**Breed Group E**ff**ects on Complaints about Canine Welfare Made to the Royal Society for the Prevention of Cruelty to Animals (RSPCA) Queensland, Australia**

#### **Hao Yu Shih 1,\*, Mandy B. A. Paterson <sup>2</sup> and Clive J. C. Phillips <sup>1</sup>**


Received: 20 May 2019; Accepted: 24 June 2019; Published: 26 June 2019

**Simple Summary:** This retrospective study involves 107,597 dog welfare complaints received by the Royal Society for the Prevention of Cruelty to Animals (RSPCA) Queensland from 2008 to 2018. Results show that, compared to pure breed dogs, cross-breed dogs were more likely to be reported in welfare complaints. Poisoning, lack of veterinary support, abuse, and being left unattended in a hot vehicle were common complaints in pure breed dogs; while insufficient shelter, exercise and food/water, as well as overcrowding and abandonment, were more commonly reported in cross breed dogs. Utility breeds, terriers and working dogs were most likely to be reported, while toy, non-sporting breeds and gundogs were least likely to be reported. Common complaint types for utility dogs were: insufficient food/water, shelter and exercise, and poor living conditions; for terriers: abandonment, intentional abuses and killing or injuring another animal; for working dogs: insufficient food/water, shelter and exercise; for toy dogs: lack of veterinary care, overcrowding and staying in a hot vehicle alone; for non-sporting dogs: lack of veterinary care, being left in a hot vehicle unattended and poor body conditions; and for hounds: killing or injuring another animal, intentional abuses and poor body conditions.

**Abstract:** Cruelty- and neglect-related canine welfare concerns are important welfare and social issues. Dog breed has been identified as a risk factor for bad welfare, and yet its role in different types of canine welfare concerns has not been fully investigated. We conducted a retrospective study of 107,597 dog welfare complaints received by RSPCA Queensland from July 2008 to June 2018. The breed of the dog involved in the incident was either recorded as stated by the complainant or by the inspector attending the case. Dog breed was divided into groups following the Australian National Kennel Club nomenclature. Dogs of a non-recognised breed were more likely to be reported in welfare complaints than recognised breed dogs. Recognised breed dogs had a greater risk of being reported with poisoning, lack of veterinary support, abuse and being left unattended in a hot vehicle; while non-recognised breed dogs had greater risk of being reported with insufficient shelter, exercise and food/water, as well as overcrowding and abandonment. Utility breeds, terriers and working dogs were most likely to be reported, while toy, non-sporting breeds and gundogs were least likely to be reported. Common complaint types for utility dogs were: insufficient food/water, shelter and exercise, and poor living conditions; for terriers: abandonment, intentional abuses and killing or injuring another animal; for working dogs: insufficient food/water, shelter and exercise; for toy dogs: lack of veterinary care, overcrowding and staying in a hot vehicle alone; for non-sporting dogs: lack of veterinary care, being left in a hot vehicle unattended and poor body conditions; and for hounds: killing or injuring another animal, intentional abuses and poor body conditions. Breed groups rather than breeds may be the best method of breed identification in a public reporting system as they group similar breeds together, and as our research shows, they relate to types of animal welfare complaints. Understanding the relationship between breed group and canine welfare complaints

may help authorities improve public education programs and inform decision-making around which breed a new owner should choose.

**Keywords:** canine welfare; breed; canine cruelty; neglect; RSPCA

#### **1. Introduction**

Animal cruelty can be defined as any socially [1] or legally [2] unacceptable behaviour causing unnecessary pain, discomfort or distress to animals. It is an important social issue, affecting not only animals but also the entire society, for example, there appears to be a strong relationship between animal cruelty and other criminal activities, such as domestic violence [3]. Animal cruelty can happen to any animal, and the dog (*Canis familiaris*) is one of the most common species reported for suspected animal cruelty [4].

For over 15,000 years, dogs have been domesticated to fit into our society with humans intentionally taming them, which facilitates the development of close bonds between humans and dogs [5,6]. According to the Australian National Kennel Club (ANKC), there are 208 recognised breeds that are categorised into seven breed groups—toys, terriers, gundogs, hounds, working dogs, utility, and non-sporting—on the basis of their physical characteristics, temperaments, behaviours, and functions [7]. For example, greyhounds were originally bred to chase and hunt in the Egyptian deserts 5000 years ago, and are categorised as hounds by the ANKC [7,8]. Border collies were selected for their sheep herding abilities in western Europe, northern England and Scotland over 100 years ago, and are classified as working dogs by the ANKC [7,9,10]. However, the original selection traits are often of little value today and their purpose has broadened to become purely aesthetic or related to entertainment (including gambling), causing them to experience some breed-specific mistreatments. For instance, greyhounds [11] and huskies [12] are used in the racing industry, and greyhounds have been widely reported to experience welfare issues associated with their training, living conditions and injuries during racing. The greyhound racing industry has been associated with "live baiting", where a live animal, such as a chicken or possum, is used as a lure to train greyhounds to race [11–13]. Border collies are popular for companionship, but their herding instincts may not be fulfilled in a home environment, which predisposes them to behavioural problems, such as bike or runner chasing, and can finally lead to unsuccessful ownerships [14]. Pit bull type dogs, such as the American pit bull terrier, American Staffordshire terrier and Staffordshire bull terrier are often thought of as aggressive [15]. These dogs may be made to participate in illegal dog fights or used for pig hunting, and animal management laws may be biased against them. In Australia, the importation of American pit bull terriers is banned [16] and the law requires all American pit bull terriers currently in Australia to be sterilised. In addition, these dogs must be kept strictly confined when at home, and when taken out be muzzled and wear easily identifiable collars [15,17]. Similarly, in America, some states (e.g., Indiana and Louisiana) require pit bull owners to obtain a special license and maintain \$100,000 to \$300,000 in liability insurance to cover any potential injuries caused by the dogs [18]. Many studies highlight the negative welfare that may be experienced by racing and fighting dogs [17,19]. The welfare issues they experience are not only breed specific, but are directly related to specific industries (e.g., dog racing and fighting) [20]. However, neglect-related issues, such as failing to provide suitable food and water, veterinary support and suitable living conditions, are more common, yet less discussed [3,21]. To our knowledge, there has been little consideration of the correlation between canine breed and different forms of animal welfare concerns, particularly those related to neglect.

Accurate breed identification is useful in many areas including in shelters, veterinary clinics, research, and even the media [22]. How dogs are identified influences the way they are perceived and how people interpret their behaviours [23–25]. For instance, dogs identified as terriers, especially American pit bull terriers and Staffordshire bull terriers are perceived as playful, curious, fearless, chase prone and aggressive; however, dogs identified as toy breeds are seen as sociable [25]. Current breed identification in the majority of facilities is based on observation, but such visual identification of breed is problematic and often inaccurate [22,26]. In a laboratory-based experiment, 986 people engaged in dog-related professions were asked to visually identify the breed of 20 dogs using video clips. The visual identification was later compared with DNA identification. The results showed that over 50% of participants failed to visually identify dog breeds that matched the DNA identification, and agreement by over 50% of participants was only found with 35% (*n* = 7/20) of dogs [22]. Another study exploring inter-observer agreement among shelter staff differentiating pit-bull-type dogs versus non-pit-bull-type dog revealed moderate reliability (76%–83%) [26]. Consequently, a better breed identification method is needed, and identifying dogs per group or type (e.g., pit-bull type), rather than by specific breed, may result in higher agreement among individuals and be more useful.

In Queensland, Australia, animals are protected by the Animal Care and Protection Act 2001 (ACPA) [2]. This state-based legislation appoints inspectors, some of whom are employed by the Royal Society for the Prevention of Cruelty to Animals, Queensland (RSPCA Qld), to investigate potential breaches of, and enforce compliance with, the Act [2]. There are two main offences under the ACPA: failure to fulfil duty of care responsibilities and cruelty. There are a number of other specified offences. The Act recognises that a person who has charge of an animal owes that animal a duty of care. Failure to provide such care is the basis of the "breach of duty of care" offence. This offence covers such actions as not providing sufficient food, water, exercise, veterinary care and suitable living conditions. It is not only the owner that has a duty of care towards an animal. Anyone who is even temporarily in charge of an animal has a duty of care. The second major offence is "animal cruelty" and according to Section 18 of the Act, cruelty describes any action that causes unjustifiable and unnecessary physical and mental discomfort to animals, inappropriate confinement or transport, unreasonable injuries and inhumane death [2]. A cruel act can be committed by anyone towards an animal, whether it is their own animal, another domestic animal or even a wild animal [2]. It is important to note, that under the ACPA, the intention of a person to be cruel is not a necessary element of a cruelty offence to be proven in Queensland. If an action carried out by a person causes pain and suffering and the action was intentional, the person may be charged with cruelty. The intention to carry out the action must be proved but not the intention to be cruel. If a lack of action deprives an animal of its fundamental needs, then the person who has a duty of care towards the animal may be charged with a breach of their duty of care or cruelty depending on the circumstances. Intention may be considered during sentencing however [2]. Other offences under the Act include unreasonable abandonment or release, the carrying out of prohibited surgical procedures (e.g., tail docking, ear cropping, debarking, etc.), being involved in, or having items used for, a prohibited event, such as dog or cock fighting, and allowing an animal to injure or kill another animal [2].

The public can report suspected welfare concerns to the RSPCA via a "Cruelty Complaints" telephone number, which operates 24 h a day, seven days a week, or by email. In addition, complaints can be made by veterinarians and veterinary nurses, council officers, and other government and non-government employees visiting a location as part of their duties. Finally, animals entering a shelter may trigger an investigation if cruelty or neglect is suspected.

In this study, we aimed to evaluate whether breed was an important factor in relation to canine welfare concerns. This report is the second in a series relating to the analysis of RSPCA Qld canine welfare complaint data [21,27]. We hypothesized that certain breeds would have a higher risk of being reported. We also hypothesized that some breeds would be at higher risk of suffering specific welfare issues than others. Other risk factors, age of the dog [21] and socioeconomic status of the complainant are the subject of other papers [21,27].

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

From July 2008 to June 2018, RSPCA Qld received 129,036 canine welfare complaints. Some involving more than one dog were recorded as multiple complaints sharing the same case number, while others were recorded as one complaint with multiple animals. To avoid sample bias due to multiple entries, we only retained the first complaint of case numbers with multiple entries, discarding 21,439 entries as a result. There remained 107,597 canine welfare complaints for this retrospective study. The data analysis was originally undertaken on the entire dataset and then redone with the reduced number. Finding the complaint distribution and demographics to be similar, we opted for the reduced dataset to avoid problems with pseudoreplication. Animal welfare complaints that fell within the geographical zone of responsibility of RSPCA Qld (determined by a Memorandum of Understanding between RSPCA Qld and Biosecurity Queensland, the Government Department tasked with the administration of ACPA) were investigated by RSPCA Qld inspectors. All other complaints were referred to Biosecurity Queensland to be investigated by their inspectors. However, all complaints coming into RSPCA Qld were included in this analysis.

All complaints were recorded in ShelterBuddy® (RSPCA, Queensland, Australia), the RSPCA Qld database. The following information was requested from the reporter of each incident at the time of taking the complaint: the number of dogs involved and their age, breed(s) (if known), the "complaint code(s)", the suburb, postcode, and in addition, the date was recorded. All cases were investigated either by RSPCA Qld inspectors (*n* = 100,432) or Biosecurity Qld inspectors (*n* = 7165). It is recognized that some of the calls did not relate to a breach of the ACPA or to a genuine welfare concern. The outcome data for these complaints was not analysed in this research. This research is focused on the complaint calls coming in to RSPCA Qld.

Dogs were classified according to two broad age ranges, being dog and puppy, based on reporters' interpretation. It is important to recognise that the information recorded from the complainant may be inaccurate or inaccurately interpreted, e.g., a small dog is commonly referred to as a puppy in Queensland. Records regarding breed and the number of dogs involved were based on either complainants' initial reports or comments from trained inspectors, again recognising inaccuracies with identification of the breed. The "complaint code" was selected by the staff member receiving the call or email from a drop-down menu of 18 possible complaints (Appendix Table A1) [21]. Multiple "complaint codes" were able to be selected for each case according to the description of what was alleged to have happened to the dog(s), and each was treated as a separate code for analysis.

#### *2.1. Dog Breeds*

The distribution of breeds was compared to the breeds of registered dogs obtained from the councils of two cities situated close to the RSPCA Qld headquarters, namely Ipswich City Council and Gold Coast City Council for the same period. Any breed in our data that was documented in any of the following kennel clubs—Australian National Kennel Council (ANKC) [7], New Zealand Kennel Club (NZKC) [28], American Kennel Club (AMKC) [29] and United Kennel Club (UKC) [30]—was considered a recognised breed (RB) and was added to our breed list (Appendix Table A2). Any breed in our data that was not recognised by at least one of the major kennel clubs listed above was classified as a non-recognised breed (N-RB), including all crossbred dogs without any identified breed. In our dataset, it was decided that if more than one dominant breed was listed, the first breed mentioned would be used. For instance, Great Dane × Bull Arab was categorized as Great Dane (Appendix Table A2).

To achieve a secondary representation of breed recognition, RB breeds were amalgamated into the following seven breed groups based on the breed inclusion categories of the ANKC: toys, terriers, gundogs, hounds, working dogs, utility, and non-sporting. Breeds not listed by the ANKC, but recognised by the NZKC, AMKC, or UKC, were categorized into one of the seven groups based on the description of each kennel club. Some breeds (e.g., Australian Koolie and Bull Arab), though listed by the council registrations and thus on the breed list (Appendix Table A2), were not recognized as breeds

by any major kennel club worldwide. Therefore, these breeds were categorized as N-RB. If the breed description was left blank, the dogs' breed was considered unknown (*n* = 15,576/107,597), and these complaints were excluded from any data analysis related to breed factors.

#### *2.2. Statistical Analysis*

Data was analysed using the statistical package Minitab® 17.3.1. (Minitab, LLC., State College, PA, USA) Descriptive analysis was first used to investigate the distribution of RB/N-RB and the seven breed groups. Complaints reported in July 2017 and June 2018 that contained breed information provided by RSPCA inspectors (*n* = 95) were used to examine the agreement of breed identification between the complainant and inspectors. Apart from simple percent agreement measurements, Cohen's kappa coefficient was calculated. Cohen's kappa is a statistical method measuring agreement with qualitative assessments among different raters. It is more robust than a percentage because it considers the possibility of the agreement occurring by chance [26,31]. To examine whether RB, N-RB or certain breed groups were more likely to be reported, the study group was compared with the registration data from Gold Coast and Ipswich City councils where all owned dogs, including working dogs on farms, are required to be registered [32,33]; this was done using Pearson chi-square tests. Eighteen stepwise forward binary logistic regression models were constructed to understand how breed factors correlated with each complaint code. The binary logistic regression model is a nonlinear model using a logistic function to describe the relationship of independent variables and a dependent variable with two possible values, such as yes/no, 0/1, or healthy/sick [34]. The stepwise forward selection refers to a step-by-step method of adding the most significant dependent variable into the model [35]. To determine the effect of breed (RB/N-RB or breed group) on complaint codes, breed (RB/N-RB or breed group) was entered into the binary logistic regression model as a fixed factor, using logit models with the alpha value to enter being 0.15. Complaint codes were entered into the model as outcomes. Separate models were constructed for each complaint code with the same input variable.

#### **3. Results**

#### *3.1. Dog Characteristics*

Common breeds reported by the complainants were Staffordshire bull terrier (10.5%, *n* = 10/95), American Staffordshire terrier (10.5%, *n* = 10/95), Maltese (6.3%, *n* = 6/95), and Bullmastiff (5.3%, *n* = 5/95). Overall, the agreement between complainants and inspectors of breed identification was 23.2% (Cohen's Kappa = 0.074, indicating a slight agreement [31]), and the agreement of breed group identification was 77.8% (Cohen's Kappa = 0.69, indicating a substantial agreement [31]). Therefore, breed groups were used for further analyses.

In the study group, 32.7% (*n* = 35,178) of dogs were N-RB, while only 1.7% (*n* = 1733) of dogs in the Gold Coast and Ipswich councils' data were listed as N-RB. Around 53% (*n* = 56,843) of dogs in our data and 98.3% (*n* = 99,266) of dogs in the councils' data were of RB (Table 1). The remaining dogs (14.5%, *n* = 15,576) in our database were unspecified. Thus, there was an over-representation of N-RB and an under-representation of RB in our dataset. The most common breed group to be reported for canine welfare concerns in our dataset were terriers (28.2%, *n* = 16,030), followed by working dogs (24.8%, *n* = 14,085), utility dogs (15.6%, *n* = 8,857), toy dogs (9.2%, *n* = 5223), non-sporting dogs (8.9%, *n* = 5071), gundogs (7.8%, n=4417), and hounds (5.6%, *n* = 3160) (Table 2). The most common breed group registered by the city councils were also terriers (22.2%, *n* = 22,056), but followed by toy dogs (21.0%, *n* = 20,796), working dogs (17.8%, *n* = 17,637), non-sporting dogs (14.0%, *n* = 13,915), gundogs (11.6%, *n* = 11,504), utility dogs (7.8%, *n* = 7770), and hounds (5.6%, *n* = 5581) (Table 2).



<sup>a</sup> Percent of breeds in our study/percent of breeds in the councils' registrations. 1.00 signifies equal representation in our database, and Ipswich and Gold Coast City Councils registrations. <sup>b</sup> Unable to calculate the overrepresentation coefficient because there was no dog with an unknown breed in the councils' data.

**Table 2.** Distribution of each breed group in our study, and in Ipswich City Council and Gold Coast City Council registrations, with the overrepresentation coefficient.


<sup>a</sup> Percent of breeds in our study/percent of breeds in the councils' registrations. 1.00 signifies equal representation in our database, and Ipswich and Gold Coast City Councils registrations.

#### *3.2. Predispositions of RB*/*N-RB to Welfare Complaints*

Table 1 summarizes the numbers and percentages of RB and N-RB in our data and the councils' data, along with the overrepresentation coefficients. This coefficient is a simple method to compare the two percentages (see explanation below Table 1). Our results indicate that N-RB were at a greater risk of being reported than RB (*p* < 0.001).

We further explored the association between RB/N-RB and different complaint codes. A logistic regression model was generated (Table 3). In the model, there were significant correlations between RB/N-RB and nine (*n* = 9/18) complaint codes. RB had significantly greater risk of being reported with the following complaint codes, listed in increasing order of odds ratio (OR): baiting/poisoning (OR = 0.36, *p* < 0.001), no treatment (OR = 0.59, *p* < 0.001), cruelty (OR = 0.95, *p* = 0.004), and hot animal in car (OR = 0.95, *p* = 0.043). Meanwhile, N-RB had significantly greater risks of experiencing the following complaint codes, listed in declining order of odds ratio (OR): no exercise/confined/tethered (OR = 1.32, *p* < 0.001), overcrowding (OR = 1.32, *p* < 0.001), abandonment (OR = 1.25, *p* < 0.001), no shelter (OR = 1.19, *p* < 0.001), and insufficient food and/or water (OR = 1.08, *p* < 0.001).



<sup>a</sup> Odds ratio refers to N-RB relative to RB; <sup>b</sup> Odds ratio and the 95% confidence interval (CI) of each breed group for every complaint code are presented in Table 4; <sup>c</sup> Breed factor (N-RB/RB or breed group) was not selected in the logistic regression model; <sup>d</sup> A person was reported to have abused an animal.

#### *3.3. Predispositions of Breed Groups to Welfare Complaints*

When we compared the numbers of different breed groups in our data with the councils' data (Table 2), the following breed groups were over-represented in our data, listed in declining order of overrepresentation coefficient (*p* < 0.001): utility, working dogs and terriers. The following breeds in our database listed in increasing order of overrepresentation coefficient were underrepresented (*p* < 0.001): toys, non-sporting and gundogs.

In the regression model (Table 3), twelve (*n* = 12/18) complaint codes were predicted by breed group. Detailed results related to breed groups are summarized in Table 4. Toy dogs were more likely to be the subject of: no treatment, hot animal in car and overcrowding; terriers to: abandonment, cruelty, and knowingly allow an animal to kill/injure another; utility dogs to: insufficient food and/or water, no shelter, no exercise/confined/tethered and poor living condition; non-sporting dogs to: no treatment, hot animal in car and poor dog condition; hounds to: knowingly allow an animal to kill/injure another, cruelty and poor dog condition; and working dogs to: no exercise/confined/tethered, no shelter and insufficient food and/or water.


**Table 4.** Odds ratio and the 95% confidence interval (CI) of each breed group for every complaint code.

*Animals* **2019**, *9*, 390

#### **4. Discussion**

#### *4.1. Breed Identification*

The agreement of breed recognition between the general public and RSPCA inspectors was very low, which is in line with a previous study demonstrating inconsistent breed identification [22]. In light of the inconsistency, instead of assigning a breed for each dog, a potential alternative is to group them by function or more general characteristics into a recognised breed group [36]. Various kennel clubs have adopted similar concepts in their breed group criteria. For instance, in the ANKC, dogs that are originally bred to work with livestock are classified as working dogs, and those developed to assist hunters in retrieving game are classified as gundogs [7]. In this study, breed information provided by the general public and RSPCA inspectors was used to categorise dogs into ANKC breed groups. Consequently, the agreement increased from 23.2% to 77.8%, with substantial reliability (Cohen's Kappa = 0.69) between the public and trained inspectors. This result suggests that both the general public and inspectors are able to recognize some important and obvious characteristics of the dogs. Hence breed groups, rather than breeds, may be a better and more practical way of classifying breed for shelter research.

#### *4.2. RB versus N-RB*

This study examined the relationship between breed and canine welfare complaints. Specifically, we examined RB/N-RB and different breed groups with respect to their predisposition to be reported for welfare problems. We found there was a greater proportion of RB in the councils' data compared with RSPCA's, which might reflect a low rate of registration of N-RB, mainly crossbred dogs, in the two Queensland regions or that RB dogs were less likely to be involved in poor welfare or cruelty. These findings are supported by a previous study showing that crossbred dogs are at a higher risk of non-accidental injuries as a result of physical abuses, such as beating, throwing and burning, than pure bred dogs [37]. A further analysis of different complaint codes revealed that RB were predisposed to complaints related to "gaming", where owners allowed these dogs to engage in racing, fighting and blood sports. The number of dog fight cases was relatively small [21], so we should interpret them cautiously, although it is likely that dog fighting occurs more frequently than is reported. Credible and actionable evidence of such events is rarely received. One surprising finding was RB dogs were reported more often than N-RB for not receiving adequate veterinary care. Previous research into pet ownership and attitudes to pet care found that owners of shelter-acquired pets, usually mixed breed [38], took their animals for veterinary care more often and were equally willing to spend over \$1000 on medical treatment for their pets [39] than pets acquired by other means. Finally, our data suggest that many N-RB dogs are not registered, which may make them more likely to be surrendered to a shelter or abandoned when medical care is required. Previous studies have reported higher rates of surrender of N-RB dogs [38]. N-RB dogs were more likely to be involved in complaints related to husbandry practices and abandonment.

#### *4.3. Breed Groups*

RB dogs were divided into breed groups and strong correlations between the groups, characteristics of the breeds and reasons for being reported were observed. For instance, toy dogs are small and possibly travel with owners more often, and thus, as found in our study, were more likely to be left alone in a car in hot weather. Previous research reports that smaller breeds of dogs are popular in Australia [40], which might help explain the number of complaints about toy dogs being left unattended in hot vehicles, as reported previously [21]. However, increased awareness of the dangers for dogs in hot cars through regular campaigning on this issue may also explain the high number of reports received [41]. Many terriers, especially those with some pit bull type characteristics (e.g., Staffordshire bull terrier, American Staffordshire terrier, and pit bull terrier), are considered aggressive and dangerous [15], therefore would be predisposed to being reported for abandonment, dog fights and cruelty [17,19,37,42]. Finally, utility and working dogs are mostly bred for guarding, rescuing or herding functions [7], and are generally energetic and require exercise [43,44]. These breeds were reported for not receiving adequate exercise. In contrast, toy breeds were the least likely to be reported for insufficient exercise, which is in agreement with previous research that found that smaller dogs were likely to have their exercise needs met even though they were walked less frequently [45].

#### *4.4. Practical Application*

This study provides fundamental information about the relationship between breed groups and various types of welfare complaints in dogs. The information can be used to develop education campaigns to increase awareness of what is involved in adequately and appropriately caring for dogs. Specific breeds have specific needs with respect to, for example, exercise requirements and cognitive enrichment to ensure their welfare is good. Information could be made available to prospective new owners of specific breeds to improve their understanding of the breed and its care requirements. Such information could also inform decision-making around breed choice as requirements of the breed could be matched with the ability of the owner to provide these needs.

#### *4.5. Limitations and Need for Future Research*

This was the first study providing fundamental information of the relationship between dog breed and welfare complaints made to a welfare organization with the responsibility of administering the Animal Welfare Act. Future research could focus on common breeds and explore the welfare issues in more detail.

There were several limitations of this present study. First, N-RB dogs contributed 32.7% of our dataset but only 1.7% of the total population in the councils' data. This major difference might indicate that N-RB dogs were indeed more susceptible to animal welfare concerns, but might also be affected by: (1) the difficulty of breed recognition [22], (2) the potentially different criteria of breed classification in our data and the councils' data and (3) the possibly lower registration rate of N-RB. Second, complaint codes were made based on public reports, which are likely to be inaccurate, at least in some cases. Third, we compared our data with reference data from the Gold Coast and Ipswich City Councils, urban areas in South East Queensland. The RSPCA cases were collected from a broad geographical area including both urban and rural areas, which may cause some regional bias. There were only 0.76% (*n* = 814/107,597) of cases from the Gold Coast and Ipswich regions. If we only compare the cases in these two regions, our results may be skewed by some less-common breeds that are either reported in our data or the councils' registration. Our data cover the Queensland regions along the East Australian seashore, which are nearly identical to the previous research [46]. Therefore, we decided to use similar methods by comparing our entire data with the councils' data. Given these limitations, the data reported here included canine welfare complaints only in Queensland, and national or global generalization should be made with caution.

#### **5. Conclusions**

Dog identification classified on the basis of breed groups rather than specific breed had a higher agreement between the public and shelter staff, and thus may serve as a better method of describing dogs involved in welfare reports. N-RB dogs, mainly crossbred dogs, were significantly more likely to be reported for alleged animal welfare concerns, especially poor living conditions and abandonment than RB. In addition, the characteristics of specific breeds, such as size, physical traits and exercise demands, were correlated to the reported complaints. Our results can help to improve public education and awareness raising. Finally, future studies are encouraged to explore in more detail the relationships between breed and welfare issues in dogs.

**Author Contributions:** Conceptualization, C.J.C.P., H.Y.S. and M.B.A.P.; methodology, H.Y.S., C.J.C.P. and M.B.A.P.; software, H.Y.S. and C.J.C.P.; validation, H.Y.S., C.J.C.P. and M.B.A.P.; formal analysis, H.Y.S.; investigation, H.Y.S.; resources, M.B.A.P. and C.J.C.P.; data curation, C.J.C.P., H.Y.S. and M.B.A.P.; writing—original draft preparation,

H.Y.S.; writing—review and editing, C.J.C.P., M.B.A.P. and H.Y.S.; visualisation, H.Y.S.; supervision, C.J.C.P. and M.B.A.P.; project administration, C.J.C.P. and M.B.A.P.

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

**Acknowledgments:** We thank RSPCA, Qld, for providing the database of canine welfare complaints, RPSCA inspectors for the consultation, and Gold Coast and Ipswich councils for providing the breed registration data.

**Conflicts of Interest:** Mandy Paterson is employed as the principal scientist by RSPCA, Qld, but none of the authors received any interest or financial support from people or organisations that inappropriately influenced this study.

#### **Appendix A**


**Table A1.** Description of each complaint code alleging a welfare issue.

<sup>a</sup> Prohibition order: A prohibition order is given by the court when a person convicted of an animal welfare offense must not possess any or specific animal for a prescribed period of time [2].

#### **Table A2.** Breed list.


#### **Table A2.** *Cont.*


**Table A2.** *Cont.*


**Table A2.** *Cont.*


#### **Table A2.** *Cont.*


**Comment from the Public Breed List Breed Group** Sloughi Sloughi Hounds Small terrier cross Terrier Terrier Smithfield cattle dog Cross breed N-RB Soft coated wheaten terrier Soft coated wheaten terrier Terrier Spaniel Spaniel Gundogs Spanish water dog Spanish water dog Gundogs Spitz Spitz Non-sporting Spoodle Cocker spaniel Gundogs Staffordshire bull terrier Staffordshire bull terrier × labrador American Staffordshire bull terrier Terrier Staghound Staghound N-RB Swedish vallhund Swedish vallhund Working dogs Tenterfield terrier Tenterfield terrier Terrier Terrier Terrier Terrier Thai ridgeback Thai ridgeback Hounds Tibetan mastiff Tibetan mastiff Utility Tibetan spaniel Tibetan spaniel Toys Tibetan terrier Tibetan terrier Non-sporting Timber shepherd Cross breed N-RB Weimaraner Weimaraner Gundogs Welsh springer spaniel Springer spaniel Gundogs Welsh terrier Welsh terrier Terrier West highland white terrier West highland white terrier Terrier Whippet Whippet Hounds White Swiss shepherd dog White Swiss shepherd dog Working dogs Wirehaired fox terrier Fox terrier Terrier Xoloitzcuintle Xoloitzcuintle Non-sporting Yorkshire terrier Yorkshire terrier Toys

**Table A2.** *Cont.*

ANKC: Australian National Kennel Council (http://ankc.org.au/); AMKC: American Kennel Club (https://www.akc. org/dog-breeds/); UKC: United Kennel Club (https://www.ukcdogs.com/breed-standards); NZKC: New Zealand Kennel Club (https://www.dogsnz.org.nz/home/home).

#### **References**


© 2019 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* **Determination of Prolactin in Canine Saliva: Is It Possible to Use a Commercial ELISA Kit?**

#### **Jara Gutiérrez \*, Angelo Gazzano, Beatrice Torracca, Valentina Meucci and Chiara Mariti**

Dipartimento di Scienze Veterinarie, Università di Pisa, 56124 Pisa, Italy

**\*** Correspondence: jara.gutierrez@vet.unipi.it

Received: 13 May 2019; Accepted: 2 July 2019; Published: 4 July 2019

**Simple Summary:** Prolactin is considered a remarkable index of stress response, both acute and chronic, in several species. Some studies have investigated the possibility of measuring prolactin in saliva in human beings and Rhesus macaque. The possibility of measuring it in dog saliva would provide a non-invasive, helpful tool for the assessment of dog welfare. The aims of this research article are to study (1) the possibility of quantifying canine prolactin in saliva using a prolactin canine ELISA (enzyme-linked immunosorbent assay) kit validated for measuring prolactin in canine blood and (2) the potential presence of a correlation between prolactin levels in saliva and plasma.

**Abstract:** Prolactin has been reported to be a remarkable index of stress response, both acute and chronic, in several species. The use of biological matrixes other than blood is receiving increasing interest in the study of hormones, due to the lower invasiveness in collection. This research aimed to investigate the possibility of using a commercial ELISA (enzyme-linked immunosorbent assay) kit for measuring canine prolactin in blood for the quantification of canine prolactin in saliva. Study 1 consisted of a validation protocol, using saliva samples collected from lactating and non-lactating dogs. Study 2 was conducted to investigate a possible correlation between prolactin concentration in saliva and plasma in sheltered dogs by using the same kit. Prolactin values were reliably read only when they came from blood samples, not from saliva, but tended to be low in most of the cases. Study 1 showed that saliva had a matrix effect. In study 2, saliva prolactin levels were low and in 42.9% of cases, not readable. No correlation between prolactin values in plasma and saliva was found (ρ = 0.482; *p* = 0.274). These findings suggested that the determination of prolactin in dog saliva through an ELISA kit created for measuring prolactin in dog blood was unreliable.

**Keywords:** blood; dogs; prolactin; saliva; stress

#### **1. Introduction**

The best-known role of prolactin in the dog is the stimulation of the growth of the mammary gland and the lactation processes. Nevertheless, prolactin has over 300 separate biological activities and it plays multiple homeostatic roles and physiological functions in the organism, such as the electrolyte balance, luteal function, regulation of the immune system, osmoregulation, angiogenesis, maintenance of the inter-oestrous interval, etc. [1].

In addition, prolactin is considered an index of acute stress in some species. For instance, prolactin concentrations increased after various stressful stimuli in humans [2] and rats [3–6], including male rats [7]. In lactating rats, hyperprolactinemia seemed to be a significant factor for the decrease of plasma oxytocin response to acute stress [8]. Furthermore, prolactin seems to reduce anxiety-related behavior in both female and male rats, maybe because prolactin acts as an endogenous anxiolytic, both in males and non-pregnant female rats [9]. The decrease in stress-induced secretion of ACTH from the adenohypophysis and reduced corticosterone secretion from the adrenal gland in lactating rats is also manifested in mid-gestation, from day 15 until day 21 of pregnancy [10].

In human beings, prolactin-releasing stimuli include suckling, perception of visual, acoustic, and olfactory stimuli and stress [11]. For instance, it has been suggested that human serum prolactin concentrations may be elevated by psychological stressors, as well as by psychosocial stress (Trier Social Stress Test) [12]. In addition, in human beings, it is known that prolactin is in response to severe experimental stress induced by hypoglycaemia [2], surgery [2], parachute jumping in military recruits [13], and compulsory swimming in non-swimmers [14]. A significant correlation was found between day-to-day changes in anxiety measured by questionnaires as well as by stress hormones, cortisol, and prolactin [15].

Prolactin also increases in response to psychological stressors like restraint and transport in rats [16], heat stressin both rats and domestic ruminants [16,17], and stressful situations in donkeys [18], dromedaries [19], cattle [20], and sheep [21].

In dogs, both cortisol and prolactin decrease immediately after parturition [22]. Moreover, prolactin increases just after delivery due to the pups' suckling stimulation [22]. In lactating bitches, high levels of prolactin and increased expression of prolactin receptors in the paraventricular nucleus may produce a decrease in the stress response during lactation [23,24]. In addition, anxious dogs displaying signs such as stereotypes, displacement activities, various autonomic disorders, and fear aggression, have an increase in prolactin blood levels [25]. Assistance dogs seem to have higher mean prolactin blood levels than pet dogs, suggesting a possible role of canine blood prolactin as an index of stress-related responses in dogs [26].

When quantifying certain physiological parameters, blood has often been used as the best body fluid to evaluate different biomarkers. However, in recent years, the replacement of blood by saliva samples has achieved growing interest due to its reduced invasiveness, cost, and risk of infection compared to blood collection [27]. Nevertheless, the concentration of these specific biomarkers may often differ between blood and saliva [28]. When multiple human biomarkers were compared in plasma and saliva samples, consistent correlations were found between both types of saliva sampling (passive drool and filter paper), but little correlation was found between plasma and saliva [28]. Korot'ko and Gotovtseva [29] found that human prolactin levels were lower in saliva than serum. In other cases, some methods were reported to be unable to quantify a certain biomarker in the saliva matrix, such as it being reported for salivary oxytocin when measured by immunoassay [30].

As prolactin is usually measured in blood to reduce the stress of blood sampling, it would be useful to find different biological matrixes in which prolactin could be reliably measured. The possibility of quantifying prolactin in human saliva has been evaluated using four commercially available methods different from ELISA (enzyme-linked immunosorbent assay), but none of which could detect it [31]. Salivary prolactin has instead been successfully measured in Rhesus macaques (*Macaca mulatta*) by radioimmunoassay, but a positive correlation between serum and salivary prolactin was not found, perhaps because the pulsatile variation of prolactin in the blood may be an impediment to detecting such a correlation [32]. This pulsatile secretion for prolactin in blood has also been found for dogs [33].

The aim of this study was to evaluate whether a commercially available ELISA kit for measuring canine prolactin in blood is also suitable for measuring canine prolactin in saliva. To do that, the research was divided into two parts: Study 1) consisted of a validation protocol using saliva samples collected from lactating and non-lactating dogs, and study 2 investigated the possible correlation between prolactin concentration in saliva and plasma in sheltered dogs.

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

The procedure was communicated to the Ethics Committee of the University of Pisa, Italy (OPBA, Organismo Preposto per il Benessere Animale) and received a favorable opinion with Decision N.09/2018.

#### *2.1. ELISA Kit*

The prolactin ELISA kit used in this study is an enzyme immunoassay for the detection of canine prolactin in serum developed by Demeditec Diagnostic GmbH (Kiel, Germany). For this research, 2 kits were used and the assay was performed according to the manufacturer's instructions.

The microplate was coated with a monoclonal antibody specific for canine prolactin. Calibrators and samples were placed in front of the 96-well plate. A total of 25 μl calibrators and samples were added in triplicate in successive wells and incubated for 2 hours at room temperature. Endogenous canine prolactin in the sample binds to the antibodies fixed on the inner surface of the wells. Non-reactive sample components were removed by a washing step.

Afterwards, a second polyclonal horseradish peroxidase-labeled antibody, directed against another epitope of the prolactin molecule, was added. During an hour of incubation, a sandwich complex consisting of the 2 antibodies and the canine prolactin was formed. After incubation, the plate was washed with the provided wash buffer and a substrate solution (3, 3 , 5, 5 -tetramethylbenzidine) was added, followed by another 30 min of incubation time. Finally, stop solution (hydrochloric acid) was added and the absorption was measured at 450 nm within 30 min with a Multiskan™ FC Microplate Photometer (ThermoFisher Scientific, Waltham, MA, USA). Concentrations of prolactin were estimated from a calibration curve obtained by plotting the optical density versus the concentration for each one of the calibrators (80.0, 40.0, 20.0, 10.0, 5.0, and 2.5 ng/ml).

#### *2.2. Saliva and Plasma Samples Collection*

For study 1, saliva samples from 21 healthy adult dogs (7 males and 14 females in anoestrus, 1–11 years old) were collected and pooled to establish a sample with regular values (non-lactating saliva, NLS). Saliva samples from 5 bitches in a lactation period (1–7 years old) were collected and pooled to establish a sample with assumed high levels of prolactin (lactating saliva, LS). Saliva was always collected in the morning by the same person, except for some lactating females, whose samples were collected by the person in charge. Saliva was collected using flat cottons (Salivette®Cotton swab, neutral 51.1534, Sarstedt) and gloves, in order to avoid variability in the results and contamination. Samples remained in a cold chain (0 to +4 ◦C) until they reached the laboratory, where they were centrifuged two consecutive times (7000 rpm, 10–15 min). Saliva was obtained after centrifugation and stored at −20◦C until analysis.

For study 2, saliva and plasma samples were collected from 10 healthy adult mixed breed dogs (1 female and 9 males, 1–11 years old) single-housed in a municipal shelter, between 11:00 and 12:00 a.m. and always followed the same order: First blood, and then saliva, with an interval no longer than 2 minutes between them. With the exception of 3 dogs, from which we did not get enough saliva, both saliva and blood were analyzed for each dog. The extraction of saliva was carried out following the same protocol described for study 1.

#### *2.3. Validation Parameters*

The possible application of an ELISA kit for canine blood prolactin to canine saliva samples (study 1) was determined by evaluating the linearity, limit of quantification, matrix effect, and spiking recovery.

The lyophilized master calibrator (80 ng of lyophilized in serum/buffer matrix containing highly purified canine prolactin) was reconstituted with 1 ml of distilled water. To evaluate linearity, this prolactin standard solution was diluted with the provided sample diluent to obtain the solutions for the calibration curves (40.0, 20.0, 10.0, 5.0, and 2.5 ng/ml).

Previous literature shows that in dogs, blood prolactin levels tend to be low. In addition, unpublished data obtained by the authors of the current study showed that salivary prolactin levels are commonly low. Consequently, for this validation method, linearity could not be reliably assessed with serial dilutions of normal samples. In order to reach higher values that could improve the reading of the

kit, it was decided to dilute standard solutions to obtain different final-added prolactin concentrations in artificial saliva (AS20, AS10, AS5, AS2.5; Pickering Laboratories©, Space Park Way, Mountain View, CA, USA),and in non-lactating saliva (NLS20, NLS10, NLS5). Parallelism between the curves obtained with standard solutions and the ones obtained with the spiked saliva pool was also assessed.

Lactating saliva pool (LS), with assumed high prolactin concentration, was diluted (2:3 and 1:2 for kit 1; 2:3 for kit 2) using artificial saliva as a diluent to evaluate linearity. The goal of analyzing lactating dog saliva as such (LS) and in two different dilutions (LSd1, LSd2) was to find out the minimum concentration of salivary prolactin that could be read by the kit.

To assess the possible matrix effect and recovery for each kit, 3 repetitions of a lyophilized control from the same batch (Demeditec Diagnostics GmbH®, Kiel, Germany) were reconstituted (1 ml) with water (control reconstituted in water = CW; as specified by the manufacturer instructions), AS (control reconstituted in artificial saliva = CAS),and NLS (control reconstituted in saliva of non-lactating dogs = CNLS). For the first kit, 3 repetitions for each reconstitution were made using a control batch corresponding to 15.14 ng/ml, and for the second kit, a control batch of 8.16 ng/ml was used (Appendix A).

The lower limit of sensitivity was determined as the mean concentration obtained interpolating the optical density plus 2 SDs for all replicates of the 0 standard.

#### *2.4. Statistical Analysis*

The possible presence of correlation between the prolactin values in blood and in saliva for study 2 has been analyzed through the Spearman's Rho test (*p* < 0.05).

#### **3. Results**

Linearity was acceptable for all the regression curves for prolactin concentrations in sample diluent, in AS and in NLS, for both kits (R2 values: Sample diluents kit 1 = 0.995, AS kit 1 = 0.99, NLS kit 1= 0.994; sample diluent kit 2= 0.996, AS kit 2 = 0.992, NLS kit 2 = 0.993). Slopes between these regression curves were parallel for each kit, respectively (sample diluent kit 1 = 0.033, AS kit 1 = 0.034, NLS kit 1 = 0.037; sample diluent kit 2 = 0.032, AS kit 2 = 0.035, NLS kit 2 = 0.037) and R<sup>2</sup> coefficients for the same curves were acceptable for both kits, meaning that kits ran similarly.

Levels of prolactin measured in non-lactating (NLS) and lactating (LS) saliva pools were similar and, in both cases, low (mean ± standard deviation: NLS = 0.34 ± 0.30 ng/ml; LS = 1.03 ± 1.22 ng/ml for kit 1; NLS = 1.07 ± 0.20 ng/ml; LS = 1.17 ± 0.07 ng/ml for kit 2). Due to these low levels, it was not possible to conduct a linearity analysis based on physiological sample dilutions.

Values for the non-lactating saliva pool with the addition of the standards to obtain different final added concentrations (kit 1: NLS20 = 23.4 ng/ml, NLS10 = 11.4 ng/ml, NLS5 = 6.4 ng/ml; kit 2: NLS20 = 24.1 ng/ml, NLS10 = 12.3 ng/ml, NLS5 = 6.4 ng/ml) were higher than the normal added prolactin concentrations (20, 10, and 5 ng/ml).

Prolactin concentration values obtained in CW, CAS, and CNLS (Table 1) were within the target ranges, except for CNLS in kit 2 which was higher (12.1 ng/ml). CAS obtained virtually the same value than CW in both kits. Prolactin concentration values in CNLS were slightly higher than CW, in both kits (kit 1: CNLS–CW = 18.7–16.5 = 2.2 ng/ml; kit 2: CNLS – CW = 12.1–9.7 = 2.4 ng/ml). Values of CNLS were slightly higher than CAS (Table 1).


**Table 1.** Expected and observed prolactin values when control was dissolved in water (CW), artificial saliva (CAS), and non-lactating saliva (CNLS). Mean and standard deviation for the observed values and recovery percentage.

The lower limit of sensitivity was calculated as the mean concentration obtained interpolating the optical density plus 2 SDs for all replicates of the 0 standard was 1.10 ng/ml.

A reliable inter-assay repeatability was recorded for values whose concentrations were higher than 5 ng/ml (e.g., AS20, AS10, AS5, F10, F5), with coefficients of variation lower than 12% (mean %CV ± standard deviation = 8.13% ± 0.03), whereas for lower concentrations (e.g., AS2.5, LC, LCDil1, F20, F), high coefficients of variation were recorded (mean %CV ± standard deviation = 71.54% ± 0.32).

Prolactin concentration values in saliva and plasma samples obtained in study 2 are shown in Table 2. Prolactin values from plasma samples were below the limit of the kit's detection (0.4 ng/ml) in 20.0% of cases (mean ± standard deviation = 4.69 ± 5.37 ng/ml), and were especially high for dog 1 (17.8 ng/ml). Saliva prolactin levels were low too (mean ± standard deviation = 1.94 ± 1.96 ng/ml) and, in 42.9%of cases, not detectable. In fact, prolactin concentration in saliva samples without any additions resulted in very low values in both study 1 and 2.


**Table 2.** Prolactin concentrations in saliva and plasma samples (ng/ml) using a canine prolactin ELISA (enzyme-linked immunosorbent assay) kit. (BLD = below the limit of detection).

Values from Table 2 clearly show that there is no correspondence between prolactin concentrations in plasma and in saliva: On one hand, dogs 1, 3, and 7 had the highest values of plasma prolactin, but low saliva concentrations, and on the other hand, dog 4 had a relatively high value of saliva prolactin, but prolactin in plasma was not even detectable. This was confirmed by a lack of correlation between plasma and saliva values (ρ = 0.482; *p* = 0.274).

#### **4. Discussion**

Both kit 1 and 2 were able to read the control within the optimal target ranges. Artificial saliva (AS) did not seem to interfere with the reading of the kit, as the reading of control joined to artificial saliva (CAS) was almost the same for the reading for control added to water (CW). In other terms, artificial saliva did not seem to have a matrix effect. However, saliva is a complex matrix and for this reason further measurements were done using real saliva, which will be discussed later.

In both studies, prolactin concentration in saliva samples without any additions resulted in very low values, often below the limit of detection of the kit, meaning that the kit cannot feasibly read prolactin concentrations in natural saliva samples.

A difference emerged between the values of CW and the values of control joined to non-lactating saliva pool (CNLS). It could be hypothesized that this difference was due to the fact that the kit was reading the prolactin present in the non-lactating saliva pool. However, this does not seem to be a justifiable explanation, since the concentration of prolactin in the non-lactating saliva pool alone (NLS) was lower than the difference between CNLS and CW. Since NLS was used as an example of natural saliva, it can be deduced that such a difference was due to a matrix effect of the natural canine saliva itself, and consequently the kit could not read saliva samples properly. Sometimes, a matrix effect can disappear by diluting the sample serially until a linearity of the results is obtained. However, serial dilutions of the lactating saliva pool (LS) did not exhibit a linear dilution, indicating that a matrix component was interfering with an accurate detection, causing a loss in reading sensitivity.

The variation between the values for non-lactating saliva pool with addition of the standards to obtain different final added concentrations (NLS20, NLS10, NLS5) and those of the corresponding standards (20, 10, and 5 ng/ml) are not due to the presence of natural prolactin, since calculated NLS values were different when using the different additions. For this reason, natural prolactin could not be reliably determined with the addition of the standards and the observed difference was likely not reflecting the real value of prolactin, due to the presence of a matrix effect of the saliva. In other terms, although the addition of the prolactin standard allowed higher values to be obtained and, therefore, a reading of prolactin results feasible, a large matrix effect was observed, preventing the assertion that the value obtained was reliable.

Furthermore, we expected prolactin concentrations in non-lactating saliva to be lower than prolactin concentration of saliva samples in lactating dogs, since prolactin levels in blood have been reported to have an increase in lactating dogs [33–35]. Indeed, saliva from lactating dogs had values that were only slightly higher than those of non-lactating dogs and with a high variation. In addition, the reliability of NLS and LS readings was not adequate because their coefficients of variation were quite high.

In most of the cases, the kit was not able to detect prolactin in saliva samples as obtained values were very low and some of them did not reach the limit of the kit's detection. In study 2, when prolactin in saliva was detected by the kit, most of these values were higher than the ones from the lactating saliva pool, as mirrored by mean values. A possible explanation might be that shelter dogs are often exposed to stressful conditions, leading to higher prolactin circulating levels [2,14,15]. However, the lack of correlation with plasma concentrations did not allow us to draw reliable conclusions.

In study 2, plasma prolactin concentrations, when readable, agreed with values reported by other authors for canine prolactin in serum, ranging 1–6.3 ng/ml [36–39]. One exception in the current study is represented by dog 1, with high plasmatic prolactin values, probably due to its behavioral problems since it showed a phobia of numerous stimuli, and possibly a state of anxiety (not assessed as it was out if the scope of this study). The relationship between phobia, anxiety, and prolactin has been investigated by Pageat, 2007 [25], however further research is needed for a better understanding of their possible links.

A correlation between plasmatic and salivary prolactin levels from paired single samples was not found. This fits with the results of Lindell et al. [32], who did not find a positive correlation between values of prolactin in the serum and saliva of Rhesus macaques, discussing the possibility that the pulsatile variation of prolactin in serum impedes the detection of a significant relationship with the pooled saliva source, since single blood samples were used, eventually suggesting that a repeated blood sampling might lead to a significant correlation [32].

In this study, blood and saliva were only collected from non-lactating stressed dogs at the shelter, while for the two pools used for validation (lactating dogs and non-lactating dogs), individual samples were not collected. Although having the results of blood prolactin levels from the same samples of

saliva used for the validation assay would have been optimal, blood collection was avoided in lactating bitches for ethical reasons. We also avoided collecting multiple saliva samples from lactating bitches, in order to minimize interference with their nursing and lactation. Due to the low values obtained in study 1 for both non-lactating and lactating dogs, for study 2, it was decided to use individual samples of blood and saliva from sheltered dogs, which possibly had higher prolactin values due to stress.

Moreover, the use of Salivette for collecting saliva samples may have caused a bias in the results. The different available saliva collection methods were reported to possibly interfere in the biomarker of interest present in the saliva. For this reason, in certain cases, a correspondence between methods of collection was found. For example, Salivette was reported to be a reliable and predictable method of total and quantified free serum cortisol levels [40]. However, in other cases this correspondence was not found. For instance, nephelometrically determined IgA concentrations were significantly lower in saliva when collected by the Salivette than by a suction or spitting method [41]. When quantifying cortisol and dehydroepiandrosterone (DHEA), results using two different saliva collection methods ('passive drool' method and a citric acid-treated salivette) correlated highly with plasmatic levels and with each other, whereas results using Salivette did not correlate significantly with plasma values [42]. Future research using other methods of saliva collection may be useful for knowing if the method of collection interferes with the measurement of prolactin in saliva.

Taken together, all these results suggest that prolactin cannot be reliably measured in saliva by the ELISA kit used in the study, which is validated for measuring prolactin in dog blood, meaning that canine saliva was not a suitable matrix for the kit. The quantification of prolactin though ELISA kits out of the intended use needs caution. In fact, a previous study found that prolactin in horses cannot be reliably measured using a canine prolactin ELISA kit, due to a matrix effect [43].

#### **5. Conclusions**

In summary, for saliva the study found: A matrix effect, very low prolactin concentrations (often under the limit of detection), and a lack of correlation with prolactin plasmatic levels. Validation of the kit showed that prolactin in saliva could be read under certain conditions (standard addition) but without reliability (matrix effect). In the absence of standard addition, prolactin values were too low to be read.

These results suggest that saliva was not a suitable matrix to measure prolactin levels using an ELISA kit created for measuring canine blood prolactin concentrations.

**Author Contributions:** Conceptualization, A.G. and C.M.; Data curation, J.G.; Investigation, J.G. and C.M.; Methodology, B.T., V.M. and C.M.; Supervision, C.M.; Writing—original draft, J.G. and B.T.; Writing—review & editing, J.G., A.G., and C.M.

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

**Acknowledgments:** We appreciate the collaboration of the dog shelter Canile Soffio di Vento(Pisa, Italy) and to the dog breeding centers *Allevamento della Maschera di Tutankhamon* (Palaia, Pisa, Italy) and *Allevamento Labrador Sindia's Labradors* (Pistoia, Italy), as well as the group of dog owners who collaborated altruistically for this study.

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

#### **Appendix A**

**Table A1.** Outline distribution adopted for the validation procedure. Details of the wells for kit 1 and kit 2. The only difference is that the second dilution for lactating female saliva (LSdil2) was replaced in kit 2 by saliva from a different pool of domestic dogs (saliva as such -P-, and two dilutions of it -P20 and P10-). Caption: A-G (reconstitute lyophilized canine prolactin master calibrators), CW (control dissolved in water), CAS (control dissolved in artificial saliva), CNLS (control dissolved in non-lactating dogs saliva), AS20-AS2.5 (artificial saliva with addition of the standards), LS (lactating dogs saliva pool), LSd1-LSd2 (dilutions of lactating dogs saliva pool), NLS (non-lactating dogs saliva pool), N20-N10-N5 (non-lactating dogs saliva pool with addition of the standards, respectively 20, 10 and 5 ng/ml), NLS+CNLS (non-lactating dogs saliva pool + control dissolved in saliva, 1:1).


#### **References**


© 2019 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* **The Perceived Value of Behavioural Traits in Australian Livestock Herding Dogs Varies with the Operational Context**

### **Jonathan Early 1,\*, Elizabeth Arnott 1, Bethany Wilson 1, Claire Wade <sup>2</sup> and Paul McGreevy <sup>1</sup>**


Received: 25 June 2019; Accepted: 11 July 2019; Published: 16 July 2019

**Simple Summary:** Information on Australian livestock herding dogs and their handlers and breeders is limited. This study aimed to collate baseline information on how handlers and breeders value various behavioural traits relevant to the work of these dogs. A survey was presented to explore herding dog behaviour in four contexts including work and competition. The behavioural traits were divided into three groups: working manoeuvres, working attributes and general attributes. Data from 811 respondents revealed that several behavioural traits were of high and low value to handlers and breeders across all contexts, while others were unique to only one or two contexts. For example, *cast*, *force*, *gather*, *trainable*, *confident* and *friendly* were of most value; whereas *bite*, *bark* and *back* were of less value. Further analysis revealed that respondents can be considered as coming from two main groups: firstly, handlers with a preference for specialised dogs in the utility context and, secondly, handlers focussed on the yard context, who need dogs that have a broad range of skills and that are easy to work with. This information may assist in matching handlers with suitable dogs. Future research should clarify handlers' understanding of innate and learnt behaviours.

**Abstract:** This study investigated the value that handlers and breeders assign to various behavioural traits in Australian livestock herding dogs. Data were obtained from 811 handlers and breeders through the 'Australian Farm Dog Survey'. Respondents were asked to consider dogs within four contexts: utility (livestock herding in both paddocks and yards), mustering (livestock herding in paddocks and along livestock routes), yards (in and around sheds, sale-yards and transport vehicles), and trial (specifically a standard 3-sheep trial), and to rate the value of 16 working manoeuvres (movement sequences used in herding), 11 working attributes (skills or attributes used in herding) and five general attributes (personality traits ascribed to an individual dog). The most valued working manoeuvres were *cast*, *force* and *gather*. *Bite*, *bark* and *backing* were considered of little value in certain contexts, notably the trial context. Across all four contexts, the general attributes most valued in dogs were being *trainable*, *motivated*, *confident* and *friendly*, while *control* and *trainability* were the working attribute traits considered to be of most value. *Excitability* was revealed to be a 'Goldilocks' trait in that respondents preferred not too much or too little but a 'just right' amount in their preferred dog. Analysis indicated a handler preference for either specialised dogs for the utility context or dogs who are easy to work with because of a broad range of traits favoured in the yard context. These results reveal both generalities across and the need for specialisation within these four herding contexts. Further investigation may help to reveal how well handlers distinguish between innate and learnt behaviours when selecting and training livestock herding dogs. Identifying which group handlers fit into optimally may assist in selecting suitable dog–human dyads.

**Keywords:** herding; livestock; working dog; survey; traits; boldness

#### **1. Introduction**

The national population of working livestock herding dogs in Australia has been estimated at more than 94,000 individuals [1]. Over an average working career, the estimated economic value of these dogs' work is over A\$40,000 per individual [2]. Successful partnerships between dog and handler reflect the quality of the match between the personality and behaviour profile of the dog and the preferences, experience and skills of its handler as well as the perceived financial value of the dog [3].

Livestock herding dogs are routinely used to move livestock in three over-arching contexts that are also used to label specific working skill-sets: utility (both paddock and yard), mustering (paddock and livestock routes) and yard (in and around sheds, sale-yards and transport vehicles). They are selected primarily for performance and health rather than morphological traits [4], an approach that has resulted in the prevalence of a suite of behaviours thought to be stylised elements of the predatory sequence exhibited by *Canis lupus familiaris* [5–7].

Natural instinct, combined with opportunities to regularly practise and be reinforced for herding behaviours, is fundamental to any dog's performance in a herding task. The unique triadic interaction of humans, dogs, livestock and sometimes handlers on horseback has been referred to as a 'mutually adjusted system' [8]. Insufficient or poor quality training may jeopardise dog and livestock welfare and compromise learning outcomes [9–11]. Investigations into handler–dog interactions during livestock herding training have focussed on moderating access to livestock through negative punishment (interrupting access to livestock) or positive reinforcement (allowing continued access to livestock) [5,12]. These operant techniques reveal the reinforcing value of access to livestock in dogs that have been selected to relish this sort of work [13].

Data on the ease or difficulty with which livestock herding dog handlers can condition dogs to perform certain working behaviour traits may reveal areas in which trainer education may be especially beneficial. They may also identify which traits deserve a particular focus in breeding and selection to ensure livestock herding dogs can perform the task for which they are being bred.

Peer-reviewed studies on the behaviour of livestock herding dogs are rare (*see* [5,8,12,14–16]) and, among them, few are easily transferable to the Australian context. Importantly, only two studies [5,8] defined the dogs used in their studies as working dogs or from working dog lines, rather than companion dogs of herding breeds. One Australian study, which was not subject to peer-review, examined the inheritance of the behavioural trait *eye* (commonly defined as a dog's ability to hold sheep together by staring at them) [17]. It reported that, when using a six-point scale to score *eye*, the 28 dogs tested were most likely to be scored as intermediate to or aligned with one of their parent's scores.

Popular literature on livestock herding in Australia suggests that certain behavioural traits such as *eye*, *force*, *boldness*, *anticipation* and *cast* are pivotal to successful herding ability [18,19]. However, the definition, interpretation, perceived value and relevance of these and other traits varies among authors [17,18,20–25]. For example, a recent study of eight Australian herding manuals identified a significant discordance in the frequency of the use of such popular terms [26].

The current study used a questionnaire to identify the ideal position for a livestock herding dog on the shyness–boldness continuum, the value that handlers place in the ideal dog on the expression of five general attributes; 11 working attributes; 16 working manoeuvres and the ease with which dogs can be trained to show each of these in four distinct herding contexts. A central hypothesis for the current survey study was that respondents would, having some knowledge of innate versus learnt behaviours, report innate behaviours as being more difficult to train.

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

The 'Australian Farm Dog Survey' was designed to investigate the distribution of farm dogs in Australia, their usage, their management and the views of their owners along with demographic information relating to the breeder/handler (for other publications using this survey *see* [2,3,27]). In particular, respondents were asked about five general attributes, 11 working attributes and

16 working manoeuvres in their dogs. The questions were grouped within four herding contexts: utility, mustering, yard, and competition (3-sheep trial).

Prior to publication of the survey, popular working-dog training manuals were consulted and advice was sought from members of the Working Kelpie Council of Australia to ensure that the terminology in the survey was appropriate for the target audience. A pilot distribution of the survey to 125 participants led to some minor modifications prior to widespread distribution.

The online version of the survey was available over a three-month period from 10 March to 10 June 2013. All promotional materials relating to the survey indicated that a hard copy of the survey with a reply-paid envelope would be provided to participants if they requested one by telephone. Approval for this study was granted from the University of Sydney Human Research Ethics Committee (Approval number 15474).

The target population for the survey was livestock herding dog users across Australia. Participation was encouraged with entry into a prize draw to win commercial working-dog food at the end of the survey period. An introductory message gave participants the option to respond anonymously with an assurance of confidentiality were they to choose to leave their details to enter the prize draw.

A link to the online survey was posted on the websites of the University of Sydney [28] Meat and Livestock Australia [29] and the Working Kelpie Council of Australia (WKCA) [30]. It was advertised through stories in multiple regional newspapers, on three nation-wide television programs and in two national agricultural magazines. The committee of the 2013 Casterton Kelpie Auction (one of Australia's leading livestock herding dog auction events) promoted the survey in a mail-out to past and current vendors and purchasers. The researchers also recruited survey participants, in person, at livestock herding dog trials during the study period.

The online version of the survey was constructed using the survey system QSmart (Torque Management Systems Limited, Auckland, New Zealand). The entire questionnaire had a maximum of 143 items assigned to 10 sections. However, participants needed to answer fewer questions if they responded in the negative to questions about certain activities, such as breeding or trialling of dogs. The logic system of the online survey permitted the redirection of participants to questions of relevance. (To view the complete survey, *see* [31].

In Early et al. [26], we defined working manoeuvres, working attributes (referred then as working skills) and general attributes: working manoeuvres represent a sequence of movements used in herding; working attributes reflect an ability used in herding; general attributes are personality traits ascribed to an individual. Where a trait might fit into both working and general attributes or might lie on the same spectrum (e.g., *boldness* and *cautiousness*), the authors made a decision into which category it should be included, based on whether testing and exploring these relationships would identify, statistically, if they are valued in different ways.

Respondents were asked to indicate the ease with which 16 working manoeuvres in the typical working dog can be trained: *cast*, *force*, *gathering*, *cover*, *backing*, *bark*, *bite*, *heading*, *hold*, *balance*, *drive*, *break*, *width*, *pull*, *lift* and *draw*. They answered using a semantic differential-type 5-point rating scale. Descriptive phrases 'extremely easy' and 'almost impossible' were used at either extreme of the scale. Respondents were advised to not provide a rating for any working manoeuvre terms that they were unfamiliar with.

The same 16 working manoeuvres were used again in the next question, which asked the respondents to indicate how valuable they considered these behaviours in livestock herding dogs. Respondents were asked to answer this question separately for up to four herding contexts in which they had experience. These included three types of work, namely, utility (generally known among trainers as all-round), mustering, and yard, and one competition context referred to as trial (i.e., working trials, generally known among trainers as arena or 3-sheep trials). Answer options included a semantic differential-type 5-point rating scale, ranging from 'no value' to 'highly valuable'. Respondents unfamiliar with any working manoeuvre terms were advised to not provide a rating.

The handlers were asked to score the value of 11 working attributes—*shows eye*, *control*, *initiative* (which we described as including working independently, *keenness* and willingness), *trainable* (including tractable), *intelligent* (including sagacious, brainy and clever), *calm*, *firmness* (including strength and power), *style of work* (including width), *physical suitability* (including stamina and durability), *anticipation* and *boldness*. These attributes' values were recorded for each working and competition environment within which the respondent had handled livestock herding dogs. Answer options included a semantic differential-type 5-point rating scale. Descriptive phrases 'no value' to 'extremely valuable' were used at either end of the scale. Respondents unfamiliar with any working attribute terms were advised to not provide a rating.

Respondents were asked to indicate the degree of expression of five general attributes they would expect to be present in the ideal dog for the working and competition environments in which they had experience. These attributes were *excitability*, *trainability*, *motivation and confidence*, *friendliness* and *cautiousness*. They were drawn, and slightly modified from, the "Big Five" personality traits identified by Ley et al. [32]. Respondents answered using a semantic differential-type 5-point rating scale ranging from 'none' to 'a very high degree' at each end point with 'a moderate degree' at the midway point. Respondents unfamiliar with any general attribute terms were advised to not provide a rating.

The final question asked handlers to indicate, on a 100-point visual analogue scale, the balance of shyness–boldness expression they would expect the ideal dog to exhibit for each herding context in which they had experience, where zero was maximum shyness and 100 was maximum boldness. Specifically, they were asked: *Please indicate, by moving the sliding scale*/*marking on the scale, the balance of shyness and boldness that the ideal dog would exhibit*. As a guide, the descriptive phrases 'extremely shy' and 'extremely bold' were used at either extreme of the scale. Respondents unfamiliar with the concept of shyness–boldness expression were advised to not provide a rating.

Genstat Version 16 (VSN International, Hemel Hempstead, UK) was used for statistical analysis of shyness–boldness expression results. REML analysis was performed using the variate means of shyness–boldness expression for each herding context: utility, mustering, yard and trial. This permitted assessment of whether differences in ideal shyness–boldness expression across the four contexts were significant.

Individual handler optima on the shyness–boldness expression in the ideal dog for each of the four herding contexts were gathered and descriptive statistics collated. The same approach was taken to the amount of five general attributes in the ideal dog; the 11 working attributes in their dogs; the 16 working manoeuvres and the training ease of each working manoeuvre.

To explore the influence of context on handler preferences, a hierarchical cluster analysis based on Gower distance (as all variables in this study were ordinal; this corresponded to Manhattan distance) was conducted using the hclust function of the R statistical and computing software [33].

A Pearson's chi-squared test was performed to assess the significance of the results.

#### **3. Results**

#### *3.1. Respondent Demography*

Of the 812 livestock herding dog handlers and breeders who completed the survey, 563 were male and 249 were female. Most respondents were aged 50–59 years (*n* = 213), followed by 60–70 years (*n* = 182), 40–49 years (*n* = 165), 30–39 years (*n* = 120), 20–29 years (*n* = 120), and over 70 years (*n* = 44). There were more respondents who did not breed dogs (non-breeders, *n* = 451) than who did (breeders, *n* = 361). More than half of respondents had acquired knowledge of dog training beyond 'on-the-job' experience (experience only, *n* = 302; further education, *n* = 510), but only 19 of these respondents had completed a certified course. The remainder reported having attended dog-training schools and/or read dog-training books.

#### *3.2. Herding Context Experience*

Among the 811 livestock herding dog handlers and breeders who selected having experience in one or more of the four herding contexts, the following contexts (and combinations of contexts) were selected: mustering/yard/utility (*n* = 241), mustering/yard (*n* = 169), utility (*n* = 125), mustering/yard/trial/utility (*n* = 96), mustering (*n* = 85), mustering/utility (*n* = 33), mustering/yard/trial (*n* = 14), yard (*n* = 13), mustering/trial (*n* = 10), mustering/trial/utility (*n* = 10), trial (*n* = 5), yard/utility (*n* = 4), trial/utility (*n* = 3), yard/trial (*n* = 2), and yard/trial/utility (*n* = 1). Totals for each herding context were: mustering (*n* = 658), yard (*n* = 540), utility (*n* = 513) and trial (*n* = 141).

#### *3.3. Preferred Balance of Shyness–Boldness Expression in the Ideal Dog*

The highest boldness expression in the ideal dog selected by respondents was in the yard context, followed by utility, mustering and trial (means of 79.12, 72.11, 69.77, and 65.43, respectively). Between contexts, differences in mean value were statistically significant between trial and mustering (*p* = 0.004), mustering and utility (*p* = 0.006), highly significant between utility and trial (*p* < 0.001) and yard and all other contexts (mustering/trial/utility, *p* < 0.001) (see Figure 1).

**Figure 1.** Boxplot of shyness–boldness expression in the ideal dog per herding context: the box spans the interquartile range of the values; middle 50% of the values lie within the box with a diamond indicating the mean. The whiskers extend beyond the box to represent the range of the data. Respondents marked shyness–boldness expression on a visual scale that used the descriptive phrases 'extremely shy' and 'extremely bold' at either end. \* *p* = 0.006 \*\* *p* < 0.001 \*\*\* *p* = 0.004.

#### *3.4. Amount in the Ideal Dog*

#### 3.4.1. Amount of General Attributes in the Ideal Dog for Utility Work

For the utility context, most respondents selected 'a very high degree' for the general attributes of being *trainable* (*n* = 351; out of 492) and *motivation and confidence* (*n* = 320; out of 491). Meanwhile, over half the respondents selected 'a moderate degree' (midway on the five-point scale) for *excitability* (*n* = 258; out of 491) and *cautiousness* (*n* = 245; out of 485) (see Figure 2).

**Figure 2.** Amount of five (5) general attributes in the ideal dog across the four herding contexts. Respondents' ratings: **A**—utility; **B**—mustering; **C**—yard; **D**—trial.

#### 3.4.2. Amount of General Attributes in the Ideal Dog for Mustering Work

For the mustering context, similar to the utility context, most respondents selected 'a very high degree' for*trainable*(*n* = 397; out of 619) and *motivation and confidence*(*n* = 406; out of 621). In comparison, for *excitability* (*n* = 317; out of 620) and *cautiousness* (*n* = 302; out of 616), no respondent selected more than 'a moderate degree' in the ideal dog (see Figure 2).

#### 3.4.3. Amount of General Attributes in the Ideal Dog for Yard Work

Most respondents who supplied data for the yard context, similar to both utility and mustering contexts, selected 'a very high degree' for *trainable* (*n* = 313; out of 485) and *motivation and confidence* (*n* = 348; out of 485). Unlike the utility and mustering context results, respondents selected increased amounts of *excitability* and similar *cautiousness* in the ideal dog (see Figure 2).

3.4.4. Amount of General Attributes in the Ideal Dog for Herding Trials

In the trial context, 107 out of 129 respondents selected 'a very high degree' for *trainability*. Additionally, more than half of respondents selected 'a very high degree' for *motivation and confidence* (*n* = 88; out of 127). *Excitability* was considered least useful in the trial context with 'none' selected relatively more by respondents (*n* = 40; out of 128) than the working contexts: mustering (*n* = 70; out of 620), yard (*n* = 41; out of 484) and utility (*n* = 32; out of 491) (see Figure 2).

#### *3.5. Value of Working Attributes*

#### 3.5.1. Value of Working Attributes—Utility

Respondents scored five working attributes as 'extremely valuable': *control* (*n* = 355; out of 490), *intelligent* (*n* = 350; out of 491), *initiative* (*n* = 349; out of 491), *trainable* (*n* = 347; out of 490), and *physical suitability* (*n* = 343; out of 490). *Style of work* received the fewest ratings of 'extremely valuable' overall (*n* = 186; out of 461). Most respondents selected one of the two highest ratings for *boldness* ('extremely valuable' *n* = 208, followed by the next point on the scale (unlabeled) *n* = 191; out of 485) (see Figure 3).

**Figure 3.** The value of eleven (11) working attributes across the four herding contexts. Respondents' ratings: **A**—utility; **B**—mustering; **C**—yard; **D**—trial.

#### 3.5.2. Value of Working Attributes—Mustering

*Initiative* and *Intelligence* were considered 'extremely valuable' by 489 and 476 (both out of 636) respondents. Of the 611 respondents who assigned scores to this context for *boldness*, 475 selected the two highest ratings (see Figure 3).

#### 3.5.3. Value of Working Attributes—Yard

Respondents considered the most valuable working attributes in the yard context *trainable* ('extremely valuable' *n* = 342; out of 501), *control* ('extremely valuable' *n* = 339; out of 509), *intelligence* ('extremely valuable' *n* = 326; out of 509), *firmness* ('extremely valuable' *n* = 323; out of 502) and *physical suitability* ('extremely valuable' *n* = 323; out of 506). They considered *eye* and *style of work* as being of least value, with responses more evenly spread across the ordinal scale compared to all other working attributes (see Figure 3).

#### 3.5.4. Value of Working Attributes—Trial

*Control* (*n* = 118; out of 130), *trainable* (*n* = 113; out of 131) and *calm* (*n* = 110; out of 131) were scored as 'extremely valuable' by most respondents when describing the ideal trial dog. *Boldness* was rated the least valuable of the working attributes assessed (see Figure 3).

#### *3.6. Value of Working Manoeuvres*

#### 3.6.1. Value of Working Manoeuvres—Utility

Respondents describing the ideal utility dog considered *cast* to be of highest value ('highly valuable' *n* = 365; out of 485 responses), and *bite* received the highest score for 'no value' ('no value' *n* = 177; out of 456). *Break*, *width*, *pull*, *lift* and *draw* received fewer than 300 responses, likely indicating that knowledge of these terms was limited to a sub-set of survey respondents (see Figure 4).

**Figure 4.** The value of sixteen (16) working manoeuvres across the four herding contexts: Respondent's ratings: **A**—utility; **B**—mustering; **C**—yard; **D**—trial.

#### 3.6.2. Value of Working Manoeuvres—Mustering

Six working manoeuvres were considered 'highly valuable' by most respondents in the mustering context: *Cast* (*n* = 492; out of 616), *gathering* (*n* = 415; out of 591), *cover* (*n* = 259; out of 463), *heading* (*n* = 344; out of 561), *hold* (*n* = 333; out of 575) and *balance* (*n* = 255; out of 458). *Bite* was considered to be of 'no value' (*n* = 242) or of limited value (*n* = 102) by most respondents (total *n* = 584). Responses on the value of *backing* were spread across the five-point scale ('extremely valuable' *n* = 117, *n* = 77, *n* = 77, *n* = 100, 'no value' *n* = 172; out of 543) (see Figure 4).

#### 3.6.3. Value of Working Manoeuvres—Yard

In the yard context, *force* and *bark* were considered 'highly valuable' (*n* = 404; out of 513, *n* = 295; out of 503) by most respondents. No one considered *back* 'highly valuable' (*n* = 0; out of 395). *Bite* was

scored as having 'no value' by nearly half of respondents (*n* = 208; out of 474). Similar to the mustering context results, *break*, *width*, *pull*, *lift* and *draw* each received fewer than 262 responses, compared to more than 361 each for the other traits (see Figure 4).

#### 3.6.4. Value of Working Manoeuvres—Trial

The highest value traits in the trial context were *cast* ('highly valuable' *n* = 126; out of 134), *balance* ('highly valuable' *n* = 106; out of 127), *cover* ('highly valuable' *n* = 106; out of 128), *heading* ('highly valuable' *n* = 96; out of 128) and *hold* ('highly valuable' *n* = 94; out of 127). Most respondents scored *bite*('no value' *n* = 74; out of 129) and *bark* ('no value' *n* = 68; out of 122) as 'no value' (see Figure 4).

#### *3.7. Trainability of Working Manoeuvres*

Among the 16 working manoeuvres, all but two traits were scored by respondents at the midway point between 'extremely easy' and 'almost impossible'. These were *force* (first two points at the end of the scale including 'extremely easy' *n* = 387; out of 748) and *heading* (first two points at the end of the scale including 'extremely easy' *n* = 371; out of 684), indicating respondents' possible awareness of these traits being innate behaviours. *Break* (*n* = 435), *width* (*n* = 397), *pull* (*n* = 366), *lift* (*n* = 361) and *draw* (*n* = 363) received fewer than 435 responses. For all the other traits measured, there were at least 569 responses each: *Cast* (*n* = 741), *force* (*n* = 748), *gather* (*n* = 730), *cover* (*n* = 588), *back* (*n* = 679), *bark* (*n* = 691), *bite* (*n* = 604), *heading* (*n* = 684), *hold* (*n* = 708), *balance* (*n* = 569), *drive* (*n* = 650) (see Figure 5).

**Figure 5.** Reported ease of training for sixteen (16) working manoeuvres.

#### *3.8. Cluster Analysis*

The hierarchical cluster analysis resulted in three groups: Group Three (most common preference pattern), Group Two (the smallest preference pattern) and Group One (intermediate to Groups Two and Three) (see Figure 6).

Group One owners prioritised cast, force, gathering, cover, heading, hold, balance, firmness, calmness, intelligence, trainability, initiative, control, anticipation, physical suitability and confidence (see Figure 7).

Group Two owners prioritised *cast*, *gathering*, *cover*, *heading*, *control* and *intelligence*. *Back*, *bark* and *excitability* were clearly not preferred in this group (see Figure 7).

Group Three owners prioritised a more balanced approach to the value and amount of each trait. They were less concerned with *draw*, *lift*, *pull*, *width* and *break* (see Figure 7).

Across all three groups, *bite* was consistently less in demand than the other traits analysed.

**Figure 6.** Cluster dendogram showing hierarchical, agglomerative clustering of Euclidean Distances of preference scores for analysed traits. The type of work or competition for which a trait was favoured by respondents is indicated by: Green = utility, Red = mustering, Blue = yard, Yellow = trial. Dendogram shows, from left to right, Group Three, Group Two, Group One.

**Figure 7.** Cluster group preferences by trait: Boxplot demonstrating the different medians, interquartile ranges and whisker lengths of preferences for each trait in the three clusters of respondents shown in Figure 6.

When the three groups were analysed across the four herding contexts (see Figure 8), using Pearson's chi-square test (chi-squared = 136.21, df = 6), highly significant (*p* < 0.01) preferences

were apparent for particular contexts. Group One's trait preferences were overrepresented for utility and underrepresented for mustering. Group Two's trait preferences were overrepresented for mustering and trial, while underrepresented for yard and utility. Group Three's preferences were overrepresented for yard and underrepresented for mustering and trial.

**Figure 8.** Associations of each cluster group with working and competition contexts. Each group would be expected to have equal representation across each herding context; the chart indicates which group's preferences were over or underrepresented for each herding context.

#### **4. Discussion**

Although surveys of owners and/or experts have been previously used to develop behavioural profiles of companion dog breeds [34,35], to our knowledge, this is the first to assess the relative value of personality and working traits in livestock herding dogs.

By identifying the most valuable working and personality traits across multiple herding contexts, the current results help to show how these traits influence successful movement of livestock while also identifying traits that could enhance both context-specific and general breeding programs. Genetic analysis from the hunting dog sector in Sweden has shown that, for at least six traits, if Best Linear Unbiased Prediction breeding values were used instead of phenotype, genetic gain would be 89% higher [36]. With the sequencing of the canine genome [37], molecular genetics provides the opportunity to identify suitable livestock herding dogs at an earlier age than behavioural assessment currently offers.

Of the four herding contexts surveyed, mustering was selected as the context in which most respondents had experience with handling livestock herding dogs. This was followed by yard and utility, with only a small number of respondents involved in the trial (competition) environment. These distributions reflect prevalent Australian working conditions in that livestock are often collected over large areas of farmland prior to being managed within yards. Whether this Australian distribution of potential respondents and the current resultant data are similar in other large-scale livestock-producing countries requires further investigation.

#### *4.1. Shy–Bold*

The ideal reported balance of shyness–boldness expression significantly differed across the four herding contexts. Livestock herding dogs working in the yard context were considered by respondents to require a higher level of *boldness* than in the utility and mustering contexts, followed by the trial context. However, the middle 50% of responses overlapped between each context. This finding reflects the necessary blend of confidence, motivation, composure and resilience that presumably make up the term boldness, the so-called super trait that yard dogs require to be successful in this close-up, threatening, high risk environment. The expression of *boldness* appears to be a simple way for respondents to differentiate between the type of dog required between each herding context.

#### *4.2. General Attributes*

For general attributes, handlers and breeders across all herding contexts reported that the ideal dog has a high degree of *trainability*, *motivation and confidence* and *friendliness*. The cluster analysis also identified similarities across the three groups for the amount preferred of these traits. The results for *excitability* highlighted one of the more interesting findings for the general attributes group. Across all contexts and the three cluster analysis groups, respondents identified this as the 'Goldilocks' trait—not too little, not too much, but 'just right'; the term Goldilocks principle or effect is referred to in other research fields including economics and education to describe contexts that seek balance [38,39]. Finding a balance for *excitability* is not unique to livestock herding dogs, for example it is also noted in guide dogs [40]. However, improvements in selective breeding for this trait may assist future research about behavioural tendencies being undertaken among service dogs [41,42].

Respondents' preference for a dog that is easy to work with, through long hours, and often as the handler's only companion in conducting their work has implications for the welfare of these dogs. Specifically, livestock herding dogs that do not meet handlers' expectations of being 'trainable, motivated, confident and friendly' could be at more risk of becoming so-called behavioural wastage (being discarded from the industry because of poor performance related to behaviour, rather than physical inadequacy) [43]. 'A moderate degree' of *excitability* and *cautiousness* was preferred, reflecting respondents' awareness of the potential for high levels of these traits to compromise the successful working performance of a dog. This preference for trait-specific expressions of certain qualities is a complex finding. A common difficulty in breeding or selection of livestock herding dogs is consistency between what one handler or breeder and another considers to be 'the right amount' of a trait expressed in a dog.

#### *4.3. Working Attributes*

As far as working attributes are concerned, the current study revealed both similarities and differences between the four herding contexts and confirmed, by the cluster analysis, the key working characteristics preferred and valued by handlers. Across all contexts, the high values assigned to *control* and *trainable* indicate the importance of these traits in allowing handlers and breeders to breed, rear and train the best dogs for each herding context. While dogs working at a distance from their handler are not unique to livestock herding, what is specialised to livestock herding dogs is the concurrent gathering and movement of livestock to the handler. The understanding of respondents that dogs in this context pose a risk to themselves (e.g., placing themselves in dangerous positions leading to injury) and the livestock (e.g., injuring livestock due to poor herding technique) should not be underestimated. For both mustering and utility contexts, respondents reported that dogs with *initiative*, *intelligence* and *physical suitability* were of most value to them. These traits reflect the complex and demanding nature of work in these contexts and the requirement for handlers to both direct their dogs, when needed, and rely on them to perform independently, as required. In the yard context, most respondents also assigned high value to *firmness* and *physical suitability*. *Boldness*, a personality trait often referred to in the peer-reviewed canine behaviour literature (e.g., [44,45]), was considered of less value than

the other working attributes in the utility, mustering and trial contexts, yet was one of the highest value attributes in the yard context. These results reflected those obtained from the shyness–boldness expression question. Similarly, *shows eye*, a core attribute of herding dogs that has clear analogues in the predatory sequence [8], was considered of high value only in the trial context. These apparent anomalies suggest that the demands of herding work peak within the broad yard context, which often requires dogs to move fearful livestock at a close distance, through tight spaces, repeatedly.

#### *4.4. Working Manoeuvres*

The value of the 16 working manoeuvres across the four herding contexts revealed some key similarities between the four herding contexts. In the utility, mustering and trial contexts, *cast*, *force*, *gathering*, *cover*, *heading*, *hold*, *balance* and *drive* were consistently of high value to respondents whereas *force* and *bark* were of extreme value to respondents compared to the other manoeuvres. While these results are not surprising, they represent a selection of manoeuvres on which breeders or handlers may focus to ensure their dogs meet expectations when working. Notable exceptions were *bite* and *backing*. Across all four contexts, many respondents assigned 'no value' for *bite* while *backing* was assigned 'no value' by many respondents in the mustering and trial contexts, reflecting their relative unhelpfulness in these contexts. Additionally, throughout the cluster analysis, *bite*, being the trait of least value, revealed awareness among respondents that the ideal dog should show limited expression of this trait.

The current study clearly identified manoeuvres whose reported value differed from one context to the next. For example, *cast* was rated as the most valuable in utility, mustering and trial contexts, but as only of general value in the yard context. Many respondents consistently rated *bite* as having 'no value' across each context.

The cluster analysis provided further insight into respondents' thoughts on working manoeuvres. Among handlers, there appear to be two main groups who require all-rounder type dogs. The first group (Group One) were those who have a clear preference for a select group of traits they need to perform the work with a focus on the utility context. This group also appears to have a keen interest in the skills their dogs need to be successful. The second group (Group Three) were those handlers who want 'jack of all trades' dogs that possess a broad range of working manoeuvres and consider most manoeuvres to have only moderate value. This group reported primarily on the yard context. Further studies are required to differentiate between the first group, who may be experienced students of livestock herding dogs with well-developed ideas of the most valuable traits to have a successful dog, and the second group, who seem to represent the majority of yard-based Australian livestock herding dog handlers who are primarily outcomes-focused. These results have provided an additional layer to this survey's findings: that, while there appears to be a move towards a need for specialist livestock working dogs on the whole, generalities remain and there is still a need for those who primarily work in the yard context to have dogs with a broad, albeit average, skill set.

#### *4.5. Ease of Training Working Manoeuvres*

When respondents were asked how easy or difficult it is to train a group of working manoeuvres, the hypothesis was that respondents of the survey would select innate behaviours as being more difficult to train. The results indicate that, overall, the manoeuvres were neither easy nor difficult to train. This may reflect our use of scales that had insufficient granularity to detect real differences between ease of training. Two manoeuvres that appear to be elements of the predatory sequence, *force* and *hold*, were reported by respondents to be easier to train than the other manoeuvres. It is possible that some handlers with a good understanding of livestock herding dog behaviour may find innate behaviours easier to train or shape, while others may credit their own training skills for any apparent ease of training certain innate behaviours. For accomplished trainers, creating scenarios for a livestock herding dog to practice and trigger these innate behaviours may make them relatively easy to train by simply fine-tuning a natural behaviour. In Australia, mean failure rates of herding dogs in training have been estimated to be at least 20% [2]. Improving understanding of the practical

application of behavioural science findings as they pertain to livestock herding dogs may boost training outcomes [46]. This approach has met with considerable success in horse training through the nascent discipline of equitation science [47]. Additionally, identifying consistently successful trainers with low failure rates, who can provide assistance to fellow trainers through sharing their knowledge, may also have merit.

While the current results provide some helpful information on the relative value of working and personality traits in the four herding contexts, they do contain some limitations. Among all three non-competition contexts, the manoeuvres *break*, *width*, *pull*, *lift* and *draw* were unfamiliar to many respondents in the working context. This is most probably because these terms are often used in competition and technical literature but rarely in the practical, working contexts. An additional limitation is that the current survey did not explore the specific reasons for why individual manoeuvres were either of high or low value to each respondent or why some respondents did not assign a rating or score to a given quality. While the terms presented in the current survey were identified as those most commonly used in relevant manuals [26], there is scope to further investigate why these working manoeuvres provide value within each herding context.

Furthermore, due to the large number of questions, it was apparent that respondents missed answering parts of questions, developed survey fatigue or were not familiar with some of the terms or traits presented. Improved survey design, including attempts to verify whether respondents deliberately failed to answer part of any given question, may assist in reducing missing data. A further limitation that we acknowledge as a potential source of lost detail among the data arose in the general attributes questions where, for consistency with Ley et al. [32], the personality dimensions *motivation* and *confidence* were grouped together. Because of this, it was not possible to determine whether respondents were selecting just one or both terms in their response. As such, this result should be interpreted with some caution. The question on training ease of the 16 working manoeuvres attempted to tease out respondents' understanding of the difference between innate and learnt behaviours in relation to livestock herding. While the results indicate some variability in understanding by respondents, a published analysis of a different section of results from the same survey, focusing on dogmanship, found a general lack of understanding of learning theory and training principles among the current respondents [27]. Further investigation into survey respondents' understanding of the behavioural underpinnings of these working manoeuvres may assist interpretation of this section of the survey.

There is limited research analysing the effect of sex on behavioural traits in livestock working dogs. While data on sex in relation to the behavioural traits in this study was not collected, *confidence* and *boldness* are examples where sex differences could be explored in future studies. Additionally, sex differences in training ease among livestock working dogs may provide interesting comparisons to other working dog sectors [48].

#### **5. Conclusions**

This survey identified preferred levels of boldness and those general attributes, working attributes and working manoeuvres of greatest value to most Australian handlers and breeders across four herding contexts. These results highlight similarities in the attributes and manoeuvres valued across these contexts but also the need for dogs working in an individual context to develop specialised skills. However, there was, among the respondents, a sub-group of handlers with a focus on the yard context, who need dogs with a broad range of skills that are easy to work with. The most valued general attributes were *trainable*, *friendly*, *motivation* and *confidence*. The most valued working attributes were *control* and *trainable* while the most valued working manoeuvres included *cast*, *force*, *cover* and *gathering*. Further investigation is required to explore handlers' understanding of the distinction between innate and learnt behaviours in training livestock herding dogs.

**Author Contributions:** Conceptualization, J.E., E.A., C.W. and P.M.; Data curation, J.E. and E.A.; Formal analysis, J.E., E.A., B.W., C.W. and P.M.; Funding acquisition C.W. and P.M.; Investigation, J.E., E.A., Claire Wade and Paul McGreevy; Methodology, J.E., E.A., B.W., C.W. and P.M.; Project administration, C.W. and P.M.; Resources, J.E., E.A., C.W. and P.M.; Supervision, C.W. and P.M.; Validation, J.E., E.A. and B.W.; Visualization, J.E. and B.W.; Writing—original draft, J.E.; Writing—review & editing, J.E., E.A., B.W., C.W. and P.M.

**Funding:** This study was supported by a grant from Agrifutures (previously the Rural Industries Research and Development Corporation) (Grant number, PRJ 007 806), including Meat and Livestock Australia.

**Acknowledgments:** In-kind support from the Working Kelpie Council of Australia. The authors also thank the 2013 Casterton Kelpie Auction Committee for their support of this study.

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

#### **References and Notes**


© 2019 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/).

## **Development of a Spatial Discount Task to Measure Impulsive Choices in Dogs**

**Paolo Mongillo 1, Anna Scandurra 1,2, Carla Jade Eatherington 1, Biagio D'Aniello <sup>2</sup> and Lieta Marinelli 1,\***


Received: 12 June 2019; Accepted: 13 July 2019; Published: 23 July 2019

**Simple Summary:** Impulsivity is believed to play a role in problematic behaviors in dogs. In this study, we developed a test to assess dogs' tendency to make impulsive choices, that is their preference for smaller immediate reward instead of larger, but harder to obtain ones. Dogs were first trained that a bowl presented on a certain side always contained a large food amount, whereas the one presented on the opposite side (although at the same distance from the dog) contained less food. Then, the bowl with less food was progressively placed closer to the dog. As expected, dogs' choices to feed from the bowl with less food increased as the distance of the latter decreased. Choices did not depend on factors that could interfere, such as dogs' level of motivation for food, training experience, or learning ability. This indicates that the test is likely to be actually assessing impulsivity, not other traits. Also, female dogs were more likely to make impulsive choices than males, in accordance with what is known in humans and rodents, supporting the validity of the test. The test was completed in less than 1 h, making it a valid option to assess impulsivity in dogs in various contexts.

**Abstract:** Impulsive choices reflect an individual's tendency to prefer a smaller immediate reward over a larger delayed one. Here, we have developed a behavioural test which can be easily applied to assess impulsive choices in dogs. Dogs were trained to associate one of two equidistant locations with a larger food amount when a smaller amount was presented in the other location, then the smaller amount was placed systematically closer to the dog. Choices of the smaller amount, as a function of distance, were considered a measure of the dog's tendency to make impulsive choices. All dogs (N = 48) passed the learning phase and completed the entire assessment in under 1 h. Choice of the smaller food amount increased as this was placed closer to the dog. Choices were independent from food motivation, past training, and speed of learning the training phase; supporting the specificity of the procedure. Females showed a higher probability of making impulsive choices, in agreement with analogue sex differences found in human and rodent studies, and supporting the external validity of our assessment. Overall, the findings support the practical applicability and represent a first indication of the validity of this method, making it suitable for investigations into impulsivity in dogs.

**Keywords:** dog; behavioral test; impulsivity; sex differences; learning; validation

#### **1. Introduction**

Impulsivity is generally referred to as the tendency to act prematurely, without forethought or consideration of the consequences [1], or as the failure to defer gratification [2]. In humans, impulsivity has been indicated as a vulnerability factor for a range of maladaptive behaviours, including substance abuse, gambling, or pathological conditions such as attention deficit and hyperactivity disorders [3,4]. Although impulsivity is sometimes measured as a single dimension of personality, it is best described

as a multidimensional trait [4]. Many studies converge on the recognition of two broad classes of impulsive behaviour, namely impulsive actions and impulsive choices [5,6]. The former is regarded as the result of an inability to inhibit or stop a motor act in response to prepotent stimuli. Behavioural paradigms such as the go/no-go task, or the stop-signal reaction time task, analogue versions of which exist for humans and rodents, are designed to pinpoint this behavioural facet of impulsivity [7]. Impulsive choices instead reflect an individual's preference for smaller immediate gratifications over delayed ones of greater value or quantity [8]. This dimension of impulsive behaviour is typically assessed in delay-discounting tasks, which measure the maximum delay tolerated by individuals who are informed of (for humans) or trained to expect (for animal paradigms) the possibility to obtain higher value rewards if they can wait for sufficiently long time-intervals [7,9]. These tasks do not simply represent different measures of a single construct. There is evidence that these two measures are independent [10,11], and that they are underpinned by different neurobiological mechanisms [12–14]. Also, they are differently related to individual characteristics, such as sex and age. For instance, while a tendency to perform impulsive actions is blandly, if at all, associated with male sex, robust associations exist between sex and impulsive choices, where females discount more steeply than males in both humans and rodents [15]. Sex differences are believed to root in differential activation of the dopaminergic signalling system between sexes, which mediate subjects' sensitivity to rewards. Less clear is the interplay between these mechanisms and circulating gonadal hormones, the role of which impulsive choice behaviour still has to be clarified [15]. As regards age, evidence indicates a higher tendency to express impulsive choices during adolescence/young adulthood, than later in life [16].

The current knowledge about impulsivity comes mostly from studies in humans and rodents. However, the same construct has been tentatively applied to dogs, where high impulsivity is thought to be a correlate of different maladaptive behavioural manifestations or cognitive processes. For instance, impulsivity may play a role in aggression [17,18], and more generally in the expression of behavioural problems [19]. Furthermore, some evidence suggests that impulsivity is associated with lower problem-solving abilities [20,21]. As it occurs in the human literature, methods used to assess impulsivity in dogs vary in scope and methodology. A plethora of tasks that were proposed as assessments of dogs' impulsivity actually represent (tentative) measures of impulsive actions, including reversal learning tasks, the A-not-B task, the cylinder task the middle-cup task, the wait-for-treat task, and buzzer tasks [21–26]; although a thorough description of these paradigms fall outside the scope of this paper, all encompass the necessity to withhold a prepotent response, either spontaneous or learnt. A much smaller variety of tasks assess impulsive choices. Although with some variations in the nature or the source (social or non-social) of the reward, these methods are based on the same general paradigms which measure dogs' ability to tolerate temporal delays on the expectation of a larger/more valuable reward [17,25,27–29]. A common disadvantage of these delay-discounting tasks is that they generally require dogs to undergo a long training (in most cases lasting more than one day), which also makes it difficult to complete the test as proved by a relatively low success rate (e.g., 58.8% [19]; 51.4% [30]). This obviously represents a strong limit to the practical applicability of these tasks, and to the possibility of administering them routinely to large dog samples.

There is accumulating evidence that measures provided by these methods are in many cases uncorrelated [23,30]. Moreover, there is variability in terms of how the outcomes of these tasks relate to the broader, indirect assessment of impulsivity provided by owners' answers to a questionnaire (Dog Impulsivity Assessment Scale, DIAS [18]), which range from no correlation [29,30], to correlation with one of the DIAS subscales [19] or with the overall DIAS score [19,30]. Although this lack of consistency may reflect the complex, multidimensional nature of the construct, it nonetheless prompts us to question which of these tasks provide a valid and easy measure of impulsive behaviours in dogs. In only a few cases, attempts have been made to assess impulsivity as a function of external variables known to influence impulsivity measures in other species, such as sex or age (see for instance [30]). However, none of the aforementioned studies addressed problems of potential intervening variables,

including dog's motivation for food, learning abilities, and previous experience, which may represent confounds in the outcome of the assessments, as highlighted by some of the very same authors [30].

Upon these premises, in this study we aimed to develop and validate a task to assess dogs' tendency to express impulsive choices. In view of the possibility to administer the task to large dog samples, one of the requisites of the task was to be successfully easily completed by most dogs, and within a reasonably short time (i.e., a single session, no longer than 1.5 h). To circumvent the difficulties associated with training the dogs to wait in the classical delay-discounting tasks, here the immediacy of the possibility to obtain the smaller reward was operationalized as a smaller space to travel, rather than as a shorter time to wait (although space differences inherently imply a time difference [31]). In the lack of a gold standard that could provide an external validation measure, we aimed at providing a first assessment of the tasks' validity, by (a) looking at the psychometric relationship between the task contingencies and dogs' performance, (b) excluding effects of other intervening factors, namely the dogs' previous training history, level of food motivation, and the learning requirements of the task, (c) assessing the tasks sensitivity to biological factors that are known to influence impulsive choices in other species, namely sex, reproductive status and age, and (d) looking at the relationship between the outcomes of the task and indirect measures of impulsivity provided by the DIAS questionnaire.

During the writing of this paper, results of the development of an analogous paradigm, independently developed by Brady and collaborators [32], came to our attention. Like the method described in the present paper, the task was a spatial version of the classical delay-discounting task. The procedure involved a single test session, preceded a short pre-training phase, and was completed by dogs in one day. The short time requirement, and a training success rate of 96% (24 out of 25 dogs), provide excellent indications in terms of feasibility of this kind of procedure. As far as validation was concerned, the primary means of validation reported in the study were the assessment of test-retest reliability and correlations with a score of the DIAS. On the other hand, the study did not look at factors included in our investigation, and highlighted by the very same authors as potential confounds in their results. In this sense, the results reported in the present study represent fundamental additional indications about the validity of this spatial-discounting task.

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

#### *2.1. Subjects*

Forty-eight pet dogs were recruited for this study through advertisement in veterinary clinics and the University of Padua. Apart from being healthy, no specific criteria for inclusion in the study were required. The sample included 15 mongrels of small (≤30 cm at the withers, N = 2), medium (>30 and ≤55 cm; N = 9) and large size (>55 cm; N = 4), and 33 pure breed dogs (N = 7 Border Collies, N = 4 Australian Shepherds, N = 3 Golden Retrievers, N = 2 Beagles, N = 2 Cocker Spaniels, N = 2 Labrador Retrievers, N = 1 American Staffordshire Terrier, N = 1 Bernese Mountain Dog, N = 1 Breton, N = 1 Czechoslovakian Wolfdog, N = 1 Dachshund, N = 1 German Shepherd, N = 1 Greyhound, N = 1 Hovawart, N = 1 Labradoodle, N = 1 Newfoundland, N = 1 Rhodesian Ridgeback, N = 1 Samoyed, N = 1 Siberian Husky). Recruitment was aimed at forming four groups of equal size based on the dogs' sex and reproductive status, namely: non-orchiectomized males (mean age ± SD: 4.4 ± 3.2 years, min = 1, max = 12), non-ovariectomized females in dioestrous or anoestrous phase (mean age ± SD: 4.7 ± 2.7 years, min = 1.5, max = 11), orchiectomized males (mean age ± SD: 4.8 ± 2.3 years, min = 1, max = 10) and ovariectomized females (mean age ± SD: 4.8 ± 1.5 years, min = 2, max = 9). Dogs of the last two groups had their gonads removed at least 6 months prior to participating in the study. The owners were asked to indicate if their dogs had any previous experience of training, choosing between four options (no training, basic training with no professional support, obedience training with a professional trainer, training to specific activities with a professional trainer). Finally, owners were asked to evaluate their dog's food motivation, as high (would always eat if given the chance, eats most types of food, never leaves food in the bowl, fights for food), medium (sometimes leaves

food in the bowl, eats many, but not all types of food, does not fight for food), or low (always leaves some food in the bowl, only eats some specific types of food, never fights for food). The distribution of training history and food motivation within each of the four experimental groups is reported in Table 1. Owners were asked to not feed their dogs on the day of the experiment.


**Table 1.** Distribution of categories of Training history and Food motivation within groups of dogs of different sex and reproductive status.

#### *2.2. Impulsivity Evaluation Questionnaire*

Owners were asked to fill out an Italian translation of the DIAS. This required owners to indicate their degree of agreement with the proposed statements, according to a score scale from 1 (complete disagreement) to 5 (complete agreement). For each dog, an Overall Questionnaire Score (OQS) was calculated as the average score obtained in all items. Moreover, for the sake of comparison with other studies, average scores were calculated for three sub-scales corresponding, in terms of item composition, to the three factors described by Wright and collaborators [18]. However, a factor analysis performed on the data collected in the current study resulted in a very different factorial structure (data not reported), thus the sub-scales used in this study could not be described using the same names adopted elsewhere.

#### *2.3. Experimental Setting*

Tests were conducted at the Laboratory of Applied Ethology (Department of Comparative Biomedicine and Food Science, University of Padua) in a room of approximatively 5 × 5 m, equipped with a chair behind a curtain (140 cm high and 160 cm wide) and two plastic panels (24 × 38 cm), placed vertically at a maximum distance of 360 cm from the chair (the actual distance depended on the experimental phase, as detailed below) and 80 cm apart (Figure 1). The panels represented placeholders for positioning food bowls (circular metal bowls, 20 cm in diameter) during the experiment and concealed the bowls from the dog's view, while the curtain served to temporarily conceal the actions of the experimenter from the dog's view during the experimental procedures.

#### *2.4. General Procedure*

The test was based on a two-alternative forced choice, between two different quantities of food, in a ratio of 1 to 7. Prior to beginning the test, the dog and owner were taken into the room, and the dog was left free to explore and familiarize itself with the experimental setting and the experimenter for approximately 5 min. During this time, an experimenter explained the procedure to the owner. Then, the owner was invited to attach the leash to the dog and sit on the chair, and the experimental procedure began.

The experiment comprised a Pre-training phase, a Training phase, and a Test phase. All phases were composed of a number of consecutive trials following a similar procedure: the owner sat on the chair behind the curtain, holding the dog next to him/her. In a separate room, the experimenter baited the bowls with 7 food pieces (each being 1/4 of a ring of Frolic®, a commercial semi-humid dog food) of in one (S+) and 1 piece in the other (S−). Then, she entered the experimental room, placed the bowl(s) behind each plastic panel, walked towards the curtain and opened it, allowing the dog to see the two plastic panels. At that point, the experimenter walked behind the owner and placed a hand on the owner's shoulder, which signalled that the dog could be released. The dog was allowed to reach only one of the two bowls, so as soon as the dog approached one bowl, the experimenter removed the other bowl, preventing the dog from eating its content. As soon as the dog ate the food, the selected bowl was also removed. Finally, the owner took the dog back to the starting position, and the curtain was again lowered, the experimenter went into the separate room to prepare the bowls for the next trials. If the dog did not make a choice within one minute, bowls were removed and the trial was considered null.

**Figure 1.** Experimental setting. Representation of the experimental setting, illustrating the owner's and dog's position behind the curtain (large horizontal grey bar) at the start of presentations, and the position of the bowls containing the larger (S+) and the smaller amount of food (S−) during training (P0) and test trials (P0 for S+, P0 to P4 for S−).

#### *2.5. Pre-Training Phase*

The aim of the Pre-training phase was to allow the dog to familiarize with the experimental procedure and experience that bowls in different location contained different amounts of food. This phase consisted of 6 trials, which followed the procedure described above, with the difference that only one food bowl was presented in each trial (S+ was presented on 3 trials, and S− on the other 3). In this phase, the food bowls were placed at the distance of 350 cm from the dog. For any given dog, S+ was always presented on the same side thorough the test, and S− on the opposite side. The side of presentation varied between subjects, and was counterbalanced within each of the four experimental

groups. To be admitted to the training phase, dogs needed to promptly eat the food from the presented bowl in each of the 6 trials.

#### *2.6. Training Phase*

This phase was meant to teach dogs to choose the bowl containing the larger amount of food when both S+ and S− were presented simultaneously. Both S+ and S− bowls were placed at the same distance (P0, 350 cm from the dog). For each dog, S+ and S− were placed on the same side as in the Pre-training phase. A maximum of 30 trials were presented, and the criterion for passing this phase was to choose S+ in 6 consecutive trials. If a dog did not reach the learning criterion within the 30 trials, it was excluded from further testing. Before the test phase began, the owner was allowed to walk outdoors with her/his dog for 10 min.

#### *2.7. Test Phase*

The test phase was aimed at verifying the effect of increasing proximity of the smaller amount of food on dogs' choice. The rationale for the test was that lower levels of impulsivity would result in dogs' higher ability to choose the larger amount of food, despite the progressively higher proximity of the smaller food amount. The test phase consisted of 14 trials, which followed the general procedure, with the exception that, while S+ was always placed at the distance of 350 cm, the proximity of S− from dogs was systematically increased along a geometric progression. Specifically, there were three levels of increasing proximity: P1 (proximity increased by 40 cm compared to P0; distance of S− from the dog: 310 cm); P2 (proximity increased by 80 cm; 270 cm from the dog); P4 (proximity increased by 160 cm; 190 cm from the dog). Each of these three levels was presented three times among the fourteen trials; in the remaining 5 trials the distance of S+ and S− from the dog was the same (P0, 350 cm from the dog) as in the Training phase. The trials were randomly presented, with the constrain that S− could not be presented at the same distance in consecutive trials.

#### *2.8. Data Collection and Analysis*

All experiments were recorded by two ceiling-mounted cameras and coded with the Observer XT software (Ver.12.5, Noldus, Gröeningen, The Netherlands). In the Training and the test phases the dog's choices were codified as S+, S−, or null.

The analysis of dogs' choices in the Test phase aimed to provide an indication regarding the validity of the procedure. To this aim, the analysis was meant to verify that dogs' ability to choose the larger food amount decreased as a function of the proximity of the smaller food amount. In addition, to obtain an indication about the specificity of the measure, the analysis was meant to exclude that the dogs' performance reflected non-impulsivity related factors, such as different levels of motivation towards food, the dogs' learning ability in acquiring the initial discrimination task, or the dogs' training level. Finally, the analysis was aimed at highlighting possible differences in performance linked to the dogs' age, sex and/or reproductive status, in accordance with associations between these factors and impulsivity reported in the literature, as an indication of the external validity of the procedure.

Training history was unevenly distributed across groups of different sex and reproductive status, making it impossible to include the variable in the model described below. To achieve a better distribution, we recoded the variable using the following two levels: "non professionally trained dogs", which included untrained dogs, and dogs trained without support of a professional trainer, and "professionally trained dogs", which included all other dogs. Prior to such recoding, we ascertained that training history had no main effect on dogs' probability to make impulsive choices. To this aim, a Generalized Linear Mixed Model (GLMM) was used, which included the dog's choice of S+ or S− as a binary dependent variable, the dogs' ID as a random variable accounting for the repeated measurement within each dog, and training history as a four-level factor. As the GLMM revealed no significant main effects of training (*p* = 0.217), the variable was recoded which was used in the analysis described below.

To ascertain the specificity and external validity of our task, a GLMM was used, which included the dog's choice of S+ or S− as a binary dependent variable, and the dogs' ID as a random variable accounting for the repeated measurement within each dog. Separate models were run to investigate the effect of sex and reproductive status: one model was run on data collected from non-gonadectomized males and females and included the dog's sex as fixed factor; the other two models were run on data collected respectively from females and males and included the reproductive status as a fixed factor. In addition to sex or reproductive status, the model included distance of S−, the dogs' training history, and food motivation, as two-level fixed factors, and the dog's age and number of trials to reach the learning criterion in the Training phase as covariates. First-order interactions between S− distance and each of the other fixed factors were also included in the model. A stepwise backwards elimination procedure was used to eliminate non-significant interactions. Post-hoc comparisons were run between factor levels when a significant effect was found for a factor, applying a sequential Bonferroni correction.

As the analysis revealed a significant effect of sex on dogs' choices of S+ in the Test phase (see Results), a one-way ANOVA was performed to ascertain that there were no differences between dogs of different sex or reproductive status on their ability to acquire the initial Training phase, as measured by the number of errors made and in the number of trials needed to reach the learning criterion in such phase.

In order to further exclude that the dogs' performance in the test reflected their learning ability in initial discrimination training, Pearson's correlations coefficients were calculated between the number of trials needed to reach the learning criterion in the Training phase and the percentage of choices of S+, both across the entire test and at each different distance of S−.

Finally, as a further way to assess the relationship between the measure provided by the proximity test and other putative measures of impulsivity, Pearson's correlation coefficients were calculated between the DIAS OQS, and the DIAS sub-scales scores, and the number of trials to reach criteria in the Training phase, the percentage of choices of S+ in the Test phase.

All analysis was run with SPSS (ver. 23, IMB, Armonk, NY, USA). A value of 0.05 was adopted as threshold for statistical significance.

#### **3. Results**

#### *3.1. Pre-Training and Training Phases*

All dogs involved in the study successfully passed the Pre-training and the Training phase. In the latter, dogs reached the training criterion with an average (± SD) of 10.8 ± 6.6 trials and made an average of 7.7 ± 3.0 choices of S+ and 3.1 ± 4.7 choices of S−. The mean ± SD number of errors (choices of S−) and of trials required to reach the learning criterion by dogs split by sex and reproductive status is reported in Table 2; the one-way ANOVA indicated that there were no differences between sexes in the number of errors (F = 1.09; *p* = 0.36) or trials required to reach the learning criterion in this phase (F = 0.56; *p* = 0.64). No null trials (i.e., a dog not approaching any of the two bowls) were observed by any dog. The Training phase was completed in an average of 14.6 ± 6.5 min (min: 5.5; max: 28.1).

**Table 2.** Mean ± SD number of errors (choices of S−) and trials required to reach the learning criterion (TTC) in the Training phase by dogs of different sex and reproductive status.


#### *3.2. Test Phase*

Overall, dogs chose S+ in 75.8% of trials (mean N of trials ± SD: 3.79 ± 1.47 out of 5) when S− was presented at distance P0, 61.8% of trials (1.85 ± 1.15 out of 3) at P1, 35.4% (1.06 ± 1.25 out of 3) at P2, and 25.7% (0.77 ± 1.17 out of 3) at P4. The average of S+ choices for each of the four groups of different sex and reproductive status, at the different S− distances are summarized in Table 3. No null trials were observed in this phase. The Test phase was completed in an average of 15.4 ± 2.7 min (min:10.6; max: 24.4).

**Table 3.** Mean ± SD number of S+ choices in the Test phase by dogs of different sex and reproductive status. In brackets: mean ± SD percentage of S+ choices on the total number of trials for each distance (i.e., 5 for P0, 3 for P1, P2, and P4).


Table 4 summarizes the results of the three GLMM models, investigating the effect of S− distance, speed of acquisition of the Training phase, and dogs' age and training history, food motivation and sex/reproductive status, on dog's probability of choosing S+.

**Table 4.** Results of the Generalized Linear Mixed Models investigating the effect of the distance of the bowl with the smaller amount of food, the dog's sex or reproductive status (investigated in separate models, and with different data subsets), age, food motivation, type of training received, and number of trials needed to reach the learning criterion in the Training phase (TTC). Only significant first-order interactions between distance and other factors are reported. IF = intact females, IM = intact males, OF = ovariectomized females, OM = orchiectomized males. Subscript numbers indicate the numerator and denominator degrees of freedom, respectively.


All models evidenced an effect of the distance of S− on dog's probability of choosing S+, which generally decreased as S− was placed closer to the dog. When data from intact dogs were analysed, an effect of the interaction between the distance of S− and the dog's sex was found (Figure 2). Post-hoc analysis revealed a significant difference between males and females in the probability of choosing S+ at distance P1 and, while in females the probability already decreased when S− was moved from P0 to P1, in males the first significant drop in probability was only observed when S− was moved from P1 to P2.

**Figure 2.** Choices of S− as a function of sex and distance. Generalized Linear Mixed Model (GLMM) mean estimates of the probability of choosing S+ as a function of the distance of S−, by intact male and female dogs. Shaded areas represent the lower and upper 95% confidence intervals. Different capital letters indicate significantly different probabilities between sexes and different levels of proximity of S− (*p* < 0.05) after sequential Bonferroni corrected post-hoc comparisons.

Models using data from the whole group of male dogs, and from the whole group of female dogs, revealed no effect of reproductive status on the probability of choosing S+ as a function of the distance of S−. None of the three models found any effect of the dog's age, training history, food motivation, or speed of acquisition of the Training phase.

#### *3.3. Correlations of Test Outcomes with Training Phase Performance and DIAS Scores*

The DIAS questionnaire resulted in a mean ± SD of 0.51 ± 0.10 (range: 0.31–0.77) for the OQS, 0.48 ± 0.14 (0.28–0.88) for Factor 1, 0.45 ± 0.09 (0.28–0.68) for Factor 2 and 0.57 ± 0.09 (0.36–0.80) for Factor 3. Results of the correlation analysis between choices of S+ in the Test phase and both the speed of learning of the Training phase and the DIAS scores are reported in Table 5. No correlation was found between any of these variables. However, the number of trials to reach the criterion in the Training phase correlated positively with the DIAS OQS (Pearson's correlation: 0.42, *p* < 0.01) and its score for Factor 1 (r = 0.44, *p* < 0.01) and Factor 2 (r = 0.40, *p* < 0.01), but not Factor 3 (r = 012, *p* = 0.44).

**Table 5.** Pearson's correlations coefficients between the percentage of choices of S+ in the test phase, both at different S− distances (P0 = 350 cm, P1 = 310 cm, P2 = 270 cm, P4 = 190 cm) and across the whole test, and the number of trials needed to reach the learning criterion in the Training phase (TTC), the DIAS overall score (OQS), the score of the DIAS' Factor 1, Factor 2 and Factor 3.


#### **4. Discussion**

In this study, we devised a behavioral test for the assessment of dogs' tendency to make impulsive choices, which was conceived as a spatial implementation of the conventional delay discounting paradigm. All dogs who participated in the study successfully achieved the initial training, which required them to consistently select the larger of two food quantities presented at the same distance. In the subsequent test phase, as expected, dogs expressed a higher probability to choose the smaller amount of food, as the latter was positioned increasingly closer to the dog. The entire assessment procedure was completed in less than approximately 1 h. Overall, the findings represent a good indication of the feasibility of the paradigm, and its better suitability for the assessment of impulsive choices in dogs, compared with lengthier and harder-to-complete delay discounting tasks. A spatial discounting test analogous to the one presented in this study was independently developed and recently presented by Brady and collaborators [32]. This study also reports a high success rate, and an outcome which conformed to expectations (i.e., choices of the larger food amount dependent on the relative distance). Thus, in agreement with this study, we converge on this paradigm's ease of application, which makes it a good candidate for the assessment of impulsivity in large dog samples.

Besides evaluating the feasibility of the procedure, we aimed at providing a first validation of the task as a measure of impulsive choices, by assessing its specificity and its external validity. To the first aim, we ascertained that dog's performance in the spatial discounting task could not be explained by factors different from impulsivity. As our task was based on the acquisition of food, one of our first concerns was to exclude that the dogs' performance did not reflect their motivation towards food rather than their impulsivity. The interplay between impulsive behaviour and motivation to obtain food is certainly a complex one [33]. In fact, while impulsivity and sensitivity towards food are independent traits, they interact to determine food-related behavioral outcomes in humans and rats [33,34]. To the best of our knowledge, the role of food motivation was seldom taken into account in dogs' impulsivity studies. Brucks and collaborators [30] report that varying the quantity and the quality of food-rewards affects dogs' ability to tolerate delays in a delay of gratification task. The same authors highlight the potential confounds represented by food motivational factors on impulsivity measures in dogs. Therefore, the finding that food motivation was not a significant predictor of the dogs' performance in our task provides a first indication in favour of the tasks' specificity. One caveat in the interpretation of these finding is that no dog was present with low levels of food motivation, restricting the validity of this claim to dogs with medium to high food motivation levels.

Past training was another factor that could potentially interfere with dogs' performance in our tasks; for instance, dogs with experience of prolonged training may be more accustomed to sustained work and be less susceptible to mental fatigue, thereby performing better than untrained dogs in the test phase of our procedure. The finding that training had no effect in explaining dogs' choices of the larger food amount was therefore another indication in favour of the tasks' specificity as a measure of impulsivity. Importantly, while this result indicates that our assessment is unaffected by differences in training history, it does not negate that some forms of training may improve dogs' ability to exert self-control. Recent findings suggest that specific forms of training can improve some measures of impulsive control, such as impulsive actions [26]. Moreover, deliberate training for self-control can lead to a generalized increased ability in different forms of impulsivity in humans, while extensive and specific training brings improvement in impulsive choice in animals [35]. However, data about the dogs' training history is seldom reported in previous studies on dog impulsive choices, thus it is difficult to make conclusions about the role of such experience on this facet of impulsivity. The availability of an easily applicable procedure for the assessment of impulsive choices, like the one presented in the present study, will allow to study the role of specific training history on canine impulsivity.

A related finding was that the number of errors (S− choices) made by dogs before reaching the learning criterion in the training phase did not explain their choices in the test. Previous research highlighted how the task's learning requirement may represent a confound in measures of dog impulsivity. In fact, the idea that alleged measures of impulsivity may actually reflect the dogs' learning ability was presented as a potential explanation for the lack of consistency across tasks [29]. In view of such concerns, the finding that dogs' performance in our assessment was not affected by the dogs' ability to learn the initial tasks represents an important indication of specificity. Another concern that relates to the learning requirements of delay discounting tasks, is that the necessary initial training is often achieved only by a fraction of dogs, producing an inherent bias in the selection of dogs who undergo the actual assessment. This does not seem to apply to spatial discounting tasks, as the training phase was acquired by all dogs who participated in our study, as well as by nearly all those who took part in the task developed by Brady and collaborators [30].

As an indication of external validity, we investigated how dogs' performance in our task was affected by age, sex, and reproductive status. Age had no effect on dogs' probability to make impulsive choices. Considering the majority of our dogs were adults, the result is in line with human studies, where evidence indicates a stabilization of impulsive choice behavior after adolescence/young adulthood [16]. The performance of the dogs in our task showed a clear dimorphic pattern: females discounted more steeply, as their probability to choose the larger food amount decreased significantly as soon as the bowl with the smaller amount was moved closer to the dog. Many sex related behaviors have been described in dogs [36]. In the present study, analysis of sex differences was undertaken to provide an indication of the tasks' goodness as a measure of impulsive behavior. In fact, our results conform to the what is reported in both humans and rodents, where steeper discount curves are generally found in females than in males [15]. No difference in performance was found between our intact and gonadectomized females. On the one hand this suggests that the main contribution to the observed sex difference is due to organizational effects of sex hormones, rather than by these hormone's circulating levels. On the other hand, as our intact female dogs were in the anœtrous phase (based on the report of the owners on the date of their last manifestations of œstrous) it cannot be excluded that the performance of intact female dogs may have been different, had females been tested in other phases of the oestrous cycle, as seen in other species [37,38]. Our current data cannot elucidate the mechanisms underlying the observed differences, Thus, we cannot tell whether dopamine transmission is involved in these differences, as suggested for other species. To the best of our knowledge there is no data about sex differences or the role of ovarian hormones in dopaminergic transmission in dogs. However, it is worth noting that sex differences are consistently found in dogs' spatial learning tasks [39–41], where dopamine plays a crucial role [42]. Regardless of the mechanism, our results indicate that the phenomenon our task is measuring is subject to the same biological influence seen in other species, providing an indication of the tasks' external validity.

Finally, no correlation was found between our dogs' performance in the test, and the score obtained by dogs in a putative assessment of impulsivity made through the DIAS questionnaire, either in terms of its overall score or the score of its subscales (calculated as described in the validation study by Wright and collaborators [18]). On the one hand, the finding clashes with the significant correlations between the DIAS score and the measures of impulsivity obtained in the spatial discounting task presented by Brady and collaborators [32], or in a delayed reward paradigm [19]. On the other hand, several other studies on dogs' impulsive choices report no association with the DIAS score [25,29], or correlations in

opposite directions than expected [30]. Although the reason of these discrepancies is not immediately clear, it must be considered that the DIAS was developed to assess impulsivity as a generic personality trait, rather than to pinpoint a specific facet of the phenomenon. As already highlighted by others [30], expressions of impulsivity are highly context specific and it is possible that the questionnaire and our task are assessing different facets of the same phenomenon. Alternatively, it is possible that they assess completely independent traits. In fact, our finding of a positive correlation between dogs' speed of learning of the initial training phase and the questionnaire scores suggests that the latter reflects the dogs' learning ability rather than their impulsivity. Moreover, questionnaires are based on indirect evaluations of the animals' behavior made by their owners, which incorporates a considerable degree of subjectivity in the assessment. Such individual variability could be further amplified by cultural differences, and translation-related nuances. In fact, while significant correlations between the DIAS and impulsivity measures were reported by studies conducted in the UK, the opposite was generally true for studies made in non-Anglo-Saxon countries, either using the original English version (e.g., [25]) or a translated version of the questionnaire [29,30], as in the current one.

#### **5. Conclusions**

In this study we presented a spatial discount task, aimed at assessing impulsive choices in dogs. A similar task was independently developed by Brady and collaborators [32] at around the same time. Both studies converge on the ease of application of the task, which advocates the procedure as a good candidate for larger-scale studies on impulsivity. We ascertained the lack of effect of several factors which may have interfered with the dogs' measure, thereby providing indications of the procedure's specificity. In addition, we provided indications about its external validity by showing a susceptibility of the assessment to sex differences, similar to those already observed in humans and rodents. Overall, the task seems to be promising as a valid, easily applicable procedure for the assessment of impulsive choices.

However, although these findings, together with those of Brady and collaborators [32], provide indications about the goodness of this assessment, other steps would be needed to provide conclusive evidence of its validity, as well as to fine-tune the procedure. For instance, it would be important to determine how the present assessment relates to the outcome of other procedures, that are assumed to measure other facets of impulsivity, such as tasks assessing dogs' tendency to express impulsive actions. Moreover, in view of a potential application in large-scale or cross-cultural studies, it would be important to extend the assessment to larger representation of size and age than those included in this study, as well as to ascertain the reproducibility of the assessment across different laboratories. Considering the ease of administration of the procedure, it is foreseeable that the same would be applied as a screening/selection tool in clinical (e.g., for the identification of pathological impulsivity) or other professional contexts (e.g., for the selection of dogs to be trained for specific activities); to this aim, evaluation of the applicability of the procedure in non-experimental settings and of its predictive validity for expected outcomes, would be required. Finally, considering the known interplay between training and impulsivity, the procedure could be used to assess the efficacy of specific forms of training, including its applications as a therapeutic intervention, in reducing impulsive behavior.

**Author Contributions:** The study was conceived by L.M. and P.M.; the experiments were conducted by A.S.; statistical analysis was performed by P.M.; drafting was done by P.M., A.S., and C.J.E.; interpretation of results was done by P.M., A.S., B.D., and L.M.; revision and editing were done by B.D. and L.M. All authors have approved the submitted version and agree to be personally accountable for the author's own contributions and for ensuring that questions related to the accuracy or integrity of any part of the work, even ones in which the author was not personally involved, are appropriately investigated, resolved, and documented in the literature.

**Funding:** This research was supported by a research grant from Università degli Studi di Padova (to P.M., Grant Nr. CPDA144871/14), two postdoc grants awarded by Università degli Studi di Padova (to A.S., Grant Nr. CPDR148844/14; to C.E., Grant Nr. BIRD178748/17). The APC was funded by Università degli Studi di Padova, Dipartimento di Biomedicina Comparata e Alimentazione (Grant Nr. DOR1809748/18) and Università degli Studi di Napoli Federico II, Dipartimento di Zoologia (Grant Nr. PRD2019).

**Acknowledgments:** We are very grateful to Carlo Poltronieri for his technical assistance and to all the dogs' owners for volunteering to take part.

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

**Ethical Statement:** The study was conducted in accordance with relevant legislation about research involving animals, and, for the type of procedures involved, no formal ethical approval was required.

#### **References**


© 2019 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*
