Effects of personality and rearing-history on the welfare of captive Asiatic lions (Panthera leo persica)

Research permit and ethical considerations

Research permit for this study was granted by the Gujarat Forest Department, India (Permit no: WLP/28/A/1316-21/2015-16). This study complies with the regulations of zoo animal welfare standards set by the Central Zoo Authority, Government of India.

Study area

We conducted the study at Sakkarbaug Zoological Garden (SZG), which is situated within the natural range of Asiatic lions (Panthera leo persica). SZG is the coordinating zoo for the Asiatic lion conservation-breeding programme in India and hosts the largest captive population (N = 60) with the highest reported number of wild founders (Srivastav, 2014) The conservation breeding programme aims to stock a healthy population of captive Asiatic lions for possible repatriation to lost range habitats. The zoo has a separate off-display conservation breeding facility, which houses 47 Asiatic lions. A map (unscaled) of the off-display conservation breeding enclosures of SZG is provided in Fig. S1.

Subjects and housing

We collected data from 38 (Male = 15, Female = 23) healthy Asiatic lions housed in the conservation breeding facility of SZG. During the study we removed three individuals (Male = 1, Female = 2) due to ongoing veterinary treatments, reducing our sample size to 35 individuals (Male = 14, Female = 21) (Table 1). The study subjects were either born in captivity (N = 16; Male = 3, Female = 13) or rescued from wild (N = 19; Male = 11, Female = 8). Individuals born in the zoo (N = 14) and rescued as cubs (N = 2) were categorized as captive-raised since they have similar life experiences. Lions that were rescued at a young age and spent most of their lives in captivity cannot have similar life-experiences as adult wild-rescued lions and hence were grouped in the captive-raised category. Most wild-rescued lions were rehabilitated as adults for treatment of injuries incurred due to infighting, and after making full recovery were assimilated in the conservation breeding programme. Some wild-rescued animals (N = 3) were rescued to ameliorate conflict caused by livestock depredation. All subjects were either housed in pairs (N = 17) or housed in a sex ratio of 1:2 (N = 18). All subjects (including the wild-rescued lions) were in socially cohesive groups and were housed in the same enclosure (with the same enclosure mates) for at least a year prior to the commencement of the study. This facility provided us with a unique opportunity to study the behaviour of wild-rescued and captive-raised lions under similar housing conditions.

Subjects were housed in 15 naturalistic enclosures spread across 8-ha area resembling the habitat of Asiatic lions. All enclosures were similar in design, devoid of enrichment devices, evenly populated with leafy trees (for shade and cover) and provided similar enclosure space per animal (400 m²), ensuring uniformity of housing conditions for all subjects. Due to the absence of complexity and an active enrichment intervention programme, all enclosures were deemed functionally barren to the subjects. The enclosure sizes ranged from 1,100–6,542 m², with an average size of (M = 1,970 m², SD = 1,685.24m²). Only one enclosure was 6,542 m² in size, and most other enclosures were similar in sizes (M = 1,424 m², SD = 224 m²). All enclosures included outdoor (paddocks) and indoor (retiring/feeding cells) areas (3 m × 3 m × 2 m dimensions) with continuous access to drinking water. Enclosure barriers consisted of v-shaped dry moats with walls at the proximal side and chain-linked fences with dual overhangs (4 m high) on the other three sides. Adjacent enclosures were separated by visual barriers in the form of dense bamboo thickets. Subjects were confined to feeding cubicles only during feeding time and had free access to all enclosure areas (including feeding cubicles) for the rest of the day. Subjects were fed separately at the indoor cubicles between 1,700–1,900 h six days a week, with a fast on Sundays. Subjects were fed in-house slaughtered and quality-inspected buffalo meat. The average meat consumption was 3.5 kg (SD = 0.5 kg) for females (N = 21) and 4.9 kg (SD = 1.4 kg) for males (N = 14). Most subjects were group-housed (1:2) (N = 18) or pair-housed (1:1) (N = 20). Four subjects were iso-sexually paired which included two male lions (2:0) and a mother-daughter dyad (0:2).

A group of animal keepers carried out all husbandry work for the subjects on a rotational basis, which meant that all subjects were accustomed to the same group of keepers. Since the conservation breeding area is off-display and restricts access to unauthorized personnel, subjects’ interactions with humans were limited to keeper interactions. The animal-keepers had trained the subjects to respond to their house names and vocal instructions for moving in and out of the feeding cubicles.

Study design

We aimed to answer two broad research questions in this study; (a) how differences in bold/shy personality traits, rearing history, sex, and social grouping are associated with variations in behavioural welfare outcomes in a group of captive Asiatic lions? (b) How are the behavioural welfare indices (viz., enclosure usage, behaviour diversity, and aberrant repetitive behaviours) interlinked?

To answer the above questions, we categorized subjects and recorded outcomes of welfare measures. The detailed design for the study is given below.

Personality assessment

We adopted a combination of keeper-rating and behaviour-coding techniques to reliably assess personality traits (Funder & Colvin, 1988; Funder, 1995; Gosling & Vazire, 2002; Highfill et al., 2010; Gartner & Powell, 2011). We separately interviewed three animal keepers with at least ten years of work experience to rate 38 subjects (15:23) on a scale of 1–9 (1-very low and 9- very high) for pre-selected bold (N = 10) and shy traits (N = 10) (Table S1). We found that all keepers agreed on their ratings for subjects (high inter-rater reliability, Cronbach’s alpha > 0.8). First, we averaged all keeper ratings (N = 3) subject-wise for all personality traits. Next, we calculated the average rating on bold (N = 10) and shy (N = 10) traits for each subject. Subjects that received an average score above seven on bold traits were categorized as bold, whereas an average rating above seven on shy traits were categorized as shy.

To validate keeper ratings for personality traits of subjects (bold/shy), we conducted novel-object tests in day kraals for ten minutes using video recorders in the absence of keepers and observers (Highfill et al., 2010). Naive observers (N = 3) with no prior exposure to study subjects, recorded the latency of subjects to interact with novel objects (Sih, Bell & Johnson, 2004; Frost et al., 2007) and calculated the percentage of bold vs shy behaviours (Powell & Svoke, 2008; Gartner & Powell, 2011; Corsetti et al., 2018) performed by the subjects during these tests (Tables S2 & S3). During these tests, the subjects were exposed to (a) unknown conspecifics, (b) unknown person and (c) non-food novel objects (lion-sam ball and bungee cord). All novel-object tests were conducted in an open-air day kraal adjacent to the paddock area of the enclosures. For the first test, we simultaneously released two subjects (same sex but unknown to one another) at adjacent day kraals and recorded their reactions to encountering a same-sex unknown conspecific. The latency counter was started as soon as both lions were released inside their respective day kraals. For the second test, we released the subject inside the day kraal and had a volunteer (unknown to the lion and not wearing a keeper’s uniform) approach the kraal, and stop at the median section facing the day kraal for ten minutes. The volunteer did not make any eye contact or vocal communication with the animal. The latency counter was started as soon as the volunteer reached the day kraal. For the final test, we placed a novel object (lion-sam ball or bungee cord) at the centre of the enclosure, and then released the subject inside the day kraal. The latency counter was started when the subject was released inside the day kraal. Observers used focal animal sampling (Altmann, 1974; Martin & Bateson, 1993) to calculate duration of all behavioural states and events performed by subjects during each of these tests. These focal observations were used to calculate the percentage of bold and shy behaviour performed by subjects (Table S3). We tested each subject separately to avoid confounding personality with dominance. We conducted the latency tests simultaneously for 12 individuals daily between 0900-1100 h. Since we did not want to overwhelm the animals with multiple novel stimuli on a single day, three sessions of novel-object tests were conducted for each subject on consecutive days. The novel object tests were conducted for ten minutes, and if subjects failed to approach the novel object after five minutes, they were categorized as shy. We repeated the novel object tests with unknown human and novel object after a month to check for trait consistency and calculated the average latency values for each subject. In the first session, lion-sam ball was used as the novel object which was replaced in the second session with a hanging bungee cord. The order of the latency tests was kept the same for all subjects. Three Asiatic lions undergoing veterinary treatments for physical injuries (Male = 1, Female = 2) showed inconsistencies in trait measures across different sessions. We excluded these animals from the study, thus reducing the number of subjects to 35 individuals (Male = 14, Female = 21). We found that keepers (N = 3) and observers (N = 3) reliably agreed on the personality type (Cronbach’s alpha > 0.9) of these 35 subjects. These subjects were further categorized based on rearing history (wild-caught = 19, captive-raised = 16), sex (Male = 14, Female = 21), and social grouping (pair-housed = 17, group-housed = 18).

Behaviour data collection

For behaviour data collection, we used pre-existing ethograms for felids (Powell, 1995; Stanton, Sullivan & Fazio, 2015) and modified them to include unique behaviours displayed by subjects (Table 2). Two independent observers collected all behaviour data. To minimize inter-observer bias, behaviour recording was commenced after inter-observer reliability reached satisfactory levels from the same group of animals (Cronbach’s α > 0.9) (Caro et al., 1979; Gliem & Gliem, 2003). We recorded ten hour-long behaviour observation sessions (for four subjects) and found one-minute instantaneous scans (Altmann, 1974; Martin & Bateson, 1993; Amato, Van Belle & Wilkinson, 2013) were comparable to focal animal behaviour observation data (Altmann, 1974; Gilby, Pokempner & Wrangham, 2010; Amato, Van Belle & Wilkinson, 2013) in recording behavioural states and events for multiple subjects. We chose instantaneous scans as it provided a good balance between data-accuracy and observer fatigue. We recorded behaviour at three different time periods: 0500-1100 h, 1300–1800 h and 2200-0500 h in six-hour blocks. During each six-hour block, we conducted four one-hour sessions of instantaneous scans at one-minute intervals for one-hour duration followed by a 15-minute rest. During each scan, we recorded the behavioural state of the subject and its location in the enclosure. All occurrences of behaviour events were recorded separately. We used the frequencies of behavioural states and events to measure behaviour diversity of each subject during an observation session (one hour). We measured the directionality of all social interactions between subjects to gain a better understanding of the social cohesiveness of each enclosure group. We also video recorded behaviour observation sessions, which were used to fill any potential gaps in observer recording of instantaneous scans. We gathered a total of 2,009 h of behaviour observation data (average data of 57 h/subject) across 486 observation days. We collected information on the following behavioural welfare indices:

Enclosure usage

Enclosure use is a critical behavioural parameter that is influenced by the biological relevance of different zones of the captive environment (Traylor-Holzer & Fritz, 1985; Plowman, 2003; Rose & Robert, 2013; Ross & Shender, 2016). Homogenous usage is indicative of a complex and novel enclosure design (Ross & Shender, 2016; Mallapur, Qureshi & Chellam, 2002; Rose & Robert, 2013) which are considered more important drivers of welfare than enclosure area (Traylor-Holzer & Fritz, 1985). We divided each enclosure into ten equal zones, which included three broad zones viz. (a) proximal, (b) medial, and (c) distal zones. Each of these broad zones was further subdivided into three smaller zones as (i) left, (ii) middle, and (iii) right. The tenth zone was the paddock area next to the retiring cell (Fig. 1). We recorded the enclosure zone location of subjects during each scan. We calculated the spread of participation index (SPI) (Plowman, 2003) of enclosure usage for all 35 subjects across 486 observation days using instantaneous scan data. We calculated the SPI values using the following formula:

SPI= $\frac{\sum |fo-fe|}{2\left(N-femin\right)}$

where f_o stands for the observed frequency of usage of enclosure zones, f_e stands for the expected frequency of enclosure usage. N stands for gross observations in all zones of the enclosure and f_emin stands for the expected frequency of observation for the smallest zone (Plowman, 2003). SPI measures indicate the homogeneity of space usage. A high SPI value (close to 1) indicates that subjects are biased towards certain areas of the enclosure, while a lower SPI value (close to 0.5 or lower) indicates that lions use most areas of the enclosure equitably.

It is noteworthy to point out that social animals like lions are likely to have hierarchies and dominant animals are likely to monopolize preferred areas, but an ideal enclosure should provide equal opportunities for exploration and free movement to all individuals. In this study, we aimed to measure the enclosure zone usage pattern of each subject in a social configuration to ascertain how it met individual welfare requirements and related to other welfare indices.

Data analysis

We tested the following hypotheses in this study:

1. There would be no variations in behaviour indices between male and female subjects and the measures would perform uniformly for both sexes.

2. The wild-rescued lions would display higher behaviour diversity than captive-raised animals.

3. Bold individuals would differ in behavioural welfare indices compared to shy individuals.

4. There would be no difference in behavioural welfare parameters between pair-housed and group-housed subjects.

5. There will be no significant prediction of behavioural diversity by the proportion of aberrant repetitive behaviour and the enclosure usage patterns of subjects.

We used R statistical software version 3.4 and 3.5.2 through RStudio (RStudio, 2015) using packages, dplyr (Wickham, 2016), ggplot2 (Wickham, 2016), lubridate (Spinu, Grolemund & Wickham, 2018), tidyverse (Wickham, 2017), and funModeling (Casas, 2019). For exploratory data analyses, we used Shapiro–Wilk test and Levene’s test to ascertain the normal distribution and homogeneity of variance in SPI, SWI, latency and ARB values, respectively. We conducted bivariate Pearson’s correlation to ascertain the strength of association between four above-mentioned welfare indices. We also checked for correlation between enclosure area and zone usage bias.

For statistical analysis, we had four categorical predictor variables (personality trait, rearing history, sex, and social grouping), each with two levels (viz. bold & shy, wild-rescued & captive-raised, male & female and pair-housed & group-housed). We compared welfare indices across groups (categorical predictor variables) using independent samples t-tests for normally distributed dependent variables (enclosure usage, behaviour diversity, and ARBs). We used non-parametric Kolmogorov–Smirnov test for latency measures and the proportion of bold and shy behaviours performed by subjects during the novel object test. When comparing between means of two groups with different sample sizes, it is important to report effect sizes in addition to p-values to indicate the scale-independent degree of difference. We calculated effect sizes to quantify differences in welfare measures between groups (Cohen, 1992; Lakens, 2013).

Finally, we conducted multiple regression analysis to understand how the behaviour diversity of captive animals were predicted by their enclosure usage patterns and ARB levels. Before conducting a regression analysis, we checked for multicollinearity between independent variables using measures of VIF (variance inflation factor). Since enclosure usage and ARBs were not highly correlated, we used them as predictors for behaviour diversity in regression analysis.

Validation of keeper ratings

The mean latency times for all subjects after averaging both trials was 47.76 s (SD = 46.85). Subjects categorized as bold by keepers (M = 11.13, SD = 3.65, N = 21) showed significantly lower latency values (z = 2.89, p < 0.01, Cohen’s d = 7.28) compared to subjects categorized as shy (M = 102.71, SD = 17.4, N = 14)(Table 3, Fig. 2). Bold subjects also showed significantly higher percentage of bold behaviours (M = 87.24, SD = 8.74) than shy individuals (M = 15.86, SD = 9.5) (t(33) =  − 10.57, p < 0.01) (Table S3). These results validate the keeper rating of subjects on the bold-shy scale.

Inter-relationship between welfare indices

Latency was positively correlated (Fig. S2, Table 4) to enclosure zone bias (r = 0.67, N = 35, p < 0.01), and proportion of ARBs (r = 0.70, N = 35, p < 0.01). Latency was negatively correlated to behaviour diversity (r=−0.67, N = 35, p < 0.01). Enclosure zone bias was positively correlated with latency values (r = 0.66, p < 0.01), and proportion of ARBs (r = 0.66, p < 0.01) (Fig. S2, Table 4), but was negatively correlated with behaviour diversity (r =  − 0.71, p < 0.01). We found that enclosure zone bias was weakly positively correlated to enclosure size (r = 0.36, p = 0.05). Behaviour diversity was negatively correlated with latency to novel objects, (r =  − 0.67, p = 0.01), ARBs (r = 0.91, p = 0.01), and enclosure usage (r =  − 0.71, p = 0.01) (Supplementary Fig. 2, Table 4). ARB was positively correlated with latency to novel objects (r = 0.70, p = 0.01), and enclosure usage (r = 0.66, p = 0.01) but was negatively correlated with behaviour diversity (r =  − 0.91, p = 0.01) (Supplementary Fig. 2, Table 4). Multiple regression analysis (Table 5) indicated that ARBs and space usage homogeneity explained 85% of the variance in the behaviour diversity (R² = 0.85, F(2, 32) = 90.92, p < 0.01) (Table 5). The predicted regression equation is Behaviour diversity =1.8 + (−0.046)x(ARB) + (−0.46)x (Enclosure usage). The results from the regression indicate that subjects that show less ARB and use enclosure space more homogenously are likely to have higher behaviour diversity.