Can dogs sense vulnerable people? Surprising Answer

Signs a Dog Can Feel Vulnerable

The shelters are littered with vulnerable pooches and honors need to be given to the many people that set up rescue centers for animals in need. Their call to action has saved the lives of countless dogs giving them hope and a new address to call home. They are the unsung heroes who pick up the pieces of human cruelty with a fostering heart.

Dogs that are abandoned, neglected, or abused find their way to the shelter on a daily basis with a battered soul and vulnerability, that sees their tails tucked between their legs, ears dropped, and no spark of life in their eyes. The day begins with volunteers trying to work their magic on these helpless mutts that whine and howl in distress. Each dog is as precious as the next and volunteers like nothing better than to see a cowering pup feel confident.

Shelter dogs often have a rags-to-riches story as they arrive undernourished, beaten down, and petrified. Some are so depressed, they avert the volunteer’s eyes and shake continuously. Once picked up by rescuers, they have a chance at a new life and for some, the journey to happiness proves impossible, while for others, like Benny, a shy reserved pup waiting at a care center – a brand new start takes place.

The story of Benny went viral on YouTube as he sat quietly in his pen while all the dogs were barking in unison. His name was suddenly called and he began wagging his tail with excitement. A lead was put on and Benny, realizing he had been adopted, started jumping around with his new owners in tow. The sheer exhilaration this pup felt was a joy to see.

Signs a dog can feel vulnerabile are:

Things that cause dogs to feel vulnerable include:

  • Finding Themeselves In A Shelter
  • Being Abandoned
  • Being Mistreated
  • Can dogs sense vulnerable people?

    The first animal shelter in the U.S was thanks to an enlightened and well-educated animal lover named Caroline Earl White. Having learned of a prominent New York businessman who set up The American Society for the Prevention of Cruelty to Animals in 1866, this animal welfare advocate was inspired to start a similar organization in Pennsylvania.

    At the time, horses were used to pull carts, and seeing these animals beaten drove this angel of mercy to enforce changes in the way animals were treated. Today, vulnerable dogs are taken to the Women’s Humane Society, started in 1869 by a woman who cared enough to make a change.

    As shelters all around the world do their utmost to help and find dogs a forever home, it is ironic that some breeds could soon become extinct. Looking back at the early days of dog domestication, its evident there were dogs breeds made redundant in mans bid to create perfection. Once upon a time, there were small Beagles that stood around 9” in height, nicknamed “Pocket Beagles” as they literally squeezed into a hunters jacket pocket. These cute pups were replaced by a larger version of the breed and were defunct by 1901.

    Now, we see a current list of “vulnerable dogs” presented by The Telegraph, UK that may shock dog lovers. Who would think the Bloodhound, famous for tracking Jack the Ripper, could be on the shelf very soon. Others include the Smooth Collie, Irish Red and White Setter, Field Spaniel and Sussex Spaniel. All these breeds have low registrations.

    Mankind is fickle and like the latest mobile phone is replaced by a newer model, our dogs are caught in the crossfire. Their usefulness is creating vulnerability in various breeds and like the dinosaur, we could see a time that many of man’s best friends are extinct.

    The blatant breeding in puppy mills of designer dogs is an exercise in greed vented by vanity. In an era of social media, the family dog could be lost in the translation. Our reliance on technology could lead to a dependence that denies the beauty of a loyal companion who offers unconditional love – a rare commodity in this day and age.

    6 Testing: Within person baseline and stress sample discrimination

    During testing, dogs were presented with each participant’s samples using the two-phase procedure outlined previously. At this stage, the target sample was the combined breath and sweat taken immediately after a participant completed the MAT and the distractor sample was the combined breath and sweat taken from the same participant at baseline, prior to taking part in the MAT (Fig 5). As before, each session consisted of 10 training trials in Phase One (target, blank, blank) and 20 discrimination trials in Phase Two (target, distractor, blank). For all training and testing, samples were taken within three hours of being shown to the dog and were stored at room temperature. Each participant’s samples were used in a single session (shown to one dog only) and then disposed of. During this stage, if a dog made an incorrect indication, they were recorded as giving a false alert on that trial, called away from the apparatus, and the next trial was prepared. As such, dogs were never assisted into making a correct alert and the handler was blind to the location of the target. Download:

    During sample collection, and for all handling of the ports, nitrile powder free gloves (Unicare non-sterile, single use examination gloves) were worn. During discrimination training and testing, the dogs were presented with fresh samples at the start of the Phase Two (discrimination trials). To do this, multiple samples were taken at the same time point. During participant testing, three blanks were made (B1, B2, B3), one distractor at baseline was taken (D1), and two targets at the end of the MAT were taken (T1, T2). The order that these were presented to the dog were: Phase One: T1, B1, B2, Phase Two: T2, D1, B3 (Fig 5). In doing so, when the discrimination trials started, all three samples were fresh, which eliminated the possibility that a dog could follow their own scent from Phase One or rely on odour differences associated with when the sample had been opened. A dog following their own odour can be described as them ‘tagging’ the port, meaning that they are aiding their discrimination by adding additional cues outside of the experimental design (e.g., licking the target port). To further reduce the likelihood of tagging, ports were wiped down with isopropanol (diluted with distilled water to 70% concentration) between each trial. In both training and testing, no samples were re-used or shown to more than one dog. This safeguards from potential issues of odour degradation or dogs incorporating the scent profile of the other study dogs into their decision making. These controls relate to potential extraneous cues associated with the samples themselves. A further area of concern is the possibility of extraneous cues coming from the apparatus itself, for example, the port that is designated to hold the target accumulates a distinctive odour over time, or there is a visual difference specific to that port that the researchers are unaware of. To control for this, all arms and ports of the apparatus were replaced with a new set once during the study. Because the fourth dog reached testing after the apparatus was replaced, one control session was conducted in which target and distractor arms were switched: the arm/port that previously held the target sample (the dog would get rewarded for indicating on this) was switched to holding the distractor sample (the dog would not get rewarded for indicating on this and it would be considered a false alert) and vice versa. If dogs were relying on extraneous cues provided by the apparatus, we would expect to see a decrease in performance after implementing the new apparatus (for three dogs) or switching which arm held the target sample (for one dog). Possible impacts on the dogs’ performance due to these controls were analysed.

    During within person training, and throughout testing, a double-blind procedure was used. Here, the first researcher (hereby referred to as the handler) was visible to the dog but the handler did not know where the target odour was located. The dog indicated their choice and the handler relayed this information to the second researcher who was out-of-sight (behind a 260x180cm three-panel room divider). The out-of-sight researcher confirmed this alert as correct or incorrect and the handler activated the clicker and rewarded the dog, or ended the trial, depending on the researcher’s response. This system prevented the handler from providing any unconscious cues to the dogs, and decreased the likelihood that the dogs would pick up on extraneous visual cues that may influence their performance.

    The participants’ VAS scores were downloaded from Qualtrics and inputted to Microsoft Excel version 16.44. Each participant’s pre-task VAS stress score was subtracted from their post-task VAS stress score to establish whether their score had increased at least two points as a result of the MAT.

    Physiological data were scored using AcqKnowledge software (Biopac Systems, Goleta, CA). Heart rate and mean arterial pressure (i.e., blood pressure) were calculated by taking the mean of the last two minutes of the baseline period and the mean of all three minutes of the MAT period. The first minute of the baseline period was excluded from the analysis on the basis that there may be spurious readings as a result of the participant recently standing up and moving while the sensors were put on, and small movements in the first minute of sitting down while they became accustomed to the sensors. Physiological reactivity was determined by subtracting averaged baseline activity from the MAT period activity. Whether the participant met both the self-report and physiological criteria to use their samples in testing was assessed using these methods immediately after completing each participant’s session, as this information was required to determine whether to show the samples to the dog, or whether they should be disposed of.

    Dogs’ performances were analysed using binomial probabilities to determine if they could detect the target odour at levels above chance. The probability of choosing the target odour by chance on any given trial was 0.33. The alpha level was set at 0.05. To examine the extent that dogs may be learning within the discrimination trials, the first discrimination trial only was additionally assessed (each dog’s first exposure to each participant’s T2, D1, and B3 samples).

    To assess for the possibility that dogs were using extraneous cues specific to the apparatus to aid in their indication decisions, we analysed the dogs’ performances on the sessions prior to- and post- the odour control intervention (replacing the arms for three dogs and switching the ‘correct’ arm for one dog). Trial scores and mean performance are reported. A repeated measures t-test was run on the number of correct trials on each dog’s last session using the original apparatus and their first session using the odour control. SPSS version 28 was used for this analysis with the alpha level set at 0.05.

    Fifty-three participant’s samples were collected: 13 remotely and 40 in-person. Of those, 11 were excluded because the participant did not meet the criteria of experiencing negative stress in response to the MAT (two remote participants and nine in-person participants). The basis of exclusion due to the stress criteria were as follows: five reported no, or less than a two-point, increase in their self-report experience of stress following the MAT, five showed a decrease in BP following the MAT (indicative of a positive stress response, or “challenge” [15]), and one showed a decrease in HR following the MAT (see S1 Dataset for full list of samples taken and details on their exclusion). A further five participants’ samples were excluded: three because they were surplus to the number of dogs able to be shown samples to that day, one because the participant did not meet our criteria regarding food (had chewing gum in their mouth), and one because the participant withdrew during the task. One test session ended after ten trials because the dog was unwell at testing. In total, 36 participants’ samples were tested.

    The 36 tested samples were collected from 30 females and six males with a mean age of 25.42 years (Min = 18, Max = 57). Participants included 30 people who identified as White, three as Asian or Asian British, two as Mixed or multiple ethnic groups, and one as Black, African, Caribbean or Black British. In the baseline condition, the mean self-report VAS stress score (minimum possible score: 0, maximum possible score: 10) was 1.98 (Min = 0.00, Max = 6.51). The mean self-report VAS stress score immediately after the MAT was 7.00 (Min = 3.00, Max = 10.00). For the 25 in-person samples for which physiological data were additionally recorded, the mean heart rate during baseline was 90.54bpm (Min = 66.61bpm, Max = 115.58bpm), and during the MAT was 104.91bpm (Min = 81.30bpm, Max = 130.50bpm). Mean blood pressure (i.e., mean arterial pressure) during the baseline period was 114.50mmHg (Min = 92.66mmHg, Max = 164.95mmHg) and during the MAT was 123.21mmHg (Min = 92.81mmHg, Max = 172.34mmHg) (see S1 Dataset for details of HR and BP per participant).

    Per test session, each dog completed 10 Phase One trials (T1, B1, B2) and 20 Phase Two discrimination trials (T2, D1, B3). The focus of the reported results will be the 20 discrimination trials where the dogs were discriminating between each participant’s baseline and stress samples (for results of the Phase One trials see S1 Dataset). Due to the availability of collected participant samples coinciding with the dog owners’ schedules, each dog participated in a different number of test sessions. Each session indicates a single participant’s samples (e.g., 36 sessions = 36 within-subject tests). The dogs’ overall performances can be seen in Table 1. Each dog’s performance in individual sessions can be seen in the S1 Dataset. As a cohort (N = 36 sessions, 720 discrimination trials), the dogs alerted to the stress sample in 93.75% of trials. A binomial test, where the probability of success on a single trial is 0.33, found each individual dog’s proportion of correct trials was greater than that expected by chance (p < 0.001), and the combined cohort proportion of correct trials (675/720) was greater than that expected by chance (p < 0.001) (Table 1). When looking at each dog’s first discrimination trial performance only, Treo scored 100% (16/16 correct identification of the stress sample), Winnie scored 100% (2/2), Fingal scored 90.91% (10/11, one false alert on a participant’s baseline sample) and Soot scored 85.71% (6/7, one false alert on a participant’s baseline sample). Overall, dogs correctly alerted on the stress sample in 94.44% (34/36) of first discrimination trials. Download:

    The dogs’ performances were compared pre- and post-odour controls to assess for the potential impact of confounding odour cues informing their performance. The original apparatus and arms had been used for all training sessions prior to testing (with the exception of the “food port” arms which had been excluded and stored away once the dog had progressed to “clean” ports). Treo, Fingal, and Soot had the arms replaced after twelve, two and one testing sessions, respectively. As Winnie joined the study after the apparatus had been replaced, her session scores pre- and post-switching which arm held the target sample are reported. Each dog’s performance pre- and post-odour control can be seen in Table 2. As a cohort, results of a repeated-measures t-test showed a non-significant difference in the number of correct trials in the last session pre-odour control (M = 18.25, SD = 1.50) and the first session post-odour control (M = 19.00, SD = 0.82), t(3) = -1.00, p = 0.391. These results indicate that the dogs were not relying on additional odour cues to inform their decisions. Download:

    This is the first study to use a controlled olfactory paradigm to assess if dogs can discriminate between human odours (combined breath and sweat samples) taken at baseline and when experiencing experimentally induced negative psychological stress. To test this, we trained dogs on a two-phase, three-alternative forced-choice paradigm of increasing difficulty, initially on odour discriminations with known VOC differences: between people discrimination, progressing to within person discrimination (the same person at two times of day). Performance at above chance level (80% correct, p < 0.001) was required at these training stages before reaching testing. By doing this step-wise method, we could assume that if a dog’s performance dropped to chance at the testing stage, it was because the stress and baseline samples were indistinguishable to the dog, and not because the dog did not know how to complete the task. The results on the test samples showed that, in tests of discrimination, the dogs’ performances were consistently above chance, ranging between 90% to 96.88%, with a combined performance of 93.75% correct trials across sessions.

    To analyse the significance of the dogs’ responses using paradigms such as this, multiple trials are required to gain statistical power (the probability of a dog guessing correctly on a single trial is one in three). As such, each dog carried out twenty discrimination trials within each session to assess their ability to discriminate between the samples. However, of additional interest is each dog’s first exposure to the three, newly opened, samples at the beginning of the discrimination phase (T2, D1 and B3). Analysing this trial in isolation provides information on whether the dogs could discriminate between an individual’s baseline, stress and blank sample by recognising that the newly opened stress sample (T2) is the same odour profile that was reinforced in the learning trials (T1), while concurrently recognising that the baseline sample (D1) is distinct from what they have previously been rewarded for and should be passed over. We found that the dogs were highly successful in the first trial of each session’s discrimination phase, and correctly alerted on the stress sample in 94.44% of first exposure trials. Indeed, the dogs incorrectly alerted on the baseline sample in their first exposure on only two occasions. Even if the dogs had shown high rates of false alerts in their first exposure but had gone on to learn to discriminate throughout the twenty trials, this would still indicate that the two odours are able to be distinguished, however, this could raise questions into whether the dogs were developing extraneous cues (e.g., adding their own odour to the sample) to aid in their discrimination across the repeated trials. However, our finding that the dogs were able to discriminate on first exposure provides strong evidence that the samples conferred distinctive odour profiles. Overall, the dogs’ performances indicate that each participant’s samples were distinct at baseline compared to after the stress induction. Furthermore, the results of the odour control procedures (replacing the apparatus and switching the target arm) suggest that the dogs were not relying on extraneous or confounding cues to aid in their discrimination of the samples. These results corroborate the findings of other studies (e.g., [7, 28]) suggesting that dogs are able to detect human physiological changes associated with psychological states and, further, highlight the importance of considering transmission of odour cues in both pet dog ownership and service dog training.

    The current study adds to the field by controlling for potential odour confounds that have not been explicitly controlled in previous studies. For example, D’Aniello et al. [7] found that dogs showed more stress-indicative behaviours when in a room with a stranger and exposed to odours samples from people experiencing “fear” compared to when the dogs were in the same room with a stranger and exposed to odours of people when they were “happy”. There are two areas to consider; first, no physiological measures were used to confirm the emotional states of people donating the biological samples, and second, the samples that represented the different emotional conditions were donated one week apart, thus increasing the likelihood of confounding odours (e.g., due to diet, medication, changes etc.), regardless of the emotional states they were meant to represent. The current study addressed both of these areas by corroborating the participant’s self-reported stress with physiological measures (for 25 of 36 samples) and imposing strict odour controls to minimise the potential impact of extraneous variables assisting in the dog’s discrimination. For example, samples were taken from each participant in the same room within four minutes of each other, reducing the likelihood that dogs were able to inform their indication decisions by using background VOCs from the air in the room or exogenous VOC changes in the participant due to time passing. It should be noted, however, that the focus of D’Aniello et al.’s [7] study was emotional contagion, which was not our focus, as we trained dogs explicitly for discrimination purposes. D’Aniello et al. [7] reported more stress-related behaviours exhibited by dogs in the condition where they were presented with human sweat samples taken when they reported to be experiencing “fear”, suggesting that, not only could detect an odour, but it also had a mirroring effect on the dog’s own emotional state. It is possible that dogs in the current study were able to recognise the odour of stress as having an emotional context, however the positive reinforcement training likely interfered with this interaction. Although dog behaviour was not coded in this study, it can be anecdotally noted that no dog showed signs of distress when encountering the human stress samples. On the contrary, dogs appeared excited when they came to the stress sample as they were anticipating the clicker and food reward for a correct alert. Future studies may wish to examine the interaction between the emotional contagion of stress and positive reinforcement directly to add insight into this area. The current study does, however, provide evidence that dogs can detect an odour associated with acute stress in humans from breath and sweat alone, which provides a strong foundation for future investigations into areas such as emotional contagion knowing that there is a confirmed odour component to acute negative stress that can be detected in the absence of other visual or vocal cues.

    The results of this study add to previous research identifying VOC changes in humans who are experiencing acute stress (e.g., [52–54, 57]) by further demonstrating that dogs can detect this VOC profile change. As posited by Turner et al. [53], there are likely VOC changes associated with increased breathing rate, heart rate and blood pressure. These responses are, in part, due to a cascade of hormonal release associated with stress. The most well-known is the secretion of cortisol, but there is also the stimulation of gluconeogenesis, glycogenolysis, and lipolysis, and increased levels of renin and angiotensin II enzyme [53, 63]. Many of the physiological markers associated with a stress response are influenced by factors other than stress, for example, physical exercise, tobacco, alcohol, and the time of day [64, 65]. Because of this, a within-subject design utilising exclusion criteria minimises the potential for these physiological markers being impacted by anything other than the stress-inducing task.

    We did not attempt a definition of stress from set values of HR and BP across individuals (e.g., “HR must raise to above 100 beats per minute”) because each individual’s baseline HR and BP is contributed to by many factors and imposing such measures would be largely arbitrary. However, we did see notable differences in the actual values of baseline and stress condition HR and BP between participants. It is possible that some participants were emitting bio-markers of stress during the baseline condition in addition to after the MAT. However, given that we required a two-point increase on their self-report stress scale and an increase in HR and BP, this meant that we excluded both participants who were “not stressed” at baseline and after the MAT, and also those who were “stressed” at baseline and after the MAT, as neither group would have shown change between conditions. The varying baseline values of HR and BP do, however, raise interesting questions on what specific biomarkers the dogs may be detecting. Future studies combining VOC analysis using GC-MS and dog detection using samples taken at the same time point from the same participants would be beneficial. It should be noted that in Santos et al.’s [54] study assessing VOCs in participant’s breath after a stress-induction task, additional breath samples were taken at two time points after the task had ended: five minutes, and one hour, afterwards. The observations related to the samples obtained at these time points showed no distinguishable VOC pattern between relaxed and stressed conditions. This finding suggests that the VOC pattern observed was associated with metabolic pathways of the acute stress response, which ceased to become detectable once this process was no longer taking place. Cortisol is the most abundant glucocorticoid released in response to a stressor, and is generally acknowledged as the ‘stress hormone’. However, cortisol levels peak approximately 10 to 30 minutes after an acute stressor [66], raising questions into what metabolic processes the dogs are actually detecting. It is likely, as our samples were taken immediately upon finishing the stress inducing task, that dogs were utilising VOC changes associated with the acute stress response, however, further investigations into dogs’ performances using samples taken five minutes to one hour after a stress inducing task had ended would add insight into the timeline of the physiological response. Establishing at what point dogs can no longer distinguish between samples may add to our understanding of the scent profile of stress.

    It must be noted that our sample consists of only four dogs. This sample size is, however, in-line with other bio-detection studies due to the time-consuming nature of training highly specialised dogs (e.g., Cornu et al. [41]: one dog; Bomers et al. [43]: one dog; Taverna et al. [42]: two dogs; Murakra et al. [40]: four dogs, Kantele et al. [45]: four dogs). Importantly, as this is a proof of principle study, a small sample size does not compromise the findings, as the goal of the study is not to generalise the findings to all dogs, but rather to demonstrate that some, carefully selected and highly trained dogs can successfully discriminate between the samples. To provide evidence that a small number of dogs can detect odour differences in baseline and stress samples suggests that an odour difference exists.

    Moreover, while the two-phase within-subject paradigm utilised here is not widely used in the existing bio-detection literature, it provides well-controlled evidence that two odours are distinguishable from each other while minimising the need for a large number of human samples required to teach a generalised odour concept. With this in mind, the two-phase within-subject protocol could be recommended for future studies investigating physiological changes within an individual if access to a large number of samples is not feasible, or the ability to store samples for later use is limited. In terms of participant samples, a total of 36 is strengthened by the within-subject nature of the design. Each participant acted as their own control, minimising potential issues of variance from background VOCs associated with age, sex, ethnicity, diet and lifestyle.

    When considering the stress-inducing task itself, we should highlight that it failed to induce negative reactions to stress in 11 participants (21%). This may be due to the participants’ perceptions of the stress-inducing task. The biopsychosocial model of challenge and threat posits that stress may lead to either positive or negative outcomes depending on how an individual perceives the stressor [14–16]. Negative responses to stress (threat) were observed in the majority of our sample, and occur when an individual does not believe they have the resources to overcome a stressor and is accompanied by increased heart rate and blood pressure. However, of the 21% who did not meet our criteria for stress, five did not interpret the task as stressful (i.e., did not report an increase in VAS score of at least two points), and a further five appeared to have experienced positive stress (challenge), as they showed an increase in VAS score and HR, but a decrease in BP [15]. If future studies wished to replicate, or extend, our findings we recommend examining how these different stress responses may impact detection of VOCs, or explore manipulations to induce negative stress more consistently (e.g., Trier Social Stress Test [67]), saving researchers time and expense on unusable samples.

    While the potential issue of confounds due to unintentional visual cues have been highlighted and addressed in many canine bio-detection studies (e.g., [24, 41]), and discussed in review papers [68], it is less common to see implementation of additional odour controls. It is possible that, over time, the dog learns to integrate extraneous cues, relating to either the samples or the apparatus. Lazarowski et al.’s [69] methodological review highlights this issue (although broadly focussed on odour generalisation paradigms), and we reiterate the importance of attention brought to potential extraneous cues, beyond visual cues, in olfactory paradigms. We believe that it is important for future papers to include information specific to the reduction of extraneous odour cues, and to implement odour controls as consistently as visual controls.

    The results of this study contribute to our understanding of human-dog relationships and add insight into how dogs may be interpreting their environment and interactions with humans as informed by their olfactory capabilities. Our findings demonstrate that there is a detectable odour associated with acute negative stress that is distinct from odours at baseline. While the dogs in this study underwent training in order to communicate that they were able to distinguish between odours, the found performances on this task suggests that there are VOC changes induced by acute negative stress that are detectable by dogs. Having established that there is an odour difference, assessing how untrained dogs may recognise and interpret these odours could be of interest in future studies. The results of the current study could have further applications to the training of anxiety and PTSD service dogs, that are, currently, predominantly trained to respond to visual cues [70]. Knowing that there is a detectable odour component to stress may raise discussion into the value of olfactory-based training (e.g., taking samples from a person when relaxed and experiencing stress) and positively reinforcing the dog to attend, or perform attention seeking behaviours in response to, this odour (similarly to how Medical Assistance Dogs are trained). As the current study was laboratory based, such methods would need to be tested in applied situations for this premise to be verified. It is important to consider the welfare of service dogs tasked with performing these types of roles, especially in light of potential emotional contagion between owner and dog. A recent study by van Houtert et al. [71] tested whether service dogs supporting in stress reduction for veterans showed higher levels of hair cortisol (a proxy measure of stress) as compared to companion dogs. They found that cortisol values did not differ between service and companion animals, which may seem at odds with the results of Sundman et al.’s [28] study which found that owner and dog cortisol levels mirrored each other. Crucially, service dogs receive specific positive reinforcement counter-conditioning training that is designed to counteract any untrained stress contagion. Indeed, van Houtert et al.

    How Dogs Can Recognize a Bad Person (And Other Dog Incredible Abilities Explained)

    New Scientist Default

    I want a cuddle

    Gary John Norman/Getty

    Dog owners swear that their furry best friend is in tune with their emotions. Now it seems this feeling of interspecies connection is real: dogs can smell your emotional state, and adopt your emotions as their own.

    Science had already shown that dogs can see and hear the signs of human emotions, says Biagio D’Aniello of the University of Naples “Federico II”, Italy. But nobody had studied whether dogs could pick up on olfactory cues from humans.

    “The role of the olfactory system has been largely underestimated, maybe because our own species is more focused on the visual system,” says D’Aniello. However, dogs’ sense of smell is far superior to ours.

    D’Aniello and his colleagues tested whether dogs could sniff out human emotions by smell alone. First, human volunteers watched videos designed to cause fear or happiness, or a neutral response, and the team collected samples of their sweat.

    Next, the researchers presented these odour samples to domestic dogs, and monitored the dogs’ behaviours and heart rates.

    Dogs exposed to fear smells showed more signs of stress than those exposed to happy or neutral smells. They also had higher heart rates, and sought more reassurance from their owners and made less social contact with strangers.

    We’ve always known that dogs collect information about their social partners through different sensory channels to decide how to respond to situations, says Márta Gácsi of Eötvös Loránd University in Budapest. “However, it is not easy to investigate such processes so that we can unfold the mechanisms and separate the channels,” as this study has done, explains Gácsi.

    D’Aniello’s study suggests humans can inadvertently hijack their dogs’ emotions by releasing smells. A second study suggests dogs can return the favour, using their expressive faces.

    Juliane Kaminski of the University of Portsmouth, UK, and her colleagues have found that dogs’ faces are most expressive when they know people are looking at them.

    The researchers introduced dogs to a human who was either looking at them or facing away, and either presenting food or offering nothing. The team analysed how much the dogs’ facial movements varied in the four scenarios.

    They found that the dogs’ facial expressions varied the most when the person was looking at them. In contrast, Kaminski says there was no sign of a “dinner table effect”, “which would predict that dogs try and look super-cute when they want something from the humans.”

    “This adds to a growing body of evidence suggesting that dogs are very sensitive to human attention,” says Kaminski.

    It’s not clear precisely how dogs visually signal us and how we respond, says Monique Udell of Oregon State University in Corvallis. “This kind of research is needed to fully understand the bidirectional nature of the human-dog relationship.”

    However, there is evidence that we are susceptible to these signals. Kaminski found that when dogs were being watched they often raised their eyebrows in a particular way. This eyebrow raise is known to give shelter dogs a better chance of being rehomed. It may make the dogs’ eyes look “sad” or infant-like, creating an empathetic response.

    It is not clear what role, if any, the domestication of dogs played in the development of these behaviours. It has been suggested that dogs’ striking emotional intelligence towards humans is a product of the thousands of years we have spent with them.