Relationship quality affects social stress buffering in dogs and wolves

https://doi.org/10.1016/j.anbehav.2021.06.008 0003-3472/© 2021 The Authors. Published by Elsevie license (http://creativecommons.org/licenses/by/4.0/) Social relationships can be described by a series of components, all having putatively different functional roles in the lives of humans and other social species. For instance, certain relationship characteristics can strongly influence how individuals deal with stress, ultimately influencing their fitness. However, species vary highly in regard to which components of their relationships influence stress buffering and how. Variation in species’ social organization could explain such differences. Comparing closely related species subjected to different ecological constraints can be especially informative when investigating this hypothesis. Here, we compared whether relationship quality differently influences how grey wolves, Canis lupus, and domestic dogs, C. l. familiaris, react to a series of stressors. We tested the role of various relationship components (i.e. two affiliation indices and two aspects of dominance rank) in mediating stress reactivity, social support seeking and social referencing in dyads of pack-living animals. To do so, we conducted systematic long-term observations of the social interactions between animals and an experimental test battery exposing animal dyads to a series of stressors (e.g. novel environment exploration, separation from and consequent reunion with the partner, exposure to a novel object and a threatening human). We found that a large rank distance and high affiliation index based on the number of friendly behaviours exchanged during everyday life (but not dominance status or the affiliation index based on the time spent in body contact) were related to a dampened stress response in both species. These results suggest a functional role of these two relationship components in the stress buffering of both dogs and wolves. © 2021 The Authors. Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour. This is an open access article under the CC BY license (http://creativecommons.org/licenses/ by/4.0/).

Social relationships can be described by a series of components, all having putatively different functional roles in the lives of humans and other social species. For instance, certain relationship characteristics can strongly influence how individuals deal with stress, ultimately influencing their fitness. However, species vary highly in regard to which components of their relationships influence stress buffering and how. Variation in species' social organization could explain such differences. Comparing closely related species subjected to different ecological constraints can be especially informative when investigating this hypothesis. Here, we compared whether relationship quality differently influences how grey wolves, Canis lupus, and domestic dogs, C. l. familiaris, react to a series of stressors. We tested the role of various relationship components (i.e. two affiliation indices and two aspects of dominance rank) in mediating stress reactivity, social support seeking and social referencing in dyads of pack-living animals. To do so, we conducted systematic long-term observations of the social interactions between animals and an experimental test battery exposing animal dyads to a series of stressors (e.g. novel environment exploration, separation from and consequent reunion with the partner, exposure to a novel object and a threatening human). We found that a large rank distance and high affiliation index based on the number of friendly behaviours exchanged during everyday life (but not dominance status or the affiliation index based on the time spent in body contact) were related to a dampened stress response in both species. These results suggest a functional role of these two relationship components in the stress buffering of both dogs and wolves.
© 2021 The Authors. Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour. This is an open access article under the CC BY license (http://creativecommons.org/licenses/ by/4.0/). Social relationships are among the most important features of a social animal's life. Despite a long history of research aiming at better understanding their patterning, importance and functions, social relationships remain among the least understood topics in the field of animal behaviour (Massen, Sterck, & de Vos, 2010). It has been shown that the number, strength and quality of social relationships can be associated with an increase in an individual's fitness (Silk et al., 2010), probably through mechanisms such as social support, thermoregulation or increased safety from predation (Silk, 2007). These mechanisms might reduce the physiological stress response further contributing to the individual's fitness by reducing the consequences of long-term stress (stress buffering; see Rault, 2012 for a review). However, relationships with various partners differ in quality, and not all of them have a positive effect on an individual's fitness; thus, it has been suggested that only specific relationship components (e.g. affiliation) might play a role in stress buffering (Kiyokawa & Hennessy, 2018;Wittig et al., 2016). The link between various relationship components, fitness and stress buffering is poorly understood, despite its potential to shed light on the varying functions of social relationships, both within and across species. Furthermore, the paucity of studies adopting a comparative approach makes it difficult to disentangle whether (and to what extent) the functions of different relationship components are conserved across evolution and/or reflect adaptations to specific environmental characteristics (Kiyokawa & Hennessy, 2018).
Affiliation and dominance are the most used components to describe relationships in animals (Silk, Cheney, & Seyfarth, 2013). Affiliation, operationally based on the number of sociopositive interactions exchanged and/or the time spent in association/proximity with the partner (Massen et al., 2010;Silk et al., 2013), has been associated with stress reduction in a variety of species (Massen et al., 2010;Rault, 2018). This effect can be due to direct positive physiological changes deriving from affiliative interactions (e.g. at a proximate level, grooming can reduce heart rate: Boccia, Reite, & Laudenslager, 1989;Aureli & Smucny, 2000). Also, behavioural mechanisms such as physical presence or closeness of a potential ally (leading to a reduction in cortisol levels, Hennessy, Mendoza, Mason, & Moberg, 1995), receiving social support during stressful events (e.g. intergroup conflicts, , obtaining valuable information from a more knowledgeable partner (Merola, Prato-Previde, & Marshall-Pescini, 2012) or behavioural coordination (partners acting together and thereby better dealing with a stressor, such as a predator's presence, Duranton & Gaunet, 2016) can ultimately lead to stress reduction.
Importantly, the presence of affiliative partners affects stress levels differently in species with diverse social organizations (Hennessy, Kaiser, & Sachser, 2009), even when the species are closely related. For instance, social contact has positive effects in monogamous but not in polygynous rodent species (Glasper & DeVries, 2005), and stress reactivity is dampened by the presence of a bonded mate in a New World monkey species characterized by long-term pair bonding (titi monkeys, Callicebus moloch) but not in another species (squirrel monkeys, Saimiri sciureus) characterized by larger mixed-sex groups (Hennessy et al., 1995).
The dominance component of a relationship (described by the asymmetric exchange of submissive and/or dominance signals, De Waal & Luttrell, 1989) between individuals living in the same group seems to be also related to stress coping, wellbeing and fitness (Creel, 2005;Majolo, Lehmann, de BortoliVizioli, & Schino, 2012). For instance, the presence of a dominance hierarchy can help members of a group deal with environmental stressors (Bonanni et al., 2017;S arov a et al., 2013). Specifically, an age-graded hierarchy facilitates group coordination, with more dominant (and older) individuals being better informed and making decisions, and more subordinate individuals benefiting from following the dominant's decisions (Bonanni et al., 2017;McComb et al., 2011;S arov a et al., 2013). However, whether dominant or subordinate animals in a group have higher cortisol levels (Sands & Creel, 2004;Sapolsky, 2005) depends on the type of stressor, the species and what type of stress-coping strategies the individuals adopt. For instance, in stressful situations, subordinate individuals might actively look for social support or avoid dominant individuals, depending on the species-specific dominance (i.e. whether more or less despotic) and breeding style (e.g. whether cooperative or not; see Creel, Dantzer, Goymann, & Rubenstein, 2013 for a review). Comparative studies investigating the potentially different effects of relationship components such as affiliation and dominance on stress buffering are still scarce (but see Koski, Koops, & Sterck, 2007 for an example), even though they would help us understand how different socioecologies shape the functions of such relationship components (Creel et al., 2013;De Waal & Luttrell, 1989;Thierry, 2000). Moreover, it remains to be investigated whether stress buffering is different for individuals of different rank.
Here we set out to investigate whether affiliation and dominance have a functional role in buffering environmental stress in two canid species: the grey wolf, Canis lupus, and the domestic dog, C. l. familiaris. We chose to compare these two species because wolves and dogs are closely related and still interfertile (Lindblad-Toh et al., 2005;Wang et al., 2016), but they have gone through a (relatively recent) differential evolution and presently have different socioecologies (Boitani & Ciucci, 1995;Marshall-Pescini, Cafazzo, Vir anyi, & Range, 2017;Udell & Wynne, 2010). While many studies have aimed at describing dogs' and wolves' social organizations, it is still unclear whether social relationships continue to have a similar functional role in the two species.
Both dog (living as free-ranging: Cafazzo et al., 2010;as pets: van der Borg, Schilder, Vinke, & de Vries, 2015;Trisko, Sandel, & Smuts, 2016; in captivity: Dale, Range, Stott, Kotrschal, & Marshall-Pescini, 2017; Essler et al., 2017;Marshall-Pescini, Schwarz, Kostelnik, Vir anyi, & Range, 2017) and wolf groups (living in the wild: Mech, 1999;Peterson, Jacobs, Drummer, Mech, & Smith, 2002;Sands & Creel, 2004;in captivity: Cafazzo, Lazzaroni, & Marshall-Pescini, 2016;Dale et al., 2017;Essler et al., 2017;Mazzini, Townsend, Vir anyi, & Range, 2013;Romero, Ito, Saito, & Hasegawa, 2014) are characterized by similar age-graded linear dominance hierarchies and preferential affiliative bonds between individuals. Moreover, the two species share similar formal signals of submission and dominance (Schenkel, 1967). Despite this, a series of differences in the social organization of the two species could suggest that the functions of their relationships might have changed during the process of domestication. In fact, wolves generally live in family groups composed of a breeding pair and its offspring, and they strongly rely on cooperative breeding and cooperative hunting for their survival (MacNulty, Tallian, Stahler, & Smith, 2014;Mech & Boitani, 2003). Free-ranging domestic dogs (which make up 70e80% of the world dog population, Lord, Feinstein, Smith, & Coppinger, 2013), on the other hand, have a more prominently scavenging feeding style, a promiscuous mating system (Boitani & Ciucci, 1995;Boitani, Ciucci, & Ortolani, 2007;van Kerkhove, 2004) and a general absence of cooperative breeding (see Marshall-Pescini, Cafazzo, et al., 2017 for a review). They form stable packs only under certain conditions and often roam alone or in pairs, which is why dogs have been described as facultatively social (Boitani & Ciucci, 1995), posing questions about the functional role of their sociality and how this may differ from that of wolves. The lack of studies specifically investigating the link between various relationship components and factors strongly connected to fitness (e.g. stress buffering) does not allow us to disentangle to what extent the differences in the socioecologies of dogs and wolves shape the functions of their social relationships.
Only one study has investigated comparatively how the presence of a specific pack mate influenced how wolves and dogs behave when facing an environmental stressor (i.e. a novel object; Moretti, Hentrup, Kotrschal, & Range, 2015). This study found that the presence of pack mates favoured object exploration, suggesting social stress buffering. Some species differences have also emerged, with rank distance influencing wolves' (the smaller the rank distance, the longer the subjects investigated the object) but not dogs' investigation of novel objects (Moretti et al., 2015). However, in this study, animals were tested in a single context, the effect of only one relationship component (i.e. ordinal rank in the pack) was investigated, and the behavioural analyses focused on interactions with the novel object, rather than social behaviours directed towards the partner. Therefore, the study provides a limited insight into the role of social relationships in moderating dogs' and wolves' reactions to environmental stressors.
The aim of the present study was to investigate how the dominance and affiliative components of a relationship may modulate the behavioural and physiological response to potentially stressful situations wolves and dogs may encounter in their daily lives. To this aim, we conducted the present study on similarly raised, captive dogs and wolves living in conspecific packs at the Wolf Science Center near Vienna (Austria). Based on systematic long-term observations of the social interactions between animals, we first characterized the relationship between individuals in terms of dominance and affiliation. We then experimentally tested each dyad using a recently developed test battery (Cimarelli, Marshall-Pescini, Range, & Vir anyi, 2019) that includes a series of contexts expected to create mild stress in the animals. The battery includes exploring a novel environment in the presence of a partner, being separated from (and being subsequently reunited with) the partner in a novel environment, encountering a novel object together, and together facing a potential conflict with an unfamiliar subject. During the tests, we recorded to what extent individuals in the dyad showed a behavioural and physiological stress activation, sought social support (e.g. by staying near each other) and coordinated their behaviours (e.g. by looking towards one another). To evaluate the physiological stress response of the animals (using changes in salivary cortisol as a proxy), we collected saliva samples before and after the testing procedure.
Given the results from previous studies in which affiliation seems to play a similar role in dogs and wolves, positively modulating synchrony and communication in stressful situations (e.g. territorial defence, Cassidy, MacNulty, Stahler, Smith, & Mech, 2015), we predicted that in both species, affiliation would promote separation distress (Mazzini et al., 2013) but dampen stress reactivity to other stressors (both behaviourally and physiologically, Rault, 2012) while being together with the partner. This may occur, for instance, by means of social support seeking (measured as proximity seeking and body contact in a potentially stressful situation, Dale et al., 2017;Marshall-Pescini, Schwarz, et al., 2017), social information seeking for coordination (measured as gaze duration and gaze alternation between the partner and a novel stimulus, Merola et al., 2012) and movement coordination during exposure to the stressor (Duranton & Gaunet, 2016). However, given wolves' higher dependence on pack mates (i.e. for hunting, high-risk territorial defence and puppy rearing, (Marshall-Pescini, Cafazzo, et al., 2017), we expected such stress-mediating effects to be stronger in wolves than in dogs (Glasper & DeVries, 2005;Hennessy et al., 1995).
Regarding dominance, we predicted that in both dogs and wolves, subordinate individuals would be more stressed and actively look for social support more from the dominant partner than vice versa (Essler et al., 2017;Mazzini et al., 2013;McLeod, Moger, Ryon, Gadbois, & Fentress, 1996). Thus, subordinates should follow and seek proximity with their dominant partners ( Akos, Beck, Nagy, Vicsek, & Kubinyi, 2014;Bonanni, Cafazzo, et al., 2010;Peterson et al., 2002) and carry out more social referencing (i.e. by looking longer and alternating the gaze more often between novel stimuli and their partner) towards the dominant partner than vice versa (assuming that the more dominant partner is seen as better informed, Peterson et al., 2002;Bonanni, Cafazzo, et al., 2010). However, considering that dogs might form more despotic social groups than wolves (Range, Ritter, & Vir anyi, 2015), we also expected subordinate dogs to be generally more stressed than subordinate wolves.
Further, since previous studies have shown that wolves closer in rank distance are more likely to coordinate with each other, but no or an opposite effect was found in dogs (Moretti et al., 2015;Marshall-Pescini, Schwarz, et al., 2017), we expected that rank distance would also affect stress mediation differently in the two species. As such, the smaller the rank distance the higher the stress-buffering effect of a partner (hence, the lower the cortisol response) and the more time spent in proximity in wolves, with the opposite in dogs.

Ethical Note
The experimental procedures were approved in accordance with Good Scientific Practice guidelines and national legislation by the Ethical Committee for the use of animals in experiments at the University of Veterinary Medicine Vienna (Ref: 19/04/97/2014).

Subjects
All subjects tested in the present project were raised and lived in packs at the Wolf Science Center, Austria (www.wolfscience.at). Eleven grey wolves, all originating from captive populations living in game parks in Austria, the United States and Canada, participated in the study. The dogs originated from shelters in Hungary (N ¼ 8) or were bred at the Wolf Science Center from two females and external, mixed-breed males (N ¼ 6, 2014 generation). Subjects were hand-raised in peer groups from the age of 10 days on or, in case of the dogs born in 2014, spent at least 4 h per day with humans and other puppies (without the mother). After 5 months, all animals started to live full-time in their packs in large 2000e8000 m 2 enclosures, but they continued to take part in daily training and various behavioural experiments and were walked regularly by their trainers with whom the animals had a close and trustful relationship (see Range et al., 2015 for more details). Pack composition had been stable for at least 1 year when pairs of animals living in the same pack participated in the current study. We tested dogs belonging to four different packs (three formed by three individuals and one formed by six individuals), while wolves belonged to three different packs (two formed by three individuals and one formed by five individuals). Fourteen dogs in 15 dyads (nine males, five females; mean age ± SD ¼ 31.93 ± 17.49 months) and 11 wolves in 14 dyads (seven males, five females; mean age ± SD ¼ 53.68 ± 20.13 months) participated. Five dog dyads were composed of siblings and three wolf dyads of siblings or cousins. See Table 1 for a complete list of the dyads tested.

Observations
Packs were regularly observed using 10 min focal observations (Pocket Observer program, 3.2 Software) and data were then imported into the Observer XT 10.5 program (both from Noldus Information Technology, Wageningen, The Netherlands). Agonistic and affiliative interactions were recorded using an 'all occurrence method' (Altmann, 1974) when all individuals of the pack were present. Data were then analysed to identify the dominance and affiliative relationships between all individuals in the pack. The frequency of agonistic interactions (formal signals of dominance and submission as in Dale et al., 2017;Essler et al., 2017; see Table 2) were used to calculate each individual's David's score, Gammell, de Vries, Jennings, Carlin, & Hayden, 2003). David' scores were calculated based on the behavioural matrix that showed the best coverage, linearity and directional consistency index (i.e. frequency of submissive or dominant behaviours, and in some cases both frequencies combined, Gammell et al., 2003). Based on the David's score we assessed for each dyad which was the dominant partner (dominance status) and calculated the absolute rank distance between the two individuals (i.e. the absolute value of the difference between individual A's and B's David's score, as a measure of disparity between the respective dominance positions, De Waal, 1991). We used both measures, since dominance status can provide information at a dyadic level about who is dominant over whom, while rank distance also quantitatively takes the relative 'strength' of each individual in relation to its partner into account while also considering their relationships to all other pack members (more directly linked to competition, Marshall-Pescini, Schwarz, et al., 2017). The frequency and/or duration of affiliative behaviours (number of times in which the two individuals were seen in body contact, time spent in body contact, number of times an individual approached the partner in a friendly way, number of times an individual lay down or stood in a friendly manner close to a partner, number of times an individual groomed another one, time spent grooming the partner; see Table 2) were first normalized per dyadic observation time and then used to calculate dyadic affiliation scores by conducting a principal component analysis (PCA, Silk et al., 2013).

Procedure
Testing took place in unfamiliar outdoor enclosures at the Wolf Science Center (average size ¼ 3000 m 2 ). In each test, we tested a dyad, and each subject was tested at least twice with two different partners, in two different enclosures to avoid habituation. The experiment included five tests: 3 min of exploration of the unfamiliar enclosure, 3 min of separation from the partner, 3 min of reunion phase, 3 min of exposure to a novel object and a social threat from a human ('social threat') lasting no longer than 30 s (Fig. 1).
After the first saliva collection (see below), the animals were walked by the animal trainers to the testing area using collar and leash. Upon arrival, both animals were simultaneously released in the testing area which they could freely explore (exploration). After 3 min, the trainers called the animals by name directing them into two smaller enclosures that were at least 10 m apart and physically and visually separated from each other (separation). After 3 min, the animals were again released into the main testing area and could reunite (reunion). After 3 min, both animals were again called by the trainers and directed to enter into a smaller area, thereby allowing the experimenter to place, unseen, the novel object (e.g. a child's plastic castle) in the main testing area. Then the animals were again released into this area and could explore the novel object freely (novel object). The novel object was moved up and down by an experimenter present outside the testing enclosure by means of a rope attached to it. After 3 min, the animals were called into a separate smaller area for the social threat test. Here, the experimenter approached them wearing a costume that covered her face and her whole body. The experimenter and the animals were separated by a fence to avoid direct contact. The experimenter showed up about 10 m away from the fence and approached the dyad making weird movements such as jumping and agitating the arms. She walked until 2 m away from the fence, moved up and down with threatening movements for about 10 s and then went back to the starting position. The test ended when the experimenter was no longer in sight of the subjects. For more detailed description of the experimental procedure see Cimarelli et al. (2019). During the test, the trainers remained near the testing area but did not interact with the animals (except when they needed to move them from one area to the other as the test required).

Behavioural Analysis
The experiment was video recorded using two hand-held cameras and then coded offline using Solomon Coder Beta 15.01.13 (Andr as P eter, http://solomoncoder.com). A second coder, blind to the relationships of the animals and the hypotheses of the study, analysed 20% of the videos to assess interrater reliability (depending on the variable, agreement was good or excellent; intraclass correlation coefficient ranging between 0.60 and 1.00, all P < 0.05). We coded stress-related behaviours, fear-related behaviours during the novel object test and the social threat, escape attempts during the separation phase and the social threat, proximity between the two partners, active proximity seeking, attempts to stay in body contact with the partner during the social threat, attempts to hide behind the partner during the social threat, gaze towards the partner, alternation of gaze between the partner and the novel object or between the partner and the masked experimenter and synchronized movements. See Table 3 for definitions of all coded variables.

Cortisol Analysis
Before the experiment, while the animals were still in their home enclosures, a saliva sample was taken orally using Salivette (Sarstedt, Ges.mbH, Wr. Neudorf, Austria). Food (e.g. cheese) was given to the subjects to increase salivation and acceptance of the swab in the mouth. A second and third saliva samples were taken immediately after the test and also 15 min later. All saliva samples were stored at À20 C until analysis. A cortisol enzyme immunoassay (for details including cross-reactivity of the antibody see Palme & M€ ostl, 1997) was used to analyse cortisol levels. The assay was validated and successfully utilized in a series of previous studies conducted on canines (e.g. Haubenhofer and Kirchengast, 2007;Glenk et al., 2013Glenk et al., , 2014Affenzeller, Palme, & Zulch, 2017).

Statistical Analysis
A PCA on the affiliative behaviours shown by the subjects in their home enclosure during everyday life (body contact, grooming, lying down or standing in a friendly manner, approaching in a friendly manner) was conducted using an Oblimin rotation method. Two affiliation components emerged ('BC (body contact) affiliation' and 'FB (friendly behaviour) affiliation', see below) and were used as predictors in the models.
To investigate whether the behaviours shown during the test or the cortisol levels were influenced by the affiliative and dominance relationships of the subjects and/or the species, we fitted general (LMMs) or generalized linear mixed models (GLMMs, Baayen, 2008), depending on the distribution of the response variable. Considering the imbalance of related versus nonrelated dyads (8 versus 21) we did not investigate whether this variable influenced the response variable of interest, but we included the predictor 'relatedness' (yes versus no) in all models to control for it. Similarly, we included sex (male versus female) or dyadic sex combination (maleemale, femaleefemale, maleefemale) and pack type (whether small (N ¼ 3) or big (N ¼ 4e6 individuals)) in all models to control for them, but we did not consider their significance.
For variables coded in more than one test, we fitted LMMs or GLMMs (depending on the distribution of the response variable) with the identity of the subject and partner, the dyad and the pack

Exploration
Novel object Reunion Separation Social threat  distribution and log link function) shown during the experimental task were included as dependent variables. Explanatory variables included in the model were the two affiliation components ('BC affiliation' and 'FB affiliation', see below), rank distance, dominance status, species, test, sex and relatedness as well as the follo wing three-way interactions: BC affiliation*species*test, FB affiliation*species*test, rank distance*species*test, dominance status-*species*test (and all lower order terms these encompass). For the variable proximity (bidirectional), we conducted an LMM with the proportion of time spent in proximity as the response variable (only one value per dyad) and the same random and fixed effects as the other models (but using sex combination instead of sex), but without including dominance status (and relative interactions). For variables coded only during the social threat test ('hiding behind the partner' and 'start body contact'), we fitted GLMMs (with binomial error distribution) with the identity of the subject and partner, the dyad and the pack as random effects (including all   theoretically identifiable random slopes). Explanatory variables included in the model were the two affiliation components, rank distance, dominance status, species, sex and relatedness as well as the following two-way interactions: BC affiliation *species, FB affiliation*species, rank distance*species, dominance status-*species (and all main effects these encompass).
For cortisol levels, we calculated the delta cortisol between the log-transformed values of the second and the first sample (as a proxy of the stress reactivity to the first four tests) and between the third and the second sample (as a proxy for the stress reactivity to the social threat test). We then conducted LMMs built with the same predictors as the other models (the two affiliation components, rank distance, dominance status, species, sex and relatedness as well as the following two-way interactions: BC affiliation*species, FB affiliation*species, rank distance*species, dominance status*species and all main effects these encompass) with the raw cortisol pretest levels (log transformed) and the delta cortisol between the two samples as response variables.
For all variables, we compared the respective full model with a null model (Forstmeier & Schielzeth, 2011) containing the same random effects and the control variables (sex or sex combination, relatedness, pack type) as the respective full model, and we used a likelihood ratio test (Dobson, 2002) for the fullenull model comparison. We used a likelihood ratio test also to determine the significance of individual effects by comparing likelihoods of models containing, or not, the specific effect (Barr et al., 2013). In all models, we z-transformed all continuous predictors to a mean of zero and a standard deviation of one to obtain more easily interpretable model estimates (Schielzeth, 2010) and ease model convergence. Considering the high complexity of the models given the small sample size, we decided to investigate individual effects in all cases in which the nullefull model comparison revealed a P value lower than 0.1. However, we still considered an individual effect to be significant at the P value threshold of 0.05. Normality and homoscedasticity of residuals of LMMs were ascertained via residual distribution plots. For GLMMs with Poisson distribution, we also assessed overdispersion, which was acceptable. Model stability was determined by removing individual cases one at a time and comparing model estimates obtained from each subset to those obtained for the full data set. Confidence intervals were determined based on parametric bootstrapping (N ¼ 1000 bootstraps; function bootMer of the package lme4, Bates, M€ achler, Bolker, & Walker, 2015). The analyses were conducted using the package lme4 (version 1.1e21, Bates et al., 2015) in R (version 3.6.2, R Core Team, 2019). All models including complete random-effects structure, sample size confidence intervals and stability are reported in the Supplementary Material.

Affiliation Scores
The PCA (KaisereMeyereOlkin, KMO: 0.56, Bartlett's sphericity test: K 2 15 ¼ 203.175, P ¼ 0.00) conducted on the affiliative behaviours recorded during regular observations of spontaneous interactions revealed two components: one based on frequency and time spent in body contact ('BC affiliation') and one based on grooming and other friendly behaviours ('FB affiliation'), explaining 48.66% and 31.30% of the total variance, respectively (see Table 4 for the variable loadings to each component). The two components were not correlated with one another (Pearson r ¼e0.10).

Variables Analysed during the Behavioural Test
We report here the results from the variables for which the nullefull model comparison revealed a significant result. Complete results can be found in the Supplementary Material. All significant results are summarized in Table 5.

Stress reactivity
The experimental task activated a stress response, since most of the tested animals showed an increase in cortisol in the post-test samples in comparison to the pretest ones (79.31%). In both dogs and wolves, increasing FB affiliation (LMM: X 2 1 ¼ 5.02, P ¼ 0.03; Fig. 2) and rank distance (LMM: X 2 1 ¼ 4.07, P ¼ 0.03; Fig. 2) were significantly associated with a smaller difference in cortisol between the two post-test samples.
Similarly, both dogs and wolves were less likely to show stressrelated behaviours when separated from a partner with whom they had higher FB affiliation scores but also more likely to show stressrelated behaviours during the social threat test when with a partner with whom they had higher FB affiliation scores (interaction between FB Affiliation and test: GLMM: X 2 4 ¼ 9.92, P ¼ 0.04; Fig. 3). Similarly, both dogs and wolves tried to escape more often during the separation and the social threat test if they were tested with a partner with whom they had higher scores of BC affiliation (GLMM: X 2 1 ¼ 8.25, P ¼ 0.004; Fig. 4) and who were further away in the hierarchy (GLMM: X 2 1 ¼ 6.15, P ¼ 0.01; Fig. 4). Generally, wolves showed more stress-related behaviours (GLMM: X 2 1 ¼ 5.16, P ¼ 0.02) and tried to escape more often than dogs (GLMM: X 2 1 ¼ 13.12, P ¼ 0.0003), but the two species did not differ in how the various relationship components affected the likelihood of showing stress-related behaviours or the number of escape attempts (all interactions including species with P > 0.05). Overall, both dogs and wolves tried to escape more during the separation than during the social threat test (GLMM: Social support seeking Social support seeking was differently modulated by relationship quality in dogs and wolves. In fact, for dogs, the lower the rank distance, the more time was spent in proximity during the exploration phase while the effect was the opposite during the reunion and the novel object tests. In contrast, for wolves the lower the rank distance, the less time was spent in proximity during the reunion phase (interaction between rank distance, species and test: LMM: X 2 2 ¼ 7.12, P ¼ 0.03; Fig. 5). During the social threat test, we found that the likelihood that dogs started to be in body contact with the partner increased with rank distance, whereas the opposite was true for wolves (interaction between rank distance and species: GLMM: X 2 1 ¼ 4.56, P ¼ 0.03; Fig. 6). However, in both dogs and wolves the higher the BC affiliation score, the more attempts were made to initiate body contact during the social threat test (GLMM: X 2 1 ¼ 10.03, P ¼ 0.002; Fig. 7).

Reference/Coordination
Both dogs and wolves synchronized their movements more often when tested with a highly affiliative partner especially in the exploration phase (interaction between FB affiliation and test: GLMM: X 2 3 ¼ 9.31, P ¼ 0.03; Fig. 8). Generally, dogs synchronized their movements with those of their partners more than wolves (GLMM: X 2 1 ¼ 8.45, P ¼ 0.004) but the two species did not differ in how the various relationship components affected how often they synchronized (all interactions including species with P > 0.05).

DISCUSSION
The aim of the present study was to investigate the role of affiliative and dominance relationship components in shaping wolves' and dogs' reaction to stressful situations. Overall, our results provide evidence that affiliation and rank distance influence the behavioural and physiological responses of dog and wolf dyads facing stressful stimuli, with few differences between the two species. Our results help shed light on the function of these relationship components (and relationships in general) in canines.
While previous studies on dogs and wolves have calculated affiliation using a single measure (e.g. Essler et al., 2017;Marshall-Pescini, Schwarz, et al., 2017), in our sample two uncorrelated components emerged: one based on grooming and other friendly behaviours and one based on body contact. Also for other species, affiliative bonds have been described along different components, which do not always correlate with each other (e.g. proximity remains distinct from other affiliative behaviours such as grooming, Massen et al., 2010). Considering that in the present study these two components mediated different behaviours, we hypothesize that they reflect different animals' motivations and have different functional roles. As predicted, the affiliation component based on friendly behaviours such as grooming was mostly involved in stress regulation, being associated with reduced cortisol level responses and with more frequent synchronized movements. These results suggest that this component might be linked to a stress-management strategy, in which animals coordinate their actions to deal with an environmental stressor together (at least during the exploration and the novel object test). In fact, we could show that both dogs and wolves tested with a partner scoring higher on this component showed a lower stress reactivity (i.e. lower cortisol increase) than subjects tested with a less affiliative partner. Considering that this affiliation component was also associated with movement synchronization, we suggest that synchronized movements might have mediated the effect of affiliation on stress buffering (Cimarelli et al., 2019;Duranton & Gaunet, 2016). These results are in line with previous evidence showing that close affiliative partners are more likely to act together when dealing with external threats (e.g. when defending the territory from intruders, , suggesting a functional role of affiliative relationships during group defence in both dogs and wolves. These results are also in agreement with previous studies conducted on a variety of species confirming the role of affiliation in mediating stress buffering (see Rault, 2012 for a review).
Differently to our predictions (and to previous studies, e.g. Mazzini et al., 2013) we also found that higher scores of affiliation were associated with a lower likelihood of showing stress-related behaviours during the separation phase and with a higher likelihood of showing stress-related behaviours during the social threat test. A higher likelihood of producing stress signals when accompanied by a closer affiliative partner would seem to be contrary to the stress-buffering hypothesis of affiliation (Kiyokawa & Hennessy, 2018). However, this apparent incongruence might be explained by the fact that some stress-related behaviours (e.g. lip licking) might have gained a communicative function, in addition to expressing a negatively aroused internal state (as also suggested by Kaminski, Hynds, Morris, & Waller, 2017). In line with this, considering previous evidence that partners with a stronger affiliative bond are more likely to cooperate in a defensive context , and the other associations we found between affiliation and movement synchronization, it is possible that during the social threat test the function of these behaviours was to better coordinate with a closely bonded affiliative partner.
Based on the idea that wolves, which are cooperative hunters and cooperative breeders, might benefit more from the presence of an affiliative partner than dogs, which are mostly scavengers and mostly do not engage in cooperative breeding (Marshall-Pescini, Cafazzo, et al., 2017), we expected a stronger association between affiliation and stress buffering in wolves than in dogs. Nevertheless, there was no difference between the two species and stress reactivity was dampened in both. These results support previous evidence showing that both dogs and wolves cooperate to defend their territory and they coordinate their movements to deal with potential threats (e.g. intruders: Cassidy et al., 2015;Pal, 2015), and that affiliative bonds with the individuals present affect dogs' choice to engage in or retreat from an intergroup conflict . Hence, we suggest that the function of affiliative relationships in such 'defensive' contexts (i.e. when a potential danger may be present, as in the social threat test) might be conserved across social canid species, and might not have changed in dogs during the process of domestication.
Rank distance was also confirmed to play a role in stress buffering: the bigger the distance to the partner in the dominance hierarchy, the more tempered the physiological stress reaction was in both species. Closeness of rank between individuals might carry uncertainty (less clear who is dominant over whom, De Waal, 1991), thereby reducing the stress-buffering effect of a social partner or even adding to the stress created by the situation. Different to previous studies where rank distance mediated coordinated activities in dogs and wolves differently (with wolves closer in rank being more likely to coordinate and vice versa in dogs, Marshall-Pescini, Schwarz, et al., 2017;Moretti et al., 2015), we did not find that rank distance modulated stress reactivity differently in the two species. However, we found that rank distance potentially affects the link between proximity and stress buffering in opposite ways in dogs and wolves, reflecting different stress-coping strategies, both leading to stress reduction. In dogs, individuals further apart in the hierarchy spent more time in proximity (especially during the reunion and novel object test) and were more likely to look for body contact during the social threat test. This suggests that physical proximity with individuals more distant in rank may help to mitigate stress in dogs. Instead, in wolves, we found that individuals closer in rank were more likely to spend time in proximity during the reunion phase and seek body contact during the social threat test than those more distant in rank. These results seem to contradict the finding that being tested with a partner closer in rank resulted in a higher increase in cortisol, suggesting that the link between rank distance and proximity might not be necessarily related to stress buffering in wolves. In fact, wolves closer in rank were more likely to be in proximity than individuals further away in the hierarchy both in potentially stressful contexts (as in the present study and in Moretti et al., 2015) and in a cooperative task involving food (Marshall-Pescini, Schwarz, et al., 2017). Hence, rank distance in wolves might promote physical proximity in different contexts and, possibly, more generally reflect a form of social monitoring (McNelis et al., 1998), rather than of social support seeking. Therefore, the stressbuffering effect of partners with distant rank is likely to be mediated by mechanisms other than physical closeness. Future studies are needed to clarify the connection between rank distance and physical proximity in wolves, but also to investigate the role of rank distance depending on the individual's dominance status (that is, whether rank distance has a differential effect in subordinate or dominant individuals). The other affiliation component identified here, based on the time spent in body contact during everyday life, was mostly associated with behaviours related to physical proximity during the test: in fact, this component positively influenced the number of escape attempts (especially during the separation test) and body contact seeking during the social threat test. However, we found no clear evidence that this relationship component affected coping with stress. Therefore, the finding that the animals tried to escape separation more often and initiated body contact more frequently when tested with partners with which they had high body contact affiliation scores may reflect their attempts to follow their usual patterns of body contact with their different partners rather than a stress-coping strategy. Future studies will be needed to better comprehend the reasons behind the variation between dyads along this component and its functions. We found no evidence that dominance status affected social stress buffering. We predicted such an effect based on previous studies conducted in the wild showing that subordinate dogs and wolves follow their dominant partners' movement initiations (in relatively relaxed contexts) more than vice versa Peterson et al., 2002). A potential reason for finding no such effect in the short-term stressful contexts applied here might be that dominance status does not have a functional role in stress regulation (at least when induced by environmental stimuli). The only other study in which the influence of dominance status on the behaviour of dogs facing a potential threat was investigated reported mixed evidence in this regard: dominant dogs seemed to be more likely to intervene in intergroup conflicts but without being more likely to occupy frontal positions than low-ranking individuals . Overall, these results suggest that in dogs and wolves the dominance status might be more important for resource management than social buffering (Dale et al., 2017;Essler et al., 2017), but additional studies are needed to test this hypothesis. Interestingly, we found that none of the relationship components analysed here influenced reference behaviours (i.e. gazing towards the partner). Social referencing has been reported to occur in pet dogs when tested with their owners (Cimarelli et al., 2019;Merola et al., 2012) and with conspecifics (although significantly less than with their owners, Cimarelli et al., 2019). Although shown in primates (e.g. Micheletta and Waller, 2012;Shepherd et al., 2006), an effect of affiliation or dominance on gazing towards the partner might not occur in canids. The present study was the first to investigate whether social relationships play a role in modulating such behaviour in dogs and wolves, but the actual situations presented here might have been novel enough to induce looking at each other in both individuals, independently from their relationship (suggesting a ceiling effect). The only studies comparing dogs and wolves' gazing behaviours have focused on the ability to follow gaze in very young animals (Range and Vir anyi, 2011;Werhahn et al., 2016), with no evidence that the relationship with the demonstrator modulates such behaviour. The lack of studies specifically investigating what factors modulate gazing behaviours with conspecifics makes it impossible to draw a conclusion and calls for further investigations.
One could argue that the fact that the single tests were carried out one after the other might have led to a potential accumulation effect (animals being more and more stressed throughout the experiment) which might have overshadowed the effect of the other variables. However, including the variable test in the models allowed us to control for this element. Only one variable, 'escape attempts', was affected by test, and interestingly more escape attempts occurred during the separation test than the last, social threat test. Thus, a stress accumulation effect seems unlikely. Still, further studies are required to investigate whether a longer delay between the single tests would influence the animals' behaviours.
We recommend caution in generalizing the results of the present study to other dog and wolf populations, since both dogs and wolves involved in the present study deal with the same environmental challenges (i.e. same captive setting) and because of the limited sample size. Further studies will need to confirm whether affiliation and dominance similarly regulate how free-ranging animals deal with stressful situations that occur in their natural environment.

Conclusions
Both dogs and wolves' social relationships can be described in terms of formal dominance and affiliation. Importantly, affiliation appears to be represented by two uncorrelated components, one based on friendly behaviours and the other on body contact. Although the two species have different social ecologies, their affiliation components similarly regulate the way in which they deal with stressful situations. We showed a positive link between affiliative relationship components and stress buffering, as well as social support seeking in both species, and a similar regulatory role of rank distance in stress reactivity and social support seeking. Still, a few differences also emerged, suggesting that the rank distance between two individuals regulates social support seeking in dogs and wolves differently: dogs were more likely to look for social support from a partner far away in rank, while the opposite was true in wolves. Taken together, these results suggest that the different social organizations of the two species do not seem to influence how affiliation and dominance affect how the animals cope with stressful situations.

Author Contributions
G.C., S.M.P., F.R. and Z.V. designed the study; G.C. and Z.V. collected the data; G.C. and A.B. analysed the data; all authors interpreted the data. G.C. wrote the first draft of the paper; S.M.P., F.R., A.B. and Z.V. critically revised the manuscript. All authors gave final approval for publication and agree to be held accountable for the work described therein.

Declaration of Interest
The authors declare no competing interests.

Acknowledgments
The Wolf Science Center (WSC) was established by Kurt Kotrschal, Friederike Range and Zs ofia Vir anyi, and we thank all the helpers who made this possible hence indirectly supporting this research. Additionally, we thank all the staff and students of the WSC for their help with the tests and care of the animals. We further thank many private sponsors, including Royal Canin, for financial support and the Game Park Ernstbrunn for hosting the WSC. We are grateful to Roger Mundry for his help with the statistical analysis, Julija Podbev sek for video coding and Martina Lazzaroni for helpful discussions. We also thank Professor Rupert Palme, Samy El Marakem and the Unit of Physiology, Pathophysiology and Experimental Endocrinology (Department of Biomedical Sciences, University of Veterinary Medicine, Vienna, Austria) for the cortisol analyses. This work was supported by the Austrian Science Fund (FWF) project I 1271-B24 and 30704-B29 and the Hungarian Scientific Research Fund project OTKA-ANN 107726.