The effect of arousal during and post-training on memory consolidation in detection dogs

Abstract

Memory consolidation is influenced by the environment. In companion dogs, treadmill exercise and social play are examples of post-learning arousal which improve retention performance. In humans, post-learning arousal is only beneficial when arousal is also experienced during learning. In dogs, the effect of arousal on performance during learning is dependent on dogs’ arousal level, with low arousal dogs benefiting from arousal during learning and high arousal dogs’ performance being negatively impacted by arousal during learning. This study investigated the effect of post-learning activities (social interaction, nonsocial toy access, control) on memory consolidation in candidate detection dogs learning an odor detection task. Physiological measures of arousal (cortisol and heart rate) were collected, and reward arousal scores from a pre-training temperament assessment were used as measures of steady state arousal. There was a three-way interaction between post-learning activity, training heart rate, and reward arousal on 24-hour hit rate (the number of correct alerts divided by number of target exposures). Dogs with low reward arousal in the social interaction group performed better with higher training heart rates, but dogs in the social interaction group with high reward arousal performed worse with higher training heart rates. Results of this study emphasize the importance of selecting training methods to meet the needs of individual dogs.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-09902-2.

Introduction

Memory is traditionally understood to occur in three stages: encoding, consolidation, and retrieval. Memories are stabilized via consolidation processes. Consolidation is a result of neural changes which occur over a more extended time^1,2. Consolidation is a slow process which allows the strength of neural connections to be modified by experiences occurring during and after learning¹. While stress impacts encoding and retrieval negatively, memory consolidation processes benefit from stressful conditions³. Stressful events stimulate the hypothalamic-pituitary-adrenal (HPA) axis, initiating cortisol secretion, and the sympathetic nervous system, which involves epinephrine and norepinephrine^4,5. The release of epinephrine, norepinephrine, and cortisol during stressful events can modulate memory processes^5–7.

Post-learning events which trigger spikes in salivary cortisol in humans^3,8, and pharmacological elevations of cortisol in human and non-human animals^9,10, improve memory consolidation. Similarly, post-learning injections of epinephrine also improve memory consolidation in humans and rodents^11–15. Cortisol and epinephrine influence consolidation processes by acting on the amygdala which releases norepinephrine^6,9,11,13,16. Positively arousing experiences have similar beneficial effects on consolidation but have received much less attention in research despite greater potential application to learning environments.

Research on dog memory consolidation is limited, and most research involves training visual discrimination tasks or command learning tasks, which involve operant learning with verbal discriminative stimuli. Previous studies indicated that variations in training schedules, such as the duration and frequency of training sessions, influenced acquisition rate but did not influence long term memory¹⁷. However, post-training activities were found to affect consolidation in dogs^18–20. For example, sleep improved consolidation, with EEG spectrum power and spindle occurrence during sleep having a positive correlation with performance on a command learning task^19,21. Post-learning treadmill exercise also improved memory performance after 24 hours in aging dogs²⁰. Different post-learning activities affected consolidation differently and resulted in varied learning patterns¹⁹. The process of training a naive dog to proficiency in detection of many diverse target odors is lengthy and challenging. A better understanding of how post-learning activities influence memory consolidation could help to shorten detection dog training timelines and improve long term performance²². Working dogs receive frequent training in scent detection tasks, often training on numerous odors. Training dogs on many odors can be time intensive, so it is important to find methods to shorten training timelines to reduce training costs²³.

Unlike research in other species which focuses primarily on negatively arousing post-learning interventions, dog research has mostly investigated effects of positively arousing post-learning interventions such as play on consolidation¹⁸. Post-learning play improved performance of a visual discrimination task 24 h after the intervention, with effects lasting up to one year later^18,24. Despite previous research suggesting that positively arousing intervention such as play would influence consolidation processes through activation of the same stress systems activated by negative arousal¹, play was associated with a decrease rather than an increase in salivary cortisol concentrations. These findings indicate that beneficial effects of play on memory may result from the activity of other stress hormones such as norepinephrine. Social play increased heart rate, which is often used as an indirect measure of norepinephrine, epinephrine and sympathetic activity¹⁸. The social aspect of play likely offsets typical exercise-associated cortisol release but social play may still activate the noradrenergic system of the amygdala involved in emotional regulation of memory consolidation through the sympathetic nervous system. This activation results in the expected improvement of memory consolidation processes with arousal despite opposing cortisol effects¹⁸.

There is little research on how nonsocial play influences stress systems. In rats, chewing on a stick reduced the effects of a stressful event, by reducing plasma corticosterone levels and improving performance on a spatial memory task²⁵. The presence of chew toys was found to benefit welfare of shelter dogs by reducing inactivity compared to no toy conditions²⁶. However, there is little research into how nonsocial play with a toy such as a chew toy, is related to any stress biomarkers such as cortisol in dogs. Individual play had a beneficial effect on memory consolidation for a command learning task after one week, but its effects were comparable to rest and walking conditions, suggesting that nonsocial play may not benefit memory consolidation as effectively as social play¹⁹. Chewing behavior affected working memory in kenneled working dogs, with frequency of chewing on a toy improving performance for dogs with higher levels of anxiety²⁷. Nonsocial play and social play may have different effects on memory consolidation due to different activations of stress hormones such as cortisol and epinephrine, which are known to interact with memory consolidation processes.

The degree of arousal or emotion experienced during learning and encoding may influence how effectively post-learning arousal enhances consolidation of that experience. Human research revealed an interaction between arousal at the time of encoding, based on the emotional valence of the presented picture slides, and the enhancing effects of post-learning stress-induced salivary cortisol elevations⁸. Memory performance was improved one week later for emotional but not neutral stimuli in participants who experience cold pressor stress after learning. Higher heart rate during encoding also predicted improved recall when a post-learning epinephrine injection was administered¹⁴. The interaction between arousal during encoding and arousal post-learning has not been investigated in dog memory consolidation research. In companion dogs learning commands, sleep only had a beneficial effect on retest performance for dogs experiencing a more rewarding learning environment due to a positive expectancy violation (i.e., outcome was more favorable than expected), but sleep was not compared to any post-learning control condition²⁸. However, the impact of post-learning activities which increase arousal, such as play, and their interaction with more rewarding or arousing learning environments has not been investigated. Arousal during encoding affects performance during learning, but its effect is mediated by dogs’ arousal levels²⁹. Guide dogs, which are bred and selected for low levels of arousal, performed better on an inhibitory control task when their arousal level was increased. In contrast, companion dogs that exhibited a higher levels of arousal, based on tail wagging rate, performed worse when arousal was increased²⁹. This finding suggests that arousal during learning does impact dog performance on cognitive tasks, but that temperament may influence its effect. The effect of arousal during learning on retention performance was not examined. In detection dogs, a population considered high arousal compared to guide dogs and companion dogs, reward arousal scores predicted dogs’ suitability for operational careers^30,31. In these cases, reward arousal scores are based on a temperament assessment of dogs’ behavior in the presence of a toy that is out of reach while restrained on-leash. Furthermore, reward arousal scores predicted training progress on a detection task³². It is unclear how physiological arousal during learning interacts with temperament assessments of arousal, such as reward arousal, and post-learning arousal to effect memory consolidation processes.

This study investigated the effect of two types of post-learning activity, social interaction with a human experimenter and nonsocial interaction with a chew toy, on memory consolidation following acquisition of an olfactory discrimination task. Dogs were assigned to either a social interaction, nonsocial toy, or a control condition for comparisons between the three conditions. Saliva samples were collected to measure physiological arousal and stress system activation via salivary cortisol concentration. Saliva sampling is a minimally invasive method to measure stress-induced cortisol responses, and salivary cortisol is a common and well-validated marker of stress in dog research^33,34. Heart rate was monitored throughout the study as an indirect measure of sympathetic activity^4,18. The social component of social play, combined with its inherent physical exercise, was hypothesized to enhance memory consolidation by activating the sympathetic nervous system and noradrenergic systems. Specific hypotheses were that the social interaction group would have the greatest activity level, measured with activity monitoring collars, during the intervention and a decrease in cortisol and an increase in heart rate following the intervention. It was expected that arousal during training would interact with post-learning arousal such that the effect of social interaction would be most prominent for dogs with high training heart rate. Nonsocial toy play was also hypothesized to result in a decrease in cortisol and an enhancement in performance relative to the control condition. Dogs in the control condition were monitored to control for individual differences in activity/rest in the absence of play and to determine effects on performance.

Results

Pre-training group comparisons

Time of day (morning or afternoon) did not have a significant effect on pre-training salivary cortisol concentration (p = 0.64), so time of day was not included in any other models with salivary cortisol. Groups did not differ in reward arousal scores (M = 1.84, SEM = 0.10) and reward value scores (M = 4.05, SEM = 0.07) which were used to balance group assignment. Dogs in this study had reward arousal scores ranging from 1 to 3 when reverse scored and reward value scores ranging from 3.33 to 5. Reward arousal scores were reverse scored for all analyses such that higher scores indicate higher levels of arousal. Groups did not differ on any pre-training physiological measures taken prior to the start of training: heart rate (M = 144.32 bmp, SEM = 3.08) and salivary cortisol (M = 0.26 µg/dL, SEM = 0.03). There was also no difference between groups in training performance metrics: trials to criteria (M = 15.84, SEM = 0.62), hit rate (M = 0.93, SEM = 0.012), false alert rate (M = 0.03, SEM = 0.01), and accuracy (M = 0.96, SEM = 0.01). Finally, there were no differences between groups in baseline testing performance metrics: accuracy (M = 0.91, SEM = 0.01), hit rate (M = 0.75, SEM = 0.04), and false alert rate (M = 0.04, SEM = 0.01). There were no significant differences between groups for any of the listed physiological or performance metrics (ps > 0.06).

Intervention effects

Given that groups were equivalent in their pre-training physiological metrics, reward arousal scores, and their training and baseline testing performance metrics prior to the intervention, the social interaction group, control group, and nonsocial toy group could be compared for differences during and post-intervention. The social interaction group had the highest average activity rate per minute (M = 87.54, SEM = 4.32) followed by the nonsocial toy group (M = 55.28, SEM = 3.67) and the control group (M = 41.82, SEM = 3.75). The social interaction group’s average activity level which was significantly higher than both control (LM: t = 8.23, p < 0.01) and nonsocial toy groups (LM: t = 5.81, p < 0.01). The nonsocial toy group was also significantly more active than the control group (LM: t = 2.38, p = 0.02). Figure 1 illustrates significant differences between groups on average Fitbark activity per minute during the post-learning intervention.

Fig. 1.

Average Fitbark Activity per minute of each group. Black squares indicate the means, horizontal lines represent the medians, boxes represent the interquartile ranges, whiskers indicate the range of values within 1.5 X interquartile range, and dots are the individual dogs for each group. Asterisks indicate a significant difference between groups (p < 0.05)

Figure 2 illustrates a significant interaction between group and timepoint on heart rate (p < 0.01) such that heart rate increased from pre- to post- intervention for the social interaction group (LME: t = 3.71, p = 0.003) and decreased from pre- to post- intervention for the control group (LME: t = -2.31, p = 0.04). There was no difference between pre- and post-intervention heart rate for the nonsocial toy group (LME: t=-1.56, p=0.15). Post-hoc analyses also revealed that the social interaction group had a significantly higher average post-intervention heart rate than the control group (LME: t = 2.41, p = 0.02). However, there was no interaction between timepoint and group on salivary cortisol (Control: M = 0.19, SEM = 0.02; social interaction: M = 0.19, SEM = 0.01; nonsocial toy: M = 0.25, SEM = 0.04; Pre-intervention: M = 0.20, SEM = 0.03; Post-intervention; M = 0.18, SEM = 0.02; p > .38).

Fig. 2.

Average pre- and post-intervention heart rate for each group. Black squares indicate the means, horizontal lines represent the medians, boxes represent the interquartile ranges, whiskers indicate the range of values within 1.5 X interquartile range, and dots are the individual dogs for each group. Asterisks indicate a significant difference between groups or timepoint (p < 0.05).

Correlating arousal measures

Mid-training heart rate, which was measured between Stage 2 and 3 of training, was positively correlated with post-training heart rate which was measured following Stage 4 of training (r(35) = 0.76, p < 0.01) therefore, post-training heart rate was utilized as a metric of training heart rate in subsequent analyses. Reward arousal scores were not significantly correlated with mid-training heart rate (r(35) = 0.20, p = 0.22) or post-training heart rate (r(35) = 0.13, p = 0.44) therefore, it was included with heart rate measures in subsequent analyses. Pre-intervention salivary cortisol concentration was significantly positively correlated with mid-training heart rate (r(35) = 0.47, p = 0.003) and post-training heart rate (r(35) = 0.55, p < 0.001). Training activity was significantly correlated with post-training heart rate (r(35) = 0.60, p < 0.01). All other correlations were not significant (ps > 0.05). Figure 3 illustrates the correlation matrix between arousal measures.

Fig. 3.

Correlation matrix between mid- and post-training heart rate, reward arousal score, pre-intervention salivary cortisol concentration, training Fitbark activity, and training trials to criteria. Correlation coefficients are displayed only for significant relationships (p< 0.05).

Effects of arousal on performance

Effects of individual differences in reward arousal scores on baseline test performance (prior to intervention) were found, with higher reward arousal scores (measured in a pre-study temperament assessment) significantly predicting higher accuracy in the baseline test immediately after training when either mid-training heart rate (LM: t = 2.40, p = 0.02) or post-training heart rate (LM: t = 2.46, p = 0.02) were included in the model (Fig. 4)

Fig. 4.

Baseline test accuracy as a function of reward arousal score across groups. Confidence band represents 95% confidence interval

Figure 5 shows a significant 3-way interaction between post-training heart rate, reward arousal score, and intervention on hit rate during the 24-hour retention test therefore, we analyzed the interaction between post-training heart rate and reward arousal score for each group separately. For the social interaction group only, there was a significant interaction between post-training heart rate and reward arousal score such that higher post-training heart rate was associated with higher 24-hr hit rates for dogs with low reward arousal scores, and lower 24-hr hit rates for dogs with high reward arousal scores (LM: t =-3.48, p < 0.01). This effect was not observed for 1-week retention testing hit rate. There were no significant effects of post-training heart rate, reward arousal score, or intervention on retention accuracy (combined measure of sensitivity and specificity) or false alert rate at 24-hour or 1-week retention testing.

Fig. 5.

Hit rate at the 24-hr retention test as a function of post-training heart rate for each group. Data is also grouped by reward arousal scores. Data is graphed by reward arousal in 3 groups: high reward arousal was calculated as mean + 1 SD, mean reward arousal, and low reward arousal calculated as mean – 1 SD. Confidence band represents 95% confidence interval.

Discussion

This study investigated the role of post-training activities on purpose-bred detection dogs’ retention of an odor detection task. Dogs’ arousal during and after training was measured by heart rate, salivary cortisol concentration, and physical activity level. Post-learning arousal was manipulated by participation in one of three post learning interventions: social interaction, nonsocial toy, or a control condition. While previous research has found that social play improves memory consolidation processes and retention performance compared to control conditions or solitary play post-learning^18,19, this study found that beneficial effects of social interaction were dependent on individual differences in reward arousal levels measured prior to the study and amount of physiological arousal during training.

Groups were compared prior to the intervention to determine if there were any differences that could not be attributed to the intervention. Because pre-training heart rate and pre-training salivary cortisol concentration were comparable across groups, any differences in heart rate and salivary cortisol concentration measured post-intervention can be attributed to the intervention condition of the group. Similarly, acquisition and baseline test performance were equivalent across groups, supporting the role of the interventions in affecting retention performance.

Differences observed in activity during the intervention and heart rate post-intervention suggest that the intervention activities had different effects on dogs physically and physiologically. The social interaction group had significantly higher activity than both control and nonsocial toy groups, and the nonsocial toy group was more active than the control group, in line with expected hypotheses and validating the purpose of the three intervention groups to achieve various levels of activity and physical arousal. Dogs in the social interaction group were generally interested in engaging in social interaction with the experimenter through fetch or tug and petting. While dogs in the control group rarely achieved calm states of rest, often engaging in search behavior around the room, the absence of the toy and experimenter engagement led to the observed lower level of physical activity. Finally, there were mixed levels of engagement with the chew toy in the nonsocial toy group with some dogs showing high engagement with the toy, carrying and running with the toy, and some dogs behaving more similarly to the control group, engaging in search behavior around the room without interest in the toy. It is likely that the dogs that did interact with the toy led to the difference in activity observed between the control and nonsocial toy group.

As hypothesized, the social interaction group also had a significant increase in heart rate from pre-intervention to post-intervention. This increased heart rate suggests that the sympathetic nervous system was activated in dogs in the social interaction group, achieving successful post-learning arousal. While the nonsocial toy group had significantly higher activity than the control group, there was no change in heart rate from pre-intervention to post-intervention, so activity increase was not associated with an increase in physiological arousal in this condition. Finally, the control condition had the lowest activity level, with a significant decrease in heart rate from pre-intervention to post-intervention, suggesting that this group experienced a decrease in physiological arousal as measured by sympathetic nervous system activity. Similar results were found in companion dogs with average heart rate increasing with play and average heart rate decreasing in a resting group¹⁸. The experimenter measuring heart rate was more familiar with the dogs in the social interaction group, but all of the dogs in the study are very sociable and excitable when any person is in close contact, so it is unlikely that this had any major effect on the dog’s observed heart rates. The experimenter was not blind to the group conditions, having observed all dogs in their post-learning interventions, so there is potential for bias in heart rate measurement, but the activity results confirm the heart rate findings, suggesting that the heart rate measures are valid.

In contrast to heart rate, there were no changes in salivary cortisol concentration from pre-intervention to post-intervention in any of the intervention groups. Previous studies in companion dog populations have demonstrated a significant decrease in cortisol resulting from social play¹⁸. Results of this study may differ due to differences in population characteristics. For example, social interaction may not activate the dogs’ HPA axis. Alternatively, salivary cortisol, collected only at specific timepoints, may not be sensitive enough to detect changes in HPA axis activity in a working dog population. Additionally, a salivary cortisol concentration of 0.4 µg/dl is the typical cut off used to distinguish between stressed and non-stressed dogs^35–37. In this study, only two dogs started above this cut off at the pre-intervention timepoint, while in previous studies the average group salivary cortisol concentrations started above 0.4 µg/dl prior to play¹⁸, so there may have been a floor effect in this study.

Retention performance was influenced by post-training heart rate, intervention, and reward arousal, measured in a pre-training assessment, as hypothesized. Higher post-training heart rate followed by social interaction was only beneficial for dogs with low reward arousal. The effect was opposite for dogs with high reward arousal scores, with higher post-training heart rate and social interaction being detrimental to success. These effects were not observed in the nonsocial toy or control groups. It is possible that using a toy in the social interaction group during training and testing, influenced these results. All dogs received reinforcement in the manner of play with a ball during training and retention tests, but the dogs in the social interaction group received additional interaction with the ball during their post-learning intervention. Hence, it is possible that experience with the same toy during and post-training may explain the results, rather than arousal during and post-training. Further research utilizing alternative reinforcers or toys may help separate this effect, but the increased activity level and heart rate of the social play group post-learning suggests that arousal may play the critical role in this group’s performance effects. The lack of an effect of the toy in the nonsocial toy group may be explained by limited engagement with the nylabone chew toy by the dogs in that group. Further, dogs in the study population did not represent the full range of possible reward arousal scores, with the maximum reward arousal score being a three, so a greater effect of the control and nylabone condition might be expected with higher arousal dogs.

The finding that social interaction was more beneficial with increased training-related arousal confirms previous memory consolidation research in humans suggesting that arousal during learning interacts with arousal post-learning^8,14,38. However, the effects were dependent on reward arousal levels, as measured by a behavioral test prior to the study, where high arousal levels following training were only beneficial for dogs with low reward arousal scores. In contrast, increased arousal from training led to decreased detection performance for dogs with high reward arousal scores. This effect is consistent with the Yerkes-Dodson law which predicts optimal performance with moderate levels of arousal, and reduced performance with either high or low levels of arousal, the effect of which depends on temperament levels of arousal²⁹. This effect has been demonstrated with guide dogs, which are bred for low-arousal temperaments, performing better on a self-control task when in a high state of arousal whereas high-arousal companion dogs performed worse when their arousal was further elevated²⁹. The current study extends these findings to demonstrate effects on memory consolidation and retention. However, dogs in this study did not have scores indicating extreme over-arousal, so future studies including dogs with higher levels of arousal are needed to further investigate the role of post-learning activities on dogs with more extreme arousal.

Unlike retention hit rate, retention accuracy (overall measure of combined sensitivity and specificity) and false alert rate were not affected by intervention or reward arousal level. Accuracy and hit rate findings likely differ due to the subjects’ experience with the task. Because the dogs tested in this study had previously been trained on odor line up tasks, but never on the target odor used in this study, it is likely that accuracy had less individual variability while hit rate was testing their learning a new odor with more sensitivity, allowing for different detection of effects.

Heart rate during and after training correlated with each other, so post-training heart rate was utilized in subsequent analyses, but they were both considered measures of training-related arousal. To explain possible sources of training arousal and reasons for its effects on retention performance, correlations between both heart rate metrics, other measures of physiological arousal during training (salivary cortisol and Fitbark activity), and training trials to criteria were analyzed. Both training heart rate measures were positively correlated with salivary cortisol, further emphasizing that dogs’ rising heart rate during training was associated with physiological arousal. While heart rate is associated with sympathetic nervous system activation and cortisol is associated with the HPA axis activation, the correlation between both systems’ activity indicates that dogs’ stress and arousal systems are being activated simultaneously in training when they are physiologically aroused. Reward arousal scores were not, however, correlated with any of the physiological arousal metrics measured during training. These findings suggest that a dog’s level of arousal in response to an unobtainable reward while restrained on-leash, measured during a pre-test temperament assessment, is not necessarily tied to how physiologically aroused they become during training. A previous study by our group with a similar population of dogs also suggested that while reward arousal scores (assessed in the same way as the current study) predicted training performance, it differed from other measures of arousal and activity level during training breaks³². One possibility is that this measure of arousal (measured by dogs’ arousal when presented with an unobtainable reward while restrained on-leash) may not be a valid measure of dogs’ general arousal across contexts, measuring a specific aspect of reward-direct arousal. Future detection dog research may benefit from behavioral and physiological characterization of different types of arousal, and effects of stable and context-specific arousal levels on dogs’ performance.

Unlike retention performance, baseline test performance was not influenced by training arousal. Rather, baseline test accuracy was significantly positively associated with reward arousal scores. This finding is supported by a previous study finding that reward arousal facilitates acquisition of a detection task in working dogs³². However, the maximum reward arousal score for dogs in this study was a 3, so it is possible that scores indicating over-arousal would be more detrimental to success, as predicted by the Yerkes-Dodson Law. Further research is needed to represent the full range of arousal levels. Reward arousal scores and physiological training arousal (measured with heart rate) seem to have different levels of importance in influencing baseline test performance and retention performance. It appears that training arousal measured by physiological activity may have a larger impact on consolidation and retention performance, whereas reward arousal levels may have greater influence on performance on the task measured at baseline before memory becomes a factor. Because training heart rate did not influence baseline test performance, this result suggests that training heart rate is influencing consolidation processes rather than dogs’ overall ability to perform the task.

This study had a limited number of subjects, with only 12 subjects in each group, which resulted in low power for some of the larger models included in the analyses. Including more subjects in future studies will increase the ability to detect interactions between the factors in the models. The ability to collect continuous heart rate data would also improve understanding of heart rate variability during training and the role of arousal during training. Due to inconsistency of continuous heart rate monitors which are typically designed for humans and utilized for collecting data on dogs, manual heart rate measurement was conducted at discrete timepoints in this study. While this provided more consistent data collection, there is likely a lot of information lost without the ability to see how heart rate changes throughout a training session for individual dogs. Furthermore, mid-training heart rate and post-training heart rate models occasionally resulted in slightly different results, indicating that there may be more information to gain from monitoring heart rate variability during training. To further understand the role of arousal during learning, continuous heart rate data will help illuminate individual differences in heart rate during training.

Future research in detection dog training should continue to incorporate physiological measures of arousal. While temperament metrics such as reward arousal levels clearly play an important role in determining dogs’ performance, these metrics seem to be measuring a separate mechanism than the physiological arousal metrics such as heart rate and cortisol. Further research is needed to determine how physiological metrics of arousal and temperament metrics of arousal are related and interact to affect performance and learning in detection dogs.

Overall, results of this study indicate that level of arousal both during and immediately after an odor detection training session play an important role in memory consolidation processes and retention of learning. Individual differences in reward arousal levels were also important, with the direction of the effect of training-related arousal on performance and memory differing for dogs with high versus low levels of reward arousal measured in a separate context. Dogs with lower reward arousal scores will benefit from increasing their arousal levels during and after training, while dogs with higher reward arousal scores will have worse performance with increased arousal during and after training. These findings have implications for detection dog training, emphasizing the need to assess dogs’ levels of arousal in response to different stimuli, such as rewards, prior to training and to tailor training methods to meet specific dogs’ needs by varying the arousal level of the training environment and selecting appropriate post-training activities.

Methods

All methods are reported in accordance with ARRIVE guidelines (https://arriveguidelines.org).

Subjects

Thirty-seven Labrador Retrievers bred and trained for odor detection work by AUCPS participated in this experiment. Dogs were split into three post-learning activity groups: social interaction (n = 13), nonsocial toy (n = 12), control (n = 12). Sample size was based on previous experiments on memory consolidation in dogs³². All dogs had previous training experience in odor detection, described generally in Lazarowski et al. (2018). All methods were approved by the Auburn University Institutional Animal Care and Use Committee. The Auburn University College of Veterinary Medicine is accredited by AAALAC International. This study was carried out in accordance with the Public Health Service Policy on Humane Care and Use of Laboratory Animals and the Animal Welfare Act.

Stimuli

Dogs were trained to detect a target odor of 1-bromooctane, a unique chemical compound that is used as a training odor due to its chemical properties³⁹. Given that these dogs already had experience with previous odors, distractor odors, listed in Table 1, were introduced early to increase the difficulty of the task and used at multiple stages of training and testing. Each stage of training and testing are described in more detail below in Training Stages and Retention Testing. Each stage introduced 3 novel distractors and 2 matched blanks which were cotton rounds with no additional odor added. Chemical distractors are similar intensity to the target odor. The target odor and all distractors were diluted in mineral oil to −1 log and 1 ml of solution was applied to a cotton round. The cotton rounds were placed into metal tins to prevent transfer of any chemical and odor residue to the boxes used for the detection task, and the tins were placed into black plastic odor presentation boxes. The boxes (19.7 cm³) were hinged at the top to allow placement of the odor tin inside, with a 6.4-cm diameter nose hole in the top to allow for sniffing the contents while preventing visual access or physical contact. These boxes were novel to the dogs but were similar to boxes the dogs encountered during previous training.

Table 1.

Distractor odors.

Stage 3	Stage 4	Baseline test	24-hour test	1-week retention test
2-phenylethanol	1-pentanol	Allyl butyrate	Cinnamaldehyde	Ethyl isobutyrate
(S)-(-)-limonene	(R)-(+) limonene	Ethyl valerate	4-allylanisole	1-hexanol
Mineral oil	2-ethylhexanol	Ethyl propionate	(R)-(-)-Carvone	Isobutyl propionate

Table 1. Novel odor distractors used in each stage of training and testing. Stage 3 of training is the first stage of training which distractors introduced. Stage 4 of training is the final stage of training and introduces a new set of distractors to prepare dogs for testing. Baseline test, 24-hour test, and 1-week retention test refer to the three testing sessions. More information on the training and testing sessions is provided below. All distractors were diluted in mineral oil to −1 log and 1 ml of solution was applied to a cotton round.

Group assignment

Prior to participating in the current study, dogs participated in a temperament assessment as part of a routine behavioral test of detection career suitability, unrelated to the current study, around 10 months of age. During the assessment, they are scored on multiple measures of toy engagement including reward arousal (level of physical and behavioral agitation, such as barking and jumping, in response to an inaccessible toy while leash restrained), reward focus (ability to maintain physical orientation towards a toy in the presence of an auditory distraction), reward persistence (level of persistence towards obtaining an inaccessible toy inside of a transparent, locked plastic container), and reward possession (desire to maintain physical possession of toy). Definitions and scoring for these measures are described in more detail in Lazarowski et al. (2021). Dogs were scored on a scale of 1–5, with higher scores reflecting higher levels of reward motivation. The definitions for each score for each measure are provided in supplementary table S1. Group assignment for the current study was balanced by these metrics to ensure all groups were similar in reward motivation levels that could influence training. Scores for reward possession, reward persistence, and reward focus were averaged to generate a single “reward value” score. Dogs were ranked first by their reward value score and secondly by their reward arousal score. Within each litter dogs were then assigned to groups using alternating assignment, alternating for each litter whether alternating assignment began with the highest or lowest ranked dog.

General procedure

Dogs were trained on an odor detection lineup task with the target odor until they met a pre-determined criterion. Following the training session dogs completed a baseline test and then participated in their designated post-learning interventions for fifteen minutes. Retention was measured at three time points: a baseline test following training and prior to the intervention, 24 h following the intervention and one week following the intervention. Saliva sampling occurred pre-training, pre-intervention, and 20 min post-intervention. Heart rate measurement occurred pre-training, mid-training, post-training, pre-intervention, and post-intervention. Training and testing occurred in the same training room measuring 4.9 × 5.6 m. Post-training interventions occurred in a separate neighboring room measuring 3 × 6 m. Time of day of training sessions was balanced by group across litters to account for possible daily fluctuations in salivary cortisol concentrations⁴.

Training

Dogs had previous experience in searching a lineup of boxes and performing a sit response at the position containing a target odor. In this study, dogs were trained using a lineup with six odor presentation boxes arranged in a line. Dogs began at one end of the line of boxes and checked each box as they worked, on leash, toward the opposite end of the line. Dogs were allowed to search one pass of the lineup on each trial unless they did not search any boxes on the first pass, then they were allowed to search from the end of the lineup back to the beginning one time.

Training was completed in a single session with short breaks (approximately 5 min) to prevent fatigue or overarousal. Training occurred in four stages within the session. In all stages an experimenter recorded the number of trials to final criterion, number of false alerts, number of misses, number of correct rejections, and number of correct alerts (hits). False alerts were defined as an incorrect sit response at any box which did not contain the target odor. Misses were defined as incorrect trials in which the dog reached the target odor, checked the box, and continued past that box without alerting. Checking the box was defined as placing their nose into or within an inch above the hole in the top of the box. Dogs were allowed to pass the box and return to alert if they only moved one position beyond the target box. Once they moved beyond one box position from the target, a miss was noted, and the trial was counted as incorrect. Correct rejections were counted every time the dog checked a box without the target odor present and continued past the box without alerting. Correct alerts were recorded when the dog offered a sit response at the target odor location.

Training stages

• Stage 1: Assisted alerts. This stage consisted of three trials and was designed to teach the dog to detect and alert to the target odor. In this stage, the target odor was paired with reward by having a ball present in the box along with the target odor, in order to elicit the dog’s attention to the target odor and establish an association. When the dog checked the box containing the target odor, the handler assisted the dog to alert. Handler assistance involved the handler verbally prompting the dog to sit with some physical assistance such as leash tension if the dog did not respond to the verbal command. Dogs were rewarded with a ball by the handler after a sit response was performed at the target odor position. The target odor was presented in the first box for the first two trials and moved to the second box position for the third trial. There were no distractors present during the first stage. If the dog offered a false alert, the trial ended and was marked incorrect.

• Stage 2: Independent alerts. The ball was removed from the box in this stage and the target odor was present alone in the box. The target odor position in the lineup was pseudorandomized such that it did not appear in the same position more than two trials in a row. The handler assisted the dog to alert if they observed a change of behavior, defined as a distinct change in posture or behavior indicative of odor recognition that did not result in a full alert response⁴⁰. Assisted trials were not counted toward criterion and the count of continuous correct trials was reset to zero. This stage did not have a set number of trials but continued until dogs met a set criterion of three trials in a row with correct, unassisted alerts.

• Stage 3: Introduction of distractors. The target odor location was pseudorandomized as above, and a different distractor was added to three of the five non-target boxes. On each trial, a distractor was moved to the previous location of the target. The other two boxes served as matched blanks containing empty tins and clean, un-odorized cotton rounds. This stage consisted of a set of three assisted trials because dogs had no previous experience with these distractors.

• Stage 4: Independent alert with distractors present. These final trials were designed to confirm that the dog could consistently perform an unassisted alert to the target odor while discriminating it from distractor odors. The three previous distractors from Stage 3 were replaced with three novel distractors. If the dog offered a false alert the trial ended and was marked incorrect. The criteria to complete this stage required three consecutive unassisted alerts. The target odor location was pseudorandomized as above. Handlers assisted the dog to alert if they observed a change of behavior not followed by an alert response at their discretion. Assisted trials did not count toward criteria and the count of continuous correct trials was reset to zero.

Dogs were given brief breaks every 10 trials and between each stage of training. Dogs were removed from the testing area and returned to their crate on the transport vehicle during breaks to reduce arousal from training stimuli. Breaks lasted approximately five minutes to allow for set up for the next stage of training.

Post-learning interventions

Immediately following training and saliva sample collection, dogs participated in their assigned post-training activity of social interaction, nonsocial toy access, or control. Interventions lasted fifteen minutes. The same female experimenter was present for each intervention condition. The experimenter who conducted the interventions was not the handler who trained the dog on the detection task.

• Social interaction. Dogs in the social interaction group were brought to a room separate from the training room to engage in social play with an experimenter. There were no other dogs or humans in the room and noise heard outside the room was minimized by limiting human and dog activity outside of the building. The same experimenter carried out the play sessions for all dogs in this condition. Toy choice, tug or ball, was based on individual preference of the dog and was determined during training and by the counsel of trainers. During play sessions, the experimenter threw the toy at least once and attempted to take the toy from the dog at least once. Otherwise play style was not restricted and tugging, fetching, or any other cooperative play styles were used such that the play experience was tailored to the behavior and preferences of the dog. The experimenter continued to encourage play even if the dog seemed disinterested. Dogs were not forced to play, but the experimenter continued interacting with the dog via verbal engagement and/or physical interaction throughout the session. To prevent overlap with the experience of the nonsocial toy group, the experimenter tried to redirect any attempt by the dog to engage with the toy individually (e.g., stationary chewing). The play session was structured into three five-minute segments. The first five minutes consisted of playful interaction with the toy as described above. The second five minutes provided a break from playing and instead included five minutes of petting to prevent overheating. The experimenter removed the toy from sight, if possible, but if the dog did not willingly surrender the toy, the dog could keep it while being petted. The final five minutes involved return to social play with the same parameters as the first five minutes.

• Nonsocial toy. Dogs assigned to the nonsocial toy intervention were brought into the same room used for social interaction intervention. Dogs were given a Nylabone chew toy. Dogs were left to interact with the toy for 15 min and monitored by an experimenter. The experimenter did not interact with the dog and stood passively in the room, only intervening if there was a risk of injury with the toy. The experimenter said the dog’s name and reintroduced the toy every 5 min if the dog was not interacting with the toy.

• Control. Dogs in the control group were moved to the same intervention room for fifteen minutes with a passive experimenter. If the dog approached the experimenter, the experimenter remained still, made no eye contact, and ignored the dog.

Retention testing

Retention testing occurred at three time points: immediately following training prior to intervention (baseline), 24 h after the intervention, and one week after the intervention. All retention time points followed the same procedure as Stage 4 of training. Dogs returned to the original testing location where they participated in a set of ten trials with no assistance from handlers. There were three novel odor distractors and dogs were scored on the number of hits, false alerts, correct rejections, and misses. The handler was blind to the target location and experimenters were out of sight.

Salivary cortisol

Saliva collection occurred at three timepoints to assess salivary cortisol levels: a pre-training sample before training, a post-training sample immediately after the first retention test, and a post-intervention sample at 20 min after the intervention ended. Training start times were balanced across groups to account for differences in cortisol concentration. Samples were collected after 0800 to avoid early morning cortisol spikes⁴¹. The second and third salivary samples were collected within two hours of the initial timepoint.

Food was provided at least one hour prior to pre-training saliva sampling to minimize sample contamination. Water was restricted for 30 min prior to saliva sampling to prevent dilution of the samples. Water access was provided after sample collection and throughout training during breaks.

Saliva was collected with a Salimetrics collection kit with an oral swab (SalivaBio Children’s Swab Device, Salimetrics) which was held in the mouth for 1 min⁴¹. Saliva samples were stored immediately on ice and placed in a freezer for storage within 3 h of collection. Cortisol samples were assayed at the Salimetrics’ SalivaLab (Carlsbad, CA) using Salimetrics Salivary Cortisol Assay Kit (Cat. No 1-3002), without modifications to the manufacturers’ protocol. This assay has been used previously in dogs.⁴⁶ Prior to testing, samples were stored at −80 °C before being shipped on dry ice to the Salimetrics SalivaLab (Carlsbad, CA). Samples were thawed to room temperature, vortexed, and then centrifuged for 15 min at approximately 3,500 RPM (1,500 x g) immediately before performing the assay. Samples were tested for salivary cortisol using a high sensitivity enzyme immunoassay (Cat. No. 1-3002). Sample test volume was 25 µl of saliva per determination. The assay has a lower limit of sensitivity of 0.007 µg/dL, a standard curve range from 0.012 to 3.0 µg/dL, and an average intra-assay coefficient of variation of 4.60%, and an average inter-assay coefficient of variation of 6.00%, which meets the manufacturers’ criteria for accuracy and repeatability in Salivary Bioscience and exceeds the applicable NIH guidelines for Enhancing Reproducibility through Rigor and Transparency.

Heart rate

Heart rate of the dogs was measured at intervals during the experiment as a measure of arousal and an indirect measure of sympathetic activity. Specifically, heart rate was measured pre-training, mid-training, post-training, and pre- and post-intervention. The experimenter counted the pulse in the femoral artery and calculated the rate.

Activity monitors

All dogs were fitted with a FitBark 2 (FitBark^®) activity monitor which was attached to their collar. FitBark monitors are accelerometers specifically designed to measure dog activity and are validated for off-leash dog activity measures⁴². Data were extracted for all periods of data collection with specific attention given to activity levels during training and post-learning activities.

Statistical analyses

All statistical analyses were performed using R statistical software (Version 4.2.1, R Core Team 2022) and using the lme4 package (Bates, 2015).

Pre-training arousal measures

Pre-training heart rate was analyzed with one-way ANOVAs with group as a between subject factor to assess differences in heart rate between groups prior to the interventions. Pre-training salivary cortisol was also analyzed with a general linear model with group and time of day as fixed factors to assess differences in salivary cortisol between groups prior to the interventions. Cortisol values were log transformed for analyses. Time of day was included in the cortisol model to determine if being tested in the morning or afternoon influenced dogs’ pre-training cortisol levels. Cortisol samples were duplicate tested, but because some samples did not have sufficient quantity for a second assay, only the first replicate assay was used for all dogs. Finally, to ensure that reward arousal, was balanced across groups prior to experimental manipulations, reward arousal scores were analyzed with one-way ANOVAs with group as a between-subjects factor. Reward arousal scores from the pre-training temperament assessment were also used in later analyses. Reward arousal levels are predictive of working dog success and have been shown to be stable between 11 months and 3 years of age³¹so we utilized reward arousal scores as a stable measure of dogs’ arousal levels. This task involves tossing a ball while the dog is held on leash at a distance where they cannot obtain the reward³³. The dog’s on-leash behavior in the presence of the unobtainable reward is observed for 25 s and then the dog is allowed to engage with the toy.

Training acquisition and baseline test performance

Hit rate, false alert rate, and accuracy were calculated from the measures collected during training and test performance: hits, misses, false alerts, and correct rejections. Hit rate was calculated as the number of hits divided by the number of target exposures. False alert rate was calculated as number of false alerts divided by the number of false alerts and correct rejections. Accuracy was calculated as the number of correct rejections and hits divided by the number of false alerts, misses, correct rejections, and hits. To assess differences between groups in task acquisition prior to intervention, the sum number of trials across all stages needed to reach final training criteria was calculated for each dog and analyzed by a one-way ANOVA with group as the between subject factors to determine if all groups had equal training acquisition prior to the interventions. The number of false alerts, accuracy, and hits were also calculated for each dog for their performance aggregated across all stages of training and analyzed with a one-way ANOVA with group as a between subjects factor. Only 7 out of the 37 dogs tested required any assistance during Stage 4 of training, with only two dogs needing more than one assisted trial and no dogs needing more than two assisted trials, so the influence of assisted trials on performance was not assessed. To assess testing performance prior to any experimental manipulations, baseline test hit rate, accuracy, and false alert rate were also compared between groups with a general linear model with group as a fixed factor.

Intervention effects

To assess the difference in activity level achieved by dogs in each intervention, average FitBark activity per minute was calculated and analyzed with a general linear model with group as a fixed factor. To assess physiological changes due to the interventions, change in heart rate and salivary cortisol concentration from pre-intervention to post-intervention were analyzed with general linear mixed effect models with fixed factors of group (control, social interaction, nonsocial toy) and timepoint (pre-intervention, post-intervention) and an interaction between group and timepoint with a random effect of dog.

Retention performance

To determine how various arousal measures were related, a Pearson’s correlation was conducted between mid-training heart rate, post-training heart rate, reward arousal, pre-intervention cortisol, training activity, and training trials to criteria. Reward arousal was reverse scored for all analyses so that higher scores would indicate higher levels of arousal.

To assess the effect of training arousal and reward arousal score on baseline test accuracy, hit rate, and false alert rate, a series of general linear models were conducted with fixed effects of mid-training or post-training heart rate and reward arousal score and an interaction between the two. If the interaction was not significant it was removed from the model. To assess the role of training arousal and post-learning interventions on retention performance, a series of general linear models were conducted with training heart rate, reward arousal score, and group as fixed effects, all two-way interactions, and a three-way interaction between post-training heart rate, reward arousal score, and group. Non-significant interactions were removed from the model in a reverse stepwise manner. These models were conducted separately for 24-hour and 1-week retention accuracy, hit rate, and false alert rate.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

CCP: Conceptualization, Methodology, Formal Analysis, Investigation, Writing – Original Draft, Writing – Review & Editing, Visualization, Project Administration. SK: Conceptualization, Methodology, Formal Analysis, Funding Acquisition, Writing – Review & Editing. LL: Conceptualization, Methodology, Funding Acquisition, Writing – Review & Editing. JSK: Conceptualization, Methodology, Formal Analysis, Supervision, Writing – Review & Editing.

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Additional information

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.