A subtle computational glitch deep in the brain’s reward circuitry may hold the key to one of psychiatry’s most stubborn mysteries: why millions of people with depression lose the capacity to feel pleasure, and why no drug has ever been approved to restore it. After two decades of meticulous cross-species research, scientists are now pointing to an objective, mathematically rigorous test—the Probabilistic Reward Task—as a translational Rosetta Stone that could finally unlock treatments for anhedonia. A comprehensive review published in Nature Mental Health weaves together 20 years of findings and argues that this single task, usable in near-identical form from mice to humans, provides the most robust biomarker platform yet for reward learning deficits that cut across neuropsychiatric disorders.
Anhedonia, the reduced reactivity to pleasurable stimuli, is a core symptom of major depressive disorder and predicts some of the worst outcomes: poor response to antidepressants and psychotherapy, a more chronic disease course, severe psychosocial impairment, and elevated suicide risk. Despite a torrent of research, no approved pharmacological or behavioral intervention directly targets anhedonia. The impasse stems, in large part, from two persistent failures. First, animal models and human studies have relied on vastly different ways of measuring reward sensitivity—a rodent pressing a lever for sugar water tells a very different story than a patient filling out a questionnaire about how much they enjoyed their morning coffee. Second, clinical scales in humans tend to collapse distinct subdomains of reward processing—wanting, liking, learning—into a single coarse score, obscuring the very processes that go awry in disease.
In 2005, clinical neuroscientist Diego Pizzagalli and colleagues published the first demonstration of the Probabilistic Reward Task in individuals with elevated depressive symptoms, planting a flag for a new approach. Rooted in signal-detection theory, the task strips reward learning down to a clean computational signal. Participants view a series of briefly presented visual stimuli—typically a cartoon face with either a short or long mouth—and must indicate which stimulus appeared. Unbeknownst to them, one of the stimuli is designated the “rich” target and is followed by a monetary or social reward far more often than the other. Healthy volunteers quickly develop an implicit response bias toward the richly rewarded stimulus; they start to preferentially identify ambiguous or degraded versions of that stimulus as having been the rich one. This shift, quantified by a signal-detection metric known as response bias, indexes the individual’s propensity to modulate behavior as a function of reward history, independent of their raw perceptual acuity.
The genius of the paradigm lies in its cross-species isomorphism. Functionally identical versions of the PRT have been built for mice, rats, nonhuman primates, and humans. In every species, the stimuli are perceptually symmetric, reward contingencies are probabilistic, and data are fed into exactly the same signal-detection equations and computational reinforcement-learning models. A mouse navigating a touchscreen to distinguish between two grating patterns with asymmetric water rewards yields the same response-bias parameter as a human volunteer in a dimly lit testing booth. This analytic unity erases the translational chasm that has plagued psychiatric neuroscience, allowing researchers to directly compare neural circuit manipulations, pharmacological probes, and genetic risk factors across the evolutionary tree.
The new review organizes the ensuing deluge of data according to multiple dimensions of validity. Construct validity is high: the PRT reliably captures the latent process of reward learning, and blunted response bias is consistently linked to anhedonic symptoms rather than general distress. Clinical or diagnostic validity emerges from studies showing that depressed samples, particularly those with pronounced anhedonia, exhibit a sluggish or absent reward-induced bias compared with healthy controls. Prognostically, a muted response bias at baseline forecasts poor response to both selective serotonin reuptake inhibitors and behavioral activation therapies, while predictive validity points toward its ability to stratify individuals who will benefit from dopaminergic or glutamatergic interventions. Strikingly, susceptibility validity is also on the table: several longitudinal studies suggest that a low response bias in never-depressed adolescents or young adults predates first-onset depressive episodes, marking it as a potential premorbid vulnerability indicator.
On the biological front, the PRT’s reward-learning signal has been tethered to concrete neural and molecular substrates. Neuroimaging studies consistently implicate the ventral striatum, medial prefrontal cortex, and midbrain dopamine regions, while pharmacological challenges show that the response bias is boosted by dopamine agonists and blunted by dopamine blockers or chronic mild stress in animals. Genome-wide association and candidate gene studies point toward polygenic risk scores for depression and specific variants in the dopamine D2 receptor gene, among others. This convergence elevates the response-bias metric from a mere behavioral score to a bona fide biomarker—a quantifiable process-level indicator with known biological anchors. External validity studies further demonstrate that PRT performance correlates with real-world motivated behavior, such as effort expenditure for rewards and daily-life ecological momentary assessments of pleasure, closing the loop between lab and life.
Psychometrically, the task is well characterized. It shows acceptable to good test-retest reliability over weeks to months, making it suitable for tracking change in clinical trials. Its context of use has been specified for early-phase drug development, where it can serve as a phenotypic screen for anhedonia-targeting compounds, as well as for experimental medicine studies probing the functional integrity of the brain’s reward system. Because the task relies on implicit bias rather than introspection, it bypasses the cognitive biases and demand characteristics that corrupt self-report scales, giving pharmaceutical and academic researchers a hard outcome measure that translates directly across species.
The review’s synthesis of two decades of PRT research arrives at a pivotal moment. With the global burden of anhedonia mounting and precision psychiatry still in its infancy, the task offers a standardized assay to chart the reward-learning circuitry that goes silent in depression, schizophrenia, and other conditions. Large-scale consortia are now combining the PRT with high-field neuroimaging, dense phenotyping, and drug repurposing screens. The next chapter will likely see the response bias deployed as a primary endpoint in registration trials for novel anhedonia therapies—a prospect that felt like science fiction when the first small study appeared twenty years ago. If the task can deliver on its promise of bridging the translational gap, clinicians may finally have a tool not just to diagnose anhedonia, but to measure its retreat under treatment with the same precision a cardiologist tracks cholesterol.
Subject of Research: Reward learning deficits across species using the Probabilistic Reward Task
Article Title: Probing biomarkers and clinical utility of reward learning across species using the Probabilistic Reward Task: 20 years of findings
Article References: Pizzagalli, D.A. Nat. Mental Health 4, 1066–1087 (2026). https://doi.org/10.1038/s44220-026-00631-7
Image Credits: AI Generated
DOI: 10.1038/s44220-026-00631-7
Keywords: anhedonia, reward learning, Probabilistic Reward Task, depression, signal-detection theory, cross-species translation, biomarker, translational neuroscience
