Impacts of illegal trade on socio-emotional and behavioural skills in macaques

Study site and population

Lao Conservation Trust for Wildlife (LCTW) operates in Lao PDR, a key corridor for the illegal trade in Southeast Asia, as a gateway between Thailand, Myanmar, Vietnam, Cambodia, and China. LCTW is registered by the Lao Government (under number 326/MoHA) and in the United Kingdom (under the number 1182501). Since 2018, this organisation has been engaged in the rescue, care, and release of native species victimised by illegal trade, currently providing shelter to over 400 animals across more than 26ha of land. Prior to this date, from 1994 to 2016, this centre was known as “Lao Zoo”, a place where visitors could interact and feed the resident animals, all of which were rescued from illegal trade. The majority of arrivals, both then and now, primarily consist of macaques rescued from the pet trade, where they were found in family settings, temples, or establishments like resorts. Regrettably, relevant information of the rescued macaques was missed between 1994 and 2017, which limits our knowledge of their life experiences. In order to link the background to a lack of social skills, a high expression of abnormal behaviour, a low score of welfare, and certain personality traits, we exclusively selected macaques with meticulously documented life experiences labelling them “focal” individuals. We collected data of abnormal and anxiety-like behaviours defined in the ethogram (Suppl. Tables 3-4 in the extended data), and we administered questionnaires (personality, welfare, social responsiveness) to these focal individuals only. Nonetheless, we gathered social behaviours for both focal and non-focal animals, as social interactions cannot be restricted to certain animals (Suppl. Table 5 in the extended data).

In this study we investigated 53 focal subjects selected from a larger sample of 88 macaques. These focal animals, aged between 1 and 18 years old (mean age ± SD = 8 ± 5 years), were distributed across eleven groups/enclosures (Suppl. Table 1 in the extended data for biographic information). The enclosures for the macaques range in size from 51.72 m² to 1,236.63 m², with an average of 594.47 m². Group sizes vary between 1 and 19 individuals, with an average of 7.53 individuals per group. Across all groups, there are 38 females and 42 males, distributed in different sex ratios. The available space per individual ranges from 7.39 m² to 109.31 m², averaging 64.22 m² per individual, ensuring adequate spatial distribution for each macaque according to group composition (Table 1).

During the course of this research, the composition of macaque groups changed due to the frequent arrival of rescued individuals at the centre. Some of these new arrivals were initially housed separately and gradually introduced to the most compatible group. Others, following unsuccessful introductions, were relocated to quarantine, pending future attempts (see Suppl. Table 2 in the extended data for more details). All the enclosures, except for P10, BP1 and BP3 are naturalised, free ceiling spaces equipped with an electric fence, available wild trees, two holdings for introducing new members and addressing medical issues, a swimming pool, and platforms for the macaques. P10, on the other hand, is a sizable cage with a natural floor that includes platforms, enrichment to hide, four holdings and one swimming pool. The BP enclosures consist of three interconnected 50m² cages (BP1, BP2 and BP3) that include a concrete floor, swimming pools and one platform each. Resident macaques in BP1 and BP3 share the middle cage BP2, enabling each group to use the additional space in rotational shifts every two days. There are no indoor facilities, and the animals remain outdoors with the whole group, except when necessary for specific reasons, such as medical interventions, cleaning, or repairs. As a consequence of the absence of indoor enclosures, keepers, and staff may enter the enclosures as needed, for tasks such as cleaning, maintenance, or medical procedures. Macaques are fed twice per day with seasonal vegetables, fruits, leaves and seeds, from 9:30 to 10:00 in the morning and from 15:00 to 15:30 in the afternoon.

Categories

We established several categories with the information collected on ZIMS (Zoological Information Management Software)⁶⁴ or provided by oral testimonies from LCTW staff about the previous traumatic events of the subjects in order to study which type of early adverse experience or stressful history may impact dramatically on the development of socio-emotional and behavioural skills and personality of the subjects. The information gathered may be incomplete, ambiguous, or scarce, specially of those who arrived at the centre before 2018, which was obtained by former workers (ZIMS) and one-time keepers at Lao Zoo (oral testimonies). Ten categories were created, three of them not being related to the background: sex, current age, and species. Seven categories were referred to the subject’s background: origin, type of rearing, life experience, social exposure during infancy, social exposure during juvenility, mother separation before 14 months old, and age of arrival at the centre (for categories see Suppl. Table 1 in the extended data, for codes’ meaning and details see Table 2).

Procedure and data collection

We combined two methods: questionnaires and behavioural observations. Socio-emotional and behavioural skills have been structured in five domains, following the BESSI [Behavioral, Emotional, and Social Skills Inventory]⁸⁷ proposal, defined in Table 3. The present study has been focused on three of these domains: (1) social engagement, (2) cooperation, and (3) emotional resilience.

Behavioural observations

We collected behavioural data through observations,⁸⁸ only while macaques had access to their outdoor enclosures, in other words, when they were not in the holding or hospital. Data on macaques’ behaviour was collected from November 14, 2022 to March 22, 2023 (for further information on the collected behaviours, see Suppl. Tables 3-5 in the extended data). We evenly distributed observation sessions of 20 minutes between 6:30 am and 17:00 pm on randomised days (Monday to Sunday). Each troop has been sampled for 12 ± 0.1 hours (min 11.67 hours, max 12.33 hours). Abnormal, anxiety-like, social (affiliative, sexual, agonistic and aggression-related) behaviours were recorded continuously with an all occurrences [multifocal] untimed-event strategy, whereas, the duration and frequency of grooming were recorded with a continuous [multifocal] timed-event strategy.⁸⁹ The duration of grooming collected will not be used in the present study. For data collection, the observer (first author) used a Sony ICD-PX370 voice recorder in three enclosures (P4, P5, P9) and Zoomonitor software⁹⁰ in eight enclosures (P1, P2, P6, P7, P8, P10, BP1, BP3), due to a variety of factors such as the lack of visibility, the number of individuals per group and the frequency of behaviours they exhibited. Following a bout recording strategy, we collected behaviours in bouts rather than the single repetition. For instance, in the case of repetitive and odd behaviours, we observed that “hit-self” or “self-bite” behaviours were seldom shown only once, but were performed in a set of repetitions or events. As an example, see Suppl. Video 1 in the extended data, in which an individual (Chock, P9) is exhibiting a “bout” that consists of: float limb, hit-self and self-bite amongst others, such as abnormal behaviours (e.g., self-pinch and abnormal displacement).

Questionnaires: raters

Three raters were carefully selected based on their substantial experience and significant time dedicated to working with the macaques. The first rater has spent more than four years continuously working with the macaques as a veterinarian and animal management. The second rater had an eight-month period of continuously working with the macaques as an enrichment coordinator. The third rater (first author) engaged with the macaques for a duration of six months, during which she collected the behavioural data for the present research (132 hours observation/total). Raters were explicitly instructed to refrain from discussing their assessments with other participants. We also provided comprehensive guidance on completing the three questionnaires, including discussions and clarifications on concepts associated with animal behaviour and welfare. Finally, we requested raters to respond according to their thoughts and current animal context.

Social Responsiveness Scale

We used the Social Responsiveness Scale (SRS), previously validated with adults⁹¹ and juvenile macaques,^92,93 to evaluate social engagement and cooperation skills. The SRS [short version] scale comprise 14 items, which are associated with statements that need to be scored by a human rater (e.g., “Seems self-confident when interacting with others”) using a Likert rating scale between 1 and 5 (1 = not true 0%, 2 = sometimes true 25%, 3 = often true 50%, 4 = almost always true 75%, and 5 = always true 100%) As described by Balint and colleagues (2021),⁹² the scoring of the items 1, 5, 7 and 14 were reversed, so that higher scores reflected greater social deficiency for each item.

Animal Welfare Survey

We employed the Animal Welfare Survey US [AWS]⁹⁴ to evaluate the emotional and resilience skills domain. This questionnaire consists of 12 items, each one with a statement or a question that needs to be scored or replied by a human rater (e.g., “How often this individual display signs of positive welfare?”) in a Likert rating scale of 1 to 5 (e.g., 1 = never, 2 = rarely, 3 = occasionally, 4 = frequently, 5 = constantly). It includes positive and negative indicators of welfare and well-being, validated with Rhesus macaques, capuchins and chimpanzees^95–97

Personality questionnaire

We used an adaptation of Cattell’s 16 Personality Factors Questionnaire^98,99 to assess the impacts of traumatic experiences on the development of personality. The questionnaire has been previously administered in chimpanzees⁶² and comprises 16 items, rated on a Likert scale 1–7. Each item was bipolar and the scores of raters described the subject evaluated closer to one pole or to the other.

Behavioural analysis

From the collected data, we calculated the frequency of each behaviour included in the ethogram per individual. Then, we calculated the rate⁸⁹ for anxiety-like, abnormal and social behaviours (grooming, maternal care, other affiliative, other agonistic, social play and sexual behaviours) per subject based on frequency/observation time. For each group, we create matrices of directed dyadic grooming interactions.

Rank was calculated with the “EloRating” package¹⁰⁰ in R,¹⁰¹ considering all dyadic agonistic interactions (dominance and submission) with a winner-loser outcome. Every macaque in each group was assigned a value between 0 (lowest ranking) and 1 (highest ranking).¹⁰²

Questionnaires

First, we assessed the interrater reliability of the items of each questionnaire via intra-class correlation coefficients (ICC): ICC (3,1), to evaluate the reliability of individual ratings, and ICC (3,k), which indicates the reliability of mean ratings¹⁰³ with JASP 0.17.3 software.¹⁰⁴ To determine the social responsiveness, animal welfare and personality dimensions, we conducted an Exploratory Factor Analysis using a Robust Unweighted Least Squares (RULS) for factor extraction.¹⁰⁵ We applied an orthogonal normalised Equamax rotation to generate uncorrelated factors.^106,107 We based our analysis on polychoric correlations (adequate to Likert-scale ordinal data with asymmetric or with excess of kurtosis data) to achieve factor simplicity and determine factorial structure and goodness of fit.^106–109 We calculated the correction for robust Chi-square with LOSEFER empirical correction.¹¹⁰ We considered factor loadings of the rotated loading matrix as significant when they were 0.5 or higher, in accordance with previous research.^102,111 Finally, we determined the number of factors following two procedures. First, we applied the “latent root criterion” (i.e., eigenvalues above 1)⁹²; and second, we used the optimal implementation of Parallel analysis based on minimum rank factor analysis.¹¹² We assessed the robust goodness of fit using the root-mean-square error of approximation (RMSEA). We considered RMSEA values between.05 and.08 as fair.¹¹³ We conducted all the analysis using FACTOR 12.04.01.¹¹⁴

We computed unit-weighted factor scores for each individual, following the procedure described by Weiss and colleagues (2009).¹¹⁵ This calculation involved taking the mean of all the items with salient loadings (>0.5). Items with positive salient loadings were assigned a score of +1 and items with negative salient loadings were assigned a score of -1. Thus, the score for each individual within a particular factor represents the weighted average of that individual’s scores on all the items related to the factor.

Influence of background on observed behaviours, social responsiveness, welfare, and personality

We assessed the effect of each background-related category on the dependent variables or individual measures using generalised linear models (GLM). We created a total of 11 models, one per each individual measure. As dependent variables we used the rate of (1) social, (2) anxiety and (3) abnormal behaviours, (4) the rank, (5-6) the social responsiveness, (7-9) personality, and (10-11) welfare domains. We included as fixed factors in our full models (a) sex, (b) estimated current age, (c) origin (d) species and background [(e) rearing, (f ) life experience, (g) infancy and (h) juvenile exposure, (i) mother separation, and (j) estimated age at the arrival).

Model interference and the selection of the subsets of best models were performed using dredge function, which is based on the Akaike Information Criterion corrected for small sample sizes (AICc).¹¹⁶ From all models tested, we considered the best explanatory model per each dependent variable those with the lowest AICC or the highest ΔAICc compared to the full model containing all the predictor variables. To assess the collinearity, we examined the value of the variance inflation factor (VIF), with a model considered acceptable when the VIF < 5 between predictor variables. This analysis was conducted in R,¹⁰¹ where we performed GLMs using the “MuMln” package¹¹⁷ and related analysis, including VIF calculations using the “performance” package.¹¹⁸ Plots were generated using the “ggplot2” package.¹¹⁹ An alpha level of 0.05 was used as a cut-off for significance.

Behavioural analysis

We divided the range of behaviours into 3 categories: social, abnormal and anxiety-like behaviours. We used the rates for anxiety-like, abnormal, and social behaviours (Suppl. Tables 6-11 in the extended data) to build the individual behavioural profiles.

Social behaviours include grooming interactions (both sender and receiver), social play (involving players regardless of whether they have started the game), other affiliative behaviours such as initiating contact (e.g., eye gaze, touch, following) or reciprocal affiliation (embrace, mutual teeth chattering, mutual touch), other agonistic behaviours (e.g., consolation, requesting/giving support), maternal care for behaviours directed towards unweaned infants, and socio-sexual behaviours for initiators only (Suppl. Table 7 in the extended data). Within social behaviours, it is noteworthy from our results that other agonistic behaviours such as appeasement or consolation, social play, and socio-sexual behaviours were observed at the lowest frequency in the majority of the enclosures, with social play being absent in P1 and P9, and socio-sexual in P6 (Suppl. Table 11 in the extended data). Only two groups exhibited a high rate of social play: P4, which had the highest number of infants and juveniles, and P10. No group exhibited a high rate of sexual behaviours (Suppl. Table 11 in the extended data). Maternal care was naturally observed only in those groups with unweaned infants (P8 and P5).

The anxiety-like category consists of four behaviours: genital self-inspection (including masturbation), scratching/rubbing, others self-directed behaviours, and yawning (Suppl. Table 8 in the extended data). As part of our predictions, we expected to find a high rate of anxiety-like behaviours in all groups. Indeed, our results show that not only was the rate of anxiety-like behaviours high, but also these behaviours were predominant over social and abnormal behaviours in the vast majority of the enclosures (Suppl. Table 10 in the extended data).

Abnormal behaviours were divided into six subcategories due to the wide range of such behaviours included in the ethogram. These subcategories comprised self-directed behaviours (e.g., poke body, grooming stereotypically, self-suck), postural (limited to leg-lift), self-abuse (e.g., self-bite, hit-self, trichotillomania), kinetic (e.g., float limb, pacing, twist), oral (e.g., regurgitation, reingestion, pica), and miscellaneous (e.g., touch urine stream, other abnormal behaviour not included in the ethogram) (Suppl. Table 9 in the extended data). Our findings reveal that the rate of abnormal behaviours was notably high in several groups, being higher than social behaviours in BP3, and slightly lower in P9 and P7 (Suppl. Table 11 in the extended data). All the groups exhibited abnormal behaviour, with the lowest rate observed in P4, which is again the group with more infants and juveniles who therefore arrived at the centre at an early age (Suppl. Table 11 in the extended data).

Social responsiveness

The reliability of individual ratings (3,1) ranged from 0.29 (Species typical reaction) to 0.76 (Socially tense) with a general mean of 0.52. The reliability of mean ratings (3, k) for the traits ranged from 0.55 (Species typical reaction) to 0.92 (Socially tense) with a mean of 0.75. There were no items with zero or negative values. The inter-rater reliabilities of all 14 items are presented in Suppl. Table 12 in the extended data.

Based on the normed MSA (Measure of Sampling Adequacy) all the items obtained values above 0.5, indicating its adequacy in representing the underlying constructs (Suppl. Table 13 in the extended data). Therefore, we retained all the items in the exploratory factor analysis. Based on the latent root criterion, we identified 2 factors to retain (Suppl. Table 14 in the extended data). The two factors accounted for 73.66% of the variance (Suppl. Table 15 in the extended data). According to the RULS, the value of the Kaiser-Meyer-Olkin (KMO) test was 0.66 (mediocre) [CI 0.364, 0.561] and Bartlett’s Test of Sphericity was significant (B=1755.7; df=91, p<0.001), thus indicating the adequacy of the correlation matrix. RMSEA fit was fair (0.078; [Bootstrap 95% CI 0.055, 0.068]).

According to the latent root criterion and an adequacy load of 0.5, two factors were loaded with the majority of the 14 items (Suppl. Table 14 in the extended data).

We interpreted the load of reverse items (1, 5, 7 and 14) on the factors as negative. Five items positively loaded on the first factor (F1): Socially tense (0.902), Social avoidance (0.875), (Not) Eye contact (0.760), Socially awkward (0.694), and Lonely (0.575). Furthermore, three reverse items scored negatively on this factor, Socially confident (0.885), Playful (0.701), and Communication skills (0.646). This factor was denominated as Social Reluctance. The second factor was labelled as Inappropriate Behaviour, scoring positively with seven items: Bizarre behaviour (0.825), Stereotypes (0.804), Restricted interests (0.784), Stares into space (0.774), Socially awkward (0.691), No physical coordinated (0.656), (Not) Eye contact (0.553); and negatively with Species typical reaction (0.787) and Communication skills (0.627).

Welfare

The reliability of individual ratings (3,1) ranged from 0.29 (Control of physical environment) to 0.68 (Number of relationships’ satisfaction) with a general mean of 0.47. The reliability of mean ratings (3, k) for the traits ranged from 0.55 (Control of physical environment) to 0.87 (Number of relationships’ satisfaction) with a mean of 0.72. There were no items with zero or negative values. The inter-rater reliabilities of all 12 items are presented in Suppl. Table 16 in the extended data.

Based on the normed MSA (Measure of Sampling Adequacy), all the items obtained values above 0.5 indicating its adequacy in representing the underlying constructs (Suppl. Table 17 in the extended data). Based on the latent root criterion, we identified 2 factors to retain (Suppl. Table 18 in the extended data). The two factors accounted for 76.65% of the variance (Suppl. Table 19 in the extended data). According to the RULS, the value of the Kaiser-Meyer-Olkin (KMO) test was 0.91 (very good) [CI 0.596, 1.200] and Bartlett’s Test of Sphericity was significant (B=1763.4; df=66, p<0.001), thus indicating the adequacy of the correlation matrix. RMSEA fit was close (0.03; [Bootstrap 95% CI 0.049, 0.062]).

According to the latent root criterion and an adequacy load of 0.5, two factors were loaded with the majority of the 12 items (Suppl. Table 20 in the extended data). One single item was loaded in the first factor (F1) Psychological stimulation (0.785); therefore, we labelled this factor Psychological Stimulation. The second factor was positively related to Cope with the stress (0.919), Impact of experiences (0.820), Balance of the experiences (0.707), Control of physical environment (0.560) and Physical health (0.544), and negatively with Negative welfare indicators (-0.886) and Stress frequency (-0.834); and this factor was named Welfare.

Personality

The reliability of individual ratings (3,1) ranged from 0.005 (Sensitivity/Objectivity) to 0.61 (Social boldness/Shyness) with a general mean of 0.37. The reliability of mean ratings (3, k) for the traits ranged from 0.015 (Sensitivity/Objectivity) to 0.92 (Social boldness/Shyness) with a mean of 0.59. There were no items with zero or negative values. The inter-rater reliabilities of all 16 items are presented in Suppl. Table 21 in the extended data.

Based on the normed MSA (Measure of Sampling Adequacy) three of the items (Sensitivity/Objectivity, Abstractedness/Pragmatism and Perfectionism/Flexibility) obtained values below 0.5 suggesting that they correlated with other items and failing its adequacy in representing the underlying constructs (Suppl. Table 22 in the extended data). Thus, we excluded these items during the exploratory factor analysis. In the second round, all the MSA values were above 0.5 (Suppl. Table 22 in the extended data). Based on the latent root criterion, we identified 3 factors to retain (Suppl. Table 23 in the extended data). The three factors accounted for 76.55% of the variance. According to the RULS, the value of the Kaiser-Meyer-Olkin (KMO) test was 0.833 (good) [CI 0.362, 0.862] and Bartlett’s Test of Sphericity was significant (B=1771.1; df=78, p<0.001), thus indicating the adequacy of the correlation matrix. RMSEA fit was mediocre (0.087; [Bootstrap 95% CI 0.055, 0.073]).

According to the latent root criterion and an adequacy load of 0.5, three factors were loaded with the majority of the 16 items (Suppl. Table 24 in the extended data). Items were previously defined by two adjectives, first one corresponding to the lowest score (1) and second one the highest score (7). We selected the second adjective to label the obtained factors. On the first factor (F1), the items that positively loaded were Pragmatism (0.809), Apathy (0.736) and Conventionalism (0.675), and those that negatively loaded were Unruliness (-0.751) and Openness (-0.610). Thus, we labelled this factor as Introversion. On the second factor (F2), the items positively loading were Self-assurance (0.683), Carelessness (0.563), and Openness (0.529) and the items negatively loading were Flexibility (-0.671) and Shyness (-0.589), thus we named this factor Calmness. Finally, the last factor was related to Detachment (0.792), Cooperation (0.598), Conventionalism (0.587), Emotional unsteadiness (0.559) and Shyness (0.538), and negatively with Affiliation (-0.707), Carelessness (-0.562) and Self-assurance (-0.526); Therefore, this factor was named Unfriendliness.

Influence of background on observed behaviours, social responsiveness, welfare, and personality traits

Two of the initial categories had to be removed from the analysis, due to the lack of variability amongst the sample: Mother separation (97% of the subjects were separated from their mothers and only 3% were not) and Origin (97% of the subjects were born in the wild, 3% were captive conceived). In the same line, c3 category (Species-Macaca fascicularis) was only represented by two subjects, which was not enough data to perform a generalised linear analysis. Therefore, we had to exclude the long-tailed macaques from the GLM analysis, although the description of their socio-behavioural profile is still included in this study (Suppl. Table 6-11 in the extended data). Tables 4 and 5 contain the comparison between the best explanatory model and the full model per each response variable, significant ones being in bold. The “Best Model” was selected according to the lowest AICc.^120,121 The collinearity between the predictors for those best models that seem to be influenced by two variables or more is less than 5 in all cases. Tables 6 and 7 show the best explanatory model per measure, significant predictive variables being in bold.

Regarding GLM analysis results for the observed behaviours, the best model that predicts the rate of social behaviours includes Juvenile Social Exposure and Sex, only the first one being significant ( Table 6). As shown in Table 2, the lowest value of this category (h1) corresponds to “Accompanied” whereas, h3 and h4 mean “Alone” and “Unknown” respectively. The direction of the prediction is inverse for h3, therefore subjects who spent their juvenility alone may be less social in the near or later future than those that were accompanied ( Figure 1). Pairwise contrasts of the significant category revealed that h3 or “Alone” subcategory is the predictive variable for lower display of social behaviour (h1-h3 p-value = 0.05; h2-h3 p-value = 0.04). The anxiety-like best model includes Estimated Age at Arrival, Species and Sex ( Table 6). According to this model, the later an individual arrives at the centre (j3), the higher the rate of anxiety behaviours will exhibit (estimate = 0.60) ( Figure 2) (j2-j3 pairwise contrast: p-value = 0.04, estimate = -0.963), males (a1) being more likely to be anxious than females (a2). In addition, the rate of anxiety-like behaviours seems to be significantly lower in rhesus (c5) or pig-tailed macaques(c4) than in stump-tailed (c1), with pairwise contrast being significant in c1-c5 (p-value = 0.003) and c1-c4 (p-value = 0.006) ( Table 6). Third, the best model that predicts the rate of abnormal behaviours consists of Rearing and Sex ( Table 6), with Rearing predictor being the only significant. The direction of e2 predictor’s influence is positive, meaning that macaques that were raised by humans are more likely to exhibit abnormal behaviour than those that were raised by their own parents or by foster parents ( Figure 3). Lastly, Sex is the significant parameter that may predict the acquisition of the rank in the hierarchy ( Table 6). The direction of the influence is negative for a2, meaning that males are more likely to hold a higher rank.

Figures 1-3. Boxplots representing the influence of background on observed behaviours.

Influence of Juvenile Social Exposure on the rate of social behaviours (Best model 1), influence of Estimated Age at Arrival on the rate of anxiety-like behaviours (Best model 2), and influence of Rearing on the rate of abnormal behaviours (Best model 3). Black inverse triangles represent outliers. For definition of the codes see Table 2.

In relation to the predictive models for the results of the questionnaires, Juvenile Social Exposure is the significant variable that seems to predict the first social responsiveness domain, Social Reluctance ( Figure 4), whereas, Rearing is the predictor for the second domain, Inappropriate Behaviour ( Table 7). Best model 5 consists of Juvenile Social Exposure only, the direction of the influence being positive with h3 and pairwise contrasts being significant for h1-h3 (p-value=0.003) and h2-h3 (p-value=0.03) Therefore, social withdrawal during juvenility may result in higher Social Reluctance in the near or later future. Best model 6 involves Rearing only, with e2 or hand-rearing being the significant variable to predict a higher item score for Inappropriate Behaviour ( Figure 5). Regarding personality questionnaires, we obtained three significant predictive models corresponding to the three personality resulting domains from Cattell 16PF questionnaire ( Table 7). The best explanatory model for Introversion includes Juvenile Social Exposure, Life Experience, Sex and Species. According to GLM results, subcategories f2 (macaques used for human entertainment), f4 (macaques that were rescued from the trade whose past is not exhaustively known), and a2 (females) positively influence this domain ( Figure 6). In contrast, subcategories h4 (unknown juvenile social exposure) and c2 (Assamese macaques) negatively predict this domain, meaning that Assamese tends to be less introverted than the rest of studied species ( Figure 7). However, pairwise contrast shows that within Juvenile Social Exposure, h3 is also a predictive subcategory for lower introversion (h2-h3 estimation=-5.15, p-value=0.007), only f4 would be a significant predictor for this variable (f1-f4 p-value=0.02; f3-f4 p-value=0.04), and any pairwise contrast is significant for Species category. Best Model8, that may predict Unfriendliness, consists of Rearing (e2) whose estimation is positive, meaning that hand-rearing predicts higher item score for this domain ( Figure 8). In contrast, e2 or hand-rearing is the only significant variable that negatively predicts the last personality domain, Calmness ( Table 7 and Figure 9). Regarding the last questionnaire, the best model to predict the Welfare domain consists of Rearing (Best model 10), the influence of e2 predictor being negative for welfare score ( Table 7 and Figure 10). Finally, Best model 11 involves Infancy Social Exposure and Life Experience predictors. Subcategory g3 seems to negatively influence the Psychological Stimulation domain, thus macaques that spent their infancy alone are more likely to be rated lower in welfare than those that were accompanied (g1-g3 p-value = 0.03) ( Table 7) ( Figure 11). Moreover, macaques used for entertainment (f2) and trade (f4) seem to be predictive of the item score for this domain as well ( Figure 12). Nevertheless, pairwise contrast revealed that only f1-f4 comparison is significant at p-value = 0.02 with estimate = 1.276.

Figures 4-5. Boxplots representing the influence of background on social responsiveness domains.

Influence of Juvenile Social Exposure on Social Reluctance domain (Best model 5) and influence of Rearing on Inappropriate Behaviours domain (Best model 6). Black inverse triangles represent outliers. For definition of the codes see Table 2.

Figures 6-9. Boxplots representing the influence of background on personality traits domains.

Influence of Juvenile Social Exposure and Life Experience on Introversion domain (Best model 7) and influence of Rearing on Unfriendliness (Best model 8) and Calmness (Best model 9) domains. Black inverse triangles represent outliers. For definition of the codes see Table 2.

Figures 10-12. Boxplots representing the influence of background on Welfare domains.

Influence of Rearing on Welfare domain (Best model 10), and influence of Infancy Social Exposure and Life Experience on Psychological Stimulation domain (Best model 11). Black inverse triangles represent outliers. For definition of the codes see Table 2.

Social engagement and cooperation skills domains: social skills and rank

According to our GLM analysis findings, the rate of social behaviours is significantly influenced by Juvenile Social Exposure ( Table 6). Macaques who experienced social isolation during juvenile years appear to exhibit less social behaviour. Similarly, the social responsiveness domain “Social Reluctance” is influenced by the social withdrawal during juvenility ( Table 7). This domain includes avoidance of social interactions, a lack of social self-confidence, diminished playful interest, and communication skills. Therefore, these results suggest a link between social impairment and social anxiety and the deprivation of social stimulation with conspecifics during the critical period from 14 months to 36 months of age in macaques. For decades, early adverse experiences, such as mother deprivation or infant isolation, have been considered crucial for the development of social skills in non-human primates.³³ Our results do not contradict these established facts; instead, they highlight the significance of social stimulation during later developmental stages in shaping social skills, especially in the realms of cooperation (social warmth) and social engagement skills (sociability). Additionally, multiple studies have stressed the critical role of the mother and peers during pre-adolescence in shaping the behavioural profile of primates.¹²⁴ Likewise, our findings indicate that the Inappropriate Behaviours domain seems to be influenced by the Rearing category, mirroring the predictor models for Welfare and the rate of abnormal behaviours.⁶ This social responsiveness domain comprises bizarre behaviours and stereotypes amongst other non-typical behaviours of these species. Moreover, it includes items related to communication skills and physical coordination deficiencies. In substance, a high score in this domain means that an individual displays oddly in a social context, closely associated with abnormal behaviour and social anxiety, and indicative of compromised welfare.⁶³

Regarding rank, we did not observe any significant effects related to the background, as predicted. The significant factor that appeared to influence the rank was “Sex”, probably due to the male-dominant nature of the studied macaque species. Bastian and colleagues²² revealed that absence of adults and limited social interactions during early life negatively affect the acquisition of dominance rank, along with age and sex. We suggest that our results differ from previous studies due to the limited background variability in our sample. In the study conducted by Bastian and colleagues, there were three distinct groups with different rearing backgrounds. In contrast, the majority of our subjects have experienced traumatic pasts, and as a result, the predictive strength of the different “background” variables may be significantly lower compared to the influence of Sex.

Emotional Resilience skills domain: welfare, anxiety, and abnormal behaviours

Secondly, we predicted to find, amongst our sample, psychological distress based on higher rate of anxiety-like behaviours and/or higher expression of abnormal behaviour and stereotypes.^1,2,37,38 While the rate of anxiety is notably high across all enclosures, as predicted, our results reveal that the rate of anxiety-behaviours appear to be significantly influenced by the Estimated Age at Arrival, Sex, and Species only ( Table 7). In contrast to previous studies in macaques, bonobos, capuchins, and chimpanzees,^{62,125–127,} our findings indicate that males are more likely to exhibit anxiety-like behaviours compared to females. According to our findings, the manifestation of anxiety-like behaviours may vary amongst different species. Despite that the best explanatory model for abnormal behaviours does not include Species category, the rate of both behaviours seems to be significantly lower in Rhesus and pig-tailed compared to stump-tailed or Assamese. This lends support to the idea that using a single species as a model for abnormal or anxiety behaviour within the Macaca genus may not be advisable.¹²⁸

Contrary to our predictions, the social exposure in infancy or juvenility do not appear to predict anxiety in our sample, as reported in other NHP.³ Nonetheless, the j3 subcategory or arriving at the centre in adulthood seems to be significant to predict a higher rate of anxiety. Given that individuals arriving at the centre in later life or adulthood might have spent more time in the illegal trade, we interpreted that the longer an individual has been a victim of the illegal trade, the higher the rate of anxiety behaviours, regardless of the life experience or other conditions. Therefore, we claim that the set of potential distressing events associated with illegal trade, such as exposure to humans, social deprivation with conspecifics, or psychological abuse, have a discernible impact on behavioural outcomes. In essence, elevated levels of anxiety are considerably more prevalent in those macaques who experienced episodes of distress over extended periods, with the absence of peer interactions during juvenility being particularly pivotal for the development of social anxiety.

Regarding the results of the Welfare questionnaire, we named the first domain as “Welfare” because the items that positively loaded on this factor were indicative of preserved welfare, such as “good physical health” and “coping well with the stress”, while those with negative loadings were associated with compromised welfare, such as “high stress frequency”. This is consistent with the traditional use of abnormal behaviour and stereotypes as predictive factors of negative welfare as outlined by Mason and colleagues.⁶³ Several studies have highlighted the profound impact of rearing conditions on the development of behavioural profiles in laboratory macaques, particularly the exhibition of odd repetitive behaviours or stereotypes.^{12,13,16,17,21,27,28} The best explanatory model for predicting the rate of abnormal behaviours in our sample includes the type of rearing, which aligns with previous findings. In addition, this parameter is also included in the best predictor model for Welfare, as observed in the Inappropriate Behaviour domain. Consequently, hand-rearing conditions emerge as a risk factor for an individual’s inability to cope with stress in social and non-social events, with the resulting detriment to welfare. Since most victims of the illegal trade are separated from their mothers at an early age and reared by humans for purposes such as keeping them as pets or exploiting them for economic purposes, it may be challenging to prevent hand-rearing practices in these circumstances. Nonetheless, rescue centres that frequently receive unweaned rescued primates should consider the possibility of finding foster parents to rear these infants instead of opting for hand-rearing. Similarly, zoos that occasionally care for neglected newborns should contemplate fostering as an alternative to evade the potential effects of hand-rearing or, at the very least, employ both methods to minimise its impact.

Finally, the Psychological Stimulation domain seems to be negatively influenced by the lack of social exposure during infancy and the type of life experiences. EFA analysis revealed that Psychological Stimulation, represented by the item 5 of the Animal Survey Welfare questionnaire, constituted a domain in itself (Suppl. Table 20 in the extended data), suggesting the critical role of psychosocial enrichment in determining welfare in our sample. This item could be rated from “very bored” (1) to “very stimulated” (5), with a higher score indicating positive welfare. Individuals raised in social isolation during infancy for recreational purposes tend to be rated lower in this domain, regardless of the quality of the enrichment. This finding denotes that past experiences can diminish engagement skills of macaques with their environment. Ideally, the past life of the resident animals at zoos and rescue centres should be considered when designing high-quality enrichment protocols to guarantee their welfare.

Personality traits

Thirdly, we introduced Cattell’s 16PF questionnaire for the first time in macaques, which had previously been validated in humans and chimpanzees.^62,98,99 We identified three personality domains: the first domain, which we labelled as “Introversion”, showed a positive relationship with pragmatism and apathy, and a negative relationship with openness. The second domain, named “Calmness”, was associated with items opposing vigilance and apprehension. The third factor was designated “Unfriendliness” as it displayed an inverse relationship with affiliation, carelessness and self-assurance. On one hand, we expected to find similar personality traits to those in the reference study, due to the similarity of the backgrounds, despite the macaques being different species. It is important to note that although we did not divide the factor analysis by species due to sample size constraints, species differences could potentially influence the structural outcomes. Given that many of the macaque species in this study were housed in the same facility, we did not expect significant variation. However, future studies with larger sample sizes and separate analyses for each species would be beneficial in exploring these potential effects more thoroughly.

We obtained two opposite and comparable domains (Introversion-Extraversion, Calmness-Anxiety) and one non-related (Unfriendliness) to any of the factors described by Ortin and colleagues.⁶² Nevertheless, the resulting domains are comparable to those defined by Weiss and colleagues in Rhesus macaques: Dominance, Confidence, Openness (Introversion in our results), Anxiety (Calmness in our results), and Friendliness (Unfriendliness in our results).¹²⁹ On the other hand, we predicted that the personality profiles may be impacted by, at least, one of our background categories, as occurred in the baseline study.⁶² GLM analysis shows that the Introversion factor seems to be impacted by several predictive variables which were both related (Juvenile Social Exposure and Life Experience) and not related to the background (Sex and Species) ( Table 7).

Regarding Sex, females tend to be more introverted than males, which aligns with macaques’ social structure, where males have to leave their natal group and socialise to be integrated in other groups, while females typically remain in their original group.¹³⁰ Regarding Species, it is worth recalling that most of the groups are mixed-species, except for P1, P2, P8, P9 (stump-tailed only) and BP1 (three pig-tailed only). Because of that, we cannot be certain whether this influence is accurate or caused by the unequal composition of the group. Furthermore, pairwise contrast reveals that the difference between species is not significant enough to predict the introversion trait in individuals. We suggest conducting additional research to thoroughly investigate macaque personality at the species-level.

The variables related to the background that predict Introversion are j4 or unknown juvenile social exposure, f2 or entertainment life history and f4 or unknown past in illegal trade. At first sight, we could only take into account f2, which means that those macaques that were used or exploited for human entertainment are more likely to be introverted than those that were pets or born in the zoo. Nevertheless, those individuals who were raised in social isolation during juvenility are more likely to be introverted than those who were accompanied, according to the pairwise contrast analysis. We consider these significant results to be in line with our reference study in terms of a reduced social interaction in early life shapes extroversion-introversion traits in individuals.⁶² Finally, the best explanatory models for both Calmness and Unfriendliness domains are predicted by the same subcategory, e2, in opposite directions: hand-rearing individuals are more prone to be more anxious and unfriendly or less calm and friendly. We assert that individuals’ anxiety levels and social warmth may also be related to early mother separation and high exposure to humans in early life as this is what being hand-reared implies.^61,62 These findings are consistent with previous research in NHP.^{3,6,20,21,42,48,131}

Consequently, the personality structure of the study sample seems to be shaped by the adverse past according to our results. Our results show potential for the use of Cattell’s 16PF for the assessment of personality in macaques. Nonetheless, as shown in Suppl. Table 21 in the extended data, three of the items (Sensitivity-Objectivity, Abstractedness-Pragmatism and Perfectionism-Flexibility) obtained ICC 3,k values below 0.5, which indicates poor reliability. Whilst we obtained significant results consistent with our predictions, raters agreed with (1) the complexity of the adjectives to describe the personality traits of the focal macaques, (2) the need to have a deeper understanding of the focal subjects and (3) macaque behaviour to fill the questionnaires. Therefore, further studies on the use of Cattell’s 16PF questionnaire in personality assessment should be conducted to evaluate its suitability in macaques. We also suggest using a simpler and shorter questionnaire validated in Macaca fuscata¹³² or commonly implemented and validated questionnaires in macaques as the Hominoid Personality Questionnaire.¹²⁹

Overall, these results would be in concordance with (1) the baseline study of Lopresti-Goodman in rescue chimpanzees,⁶¹ which states that victims of the wildlife trade tend to exhibit psychological distress and more stereotype; (2) former research in bushmeat chimpanzees,⁶² which states that traumatic past predicts higher anxiety in the victims, and (3) in exploited macaques (Macaca leonina),⁵⁴ which states that stressful episodes related to the use and abuse of macaques for economic profits leads to their detrimental welfare.