Could Greater Time Spent Displaying Waking Inactivity in the Home Environment Be a Marker for a Depression-Like State in the Domestic Dog?

Simple Summary Stressed pet dogs, such as when deprived of their owners or after the loss of a social companion, can become inactive and unresponsive. Dogs in this condition are commonly referred to as being “depressed”, but this remains an untested hypothesis. One hallmark of human clinical depression is anhedonia—a reduction in the experience of pleasure. Here we tested the hypothesis that shelter dogs that spend greater time inactive “awake but motionless” (ABM) in their home-pen would also show signs of anhedonia, as tested by reduced responses to a treat filled KongTM. We also explored whether dogs being rated by experts as disinterested in the KongTM would spend greater time ABM (experts did not know the dogs’ actual inactivity levels). Fifty-seven dogs from 7 shelters were tested in total. Dogs relinquished by their owners spent more time ABM than strays or legal cases, and one association was found between the ABM and the dogs’ response to the filled KongTM, which was in the opposite direction that expected, so does not support the hypothesis that waking inactivity indicates a depression-like state in dogs. Dogs rated by experts as “depressed” and “bored” when exposed to the KongTM, however, spent greater time ABM; we discuss whether ABM could tentatively indicate “boredom” in dogs. Abstract Dogs exposed to aversive events can become inactive and unresponsive and are commonly referred to as being “depressed”, but this association remains to be tested. We investigated whether shelter dogs spending greater time inactive “awake but motionless” (ABM) in their home-pen show anhedonia (the core reduction of pleasure reported in depression), as tested by reduced interest in, and consumption of, palatable food (KongTM test). We also explored whether dogs being qualitatively perceived by experts as disinterested in the food would spend greater time ABM (experts blind to actual inactivity levels). Following sample size estimations and qualitative behaviour analysis (n = 14 pilot dogs), forty-three dogs (6 shelters, 22F:21M) were included in the main study. Dogs relinquished by their owners spent more time ABM than strays or legal cases (F = 8.09, p = 0.032). One significant positive association was found between the KongTM measure for average length of KongTM bout and ABM, when length of stay in the shelter was accounted for as a confounder (F = 3.66, p = 0.035). Time spent ABM also correlated with scores for “depressed” and “bored” in the qualitative results, indirectly suggesting that experts associate greater waking inactivity with negative emotional states. The hypothesis that ABM reflects a depression-like syndrome is not supported; we discuss how results might tentatively support a “boredom-like” state and further research directions.

Dogs housing conditions, home-pen observations (quantifying time spent being awake but motionless) and Kong TM anhedonia test are similar to those described in the Methods section of the main paper, with the exception of the Kong TM variables (that we refined following the pilot study). Kong TM measures for the pilot study were: the total time the dog spent interacting with the KongTM (defined as: paw or muzzle in contact with, or sniffing the KongTM, excluding time stood chewing not in contact) (Kong TM Time); the time to interact with KongTM from first placement on floor (Kong TM Latency); the number of times dog returned to contact Kong after moving >2 paces away (Kong TM Returns). Data regarding the percentage of the food mix eaten by the end of the 30 minutes test was available for the main study dogs only.
Although ABM was not normally distributed, both univariate and multivariate model residuals were visually examined and considered to be acceptable, so no transformations were applied.

Interval sample selection
In order to be time efficient, but not loose accuracy of the data, it was necessary to identify the maximum sample interval we could use that would produce a representative activity budget. Shorter sample intervals form more accurate representations of behaviour but are less time and cost effective than longer intervals (Hämäläinen, Ruuska, Kokkonen, Orkola, & Mononen, 2016).
For the first seven dogs from Shelter A scans were taken every 30 seconds, over the entire 6hr recording period. Inevitably, dogs were not always visible for the entire 6hrs of footage, with some dogs being taken out for walks during a recording period or kennels being cleaned. For this reason, the number of visible scans was summed for each period per dog, and the minimum number of visible scans for each dog identified. In this way it was confirmed that each of the initial 7 dogs had at least 160 visible scans (80 minutes) per 2hr period and 4 hours (480 scans) in total; this was the maximum number of visible scans that could be met for every dog. Therefore, while some dogs had more than 160 scans visible, we limited all dogs to 160 visible scans to ensure all dogs had the same number of visible scans and that all dogs could be compared in a comparative manner. For a dog with more than 160 visible scans in any period, 160 visible scans were identified, iteratively, from the centre of the largest consecutive block of visible scans.
An activity budget was calculated for each dog comprising of the proportion of scans seen exhibiting each point event from the ethogram. This was done iteratively for all 7 dogs for the original 30-second interval scans, then utilising every 2 nd scan (representing 1-minute intervals), every 3 rd scan (representing 1.5-minute intervals) and every 4 th scan (representing 2-minute intervals), creating 4 different datasets.
The mean proportion of scans spent in each behavioural state was calculated for each dataset, followed by the difference in the mean between the 30-second reference data and each of the 2 nd , 3 rd and 4 th scan data (the Error Proportion; EP). If the EP for the larger interval datasets for any behavioural state was less than 10% different from of the 30-second mean estimate then it was considered to have retained accuracy (Hämäläinen et al., 2016). In this way, the longest interval that produced mean behavioural estimates most similar to the 30-second reference sample, for the greatest number of behavioural states, was selected for subsequent video analysis.

Results
Time budget estimates remained very similar for all scan intervals between 30-seconds and 2 minutes. However, accuracy at longer intervals decreased for behaviour with the shortest durations. For 1-minute intervals, the only behavioural estimates that changed by more than 10% were those that were exhibited for less than 1% of the time, and at 1.5-minute intervals for those that were exhibited for less than 2% of the time (Table 1). At 1 and 1.5-minute intervals, all estimates for the behaviour essential to this study (doing nothing) were all within a 10% margin of the 30-second interval estimates. Two-minute intervals produced a greater difference in estimates and a marked loss of accuracy for behaviour including lying down head or ears mobile, sleeping and walking. For these reasons, the 1.5-minute interval was deemed to produce acceptably accurate time budget estimates and was utilised for all subsequent home-pen video analysis.
Latency to interact with the Kong TM showed no variability, with all dogs except one interacting (approaching and sniffing/licking) with it as soon as it touched the floor (less than 1 second, and just 2.5 seconds for 1 dog). For this reason, only two measures from the Kong TM Test were considered in analysis: Kong TM Time (median time 10.7sec, Q1 5.2, Q3 12.5, from 0.4 to 15.3) and Kong TM Returns (median number 3.0, Q1 2.8, Q3 6, from 1 to 12).
Time spent interacting with the Kong TM (Kong TM Time) was not associated with ABM (B=-0.23, t=-0.82, p=0.423), whilst the number of times the dog returned to interact with the Kong TM was trending towards significance (Kong TM Returns: B=-0.71, t=-1.73, p=0.103), with dogs spending greater time ABM tending to return less to the Kong TM . Due to the trend towards sex being associated with ABM (p<0.1), a final multivariate model was tested, using a backwards elimination procedure, including sex and KongTM Returns. One multivariate model could be formed containing sex and KongTM Returns, both of which were significant to p<0.05 and explained a combined 37% of ABM variance (model R2 = 0.37, F = 4.47, p = 0.030).
Note on refining Kong TM measures for the main study: we excluded Kong TM Latency as this measure did not show any variability and would have, at least in theory, be influence by the original location of the dog in the kennel at the time the Kong TM was placed on the floor (an aspect we did not think of originally). We also replaced Kong TM Returns by the number of bouts X X and the duration of each bout based on ceasing physical contact with the Kong and ceased to chew food retrieved from the Kong TM (see main paper), as these aspects have proved to be easier to extract from the footage (i.e. less ambiguous) than the moving >2paces away criterion.

Sample size calculation for the main study
Sample size calculations were conducted using an online tool (www2.ccrb.cuhk.edu.hk/stat/epistudies/reg1.htm) according to methods outlined for linear regression power and sample size calculations (Dupont & Plummer, 1998). Calculations were based upon α=0.05 and β=0.2 and utilised the results of univariate linear regressions on the pilot data. Required sample sizes for the associations between ABM and the Kong TM measures and sex were: 42 for interaction bouts; 117 for total interaction time and 38 for sex differences. Whilst a sample size of 117 would be unfeasible, it was considered that a sample size of at least 42 dogs would be adequate to fully evaluate potential associations between the majority of measures for anhedonia and time spent ABM.  Table S1. Median percentage of scans, first (Q1) and third (Q3) quartiles and minimum and maximum values for behaviour based upon scans using 1.5 minute (n=18). The behaviour are ordered from the longest to the shortest median times spent displaying them, and the behaviour we hypothesise to specifically reflect depression is highlighted in bold.