The central obstacle the researchers faced was a confound familiar to anyone who studies reward in animals. When a mouse drinks water or eats food, its cerebellum is obviously active, but that activity could simply reflect the exquisite motor coordination of jaw, tongue, and forelimb movements rather than any representation of reward value. Disentangling the two requires an experimental design in which reward arrives without any consummatory behavior. The team’s solution was elegant: they trained head-fixed mice to push a robotic manipulandum, and each successful push delivered a delayed dose of dopamine directly into the brain, either through optogenetic activation of dopamine neurons in the ventral tegmental area or through electrical stimulation of the medial forebrain bundle, a classic self-stimulation pathway known to powerfully reinforce behavior. In this push-for-dopamine task, reward is stripped of its natural consummatory movements, leaving any cerebellar reward signals nowhere to hide.

While mice performed the task, the researchers used two-photon calcium imaging to watch two fundamentally different input streams of the cerebellar cortex simultaneously. The first stream consisted of cerebellar granule cells, the tiny, extraordinarily numerous neurons that receive mossy fiber inputs and form the parallel fiber system that broadcasts information across the cerebellum. The second stream was the climbing fibers, the powerful axons originating in the inferior olive that wrap around Purkinje cells and have long been associated with teaching signals in motor learning. By imaging granule cells with GCaMP indicators and climbing fiber terminals with the red calcium sensor RCaMP2, the team could record both channels of cerebellar input during the same behavioral sessions, capturing how each population responded as mice anticipated, earned, and consumed dopamine rewards.

The granule cell results were striking. Many individual granule cells encoded upcoming dopamine rewards predictively, ramping up sustained activity during the one-second delay between the completed push and reward delivery, and then terminating that activity abruptly the moment the reward arrived. This was not a brief burst locked to the action; it was a slowly building expectation signal that stretched or compressed to match the timing of the reward itself. When the researchers trained mice with two-second delays instead of one, the granule cell activations stretched correspondingly longer, and single-trial analyses confirmed that individual neural responses genuinely temporally scaled rather than simply broadening through trial averaging. When rewards were occasionally omitted, the anticipatory ramp persisted and failed to quench, exactly what one would expect from a signal tracking expected reward rather than the physical act of reaching.

Crucially, the researchers ran a battery of controls to rule out the possibility that these ramping signals were secretly encoding movement. DeepLabCut-based tracking of jaw, forepaw, nose, and hindpaw kinematics showed that the neural expectation dynamics diverged sharply from movement profiles: in some analyses, neural expectation peaked precisely when physical movement was minimal, and variance-partitioning models demonstrated that for most reward-anticipating granule cells, reward regressors explained neural activity far better than concurrent body kinematics did. The team even showed that anticipatory ramping appeared in a purely passive paradigm, in which an auditory cue predicted dopamine delivery with no instrumental action required at all. The expectation timer, in other words, does not depend on prior motor execution.

Perhaps most surprising was how the dopamine signal compared with a natural reward. In mice trained on both the push-for-dopamine task and an analogous push-for-water task, the strength of granule cell reward encoding for artificial dopamine stimulation matched or exceeded the encoding for water. Individual granule cells frequently generalized across reward types, and separate populations of cells generalized across delay durations, suggesting a common internal representation of anticipated reward value and timing that transcends the specific sensory identity of the reinforcer. Prior water training was not necessary for dopamine reward prediction signals to emerge, and animals that failed to reduce their orofacial movements upon switching from water to dopamine did not disproportionately drive the population-level anticipation effect, further dissociating reward expectation from consummatory habit.

The climbing fibers told a different and complementary story. Whereas granule cells predicted rewards, the majority of climbing fibers spiked just after dopamine delivery, firing robustly within a short latency of reward arrival on the very first day of training in naive animals. Many of the same climbing fibers also responded after water rewards, and responses to dopamine were typically stronger than responses to water in cells responsive to both. The researchers propose that this post-reward climbing fiber activity functions as an instructive, teaching-like signal, the cerebellar analogue of the dopamine system’s own reward delivery response. In classical cerebellar theory, climbing fiber activity serves as an error signal that drives synaptic plasticity at parallel fiber-Purkinje cell synapses; the new data suggest that reward arrival itself can constitute such a signal, allowing the cerebellum to strengthen the associations between actions, temporal expectations, and rewarding outcomes.

Both codes proved causally important, not merely correlational. When the researchers chronically inhibited granule cell activity with the optogenetic chloride pump stGtACR1 throughout the delay period and across multiple training days, mice learning the push-for-dopamine task were significantly impaired: they improved less in successful pushes per minute and in the percentage of completed reaches compared with normally trained controls. Importantly, the deficit persisted into a subsequent laser-off washout session, confirming that chronic granule cell inhibition disrupted the actual acquisition of the task rather than merely producing an acute motor impairment during stimulation. On the instructive side, the team asked whether climbing fiber activation alone could serve as a reward. In a remarkable demonstration, naive mice learned to push at moderate rates for delayed optogenetic activation of their own climbing fibers, receiving no dopamine and no natural reward whatsoever. Climbing fiber self-stimulation was less powerful than true dopamine or water rewards, falling into a rough motivational hierarchy of climbing fiber below ventral tegmental area stimulation below medial forebrain bundle and water, yet it was sufficient to support operant learning, accompanied by the same predictive granule cell ramping observed for genuine rewards.

Together, these results sketch a cerebellum that is deeply embedded in the machinery of motivation. Granule cells provide a predictive code, a running estimate of when and whether reward will arrive that scales with delay duration and reward type, while climbing fibers deliver an instructive code that marks rewarding outcomes and can itself reinforce actions. Because the cerebellum maintains well-characterized reciprocal connections with the basal ganglia, the thalamus, and reward-related cortical regions, these signals are positioned to interact with midbrain dopamine circuitry rather than operate in parallel to it. The authors suggest that cerebellar reward encoding may contribute to how animals learn the timing and value of their actions, integrating the prediction of future reward with the motor programs needed to obtain it.

The broader implications are considerable. Conditions ranging from Parkinson’s disease and addiction to autism and ataxia have all been linked, in various ways, to disrupted reward processing or cerebellar dysfunction, and several psychiatric and neurological disorders increasingly show cerebellar signatures in neuroimaging studies of motivation and cognition. If cerebellar granule cells and climbing fibers genuinely encode and instruct reward, then models of reinforcement learning that treat dopamine as the sole teaching signal may need revision, and the cerebellum may emerge as a target for therapeutic strategies aimed at restoring motivated behavior. At minimum, the study delivers a vivid demonstration that a structure long confined to the motor periphery of neuroscience is, in fact, keeping its own account of the brain’s most valuable currency, and using that account to help drive the actions that earn it.

Subject of Research: Cerebellar neural encoding of dopamine reward prediction and reinforcement in motivated behavior

Article Title: Predictive and instructive cerebellar encoding of dopamine reward drives motivated behavior

Article References: Predictive and instructive cerebellar encoding of dopamine reward drives motivated behavior. (n.d.). https://doi.org/10.1038/s41593-026-02449-z

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02449-z

Keywords: cerebellum, dopamine, reward prediction, granule cells, climbing fibers, motivated behavior, two-photon imaging, optogenetics, intracranial self-stimulation, reinforcement learning, neural circuits, Nature Neuroscience

The significance of self-administration models in addiction neuroscience is difficult to overstate. Unlike passive drug exposure experiments, in which the investigator simply injects an animal with a drug and observes the consequences, self-administration paradigms allow the animal itself to control when and how much drug it receives, typically by pressing a lever or poking its nose into a port. This voluntary element matters enormously. It allows investigators to dissociate the behavioral and neural mechanisms that motivate drug seeking and taking from the pharmacological effects of drug exposure itself, a distinction that lies at the heart of modern addiction research. Compulsive drug taking, relapse after abstinence, and the escalation of intake over time are all phenomena that only emerge when the animal has agency over its own drug consumption.

The intellectual lineage of this approach stretches back more than eight decades. As the authors note in their News and Views commentary accompanying the protocol, the earliest demonstrations that animals will work to obtain drugs appeared in the comparative psychology literature of 1940, when Spragg reported that chimpanzees would manipulate mechanisms to gain access to morphine. In the mid-1950s, Headlee, Coppock, and Hichols described intravenous morphine administration procedures in the Journal of Pharmaceutical Sciences, laying groundwork for controlled delivery of drugs into the bloodstream of laboratory animals. These early efforts culminated in a landmark 1962 study by James Weeks, published in Science, which established the operant intravenous drug self-administration procedure in rats—a technique that remains the workhorse of preclinical addiction research to this day.

The field built rapidly on that foundation. Thompson and Schuster demonstrated in 1964 that the principles of operant conditioning could be applied to morphine self-administration, and Risner and Jones extended the approach in 1975 to compare the reinforcing effects of different drugs. By 1991, Carney and colleagues were using self-administration procedures to study stimulant pharmacology, and the paradigm had become a standard screen for the abuse liability of novel compounds. A 2020 review by Venniro, Banks, Heilig, Epstein, and Shaham in Nature Reviews Neuroscience synthesized how decades of self-administration research had refined the field’s understanding of the neural circuits of drug reward, relapse, and the economic and social factors that influence drug taking. The historical arc is clear: each generation of methodological refinement has allowed sharper questions to be asked about why drugs are so compelling.

What the new protocol adds is the head-fixed dimension. In a conventional self-administration chamber, a mouse moves freely, presses levers, and receives intravenous infusions through a catheter implanted in its jugular vein. This arrangement is behaviorally rich but experimentally constrained. The animal’s head moves unpredictably, which degrades optical recordings, complicates the delivery of precisely timed sensory stimuli, and makes it nearly impossible to hold a microscope objective or a stimulus display in a fixed relationship to the animal’s eyes and whiskers. Head fixation solves these problems. With the skull rigidly stabilized, researchers can perform stable two-photon imaging of the same population of neurons across many sessions, deliver visual or auditory cues with millisecond precision, and apply optogenetic or pharmacological manipulations to genetically identified circuits while the animal is actively working for drug.

Combining head fixation with intravenous self-administration, however, is far from trivial, and this is precisely where the protocol makes its contribution. The authors provide step-by-step instructions for assembling the necessary hardware, integrating the head-fixation apparatus with the operant response devices and the infusion system that delivers drug through the implanted catheter. The technical challenges are considerable: the catheter must remain patent across repeated sessions while the animal is restrained; the operant manipulandum must be positioned so that a head-fixed mouse can respond comfortably; and the infusion line must be routed so that drug delivery is synchronized with the animal’s responses without introducing movement artifacts or leaks. By documenting these details, the protocol lowers the barrier for laboratories that want to adopt the technique without months of trial and error.

The timing of this methodological advance reflects a broader convergence in the field. A companion protocol by Doncheck and colleagues, also published in Nature Protocols in 2026 and cited in the commentary, indicates that head-fixed drug self-administration is emerging as a coherent methodological platform rather than the idiosyncratic setup of a single laboratory. Meanwhile, recent work published in Neuron in 2024 by Paniccia and colleagues demonstrated the scientific payoff of head-fixed approaches, and a 2021 study by Vollmer and colleagues in Frontiers in Behavioral Neuroscience explored related territory in freely moving animals. Together, these publications sketch a research landscape in which the voluntary taking of intravenous drugs can finally be observed with the full arsenal of modern circuit neuroscience.

For researchers weighing whether to adopt the technique, the trade-offs are worth understanding. Freely moving self-administration preserves the naturalistic ethology of drug seeking—the animal navigates its environment, approaches the drug source, and integrates spatial and contextual cues into its behavior. Head-fixed preparations sacrifice some of that naturalism in exchange for experimental control and optical stability. The authors’ framing suggests that the two approaches should be viewed as complementary rather than competing: freely moving studies establish the validity of a behavioral phenomenon, while head-fixed preparations dissect its circuit-level mechanisms with cellular precision. A protocol that makes the head-fixed variant accessible and reproducible therefore expands the toolkit available to the field rather than replacing the established one.

The potential applications span the major questions of addiction neuroscience. With a head-fixed mouse pressing a port for intravenous cocaine, opioid, or other drug infusions, investigators could track how neurons in the prefrontal cortex, nucleus accumbens, dorsal striatum, or amygdala change their activity as drug seeking becomes habitual, as intake escalates, or as cues that predict drug availability come to drive behavior. They could test whether manipulating specific projections alters the propensity to seek drug, and they could do so across many consecutive sessions with imaging planes held perfectly stable. Because the animals are mice rather than rats, the full power of mouse genetics—cell-type-specific reporters, conditional knockouts, and activity-dependent labeling—becomes available to self-administration research in ways that were previously cumbersome.

As with any methodological innovation, the ultimate test will be how widely the technique is adopted and what it reveals. The authors declare no competing interests, and the protocol, published as a U.S. Government work, is accompanied by supplementary information including an additional figure and references to support implementation. For a field that has spent more than sixty years refining the operant self-administration paradigm since Weeks first described it in Science, the arrival of a validated head-fixed variant marks a genuine expansion of the frontier. If it delivers on its promise, the coming years should see drug self-administration experiments in which every lever press is paired with a window into the living brain—bringing addiction neuroscience closer to the mechanistic depth that the problem of addiction demands.

Subject of Research: A protocol for performing intravenous drug self-administration in head-fixed mice to study the neural mechanisms of addiction.

Article Title: Head-fixed intravenous drug self-administration: a new frontier in addiction neuroscience

Article References: Negishi, K., D’Ottavio, G., & Shaham, Y. (2026). Head-fixed intravenous drug self-administration: a new frontier in addiction neuroscience. Nature Protocols. https://doi.org/10.1038/s41596-026-01438-7

Image Credits: AI Generated

DOI: 10.1038/s41596-026-01438-7

Keywords: addiction neuroscience, drug self-administration, head-fixed mice, intravenous drug delivery, Nature Protocols, behavioral neuroscience, two-photon imaging, operant conditioning, drug reward, relapse, NIDA, mouse models