, on average 40 ( ± 12, Study 1) and 35 ( ± 14, Study 2) minutes before the initiation of the first procedure.To evaluate changes in the concentration of cortisol in saliva in response to the procedures, another sample was collected per piglet on average 17 ( ± 9, Study 1; ± 1, Study 2) minutes after castration, and again approximately 6 hours after castration (5 hours 47 minutes ± 26 minutes, Study 1 only).A cotton swab (Salivette, Sarstedt Aktiengesellschaft & Co., Numbrecht, Germany) was cut in pieces (2.0 × 0.5 cm), soaked in concentrated apple juice (nectar from concentrated juice, minimum 60%, Rynkeby Foods A/S, Ringe, Denmark) for 1 hour, and dried in an electric oven at 60 °C for 5 hours.During sampling, the cotton swab was fixed at the end of a straight forceps and carefully introduced into the piglet's mouth, while the piglet was held in the experimenter's arms.The cotton was rotated carefully in the piglet's mouth for 30-45 seconds, with insistence around the salivary glands.This procedure was performed by one of four trained experimenters, blinded to the experimental treatments.The cotton swab was then placed in an experimental tube (provided as part of the Salivette), labeled, and stored at −18 °C until cortisol concentration determination at the departmental laboratory, according to the method described in Coutant et al. (2022a).

Test of social motivation
Immediately after castration, piglets were individually subjected to a 180-second test in an open arena, aiming to assess indicators of their motivation to reach siblings placed at the end of the arena, as described in Coutant et al. (2022c).The arena (2.4 × 0.8 m) was situated in a corridor outside the farrowing room.The behavior of piglets during testing was video-recorded using a camera (Logitech HD Pro Webcam, Logitech, Lausanne, Switzerland) placed approximately 2 m above the arena and connected to a computer.Two experimenters, blinded to the experimental treatments, analyzed the video recordings and recorded the latency to start moving and to reach the littermates (seconds) using stopwatches.In addition, the footage was analyzed using a tracking software (Lolitrack 5 -2D tracking, LoligoSystems, Viborg, Denmark) to determine the speed (cm/s) and distance walked (cm) for each piglet.Test success (later considered as a binary variable) was defined as the piglet reaching the grid separating him from his littermates within 180 seconds.

Home pen behavior
In Study 2, immediately upon return to the home pen, piglet behavior was recorded by a single experimenter standing in front of the pen, by use of a hand-held computer (PsionWorkabout, Psion PLC, London, UK).Behavior of each piglet was recorded continuously for 10 minutes from re-introduction in the pen by focal sampling (Bateson and Martin, 2021), based on an ethogram (Table 5) developed by Coutant et al. (2022b).The ethogram included behavioral states (active, active at the udder, resting, huddled up, prostrated) and events (shivering, rump scratching, self-scratching, tail wagging, seeking sow, playing).

Human-animal relationship test
In Study 1, approximately 6 hours (5 hours 42 minutes ± 25 minutes) after castration, piglets were individually subjected to an HAR test in the corridor outside the farrowing room.Each piglet was transported to an arena in a plastic box layered with straw (37 × 27 × 26 cm) and placed in the entrance area of the arena (Figure 2).At one end of the arena (measuring 3.4 × 0.8 m), an observer wearing the same color clothing as the technician who performed the castration earlier in the day was seated.The observer released the piglet into the arena by sliding the door of the entrance area (pulling a cord from the seated position approximately 150 cm from the entrance area), and the piglet was allowed 2 minutes to enter the arena (defined as two front legs out of the entrance area).If

Table 3
Description of the vocal parameters analyzed for each piglet during injection of local anesthetic, castration, or sham handling, all performed while the piglet was in a castration bench.

Parameter (unit) Description
Call proportion Proportion of time spent vocalizing during the procedure, calculated as call duration/procedure duration.Call per second (s −1 ) Number of calls per s of the procedure.

Mean call duration (s)
Average duration of a call during the procedure, calculated as sum of call durations/number of calls.Mean energy (dB) Average energy, calculated as an average of the energy of each call during the procedure.Max energy (dB) Maximum value of energy recorded for all calls during the procedure.Max power (dB) The maximum power recorded for all calls during the procedure, relative to the specific recording setup.Max amplitude (U) Maximum absolute value of sound wave (signal) recorded for all calls during the procedure.Sum of entropy (kilobits) Aggregated disorder of a call obtained by analyzing the energy distribution within each call.Higher entropy values correspond to greater disorder in the sound whereas a pure tone would have zero entropy (Charif et al., 2010).Max entropy (kilobits) Highest value of disorder recorded for all calls during the procedure.Average entropy (kilobits) Average disorder recorded within all calls.The value corresponds the average of the average entropy recorded for each call.

Table 4
Description of the leg resistance movements recorded during injection of local anesthetic, castration, or sham handling, all performed while the piglet was in a castration bench.

Category Description
Flexion Piglet vertically bends the front leg, provoking a flexion of the elbow of at least 90 degrees.

Extension
Piglet fully extends the front leg while lowering the head in the bench.May be accompanied by trembling of the leg and/or by a subtle lift of the piglet's back.

Kick
Piglet front leg performs a sudden upward movement, changing from a flexion to a tense upward position.

Blow
Piglet suddenly draws back the front leg forward or backward for at least half a bench length, from a normal upright position to an extended position, with little or no flexion of the elbow.

Leg blocked
Piglet's front leg is blocked in the bench cone, preventing movement.
the piglet did not exit the entrance area within this interval, the HAR test could not be performed, as the test started upon entrance in the arena.During the next 2 minutes, the piglet was allowed to move freely in the arena, and the frequency of each category of vocalizations was recorded using either a hand-held computer (Psion Workabout, Psion PLC, London, UK) or a pen and paper according to a pre-defined ethogram (Table 6).The two persons acting like observers during the test practiced identification of piglet vocalizations for 2 hours prior to the test and showed a high inter-observer reliability with an intra-class correlation coefficient close to 78% (95% confidence interval: 62%-88%; comparing 10 audio recordings of the test).

Handling score
Immediately after the end of the HAR test, the observer quietly stood up and walked toward the piglet.The piglet was picked up, and held in the observer's arms for 30 seconds.During this process, the observer evaluated the piglet's response to handling based on the scale defined in Table 7.The scores for vocalizations and movements, during pick up and once in the arms of the observer, were summed into a total handling score.

Data handling
Data for some piglets had to be removed before statistical analysis due to missing values (or technical issues related to the measurement, Figure 3) as missing values are incompatible with the applied multivariate method.Baseline cortisol concentration (prior to the procedures) was not included in the analysis as this was not considered a response.Considering that the two studies included different indicators, three datasets were created: Study 1 (including 406 piglets and 30 parameters), Study 2 (including 174 piglets and 35 parameters), and both studies (Study 1 + 2) combined (including 611 piglets and 21 parameters; Figure 3).

Statistical analysis
In the first step of the analysis, Spearman's correlations were calculated among all parameters and presented in a matrix.Significant correlations were defined with a significance level of 0.05, without adjustment for multiple comparisons.All correlation matrices are available in Supplementary Material.For ease of interpretation, graphical representations of these matrices were obtained in the form of heat maps, showing only Spearman's rank correlation coefficients of statistical significance calculated in t-tests (Figures 4-6).Trembling and/or quick and involuntary contraction of muscles under the skin (spasm).An event is defined as either one shivering/contraction or a set of contractions separated by less than 1 s Rump scratching Rubbing the rump against the floor or the pen walls.Self-scratching Scratching of the rump or lower belly area with a hind-leg.

Tail wagging
Up and down or side to side movement of the tail, with a range superior to body width.An event is defined as either one movement or a set of movements separated by less than 1 s Seeking sow Physically moving the body (often associated with vocalization) in direction of the sow's head.The body of the piglet is moved at least 10 cm.Playing Head shaking, jumping, running with bouncy movements, possibly involving other piglets (gentle nudging or pushing, mounting or chasing).An event is counted until the behavior is stopped for a least 1 s or until the piglet performs another type behavior.Relationships among indicator types within a dataset were then further investigated using a principal component analysis (PCA) with orthogonal (varimax) rotation.If needed, parameters were either square root transformed or logarithmically transformed prior to the PCA to not reject normality.The concentration of cortisol in saliva quantified 6 hours after castration (Study 1) was not included in the analysis due to a higher risk of misrepresentative results related to the inclusion of repeated measures in the multivariate analysis (Budaev, 2010).Parameters showing communality estimates below 0.3 were removed step-wise during the analysis until reaching a Kaiser-Meyer-Olkin measure of sampling adequacy above 0.6 (Budaev, 2010).Principal components (PC) with an absolute eigenvalue above 1 were selected for further analysis.If this method resulted in the number of PCs superior to 4, the scree plot was assessed to determine another logical cut-off point, with the objective of explaining at least 55% of the overall variation across all selected PCs.Main components for each PC were determined as the indicators having factor loadings superior to 0.6 or inferior to −0.6.
Potential effects of procedure type (castration after injection of anesthetic, castration without prior anesthesia, sham handling, see Tables 1 and 2), piglet weight, age, or time of day on each PC were then investigated.Using each PC as response, mixed effects models were carried out for each dataset with procedure type, weight, age, hour of day as fixed effects, and litter as random effect.Satterthwaite's approximation of denominator degrees of freedom was used.Pairwise comparisons among procedure types were performed with P-values adjusted for multiple comparisons using the Tukey-Kramer method.Deviations from assumption of normality and variance homogeneity were monitored visually by plotting residuals at each step.
Correlation matrices were obtained by use of the statistical software R (R Core Team, 2022).All other analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, North Carolina, USA).All data used for statistical analyses can be found in the Supplementary Material.Descriptive measures are presented as average ± standard error.

Correlations between indicators
Correlation coefficients for each pair of parameters are available in Supplementary Material.For the sake of simplicity, only strong correlations (ρ > 0.5) are reported in this section.

Study 1
Correlations among all parameters within indicator types recorded in Study 1 can be seen in Figure 4. Strong positive correlations were found between all parameters within the resistance The handling score was calculated as sum of vocalization and movements scores recorded during pick-up and once in the arms of the human.movements indicator type, at the exception of blow per second and kick per second (ρ < 0.2).Within the vocalization indicator type, strong positive correlations were found between the parameters mean call energy, max call energy, max call power, and max call amplitude.Sum of call entropy was also strongly correlated with the number of calls per second, while mean call energy was positively correlated to call proportion.Latency to reach littermates in the social motivation test was negatively correlated to piglet speed in the test (ρ = −0.8).No strong correlations were found between parameters from the resistance movements indicator type and parameters from the vocalization indicator type.Significant correlations were found between some parameters, though (e.g., calls per second and total number of resistance movements per second), but the strength of the correlations was weak (ρ < 0.2).Parameters from the social motivation test did not correlate strongly with the resistance movements nor vocalizations during castration, neither did the parameters from reporting of vocalizations during the HAR test.
Last, the post-castration saliva cortisol concentrations did not correlate strongly with any of the other recorded parameters, although significant but weak correlations were found between the number of squeals and grunt-squeals in the HAR test, the distance walked in the social motivation test, and the cortisol concentrations 6 hours after castration (ρ < 0.2).

Study 2
Correlations among all parameters recorded in Study 2 can be seen in Figure 5.As in Study 1, positive correlations were found within the parameters of the resistance movement indicator type, at the exception of blow per second and kick per second.Strong positive correlations (ρ > 0.4) were found between most of the parameters reporting vocalizations, at the exception of calls per second and mean energy, and call per second and max power.Latency to reach littermates in the social motivation test was negatively correlated to the speed while moving during the test (ρ = −0.7).None of the behaviors observed during the home pen observations showed a strong correlation to each other, though the occurrence of shivering was positively correlated to the duration of resting (ρ = 0.4), and the occurrence of rump scratching was significantly correlated to the number of tail wags observed (ρ = 0.4).
Most of the vocal parameters showed significant correlations with parameters of resistance movements during castration, at the exception of average call duration and maximum call power.Although the strength of the correlations was not categorized as strong, the correlation coefficients exceeded the ones observed in Study 1 (0.2 < ρ < 0.5).Overall, neither the parameters from the social motivation test nor the behavioral parameters obtained during the home pen observations correlated strongly with the acute parameters (indicators of resistance movements and vocalizations) recorded during castration.As in Study 1, the post-castration saliva cortisol concentration did not correlate significantly with any of the parameters recorded at castration, but weak correlations were found with some post-castration behaviors, such as distance walked in the social motivation test (ρ = 0.1), duration of huddling up (ρ = −0.1),and the occurrence of rump scratches (ρ = 0.1).

Study 1 and 2 combined
Correlations among parameters recorded in both Study 1 and 2 can be seen in Figure 6.As evident when analyzing Study 1 and 2 separately, parameters of resistance movements were strongly correlated with each other, at the exception of number of blows per second and kick per second.Vocal parameters relating call energy, call power, and call amplitude were correlated strongly, but were not all significantly correlated to parameters of call entropy.Maximum, average, and sum of call entropy correlated with the number of calls per second (ρ > 0.5).Latency to reach littermates in the social motivation test correlated negatively with the speed of movement during the test (ρ = −0.7).Again, significant correlations were found among parameters from indicators of the vocalization and resistance movement types, at the exception of average call duration and maximum call power parameters, though correlation coefficients were weak (ρ < 0.2).Similarly, some significant correlations were observed between parameters of the social motivation test (latency to reach siblings and speed of movement) and vocalization parameters, but correlation coefficients did not exceed 0.2.Post-castration saliva cortisol concentration did not correlate to any other parameter, at the exception of a weak correlation with the distance walked in the social motivation test (ρ = 0.1).

Study 1
When examining data from Study 1, the principal component analysis resulted in six PCs explaining, together, 56.2% of the total variation in piglet responses (Table 8).For the first component, PC1, the only parameters loading with absolute value above 0.6 reflected vocal intensity, i.e., max call energy, mean call energy, and max call amplitude.PC2 reflected piglets' resistance movements during the procedure, and the main loadings included all types of foreleg movements at the exception of the behavior called blow.PC3 was highly loaded with factors reflecting behavior of the piglets in the social motivation test after castration, namely speed of movement, distance walked, and latency to move and to reach littermates.PC4 reflected vocal parameters related to the number of calls emitted during the procedure, i.e., calls per second and sum of call entropy.PC5 reflected the number of vocalizations during the HAR test, with high loadings on the total number of vocalizations and the number of squeals emitted during the test.Last, PC6 was highly loaded with average call duration during castration.

Study 2
Analyzing data from Study 2, the principal component analysis resulted in six PCs explaining, together, 60.4% of the total variation in piglet responses (Table 9).Similar to Study 1, PC1 was highly loaded with parameters reflecting vocal intensity, i.e., max call energy, mean call energy, and max call amplitude, but also with the average call duration and call proportion during the procedure.PC2 reflected the resistance movements of the piglets during the procedure, at the exception of the behavior called blow.PC3 loaded the post-castration saliva cortisol concentration, the duration of activity at the udder and resting in the home pen after castration, as well as the occurrence of shivering recorded during the same period of time.PC4 was highly loaded with parameters from the test of social motivation after castration, i.e., latency to start moving in the arena, speed of movement, and latency to reach the littermates.High loadings for PC5 included the number of tail wags and scratching events recorded in the home pen after castration.Last, PC6 reflected the number of calls performed during the procedure, i.e., the number of calls per s.

Study 1 and 2 combined
Combining data from Study 1 and 2, the PCA resulted in four PCs, explaining together 60.8% of the variation in piglet responses (Table 10).PC1 reflected vocal intensity, i.e., mean call energy, max call energy, max call power, and max call amplitude.PC2 was highly loaded with parameters of foreleg resistance movements, excluding the behavior called blow.High loadings for PC3 included aspects of piglet vocalizations during castration, namely call proportion, calls per second, max call entropy, and sum of call entropy.PC4 reflected parameters of the social motivation test performed after castration, i.e., speed of movement, distance walked during the test, and latencies to start moving and to reach the littermates.

Study 1
All PCs were significantly associated with the type of procedure (castration after injection of anesthetic, CA, castration without prior anesthesia, CC, sham handling, SH; Table 11) at the exception of PC5 (HAR vocalizations) and PC6 (call duration).In PC1 (vocal intensity) and PC2 (resistance), the three procedure types differed, with highest PC values recorded in CC and lowest PC values recorded in SH.In PC3 (social motivation test), higher PC values were recorded in CA compared to SH or CC.In PC4 (vocal count), all three procedure types differed, with, this time, highest PC values recorded in CC and lowest PC values recorded in CA.In addition, PC1 was positively associated with piglet weight (P = 0.036) and PC5 with piglet age (P = 0.006).

Study 2
As in Study 1, all PCs were significantly associated with the type of procedure (Table 11), at the exception of PC3 (post-castration saliva cortisol, activity at udder and post-castration shivering) and PC6 (vocal count).In PC1 (vocal characteristics), higher PC values were recorded in CC than CA and SH.Higher PC2 (resistance movements) values were observed in CA and CC compared to SH.In PC4 (social motivation test), higher PC values were recorded in CA than SH.Higher values of PC5 (post-castration tail wag and scratching) were observed in CA compared to SH.In addition, piglet weight was positively associated with PC5 (P = 0.043).

Study 1 and 2 combined
Type of procedure was significantly associated with all PCs.High PC1 (vocal intensity) and PC2 (resistance movements) values were recorded more in CC than in CA, with the lowest values observed in SH.In PC3 (vocal rate and call entropy), all procedure types differed from each other, with highest values recorded in CA and lowest values observed in CC.Higher PC4 (social motivation test) values were recorded in CA compared to SH or CC.In addition, heavier   piglets had significantly greater PC1 values (P = 0.007).Age was significantly associated with PC2, with greater values recorded in 4 compared to 3 days old piglets.

Discussion
The aim of this paper was to get further insight into the potential relationships between different types of indicators (and parameters within) of piglets' responses to surgical castration used to interpret the level of pain and stress involved.The methods of analysis, associating correlation matrices and multivariate analysis, are exploratory, but allowed us to identify patterns across indicator types and within associated parameters.
A first noticeable finding is the few correlations observed between parameters from different types of indicators.While such results were somehow expected between certain indicators, for example between vocal parameters recorded during castration and responses in the post-castration social motivation test [as these parameters may reflect motivational elements and/or physiological states distinct from the pain experienced during the procedure (Coutant et al., 2022c)], the general lack of correlations between different indicator types was rather unexpected.For instance, acute responses to castration, such as vocal parameters and resistance movements, were expected to somewhat correlate.The two types of recordings are frequently used in combination to detect responses of piglets to castration and/or anesthetic injection and have been shown to lead to comparable patterns of response across different experimental treatments (e.g., Skade et al., 2021;Coutant et al., 2022a).Yet, despite showing some positive correlations, the two types of indicators loaded on different principal components, across analyses.This finding may suggest that foreleg resistance movements and vocal parameters reflect different parts of the variation in the piglets' responses.Similarly, behavioral observations recorded upon return to the home pen are often thought to reflect, to some degree, the level of pain experienced in response to castration (Hay et al., 2003).Yet, none of the acute parameters showed strong correlations with any of the behaviors recorded in the home pen.Although this could be partly explained by the particular methodology of behavioral recording used in the study (e.g., relatively short duration of observation), it is somewhat in line with previous work showing no difference in home pen behaviors following castration with or without anesthesia (Hansson et al., 2011;Kluivers-Poodt et al., 2013).Last, saliva cortisol concentration following castration did not show significant correlation with any of the parameters recorded during castration, despite its common interpretation as an acute indicator of arousal, pain, and stress (Molony and Kent, 1997).
Even within indicator types, the choice of parameter seems to be of importance.In the principal component analyses, clear distinctions could for instance be detected between vocal indicators related to the maximum intensity of the calls (maximum energy, power, or amplitude), the number of calls emitted (call proportion, sum of call entropy), and the average call duration.The percentage of variance explained by the principal component loading with maximum call intensity, as well as analyses of the differences between types of procedures for all principal components, may suggest that call intensity may be more specific, among these parameters, to record differences between castration treatments.Indeed, the different types of procedures did not lead to significantly different average duration of calls in Study 1, nor different number of calls in Study 2, while analysis of vocal count in Study 1 led to contradictory results as compared to the other vocal indicators.This finding is in line with previous work on piglet vocalizations during castration, suggesting that parameters describing a single event of the call, such as maximum call intensity, may be more specific and sensitive than averages (Marx et al., 2003).Similarly, the number of foreleg blow movements only represented a low loading on the principal components in the multivariate analysis, indicating that this particular type of leg movement did not significantly explain variation between piglets.Yet, this finding could partly be explained by a relatively low occurrence of this leg movement category as opposed to the other categories, flexion, extension and kick.Overall, the different behaviors recorded in the home pen showed low correlations with each other, although certain behaviors bore high loadings in common principal components.This finding was somewhat expected, as the ethogram used for behavioral observations included a combination of non-specific and castration-related behaviors, potentially affected by the procedures to a different extent.Spontaneous or non-specific behaviors such as resting, nursing, or general postures, also referred to as non-evoked behaviors (Mogil, 2020), have for instance been shown to poorly reflect piglets' response to pain mitigating strategies at castration, while castration-related behaviors, or evoked behaviors, such as prostration or rump scratching have been shown to more be specific (Sheil and Polkinghorne, 2020).
It can also be noticed that a number of indicators seem to be associated with piglet age or weight.While this result could be due to differences in piglets' responses to the procedures based on their physical characteristics, the lack of consistency of these effects across principal components, and studies, limit this interpretation.As suggested previously (Coutant et al., 2022a), a likely explanation relates to the physical capacities of the piglets to express some of the recorded behaviors.Heavier piglets may for instance have a large thoracic cage allowing the expression of more intense calls, while their weight may affect the fit in the castration bench, and, indirectly, the capacity to express foreleg resistance movements.We also note that our results are based on a limited age (3-4 days) and weight (0.9-2.7 kg) range.However, even this narrow variation affected the results.Thus, in order to reach a better comparability between studies of pain, our results stress the importance of correct determination and reporting of the age and weight of piglets involved in experimental studies.
In addition, the strength of the present correlations between different indicator types was somewhat affected by the study design and/or number of piglets involved.For instance, stronger correlations between resistance movements and vocal parameters were observed in Study 2 as compared to Study 1.While this result could be an artifact related to the number of piglets and/or the nature of the indicators incorporated in the correlation matrices, it could also relate to the nature of the treatments included in the two studies.Indeed, analysis of the acute responses to the different experimental treatments showed stronger responses in Study 1 as compared to Study 2, potentially because Study 1 included a number of anesthesia treatments shown to be poorly effective at castration (see Coutant et al., 2022a), while Study 2 included a larger proportion of sham-handled piglets and, thus overall, milder responses at castration (see Coutant et al., 2022b).Although speculative, this observation could indicate that the choice of indicators recorded in a given study should take into consideration the nature of the experimental treatments tested, and their expected degree of effect on the piglets.This warrants further study.
Altogether, the present results highlight the importance of carefully choosing indicators, as this choice may greatly influence the outcome of a particular study.In other words, the chosen indicators (and parameters) may reflect different underlying mechanisms, and are therefore probably not interchangeable.This suggestion encourages further pre-study description and justification of the reasoning behind the choice of indicators used to quantify piglets' response to castration (which, to the best of our knowledge, is often lacking in the literature) and draws attention to careful interpretation of results.The specific setup of a given study, i.e., number of piglets per experimental treatment and expected variation between treatments, should always be taken into consideration.In addition, the terms used to interpret each indicator should be carefully evaluated.Simply put, vocal parameters, resistance movements, cortisol concentrations, and home pen behaviors (to only cite those) may not be all qualified as "pain responses," as they are also expressed-maybe at a lower level-when no pain is present.In that sense, naming indicators after an emotional state or physiological state, that may be present, can be seen as one way of over-interpreting results, as discussed by Nelson (2018).Thus, interpretation and labeling as "pain," "stress," or other defined states remains complex until the parameters are further validated and judged as highly specific.This type of exploratory study and associated discussion is of importance to ensure high quality of research, and should therefore be replicated on a larger scale.Future work could include other types of pain and stress indicators used to evaluate castration, such as the piglet grimace scale (Viscardi et al., 2017), and could be applied on a wider range of contexts, for instance for tail docking of piglets or disbudding of calves.
Precise description of the phenomena underlying each indicator is out of the scope of this paper and requires further validation studies, although limitations inherent to the recording of subjective emotional states will probably remain (Weary et al., 2017).However, strong correlations between certain indicator types and parameters allow for discussion about the phenomena to which they relate.For instance, the finding that salivary cortisol concentration after castration loaded together with activity at the udder, resting and shivering in the home pen may support the hypothesis that cortisol may not record acute response to the procedure, but rather is associated to the post-surgical state.In addition, the principal component on which these parameters loaded did not differ significantly between the procedures, which may suggest that pain was not the main underlying phenomenon.Rather, the combination of responses shown by these parameters may be interpreted as a physiological reaction to the procedure, potentially involving thermoregulation and activity level.Further studies involving cortisol sampling in response to handling and/or drug administration alone are required to examine this possibility.Similarly, the finding that tail wagging and scratching loaded together on a principal component seems to contradict the hypothesis that tail wagging is indicator of positive emotion (Reimert et al., 2013) in the context of castration.Rather, the combination of these two indicators and the difference in their occurrence among procedures may be interpreted as an indication of discomfort in the hind area caused by the invasive procedures, including injection of local anesthesia and later castration.Last, the number of vocalizations performed during the human-animal relationship test, and more precisely the number of squeals recorded, did explain a small percentage of the variation in piglets' responses to castration.This principal component did not relate to any other type of indicator, and the vocalizations alone showed poor correlations with the other types of indicators.The suggestion that this test reflects piglets' experience to castration (Waiblinger et al., 2006), was therefore not confirmed by the present findings.Compared to earlier work (e.g., Tallet et al., 2019), these results may be explained by the specifics of the involved test setup (e.g., no habituation, voluntary entrance to the arena) and by the relatively young age of the piglets.

Conclusions
In the absence of a gold-standard to record pain during and after piglet castration, multi-dimensional recording approaches are of interest.This work is a first step to increase the understanding of phenomena underlying various behavioral and physiological responses recorded in studies of castration.The present findings suggest that the nature of the selected indicators and the precise behaviors recorded may influence results and conclusions.Thus, the findings, although explorative in nature, draw attention to a careful consideration of the differences that exist among different types of indicators, and the consequent importance of nuancing interpretation and avoiding simplified labeling of the responses recorded.Such knowledge is crucial in order to avoid misinterpretation of results and to improve the evaluation of any welfare effects of castration after local anesthesia and related painmitigating measures.

Figure 1 .
Figure 1.Diagram of the recordings performed during Study 1 and 2.The piglets were brought to the castration room with some of their littermates, anaesthetized and castrated one by one, then individually subjected to a social motivation test prior to being returned to their home pen.In Study 1, piglets were individually subjected to saliva sampling 17 minutes after castration, then placed back in the pen again and left undisturbed until being brought back to the corridor outside of the farrowing room for individual human-animal testing and saliva sampling 6 hours after castration.The piglets were then brought back to the home pen and left undisturbed.In Study 2, after the social motivation test, the behavior of the piglets was observed continuously for 10 minutes upon return in the home pen.After saliva sampling 17 minutes after castration, the piglets were left undisturbed in the home pen.

Figure 2 .
Figure 2. Arena used for the 2-minute human-animal relationship test, in which the piglet enters via a sliding door.

Figure 3 .
Figure 3. Flow diagram of the number of piglets tested for each type of indicator in Study 1 and Study 2, and the resulting number of piglets included in the corresponding datasets.Items with slipped corner represent the number of piglets removed at a particular stage and the underlying reasons.The number of piglets included for each type of indicator are independent from each other.

Figure 4 .
Figure 4. Heat map of the correlations among all parameters in Study 1. Pairwise correlations between parameters within main indicator types are represented as a square and grouped according to indicator types (resistance movements, vocalizations, social motivation test, human-animal relationship test, cortisol response).White coloring indicates a non-significant correlation (P < 0.05).Colored squares indicate a significant correlation.The legend on the right side of the matrix indicates the strength and direction of the correlation.

Figure 5 .
Figure 5. Heat map of the correlations among all parameters in Study 2. Pairwise correlations between parameters within main indicator types are represented as a square and grouped according to indicator types (resistance movements, vocalizations, social motivation test, home pen behavior, cortisol response).White coloring indicates a nonsignificant correlation (P < 0.05).Colored squares indicate a significant correlation.The legend on the right side of the matrix indicates the strength and direction of the correlation.

Figure 6 .
Figure 6.Heat map of the correlations among all parameters common to Study 1 and Study 2. Pairwise correlations between parameters within main indicator types are represented as a square, and grouped according to indicator types (resistance movements, vocalizations, social motivation test, cortisol response).White coloring indicates a nonsignificant correlation (P < 0.05).Colored squares indicate a significant correlation.The legend on the right side of the matrix indicates the strength and direction of the correlation.

Table 1
Description of the 13 treatment groups involved in Study 1.

Table 2
Description of the five treatment groups involved in Study 2.

Table 5
Ethogram used during continuous observation of piglet behavior upon return to the home pen.Active Behavioral state characterized by being upright and moving (walking, running, nosing, chewing) except described in another category.Active at the udder Nursing (with active milk intake), working for access to a teat, or active massaging or nosing of the udder.RestingSitting or standing motionless with the head above shoulder level, or lying, with eyes closed or open, legs not tucked under the body.Huddled upLying with at least three legs tucked under the body.ProstratedSitting or standing motionless with the eyes opened and the head down, lower than shoulder level.Social huddlingLying/resting together with other piglets in such way that part of the body or the whole body is not visible by the observer.Out Not visible for the observer.

Table 6
Ethogram used during recording of piglet vocalizations during the human-animal relationship test.Vocalization Description Short grunt Low tone of less than half a second (one note) Long grunt Low tone of more than half a second (one note) Grunt-squeal Low tone that transforms into a high tone Squeal High tone (different notes) ScreamHigh, long and loud tone, often as long as an expiration Bark Low tone that sounds like "wuff"

Table 7
Scale used to determine the handling score immediately after the end of the human-animal relationship test.

Table 8
Factor loadings of the six principal components (PC) obtained for Study 1.

Table 9
Factor loadings of the six principal components (PC) obtained for Study 2.
Factor loadings with relatively high absolute value (≥0.6 or ≤−0.6) are indicated in bold.Only eigenvalues above 1 are presented.

Table 10
Factor loadings of the four principal components (PC) obtained for the combined analysis of Study 1 and 2.
abc Different letters within a row indicate significant differences between procedure types. 1 Human-animal relationship test.