Are infrared thermography, feeding behavior, and heart rate variability measures capable of characterizing group-housed sow social hierarchies?

Abstract

Group gestation housing is quickly becoming standard practice in commercial swine production. However, poor performance and welfare in group housed sows may result from the formation and maintenance of the social hierarchy within the pen. In the future, the ability to quickly characterize the social hierarchy via precision technologies could be beneficial to producers for identifying animals at risk of poor welfare outcomes. Therefore, the objective of this study was to investigate the use of infrared thermography (IRT), automated electronic sow feeding systems, and heart rate monitors as potential technologies for detecting the social hierarchy within five groups of sows. Behavioral data collection occurred for 12 h after introducing five sow groups (1–5; n = 14, 12, 15, 15, and 17, respectively) to group gestation housing to determine the social hierarchy and allocate individual sows to 1 of 4 rank quartiles (RQ 1–4). Sows within RQ1 were ranked highest while RQ4 sows were ranked lowest within the hierarchy. Infrared thermal images were taken behind the neck at the base of the ear of each sow on days 3, 15, 30, 45, 60, 75, 90, and 105 of the experiment. Two electronic sow feeders tracked feeding behavior throughout the gestation period. Heart rate monitors were worn by 10 randomly selected sows per repetition for 1 h prior to and 4 h after reintroduction to group gestation housing to collect heart rate variability (HRV). No differences were found between RQ for any IRT characteristic. Sows within RQ3 and RQ4 had the greatest number of visits to the electronic sow feeders overall (P < 0.04) but spent shorter time per visit in feeders (P < 0.05) than RQ1 and RQ2 sows. There was an interaction of RQ with hour for feed offered (P = 0.0003), with differences between RQ occurring in hour 0, 1, 2, and 8. Higher-ranked sows (RQ1 and RQ2) occupied the feeder for longer during the first hour than lower ranking sows (RQ3 and RQ4; P < 0.04), while RQ3 sows occupied the feeder longer than RQ1 sows during hour 6, 7, and 8 (P < 0.02). Heart beat interval (RR) collected prior to group housing introduction differed between RQ (P = 0.04), with RQ3 sows exhibiting lower RR compared to RQ2 sows (P = 0.009). Overall, these results indicate that feeding behavior and HRV measures may be capable of characterizing social hierarchy in a group housing system.

Establishment and maintenance of the sow social hierarchy in group gestation housing may lead to compromised sow welfare and performance. This study evaluated whether the sow social hierarchy could be detected using precision technologies.

Introduction

Group gestation housing is quickly becoming a standard practice in commercial swine production in the United States, with approximately 33% of US producers currently employing group gestation housing in their commercial production systems (personal communication with the National Pork Producers Council, 2022). In contrast with the EU, where group gestation housing is required (Directive 2008/120/EC; European Commission, 2008), commercial pork producers in the US have traditionally housed gestating sows in individual gestation stalls. Group housing offers several benefits for sow welfare including improved behavioral variability and expression (Barnett et al., 2001; Elmore et al., 2011). However, group gestation housing can also lead to several management challenges for producers, such as injuries and poor performance, which occur after mixing and can last throughout the gestation phase (Bench et al., 2013). Aggressive behavior among sows and gilts is expected during and shortly after mixing, as sows within groups fight in order to establish a social hierarchy (Arey and Edwards, 1998; Kirkwood and Zanella, 2005).

The establishment of a social hierarchy in mammalian species results from an individual’s need for access to resources. For example, group gestation housing systems that utilize electronic sow feeding systems (ESFS) often have fewer feeding spaces than traditional individual stall systems, which creates competition for feeding access (Liu et al., 2021). The process of establishing the social hierarchy involves aggressive interactions that may cause injuries, such as lesions and lameness, as well as poor reproductive performance (Munsterhjelm et al., 2008; Greenwood et al., 2014). The aggressive interactions among sows and gilts in group housing systems can lessen over time. However, an individual sow ranked lower in the hierarchy may have limited access to feeder space throughout the entire gestation period due to higher ranked sows’ dominance over the feeder (Bench et al., 2013). This leads to unconsumed rations which can cause poor body condition, performance, and welfare outcomes (Spoolder et al., 2009; Verdon et al., 2015; Maes et al., 2016).

Currently, determining the social hierarchy in group-housed breeding females is time and labor intensive. However, the ability to quickly identify sows and gilt positioning within the social hierarchy could be beneficial to producers for mitigating any associated poor welfare outcomes. One potential option for overcoming this challenge is to introduce the use of precision livestock farming technologies, which have become useful for monitoring health and animal welfare in livestock management (Brown-Brandl et al., 2013; Vargovic et al., 2021). Infrared thermography (IRT) and radio frequency identification (RFID) feeding system technology are capable of providing real time monitoring of commercially kept livestock (Halachmi and Guarino, 2016). Additionally, with the future development of automated heart rate imaging technologies (Pai et al., 2021; Wang et al., 2021), heart rate (HR) and HR variability (HRV) measures may be a useful physiological measure for identifying the impact of social stress on swine (De Jong et al., 2000; Marchant-Forde et al., 2004).

The ability to utilize a combination of currently available and future precision farming livestock technologies to characterize the social hierarchy of group-housed sows could benefit producers in managing their breeding herd’s welfare and performance. Therefore, the objective of this study was to investigate the use of IRT characteristics, feeding behavior traits obtained from an ESFS, and HRV for detecting the social hierarchy of group housed sows.

Materials and Methods

All procedures and methods performed in this study were conducted at the North Dakota University Swine Research Unit and approved by the North Dakota State University Institutional Animal Care and Use Committee (IACUC #A21044).

Animals and housing

Thirteen Yorkshire and thirty-six Yorkshire-crossbred sows with previous experience in group gestation housing were utilized in this study. Five repetitions (n = 14, n = 12, n = 15, n = 15, n = 17, for repetitions 1–5, respectively) of the study were performed over a 12-mo period between May 2021 and May 2022. Seventeen sows were observed in two repetitions. All sows were 12–36 mo of age and ranged from 1 to 7 parities.

Three group gestation pens measuring approximately 7.30 × 7.62 m (Big Dutchman, Holland, MI, USA) were used. Each of the pens were partially slatted and included a protecting wall or a “T” configuration that created two resting areas (each measuring approximately 2.30 × 2.24 m) and a larger communal area. The communal area included two electronic sow feeders (ESF; Big Dutchman, Holland, MI, USA) equipped with RFID technology and a full body race with a rear gate. Each group gestation pen included three water cups with nipples that provided ad libitum water access. The ESF manufacturer’s recommendation for group size is 20 sows per feeder. In this study, group size varied between replicates, ranging from 12 to 17. While housed in group gestation, sows could consume up to 1.81 kg of a corn-soybean meal-based gestation diet that was formulated to meet or exceed NRC (2012) nutrient requirements.

Three to 5 d after introduction to group gestation pens, sows were estrus or “heat” checked utilizing a combination of nose-to-nose contact with boars located in an adjacent heat detector pen and physical stimulation by a caretaker, which included applying pressure to the back of the sow to assess standing estrus. Sows identified to be in standing heat were artificially inseminated in the group gestation pen. Sows were pregnancy checked by ultrasound (Xinda 8300 Veterinary Ultrasound Scanner, Medical Device Co., Ltd., Zhangjing, Wuxi, China) approximately 35 d after breeding and remained in group gestation pens throughout the entire gestation period. This standard operating procedure at our facility differs from other commercial facilities, where sows may be housed individually until pregnancy is confirmed before returning to group gestation housing.

On day 105 of gestation, sows were moved to individual farrowing stalls (1.83 m × 0.74 m; Big Dutchman, Holland, MI, USA) in a separate farrowing room within the facility. Once in farrowing stalls, sows were fed 1.36 kg of a corn–soybean meal-based lactation diet that was formulated to meet NRC (2012) nutrient requirements twice daily until farrowing. After farrowing, sows were fed to appetite with rations increasing by 0.91 kg each day until the total ration equaled a maximum of 7.26 kg/d. Sows and litters remained together in farrowing stalls for approximately 18 d. Following the lactation period, piglets were weaned from sows and relocated to a separate nursery room. Once piglets were weaned, sows were reintroduced to group gestation housing. Gestation groups consisted of sows who were previously acquainted and parity 1 sows that were previously housed in a pen consisting entirely of gilts prior to their relocation to farrowing stalls.

Behavioral video collection

Behavioral video collection occurred for all sows upon reintroduction to group gestation housing (experimental procedure day 1), until day 4 of the study. Four video cameras (Lorex LBV2531U Security Cameras, Lorex Technology Inc., Markham, Canada) were installed above each pen to ensure full coverage of the group gestation pen, including the resting areas, the side of the pen with the ESF, and above the opposite side of the pen angled toward the ESF. Recordings were transmitted and stored on a digital video recorder (Lorex D441A6B-Z DVR, Lorex Technology Inc., Markham, Canada). After video collection, all video recordings were saved to an external hard drive, merged together, and converted to MP4 format (VideoProc, Digiarty Software Inc., Chengdu, Sichuan Province, China) for behavioral coding using a commercially available coding software (The Observer XT 15; Noldus Information Technology, Wageningen, Netherlands).

Social hierarchy determination

Twelve hours of video recordings (experimental days 1 to 2; beginning as sows entered group gestation housing) were used to determine the social hierarchy within groups. This length of time was chosen based on a preliminary observational study, where pen social hierarchy structure was largely unchanged among 12, 24, and 36 h. A single observer collected all behavioral data using a continuous recording rule to document the results of agonistic, or combative, interactions between pen mates during the 12-h observation period. These data were used to calculate the relative dominance index (DI; see description below) of each sow to determine the social hierarchy within groups.

An agonistic interaction was defined as a fight or displacement with physical contact initiated by one individual that included aggressive behavior followed by any submissive behavior performed by either individual involved in the encounter (Langbein and Puppe, 2004). A previously developed ethogram of social interactions between group housed sows (Jensen, 1980) was used to categorize aggressive and submissive behaviors performed during agonistic interactions. Parallel pressing, inverse parallel pressing, head-to-head knock, head-to-body knock, levering, biting, and physical displacement were considered aggressive behaviors. Retreating during a fight, turning away from an attack, attempting to flee, or displacement from a location were considered submissive behaviors (Jensen, 1980). During an agonistic interaction, the animal that performed a submissive behavior was considered to be “defeated” while the other individual was considered the “winner” of the interaction. In order to calculate the relative social rank of each sow within each group, a DI was calculated using the number of wins and defeats for each sow (Stukenborg et al., 2011). To calculate the DI, the following equation was used:

$\mathrm{D}\mathrm{I}=\frac{(\mathrm{w}\mathrm{i}\mathrm{n}\mathrm{s}-\mathrm{d}\mathrm{e}\mathrm{f}\mathrm{e}\mathrm{a}\mathrm{t}\mathrm{s})}{(\mathrm{w}\mathrm{i}\mathrm{n}\mathrm{s}+\mathrm{d}\mathrm{e}\mathrm{f}\mathrm{e}\mathrm{a}\mathrm{t}\mathrm{s})}$

Dominance index values were used to rank sows from the highest (most dominant) to lowest (least dominant) with a DI closest to 1.00 indicating the most dominant sow and a DI closest to −1.00 indicating the most submissive or lowest ranking sow within the pen. Individuals within groups were then categorized into four rank quartiles (RQ1–4; Table 1) based on DI, where the first quartile (RQ1) represented the highest-ranked sows with the greatest DI and the fourth RQ (RQ4) represented the lowest-ranked sows with the lowest DI. Rank quartiles were utilized to account for unequal group sizes during each repetition and the DI range’s dependence on individual interactions within each group (i.e., a DI value of 0.5 could represent the highest-ranking animal in one group and the lowest-ranking animal in another, depending on the group dynamics). Additionally, given the possibility of minor hierarchical shifts between individual animals at certain points throughout the experiment following the social hierarchy determination period, grouping sows within RQs by similar DI may help minimize the effects of individual sow hierarchy changes on the study results. These RQ served as the “Gold Standard” to which precision technologies results were compared.

Table 1.

Number of sows allocated to each rank quartile (RQ) per replication (1–5) of the study based on social rank determined by dominance index value (DI). Rank quartile 1 sows ranked highest in the social hierarchy while sows within RQ4 were the lowest ranked sows. Rank quartile 2 and RQ3 sows were intermediately ranked

Rank Quartile	Repetition 1	Repetition 2	Repetition 3	Repetition 4	Repetition 5
RQ1	3 (+1.00)1	3 (+1.00)	4 (+1.00)	4 (+0.73)	4 (+0.92)
RQ2	4	3	4	4	4
RQ3	4	4	4	4	5
RQ4	3 (−0.57)2	2 (−0.19)	3 (−0.67)	3 (−0.67)	4 (−0.54)
Total	14	12	15	15	17

¹Values within parentheses indicate the DI for the highest-ranking sow in RQ1.

²Values within parentheses indicate the DI for the lowest-ranking sow in RQ4.

Infrared thermal imaging

Infrared thermal images were captured on day 3 of the experimental period and every 15 d after reintroduction to group gestation housing using an infrared thermal imaging camera (FLIR E8, FLIR Systems LLC, Wilsonville, OR). Thermal images were captured behind the neck at the base of the left ear between 0700 and 1000 h when sows were awake and standing. The base of the ear is an anatomical location previously reported to be reliable for observing body surface temperature changes in swine (Rocha et al., 2019; Farrar et al., 2020) and therefore, was considered the region of interest (ROI) in this study. Measurement parameters, such as emissivity and distance from subject, were set and remained constant throughout the experiment. Emissivity was set to 0.98 (Soerensen et al., 2014) and the distance from subject being captured remained at 1 m. Atmospheric temperature and relative humidity within the gestation pens were collected by data loggers (HOBO Temperature/Relative Humidity Data Logger, Onset Computer Corporation, Bourne, MA) and accounted for within the IRT camera setting prior to each measurement. Infrared thermal imaging camera parameter settings related to temperature and humidity were adjusted accordingly throughout the experiment. After IRT collection, images were downloaded and analyzed using an infrared imaging processing software (FLIR Thermal Studio FLIR Systems LLC, Wilsonville, OR). An ellipse was fit to the ROI to identify the maximum (MaxT), mean (MeanT), and minimum (MinT) temperature for each thermal image taken on days 3, 15, 30, 45, 60, 75, 90, and 105 of the experiment.

Feed data

Sow identification ear tags equipped with RFID transponders were used to identify each sow using the ESFS (Big Dutchman, Holland, MI, USA) within group gestation. The ESFS dispensed feed until sows exited the ESF or until the daily allotted ration was dispensed. Various feed data were collected automatically by the ESFS. The time at entrance and exit of the ESF was recorded. Additionally, the amount of feed dispensed during a visit was recorded, which allowed the software program to calculate total feed offered for the day and the remaining feed in each sow’s individual daily allotment. The amount of feed distributed is based on a calibration of how much feed is dispensed during one revolution of the auger. Therefore, only feed offered and not feed consumed can be evaluated. Daily feed allotments occurred between 0000 and 2359 h. Feeding behavior was documented beginning on day 1 of the experimental procedure until approximately day 105 of gestation. Feed data were exported from the ESFS and used to calculate various feeding behavior traits (Table 2). Due to technical difficulties with the ESFS during the fourth and fifth repetitions of the study, portions of data were removed from the data set due to failure to dispense feed (14 d worth of data were removed for repetitions 4 and 5). Additionally, the software program for the ESFS failed to save the first 40 d of data for the first repetition.

Table 2.

Feeding behavior trait definitions calculated for each sow

Feeding behavior trait	Definition
Feed offered	The amount of feed (kg) offered to a sow within the electronic sow feeder (ESF).
Occupation time	The amount of time (s) a sow remained within the ESF continuously.
Number of visits	The number of times (count) a sow entered the ESF.
Number of meals	The number of consecutive visits (count) the sow made to the ESF with less than 300 s elapsing between visits.
With feed	Refers to visits or meals in which a sow remained within the ESF and feed was dispensed into feed trough.
Without feed	Refers to visits or meals in which a sow remained within the ESF but feed was not dispensed into feed trough. This occurred when a sow entered the ESF but had previously consumed its daily feed allotment.

Feeding behavior traits

Feeding behavior traits were calculated on a daily basis (per-day feeding behavior traits) using data generated from the ESFS and are defined in Table 2. Due to the ESFS recording any time the sow’s right ear moved away from the RFID antenna as separate visits, even if she did not leave the ESF (i.e., eating from the right side of the trough instead of the left), feed data from a nonproject sow was plotted to look for an obvious break in times between visits and meals (Tolkamp et al., 2011). As a result, ESFS visits were combined and defined as one visit when there was less than a 55-s break between the end time of the first visit and start time of the next visit, while a 300-s break was used to define one meal. For each sow, combined feeding events were then used to construct feeding behavior traits per visit, meal, and day (Table 2). Because sows could enter the ESF even after they consumed their daily allotment, feeding behavior traits were also evaluated for visits per meals with feed offered and visits per meals with no feed offered. Daily traits (i.e., occupation time per day) with feed were calculated from only those visits with feed offered. Additionally, feeding behavior traits were also calculated per hour over a 24-h period (hourly feeding behavior traits), with hour 0 representing 0000 to 0059 h and hour 24 representing 2300 to 2359 h.

Heart rate variability

One week prior to relocation to group gestation housing, all sows in farrowing stalls were acclimated to wearing a telemetric HR monitor (Polar H10 Heart Rate Sensor, Polar Electro, Kempele, Finland) for 30 min/d for 3 d while remaining in individual farrowing stalls. Heart rate monitor straps were placed around the sows’ chest, behind the front legs, with electrodes over the left side of the chest, and the monitor placed behind the elbow. Electrode gel (Spectra 360 Electrode Gel, Parker Laboratories Inc., Fairfield, NJ) was applied to the HR monitor straps prior to strap placement to better facilitate contact between skin and electrodes during data collections.

After the acclimation period, 10 randomly selected sows within each farrowing group (n = 50; 10 per replicate) were chosen for HRV data collection. One day before sows were moved from farrowing to gestation housing (day 0), HR straps and monitors were fit around the chest of each sow. Flexible bandage wrap (VetWrap; 3M, Maplewood, MN) was used to secure the HR straps in place to prevent movement of electrodes. Sixty minutes of instantaneous, continuous HR data from each selected sow were collected (baseline data) and transmitted via Bluetooth to an individual iPod Touch (Apple Inc., Cupertino, CA) using the Elite HRV application (Elite HRV, Asheville, North Carolina). All baseline HRV measurement periods occurred between 0900 and 1130 h. The following day (day 1), the HRV data collection procedure was repeated for each sow as they were reintroduced to gestation penning (post-mixing). All post-mixing HRV measurement periods occurred between 0930 and 1730 h. After 240 min of HR data collection within group gestation housing, the HR monitors and straps were removed.

Collected HRV data were exported, reviewed, and errors were edited manually using previously determined methods for artifact correction (Marchant-Forde et al., 2004). No data set used for analysis had more than 5% corrected errors. Data with more than five consecutive heartbeat interval errors were not used. Three hundred seconds of continuous HR data collected during periods of sow inactivity on days 0 and 1 were used to calculate linear HRV measures for each sow using a freely available HRV software package (Kubios HRV Standard, Kubios Oy, Kuopio, Finland). The following HRV measures, and their definitions, collected for this experiment were: 1) average heart beat interval length (RR, ms), defined as the average interval between adjacent heat beats over a period of time, 2) the standard deviation of RR intervals (SDNN; ms) defined as the standard deviation of all RR intervals over a period of time, and 3) the root mean square of successive differences (RMSSD; ms) defined as the root mean square of successive RR intervals over a period of time (Shaffer and Ginsberg, 2017). Due to loss of data during the baseline HRV measurement period, data sets from 7, 7, 9, and 8 sows were analyzed for RQs 1, 2, 3, and 4, respectively. During the post-mixing period, data sets from 13, 10, 16, and 10 sows were analyzed for RQs 1, 2, 3, and 4, respectively.

Statistical analysis

Data were analyzed in SAS (v. 9.4; SAS Institute, Inc., Cary, NC). A P-value less than 0.05 was considered significant for all models.

Thermal imaging characteristics (MaxT, MeanT, and MinT) were analyzed using the MIXED procedure. Backfat depth was fit as a covariate and repetition as a random effect. Fixed effects included were day, RQ, and their interaction. A repeated measures statement with sow nested with repetition was fitted to account for the multiple day measurements. Different covariance structures were evaluated and the covariate structure resulting in the lowest AICC score was selected for use.

Feeding behavior traits were analyzed using the MIXED procedure. For per-day feeding behavior traits, age was included as a covariate, RQ as a fixed effect, and repetition as a random effect. Since some individual sows were included in up to two study repetitions, sow was also fit as a random effect. A repeated measures statement was not used since not all sows were included in more than one repetition. As a result, the inclusion of a repeated measures statement caused issues with model convergence. For hourly feeding behavior traits, the same model was used with hour and the interaction between hour and RQ included as fixed effects. The least squares means statement with the DIFF option was used to determine differences between RQ and the interactions between time (i.e., day or hour) and RQ for all feeding behavior traits.

Heart rate variability data were analyzed using the MIXED procedure. For both baseline and post-mixing HRV measurement, RQ was included as a fixed effect, age as a covariate, and sow and repetition as random effects. Each sow’s baseline HRV measurement was also used as a covariate in the post-mixing HRV models.

Results

Thermal imaging

No RQ (P > 0.24; Table 3) or RQ by day interaction (P > 0.91; data not shown) effects were found for MeanT or MinT. A day effect was found for all thermal imaging characteristics (P < 0.0001), with temperatures decreasing throughout the experimental period (data not shown). Backfat depth as a covariate was significant for MinT (P = 0.04) but had no effect on MeanT or MaxT (P = 0.18 and 0.51, respectively).

Table 3.

Least squares means of the mean (MeanT), maximum (MaxT), and minimum (MinT) thermal image temperatures by rank quartile (RQ¹)

Trait	RQ1	RQ2	RQ3	RQ4	SEM2	P-value
MaxT, °C	34.4	34.7	34.7	34.7	0.6	0.25
MeanT, °C	33.2	33.4	33.4	33.4	0.7	0.61
MinT, °C	31.4	31.5	31.7	31.6	0.7	0.41

¹Rank quartile 1 sows ranked highest in the social hierarchy while sows within RQ4 were the lowest ranked sows. Rank quartile 2 and RQ3 sows were intermediately ranked.

²Largest standard error of the mean by row.

Feeding behavior traits

Per-day feeding behavior traits

Per-day feeding behavior traits are presented in Table 4. Rank quartile affected occupation time per visit (P = 0.045). Rank quartile 4 sows had less feed offered than RQ2 sows per visit (P = 0.047) and per visit with feed (P = 0.02), with RQ1 and RQ3 sows being intermediate to but not different from RQ2 and RQ4 sows for both traits (P > 0.05). This is supported by the occupation time per visit and per visit with feed, as amount of feed offered is dependent on time in the feeder. Rank quartile 4 sows spent less time in the feeder than RQ1 or RQ2 sows per visit (P = 0.03 and 0.01, respectively) and per visit with feed (P = 0.02 and 0.001, respectively). Although there were no differences in occupation time per day between RQ (P > 0.10), RQ4 sows had a lower occupation time per day than RQ2 sows when evaluating only visits with feed offered (P = 0.01). There was also an effect of RQ on the number of visits per day and the number of visits per day with feed (P = 0.01 and P = 0.02, respectively). For both traits, higher-ranked sows (RQ1 and RQ2) had fewer visits to the feeder than RQ4 sows (P < 0.04). Rank quartile 3 sows had fewer visits than RQ1 and RQ2 sows per day (P = 0.003 and 0.03, respectively) but not per day with feed (P = 0.18 and 0.11, respectively). No other differences were found between RQ.

Table 4.

Least squares means of per-day feeding behavior traits by rank quartile (RQ¹)

Trait	Time	Feed2	RQ1	RQ2	RQ3	RQ4	SEM3	P-value
Feed offered, kg	Visit	All	0.71ab	0.71b	0.56ab	0.51a	0.20	0.20
		With	0.96	0.87	0.77	0.84	0.13	0.47
	Meal	All	1.11ab	1.16b	0.94ab	0.86a	0.09	0.08
		With	1.38	1.40	1.30	1.31	0.07	0.70
Occupation time of feeder, s	Visit	All	282bc	296c	239ab	212a	31	0.045
		With	362bc	394c	303ab	255a	31	0.01
		Without	170	183	160	181	26	0.81
	Meal	All	464	437	396	434	37	0.48
		With	587	604	538	520	51	0.43
		Without	195	222	181	233	35	0.50
	Day	All	909	1073	1155	1088	143	0.33
		With	605ab	650b	612ab	551a	31	0.04
		Without	487	592	785	715	157	0.57
Visits to the feeder, n	Day	All	4.48a	4.67a	7.00b	8.05b	1.08	0.01
		With	2.74a	2.61a	4.14ab	5.78b	0.82	0.02
		Without	2.94	3.17	4.21	3.32	0.65	0.37
Meals4, n	Day	All	3.20	3.39	4.85	4.71	0.84	0.19
		With	1.71	1.69	2.30	2.63	0.43	0.29
		Without	2.61	2.82	3.72	3.06	0.57	0.34

¹Rank quartile 1 sows ranked highest in the social hierarchy while sows within RQ4 were the lowest ranked sows. Rank quartile 2 and RQ3 sows were intermediately ranked.

²All = includes all data; With = includes only data from visits with feed offered; Without = includes only data from visits without feed offered.

³Largest standard error of the mean in each row.

⁴Meals are groups of visits that occurred with less than a 300-s break between consecutive visits.

^a,b,cDifferent superscripts indicate pairwise differences (P < 0.05) between RQ for each feeding behavior trait.

Hourly feeding behavior traits

The interaction between RQ and hour affected feed offered per hour (P = 0.003; Figure 1A). Hours with differences between RQ are shown in Figure 1B. During hour 0, RQ1 and RQ2 sows had more feed offered to them than RQ3 and RQ4 sows (P < 0.003), with RQ3 sows having more feed offered to them than RQ4 sows (P = 0.003). During hour 1, RQ1 sows had more feed offered to them than RQ3 and RQ4 sows (P < 0.02), while RQ2 sows had more feed offered to them than RQ4 sows (P = 0.02), with no other differences between RQ comparisons (P > 0.16). Sows in RQ1 had more feed offered to them than RQ4 sows during hour 2 (P = 0.049), while the opposite was true during hour 8 (P = 0.04). No other differences between RQ were seen for feed offered per hour (P > 0.05).

Figure 1.

Rank quartile by hour interaction for feed offered. Feed offered per hour (kg) over a 24 h period (A) between rank quartiles (RQ). Hour 0 represents 0000 to 0059 h. Rank quartile 1 sows ranked highest in the social hierarchy while sows within RQ4 were the lowest ranked sows. Rank quartile 2 and RQ3 sows were intermediately ranked. A RQ by hour interaction was observed (P = 0.003), where differences between RQ were detected during hour 0, 1, 2, and 8 (B) Different superscripts in (B) indicate differences (P < 0.05) between RQ for each hour.

Occupation time per hour and occupation time per hour with feed (Figure 2A and C, respectively) were similarly influenced by the interaction of RQ and hour (P = 0.01 and 0.0007, respectively). Differences for occupation time per hour were found for hour 0, 1, 6, 7, 8, and 12 (Figure 2B). Higher-ranking sows (RQ1 and RQ2) exhibited greater occupation time per hour compared to lower-ranking sows (RQ3 and RQ4) during hour 0 (P < 0.04). Rank quartile 3 sows also occupied the ESF for longer compared to RQ4 sows during hour 0 (P = 0.004). Rank quartile 1 sows continued to occupy the ESF for longer periods during hour 1 compared to RQ4 (P = 0.01). During hour 6, RQ3 sows occupied the ESF for longer periods compared to RQ1 and RQ4 sows (P = 0.005 and 0.045, respectively). Rank quartile 3 sows continued to occupy the ESF for longer periods of time compared to RQ1 during hour 7, 8, and 12 (P < 0.05). Differences for occupation time per hour during visits with feed offered were found for hour 0, 1, and 2 (Figure 2D). Higher-ranking sows (RQ1 and RQ2) also exhibited greater occupation time per hour with feed compared to lower ranking sows (RQ3 and RQ4) during hour 0 (P < 0.002). During hour 1, RQ1 sows spent more time occupying the ESF per hour with feed compared to RQ3 and RQ4 sows (P < 0.0001 and P = 0.002, respectively). Rank quartile 2 sows spent a greater amount of time occupying the ESF per hour with feed compared to RQ4 sows during hour 1 (P = 0.007). Finally, RQ1 sows spent a greater amount of time occupying the ESF per hour with feed compared to RQ3 and RQ4 (P = 0.04 and 0.03, respectively) sows during hour 2.

Figure 2.

Rank quartile by hour interaction for occupation time of feeder. Electronic sow feeder occupation time (s) by hour (A) and by hour when feed was offered (C) over a 24 h period organized by rank quartile (RQ). Hour 0 represents 0000 to 0059 h. Rank quartile 1 sows ranked highest in the social hierarchy while sows within RQ4 were the lowest ranked sows. Rank quartile 2 and RQ3 sows were intermediately ranked. Rank quartile by hour interactions were observed for occupation time per hour (P = 0.01; B) with differences between RQ at hour 0, 1, 6, 7, 8, and 12 and occupation time per hour with feed offered (P = 0.0007; D) with differences between RQ at hour 0, 1, and 2. Different superscripts in (B) and (D) indicate differences (P < 0.05) between RQ for each hour.

The differences observed in overall ESF occupation time per hour during hour 6, 7, 8, and 12 are largely driven by differences between RQ when they occupied the ESF without feed offered. Rank quartile 3 sows spent greater time occupying the ESF without feed than RQ 1 sows during hour 6 (28. 3 vs. 13.6 s ± SEM 8.4 s; P = 0.03), hour 7 (28. 2 vs. 12.2 ± SEM 8.4 s; P = 0.02), hour 8 (28. 1 vs. 13.9 ± SEM 8.4 s; P = 0.04), hour 9 (28. 1 vs. 14.5 ± SEM 8.4 s; P = 0.046), hour 11 (28. 2 vs. 12.5 ± SEM 8.4 s; P = 0.02), and hour 12 (25.2 vs. 6.7 ± SEM 8.4 s; P = 0.01). Rank quartile 3 sows occupied the ESF without feed for greater time compared to RQ4 sows during hour 6 (28.3 ± 8.4 vs. 14.5 ± 8.7 s; P = 0.045) and RQ2 sows during hour 12 (25.2 ± 8.4 vs. 11.1 ± 8.3 s; P = 0.03). Finally, RQ4 sows spent greater time occupying the ESF without feed than RQ2 sows during hour 20 (18.7 ± 8.7 vs. 5.0 ± 8.3 s; P = 0.048).

The main effect of rank quartile for number of visits and meals per hour, per hour with feed, and per hour without feed are shown in Table 5. Higher-ranking sows (RQ1 and RQ2) had fewer visits to the feeder per hour than lower-ranking sows (RQ3 and RQ4; P < 0.002). This was driven primarily by the first 8 h of the day (hour 0–7) as no differences were seen for hour 8 or later (P > 0.15, data not shown). For number of visits per hour with feed offered, RQ2 sows had fewer visits than lower-ranking sows (RQ3 and RQ4; P < 0.01) while RQ1 sows had fewer visits than RQ4 sows (P = 0.002) and tended to have fewer visits than RQ3 sows (P = 0.0504). Once again, differences between RQ for number of visits per hour with feed offered were driven by the first hours of the eating period (hour 0–8) with no differences between RQ seen for hour 9 or later (P > 0.13; data not shown). While RQ affected number of visits per hour without feed (P = 0.003), there was no clear trend between RQ across the day (data not shown). Similar to number of visits per hour, higher-ranking sows (RQ1 and RQ2) had fewer meals per hour than lower-ranking sows (RQ3 and RQ4; P < 0.0002). This was driven by meals during the first 9 h of the feeding period (hour 0–8) as no differences were seen for hour 9 or later (P > 0.13, data not shown). While higher-ranking sows had fewer meals per hour when feed was offered than lower-ranking sows (P < 0.05), there were only differences between RQ for hour 0, 1, 7, and 8. For hour 0, RQ1 sows had fewer meals with feed than RQ4 sows (P = 0.02), with no differences between any other RQ (P > 0.06). For hour 1, RQ1 and RQ2 sows had fewer meals with feed than RQ3 sows (P = 0.02 and 0.002, respectively), with no differences between other RQ (P > 0.06). Rank quartile 4 sows had more meals per hour with feed than higher-ranking sows (RQ1 and RQ2) for hour 7 (P = 0.04 for both) and for hour 8 (P = 0.02 for both).

Table 5.

Least squares means of hourly (per-hour) feeding behavior traits by rank quartile (RQ¹)

Trait	Feed2	RQ1	RQ2	RQ3	RQ4	SEM3	P-value
Visits to the feeder per hour, n	All	0.13a	0.12a	0.25b	0.25b	0.06	0.0004
	With	0.09ab	0.07a	0.14bc	0.18c	0.04	0.0009
	Without	0.05a	0.06ab	0.10c	0.08bc	0.03	0.003
Meals per hour4, n	All	0.08a	0.08a	0.18b	0.18b	0.04	<0.0001
	With	0.05a	0.04a	0.08b	0.10b	0.02	0.004
	Without	0.04a	0.05a	0.09b	0.07b	0.03	0.0005

¹Rank quartile 1 sows ranked highest in the social hierarchy while sows within RQ4 were the lowest ranked sows. Rank quartile 2 and RQ3 sows were intermediately ranked.

²All = includes all data; with = includes only data from visits with feed offered; without = includes only data from visits without feed offered.

³Largest standard error of the mean in each row.

⁴Meals are groups of visits that occurred with less than a 300-s break between consecutive visits.

^a,b,cDifferent superscripts indicate pairwise differences (P < 0.05) between RQ for each feeding behavior trait.

Heart rate variability

Baseline and post-mixing HRV traits are presented in Table 6. Rank quartile 2 sows exhibited greater RR compared to RQ3 sows (P = 0.009), while RQ1 and RQ4 sows were intermediate to- but not different from RQ2 and RQ3 (P > 0.05 for all remaining comparisons). Rank quartile was not associated with any HRV measures during the post-mixing period (P > 0.76).

Table 6.

Least squares means of heart rate variability measures prior to reintroduction to group gestation housing (baseline) and after reintroduction (post-mixing) organized by rank quartile (RQ¹)

Experimental period	Trait2	RQ1	RQ2	RQ3	RQ4	SEM3	P-value4
Baseline	RR, ms	573ab	612b	551a	583ab	22	0.04
	SDNN, ms	15.4	16.2	12.8	14.0	3.1	0.69
	RMSSD, ms	7.30	7.68	6.96	9.14	1.40	0.79
Post-mixing	RR, ms	548	568	534	523	33	0.77
	SDNN, ms	17.9	22.7	17.2	20.2	5.7	0.86
	RMSSD, ms	8.21	13.47	6.98	8.63	5.20	0.76

¹Rank quartile 1 sows ranked highest in the social hierarchy while sows within RQ4 were the lowest ranked sows. Rank quartile 2 and RQ3 sows were intermediately ranked.

²RR = the average interval between adjacent heart beats over a period of time; SDNN = the standard deviation of RR intervals over a period of time; RMSSD = the root mean square of successive RR intervals over a period of time.

³SEM = largest standard error of the mean by row.

⁴ P-value for main effect of rank quartile.

^a,bDifferent superscripts indicate differences (P < 0.05) between RQ for each trait.

Discussion

Precision livestock farming technology is capable of aiding livestock producers and researchers in detecting illness, injury, and poor welfare (Benjamin and Yik, 2019; Gómez et al., 2021). This study sought to investigate whether IRT, RFID, and HRV technologies could be used to identify group housed sow social hierarchies. The establishment of the social hierarchy is commonly observed in nature but could also be a welfare concern in group housing systems due to the occurrence of agonistic interactions and their negative impacts on sow welfare (McGlone, 1986). The social hierarchy determines sow ability to access limited resources and can leave sows at risk of poor welfare due to injury and stress (Salak-Johnson, 2017).

Infrared thermal imaging

In response to a stressor, the hypothalamic-pituitary axis is activated, and heat is generated due to increases in stress hormones (i.e., cortisol) and blood flow. The increased blood circulation near the skin results in increased heat radiated from the body’s surface (Stewart et al., 2005, 2007). Infrared thermography offers a non-invasive method for capturing infrared radiation (heat) emitted by animals, which is then converted into a computer-generated image to evaluate the changes that occur. This technology has been used to detect changes in physiological processes related to changes in body surface temperature, including illness and thermal stress (Brown-Brandl et al., 2013; Cook et al., 2015; Salles et al., 2016). Additionally, IRT has been used to evaluate painful procedures and social stress in livestock species (Stewart, 2008; Boileau et al., 2019; Travain and Valsecchi, 2021).

Internal body temperature has been shown to increase under stress (Balcombe et al., 2004) and it could be assumed that animals lower in the social hierarchy would be under more stress (Marchant-Forde, 2010; Martínez-Miró et al., 2016). As a result, lowly-ranked animals would display elevated body surface temperatures. However, this has not been a consistent finding. For example, Boileau et al. (2019) captured thermal images of swine dyads during agonistic interactions and found there were no differences in body surface temperature between winners and losers of the encounters. Interestingly, both animals involved in the dyadic contest exhibited a sharp decline in body temperature after the encounter, which indicates that a thermal response does take place during aggressive interactions.

In the present study, no differences were found between RQ on any days of the experiment other than day 105, when RQ2 had greater MaxT compared to RQ1. The lack of meaningful differences may be due to the timeline of IRT data collection, since images were not captured upon sow reintroduction to group gestation housing. There was, however, a day effect on IRT, where IRT mean temperatures decreased as the gestation period proceeded. This could be due to the decline in ambient temperature observed throughout the study and thermoregulatory processes which occur naturally. Additionally, this could also be attributed to animals becoming accustomed to one another over time and the decrease in stress which occurs after the social hierarchy is established. Future research should evaluate IRT immediately after reintroduction to group gestation housing. Moreover, different ROI should be investigated for use in precision livestock farming tool research related to sow welfare.

Feeding behavior traits

Electronic sow feeding systems were developed in the 1980s as an alternative feeding method to reduce aggression and competition over feed within groups of gestating sows (Remience et al., 2008). Although feeders within an ESFS can accommodate numerous sows based on the manufacturer recommendations, the lower number of feeding spaces can result in competition and aggression between pen mates similar to competitive feeding methods (Spoolder and Vermeer, 2015; Norring et al., 2019).

Since the social hierarchy is created in order to determine one’s access to limited resources (Meese and Ewbank, 1973; Stukenborg et al., 2011), competition to gain feeder space within an ESFS can be heightened and can increase aggression (Olsson et al., 2011; Jang et al., 2015). Sows of higher rank, or dominant sows, may have easier access to feed compared to lower ranking sows (Csermely and Wood-Gush, 1990), which can lead to body condition and weight variation within groups (Rasmussen et al., 1962).

Previous research has explored social hierarchy and its relation to feeding behavior using different feeding methods. In the current study, occupation time per ESF visit was greatest in RQ2, while RQ4 exhibited the least amount of time in the ESF. Rank quartile 1 and RQ3 sows were intermediate to RQ2 and RQ4 sows. This same pattern was also seen in occupation time per visit where sows received feed during the visit. These results indicate that higher ranking sows are occupying the feeder for longer periods of time, similar to previous research (Martin and Edwards, 1994; Salak-Johnson, 2017). It can also be inferred that these higher-ranking sows have priority access to feeding space compared to other sows due to their dominance and are able to enter the feeder before other sows, even when there are sows waiting to gain feeder access and have not consumed their entire daily allotment (Olsson et al., 2011; Brajon et al., 2021).

Rank quartile 1 and RQ2 sows made fewer ESF visits per day compared to RQ3 and RQ4 sows. Similarly, RQ1 and RQ2 sows made fewer ESF visits with feed per day compared to RQ4 sows. This provides some evidence that higher ranking sows may be consuming a majority of their ration in a smaller number of visits but continuing to occupy feeding space for longer periods of time even when their ration is consumed. In response, lower-ranking sows may have adjusted their feeding habits to times when there is little pressure to access the feeder (i.e., when higher-ranked pen mates are resting or sleeping throughout the day; Brajon et al., 2021). This has been reported in other studies utilizing both ESF and ad libitum feeding systems (Brouns and Edwards, 1994; Marchant-Forde, 2010), which indicates that lower-ranking sows are capable of adapting to their environment and location within the hierarchy when feeding resources are limited. Further evidence of their ability to adapt has been reported in other previous studies where sow feeding behavior has changed throughout gestation when ESF settings have been altered (Vargovic et al., 2021). However, since feeding behavior was not monitored via video or live observation, additional work is needed to confirm this interpretation of the present data.

The feeding behavior results in the current study demonstrate that lower ranking sows may be consuming portions of their rations over numerous visits throughout a larger portion of the day, as opposed to higher ranking sows, who consume most of their ration early in the day after the ESFS resets at 0000 h. Other studies have reported similar results, with high ranking sows feeding earlier in day (Chapinal et al., 2008). Additionally, an increase in total number of visits to the ESF has been reported for low-ranked growing pigs throughout the grow-finish phase (Hoy et al., 2012). One reason for the increased number of visits exhibited by low-ranked sows in the current study could be due to social pressure around the feeding spaces (Boumans et al., 2018). Although the feeders used within the current study were fully enclosed, sows within feeders were able to be seen by their conspecifics. As a result, the formation of a “feed queue” (Anil et al., 2006), or a line of sows surrounding the feeders, may create social pressure and stress for the sow occupying the ESF. In fact, although not quantified in the current study, queueing sows were also sometimes observed manipulating the gate that enclosed the feeding pig within the ESF. Research has found that pigs fed individually within an ESF may be able to gain access to feeder space daily; however, it is possible that some individuals may not consume their entire allotment due to intimidation from pigs waiting outside the feeder (Mendl et al., 1992).

Heart rate variability

Heart rate variability, or the variation in length between adjacent heart beat intervals over time, is a useful noninvasive measure for assessing physiological stress in livestock (von Borell et al., 2007). While mean HR is commonly used to determine the impacts of stressful stimuli on pigs and various livestock (De Jong et al., 2000; Mott et al., 2021), HRV improves upon mean HR by accounting for the variability that is present in instantaneous heart rate over time and can be used as a proxy measure of autonomic nervous system activity (Shaffer and Ginsberg, 2017). Whereas mean HR represents the net effect of parasympathetic and sympathetic activity on the heart, short term measurement (i.e., 5 min) of linear HRV measures, like resting SDNN and RMSSD, have been shown to reliably represent parasympathetic-mediated respiratory sinus arrhythmia (SDNN; Shaffer et al., 2014) or vagally mediated changes to heart rate signal variability (RMSSD; Shaffer et al., 2014). Accordingly, higher values of SDNN and RMSSD are indicative of greater parasympathetic activity (Kovács et al., 2014).

In the current study, HRV differences between RQ were only observed during the baseline period (while sows were still housed in their farrowing stalls). Rank quartile 2 sows exhibited greater RR values during the baseline period compared to RQ3 sows, whereas RQ1 and RQ 4 sows exhibited RR values intermediate to- but not different from RQ2 and RQ3 sows. A clear explanation for the relationship between resting RR during the baseline period and future social rank is not available based on our data. However, previous research indicates that baseline HRV may be capable of detecting stress susceptibility (Bachmann et al., 2003) or stress coping styles (Krause et al., 2017) in certain species, which are individualized behavioral and physiological responses to stressors that are consistent over time and can be characterized on a spectrum from proactive to reactive (Janczak et al., 2003). Proactive animals display a bolder and more dominant response to stressors, which is characterized by increased aggression, reduced flexibility to environmental changes, and greater sympathetic-adrenomedullary activity (Koolhaas et al., 2010). Reactive animals, on the other hand, respond in an opposite manner, where their response to a stressor is characterized by increased immobility, reduced aggression, and reduced sympathetic-adrenomedullary activity (Koolhaas et al., 2010). Indeed, pigs classified as having a proactive coping style exhibit greater HR and lower RMSSD than reactive pigs during rest (Krause et al., 2017).

The baseline HRV results reported here only partially support this assertion, since RQ1 sows, who might be more likely to have proactive coping styles based on their willingness to engage in aggressive interactions (Greenwood et al., 2017), did not differ in their baseline RR response compared to all other RQs. Rank quartile 2 sows did exhibit higher RR compared to RQ3, but not RQ4 sows. We would expect to see differences between high- (RQ 1 and RQ2) and low-ranked (RQ3 and RQ4) sows if baseline RR were capable of detecting differences in social rank. Previously published data indicates that lower ranking sows receive more aggressive head knocking from higher-ranked sows and instigate fewer aggressive head knocking interactions with their pen mates (Greenwood et al., 2017), which might be more indicative of a reactive coping style. Since we did not take coping style into account for the current study, it is not possible to determine whether coping style is related to the relationship between baseline HRV and future position within the social hierarchy. Future research should investigate this relationship.

The results presented here provide a meaningful basis for further investigation of precision technologies aimed at characterizing the sow social hierarchy. However, future studies on this topic should attempt to improve upon two drawbacks in the current study. Specifically, the degree of variation associated with sow age (12–36 mo) and parity (1–7) in this study is likely not common in commercial production. It is possible that this wide range of ages and parities may have affected the results differently compared to an experimental group with more limited ages and parities. Additionally, due to the smaller herd size that was utilized and limited space availability in our facility, first-parity gilts were mixed into already-existing gestation groups of less than 20 sows after the weaning period to replace cull sows. Future studies should attempt to maintain static gestation groups throughout the data collection period.

Conclusion

Precision livestock farming technology has shown great promise aiding animal caretakers in the monitoring of animal health and welfare. Due to welfare concerns which arise in group housed sows from the establishment of a social hierarchy, the objective of this study was to investigate the use of IRT characteristics, feeding behavior traits obtained from an ESFS, and HRV for detecting the social hierarchy of group housed sows. Sows of higher rank occupied the ESF for longer periods of time while the lowest ranked sows (RQ4) occupied the ESF for the least amount of time but had a greater number of visits overall. These lowest ranking sows also consumed their ration in a greater number of visits during the feeding period. Rank quartile was associated with baseline RR, but did not clearly distinguish between the highest and lowest ranked sows. Therefore, ESF feeding data and HRV technologies should be explored further as potential PLF technologies for determining group housed sow social hierarchies. Additionally, research should also investigate how these technologies could aid animal caretakers in mitigating negative welfare outcomes due to sow social rank.

Glossary

Abbreviations

DI

dominance index

ESF

electronic sow feeder (race and trough)

ESFS

electronic sow feeding system (feeder, hopper with feed auger, feed lines, software)

HR

heart rate

HRV

heart rate variability

IRT

infrared thermography

MaxT

maximum temperature of thermal image region of interest (°C)

MeanT

mean temperature of thermal image region of interest (°C)

MinT

minimum temperature of thermal image region of interest (°C)

RFID

radio frequency identification

RMSSD

root mean square of successive differences (ms)

ROI

region of interest

RQ

rank quartile

RR

average interval between successive heart beats (ms)

SDNN

standard deviation of intervals between successive heart beats (ms)

Conflict of Interest Statement

The authors declare no conflicts of interest, financial or otherwise.