**Contents**



## **About the Editor**

**Betty McGuire** is a Senior Lecturer in the Department of Ecology and Evolutionary Biology at Cornell University, Ithaca, NY, USA. She received her undergraduate training at Pennsylvania State University, and completed her doctoral degree at the University of Massachusetts Amherst. For many years, she studied social behavior, reproduction, and ecology of small mammals, including voles, rats, and shrews, using both laboratory and field approaches. More recently, her research has focused on the behavior of shelter dogs, particularly scent-marking and food-guarding behaviors, and human–dog interactions. She has been recognized for training research undergraduates, with whom she frequently publishes. Having co-authored textbooks in human biology and animal behavior, she is currently collaborating on a vertebrate biology text. She is an Editorial Board Member for *Animals*.

## **Preface to "Behavior of Shelter Animals"**

Each year animal shelters receive, care for, and rehome dogs, cats, and other companion animals. Even under the best of circumstances, however, shelters can be challenging environments for the animals admitted. Often, animals are housed in close proximity in unfamiliar locations, sometimes with limited space and, typically, high noise levels. Additionally, shelter animals interact with many unfamiliar people and experience a lack of predictability and control in their daily lives. Some animals enter shelters with behavioral problems, while others may develop problematic behaviors as a result of shelter experiences. Collaborative research programs between academic institutions and animal shelters have provided new data on the welfare of shelter animals. Data are now available on how shelter environments impact the welfare of resident animals, and the effectiveness of shelter programs in reducing stress and providing enrichment. Studies also address whether shelter behavioral evaluations are useful tools for predicting behavior and assessing adoptability. This Special Issue focuses on the behavior of dogs and cats while they are either in shelter environments or adoptive homes. The eleven papers focus on stress and behaviors associated with stress; the effectiveness of shelter enrichment programs in reducing stress; the usefulness of shelter behavioral evaluations in predicting behavior in other contexts; and human–dog interactions. The goal is to provide information that will inform shelter programs and policies, and thereby improve the welfare of shelter animals.

> **Betty McGuire** *Editor*

*Review*

## **Psychological Stress, Its Reduction, and Long-Term Consequences: What Studies with Laboratory Animals Might Teach Us about Life in the Dog Shelter**

## **Michael B. Hennessy 1,\*, Regina M. Willen 2,\* and Patricia A. Schiml <sup>1</sup>**


Received: 12 October 2020; Accepted: 4 November 2020; Published: 7 November 2020

**Simple Summary:** Experiments in laboratory animals have provided the basis for studies of stress, its reduction, and its long-term consequences in shelter dogs. Stressors often used in laboratory experiments, such as uncontrollable noise and novelty, are also inherent in shelters where they produce similar physiological reactions, including elevations of circulating levels of glucocorticoid stress hormones. We review how experiments demonstrating a social partner can reduce glucocorticoid responses in the laboratory guided studies showing that human interaction can have similar positive effects on shelter dogs. We also describe recent work in which human interaction in a calming environment reduced aggressive responses of fearful shelter dogs in a temperament test used to determine suitability for adoption. Finally, we present evidence from the laboratory that stress can produce long-term effects on behavior (e.g., reduced socio-positive behavior) that may be due to glucocorticoids or other factors, and which may not occur until long after initial stress exposure. We suggest that the possibility of similar effects occurring in shelter dogs is a question deserving further study.

**Abstract:** There is a long history of laboratory studies of the physiological and behavioral effects of stress, its reduction, and the later psychological and behavioral consequences of unmitigated stress responses. Many of the stressors employed in these studies approximate the experience of dogs confined in an animal shelter. We review how the laboratory literature has guided our own work in describing the reactions of dogs to shelter housing and in helping formulate means of reducing their stress responses. Consistent with the social buffering literature in other species, human interaction has emerged as a key ingredient in moderating glucocorticoid stress responses of shelter dogs. We discuss variables that appear critical for effective use of human interaction procedures in the shelter as well as potential neural mechanisms underlying the glucocorticoid-reducing effect. We also describe recent studies in which enrichment centered on human interaction has been found to reduce aggressive responses in a temperament test used to determine suitability for adoption. Finally, we suggest that a critical aspect of the laboratory stress literature that has been underappreciated in studying shelter dogs is evidence for long-term behavioral consequences—often mediated by glucocorticoids—that may not become apparent until well after initial stress exposure.

**Keywords:** shelter dog; stress; hypothalamic–pituitary–adrenal; cortisol; glucocorticoid; social buffering; enrichment; early-life stress; individual differences; animal welfare

#### **1. Introduction**

There is a vast literature documenting the consequences of psychological stress in laboratory animals. Many of these studies are translational in that they use rats, mice, or other species as models

to provide insight into how stress exposure in humans can impair emotional wellbeing and promote the development of mental as well as physical disorders. A common procedure is to expose animals to one or more stressors that are uncontrollable and often unpredictable. These may include isolation, noise, separation from companions, and confinement or restraint. These manipulations are often found to increase behaviors thought to share processes underlying human psychopathology such as anxiety, depression, or post-traumatic stress disorder, as well as to reduce some cognitive abilities [1–7]. Even limiting the amount of nesting material provided to lactating female rats, which then disrupts the treatment the females provide their litters, has multiple negative outcomes on the offspring at later ages [8].

Now consider the experience of a dog (or lactating bitch and pups) that suddenly find themselves confined in an animal shelter. While housing conditions vary substantially across shelters, stressors like those that contribute to serious deleterious consequences for laboratory animals are still unavoidably inherent to some degree across shelter environments. From the perspective of a laboratory that has split its efforts between basic and translational studies of psychological stress with laboratory animals and studies to measure and reduce stress and its effects in shelter dogs, the similarities between shelter conditions and laboratory paradigms designed to induce adverse emotional and behavioral outcomes are impossible to dismiss. Both shelter housing and laboratory stress paradigms induce physiological stress responses, perhaps most importantly, though not exclusively, activation of the hypothalamic–pituitary–adrenal (HPA) axis, e.g., [6,9]. Not only can HPA activation be taken as a sign of stress in shelter dogs, but the repeated or prolonged activation of the HPA system, and particularly of the glucocorticoid hormones that are the endpoint of the HPA response, may serve as a mechanism underlying many of the long-term effects of prior psychological stress [10,11].

Fortunately, basic research has also suggested means by which the impact of stressors can be minimized. Prominent among these is the process referred to as "social buffering", or the ability of a companion to moderate physiological stress responses. Social buffering is both a basic phenomenon we have studied over the years as well as a strategy we have used to attempt to improve welfare and preclude later adverse consequences of shelter confinement. In the remaining portions of this paper, we will view our work with dogs and related studies by other investigators within the broader context of basic and translational research on stress and social buffering.

#### **2. HPA Responses and Social Bu**ff**ering in the Shelter**

#### *2.1. How Stressors Like Those in the Shelter A*ff*ect HPA Activity*

It has long been known that the HPA system is especially sensitive to psychogenic stressors, that is stressors that pose no actual physical harm [12]. Events that are novel, unpredictable, or out of an individual's control indicate that the current situation is not fully understood and can suggest that harm is likely forthcoming. How this perception activates the HPA axis is complex and still not fully understood, see [13]. However, in brief, cortical regions detecting psychogenic stressors (e.g., medial prefrontal cortex—mPFC) together with associated limbic and brain stem regions including amygdala nuclei, portions of the bed nucleus of the stria terminalis and hippocampus, as well as the nucleus of the solitary tract, relay neural signals to the region of the paraventricular nucleus of the hypothalamus (PVN). Here, they excite (or block inhibition of) neurons containing arginine vasopressin and especially corticotropin-releasing hormone (CRH), which are then released into the hypothalamic–hypophyseal portal system. Upon reaching the anterior pituitary, these peptides bind receptors to spur release of adrenocorticotropin hormone (ACTH) into the general circulation. In several minutes, ACTH reaches the adrenal cortex to trigger synthesis and release of glucocorticoid hormones, notably cortisol and corticosterone. Glucocorticoids bind with Type 1 (MR) and (particularly during stress) Type 2 (GR) receptors throughout the body, including the brain, to induce multitudinous actions on target tissues. Activation of GR receptors in the hippocampus and elsewhere also mediates negative feedback to suppress HPA activation as the stressor passes. In addition, just as some cortico-limbic inputs excite

the PVN, others (from, e.g., prefrontal cortex (PFC), hippocampus) actively suppress activity of CRH and vasopressin neurons. In principle then, there are many potential routes by which social buffering can inhibit HPA activation, either by inhibiting excitatory inputs or by exciting inhibitory inputs to the PVN. Investigation of stimuli that induce HPA activation began with Selye [14]. Although social buffering is now a widely accepted concept across a number of species [15–17], investigation of the ability of social partners to reduce stress developed decades after Selye's original work.

#### *2.2. Brief History of Social Bu*ff*ering in Nonhuman Primates*

The first publication addressing social buffering of HPA activity appears to have been Hill et al.'s [18] study of surrogate-reared Old-World rhesus macaque monkeys during the first year of life. When these infants were removed from the home cage and placed into a novel cage for an hour, they had higher plasma cortisol concentrations when alone than when accompanied by their rearing surrogate. That is, the presence of the artificial mother appeared to buffer the response of the HPA axis to disturbance and exposure to novelty. If a surrogate mother could buffer cortisol responses of infants, then one would certainly expect an actual mother to be effective as well. Indeed, later studies confirmed that cortisol elevations of rhesus infants that had been handled or handled and exposed to novel surroundings were significantly reduced when in the presence of their biological mother as compared to when they were alone [19,20]. This effect of the mother's presence was not restricted to just rhesus or other Old World monkeys. In New World squirrel monkeys, both infants [21] as well as their mothers [22] had higher plasma cortisol concentrations following disturbance if tested without the other member of the dyad than if mother and infant remained together. On the other hand, not all affiliative social companions appeared capable of buffering HPA responses. In squirrel monkey troops, some of the most amicable interactions occur among juvenile peers which avidly engage in play, and in adult females which spend much time in close proximity with one another. Yet, juveniles did not reduce the cortisol response of familiar juveniles, and adult females did not reduce the cortisol response of familiar, even preferred, adult females [23,24].

Together these results suggested there was something special about the mother–infant relationship that enabled the partners to buffer each other's stress responses. The most obvious possibility was simply that the degree of social connection between partners was critical; specifically, that the intensity of the relationship between mother and infant was greater than the affiliation among other friendly partners. Notably, the data did not allow one to exclude other unspecified attributes that might be characteristic of only mothers and infants. However, a second New World primate, the titi monkey, provided some insight. Unlike squirrel monkeys, titi monkeys are monogamous, with adults typically spending long periods of time in quiet contact with their pair-mate. Additionally, whereas young squirrel monkeys ride on the back of their mothers, and never their fathers, titi infants ride on the backs of both parents, especially the father [25]. Moreover, in preference tests, the mother as well as the father more often chose to be near each other than to be near their infant, and infants, in turn, preferred being near their father rather than their mother [25]. This very unusual pattern of familial preferences (Table 1) permitted experimental dissociation of social attraction or relationship intensity from other characteristics specific to the mother–infant relationship. For these experiments, entire family groups were captured and then placed back into the home cage either alone or with a specific partner(s). While all family members showed a pronounced elevation of cortisol levels following capture when returned to the home cage alone, adult males and females showed significant reductions in cortisol only when returned with each other, i.e., their adult pair-mate, and not when returned with their infant [25]. Infants displayed a reduction in cortisol concentrations when returned only with their mothers, and a further significant reduction to the level seen in undisturbed infants when returned only with their fathers [26]. Thus, the likelihood of social buffering occurring corresponded perfectly to the strength of the social attraction between the partners (Table 1). Since the time these early experiments were conducted there have been numerous demonstrations of buffering in other species, some of which involve partners with no prior social relationship whatsoever, e.g., pairs of unfamiliar adult male

rats; [27,28]. Yet, the intensity of the positive relationship between partners remains the best predictor of whether an individual will buffer the glucocorticoid response of a companion [17,28,29].


**Table 1.** Relative preference for, and buffering by, specific titi monkey family members.

Data from Mendoza and Mason [25] and Hoffman et al. [26].

#### *2.3. Social Bu*ff*ering of HPA Responses in Dogs*

With this general principle of the importance of the strength of the relationship in mind, our first study of social buffering in dogs compared the ability of a long-term conspecific kennelmate and the human caretaker in reducing glucocorticoid elevations in adult dogs. The dogs (7–9 years old), which had been maintained in littermate pairs continuously since ~8 weeks of age, were examined in a novel environment either alone, with their kennelmate, or with their life-long human caretaker. Although we had anticipated that the caretaker might have some effect, we were nevertheless struck by the differential influence of the two companions. Whereas, the passive presence of a dog's human caretaker reduced the plasma glucocorticoid elevation to the novel environment, the dog's sibling and long-term kennelmate was without effect [30]. This finding certainly seemed to speak to the affinity that dogs have evolved for humans over thousands of generations. In addition, the results documented a tangible effect of human presence on the stress physiology of dogs that might then be leveraged to reduce stress and, therefore, improve welfare of dogs confined in shelters.

The first step, however, was to determine how a stay in an animal shelter affected glucocorticoid levels. The initial experiments confirmed what was expected: The psychological stressors encountered upon entering an animal shelter powerfully activated the HPA axis. Circulating cortisol levels were nearly three times as high as those of pet dogs sampled in their home and remained so for 3 days before gradually waning [31]. Subsequent work indicated that cortisol concentrations did not decline to levels like those of pet dogs sampled under resting conditions until sometime after 10 days [9]. Work from other laboratories has generally found cortisol levels to be elevated at least about this length of time or longer [32–37], though variability among individual dogs is common, e.g., [35,37].

In our first attempt to buffer this response [31], dogs were taken from their kennels and had a blood sample collected. They were either petted or returned to their kennel for 20 min, and then a second blood sample was collected to estimate the effect of the petting. Initial results were disappointing, showing that petting had no overall effect. However, when in a follow-up analysis dogs in the petting group were partitioned based on the sex of the individual doing the petting, those dogs petted by a man showed an increase from the first to the second blood sample, while those petted by a woman showed no change across samples [31]. It appeared, therefore, that interaction with a woman prevented the sampling procedure required to collect the first blood sample from elevating cortisol levels obtained in the second sample, but interaction with a man had no buffering effect. This differential influence was determined to depend on the nature of the petting administered by men versus women. When men were trained to pet in the more soothing and quiet manner characteristic of the women, the men were as effective as the women in preventing the initial blood sampling procedure from elevating cortisol levels in the second sample [38]. This was the second lesson we learned about social buffering in dogs: Not only are humans particularly effective in buffering the glucocorticoid response of dogs, but seemingly subtle differences in how the human interacts with the dog can determine whether or not the HPA response is reduced. Still, however, we were only effective in reducing the response of the dogs to an additional minor stressor—the initial blood sampling—rather than mitigating the response to the shelter itself.

After a series of further unsuccessful attempts to reduce the glucocorticoid response to shelter housing, we came upon what appeared to be a third lesson about social buffering in shelter dogs; that is, in addition to how you interact, the location where you interact also is critical. In this study, we were able to secure a quiet, secluded room in the rear of the shelter that was farther away from the commotion of the housing and public areas than in any of our previous studies. Here, we found that it did not matter if a person petted, played with, or passively sat near the dogs. In all cases, plasma cortisol levels were reliably reduced when a person (woman) was present [39]. If the dog was simply isolated in the secluded room, there was no reduction in cortisol levels. In other words, the secluded room alone had no effect, but a person in the room, even sitting quietly with the dog, suppressed the cortisol response to shelter housing. Ours was not the first laboratory to find human interaction to reduce the cortisol response to shelter conditions. During the time of our unsuccessful attempts, two other laboratories had found human interaction to such buffering effects [33,36]. The forms of human interaction in these studies were more complex, involving a variety of activities in different locations, but both included some time outdoors, removed from the commotion of the shelter. In all, these studies document how social buffering in the form of human interaction can readily mitigate the physiological stress response imposed by inherent features of shelter housing. Yet, this strategy has clear limits in that the effect is temporary. When dogs are returned to their kennel, cortisol levels elevate to their pre-interaction levels within an hour [40]. A recent approach that greatly prolongs the beneficial effect of interaction is to foster shelter dogs to a private home for a night or two. In shelters in which this procedure is implemented, urinary cortisol levels are reduced throughout the fostering period, though here too cortisol concentrations elevate to pre-interaction levels when returned to the shelter environment [41].

#### *2.4. Mechanism of Social Bu*ff*ering of HPA Responses*

Oxytocin appears to be the most likely mediator of social buffering of dogs' HPA response by human interaction. Release of oxytocin both stimulates, and is stimulated by, engaging social behaviors such as gentle touch and prolonged gazing [42,43]. Furthermore, while oxytocin's influence is much more complex than simply enhancing sociability, there is a wealth of data on how oxytocin can promote socio-positive or bonding-related behaviors [43–45], including those of dogs with humans or other dogs [46,47]. These effects often may be due to oxytocin reducing anxiety or wariness to engage in social activity [48]. Oxytocin can also reduce HPA activity more directly by, for instance, inhibiting excitatory input to the PVN or via inhibitory GABA interneurons connecting oxytocin neurons to corticotropin releasing hormone cells in the PVN [43,49]. Indeed, in the monogamous prairie vole, the ability of an adult male to buffer the HPA response of his female partner is inhibited by pharmacologically blocking oxytocin receptors in the PVN [50] (Figure 1, top).

**Figure 1.** Summary of findings regarding neural circuits underlying social buffering. (**Top**) Presumed neural mechanisms underlying social buffering by mates in female prairie voles. (**Middle**) Possible neural mechanisms underlying social buffering in rodent pups. (**Bottom**) Possible neural mechanisms underlying social buffering by adult conspecifics other than mother and mates. Solid and dashed lines represent pathways proposed in each experimental model. However, the pathways do not necessarily imply direct anatomical connections. Hypothetical buffering pathways are marked by asterisks. AOP, posterior complex of the anterior olfactory nucleus; CORT, cortisol or corticosterone; CRH, corticotropic releasing hormone; LA, lateral amygdala; LRN, lateral reticular nucleus; MOB, main olfactory bulb; NE, norepinephrine; NTS, nucleus of the solitary tract; OXT, oxytocin; PI, prelimbic cortex; PVN, paraventricular nucleus of the hypothalamus; VP, vasopressin. Figure redrawn from [28].

However, studies in laboratory rodents have identified a number of other potential mediators that could act independent of, or in conjunction with, oxytocin. For lactating rats, evidence indicates that the mother's buffering of HPA activity of her pups is due to inhibition of excitatory noradrenergic input to the PVN from brainstem [51]. In guinea pig pups, both the mother and an unfamiliar male can buffer HPA responses and do so through different mechanisms. The presence of the mother, even when anesthetized, reduces pups' cortisol response during exposure to novelty [52] quite possibly again by inhibiting noradrenergic input [53]. In contrast, adult males reduce pups' cortisol response in a novel environment when the male is awake and actively engaging the pup, but not, unlike the case for the mother, when then male is anesthetized [54]. The active male increases excitation in the pup's PFC, which may activate known inhibitory connections to the PVN [55] (Figure 1, middle). Finally, in adult rats and mice, the ability of companions to reduce HPA responses appears due to the companion activating olfactory connections to the amygdala or directly to the PVN [56–58] (Figure 1, bottom). Thus, at this point it would be premature to conclude that oxytocin mediates the reduction in HPA activity in dogs interacting with humans. Furthermore, as the guinea pig data above suggest, it is even possible that different forms of human interaction (e.g., soothing touch, play) suppress HPA activity through different pathways.

#### **3. Stress E**ff**ects on Behavior**

#### *3.1. The Challenge of Detecting Behavioral Consequences of Stress in Shelter Dogs*

Stress in shelters is of concern in large part because of the possibility it will increase readily apparent behaviors such as stereotypy, hyperactivity, fearful behaviors, and continual barking that will either discourage adoption or prompt recent adopters to return their dog to the shelter, e.g., [59,60]. However, the stress endured by shelter dogs may have less conspicuous effects on behavior that are more difficult to verify experimentally. Major obstacles to the necessary experiments include the impossibility of achieving random assignment, undesirability of invasive procedures, need to accommodate shelter procedures in experimental designs, the hugely divergent past experiences of dogs who end up in shelters, and the difficulty of distinguishing effects of stress as opposed to other aspects of the shelter environment. However, while effects that are unequivocally due to stress are difficult to document, the existing literature in laboratory animals clearly points to a variety of ways that psychological stressors like those experienced in the shelter may both reduce desirable behavior and lead to later emerging behavioral and emotional repercussions, at least for some dogs. To take some examples from the broader literature, juvenile rats exposed to social instability (15 days of repeated periods of isolation followed by housing with unfamiliar conspecifics) showed lower levels of social behavior, both immediately after treatment and in adulthood [61,62]. In another study, periods of maternal separation prior to weaning led to inhibited social behavior and abnormal PFC development in juvenile female rats, whereas for males these effects did not appear until adolescence [63]. Adult mice housed individually for 8 weeks performed more poorly on several measures of cognition than did mice housed in groups [64], and rhesus macaques, whose mothers had been exposed to unpredictable noise bursts during pregnancy, scored worse than controls on measures of attention and neuromotor maturation during the first 3 weeks of life [65] and played less in adulthood [66]. Findings such as these raise concern that desirable traits, such as sociality and cognition, may be compromised as a result of the shelter experience.

Other potential long-term consequences may be more insidious. Much of the current surge in laboratory studies of lasting biobehavioral effects of stress has been driven by the increasing realization that stress at a particular life stage (primarily but not exclusively during prenatal, early postnatal, or adolescent phases) can alter the course of later development, which in humans leads to increased vulnerability to a variety of mental and physical disorders [67]. These include increased susceptibility to major depression, anxiety disorders, post-traumatic stress disorder, and schizophrenia [11,68–70]. Importantly, these outcomes may not emerge in humans until years later, often after a mental or physical

challenge at the later age, a pattern commonly referred to as the "2-hit" model [68,70,71] because a second major stressor or "hit" is required to unmask the long-term effect. The first hit is thought to sensitize some aspect of underlying stress physiology so that the second stressor produces a larger, more prolonged, and/or unregulated stress response that then gives rise to the mental disturbance. These effects can be modeled in laboratory animals. For example, exposing adolescent mice to 12 days of unpredictable stress increased measures of anxiety-like and depressive-like behavior when the mice were placed in stressful situations 30 days later [72]. Similarly, two 3-h periods of isolation near the time of weaning increased depressive-like behavior of guinea pigs when isolated again in early adolescence [5]. If such a model applies to some extent to dogs confined in animal shelters, it implies that behavioral and welfare consequences of the stress of shelter housing may not occur until exposed to a subsequent stressor that then engages the now sensitized stress physiology, perhaps well after the dog has been adopted. One piece of evidence supporting such concern derives from an early study in our laboratory in which dogs were exposed to a highly novel stressful situation before and after 8 weeks of shelter housing. Whereas dogs that received regular sessions of human interaction throughout the 8-week period showed comparable plasma cortisol responses to the two stress sessions, those deprived of the supplemental interaction showed a significantly greater cortisol response to the second stressor [34] (Figure 2).

**Figure 2.** Mean per cent increase in plasma cortisol levels in response to a highly novel situation prior to and following an 8-week period in which shelter dogs either received or did not receive supplementary human interaction (5 weekly, 20-min sessions). Vertical lines represent standard errors of the means. Dogs receiving standard care, but not dogs receiving supplemental human interaction, exhibited enhanced cortisol responsiveness over the 8-week period (*p* < 0.05). Figure redrawn from Hennessy et al. [34].

Glucocorticoids are, in fact, one mediator of lasting behavioral effects of stress exposure, e.g., [73,74], including on social behavior [75]. In utero effects of stress exposure appear mediated by maternal glucocorticoids acting on the fetus [76,77]. Even glucocorticoids received through the mother's milk may influence social and cognitive development [78]. Another mediator of long-term effects is stress-induced neuroinflammatory signaling. Early-life stress upregulates central inflammatory activity in later

life [79,80] and, in humans, increased inflammatory activity promotes development of stress-related disorders [81,82]. Seemingly analogous processes have been demonstrated in laboratory rodents [83,84]. Although stress in early life can affect a variety of brain regions, threat-related circuits [85] including connections between the mPFC, amygdala, and hypothalamus appear to be critical. Among the most robust effects of early-life stress is sensitization of cortico-amygdalar circuitry [86,87]. One way in which inflammatory signaling appears to promote human psychopathology is by enhancing amygdala activity, which then appears to further increase inflammation, creating a positive feedback loop promoting greater susceptibility to stress-related disorders [86,88]. Moreover, increased amygdala activity that escapes regulatory control by the PFC can then affect hypothalamic control of the HPA axis and sympathetic nervous system [89] to further promote development of stress-related pathologies [90–93].

One would not expect all dogs to be equally susceptible to the stress of shelter housing. To the extent that outcomes such as those outlined above pertain to shelter dogs, young dogs or the unborn fetuses of pregnant bitches would be most vulnerable. Further, as others have emphasized, e.g., [94], even in dogs of the same age, we should expect substantial individual differences in stress responsiveness and vulnerability. Due to a combination of experience and temperament, some dogs react much more strongly than others to admittance to a shelter. One form of enhanced reaction is aggression.

#### *3.2. Reduction in Fear-Induced Aggression in the Shelter*

Dogs exhibit a variety of initial reactions upon entering a shelter. While most show signs of fear, for some, the fear is extreme. These dogs may tremble, cower in the back of the kennel, and keep their tail tucked firmly between their legs. They may also show signs of fear-induced aggression [95,96], a situationally dependent form of aggression that occurs in some individuals when fear is high, and escape is thwarted. As this aggression only occurs when dogs are severely frightened, such dogs may be excellent candidates for adoption as a pet in a typical home, but in the shelter they are often in great peril. With shelters understandably concerned about the injuries and damage to the shelter's reputation as a source of quality pets that an aggressive dog might cause, preventing dangerous dogs from being adopted becomes a priority. Some form of a "temperament test" is often used for this purpose. The SAFER® (hereafter SAFER) is one such instrument. Though designed to be but one of several sources of information used to determine suitability for adoption [97], busy shelters may rely solely on the outcome of the test, or a modified version of it, to determine the fate of their confined dogs. If the test is administered a few days after entry to the shelter—before initial fear may have a chance to abate—dogs exhibiting fear-induced aggression are likely to fail and be euthanized rather than adopted.

Both our own work and that of others, e.g., [98,99], suggested that some form of human contact might help reduce the fear and aggression, and ultimately the euthanasia of such dogs. Accordingly, the second author initiated an enrichment program centered around responsive human interaction for fearfully aggressive dogs at a local shelter. The program appeared successful anecdotally and so prompted an experimental evaluation of its effectiveness [100]. This enrichment was provided in a secluded room as in our earlier work [40]. Dogs also had access to toys and were given small treats. In addition, oil of lavender was misted into the room and classical music was played softly in the background in light of the reported calming effect of these forms of stimulation [101,102]. Only dogs that exhibited signs of both a high level of fear and aggression were enrolled. Enrichment was conducted by the second author, a board-certified Associate Applied Animal Behavior Specialist with extensive experience working with shelter dogs. Dogs in a treatment group received the enrichment for 15 min, twice a day, for 5–7 days. Control dogs received normal shelter care. The day following the final treatment, or on the same average day in the shelter for controls, dogs were administered the modified version of the SAFER used by this shelter, administered by shelter staff as per shelter operating procedure. This version of the test assessed the dog's reaction to eye contact, sensitivity to touch, movement and sound during play, response to having its paw squeezed and having its

food bowl manipulated during eating, and the presence of another dog. The staff were unaware that performance on the SAFER was an outcome that we measured in the study.

In an initial experiment, we found that just 10 of 30 fearful dogs in the control group passed the SAFER test, whereas 23 of 30 fearful dogs in the treatment condition passed, a difference that was highly significant [100]. These results were replicated in a second experiment in which only 2 of 16 fearful control dogs and 15 of 16 fearful dogs receiving our enrichment passed the SAFER. For comparison, we also included groups of non-fearful dogs, nearly all of which passed the SAFER regardless of whether they received enrichment or not [100] (Table 2). While we certainly do not encourage those without appropriate qualifications to work with shelter dogs exhibiting fear-induced aggression, we do believe these results are an example of enrichment centered around human interaction buffering meaningful behavioral effects of stress in the shelter environment. They also highlight the value of attending to individual differences in designing treatment strategies. In addition, they align with conclusions of others [103] advocating for the discontinuation of temperament tests for testing adoption suitability (as has been done in the shelter in which we conducted our work).


**Table 2.** Number of fearful enrichment and control dogs that passed and failed the SAFER test in the first experiment of Willen et al. [100].

**\*\*\*** differs from total fearful control dogs, *p* < 0.001.

#### **4. Conclusions**

Lasting behavioral consequences of stress exposure in translational laboratory experiments or in the dog shelter are rightfully considered to be negative outcomes if they serve as models of human suffering or pathology, or in the case of the shelter, reduce welfare or the likelihood of successful adoption. Yet many such outcomes appear to have derived from behaviors that were adaptive in natural environments. If stressful conditions experienced by a young animal, or by its mother prenatally, are an indication that conditions are likely to be stressful in that environment as the individual matures, it can be beneficial for that animal to alter its development to best address the expected future environment. Thus, behavioral plasticity, which can allow for changes in the developmental trajectory to better suit the predicted environment—whether plasticity is induced by glucocorticoids or by other means—will be subject to natural selection [104,105]. This notion of "predictive adaptive responses" is thought to describe a common evolutionary process [76,106–108]. In a stressful, competitive environment, behavioral traits such as reduced sociality, increased reactivity and aggression, and a re-focusing of cognition on skills relevant to basic survival at the expense of "higher level" skills might all be adaptive and, indeed, all have been documented to develop disproportionally following early stressful conditions [108–110]. While specific behavioral outcomes vary by species and situation, there is no reason to expect dogs to be exempt from these influences. We cannot be sure that the stress of shelter exposure or the glucocorticoid elevations or other stress-related physiological changes induced by stress encountered in the shelter are sufficient to produce such changes in behavior development. However, this remains a possibility, particularly for dogs that are especially sensitive to stressors. The concept of predictive adaptive responses might afford a useful perspective from which to consider stress in the shelter and its outcomes in future studies. From a practical point of view, the chance that stressors encountered in the shelter may be shaping later behavior in unwanted ways reinforces continued efforts to reduce shelter stress, such as reviewed here, even if a later negative behavioral or welfare outcome cannot be documented at the present time.

**Author Contributions:** Conceptualization, M.B.H.; writing—original draft preparation, M.B.H.; writing—review and editing, R.M.W., P.A.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** The writing of this review was supported by grant HD100825 from the National Institute of Child Health and Human Development.

**Acknowledgments:** The authors acknowledge the many colleagues and students participating in the experiments from our laboratory reviewed here. The continued support of the Montgomery County Animal Resource Center in Dayton Ohio is gratefully acknowledged.

**Conflicts of Interest:** The authors have no conflict of interest to declare.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Investigating the Impact of Brief Outings on the Welfare of Dogs Living in US Shelters**

**Lisa M. Gunter 1,\*, Rachel J. Gilchrist 1, Emily M. Blade 1, Rebecca T. Barber 2, Erica N. Feuerbacher 3, JoAnna M. Platzer <sup>3</sup> and Clive D. L. Wynne <sup>1</sup>**


Academic Editor: Betty McGuire

Received: 8 February 2021; Accepted: 15 February 2021; Published: 19 February 2021

**Simple Summary:** Animal shelters can be stressful places for dogs to live. Social isolation is likely one component of the environment that contributes to poor welfare but spending time out of the kennel with a person has been shown to temporarily ameliorate that stress. In this study, 164 shelter-living dogs at four animal shelters across the United States were taken on two-and-half-hour outings with a person and physiological measures of stress and physical activity captured by accelerometer devices were compared before, during, and after this short-term outing. We found that dogs' stress was higher when they were away on these field trips and their activity changed, including less time spent in low activity and more time in higher activity. While measures of physiology and activity were found to return to pre-field trip levels the following day, these results suggest that outings of this duration do not provide the same reduction in stress as previously shown with temporary fostering. Nevertheless, short-term outings may provide shelter dogs with greater adoption visibility and assist in foster recruitment and, thus, should be further explored.

**Abstract:** Social isolation likely contributes to reduced welfare for shelter-living dogs. Several studies have established that time out of the kennel with a person can improve dogs' behavior and reduce physiological measures of stress. This study assessed the effects of two-and-a-half-hour outings on the urinary cortisol levels and activity of dogs as they awaited adoption at four animal shelters. Dogs' urine was collected before and after outings for cortisol:creatinine analysis, and accelerometer devices were used to measure dogs' physical activity. In total, 164 dogs participated in this study, with 793 cortisol values and 3750 activity measures used in the statistical analyses. We found that dogs' cortisol:creatinine ratios were significantly higher during the afternoon of the intervention but returned to pre-field trip levels the following day. Dogs' minutes of low activity were significantly reduced, and high activity significantly increased during the outing. Although dogs' cortisol and activity returned to baseline after the intervention, our findings suggest that short-term outings do not confer the same stress reduction benefits as previously shown with temporary fostering. Nevertheless, it is possible that these types of outing programs are beneficial to adoptions by increasing the visibility of dogs and should be further investigated to elucidate these effects.

**Keywords:** dogs; animal shelter; cortisol; stress; welfare; human-animal interaction; activity

#### **1. Introduction**

Between 4.0 and 5.5 million dogs enter animal shelters each year in the US [1,2]. Considerable efforts have been made over the past two decades to improve outcomes for dogs facing this experience (for a review, see [3]), leading to more dogs being adopted and returned to their owners and reduced levels of euthanasia [2,4].

More recently, animal welfare organizations have begun focusing on the standard of care that dogs receive while in the shelter [5]. In part, this is in recognition of the potential stressors within the environment, including excessive noise [6–9], spatial restriction [10–12], social isolation [13], loss of attachment figures [14], loss of control [15] and lack of a daily routine [16]. One way to mitigate the impact of these stressors is through the use of enrichment interventions intended to improve welfare [17–19].

The most commonly studied enrichment intervention in sheltering is interactions with people [15,16, 20–32]. The majority of these interventions occur at the shelter but out of the kennel and are 15–45 min in duration. Often, their impacts are measured by changes in physiology and behavior.

Cortisol is one of the most widely utilized physiological markers of stress in dogs [33]. Previous studies have found elevated cortisol levels for dogs living in shelters as compared to those in homes [29,34] as well as for dogs from homes entering kennels for the first time [35]. Resting activity or sleep can also be a useful component in welfare assessment. Dogs in shelters have been observed to sleep less during the day than dogs in homes [36,37] and had more activity during their most and least active hours than owned dogs [38], suggesting a lack of restful activity for shelter dogs.

A handful of human-interaction interventions, such as those by Hennessy et al. [22], Gunter et al. [32], and Fehringer [28], describe dogs leaving the animal shelter which may provide greater relief to dogs than in-shelter interactions. Gunter et al. [32] reported the impacts of one- and two-night stays in volunteers' homes and measured dogs' cortisol levels before, during, and after those stays. Dogs were found to have lower cortisol levels in homes; and while dogs' cortisol increased upon return to the shelter, it was no higher than baseline levels. Additionally, dogs' bouts of uninterrupted rest were longest when in the foster home, but still remained longer upon return to the shelter than prior to fostering. Fehringer [28] also found that placement in a home resulted in lower cortisol compared to in-shelter levels, and dogs' cortisol steadily declined over the first three days in foster care.

Little is known about how short-term outings of a few hours in duration without an overnight stay could impact the welfare of dogs awaiting adoption. Considering that previously tested in-shelter interventions of less than one hour have been shown to reduce cortisol and improve behavior [15,16,21,26, 27,29–31], it is possible that out-of-shelter outings of a slightly longer duration could confer even greater benefits. However, it is worth noting that many of these aforementioned interventions took place in living room-like settings at the shelter and involved calm interactions with the dog lying down and being petted by a person.

Conversely, short-term outings into the community such as field trips allow for increased physical activity for the dog. Yet a prior study failed to find clear evidence of behavioral improvement following physical activity when provided to dogs at the shelter. Protopopova, Hauser, Goldman, and Wynne [39] compared interventions of exercise and reading offered for 15 min daily for two weeks. Both interventions were followed by increases in dogs' back-and-forth motion in the kennel, a locomotor behavior associated with reduced welfare and longer lengths of stay [40]. While a greater reduction in door jumping was observed after exercise as compared to the reading intervention, dogs were more often at the front of their kennels and barking less often after volunteers read to them for the same duration. Thus, it seems likely that the activity a person engages in with a shelter dog could differentially affect its welfare.

In the present study, we explored whether short-term outings with a person away from the animal shelter would influence dogs' urinary cortisol:creatinine (C/C) ratios in the afternoon of the intervention as compared to ratios collected in the shelter before and after these outings. Additionally, dogs' physical activity was monitored throughout the study to detect potential differences in activity intensity.

#### **2. Materials and Methods**

#### *2.1. Shelters*

Data were collected at four animal shelters in the United States: Spokane County Regional Animal Protection Service (SCRAPS, June 2019); Fulton County Animal Services in Atlanta, GA (FCAS, July 2019); the Regional Center for Animal Care and Protection in Roanoke, VA (RCACP, August and October 2019); and Detroit Animal Care and Control (DACC; September 2019). The shelters varied in their geographical location and annual dog intake but all were open-admission facilities (Table 1).

**Table 1.** Shelter admission type and state where located, canine intake for the prior year, and number of dogs, complete sequences, and sample details.


Note. *\** Complete sequences are dogs in which all five collection timepoints were obtained and used in the analysis. \*\* Samples were removed from data analysis when urinary cortisol:creatinine ratio values were three standard deviations above the shelter's mean.

These shelters had existing short-term outing programs prior to data collection, although their weekly usage varied, ranging from one or less at RCACP to five or less per week at SCRAPS, FCAS, and DACC. Additionally, DACC held monthly events in which typically sixty-five or more dogs would leave on short-term outings with shelter volunteers. Data collection at DACC began 19 days after their last event.

In addition to providing short-term outings, all shelters in this study had walking programs for their dogs in which dogs would spend time out of their kennels interacting with volunteers in and around the facility. Both FCAS and DACC had temporary fostering programs in which dogs would leave the shelter for stays in a caregiver's home. Dogs at all four shelters had access to elevated, cot-style beds in their kennels and also received sporadic food enrichment. At SCRAPS and FCAS, dogs interacted with each other outdoors in supervised groups.

Staff at each of the facilities determined which dogs would participate in this study. To minimize the likelihood of injury or harm to individuals carrying out these field trips, shelters selected dogs without histories of aggressive behavior. Dogs enrolled in the study at FCAS and RCACP had not previously experienced an outing, while it is possible that dogs at SCRAPS and DACC that entered the shelter prior to data collection may have done so. Dogs that were fearful during urine collection were not included due to the researchers' inability to obtain samples for cortisol analysis.

#### *2.2. Short-Term Outings*

Dogs experienced approximately two-and-a-half-hour-long outings with a person between 11:00 a.m. and 3:00 p.m., off the property of the animal shelter. (Less than two percent of outings occurred outside

of these times.) Volunteers, staff, and members of the public were eligible to take dogs on outings but were required to meet organizational requirements prior to participation, including being over 18 years of age and providing a driver's license for identification. Shelter staff provided participating individuals with handling instructions and supplies, and the authors discussed the purpose of this study. When dogs returned to the shelter, the research team asked these individuals to complete a questionnaire about their outing, and dogs were placed back into their kennels.

#### *2.3. Collection Timeline*

Study enrollment lasted three days for each dog with five collections per dog. Beerda et al. [12] found that both urinary and salivary cortisol levels of kenneled dogs were higher in the mornings than in the evenings. However, in studies in which working dogs were active in both the morning and evening as compared to dogs that only had daytime activities, this across-the-day reduction in cortisol was not present [41]. Given this uncertainty regarding the presence of a circadian influence with the cortisol levels of shelter dogs, collections occurred in both the morning and afternoon to capture potential changes over time as well as those caused by the outing.

Morning collection on Days 2 and 3 was conducted between 7:00 a.m. and 9:30 a.m. These times are consistent with those used by Gunter et al. [32]. Afternoon collections on Days 1–3 occurred between 3:30 p.m. and 6:00 p.m. Table 2 provides the experimental timeline for collection. Less than one percent of samples fell outside these collection windows due to dogs not urinating when walked or providing an inadequate volume of urine (minimum 1.5 mL). In these cases, dogs were provided a mixture of wet food and water and walked until the dog urinated.



Note. Due to the time window in which outings could occur, collection for Sample 3 commenced two hours after the dog had returned from its outing. Additionally, collection for Sample 5 was attempted at a time that was between dogs' collection times for Samples 1 and 3.

#### *2.4. Urine Collection*

The research team collected dogs' urine before and after their outings for C/C analysis. Dogs were removed from their kennels on leash and walked to locations outside of the shelter for urine collection and returned to their kennels after samples were obtained. Olympic Clean-Catch™ plastic trays taped to 36 inch (91 cm) "Pickup and Reach" tools (Harbor Freight, Calabasas, CA, USA) were used for urine collection. Samples were poured from the collection trays into 10 mL plastic tubes with snap caps for storage. Trays were rinsed with water and air-dried or wiped with sterile KimWipes™ (Kimberly-Clark, Irving, TX, USA) between collections. Samples were immediately placed in a cooler with ice after collection, and in a freezer within two hours at a temperature of −18 ◦C.

Frozen urine samples were shipped overnight on dry ice to ZNLabs Veterinary Diagnostics (Louisville, KY, USA) for C/C analysis. Analysis was conducted using an automated wet biochemistry analyzer (AU680, Beckman Coulter, Brea, CA, USA) for measurement of creatinine. Bio-Rad Liquid Human Urine Precision Chemistry Controls 1 and 2 (Bio-Rad Laboratories, Inc., Hercules, CA, USA, Control Level 1 #397, Control Level 2 #398) were run on each day of urine sample testing and stored according to

manufacturer instructions. Cortisol was measured using a commercially available product designed for an enzyme-amplified chemiluminescence assay system (Immulite 2000 XPi, Siemens Healthcare Diagnostics, Inc., Newark, DE, USA). Cortisol:creatinine ratios (measured in <sup>μ</sup>mol/L: <sup>μ</sup>mol/L) × <sup>10</sup>−<sup>6</sup> were then calculated.

#### *2.5. Activity Monitoring*

Whistle FIT activity monitors (Whistle Lab Inc., San Francisco, CA, USA) were attached to collars and placed on the dogs, allowing for collection of their movement via the triaxial accelerometer. Placement of collar-mounted monitors occurred during the morning of the dogs' first day in the study and worn for the study's duration, unless battery loss or malfunction resulted in removal and replacement with a new device. Data from the Whistle FITs were transmitted to Whistle servers via each shelter's wireless network and Bluetooth.

Yashari, Duncan, and Duerr [42] assessed the validity of Whistle devices as a measure of canine activity by comparing data generated from these devices with that of Actical, a previously validated activity monitor. They found that measurements of dogs' activity intensity and total activity were highly correlated between the two devices, 0.81 and 0.93, respectively.

For the purposes of this study, dogs' activity was calculated using the raw data recorded by the Whistle devices. These triaxial accelerometers collected the x, y, and z components of a vector representing dog movement, *M* = (*Mx, My, Mz*), at a rate of 50 times per minute. The magnitude of this vector, *M =* <sup>√</sup> *(Mx 2, My 2, Mz 2)*, was used to indicate dogs' composite movement at each time period. Magnitude calculations were then summed over one-minute epochs as an estimate of the dogs' activity during that minute. Magnitude-per-minute values ranged from 0.89 to 8102.

To characterize this activity, all magnitude-per-minute (m/m) values were categorized into one of five, evenly apportioned activity levels. Quintile thresholds were derived from the complete set of magnitude values obtained throughout the study. Each quintile thus contains approximately 37,576 records. The magnitude-per-minute thresholds and associated activity categories, Q1 (lowest) through Q5 (highest), are shown in Table 3.


**Table 3.** Activity level categorization as based on each quintile's magnitude-per-minute thresholds.

M/m values were calculated for the four hours prior to the five urine sample collections, a time window based on the previously demonstrated reflection period of canine urinary cortisol [43]. Dogs' total minutes in each activity level as well as the proportion of time spent in each of those levels were calculated. Data from dogs with at least 200 m/m values for the four hours prior to the urine collection were used in our analysis.

#### *2.6. Statistical Analysis*

To investigate whether dogs' cortisol differed across time, by shelter or in a shelter-by-timepoint interaction, we analyzed C/C ratios obtained for the dogs at our four study sites with a linear mixed model.

Dog and intercept were entered as random effects with timepoint, shelter, and a timepoint-by-shelter interaction along with the covariates of age, weight, length of stay (LOS), and the activity categories included as fixed effects. These covariates were included in the model as they have been previously found to influence cortisol [32]. A variance covariance matrix was employed, and a diagonal covariance matrix for the repeated measure of timepoint. The method of Restricted Maximum Likelihood (REML) was used for estimating parameter values.

To explore whether the number of minutes dogs spent in the five activity categories changed throughout this study, by shelter or in a timepoint-by-shelter interaction, we performed a doubly multivariate analysis with a general linear model. Dogs' weight, age, and mean LOS (across the three days of the study) were entered as covariates. For the within-subjects variable of time, the contrast was polynomial and for shelter, simple.

With both models, dogs' age and LOS were log transformed to ensure the normal distribution of variables. When post-hoc comparisons were conducted in our analyses, a Sidak correction was utilized to reduce the likelihood of false positives when multiple comparisons were made. A statistical significance level of *p* < 0.05 was used throughout.

#### *2.7. Ethical Statement*

Procedures carried out at SCRAPS, FCAS, DACC, and RCACP were approved by the Arizona State University Institutional Animal Care and Use Committee (IACUC: 17-1552R).

#### **3. Results**

#### *3.1. Descriptive Statistics*

When characterizing the dogs at SCRAPS, FCAS, DACC, and RCACP, they were more often male (56.1%), with most dogs arriving to the shelter as a stray (76.2%). Nearly one-fifth of dogs (19.5%) were surrendered by their owners or returned after a failed adoption. On average, dogs were slightly over three years of age (*M* = 39.00 months, *SD* = 30.96) and weighed 23.96 kg (*SD* = 7.55). The number of days dogs were living in the shelter at the time of the study ranged from 1.50 to 252.50 days (*M* = 38.95, *SD* = 42.45).

In total, 164 short-term outings were conducted as part of this study, with 40 dogs participating at FCAS, 41 each at both SCRAPS and DACC, and 42 at RCACP. In an effort to better understand what transpired during these outings, we collected information about the individuals that provided these outings as well as the activities they engaged in with the dogs and where those activities occurred.

Nearly two-fifths of individuals that took dogs on short-term outings were public participants (37.80%), with the remaining individuals being shelter staff and volunteers or members of the research team, and most often those that were providing outings were female (86.59%). Over half of all outings (51.83%) included just one person and the shelter dog. Over one-third of short-term outings (35.37%) did not involve any additional people interacting with the dog, such as petting, playing, or offering the dog a treat, while more than half (53.05%) involved additional interactions with 1–5 people. Shelters varied by the individuals taking dogs on these outings and their interactions (Table 4).

**Table 4.** By shelter, individuals taking dogs on short-term outings that were members of the public, that were female; outings that included more than one person participating; outings where no additional people interacted with the dog; and outings where 1–5 additional people interacted with the dog.


Note. \* Individuals responsible for taking dogs on short-term outings that were not members of the public were either shelter staff and volunteers or part of the research team. \*\* Outings in which more than five people interacted with the dog account for the remaining percentage of outings not reported here.

Over three-quarters of dogs (75.60%) spent time outdoors on their field trip, such as visiting a park, where they walked, hiked, or jogged with the person, and almost 30% of dogs (29.90%) visited a pet-friendly store or restaurant in the community. Fewer than half of the dogs (43.90%) visited the person's home while on the field trip, with only 22.00% of dogs exclusively spending time in a home during the outing.

#### *3.2. Cortisol Analysis*

Dogs at SCRAPS, FCAS, DACC, and RCACP yielded 793 cortisol values that were statistically analyzed across the five urine collections in this study to detect an effect of time of collection, shelter, or shelter-by-timepoint interactions with dogs' weight, (log) age, (log) LOS and minutes spent in each activity category added into the model as covariates.

With this model, the variables of shelter, timepoint, shelter-by-timepoint interaction, weight, and log (LOS) were significant (at *p* < 0.05), with log(age) marginally significant at *p* = 0.061. None of the activity categories were statistically significant but were retained in the model to account for the effect of activity on C/C ratios.

The main effect of timepoint tested was significant, *F* (4, 560.42) = 6.29, *p* < 0.001, demonstrating that the dogs' cortisol changed across the study. In post-hoc comparisons, dogs were found to have significantly higher cortisol values on the afternoon of the field trip as compared to the afternoon of the day before (*p* < 0.001) and the afternoon of the day after (*p* = 0.001). Figure 1 presents the estimated marginal means and standard errors of the cortisol values for the five timepoints across the three days of the study.

**Figure 1.** Estimated marginal means of dogs' cortisol:creatinine ratio values and standard errors for the five study timepoints by shelter. "Overall" represents the estimated marginal means and standard errors at each timepoint, regardless of shelter. All comparisons (shared letters: a–i) are significant at *p* < 0.05, except for comparisons b (*p* = 0.051) and g (*p* = 0.068).

A main effect of shelter was also detected, *F* (3, 149.15) = 3.19, *p* = 0.026, revealing that the estimated marginal means for cortisol varied amongst the shelters. In post-hoc comparisons, dogs at SCRAPS had the lowest C/C ratios, which were significantly different from dogs at DACC (*p* = 0.017). Table 5 includes the average estimated marginal means of C/C ratios and standard errors for the dogs at SCRAPS, FCAS, DACC, and RCACP. When this analysis was repeated, excluding cortisol values from the afternoon of the field trip to examine only pre- and post-intervention timepoints in the shelter, this effect was slightly more pronounced, *F* (3, 149.79) = 4.00, *p* = 0.009. Post-hoc comparisons and subsequent differences between dogs at SCRAPS and those at DACC (*p* = 0.007) were marginally greater. %endparacol


**Table 5.** Mean cortisol:creatinine ratio values, standard errors, *F* test statistics, and *p* values for five timepoints before and after a short-term outing at four US animal shelters.

Note. All comparisons (shared letters: a–i) are significant at *p* = 0.05 or less except for comparisons b (*p* = 0.051) and g (*p* = 0.068).

The interaction of shelter-by-timepoint was significant, *F* (12, 537.12) = 2.01, *p* = 0.022, indicating that dogs' cortisol values differed at each of the shelters at the various study timepoints. When examining these shelter-specific timepoint differences, dogs at SCRAPS had significantly higher cortisol values on the afternoon of the field trip as compared to the following afternoon (*p* = 0.043) and marginally so the morning before the field trip (*p* = 0.051). Dogs at DACC had significantly higher cortisol on the morning of the day following the field trip as compared to the afternoon of the day before the field trip (*p* = 0.020). At RCACP, dogs had significantly higher cortisol during the afternoon of the field trip as compared to the afternoon of the day before the field trip (*p* = 0.020), the morning prior to the field trip (*p* = 0.007), the afternoon of the day after the field trip (*p* = 0.005), and marginally so the morning after the field trip. Figure 1 presents the estimated marginal means and standard errors of the cortisol values at each shelter across the study timepoints.

When investigating differences between shelters at the same collection timepoint, dogs at DACC had significantly higher morning cortisol values on the days pre- and post-field trip than dogs at SCRAPS (*p* = 0.004 and *p* = 0.001, respectively). The morning values post-field trip on Day 3 were also significantly higher at FCAS (*p* = 0.017) and marginally so at DACC (*p* = 0.064) than values obtained at SCRAPS. Table 5 includes the estimated marginal means of C/C ratios and standard errors at each collection timepoint at SCRAPS, FCAS, DACC, and RCACP.

#### *3.3. Activity Analysis*

At SCRAPS, FCAS, DACC, and RCACP, 121 dogs provided 710 readings of minutes spent in each of the five activity categories. Minutes were analyzed across the five study timepoints to detect an effect of time, shelter, or a shelter-by-timepoint interaction with dogs' weight, (log) age, (log) meanLOS entered as covariates in the model.

With this model, Box's Test of Equality of Covariance Matrices was shown to be violated (*p* < 0.001), indicating that the covariance matrices of the dependent variables were not equal. As such, test statistics are reported here are using Pillai's Trace as it is considered to be the most robust test to violations of model assumptions [44]. Mauchly's Test of Sphericity was also violated for four of the five activity categories (*p* < 0.001), with the exception of activity category Q3 (*p* = 0.369), thus sphericity was not assumed, and Greenhouse–Geisser tests were used to determine statistical significance.

We found that timepoint significantly influenced dogs' minutes in the five activity categories, *F* (20, 95) = 41.78, *p* < 0.001, demonstrating that their activity varied across the three days of this study. In post-hoc comparisons, dogs spent less time in the lower activity categories of Q1 and Q2 during the afternoon of the field trip than any other time in the study (*p* < 0.001). Conversely, dogs spent significantly more time in

the higher activity categories, Q4 and Q5, during the afternoon of the field trip than all other timepoints (*p* < 0.001). Figure 2 presents the estimated marginal means and standard errors of minutes spent in the five activity categories for the five timepoints across the three days of the study.

**Figure 2.** Estimated marginal means and standard errors of minutes spent in the five activity categories during the four hours prior to each of the five urine collection timepoints.

A significant effect of shelter was also found, *F* (15, 336) = 2.20, *p* = 0.006, indicating that the minutes dogs spent in the various activity categories differed amongst the shelters; however, in post-hoc comparisons, only one difference was found between shelters. Dogs spent more time in Q3 activity at DACC than dogs at RCACP (*p* = 0.007).

A significant shelter-by-timepoint interaction was detected, indicating that the time dogs spent in each of the activity categories differed between shelters at the various study timepoints. (SCRAPS: *F* (20, 95) = 8.91, *p* < 0.001; FCAS: *F* (20, 95) = 5.97, *p* < 0.001; DACC: *F* (20, 95) = 12.53, *p* < 0.001; RCACP: *F* (20, 95) = 15.88, *p* < 0.001.) When examining shelter-specific activity levels, two patterns were apparent. Firstly, activity during the afternoon of the field trip was generally different from other timepoints; and secondly, the activity recorded in the mornings and afternoons differed from each other, mirroring the timepoint post-hoc comparisons previously reported (see Figure 2). One additional difference was seen at SCRAPS, where dogs spent less time in the lowest activity category (Q1) during the afternoon of the day after the field trip than the afternoon of the day before the field trip (*p* = 0.018).

When exploring whether dogs' activity varied across study timepoints at the shelters, the timepoint-by-shelter interaction was significant, with the exception of the morning of the day after the field trip, indicating that minutes of time spent in the various activity categories at each of the study

timepoints differed by shelter. (Timepoints 1: *F* (15, 336) = 2.39, *p* = 0.003; 2: *F* (15, 336) = 1.54, *p* = 0.088; 3: *F* (15, 336) = 2.18, *p* = 0.007; 4: *F* (15, 336) = 1.41, *p* = 0.138; 5: *F* (15, 336) = 2.10, *p* = 0.003.) We found that dogs at SCRAPS had more low activity (Q1) than dogs at DACC during the afternoon of Day 1 (*p* = 0.034), fewer Q1 minutes in the morning of Day 2 as compared to the dogs at DACC (*p* = 0.036) and RCACP (*p* = 0.013). Dogs at RCACP also had significantly more minutes of Q1 low activity during the afternoon of Day 3 than dogs at DACC (*p* = 0.004).

When exploring the second-lowest activity category (Q2), the only differences detected were during the afternoon of the field trip: dogs at FCAS had more Q2 activity than dogs at SCRAPS (*p* = 0.039) and RCACP (*p* = 0.048). With regards to moderate Q3 activity, DACC dogs spent more time in this category during the first afternoon of this study than dogs at FCAS (*p* = 0.001) and RCACP (*p* = 0.007) and marginally more time than RCACP dogs during the morning of Day 3 (*p* = 0.054). Only two differences were found in dogs' higher activity (Q4), and those were between the same shelters in the mornings before and after the field trip: dogs at SCRAPS had more higher activity than dogs at RCACP (*p* = 0.008 and *p* = 0.002, respectively). Lastly, dogs at RCACP spent more time in the highest activity (Q5) during the field trip than dogs at FCAS (*p* = 0.001).

#### **4. Discussion**

In our investigation of short-term outings, we found that this intervention increased the stress of shelter dogs and decreased their resting activity. Even after accounting for activity in our cortisol analysis, we found that dogs' C/C values were significantly higher during the afternoon of intervention as compared to the prior and following afternoons. However, cortisol did return to pre-outing levels by the next day. Additionally, we found that dogs of greater weight had lower cortisol values, and older dogs had higher cortisol values (as previously shown by Zeugswetter et al. [45] and Rothuizen et al. [46], respectively). These new findings support the previously reported relationships of dogs' weight and age to cortisol [32]. Dogs with shorter lengths of stay also had higher cortisol values than those with longer stays in the shelter.

In our analysis of time dogs spent in the five activity levels across this study, we found similar apportioning of activity in the mornings and afternoons prior to and after field trips. During the mornings, dogs were spending the most time in low activity (see Table 3). Not surprisingly, these are early morning hours prior to staff arrival when dogs are often sleeping, supporting prior findings about the activity of dogs in shelters [36,38]. Conversely in the afternoons pre- and post-outing, dogs were more active than the morning, spending more minutes in the higher and mid-activity levels of Q5, Q4, and Q3.

#### *4.1. Duration, Activities, and Location of Human Interaction with Shelter Dogs*

To our knowledge, the present study is the first investigation of an intervention in which dogs leave the shelter for a few hours and then are returned to their kennels. More commonly in the scientific literature, human interaction is provided to dogs while remaining at the shelter. In these studies [15,16,21,26,27,29–31], interactions with the person ranged from 15 to 45 min in duration and dogs' cortisol was found to temporarily decrease, a finding that was not replicated in the present study despite the increased duration (2.5 h) of human interaction.

It may not be the duration of the interaction, though, that is as important as the location and activities undertaken between the person and the dog. In the aforementioned studies where the dogs' cortisol levels were lower following the intervention, particularly those by Hennessy et al. [15], Hennessy et al. [16], Shiverdecker et al. [27], Dudley et al. [29], Willen et al. [30], and McGowan et al. [31], dogs were removed from their kennels and the interaction occurred at the shelter but in a quiet, secluded room. In the present study, less than one-quarter of the dogs returned to the person's home for their outing, while the majority of dogs spent time walking, hiking, or jogging in public. When Gunter et al. [32] and Fehringer [28]

reported reductions in cortisol in their fostering interventions, dogs were taken to homes for one and two or three nights, respectively.

When examining dogs' activity across the three days of this study, dogs were more active in the greater intensity categories during the short-term outings. Specifically, they spent nearly twice as long in Q5 activity as any other afternoon in the study. Previous research by Radosevich et al. [47] demonstrated that as the intensity and duration of exercise increased so did dogs' cortisol. This arousing effect of activity on cortisol has also been found with sled dogs [48,49] and dogs off leash at a dog park [50].

Contrariwise, cortisol reductions observed by Hennessy et al. [15], Hennessy et al. [16]; Shiverdecker et al. [27], Dudley et al. [29], Willen et al. [30], and McGowan et al. [31] were induced after dogs received calm petting, with the person even massaging and talking soothingly to the dog during the intervention (with the exception of the stranger and play conditions in Shiverdecker et al. [27]). Similarly, Gunter et al. [32] reported a relaxing experience in the home, with dogs having their longest bouts of uninterrupted rest during temporary fostering.

#### *4.2. Psychological Stress and Increases in Cortisol*

Based on the statistical analysis, however, activity alone does not explain this rise in cortisol during the afternoon of the outing, allowing for the possibility that the difference in cortisol was related to the psychological stress of the field trip. Instead of a quiet room where dogs could escape the stressors of the shelter or spend time resting in a home, dogs were active on these field trips and exposed to a variety of novel sights and sounds. If visual, auditory, and olfactory stimuli in the shelter have been found to negatively impact dogs' welfare [17], field trips that include activities such as outdoor dining, hiking, or visiting a store could be stressful, too. Future studies, where volunteers exclusively take dogs to homes or other quiet locations during the outings and calmly interact with them instead of partaking in more energetic activities may find the type of reduction in cortisol that was reported in previous studies.

Increases in cortisol, as reported here with short-term outings, do not necessarily indicate poorer welfare for dogs. Owned dogs engage in a variety of preferred activities that positively impact their welfare and are accompanied by higher cortisol levels, such as attending the dog park [50], competing in agility [51] or hunting [52]. Certainly, these activities are arousing; but we propose that it is the environments in which these dogs are living that should be considered. After such activities, owned dogs return to their homes while dogs awaiting adoption return to the stressful environment of the shelter. Zeugswetter and colleagues [45] identified that the median morning C/C ratio for healthy owned dogs is just 16 (μmol/L: <sup>μ</sup>mol/L × <sup>10</sup>−6), while in the present study, the median cortisol value in the morning was 27.1 (μmol/L: <sup>μ</sup>mol/L × <sup>10</sup><sup>−</sup>6).

#### *4.3. Proximate Welfare Interventions*

Our research and that of others [29,34] indicate that shelter dogs have highly elevated cortisol levels, and these levels likely persist for a prolonged period of time [18]. In addition to physiological measures of stress, shelter dogs spend significantly less time resting than dogs living in homes [38]. By all accounts, it would seem that instead of further arousing dogs that are already coping with an unpredictable environment, enrichment interventions should be aimed at reducing dogs' cortisol levels and other physiological and behavioral measures indicative of compromised welfare, even if these reductions are transitory. A recent review of canine enrichment in the shelter by Gunter and Feuerbacher (under review) [53] characterized a variety of interventions that can positively impact the lives of shelter-living dogs.

In addition to the previously mentioned in-shelter human interactions and fostering programs where the dog leaves the kennel and spends time with a person in an effort to improve their welfare, Gunter and

Feuerbacher [53] suggested that dogs would benefit from more enriched living conditions while in the shelter, such as beds to lie upon and objects that they prefer to chew, such as soft, plush toys. With regards to social interaction with other dogs, Gunter et al. (in prep) [54] recently found that when dogs were provided three, 15 min conspecific sessions a day, their levels of Secretory Immunoglobin-A, an immune function antibody, were lower when compared to days when no dog contact was provided, suggesting that time with other dogs could also be beneficial to their welfare.

#### *4.4. Individual Shelter Differences*

While we were able to detect an overall effect of the intervention on cortisol values across our study, we were also able to detect overall shelter differences as well as variability in how the intervention affected dogs at the individual shelters. For example, cortisol values of dogs at SCRAPS were, on average, lower than those at other shelters. Differences in the magnitude of the intervention's impact were also seen. Specifically, the arousing effect of the field trips, as indicated by increased C/C ratios, was detected most strongly at SCRAPS and RCACP, likely because of their lower in-shelter cortisol values. Yet, dogs at DACC and FCAS that also had higher afternoon cortisol values during field trips did not have statistically significant differences in pre- and post-outing comparisons.

We also found that cortisol values obtained at SCRAPS and RCACP failed to demonstrate evidence of a circadian rhythm effect: both morning and afternoon cortisol values were similar. One shelter, DACC, showed the strongest potential evidence of reduction across the day; however, a possible explanation for this reduction may be related to their operating hours. While SCRAPS and RCACP remained open until the early evening, DACC closed to potential adopters, volunteers, and most staff by 3:30 p.m. Hewison, Wright, Zulch, and Ellis [55] found that closing a shelter to the public during afternoons led to reductions in noise, increases in dogs' sedentary behavior, along with decreases in locomotor and stereotypic behaviors. Thus, the reduction in cortisol across the day at DACC may be related to an absence of late-day stressors. If so, it may be possible that shelter dogs' cortisol levels could decrease across the day, but the stimulating nature of human traffic at adoption facilities in the afternoons and early evenings may be preventing this from occurring.

#### *4.5. Distal Effects on Welfare*

While we have not found evidence here for the proximate welfare advantages of brief outings, it is possible that short-term outings may benefit shelter dogs in ways we did not measure. Outings with a volunteer, staff person, or member of the public may support adoptions and foster recruitment efforts by increasing the visibility of adoptable dogs in the community, which could benefit dogs' ultimate welfare by enabling them to permanently leave the shelter and live in a home. In an ongoing investigation evaluating the deployment and implementation of short-term outing programs in animal shelters, preliminary analyses indicate that over 5% of dogs that experience a field trip are adopted by the person taking them on that trip. Future studies that explore these distal effects would aid shelters in the evidence-based care they provide to homeless dogs by identifying which enrichment interventions improve dogs' daily experiences and which programs are associated with decreased lengths of stay and increased likelihood of adoption.

#### *4.6. Limitations*

When considering the limitations of our study, not all dogs at these shelters were eligible to participate, particularly dogs that were unable to walk on leash, fearful of the urine collection tools, or deemed unsafe to handle by shelter staff. While efforts were made to conduct this study at shelters that were representative of animal sheltering in the US, the four participating shelters were open-admission facilities. It is possible

that facilities with more managed intake policies and lower intake numbers may have dogs in different living conditions that would respond differently than the findings reported here.

Additionally, the individuals taking dogs on short-term outings, though they were most often shelter staff and volunteers and members of the research team, did vary in their dog knowledge and handling skills, which may have affected the stress that dogs experienced during their field trips. Additionally, it is possible that a dog's familiarity with the individual providing the short-term outing may further impact its experience [56]. However, it is worth noting that Gunter et al. [32] showed a consistent reduction in stress when dogs left the shelter during one and two nights of temporary fostering, which were provided by members of the public and shelter volunteers.

#### **5. Conclusions**

This study demonstrates that shelter dogs' urinary cortisol concentrations increased during the afternoon of a short-term outing, even when accounting for their activity throughout the study. During the afternoon of the intervention, dogs' high-intensity activity increased, and low-intensity activity decreased. These changes in cortisol and activity, however, were temporary, and both returned to pre-outing levels by the following day.

The magnitude of the intervention's effect on cortisol and activity varied between the participating shelters with values differing amongst shelters as well as at this study's various timepoints, suggesting that the living conditions at these facilities also influence dog welfare. In all, our findings indicate that short-term outings as tested here do not provide the reductions in stress achieved with temporary fostering in a home. Nevertheless, it is possible that outings of this type benefit shelter dogs' distal welfare by increasing adoption prospects within the community and should be investigated further to understand this effect.

**Author Contributions:** Conceptualization, L.M.G., E.N.F., and C.D.L.W.; methodology, L.M.G., E.N.F., R.J.G., and E.M.B.; validation, R.J.G., E.M.B., R.T.B., and J.M.P.; formal analysis, L.M.G. and R.T.B.; investigation, L.M.G., R.J.G., E.M.B., R.T.B., E.N.F., and J.M.P.; original draft preparation, LMG; writing—review and editing, L.M.G., R.J.G., E.M.B., R.T.B., E.N.F., J.M.P., and C.D.L.W.; visualization, L.M.G., R.T.B., and C.D.L.W.; supervision, L.M.G., E.N.F., and C.D.L.W.; funding acquisition, C.D.L.W., E.N.F., and L.M.G. All authors have read and agreed to the published version of this manuscript.

**Funding:** This research was funded by Maddie's Fund.**Institutional Animal Care & Use Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Animal Care and Use Committee of Arizona State University (Protocol 17-1552R, approved on 10 February 2017).

**Acknowledgments:** Thank you to the staff and volunteers of Spokane County Regional Animal Protection, Fulton County Animal Services, the Regional Center for Animal Care and Protection, and Detroit Animal Care and Control for their generosity and support of this research. Special gratitude to Havalah Moran, Lindsey Soffes, Meera Solomon, Michelle Harmon, Belinda Bell, Lara Hudson, Margo Butler, Dana Eldred, Renena McCaskill, Meghann Cords, Melinda Rector, and Mike Warner for their logistical and data collection support. Additional thanks to laboratory staff, graduate students, and research assistants, including Lindsay Isernia, Vlada Markov, Madeline Hadank, Bre Johnson, Becca Rafalowski, Anjali Patel, Meagan Samele, Tayo Oduyelu, Michelle Dwyer, Allie Kaufman, Chelsea Hughes, and Alysha Verba. Without the support of these individuals and organizations, this research would not be possible.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of this manuscript, or in the decision to publish the results.

#### **References**





**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **E**ff**ects of Olfactory and Auditory Enrichment on Heart Rate Variability in Shelter Dogs**

#### **Veronica Amaya 1,\*, Mandy B. A. Paterson 1,2 , Kris Descovich <sup>1</sup> and Clive J. C. Phillips <sup>1</sup>**


Received: 20 July 2020; Accepted: 6 August 2020; Published: 10 August 2020

**Simple Summary:** Many pet dogs end up in shelters, and the unpredictable and overstimulating environment can lead to high arousal and stress levels. This may manifest in behavioural problems, and decreased welfare and adoption chances. Heart rate variability is a non-invasive method to measure autonomic nervous system activity, which plays an important role in the stress response. The sympathetic nervous system is responsible for increasing the dog's arousal in response to stress and the parasympathetic nervous system is responsible for counteracting the arousal and calming the dog. Environmental enrichment can help dogs to be more relaxed, which is likely to be reflected by increased parasympathetic activity. Dogs' heart rate variability responses to three enrichment methods capable of reducing stress—music, lavender and a calming pheromone produced by dogs, dog appeasing pheromone and a control condition (no stimuli applied) were compared. Exposure to music appeared to activate both branches of the autonomic nervous system, as dogs in that group had higher heart rate variability parameters reflecting both parasympathetic and sympathetic activity compared to the lavender and control groups. We conclude that music may be a useful type of enrichment to relieve both the stress and boredom in shelter environments.

**Abstract:** Animal shelters can be stressful environments and time in care may affect individual dogs in negative ways, so it is important to try to reduce stress and arousal levels to improve welfare and chance of adoption. A key element of the stress response is the activation of the autonomic nervous system (ANS), and a non-invasive tool to measure this activity is heart rate variability (HRV). Physiologically, stress and arousal result in the production of corticosteroids, increased heart rate and decreased HRV. Environmental enrichment can help to reduce arousal related behaviours in dogs and this study focused on sensory environmental enrichment using olfactory and auditory stimuli with shelter dogs. The aim was to determine if these stimuli have a physiological effect on dogs and if this could be detected through HRV. Sixty dogs were allocated to one of three stimuli groups: lavender, dog appeasing pheromone and music or a control group, and usable heart rate variability data were obtained from 34 dogs. Stimuli were applied for 3 h a day on five consecutive days, with HRV recorded for 4 h (treatment period + 1 h post-treatment) on the 5th and last day of exposure to the stimuli by a Polar® heart rate monitor attached to the dog's chest. HRV results suggest that music activates both branches of the ANS, which may be useful to relieve both the stress and boredom in shelter environments.

**Keywords:** dog; heart rate variability; shelter; stress; arousal; lavender; dog appeasing pheromone (DAP); music

#### **1. Introduction**

Animal shelters are stressful environments due to novelty, loud noises, unpredictability and lack of control [1,2]. This overstimulating environment can lead to high arousal levels and stress in shelter animals. The stress response is multifactorial and includes activation of the hypothalamic–pituitary–adrenal (HPA) axis and the sympathetic branch of the autonomic nervous system (ANS), with behavioural [3] and physiological changes [4] produced. Behavioural responses to stress consist of increased arousal [5,6], which in turn results in heightened sensory sensitivity and alertness, the production of corticosteroids and increased heart rate (HR) [7]. Stress in animals can be monitored in various ways, such as behavioural observation, which provides external indicators of an animal's internal state [8] and the response to its surroundings, physiological measures such as the amount of circulating glucocorticoids [9] and heart rate variability (HRV) [10]. HRV is a useful indicator of ANS activity [11] and has the advantage of being measured non-invasively [10,12] (externally and without puncturing the skin).

The ANS is divided into two branches: the sympathetic nervous system (SNS), which is excitatory, and the parasympathetic nervous system (PNS or vagal), which is inhibitory [13]. When there are no apparent threats, the PNS is dominant, which helps to maintain low levels of arousal and a stable heart rate [13]. PNS activity is mediated by acetylcholine neurotransmission released by the vagus nerve [14] and it produces a rapid response in cardiovascular function [15]. The SNS becomes dominant in situations of psychological or physical stress, leading to arousal that helps the individual to respond to environmental challenges [13]. SNS activity is mediated by epinephrine and norepinephrine [14], producing changes in cardiovascular function in a slower time course than PNS [15].

HRV is the fluctuation of time intervals between successive heart beats [16] and reflects the interaction between both branches of the ANS on the sinoatrial node of the heart [10]. A healthy heart has irregular time intervals between beats [17,18], therefore a high variability in sinus rhythm suggests better health and cardiovascular adaptability [19]. Low variability can indicate abnormal cardiac activity or an ANS imbalance leading to poor adaptability to psychological and physiological challenges [19]. HRV is assessed through several time domains, frequency domains and non-linear parameters (Table 1).


**Table 1.** Heart rate variability (HRV) parameters and their physiological origin.

\* Unit abbreviations: ms: milliseconds, bpm: beats per minute, %: percentage, ms2: milliseconds squared, n.u.: normalised units.

HRV can be recorded using standard electrocardiogram (ECG) equipment, such as Holter systems, and wearable devices such as Polar® heart rate monitors [11]. The recorded RR intervals (duration between two consecutive R waves of the ECG) are then analysed using HRV software such as Kubios. The measurement of HRV can be challenging in terms of accuracy and interpretation. One key challenge is determining whether all traces are valid or if some are artefacts. Artefacts are recordings that appear like heartbeats but are not produced by sinoatrial node depolarisations and therefore are abnormal [16]. They can occur due to technical factors, such as poorly placed or fastened electrodes, movement of the subject and/or long recordings [25–27]. Artefacts can also occur due to physiological factors, such as ectopic beats, ventricular tachycardia and atrial fibrillation [25,26]. Data should be corrected for artefacts [11], as otherwise they can affect the reliability of the results [25,27]. HRV is influenced by many factors, such as respiration, posture and physical activity, and therefore the conditions under which data are collected should be standardised (i.e., stationary subject) [11] to allow treatment effects to be identified.

Studies in cattle (*Bos taurus*) [28], horses (*Equus caballus*) [20] and dogs (*Canis familiaris*) [29] have investigated the association between stress and HRV. In calves, root mean square of differences between the successive RR interval (RMSSD) was significantly reduced in those with external stress load (ambient temperature >20 ◦C and insect disturbance) and internal stress load (diarrhoea) compared to animals without obvious stress load. Standard deviation of RR intervals (SDNN) was significantly reduced in calves with internal stress load compared to those experiencing external stress load or no evident stress load [28]. In horses, there was a significant increase in HR, low frequency (LF) and the LF/HF ratio and a significant decrease in HF when they were forced to move backwards for 3 min compared when at rest or when forward walking [20]. In dogs approached by a stranger in the absence of their owner, HR increased and SDNN decreased [29]. Maros et al. [30] found that when dogs looked at their favourite toy, SDNN significantly increased, possibly indicating elevated attention. Kuhne et al. [31] found that when dogs had increased HR and decreased RMSSD compared to baseline values, they performed more appeasing and redirected behaviours. Moreover, low percentage of successive RR interval pairs that differ by more than 50 ms (pNN50) has been associated with aggression; this parameter was significantly lower in dogs with bite histories compared to dogs without them [32]. These results indicate that HRV is a useful tool to assess physiological and emotional stress.

As mentioned above, shelters can be challenging places for dogs and it is important to try to mitigate stress and arousal levels to avoid chronic stress that may impact welfare [33]. Moreover, stress and high arousal levels can increase the development of undesirable behaviours [3,34–36]. These reduce the likelihood of adoption [37], increase the risk of being returned to the shelter after adoption [38,39] and elevate the risk of euthanasia [2]. Sensory environmental enrichment, which consists of stimulating one or more of an animal's senses [40] is a useful tool to help reduce stress and arousal levels.

In humans, music has been effective in reducing anxiety in patients during hospitalisation [41], and can enhance relaxation by masking unpleasant noises [42]. Music can also reduce anxiety in patients with coronary heart disease, cancer patients and those awaiting surgery [43]. Music has been effectively used in animal studies as a form of environmental enrichment, for example, Western lowland gorillas (*Gorilla gorilla gorilla*) tended to perform more behaviours suggestive of relaxation when exposed to classical music compared to a no auditory stimulation control [44]. Additionally, Asian elephants *(Elephas maximus)* showed less stereotyped behaviours when exposed to classical music compared to a no auditory stimulation control [45]. Kennelled dogs have been experimentally exposed to different types of auditory enrichment. Kogan et al. [46] examined the effects of different types of music and found that with classical music dogs spent more time sleeping and less time barking than with heavy metal, bespoke music specifically designed for dog relaxation, or no music. Bowman et al. [47] used a variety of music genres (Soft Rock, Motown, Pop, Reggae and Classical) and found that when any type of music was played, dogs spent less time standing and more time lying (with the exception of Reggae). Wells et al. [48] played different types of music (Classical, Heavy Metal

and Pop), as well as human conversation, and found that dogs exposed to classical music spent more time resting and less time standing than dogs exposed to the other treatments. In Bowman et al. [9], the initial effects of classical music compared to a silent control, were a reduction in vocalisation and increase in time lying down, but dogs habituated to the stimuli by the second day of exposure.

Lavender exposure has been associated with increased relaxation [49] and reduced anxiety in humans [50,51], and has also been shown to have beneficial effects in different animal species. In pigs (*Sus scrofa domesticus*), lavender straw appeared to reduce the severity of travel sickness [52]. Horses exposed to humidified air mixed with lavender essential oil had lower heart rates, after an acute stress response, than horses exposed to humidified air alone [53]. In mice (*Mus musculus*), lavender was shown to have a sedative effect after inhalation, reflected by decreased motility of the animals [54,55]. Similarly, lavender has been used in dogs in different environments. Graham et al. [56] used diffused essential oils in a rescue shelter, and found that dogs spent more time lying down and less time moving when exposed to lavender and chamomile oils compared to rosemary and peppermint oil and a control (no odour). Wells [57] studied the effects of lavender for travel-induced excitement in dogs. Dogs were exposed to a lavender impregnated cloth and a control cloth (no odour) while going on car journeys. Dogs exposed to lavender spent more time resting and less time vocalising and moving. A study in sheep *(Ovis aries)* showed that lavender effects depended on the sheep temperament. Calm sheep exposed to lavender oil had a lower agitation score and vocalised less than calm control sheep, while nervous sheep exposed to lavender vocalised and attempted to escape more than nervous control sheep [58].

Dog appeasing pheromone (DAP) is a synthetic compound based on fatty acids secreted by the mammary gland of bitches after parturition [59]. The effect of the DAP diffuser use has been studied in puppies with disturbance (i.e., vocalisation and continuous door scratching at night) and house-soiling issues. Puppies exposed to DAP cried significantly less than those exposed to a placebo, but there were no effects on the number of nights that puppies soiled inside [60]. Dogs using impregnated DAP collars showed some improvement in behaviour while in car journeys. The greatest improvement was in dogs that had shown motion sickness signs (vomiting and salivating), while the least was in excitable dogs (those who had shown behaviours as barking, jumping and whining) [61]. In a veterinary clinic setting, DAP diffusers appeared to reduce anxiety signs, but there was no evidence of aggression reduction during a clinical exam with a single exposure to the pheromone [62]. In shelter dogs, DAP diffused for 7 days reduced barking amplitude and frequency when people walked by the kennels [63]. DAP collars have been used in puppies during training sessions where they appear to result in less fearful and more sociable behaviour, and improved learning [64]. This literature shows that sensory environmental enrichment can help to reduce stress and arousal signs in different settings and different animal species.

This study is part of a larger project investigating enrichment effects in shelter dogs (methodology and behaviour results are reported in Amaya et al. [65]). The first part of this project analysed the behaviour of sixty dogs when exposed to music, DAP, lavender or a control [65]. Dogs performed fewer vocalisations and increased calmer body postures when exposed to any of the treatments compared to the control, although the effect was weaker for the lavender treatment [65].

The current paper reports on the physiological data collected from the shelter dogs during the study described in Amaya et al. [65] and the aim was to determine if the stimuli have physiological effects that are detectable through HRV recordings. We hypothesized that HRV parameters influenced by vagal activity will be higher in shelter dogs exposed to music, lavender and DAP than those in the control group.

#### **2. Materials and Methods**

#### *2.1. Subjects*

The subjects enrolled in the study consisted of 60 shelter dogs; 35 males and 25 females, all desexed. Mean (± SD) dog age was 3.2 ± 2.4 years, ranging from 6 months to 11 years. They came from different sources and most were mixed breed. Their mean length of stay in the shelter was 45.9 ± 29.8 days, range 8–150 days (Appendix A). On admission to the shelter, all dogs were given a veterinary clinical examination and a standardised behaviour assessment as described in Clay et al. [66]. Each week, the RSPCA Qld Behaviour Team identified dogs for the study, with inclusion criteria being those showing high arousal-related behaviours, such as air snapping, mouthing, attempts to bite their lead or handler, excessive activity, constant vocalisation and over-reaction to other dogs. The selection was made based on information of their kennel behaviour, as provided by shelter staff working with them regularly. Shelter staff were responsible for placing the selected dogs into the study kennels; they were blind to the treatments and assigned dogs at random to each kennel as they became available.

#### *2.2. Study Environment*

This study was conducted at the Royal Society for the Prevention of Cruelty to Animals Queensland's (RSPCA Qld) Animal Care Campus at Wacol, Brisbane, Australia, between August and November 2017. Shelter activities took place as usual (cleaning, feeding and walking) and therefore shelter staff and volunteers were regularly present around the kennel blocks. Two kennel blocks were used for this study, each consisting of 16 kennels divided into two rooms of 8 kennels (two rows of four) and separated by a door. Each kennel had dimensions of 1.6 m × 4 m, and included a crate measuring 0.72 m × 1.55 m and a bed. Both sides had plastic walls that prevented dogs from seeing each other. The back of the kennel had thin metallic bars from roof to floor and the front door had a solid section at the bottom and the same metallic bars from the top of the solid section to the top of the door. For housing details refer to Amaya et al. [65]. The dogs were fed dry food twice a day and had water ad libitum. They were taken for walks twice a day by volunteers for 10 min each time (during the morning cleaning and the afternoon spot cleaning) and had occasional contact with volunteers at other times, except for the 3 h treatment period and 1 h post-treatment.

#### *2.3. Study Design*

Dogs were exposed to one of three forms of enrichment: music (*n* = 14), lavender (*n* = 15) and DAP (*n* = 16) or a control condition (no stimuli applied; *n* = 15). Dogs were exposed to the stimuli in their kennel for 3 h/day on 5 consecutive days, but the HRV measurement only took place on the final day of exposure to the stimuli. Treatments were conducted between approximately 10.30 am and 13.30 pm, depending on when all morning activities were complete. Dogs were also monitored for one-hour post-treatment.

For the music treatment, a databank of 301 songs was downloaded from Spotify (www.spotify. com/au/), and filters were applied to these songs to select music believed to be most suitable for the dogs. Songs were included if they matched the following criteria: tempo of 70 or fewer beats per minute, valence from 0 to 0.5 and energy less than 0.2 (these two last on scales of 0–1.0) [67]. The piano was the sole instrument, except in 6 tracks in which there was accompaniment by violins for part of the tracks [65]. Previous research suggests that single instruments require less neurological processing than multiple instruments [68]. The resulting selection of 51 tracks with a total 183-min duration was played with random track selection order on a mobile phone (Motorola® mobility (Google), Moto G (1st generation), Mountain View, CA, USA) connected to a mobile wireless stereo speaker (Logitech®, X300, Lausanne, Switzerland), with a set volume throughout the experiment. The speaker was placed in a plastic holder hung on the crate's door (in the middle of the kennel). The music was played at 70 dBA, measured from the kennel's door (700 cm of distance from speaker) using a sound level meter (Digitech®, QM-1589, Stanford, CT, USA) at the beginning of each treatment period.

For the lavender treatment, one ultrasonic diffuser (Select Botanicals, Gladesville, New South Wales, Australia) was placed in the crate and another at the back of the kennel. The dilution was 4 drops of 100% organic Bulgarian lavender (*Lavandula angustifolia*; Select Botanicals, Gladesville, New South Wales, Australia) in 60 mL of water. For the DAP treatment, 3 pumps of a synthetic analogue of the canine appeasing pheromone (15.72 mg/mL; Adaptil®, Ceva, Glenorie, New South

Wales, Australia) were sprayed on a bandana worn by the dog and 2 pumps on the dog's bedding as recommended by the manufacturer. Three additional pumps were sprayed at different points of the kennel's floor (1 at each of the back corners and 1 the front door). The control dogs did not receive any extra sensory stimulus.

#### *2.4. Data Collection and Analysis*

On the 5th day of every research week, the dogs were fitted with a heart rate monitor. Four human heart rate monitors were used throughout the study, randomly allocated to treatments. Two different models were used: 3 Polar® RS800CX (Polar Elctro, Kempele, Finland) and 1 Polar® V800 (Polar Elctro, Kempele, Finland). They consisted of a wearlink strap, a watch-computer and a wireless integrated network device. The Polar® RS800CX has been validated for measuring heart rate variability of dogs [69–72] and employed in studies using music as environmental enrichment [9,47,73]. The Polar® V800 has been validated for measuring heart rate variability in humans [74].

The heart rate monitor was positioned on the left side of the thorax at the third intercostal space and secured with adhesive bandages (ZebraVet®, Rocklea, Queensland, Australia). The area of attachment was shaved and cleaned with methylated spirits to allow good contact between the device and the skin. Ultrasound liquid was generously applied to the device to help with the transmission. The watch-computer was secured to the dog's collar. The heart rate monitor recorded for 20 min before commencing the treatment to capture a baseline of the heart rate. It then recorded for three hours while the dog was being exposed to the treatment. An extra hour was recorded after the treatments had finished to measure after-effects. Every 45 min the watches were checked to make sure they were still recording. If they had stopped, more ultrasound liquid was added and the recording restarted. The procedure for the heart rate monitor positioning and securing with adhesive bandages was the same for dogs in the four treatments.

Once the recording finished, data were transmitted via a bidirectional infrared interface to the Polar® Protrainer 5 software (Polar Electro, Kempele, Finland) for the Polar® RS800CX and via USB connection to the Polar® FlowSync software (Polar Electro, Kempele, Finland) for the Polar® V800. These data were then exported as text files to Kubios software (Standard Version 3.1.0. Kubios Oy (limited company) Departments of applied Physics, University of Eastern Finland, Kuopio, Finland).

Dogs were video recorded in their kennels using two mini cameras with charge-coupled devices and infrared capability (Signet®, Electus Distribution Pty. Ltd., Rydalmere, New South Wales, Australia), one at the front and one at the back of the kennel. Behaviours were recorded continuously (24 h/d during the 5 d of stimuli exposure). Observations were divided in three periods: the treatment period (3 h) 5 min observed every 15 min, i.e., 12 separate observations lasting 3600 s in total; the post-treatment period (4 h), 5 min observations every 30 min, i.e., 8 separate observations lasting 2400 s in total and the night period, 5 min of each hour were observed, i.e., 16 separate observations lasting 4800 s in total. Boris® behaviour coding software (version 6.0.4. for Windows, Torino, Italy) was used to record behaviour in an ethogram [65]. There was a single coder for all the videos and they were not blind to the stimuli as specific objects of each treatment were visible in the videos (i.e., bandana, speaker and diffusers). Time values were then transformed into % values (duration of behaviour/total observation time × 100, in s). Videos were observed for a second time during the baseline, treatment and post-treatment periods of the 5th and last day of exposure to the stimuli, when HRV was recorded, to find segments where the dogs were lying down for 5 consecutive minutes. This position was chosen as movement can interfere with the recordings and create artefacts. It has been recommended to obtain HRV during conditions when the subject is stationary, with unchanging motor activity [11,14]. As the dogs were freely moving in the kennel, the only possible segments of 5 consecutive minutes in the same position were obtained when the dogs were lying down. Five 5-min segments that fit the position criteria and had the smallest percentage artefact correction were chosen for each dog during the treatment period; this segment length has been recommended to standardise HRV studies [11,14]. Data were analysed either uncorrected or corrected using the 'very low threshold' option of this software

(0.45 s) and only segments with less than 5% corrected beats were included in the analysis following Kubios [75] recommendations. Of the 60 dogs originally enrolled in the study, 5 were excluded from the HRV analysis for the uncorrected data analysis, and 26 from the corrected data analysis. One dog was adopted the day before the HRV analysis took place. For the corrected data, the excluded dogs either did not fit the requirement of having segments with less than 5% artefact correction while lying down (*n* = 17) or did not meet the criteria mentioned above and also had missing data due to technical issues with the Polar® straps and/or watches (*n* = 8). No attempt was made to interpolate data for missing dogs. Therefore, data from 55 and 34 dogs were included in the uncorrected and corrected HRV analysis of treatment effects, respectively: music (*n* = 12 and 6), lavender (*n* = 13 and 10) and DAP (*n* = 16 and 9) or the control condition (*n* = 14 and 9). Baseline and post-treatment data were not used as only 14 and 7 dogs, respectively, had segments fitting the standard requirements. Baseline values would have been a useful measure as the dog's own control, but it was only recorded for 20 min and therefore it was hard to find 5 min segments when dogs were lying down and furthermore, with less than 5% artefact correction. Due to the large imbalance in dog numbers between treatments, the baseline data could not be statistically analysed.

#### *2.5. Statistical Analysis*

The HRV data were statistically analysed using Minitab 18 software (Minitab. LLC, State College, PA, USA). Mixed effects models were constructed using dog and heart rate monitor (HRM) as random factors, with dog nested within HRM, and treatment as a fixed factor. Dependent variables were Mean RR (ms), Mean HR (bpm), SDNN (ms), RMSSD (ms), pNN50 (%), standard deviation 1 of the Poincare Plot—short-term HRV (SD1), standard deviation 2 of the Poincare Plot—long-term HRV (SD2), LF/HF ratio, LF band (0.067–0.235 Hz) and HF band (0.235–0.877 Hz). These bands were estimated specifically for dogs by Behar et al. [22]. Both bands are expressed in absolute values of power (ms2) and normalised units (n.u.). Artefacts were also fit as a dependent variable.

Assumptions were checked via plotting, and square root transformations were used for absolute LF power (ms2), absolute HF power (ms2) and LF/HF ratio, to achieve normal distribution of residuals. Assumptions were met after transformation. R-squared values for all models were high (>68% for HRV parameters and 65% for artefacts). When omnibus tests were significant (*p* < 0.05), differences between individual treatments were examined using Tukey's tests, which adjust for multiple comparisons. Trends were considered if *p* ≤ 0.10 but >0.05.

In the first study from this project [65], treatment effects on the behaviour of dogs (*n* = 60) were analysed using mixed effects models constructed using dog as a random factor and dog number (entry time to the study), treatment and day as fixed factors. Only a subset (*n* = 34) of the dogs from that study were able to be included in the current dataset, therefore to assist in the interpretion of HRV treatment effects, the behaviour was reanalysed using the same statistical model but only using results from dogs with both behaviour and HRV data.

#### **3. Results**

#### *3.1. Artefact Correction and Model Selection*

There was no significant treatment effect on artefact correction (F3,26 = 0.75, *p* = 0.53) with mean correction levels of 1.88%, 1.42%, 1.42% and 2.23% (SED = 0.49) for music, lavender, DAP and control treatments, respectively. We selected the model that used artefact correction because that method is generally recommended by those working in this field [11,25,27]. R-squared values were high (>70%) for both uncorrected and corrected models.

#### *3.2. Treatment E*ff*ects on HRV during the 3 h Treatment Period*

Absolute LF power (ms2) was higher in dogs exposed to music compared to those in the lavender and control groups (Table 2). SD2 (ms) was higher in dogs exposed to music compared to those in the lavender group. There were trends for treatment effects on mean HR, mean RR (ms) and SDNN (ms) (*p* = 0.10, 0.08 and 0.07, respectively). Inspection of the means suggest that these trends are largely influenced by the music group, which had the lowest mean HR, and highest mean RR/SDNN of the treatment/control groups. There were no significant treatment effects for any of the other HRV parameters.

**Table 2.** HRV parameters of dogs (*n* = 34) exposed to lavender, music, dog appeasing pheromone (DAP) or a control treatment in a shelter environment, during the treatment period. For square root (√) transformed parameters, back transformed values are also reported in parentheses. Means that do not share a superscript letter are significantly different (*p* < 0.05) by Tukey's test.


Mean RR (ms; mean time duration between successive RR intervals (two consecutive R waves of the electrocardiogram (ECG)), Mean HR (bpm; mean heart rate), SDNN (ms; standard deviation of RR intervals), RMSSD (ms; root mean square of differences between successive RR interval), pNN50 (%; percentage of successive RR interval pairs that differ by more than 50 ms). LF/HF (low frequency/high frequency ratio), LF (low frequency) band (0.067–0.235 Hz) and HF (high frequency) band (0.235–0.877 Hz), both expressed in absolute values of power (ms2) and normalised units (n.u.). SD1 (ms; standard deviation 1 of the Poincare Plot—short-term HRV) and SD2 (ms; standard deviation 2 of the Poincare Plot—long-term HRV).

#### *3.3. Treatment E*ff*ects on Behaviour of Subset of Dogs (n* = *34) during the 3 h Treatment Period*

Reanalysis of the behaviour data from the previous study with the current animal cohort resulted in some differences to the statistical outcomes (Table 3 and Appendix B). For 11 of the 20 behaviours analysed, the deduced statistical significance (significant: *p* < 0.05, trend: 0.05 < *p* ≤ 0.10, or non-significant: *p* > 0.10) was the same. Two behaviours were no longer significant in the subset cohort (lie down total and sniff ground) and two lost significance and became trends (lie down-head down and body shake; Table 3) Two behaviours were no longer trends (stand and walk) and two became trends (groom and tail still). One behaviour reached criterion for significance in the subset cohort but did not in the full behavioural study (lie down-head up). It is important to note that while the HRV data was only recorded on the 5th and last day of exposure to the stimuli and only segments when dogs were lying down were analysed, the behaviour data belongs to the 5 days of treatment exposure (3 h/d) and therefore includes all the observed behaviours (i.e., standing).

**Table 3.** The behaviour of a subset of dogs (*n* = 34) exposed to lavender, music, DAP or a control treatment, during the 3 h treatment period on 5 consecutive days. For square root (√) transformed parameters, back transformed values are also reported in parentheses. Means that do not share a superscript letter are significantly different (*p* < 0.05) by Tukey's test.


In the subset of dogs, those in the DAP group laydown with their head up more than dogs in the music or lavender groups. Dogs in the control group stood more on their hind legs with their front legs on the exit door and vocalised more than dogs in the three stimuli groups. Dogs in the music group panted and wagged their tail less than those in the control group. There was a trend for treatment effects on lie down-head down. Inspection of the means suggest that these trends are influenced by the music group, which spent the most time lying down with the head down out of the treatment/control groups. In both datasets, dogs exposed to the stimuli showed more behaviours related to relaxation compared to the control group, but in the full dataset the lavender group did to a lesser extent compared to the other two stimuli.

#### **4. Discussion**

#### *4.1. Stimuli E*ff*ects on HRV Parameters*

Dogs from the music group had a higher absolute LF power (ms2) than dogs in the lavender and control groups. The interpretation of the LF band has been debated in the literature. Some studies [76–79] consider it an index of sympathetic activity only, while others suggest that this band reflects a mix of parasympathetic and sympathetic activity [23,80–90]. This second argument is based on research that shows conditions associated with sympathetic activity and therefore an increase in LF power would be expected, but instead a decrease in LF power has been observed [14], for example, during myocardial ischemia [80] and exercise [80,91,92]. Moreover, pharmacological interventions to enhance or reduce sympathetic activity in the heart do not produce consistent changes in the LF power [90,91,93]. Based on this, we have interpreted the LF band as a parameter that is influenced by both parasympathetic and sympathetic activity.

A relationship between the LF band and music has been previously established in humans. It was found that the LF component increased with the number of music sessions people were exposed to, for both calming and excitative music, and decreased when no music was played [94]. It was concluded that the LF component increases with music listening regardless of music type, and musical stimuli might activate both parasympathetic and sympathetic nervous systems as even brief exposure to music can produce perceptible cardiovascular effects [95] and the beat of music alone can cause a response in the ANS [43]. Yet, both calming music and silence produced subjectively relaxing moods [94]. These results concur with our findings, as the absolute LF power (ms2) of the music group was higher than for lavender and control, two non-auditory conditions. This suggests that this parameter reflects the presence of musical stimuli. This possibly activated both branches of the ANS due to its varied effects temporally, with different rhythms and cadences in the tracks used.

The first study of this project compared the effects of the three stimuli and a control condition on behaviour [65]. Although the HRV results suggest that music activates both branches of the ANS, dogs from that group spent significantly more time lying down with their head down and less time standing on their hind legs with their front legs resting against the exit, vocalising and panting compared to the control group (Appendix B). These results were not identical when behaviour analysis was run in the subset of dogs drawn from the first study for the present analysis. However, in both analyses dogs in the music group stood on their hind legs with their front legs resting against the exit, vocalised and panted less than the control group. In this subset of dogs there was also a trend for a treatment effect on lying down with the head down, which appeared to be highest in the music group (Table 3). All of these behaviours are associated with increased relaxation, which corresponds with the data in the previous study using the full cohort of dogs.

SD2 is a non-linear parameter that describes long-term variability and is correlated with SDNN and RMSSD [96]. It is influenced by both sympathetic and parasympathetic input [24,96–99]. Tulppo et al. [99] found that SD2 decreased during exercise after parasympathetic blockade and therefore attributed it to sympathetic activation. Consequently, an increase in this parameter is thought to indicate a decrease in sympathetic activity. Previous studies have found higher SD2 in dogs exposed to classical music compared to a silence control [9,73]. Bowman et al. [9] interpreted it as a decrease in sympathetic activity, associated to decreased anxiety in the dogs. In our study, dogs in the music group had a higher SD2 compared to lavender. As mentioned above, dogs in the music group in this subset of dogs had a trend to lie down with their head down more than the other three groups. Moreover, the lavender group showed behaviours associated with increased relaxation and reduced arousal compared to the control group to a lesser extent than the music and DAP groups in the full dataset analysis. This suggests that the difference in SD2 could be due to lower sympathetic activity in the music group or increased sympathetic activity in the lavender group. However, the lack of difference in other parameters specific to vagal activity makes any firm conclusions difficult.

In humans, several studies have tested the effects of lavender on HRV and other cardiac parameters (i.e., heart rate, systolic and diastolic blood pressure), with no significant effects [100]. However, a dog study had some significant findings. Dogs received either a dermal application of lavender or a placebo during four 3.5 h periods while monitoring HRV. In dogs exposed to lavender, there was a significant increase in HF power and a significant decrease in heart rate, but only during the 3rd and 4th periods, respectively [101]. These results suggest that topical exposure to lavender oil had some effect on vagal activity. The difference in results with our study might be due to the fact that lavender was administered through diffusers rather than on the skin, which may decrease any anxiolytic effect. Further research would be recommended on the effect of application method.

There was a trend for dogs in the music group to have lower mean HR and higher mean RR, which reflects increased vagal activity [20], and higher SDNN, which is influenced by both parasympathetic and sympathetic activity [21] and estimates overall HRV [14]. Bowman et al. [9] found a reduced mean HR, and increased mean RR and SDNN in dogs when initially exposed to classical music. They attributed these changes to a possible increase in vagal activity but also to the fact that the dogs spent a lot of time lying down while music was played. However, RMSSD and pNN50, both of which reflect vagal activity [11,16], were also increased, suggesting that the HRV changes resulted from increased vagal activity due to music exposure. In our study, HRV was only analysed when the dogs were lying down, but this was standardised across the four treatments, indicating that the trend in the music group were possibly driven by increased vagal activity compared to the other groups. This is supported by the trend of dogs in this group to show more behaviours indicative of relaxation. As shelters are very busy environments during the day, being able to rest more may indicate improved welfare [102]. Moreover, when physical activity is controlled, SDNN could be a good sign of increased attention [29,30]. This suggests that while dogs in the music group possibly had increased vagal activity, they could have more intently perceived the stimulus than dogs in the other groups, as music is constantly changing, opposed to DAP and lavender, which are constant. This increased attention could be reflected in some sympathetic activity, also inferred by the higher absolute LF (ms2) power, indicating activation of both branches of the ANS.

No significant differences between enrichment groups were found in RMSSD, pNN50 or LF/HF ratio. Köster et al. [73] did not find significant effects in RMSSD and pNN50 in dogs exposed to classical music compared to those in a silent control during a canine clinical examination practice. In that study, dogs exposed to music had higher SDNN, but lower mean RR than dogs in the control, possibly indicating that exposure to music was a novel experience rather than calming. Neither Köster et al. [73] or Bowman et al. [9] measured LF and HF bands individually, therefore it is not possible to compare directly with our results. However, they did measure the LF/HF ratio and one found it was not significant [73] while the other found it was not consistently affected by music [9].

In our previous behaviour study, dogs exposed to DAP spent more time lying down and stood on their hind legs with their front legs resting against the exit less than those in the control group. Thus dogs exposed to DAP showed more behaviours associated with increased relaxation and reduced arousal compared to the control group. The absence of any significant effects of DAP on cardiac activity is therefore surprising, but they cannot be compared with other studies, as to the authors' knowledge, no other studies have looked into DAP effects on HRV measurements.

#### *4.2. Study Limitations*

One limitation for this study was the small number of dogs with baseline data and the large imbalance in dog numbers between treatments. This baseline would have been useful as an index of each subject's autonomic state, with stressed dogs potentially having lower vagal activity before enrichment exposure [11]. The small number was in part due to the use of artefact correction, with many dogs having more than 5% artefact correction, which made them ineligible for inclusion. Having a smaller number of dogs reduced power and created imbalance across treatments, although significant differences between treatments were still apparent. Further research on the optimum level of artefact correction for studies with dogs is warranted.

Another possible limitation was the need to use an adhesive bandage to keep the heart rate monitor strap in place during the recording. Studies have shown that pressure wraps such as the ThunderShirt® (Durham, NC, USA) [103] and telemetry vests [104] help to reduce heart rate and anxiety related behaviours of dogs that wear them. The bandages could have produced an anxiolytic effect and therefore reduced treatment effects. However, all the dogs (control and treatment) had the bandages applied so if there was an effect, all groups would have experienced it. Moreover, the time that the dogs had the Polar® heart rate monitors attached to them might have been too long, allowing more technical issues and mechanical artefacts to occur. In the future, it would be recommended to have shorter HRV recording periods [105] to have more control over these issues. However, very short recordings would not be recommended either, as placing and adjusting the monitors might be stressful for the dogs and could influence recordings.

Motor activity can influence HRV and it can also mask emotional and health processes and produce more artefacts [16], therefore, HRV measures should be taken when stationary [11,14]. The correlation between Polar® heart rate monitors and echocardiogram decreases as motor activity increases in humans [106], horses [107], pigs [108] and dogs [70]. Moreover, when the aim is to compare non-motor (psychological) components of cardiac activity, only recordings obtained during similar behavioural patterns should be used [11]. As the recordings were taken from dogs freely moving around the kennel, the only possible segments of 5 consecutive minutes in the same position were obtained when the dogs were lying down, and therefore the results reflect ANS activity only for this body posture. Despite a highly standardised protocol and obtaining the HRV measure only of stationary dogs, Essner et al. [69] found that the Polar heart rate monitor missed intervals that the echocardiogram detected and therefore some HRV results can be inaccurate. Parker et al. [107] and Marchant et al. [108] also pointed out some problems with the validity and reliability of Polar® heart rate monitors and particularly when recording data in ambulatory conditions.

It is important to take into account the equipment used and its possible limitations. Following Kubios [75] advice, we used the lowest possible artefact correction level (very low threshold). However, these automatic correction levels were originally developed for human heart rate data, so there is no certainty that they can appropriately correct dog heart rate data [109].

It is possible that because HRV was only measured on the fifth day, dogs had habituated to the stimuli by then. However, based on previous behaviour observations over time [65], there was no evidence of habituation to any of the stimuli over the 5 days of exposure.

#### **5. Conclusions**

From the three stimuli dogs were exposed to, music produced the most changes in HRV, seemingly by activating both branches of the ANS and therefore producing significant changes in HRV parameters that reflect both parasympathetic and sympathetic activity. There were also trends for dogs in this group to have lower heart rate and consequently increased RR intervals. These results combined with the behaviour results from the first study and the behaviour results of this subset of dogs, indicate that dogs in the music group were more relaxed. There is evidence from the HRV that this was related to an increased vagal activity compared to dogs in the other groups. Shelters could consider using similar methods of music enrichment, as is it the easiest and cheapest stimulus to apply and produces both behavioural and physiological positive effects in dogs. It may help to relieve both the stress and boredom in shelter environments. As for the other stimuli, their effect might have not been strong enough to produce measurable changes in cardiac activity.

Wearable devices such as the Polar® RS800CX and the Polar® V800 can be useful tools to measure autonomic responses in dogs. However, many variables should be taken into account when using HRV as a physiological parameter to measure stress. These include the recording quality, the dog's motor activity while collecting the data and artefacts [106].

**Author Contributions:** Conceptualization, V.A., C.J.C.P. and M.B.A.P.; methodology, V.A., C.J.C.P. and M.B.A.P.; software, V.A.; formal analysis, V.A. and K.D.; investigation, V.A.; resources, C.J.C.P. and M.B.A.P.; data curation, V.A.; writing—original draft preparation, V.A.; writing—review and editing, C.J.C.P., M.B.A.P. and K.D.; visualization, V.A.; supervision, C.J.C.P., M.B.A.P. and K.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** We acknowledge the financial support and provision of resources by the School of Veterinary Science and the Centre for Animal Welfare and Ethics, University of Queensland, and RSPCA Queensland for lending their facilities and all the staff and volunteers who helped during the study, especially Joshua Bryson and Annie Cross. We are grateful too for the technical support of John Mallyon.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**

#### **Table A1.** Dogs (*n* = 34) included in the HRV analysis.


#### **Appendix B**

**Table A2.** The behaviour of shelter dogs (*n* = 60) exposed to olfactory and auditory stimuli or a control treatment for 3 h/d on 5 consecutive days [65]. For square root (√) transformed parameters, back transformed values are also reported in parentheses. Means that do not share a superscript letter are significantly different (*p* < 0.05) by Tukey's test.


#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
