Ethical approval
Ethical approval for human participants was granted by the NHS Health Research Authority (HRA) South-West Central Bristol Research Ethics Committee (REC) (reference: 22/SW/0134). All experiments were conducted in accordance with NHS HRA guidelines and regulations and followed the study protocol approved by the REC. Biological samples were collected, stored, and disposed in accordance with the Human Tissue Act. Animal protocols were approved by the University of Bristol Animal Welfare and Ethical Review Body (AWERB) (reference: UIN-22-0102) and conducted in accordance with the UK Animal (Scientific Procedures) Act 1986. This study is reported in accordance with ARRIVE guidelines. All human volunteers and dog owners were informed about their right to withdraw, and informed consent was obtained prior to their participation.
Experimental design
This study was split into two phases (Fig 1):
1) Collecting human odour samples
2) Cognitive bias testing in dogs
Phase 1 - Collecting human odour samples.
Sample Donors
A total of 11 volunteers (10 female and 1 male) aged 18-26 were recruited through advertisements in bulletins, notice boards and group meeting announcements at University of Bristol Vet School. Participants were in good health, were not on any medication, had no current or prior psychiatric diagnoses, reported no recent difficulty with sleep or significant stressful events and tested negative for SARS‑CoV‑2 on lateral flow on each day of testing. Participants were instructed to abstain from alcohol and smoking for at least 12hrs and from eating for 1hr before testing and to avoid spicy foods, intense exercise, using body wash, perfume, deodorants, or antiperspirants on each day of testing. Participants were informed the study involved a psychological stress test designed to induce feelings of anxiety and were advised they could stop participating at any time. Participants were compensated financially for their time and provided with snacks and refreshments at the end of each session.
Protocol
Participants attended two one-hour sessions held 1-7 days apart. The protocol was the same on both days except for the 20 min test period, during which participants were given either the Trier Social Stress Test (TSST) 51, or watched a compilation of forest and seaside videos shown to induce relaxation 55. Each session had 6 stages (Fig 2):
Stage 1: A heart rate monitor was attached. The participant rinsed their mouth and armpits with water prior to sample collection.
Stage 2: The participant sat quietly in the waiting room for 10min.
Stage 3: A pre-test saliva sample and anxiety questionnaire were collected, and odour collection cloths were attached to each armpit.
Stage 4: The participant was called into the testing room for the 20min stress test (session 1) or relaxing activity (session 2).
Stage 5: A post-test saliva sample and anxiety questionnaire were collected before the participant returned to the waiting room. Odour collection cloths were removed from the armpits, and the participant collected breath onto the same cloths.
Stage 6: The participant sat quietly in the waiting room for 10min before collecting a post-recovery saliva sample and removing the heart rate monitor before leaving.
All participants received the stress condition in session 1, but they were not informed of this until the test began. Once the stress test was complete, participants were informed session 2 would be the relaxed condition. This order allowed us to control any influence of uncertainty and anticipatory stress across the cohort, thereby maximising stress in session 1 and minimising stress in session 2.
Stress test
During session 1, the 20min TSST (stage 4) was based on the protocol by Birkett et al. (2011) 51. In brief, the primary researcher wearing a white lab coat called the participant into the testing room, where a second unexpected researcher was introduced as an expert in public speaking, there to assess the participant’s performance. The participant was told they had 10min to mentally prepare a 5-minute speech describing why they would be a good candidate for their ideal job and that a video of their speech would be recorded for assessment. A timer visible to the participant was set and both researchers left the room. After 10min, the researchers returned, and the participant was instructed to deliver their speech and told they should speak for the entire 5min. The researcher pretended to turn on the video camera and the timer was reset. If the participant stopped talking, they were prompted with the following: "You still have time remaining”. Both researchers maintained serious expressions throughout to increase social anxiety. After the 5min speech task, the participant was instructed to subtract the number 13 from 1,022 and continue subtracting 13 from the remainder until the 5min was up, verbally reporting their answers aloud. If the participant made a mistake, they were prompted with the following: "That is incorrect, please start over from 1,022." After the 5min maths task the participant was informed that the test had ended, and that the camera had not been recording.
Relaxing activity
For the relaxed condition in session 2, the primary researcher (in plain clothes) called the participant into the testing room where a bean bag and cushions were provided for their comfort. To minimise any effect of the room triggering stress from the session before, the room layout was changed and the side where the stress test occurred was obscured with a screen. Blankets and throws were used to cover the existing furniture and the room lighting was kept dim to create a more relaxing atmosphere. The participant was given a laptop and headphones and instructed to watch a 20min video compilation of forest and seaside scenes with natural soundscapes. The researcher left the room and returned once the video ended.
Following each session, participants were asked not to disclose their experience with anyone outside of the research team, as this could exclude others from participating. The full stress and relax protocols can be found in the supplementary data (S1 File).
Measuring stress response
To control for diurnal variations in cortisol, sessions 1 and 2 were held at the same time of day (±2 hrs) for each participant. To minimise changes in saliva pH that could interfere with cortisol measurement, participants were instructed not to eat or drink (other than water) for an hour prior to each session. Each participant collected a pre-test (used as baseline), a post-test and a post-recovery saliva sample (to measure delayed cortisol responses). Saliva samples were collected in 2mL cryovials via passive drool and stored at -20°C within 30min of collection for up to 1 month. Each saliva sample was analysed in duplicate using Salimetrics® Salivary Cortisol ELISA Kit according to manufacturer’s instructions. The assay was read using a Thermo Scientific Multiskan FC plate reader and cortisol concentrations were determined using a dose-response curve fitted in GraphPad Prism 9. Assay controls were used to confirm the ELISA had run correctly and sample concentrations were checked against the expected range provided by Salimetrics®. Heart rate and HRV were recorded using a Polar® H10 chest strap with data recorded in real-time via Bluetooth onto a tablet using the Polar Sensor logger app 56. Start and end times for each stage of the session were logged in real-time by the primary researcher. Data were imported into Kubios HRV software 57 where the mean HR, root mean square of the successive differences (RMSSD), and Stress Index (geometric measure of sympathetic versus parasympathetic inputs 58) were calculated for stages 2, 4 and 6 of each session (Fig 2). To assess perceived stress levels, participants completed a pre-test and post-test State-Trait Anxiety Inventory (STAI)© 59 for each session (Fig 2). State anxiety scores were used to assess within-subject responses to each test condition.
Odour Samples
Prior to entering the testing room, participants attached two 20x20cm 100% cotton cloths to each armpit using micropore tape so that they were in direct contact with the skin. The cloths remained in place for the duration of stages 4 and 5 (Fig 2), after which they were removed. Participants cut each 20x20cm cloth into four 10x10cm pieces and exhaled a full breath onto each piece before sealing each in a separate specimen bag. This gave a total of eight pieces of cloths (four per armpit) for each participant for each session: six for cognitive bias testing of dogs and two for odour analysis (not described here). All samples were double bagged and placed on ice before being stored at -20°C within 30min of collection. Samples were stored for 3-6 months until use in the cognitive bias testing. Blank cloths used in baseline testing of dogs were stored in the same conditions and for the same time as odour samples to control for “background” odour from the freezing and storage process. To maintain a within-subject design in phase 2, each dog was presented with stress and relax odours from the same donor. Since each donor collected six samples for cognitive bias testing, a maximum of six dogs could be presented with odours from the same person (without reuse of samples). To provide samples for all 18 dogs in the study, we selected three donors with the greatest stress responses across all measures during the stress test and no stress response during the relax condition (see results).
Phase 2 – Cognitive bias testing in dogs
Recruitment and protocol
A total of 26 dog-owner partnerships were recruited via advertisements in bulletins, posters and social media shared with staff, students, and members of the public at the University of Bristol Vet School and Langford Vets Small Animal Practice. All dogs were over 6 months of age, in good health, comfortable learning new tasks and had no history of fear or anxiety of novel environments or people. After expressing an interest, owners were invited to attend the first of three 40-60min sessions (Fig 1). Before starting, owners were given the opportunity to ask any questions and signed a consent form. Dogs were allowed 15 minutes to habituate to the room and equipment, during which the following information about the dog was collected: age, sex, neuter status, breed, how long they had lived with their owner, who they were most attached to (person present, another person, multiple people or no attachment) and how sensitive they were thought to be to their owner’s emotions (scale 1-5: 1 = not at all, 5 = very in tune). Following collection of this information, the first cognitive bias test began in the presence of a blank cloth (no odour). Dogs then participated in two subsequent sessions during which the stress and relax odours were introduced (Fig 1). At the start of each session, owners confirmed their dog had been well in the preceding seven days, were asked to report any recent events that may have impacted their dog positively or negatively or any recent changes to routine. Because of the potential influence of the owner’s stress on their dog, owners were also asked how relaxed they felt at the start of each session (scale 1-10: 1 = not at all relaxed, 10 = very relaxed). Sessions were held at the same time of day for each dog approximately one week (6-9 days) apart, except for on two occasions: one dog’s second session was delayed by 8 days due to owner illness and another dog’s third session was delayed by 24 days due to dog illness. Access to a water bowl was provided throughout and dogs were never separated from their owner. All dogs that completed the study received a toy reward at the end of their third session.
Dogs
Of the 26 dogs recruited, 8 were excluded during session 1: one dog was excluded due to mobility issues and signs of discomfort during training, two were excluded for behavioural reasons (one for showing signs of distress and the other showing destructive behaviour towards the equipment), four dogs were excluded for failing to engage with the task for five consecutive trials and one dog due to distractibility. All 18 dogs that completed session 1 completed the study. These were 11 male (10 neutered and 1 entire) and 7 female (4 neutered and 3 entire) dogs ranging from 8 months to 10 years old (mean = 4.73 ± 2.67 years). Breeds consisted of 2 Springer spaniels, 2 Cocker spaniels, 2 Labrador Retrievers, 2 Braque d’Auvergne, 1 Whippet, 1 Golden retriever, 1 Miniature poodle and 7 mixed breed dogs. Eight dogs were registered as teaching dogs at the University of Bristol. To control for any influence of familiarity with the study site, these eight dogs were split equally between groups and differences between teaching and non-teaching dogs explored during data analysis.
All dogs were presented with a blank cloth (no odour) during session 1, after which they were divided pseudo-randomly into three groups of six, balancing for teaching dog status. Each group was subsequently presented with odour samples from one donor (Table S1). Within each group, the order of stress and relax odour presentation was counterbalanced across dogs, so three dogs were exposed to stress odour before relaxed odour (order 1) and the other three exposed to relaxed odour before stress odour (order 2) (Fig 1) (Supplementary Table S1). We estimated a minimum of 18 dogs (3 per group) as a realistic number of recruits in the time provided for this pilot study. We used a within-subject design to increase the statistical power, however because the effect size and variability were unknown, a power calculation could not be performed. To maintain blinding, specimen bags containing odour samples were labelled with a code known only to a second researcher, not present during dog testing. A list of sample codes for each dog was provided to the primary researcher so that the odour type during testing was unknown to anyone present in the room.
Cognitive bias testing
The cognitive bias protocol use was based on Mendl et al., (2010) and Hale (2021) 42,60. During the test, owners (acting as handlers) sat at a start location with their dog and faced a screen on the opposite side of the room. The screen developed and provided by Hale (2021) 60 had five doors through which the primary researcher (ZPC) pushed a food bowl whilst remaining behind the screen (Fig 3). Each bowl position was 3.5m from the start location (Fig 3b). The screen provided a visual barrier so that the dog was unable to see the researcher preparing the bowl. When preparing an empty bowl, the researcher “mock” baited the bowl so that any auditory cues of baiting were the same whether the bowl was baited or empty. After placing the bowl through the door, the researcher gave a “release” command, and the handler released their dog from the start location, giving a verbal “go get it” or “ok” prompt. The latency to reach the bowl was measured using a manual stopwatch, which was started the moment the handler released their dog, and stopped the moment the dog’s head reached the bowl, as observed through a live overhead camera (GoPro Hero 4, GoPro, Inc., San Mateo, CA, USA) connected to a monitor. After the dog ate the reward (if present), a “recall” command was given, and the dog was called back to the start location. If the dog did not approach the bowl within the 30s, a maximum time of 30s was recorded, and the “recall” command given (even if the dog remained at the start location). The handler reset their dog’s position in preparation for the next trial. Handlers were able to choose their verbal release prompt but were instructed to keep any verbal prompt and gestures the same for all trials and all sessions. If the dog did not approach the bowl following the prompt, handlers were instructed not to repeat their prompt and to avoid encouraging/insisting that their dog approach, as this could introduce bias and alter the dog’s motivation to approach the bowl. The food rewards were Pedigree® Schmackos (turkey flavour) cut into 2cm pieces. Due to dietary restrictions one dog received 1cm cubes of cheese and another received Pedigree® Tasty minis for puppies instead. All dog received the same type and volume of food reward in all three sessions. The same bowl was used for all locations, so that no distinguishing features of the bowl itself could be used to predict the presence of a reward.
Training phase
Each session began with a training phase during which, on each trial, the food bowl was either presented at the positive training location (containing a food reward) or negative training location (empty bowl). The training phase began with two trials at the positive (P) location followed by two at the negative (N) location, after which the order of P and N trials was pseudorandomised, with a maximum of two consecutive trials at the same location. Training was repeated for a minimum of 15 trials and continued until the dog was consistently slower to approach the N location than the P location. This was measured using the following criteria based on the latencies in the preceding six trials (3x N location and 3x P location):
1) The fastest of three approaches to the N location was ≥3sec slower than the slowest of three approaches to the P location 42
2) The mean latency of the three approaches to N was >1sec longer than the mean latency of the 3 approaches to P.
A maximum of 50 trials were permitted to meet both these criteria. After passing both criteria, the dog moved on to the testing phase. To shorten the length of subsequent sessions and minimise fatigue, during session 2 and 3 the minimum number of training trials was reduced to six, after which the same criteria were used. The location of P and N bowls were counterbalanced across dogs so that half were presented with the P location on the right and the other half presented with the P location on the left. The location of P and N remained the same for all three sessions for each dog since switching sides in subsequent sessions could introduce ambiguity at these locations and impact the dog’s perception of N and P locations that are considered “unambiguous”.
Testing phase
After successfully passing the training phase, the testing phase began. During the testing phase, dogs were presented with an empty food bowl at the novel “ambiguous locations” between the P and N positions: near-positive (NP), middle (M), or near-negative (NN) (Fig 3). Each ambiguous location was presented twice per session. The order of presentation was pseudo-randomised, but remained the same for all dogs across all sessions to minimise any effect of order when comparing between sessions. Between each ambiguous bowl presentation, dogs were presented with a full bowl at P twice and an empty bowl at N twice to reinforce previous associations. Therefore, for each session there were a total of 26 trials in the testing phase (6 ambiguous and 20 re-enforcement trials).
Reward Control
To test whether the dog was using olfactory or other cues to determine if a reward was present, at the end of the testing phase, in session 3, the dog was presented with a baited bowl at the negative location (N+) and an empty bowl at the positive location (P-). This was only conducted in the final session so that the associations of P and N were consistent until the end of the study.
Introducing odour
Before each session, the assigned sample (or blank frozen cloth for session 1) was taken out of the freezer by the primary researcher wearing gloves. The cloth was removed from its bag using forceps and placed it in a glass jar and sealed immediately with an aluminium screw top. The closed jar was attached to the chair at the start location (Fig 3), the position of which was adjusted to each dogs’ nose height. The sample was allowed 30min to reach room temperature before the cognitive bias test began. The jar remained closed for the training phase of each session, so that there was no influence of odour on learning of positive and negative cues. Before being given the “release” command, the handler was given the command “jar”, which prompted them to either tap the lid of the jar (training phase) or unscrew and lift the lid off the jar (testing phase). During testing, the lid remained open for 20s to allow the odour to be released and detected by the dog. During this time the handler was instructed to keep their dog close to the jar and gently present the jar lid to the dog to sniff, allowing them to inspect and sniff the sample. Handlers were told not to insist the dog sniff the jar or forcefully push the lid towards the dog’s nose, which could be aversive. They were also instructed to remain neutral and not make any gestures or verbal comments towards the jar, so as not to create any positive or negative associations with the odour or jar itself. After 20s the experimenter repeated the “jar” command, and the handler replaced and closed the lid of the jar (or tapped the lid during the training phase). The owner was then prompted to release their dog to start the trial. After recalling their dog from their approach, the opening and closing of the jar (or tapping of the lid) was repeated before releasing them for the next trial.
Each sample or blank cloth was used only once. After each use, glass jars were put through an enzymatic wash cycle, wiped down with methanol to remove any residual organic odour, and then sterilised in the autoclave before reuse. Before each session, the windows and doors in the testing room was ventilated for 1hr. After each session, the floors were thoroughly cleaned with warm water and Odourless Disinfectant Cleaner (Safe4) at 1:100 dilution. An odourless odour eliminator (Dew Products) was then sprayed to remove any residual odours from the samples or dogs. The room was ventilated again for 1hr after cleaning. A minimum of 3hrs was allowed between sessions, with a maximum of two sessions per day to minimise odour build up in the room. There was a minimum of 24hrs “wash out” period between sessions with different sample types. If two sessions took place on the same day, the sample donor and odour type was kept the same for both sessions, or consisted of a session with a blank cloth followed by a treatment odour.
Statistical analysis
Human stress response
To calculate the change in saliva cortisol for each participant, the pre-test cortisol concentration (stage 3, Fig 1) was subtracted from the post-test concentration (stage 5) for each condition, giving a post-stress and post-relax cortisol response. To calculate the delayed change in saliva cortisol, the pre-test concentration was subtracted from the post-recovery concentration (stage 6) for each condition, giving a delayed-stress and delayed-relax cortisol response. To calculate the change in mean heart rate, Stress Index and RMSSD for each participant, the baseline value (stage 2) was subtracted from the test period (stage 4) for each condition. To calculate the change in STAI state scores in response to each condition, pre-test scores were subtracted from post-test scores for each participant. Since data for mean heart rate, STAI and Stress Index were normally distributed, paired t tests were used to compare the change from baseline to the test period/post-test between stress and relax conditions for these three measures. Since data for saliva cortisol and RMSSD were not normally distributed, paired Wilcoxon signed rank tests were used to compare changes from baseline to the test period/post-test between stress and relax conditions for these two measures. T-tests and Wilcoxon tests were run using the stats R core package.
Cognitive bias testing
Reward control
To test whether the dogs were using olfactory or other cues to determine if a reward was present, Wilcoxon tests were used to test if there was a difference in the latency to reach the N location when the bowl was baited (N+) versus empty (N), and if there was a difference in the latency to reach the P location when the bowl was empty (P-) versus baited (P).
Likelihood of approaching the bowl
To assess how fixed effects affected the likelihood of the dog approaching the bowl within the time allowed (30s)), Cox proportional hazards models were run using the coxme and survival R packages 61,62. Typically used in survival analysis, a Cox proportional hazards model incorporates both the time to an event (latency to reach bowl) and whether the event occurred or not (approached bowl within 30s or not) to calculate the likelihood of an event occurring. Combining these, it examines how fixed effects influence the rate at which an event occurs by estimating the hazard ratio (HR) compared to a reference level. In this case the HR indicates the relative likelihood of the dog approaching the bowl at any given time within the 30s. A HR <1 indicates a decreased likelihood of approaching the bowl, a HR >1 indicates an increased likelihood of approaching the bowl, and a HR = 1 indicates no effect compared to the reference level. To assess whether dogs successfully learned to discriminate between the five bowl locations (P, NP, M, NN, N) a Cox proportional hazards model was run to determine the effect of bowl location on the likelihood of the dog approaching, with the P location used as a reference level. To assess the effect of other variables within each location, the dataset was filtered by location and Cox proportional hazards models run at each location for the following fixed effects: treatment (baseline, stress, relax), session (session 1, session 2, session 3), teaching dog (Yes/No), dog attachment (person present, other person, multiple, none), dog emotional sensitivity (1-5) and handler relaxation rating (1-10). Dog ID was included as a random effect to control for individual differences in running speed. To reduce variables within the model and select the model with the best fit, stepwise deletion of terms based on Akaike’s Information Criterion (AIC) was used using the drop1 function in R. Tukey’s post-hoc pairwise comparisons based on estimated marginal means (least-squares means) were run on significant terms using the emmeans R package 63. To assess the effect of odour within each order of presentation, a stratification term of treatment order (order 1: stress – relax and order 2: relax – stress) was included. Plots were constructed using the ggplot2, ggstatsplot and ggpubr R packages 64-66 and all analyses and plots were produced using R Studio Version 2023.06.2+561.