Dopamine in Dogs: Reward, Motivation and Learning

1. What Dopamine Is and What It Does

1.1 Origin and Pathways

Dopamine is a monoamine synthesized from the amino acid tyrosine, produced primarily in two midbrain nuclei: the ventral tegmental area (VTA) and the substantia nigra pars compacta (SNc). From there, dopaminergic neurons project to the striatum (caudate and putamen), the nucleus accumbens, and the frontal cortex.

The broad anatomy is conserved across mammals, which is what licenses applying the model to dogs at all. The VTA-to-ventral-striatum pathway is associated with value-based learning; the SNc-to-dorsal-striatum pathway with motor learning and habit formation. Both pathway assignments come from rodent and primate work.

1.2 Receptor Families

Dopamine acts through five receptor subtypes grouped into two families: D1-like receptors (D1, D5), broadly excitatory, and D2-like receptors (D2, D3, D4), broadly inhibitory. The balance between them shapes how dopamine signalling translates into behavior — too little dopaminergic function is associated with impaired motivation, too much with impulsivity and repetitive behavior. These are pharmacological generalizations from other species; no study has characterized receptor-level function in behaving dogs.

1.3 Wanting, Not Liking

The most consequential correction to the popular account is that dopamine tracks wanting rather than liking. Pleasure appears to depend more on opioid and endocannabinoid signalling; dopamine drives the pursuit — the anticipation, the effort, the willingness to try again. For training, this distinction matters: a dog that works enthusiastically is not necessarily a dog experiencing more pleasure, but a dog whose expectation of a worthwhile outcome is high (a distinction that also underpins frustration and failed impulse control).

2. Reward Prediction Error: The Core Model

2.1 The Three Cases

The dominant account of dopamine's role in learning is reward prediction error (RPE): the discrepancy between the reward expected and the reward received (Schultz, Dayan & Montague, 1997). Three cases follow.

When the outcome is better than expected, dopamine neurons fire in a phasic burst — a positive RPE, which strengthens the association with whatever preceded it. When the outcome is worse than expected or absent, firing drops below baseline — a negative RPE, which weakens the association and is the mechanism underlying extinction (and why extinguished behavior tends to return). When the outcome matches expectation exactly, there is little net change and correspondingly little learning: the system has nothing to update.

2.2 The Shift from Reward to Cue

As an association is learned, the dopamine response migrates backwards in time, from the reward itself to the earliest reliable predictor of it. This is the mechanistic account of why a clicker or verbal marker becomes reinforcing in its own right: it comes to occupy the position in the sequence where the prediction error is generated.

2.3 Where the Evidence Comes From

Schultz and colleagues established RPE coding by recording directly from dopamine neurons in behaving primates, and the model has since been confirmed and causally manipulated in rodents using optogenetics and pharmacology. That is the basis of confidence in the framework. It is also the reason the framework cannot be treated as canine fact: none of these methods has been used in dogs (the prediction-error account as applied specifically to dogs).

3. What Canine Imaging Actually Shows

3.1 The First Study Was Two Dogs

Berns, Brooks and Spivak (2012) trained two dogs to lie still in an MRI scanner without sedation or restraint and compared the response to a hand signal predicting food against one predicting nothing. They observed caudate activation to the reward-predicting signal in both animals. This was a genuine methodological breakthrough — awake, voluntary, unrestrained canine neuroimaging — but as a finding about dopamine it rests on two subjects.

3.2 Heterogeneity Between Dogs

The replication is the more informative study, and it is frequently misattributed to the 2012 paper. Berns, Brooks and Spivak (2013) extended the sample to thirteen dogs and reported that eight of thirteen showed a positive differential caudate response to the reward signal, with a mean differential response of 0.09% — comparable in magnitude to human studies.

Two things follow. First, the group-level effect is real and of a plausible size. Second, roughly a third of the dogs did not show it, which is a substantial individual difference and a caution against treating any single dog as an instance of the group pattern (consistent with what temperament research finds generally).

3.3 Food Versus Praise

Cook, Prichard, Spivak and Berns (2016) scanned fifteen awake dogs, using ventral caudate activation as a measure of intrinsic reward value, and compared responses to stimuli predicting food, praise, or nothing. Thirteen of fifteen dogs showed roughly equal or greater caudate activation to the praise-predicting stimulus than to the food-predicting one. Crucially, the relative activation predicted each dog's later choices in a Y-maze task, meaning the neural measure tracked something behaviorally real and stable within individuals while varying between them.

An earlier study from the same group found caudate activation to the odour of a familiar human even without any associated reward (Berns, Brooks & Spivak, 2015), which points in the same direction: social stimuli can carry reward value in their own right.

The practical reading is that reinforcer value is individual and empirically discoverable, not something to be assumed from species-level claims.

3.4 What fMRI Cannot Tell Us

Functional MRI measures the BOLD signal — a haemodynamic correlate of neural activity — not dopamine release. Caudate activation is consistent with dopaminergic reward signalling because of what is known from other species, but it does not demonstrate it. Temporal resolution is also orders of magnitude coarser than phasic dopamine firing. Every dopamine claim in this section is therefore an inference from a proxy measure in a small sample of unusually cooperative, scanner-trained dogs.

4. Action Prediction Error: A Second Teaching Signal

4.1 The Finding

Until recently, the idea of a second dopaminergic teaching signal was theoretical. Greenstreet et al. (2025) provided direct evidence in mice: movement-related dopamine activity in the tail of the striatum encodes an action prediction error (APE) — the difference between the action taken and the extent to which that action was predicted. Causal manipulations showed this signal functions as a value-free teaching signal that reinforces repeated associations, and computational modelling plus experiments showed APE alone cannot support reward-guided learning, but paired with RPE circuitry it consolidates stable stimulus–action associations independently of reward value.

4.2 What It Might Mean for Habits

If a comparable system operates in dogs, it would offer a mechanism for several familiar observations: why behaviors become automatic with repetition even as reinforcement thins out, why established habits — desirable and undesirable alike — resist change, and why some repetitive behaviors are unusually resistant to extinction, since a value-free signal would not be undone by removing the reward.

4.3 How Far to Take It

This is now a properly published finding in a high-profile journal with causal evidence, which is a considerable upgrade over speculation. It is also entirely mouse data, in one specific circuit, on one specific task. Whether dogs have an equivalent system is untested. Training decisions should continue to rest on the RPE framework, which has broader cross-species support; APE is worth knowing about as a plausible explanation for the durability of repetition, not as a basis for method (related to how habits and automatic behavior are treated generally).

5. Translating the Model into Training

5.1 Continuous Reinforcement First

For a behavior the dog has not yet learned, reward every correct response. The association has to be built before it can be maintained, and predictability is what builds it.

5.2 The Partial-Reinforcement Caution

The common advice to move to intermittent reinforcement quickly deserves a hard look, because it has been tested in dogs. Cimarelli et al. (2021) clicker-trained two groups of naïve dogs on a novel behavior: one group received food after every click, the other after 60% of clicks. Partial rewarding did not improve learning speed, and the partially rewarded dogs subsequently showed a more pessimistic bias in a cognitive bias test than the continuously rewarded dogs.

This is a direct canine result and it contradicts a widespread training assumption. In inexperienced learners, thinning reinforcement buys nothing in speed and appears to cost something in affective state (which is precisely how emotional state and learned behavior come apart).

5.3 Variability and Surprise Once Behavior Is Established

For behavior that is already fluent, the RPE logic still applies: fully predictable rewards generate no positive prediction error, and unexpected high-value rewards generate the largest. Variability and occasional jackpots are therefore reasonable tools for maintenance — with the Cimarelli finding as the boundary condition on when to introduce them (schedules and their behavioral effects in detail).

5.4 Anticipation as the Motivating State

Because dopamine tracks anticipation rather than consumption, the structure surrounding a reward matters as much as the reward. Cues that reliably predict good outcomes, a session that builds rather than plateaus, and variation that keeps outcomes worth predicting all work with the system. Mechanical treat delivery does not.

5.5 Assess Reward Value Individually

Given that caudate response to food versus praise varied substantially between dogs and predicted their actual choices (Cook et al., 2016), reinforcer value should be established per dog. Simple preference tests — offering a choice between food and interaction, repeatedly and in a controlled way — are sufficient for practical purposes (and illustrate why behavioral measurement needs to be operationalized).

6. Dopamine and Problem Behavior

6.1 ADHD-like Dogs

González-Martínez et al. (2023) studied 58 dogs, 36 of them classified as ADHD-like following physical and behavioral assessment supported by validated questionnaires, and measured serum serotonin and dopamine by ELISA. The ADHD-like dogs showed lower concentrations of both, and levels were associated with aggression, hyperactivity, and impulsivity.

Two caveats are essential. First, serum dopamine does not reflect brain dopamine — peripheral catecholamines do not cross the blood–brain barrier, and the relationship between blood levels and central signalling is not established. Second, the design is correlational. The 2024 review from the same group frames ADHD-like expression as a gene–environment interaction rather than a fixed neurochemical deficit (González-Martínez et al., 2024). "ADHD-like" is also a behavioral construct, not a recognized veterinary diagnosis (covered in full in the ADHD-like traits article).

6.2 Compulsive Behavior

Canine compulsive behaviors — tail chasing, flank sucking, pacing, shadow chasing — are widely attributed to dopaminergic dysregulation, often with the suggestion that the behavior itself becomes self-reinforcing through dopamine release. The proposal is coherent and consistent with the pharmacology of stereotypies in other species. It should be labelled clearly as a hypothesis: the canine treatment literature centres on serotonergic medication combined with behavior modification, and the dopaminergic account of canine compulsive disorder has not been established by controlled canine trials (repetitive behavior is better approached through the flexibility literature).

6.3 Impulsivity

The relationship between dopamine and impulsivity is not monotonic. Some findings link high accumbens dopamine release to impulsive choice; others link low dopaminergic function to poor impulse control. A U-shaped relationship is plausible, in which both extremes impair regulation. In dogs, the practical complication is that impulse control is not a single measurable trait: measures do not correlate across tasks (Brucks et al., 2017), which makes population-level dopamine claims difficult to map onto individual dogs (as the work on frontal control and self-regulation shows) — and impulsivity of this kind is often what presents clinically as reactivity (seen from the neurological side).

6.4 Dopamine Does Not Act Alone

Dopamine operates inside a network with extensive feedback. The "reward cascade" — serotonin triggering met-enkephalin release in the VTA, disinhibiting dopamine neurons in the accumbens — is a didactic model drawn largely from human addiction research. It usefully illustrates that a disruption anywhere in the system, such as serotonin depletion under chronic stress, can reduce dopaminergic function downstream. It is not a verified canine circuit, and modern accounts emphasize bidirectional influence, multiple additional transmitters, and region- and receptor-specific effects that the simple cascade omits.

7. Clinical Implications

7.1 Treat Chronic Stress as a Prerequisite

Chronic glucocorticoid elevation reduces monoamine function in the models this article draws on. A dog carrying a high baseline stress load should not be expected to show normal reward processing, which makes stress management a precondition for reward-based training rather than an adjunct to it (the full account of chronic stress in dogs, and how anxiety builds on the same neurochemistry).

7.2 Aversive Methods Work Against the System

Aversive methods raise stress and are associated with a more pessimistic affective state in dogs (Vieira de Castro et al., 2020). Whatever the mechanism, they degrade the conditions under which reward-based learning operates rather than substituting for it (the neurological fallout in detail).

7.3 Recognize When Something Else Is Going On

Persistently poor motivation despite genuinely high-value reinforcers, extreme impulsivity, repetitive behavior, or an inability to learn from reinforcement warrant veterinary behavioral assessment rather than more training. Medication may restore the conditions under which learning is possible — it does not replace the learning (and arousal regulation is often the more immediate lever).

8. Summary: Dopamine in Canine Learning at a Glance

Reward prediction error (RPE) — Signal: the difference between expected and received reward. Effect: positive RPE reinforces the preceding behavior, negative RPE weakens it, zero RPE produces little learning. Training implication: build with continuous reinforcement, maintain with variability and surprise. Evidence: strong in primates and rodents including causal manipulation; canine fMRI consistent but indirect.

Action prediction error (APE) — Signal: the difference between the action taken and the action predicted. Effect: a value-free teaching signal that reinforces repetition and stabilizes habits. Training implication: none yet; explanatory rather than prescriptive. Evidence: causal evidence in mice (Greenstreet et al., 2025); untested in dogs.

Reward type — Food and praise both activate the ventral caudate, with praise equalling or exceeding food in 13 of 15 dogs, and the relative response predicting individual choice behavior. Training implication: establish reinforcer value per dog rather than assuming it. Evidence: canine fMRI, n = 15 (Cook et al., 2016).

Reinforcement schedule in naïve dogs — Partial rewarding at 60% did not improve learning speed and was followed by a more pessimistic cognitive bias. Training implication: reinforce continuously while a behavior is being acquired. Evidence: direct canine experiment (Cimarelli et al., 2021).

Dopamine and problem behavior — Lower serum dopamine and serotonin in ADHD-like dogs, associated with hyperactivity, impulsivity, and aggression. Training implication: veterinary assessment where the picture fits; not a basis for self-diagnosis. Evidence: correlational, serum rather than brain (González-Martínez et al., 2023).

9. Research Gaps and Critical Appraisal

No causal dopamine evidence exists in dogs. No study has manipulated dopamine in dogs and measured the effect on learning. The causal work — optogenetics, microdialysis, single-unit recording, receptor antagonism — is rodent and primate. Everything canine is correlational or inferred.

fMRI is a proxy, not a dopamine measurement. BOLD signal reflects blood oxygenation. Caudate activation is interpreted as reward-related because of cross-species knowledge, not because dopamine was observed.

Canine imaging samples are small and unrepresentative. Two dogs in the first study, thirteen in the replication, fifteen in the praise-versus-food study. All were volunteer pets trained to tolerate a scanner — a selected population unlikely to represent behaviorally compromised dogs.

The group effect masks real heterogeneity. Only 8 of 13 dogs showed the positive differential caudate response (Berns et al., 2013). Individual variation is not noise around the finding; it is part of the finding.

Serum is not brain. The ADHD-like dopamine result is a peripheral blood measurement. Peripheral dopamine does not cross the blood–brain barrier, so it cannot be read as an index of central dopaminergic function.

APE is mouse data. Greenstreet et al. (2025) is strong work with causal manipulation, in mice, in one striatal subregion, on one task. Its extension to dogs is speculation.

The reward cascade is a teaching model. It is drawn from human addiction research, partly contested in humans, and unvalidated in dogs.

The compulsive-behavior account is a hypothesis. The dopaminergic explanation of canine compulsive disorder is plausible but not established by controlled trials in dogs.

10. Conclusion

Dopamine is central to how dogs learn, but the useful version of that claim is narrower than the popular one. Dopamine tracks anticipation and prediction error rather than pleasure, which is why surprise and variability sustain motivation while perfect predictability flattens it — and why a dog that will not work for food in one context is usually telling you about expectation and state rather than about willingness. The framework is well founded because the underlying neuroscience is robust in other mammals; the specifically canine evidence is thinner than the confident tone of most training literature suggests, resting on a handful of small imaging studies that measure a proxy signal, plus blood measurements that do not index the brain. That is not a reason to discard the model. It is a reason to hold it as a working framework and to let the direct canine findings carry the practical weight where they exist — reinforce continuously while a behavior is being learned, establish each dog's reinforcers individually, keep outcomes worth predicting, and treat chronic stress before expecting reward-based learning to work at all.

Key Insights (Takeaways)

• Dopamine is about wanting rather than liking. It tracks anticipation and prediction error, which is why the structure around a reward — predictability, variability, surprise — matters as much as the reward itself (Schultz et al., 1997, from primates and rodents).

• The canine imaging evidence is real but small and heterogeneous. The first study used two dogs; the replication found a positive differential caudate response in only 8 of 13 (Berns et al., 2012, 2013), and fMRI measures blood oxygenation rather than dopamine.

• Reinforcer value is individual and measurable. Praise-predicting stimuli produced equal or greater caudate activation than food in 13 of 15 dogs, and the relative response predicted each dog's actual choices (Cook et al., 2016) — so assess reinforcers per dog instead of assuming food ranks highest.

• Do not thin reinforcement too early. In naïve dogs, rewarding 60% of clicks did not speed learning and was followed by a more pessimistic cognitive bias compared with continuous rewarding (Cimarelli et al., 2021). Build continuously, then introduce variability once the behavior is established.

• A second dopaminergic teaching signal exists in mice. Action prediction error is a value-free signal that reinforces repetition and may explain why habits are durable (Greenstreet et al., 2025) — but it is mouse data and should inform explanation, not method. Likewise, lower serum dopamine in ADHD-like dogs (González-Martínez et al., 2023) is a peripheral, correlational finding and does not measure brain dopamine.

References

Berns, G. S., Brooks, A. M., & Spivak, M. (2012). Functional MRI in awake unrestrained dogs. PLoS ONE, 7(5), e38027. https://doi.org/10.1371/journal.pone.0038027

Berns, G. S., Brooks, A., & Spivak, M. (2013). Replicability and heterogeneity of awake unrestrained canine fMRI responses. PLoS ONE, 8(12), e81698. https://doi.org/10.1371/journal.pone.0081698

Berns, G. S., Brooks, A. M., & Spivak, M. (2015). Scent of the familiar: An fMRI study of canine brain responses to familiar and unfamiliar human and dog odors. Behavioural Processes, 110, 37–46. https://doi.org/10.1016/j.beproc.2014.02.011

Brucks, D., Marshall-Pescini, S., Wallis, L. J., Huber, L., & Range, F. (2017). Measures of dogs' inhibitory control abilities do not correlate across tasks. Frontiers in Psychology, 8, 849. https://doi.org/10.3389/fpsyg.2017.00849

Cimarelli, G., Schoesswender, J., Vitiello, R., Huber, L., & Virányi, Z. (2021). Partial rewarding during clicker training does not improve naïve dogs' learning speed and induces a pessimistic-like affective state. Animal Cognition, 24(1), 107–119. https://doi.org/10.1007/s10071-020-01425-9

Cook, P. F., Prichard, A., Spivak, M., & Berns, G. S. (2016). Awake canine fMRI predicts dogs' preference for praise vs food. Social Cognitive and Affective Neuroscience, 11(12), 1853–1862. https://doi.org/10.1093/scan/nsw102

González-Martínez, Á., Muñiz de Miguel, S., Graña, N., Costas, X., & Diéguez, F. J. (2023). Serotonin and dopamine blood levels in ADHD-like dogs. Animals, 13(6), 1037. https://doi.org/10.3390/ani13061037

González-Martínez, Á., Muñiz de Miguel, S., & Diéguez, F. J. (2024). New advances in Attention-Deficit/Hyperactivity Disorder-like dogs. Animals, 14(14), 2067. https://doi.org/10.3390/ani14142067

Greenstreet, F., Martinez Vergara, H., Johansson, Y., Pati, S., Schwarz, L., Lenzi, S. C., Geerts, J. P., Wisdom, M., Gubanova, A., Rollik, L. B., Kaur, J., Moskovitz, T., Cohen, J., Thompson, E., Margrie, T. W., Clopath, C., & Stephenson-Jones, M. (2025). Dopaminergic action prediction errors serve as a value-free teaching signal. Nature, 643(8074), 1333–1342. https://doi.org/10.1038/s41586-025-09008-9

Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. https://doi.org/10.1126/science.275.5306.1593

Vieira de Castro, A. C., Fuchs, D., Morello, G. M., Pastur, S., de Sousa, L., & Olsson, I. A. S. (2020). Does training method matter? Evidence for the negative impact of aversive-based methods on companion dog welfare. PLoS ONE, 15(12), e0225023. https://doi.org/10.1371/journal.pone.0225023