Separation-related behavior of dogs shows association with their reactions to everyday situations that may elicit frustration or fear

Separation related disorder in dogs is a multi-faceted phenomenon. Dogs can react to the absence of their owner due to different inner states such as fear, panic or frustration. We hypothesized that dogs that are prone to frustration or fearfulness in other contexts would show a different behavioral response to separation from the owner. We investigated the association between inner states in different contexts and separation behaviors by combining a questionnaire with a separation test. Fear-related questionnaire components were rather associated with whining and the absence of barking. Dogs that received higher scores in the demanding component of the questionnaire, which might be in association of the frustration threshold of the dog, barked more and were more likely to scratch the door. Finally, dogs that were more prone to phobic reactions whined somewhat more and tried to escape. We provide empirical support for the assumption that separation-related behavioral responses of dogs might be triggered by different emotions.

Results of the PCA analysis of the behavioral variables from the separation test. As a result of the analysis of the behavior test five main components emerged (see Table 2). The components together explained 56.3% of the total variance. All were reliable when based on Cronbach's alpha values. The first component ('Chair' , Cronbach's alpha: 0.7) consisted of four variables associated to the owner's chair in the room. Results of the generalized linear models. ' Fig. 4). We did not find any significant result in the case of rearing on the door.

Discussion
Interest has been growing in the multi-faceted etiology and the possible identification of different sub-types of separation-related behavior problems in companion dogs. Our core hypothesis was that the variability of the diverse signs-at least partially-can be explained by the different underlying affective states. Though it is a relationship between two adult individuals, the human-dog social bond is thought to be analogous to filial attachment 1 . As humans provide resources to the dog-same as parents provide it to their offspring-the dog is dependent on humans and motivated to stay close to its owner. This motivation manifests itself as a stress response in the absence of the owner. However, these stress related behaviors can be various in appearance and  www.nature.com/scientificreports/ intensity in isolation; fear, anxiety, panic and also frustration might appear. While there are several questionnaire studies about the mutual occurrence of different behavior problems-such as fearfulness or phobias-and SRP, and there are also some theoretical works about the possible inner states in the background, by our knowledge, our study is the first to combine a questionnaire with a behavior test which provides evidence of the association between different behavioral responses and presumed inner states in the absence of the owner. Dogs that can be characterized with fearful tendencies based on their questionnaire-based scores, whined sooner and more often, but did not bark during the behavior test. We also found that dogs that scored lower on the 'Relaxed' component showed a tendency to start to bark sooner than the more 'Relaxed' dogs-however, the apparently relaxed behavior can be equally typical of the less fearful and less frustrated dogs as well. 'Relaxed' dogs also show less escape attempts such as door scratching or rearing. Contrary to this, dogs that received higher scores in the 'Demanding' component of the questionnaire, can be characterized with frequent and early onset barking and they also more likely and sooner scratched the door. Dogs that were told to be prone to demanding behaviors in other situations, actively and more intensively tried to get out of the room.
Items of the 'Demanding' component contain different contexts where the dog shows demanding behaviors like access for food, treats, playing or the company of the owner. When the expectation (including the absence or the delay of a reward) of an individual is violated frustration might appear 32 , thus it is reasonable to assume that those individuals who gained high scores on this component would also become easily frustrated if their goal is not fulfilled despite their persistence of demanding it. Therefore, frustration and persistence are strongly connected to each other as the failure of reaching the goal leads to frustration what prompts persistent efforts to pursue it 33 . Indeed, McPeake and colleagues 23 in their extended questionnaire about frustration-related behaviors in dogs described a component called 'Barrier frustration/Perseverance' . The fact that during their analysis questions connected to persistence and the intensive reaction when a barrier thwarted the goal of the dog formed a component is particularly interesting, as typically during separation there is a physical barrier between the dog and the owner. These results together are in line with the assumption that these dogs' behavior might be motivated by their inability to access their owner who represents an important resource.
Based on previous questionnaire studies, it has been acknowledged that there is a co-occurrence between various noise phobias and SRP 21,34 . To our knowledge our results are the first where signs of SRP were assessed in an experimental setup and a direct relationship was found between them and various, owner-reported phobic behaviors of the dogs. Interestingly, dogs that have owner-reported phobias ('Have phobias') gained somewhat higher scores in case of the 'Whining-door' component and they also can be characterized with the absence of barking (negative association with 'Bark-Wagging' component), furthermore, they gained higher scores in the 'Escape' component. This result is in agreement with the assumption that these dogs show a different, intense panic-like reaction to separation. Unlike 21 and 34 Blackwell et al. 35 found no direct association between phobias of extreme noise effects (such as thunderstorms and fireworks) and SRP, but described a connection between fear of other generic noises (that can cause startling effect in dogs) and separation-related behavioral problems. In our case, the 'Have phobias' component consists of items not only related to noise phobias but more generic fearful reactions of the dogs to startling stimuli. Thus, in agreement with 35 , our results provide support to the theory that dogs that are more sensitive to fear-eliciting effects would also show a specific behavioral response to separation. Here, the assumed connection with fear can be further supported with the result that these dogs tended to whine more but they seldom barked.
Affective states are usually characterized along two continuums. They are valenced either positively or negatively (valence), and they generate different levels of arousal (intensity). Not surprisingly, both frustration and www.nature.com/scientificreports/ fear are negative emotions, where fear and panic are thought to be higher arousal states than frustration 36 . The presumed inner states during separation are close to each other in these dimensions, particularly in the case of fear and panic where it is also known that they have a similar underlying neural mechanism 20 . While our results strengthen the theory that the cumulative experience of particular inner states also affects the reaction to separation, it is important to emphasize that naturally these mental states are not necessarily discrete, and they might overlap or change over time. Lund and Jørgensen 24 in their model proposed the possible relation and time course of the different factors and inner states influencing separation behavior. According to their study, frustration during separation leads to barking and destructive behavior, while fear provokes general signs of distress and escape behavior. Our results partly support these, however we found that escape behaviors were rather associated with frustration and panic than with fear. It is possible that in some cases dogs show escape behaviors because they just try to leave the confined space thus their behavior response is indeed provoked by fear 37 . However, it is more likely that they rather try reestablishing the contact with their owner, thus their escape attempts might be motivated by frustration. It is also possible that Lund and Jørgensen took the frustration-related damaging of the door as 'destructive' behavior, whilst in our behavior test these were labelled as escape-related reactions. They also suggested that the appearance of particular behaviors during isolation are caused by the change in the dog's emotional states (i.e. a dog at first can feel frustration, then after a while fear). Contrary to their long-term investigation in the home environment, our test lasted only 3 min and still the signs of fear and frustration could be detected in different dogs. This highlights the possibility that although dogs may experience a dynamic change in their inner state during a longer period of separation, their individual propensity for an almost immediate fear-or frustration-driven reaction can result in very different reactions to even a short interval of isolation. As we expected, all but one of the factors that emerged from the questionnaire were somehow associated with vocalizations. Vocalization is one of the most often mentioned signs of separation stress, with a strong association between particular acoustic features of the vocalizations emitted by isolated individuals and the possible inner state of the subject 26,38,39 . In other canid species-either pups or adult individuals-whining or howling is what mostly appears instead of barking [40][41][42] . Barking is a vocalization type that became abundant and acoustically versatile during the evolution of the dog 27 . It is known that it carries contextual and affective information to humans and it is emitted in several different contexts 43,44 . It was found that the particular type of barks emitted in isolation ('left alone barks') is especially easy to recognize and it is mostly characterized by human listeners as being 'desperate' and 'fearful' 44 . Based on these findings, both vocalization types have a different role during interspecific communication between dogs and humans, thus during separation they are probably emitted with an adaptive outcome by potentially changing the owner's behavior. Barking is a long-range vocalization type, thus in this case its function might be to capture the attention of the owner and it is provoked by the unpleasant frustrating situation 39 . The whining of dogs (being a short-range vocalization) might rather elicit caring behavior from the receivers once they reappear at the scene 41 .
Contrary to other studies, here we did not compare dogs with or without predetermined SRP-status, but instead we wanted to describe their reaction to separation from the owner with behavioral scores along various scales. With this method we tried to avoid the subjectivity of the owner's report about his/her dog's condition 11 . It is possible that when strong separation stress truly manifests in problematic behavior at home, this type of SRP would be mostly caused by fear. Some level of frustration is a normal reaction in most dogs to separation at any location, but it is likely that being separated from the owner at an unfamiliar place would elicit also fear from the dogs. This can result in more whining (response to fear) during the tests from those dogs that usually would rather bark at home (as being mostly frustrated).
Here we used a merged questionnaire of previous studies 12,45,46 and we also developed new questions about frustration-related behaviors of the dogs. Compared to our previous questionnaire 12 with the inclusion of 2 new frustration related questions a combined 'Demanding' component emerged that included food related behaviors, while in our previous sample these questions formed a separate component. This is not surprising as inaccessible food triggers frustration-related behaviors in dogs 47 . Meanwhile, contrary to Temesi and colleagues 45 , in our sample the questions related to phobic behaviors formed a separate group. Thus, instead of a general 'Neuroticism' component, in our case we had two independent components ('Relaxed' and 'Have phobias'). This is also in line with the assumption that there is an overlap between phobic behaviors and other fear-related problems, however these do not necessarily co-occur 21 . However, it is also possible that they are only the consequence of the slightly different statistical methods.
Interestingly, in our previous study we did not find a relationship between the behavior during an outdoor separation test and the dog's obedience 12 . However, here we found that dogs that gained higher scores in the 'Obedient' factor spent more time at the chair of the owner and they started to scratch the door later. On one hand it is very likely that these owners spend more time with their dogs thus they have a more balanced relationship with them, or these dogs are more trained which has a known beneficial effect on problematic behaviors 48,49 . These dogs stayed where their owner had spent his/her time before he/she left instead of standing at the door, thus they probably were less stressed. Alternatively, even though the owner did not give any command or verbal cue, the dog might have perceived the test situation as a task or exercise. However, in the outdoor separation test the dogs were tethered to a tree meanwhile their owner left them alone-this is a more stressful situation as the tree cannot be regarded as any form of 'secure base' , unlike the owner's chair with his/her odor on it in the indoor test.
Among the limitations of this study, we can mention that the assessment of frustration and fear-related behaviors of the subjects was done with the help of a questionnaire. Although dog owners can be considered as the closest observers of the behavior of their canine companions, they are untrained and can be often subjective source of information. At the time of our data collection there was no widely used method to describe the frustration tendencies of dogs, thus our questionnaire might not cover every aspect of it. Since then, McPeake and colleagues 23  www.nature.com/scientificreports/ future step, along with testing dogs' separation-related behavioral responses, one could try assessing the subjects' reactions to such stimuli that have the potential to elicit either fear, or frustration. This would provide a more empirical approach to test the separation-related reactions in subjects that are more prone to fear or frustration. Our results might be particularly relevant from the aspect of animal welfare, and agreeing with de Assis et al. 16 , we emphasize the importance of the recognition of the heterogeneity of this behavioral problem. Either in the case of designing empirical research, or during the development of individualized treatment strategies, one should include detailed mapping of the individual dog's behavior and the background of it, specifically the description of the prevalence of other contexts where the dog might experience similar affective states as during separation.

Materials and methods
Ethical statement. Owners of the dogs were informed about the goals and circumstances of the experimental procedure a priori and they were present during the tests. We informed them that they could interrupt the experiment and reconsider their participation if-by their judgement-the test was too stressful for the dog. Their informed consent was obtained in written form via filling and signing the Department of Ethology's standard consent form. The tests were performed in accordance with the Hungarian regulations on animal experimentation and the Guidelines for the use of animals in research described by the Association for the Study Animal Behaviour (ASAB) and ARRIVE.
All experimental protocols were approved by the Animal Welfare Committee of the Eötvös Loránd University and the National Scientific Ethical Committee on Animal Experimentation (Ref. no.: PEI/001/1056-4/2015). Informed consent was obtained from all subjects (human) for their involvement in study. Additionally informed consent was obtained from the dog owners to involve their dogs in the study.

Human ethics. This research was approved by the National Research Ethics Committee (PE/EA/55-4/2019)
and was carried out in accordance with the Declaration of Helsinki. Informed consent was obtained from all dog owners who completed our online questionnaire with an Institutional Review Board-approved protocol.
Questionnaire. The questions (see Supplement) were taken from the already published questionnaires of Temesi et al. 45 , Bálint et al. 46 and Lenkei et al. 12 . The questions related to the fearfulness were based on the consensus questionnaire of Temesi and colleagues who synthetized previous studies about different aspects of fear in dogs [50][51][52][53][54] . Besides the basic demographic information of the dog and the owner, the questionnaire contained questions about the occurrence and symptoms of SRD, and also about other possible problematic behaviors of the dog. The second section of the questionnaire contained questions about situations where the dog might act fearful (e.g.: Dog acts anxiously or fearfully when approached directly by an unfamiliar dog of smaller size) or show frustrated behavior (e.g.: If the dog wants to obtain something, it pursues persistently or even aggressively). In the case of these questions, the owner's responses were measured on a Likert-scale from 1 ("not typical at all"/"never happens") to 5 ("completely typical"/"happens all the time"). The questionnaire was available for the participants online and it was advertised via social media. We recorded a total of N = 397 entries to the questionnaire, from which N = 392 were used for the analysis (N = 5 entries were removed because of duplicate participation). We invited owners to the behavior test who indicated in the questionnaire that they would willingly participate in such events.
Behavior test (based on Konok et al. 31 ). Subjects. We tested N = 66 subjects, however N = 10 owners did not complete the questionnaire, therefore we included only those to the analysis that had both the behavioral and questionnaire data (N = 56). The subjects were more than 1 year old family dogs (mean age in years ± SD: 6.3 ± 3, sex ratio: N = 27 males, N = 29 females; see Table 3). In the sample 13 dogs had owner reported SRD that was based on a Yes/No question in the questionnaire. The methods of the behavior tests were accepted by the Animal Welfare Committee of the Eötvös Loránd University (Ref. no.: PEI/001/1056-4/2015). The owners were informed about the aims and the methods of the experiments and that they were allowed to interrupt the tests anytime they feel the situation was too stressful for their dog.
Procedure. Dogs were tested in a room (6.27 m × 5.40 m) that was empty except for a chair. The owner entered the room with the dog on leash. After taking off the leash the dog was free to explore the room during the whole test. At first the owner sat on the chair and did not initiate any interaction with the dog. After 1 min elapsed (measured with a stopwatch by the owner), he/she left the room without any interaction with the dog, leaving the dog's leash on the chair. The dog was alone in the room for 3 min while its behavior was recorded with a digital camera system and two microphones. After the 3 min elapsed the owner returned, greeted and petted the dog. We only used for behavior coding the interval that started when the owner closed the door after leaving the room and finished when he/she opened it again.
Data analysis. The behavior coding of the videos was performed by Solomon Coder (beta 17.03.22 copyright by András Péter). Table 4 shows the coded behaviors. To check the reliability of the coding method an independent observer coded 20% of randomly chosen videos. We calculated Cohens's Kappa statistics for each behavior's variables by taking 20% random samples. This procedure was repeated 100 times and we averaged the calculated values (Vocalization: k = 0.7; Distance: k = 0.8; Orientation: k = 0.664; Position: k = 0.881; Exploration: k = 0.645; Rearing: k = 0.835; Tail-wagging: k = 0.728; Scratching: k = 0.795). We also calculated an overall mean Kappa value (k = 0.76) indicating substantial agreement.  26 full sample of the analysis: N = 167). For the questionnaire, as the questions were not continuous but Likert scales, we used polychoric correlation matrix to calculate the PCA. We ran parallel analysis (paran) in each cycle to re-determine the number of the extracted components 58 . For each factor, we calculated Cronbach's alpha values. Based on the components we calculated behavior scores for each subject for further analysis. Later using the calculated factor scores we ran separate General Linear Models. In case of behaviors where normal distribution was found (whine_door, chair) and where boxcox based power transformation was applied (bark_door) we used General Linear Models (lm function), while in cases where the boxcox analysis suggested log transformation (esc, sit) we applied Generalized Linear Models (glm function) with Gaussian distribution and log link 59 . To analyze the latencies (the first occurrence of whining, barking, scratching the door and rearing at the door) Cox-regressions (survival package 60,61 ) to compare the behavior during the separation with the results of the questionnaire. For testing the normality, we ran Shapiro-Wilk test and we used logarithmic or box-cox transformation where it was necessary. We performed backwards model selection based on p values. Here we report the results of the final models.

Data availability
Data will be available upon request.   www.nature.com/scientificreports/