Search

Frontal area MOs (secondary motor area) is a key brain structure in rodents for making decisions based on sensory evidence and on reward value. In behavioral tasks, its neurons can encode sensory stimuli, upcoming choices, expected rewards, ongoing actions, and recent outcomes. However, the information encoded, and the nature of the resulting code, may depend on the task being performed. We recorded MOs population activity using two-photon calcium imaging, in a task requiring mice to integrate sensory evidence with reward value. Mice turned a wheel to report the location of a visual stimulus following a delay period, to receive a reward whose size varied over trial blocks. MOs neurons encoded multiple task variables, but not all of those seen in other tasks. In the delay period, the MOs population strongly encoded the stimulus side but did not significantly encode the reward-size block. A correlation of MOs activity with upcoming choice could be explained by a common effect of stimulus on those two correlates. After the wheel turn and the feedback, the MOs population encoded choice side and choice outcome jointly and nonlinearly according to an exclusive-or (XOR) operation. This nonlinear operation would allow a downstream linear decoder to infer the correct choice side (i.e., the side that would have been rewarded) even on zero contrast trials, when there had been no visible stimulus. These results indicate that MOs neurons flexibly encode some but not all variables that determine behavior, depending on task. Moreover, they reveal that MOs activity can reflect a nonlinear combination of these behavioral variables, allowing simple linear inference of task events that would not have been directly observable.

The striatum plays critical roles in visually-guided decision-making and receives dense axonal projections from midbrain dopamine neurons. However, the roles of striatal dopamine in visual decision-making are poorly understood. We trained male and female mice to perform a visual decision task with asymmetric reward payoff, and we recorded the activity of dopamine axons innervating striatum. Dopamine axons in the dorsomedial striatum (DMS) responded to contralateral visual stimuli and contralateral rewarded actions. Neural responses to contralateral stimuli could not be explained by orienting behavior such as eye movements. Moreover, these contralateral stimulus responses persisted in sessions where the animals were instructed to not move to obtain reward, further indicating that these signals are stimulus-related. Lastly, we show that DMS dopamine signals were qualitatively different from dopamine signals in the ventral striatum (VS), which responded to both ipsilateral and contralateral stimuli, conforming to canonical prediction error signaling under sensory uncertainty. Thus, during visual decisions, DMS dopamine encodes visual stimuli and rewarded actions in a lateralized fashion, and could facilitate associations between specific visual stimuli and actions. SIGNIFICANCE STATEMENT While the striatum is central to goal-directed behavior, the precise roles of its rich dopaminergic innervation in perceptual decision-making are poorly understood. We found that in a visual decision task, dopamine axons in the dorsomedial striatum (DMS) signaled stimuli presented contralaterally to the recorded hemisphere, as well as the onset of rewarded actions. Stimulus-evoked signals persisted in a no-movement task variant. We distinguish the patterns of these signals from those in the ventral striatum (VS). Our results contribute to the characterization of region-specific dopaminergic signaling in the striatum and highlight a role in stimulus-action association learning.

Transcriptomics has revealed the exquisite diversity of cortical inhibitory neurons1–7, but it is not known whether these fine molecular subtypes have correspondingly diverse activity patterns in the living brain. Here, we show that inhibitory subtypes in primary visual cortex (V1) have diverse correlates with brain state, but that this diversity is organized by a single factor: position along their main axis of transcriptomic variation. We combined in vivo 2-photon calcium imaging of mouse V1 with a novel transcriptomic method to identify mRNAs for 72 selected genes in ex vivo slices. We used transcriptomic clusters (t-types)4 to classify inhibitory neurons imaged in layers 1-3 using a three-level hierarchy of 5 Families, 11 Classes, and 35 t-types. Visual responses differed significantly only across Families, but modulation by brain state differed at all three hierarchical levels. Nevertheless, this diversity could be predicted from the first transcriptomic principal component, which predicted a cell type’s brain state modulation and correlations with simultaneously recorded cells. Inhibitory t-types with narrower spikes, lower input resistance, weaker adaptation, and less axon in layer 1 as determined in vitro8 fired more in resting, oscillatory brain states. Transcriptomic types with the opposite properties fired more during arousal. The former cells had more inhibitory cholinergic receptors, and the latter more excitatory receptors. Thus, despite the diversity of V1 inhibitory neurons, a simple principle determines how their joint activity shapes state-dependent cortical processing.

The use of head fixation in mice is increasingly common in research, its use having initially been restricted to the field of sensory neuroscience. Head restraint has often been combined with fluid control, rather than food restriction, to motivate behaviour, but this too is now in use for both restrained and non-restrained animals. Despite this, there is little guidance on how best to employ these techniques to optimise both scientific outcomes and animal welfare. This article summarises current practices and provides recommendations to improve animal wellbeing and data quality, based on a survey of the community, literature reviews, and the expert opinion and practical experience of an international working group convened by the UK's National Centre for the Replacement, Refinement and Reduction of Animals in Research (NC3Rs). Topics covered include head fixation surgery and post-operative care, habituation to restraint, and the use of fluid/food control to motivate performance. We also discuss some recent developments that may offer alternative ways to collect data from large numbers of behavioural trials without the need for restraint. The aim is to provide support for researchers at all levels, animal care staff, and ethics committees to refine procedures and practices in line with the refinement principle of the 3Rs.

SUMMARYThe medial prefrontal cortex (mPFC) is necessary for executing many learned associations between stimuli and movement. It is unclear, however, whether activity in the mPFC reflects sensory or motor aspects of sensorimotor associations and whether it evolves gradually during learning. To address these questions, we recorded cortical activity with widefield calcium imaging while mice learned a visuomotor task. The task involved associating a visual stimulus with a forelimb movement. After learning, the mPFC showed stimulus-evoked activity both during task performance and during passive viewing, when the stimulus evoked no action. This stimulus-evoked activity closely tracked behavioral performance across training, exhibiting jumps between training days. Electrophysiological recordings localized this activity to the secondary motor and anterior cingulate cortex. We conclude that learning a visuomotor task promotes a route for visual information to reach the prefrontal cortex, which develops responses to the relevant visual stimuli even outside the context of the task.

Correlates of sensory stimuli and motor actions are found in multiple cortical areas, but such correlates do not indicate whether these areas are causally relevant to task performance. We trained mice to discriminate visual contrast and report their decision by steering a wheel. Widefield calcium imaging and Neuropixels recordings in cortex revealed stimulus-related activity in visual (VIS) and frontal (MOs) areas, and widespread movement-related activity across the whole dorsal cortex. Optogenetic inactivation biased choices only when targeted at VIS and MOs,proportionally to each site's encoding of the visual stimulus, and at times corresponding to peak stimulus decoding. A neurometric model based on summing and subtracting activity in VIS and MOs successfully described behavioral performance and predicted the effect of optogenetic inactivation. Thus, sensory signals localized in visual and frontal cortex play a causal role in task performance, while widespread dorsal cortical signals correlating with movement reflect processes that do not play a causal role.

Sensory cortices can be affected by stimuli of multiple modalities and are thus increasingly thought to be multisensory. For instance, primary visual cortex (V1) is influenced not only by images but also by sounds. Here we show that the activity evoked by sounds in V1, measured with Neuropixels probes, is stereotyped across neurons and even across mice. It is independent of projections from auditory cortex and resembles activity evoked in the hippocampal formation, which receives little direct auditory input. Its low-dimensional nature starkly contrasts the high-dimensional code that V1 uses to represent images. Furthermore, this sound-evoked activity can be precisely predicted by small body movements that are elicited by each sound and are stereotyped across trials and mice. Thus, neural activity that is apparently multisensory may simply arise from low-dimensional signals associated with internal state and behavior.

The brain should be best able to associate distinct behavioral responses to sensory stimuli if these stimuli evoke population firing patterns that are close to orthogonal. To investigate whether task training orthogonalizes population codes in primary visual cortex (V1), we measured the orientation tuning of 4,000-neuron populations in mouse V1 before and after training on a visuomotor task. The effect of task training on population codes could be captured by a simple mathematical transformation of firing rates, which suppressed responses to motor-associated stimuli, but only in cells responding to them at intermediate levels. This transformation orthogonalized the representations of the task orientations by sparsening the population responses to these stimuli. The strength of response transformation varied from trial to trial, suggesting a dynamic circuit mechanism rather than static synaptic plasticity. These results indicate a simple process by which visuomotor associations orthogonalize population codes as early as in primary visual cortex.

To interpret the world and make accurate perceptual decisions, the brain must combine information across sensory modalities, For instance, it must combine vision and hearing to localize objects based on their image and sound. Probability theory suggests that evidence from multiple independent cues should be combined additively 1, but it is unclear whether mice and other mammals do this 2–6, and the cortical substrates of multisensory integration are uncertain 7. Here we show that to localize a stimulus mice combine auditory and visual spatial cues additively, a computation supported by unisensory processing in auditory and visual cortex and additive multisensory integration in frontal cortex. We developed an audiovisual localization task where mice turn a wheel to indicate the joint position of an image and a sound. Scanning optogenetic inactivation of dorsal cortex 8–11 showed that auditory and visual areas contribute unisensory information, whereas frontal cortex (secondary motor area, MOs) contributes multisensory information to the mouse’s decision, Neuropixels recordings of >10,000 neurons indicated that neural activity in MOs reflects an additive combination of visual and auditory signals. An accumulator model 1,12,13 applied to the sensory representations of MOs neurons reproduced behaviourally observed choices and reaction times. This suggests that MOs integrates information from multiple sensory cortices, providing a signal that is then transformed into a binary decision by a downstream accumulator.

Recording many neurons for a long time The ultimate aim of chronic recordings is to sample from the same neuron over days and weeks. However, this goal has been difficult to achieve for large populations of neurons. Steinmetz et al. describe the development and testing of Neuropixels 2.0. This new electrophysiological recording tool is a miniaturized, high-density probe for both acute and long-term experiments combined with sophisticated software algorithms for fully automatic post hoc computational stabilization. The technique also provides a strategy for extending the number of recorded sites beyond the number of available recording channels. In freely moving animals, extremely large numbers of individual neurons could thus be followed and tracked with the same probe for weeks and occasionally months. Science , this issue p. eabf4588

Functional ultrasound imaging (fUSI) is an appealing method for measuring blood flow and thus infer brain activity, but it relies on the physiology of neurovascular coupling and requires extensive signal processing. To establish to what degree fUSI trial-by-trial signals reflect neural activity, we performed simultaneous fUSI and neural recordings with Neuropixels probes in awake mice. fUSI signals strongly correlated with the slow (

During navigation, the visual responses of neurons in mouse primary visual cortex (V1) are modulated by the animal’s spatial position. Here we show that this spatial modulation is similarly present across multiple higher visual areas but negligible in the main thalamic pathway into V1. Similar to hippocampus, spatial modulation in visual cortex strengthens with experience and with active behavior. Active navigation in a familiar environment, therefore, enhances the spatial modulation of visual signals starting in the cortex.

SummaryParietal cortex is implicated in a variety of behavioral processes, but it is unknown whether and how individual neurons participate in multiple tasks. We trained head-fixed mice to perform two visual decision tasks involving a steering wheel or a virtual T-maze, and recorded from the same parietal neurons during the two. Neurons that were active during the T-maze task were typically inactive during the steering-wheel task, and vice versa. Recording from the same neurons in the same apparatus without task stimuli yielded the same specificity as in the task, suggesting that task specificity depends on physical context. To confirm this, we trained some mice in a third task combining the steering wheel with the visual environment of the T-maze. This hybrid task engaged the same neurons as the steering-wheel task. Thus, participation by neurons in mouse parietal cortex is task-specific, and this specificity is determined by physical context.

To study the dynamics of neural processing across timescales, we require the ability to follow the spiking of thousands of individually separable neurons over weeks and months, during unrestrained behavior. To address this need, we introduce the Neuropixels 2.0 probe together with novel analysis algorithms. The new probe has over 5,000 sites and is miniaturized such that two probes plus a headstage, recording 768 sites at once, weigh just over 1 g, suitable for implanting chronically in small mammals. Recordings with high quality signals persisting for at least two months were reliably obtained in two species and six different labs. Improved site density and arrangement combined with new data processing methods enable automatic post-hoc stabilization of data despite brain movements during behavior and across days, allowing recording from the same neurons in the mouse visual cortex for over 2 months. Additionally, an optional configuration allows for recording from multiple sites per available channel, with a penalty to signal-to-noise ratio. These probes and algorithms enable stable recordings from >10,000 sites during free behavior in small animals such as mice.

Summary The visual responses of neurons in the primary visual cortex (V1) are influenced by the animal’s position in the environment [1, 2, 3, 4, 5]. V1 responses encode positions that co-fluctuate with those encoded by place cells in hippocampal area CA1 [2, 5]. This correlation might reflect a common influence of non-visual spatial signals on both areas. Place cells in CA1, indeed, do not rely only on vision; their place preference depends on the physical distance traveled [6, 7, 8, 9, 10, 11] and on the phase of the 6–9 Hz theta oscillation [12, 13]. Are V1 responses similarly influenced by these non-visual factors? We recorded V1 and CA1 neurons simultaneously while mice performed a spatial task in a virtual corridor by running on a wheel and licking at a reward location. By changing the gain that couples the wheel movement to the virtual environment, we found that ∼20% of V1 neurons were influenced by the physical distance traveled, as were ∼40% of CA1 place cells. Moreover, the firing rate of ∼24% of V1 neurons was modulated by the phase of theta oscillations recorded in CA1 and the response profiles of ∼7% of V1 neurons shifted spatially across the theta cycle, analogous to the phase precession observed in ∼37% of CA1 place cells. The influence of theta oscillations on V1 responses was more prominent in putative layer 6. These results reveal that, in a familiar environment, sensory processing in V1 is modulated by the key non-visual signals that influence spatial coding in the hippocampus., Graphical Abstract, Highlights • We compared responses of mouse V1 and CA1 during navigation in virtual reality • Similar to CA1 neurons, V1 neurons are influenced by physical distance traveled • V1 responses are modulated by theta oscillations recorded in hippocampus • Positions encoded by V1 shift spatially with the phase of the theta oscillation, Fournier et al. compare responses of hippocampus and visual cortex in mice navigating a virtual reality. They find that, similar to hippocampus, activity in visual cortex is influenced by the physical distance traveled in an environment. In addition, theta oscillations modulate the activity of visual cortical neurons and their position coding.

Midbrain dopamine neurons play key roles in decision-making by regulating reward valuation and actions. These roles are thought to depend on dopamine neurons innervating striatum. In addition to actions and rewards, however, efficient decisions often involve consideration of uncertain sensory signals. The functions of striatal dopamine during sensory decisions remains unknown. We trained mice in a task that probed decisions based on sensory evidence and reward value, and recorded the activity of striatal dopamine axons. Dopamine axons in ventral striatum (VS) responded to bilateral stimuli and trial outcomes, encoding prediction errors that scaled with decision confidence and reward value. By contrast, dopamine axons in dorsal striatum (DS) responded to contralateral stimuli and contralateral actions. Thus, during sensory decisions, striatal dopamine signals are anatomically organized. VS dopamine resembles prediction errors suitable for reward maximization under sensory uncertainty whereas DS dopamine encodes specific combinations of stimuli and actions in a lateralized fashion.

Although cortical circuits are complex and interconnected with the rest of the brain, their macroscopic dynamics are often approximated by modeling the averaged activities of excitatory and inhibitory cortical neurons, without interactions with other brain circuits. To verify the validity of such mean-field models, we optogenetically stimulated populations of excitatory and parvalbumin-expressing inhibitory neurons in awake mouse visual cortex, while recording population activity in cortex and in its thalamic correspondent, the lateral geniculate nucleus. The cortical responses to brief test pulses could not be explained by a mean-field model including only cortical excitatory and inhibitory populations. However, these responses could be predicted by extending the model to include thalamic interactions that cause net cortical suppression following activation of cortical excitatory neurons. We conclude that mean-field models can accurately summarize cortical dynamics, but only when the cortex is considered as part of a dynamic corticothalamic network.

Progress in science requires standardized assays whose results can be readily shared, compared, and reproduced across laboratories. Reproducibility, however, has been a concern in neuroscience, particularly for measurements of mouse behavior. Here, we show that a standardized task to probe decision-making in mice produces reproducible results across multiple laboratories. We adopted a task for head-fixed mice that assays perceptual and value-based decision making, and we standardized training protocol and experimental hardware, software, and procedures. We trained 140 mice across seven laboratories in three countries, and we collected 5 million mouse choices into a publicly available database. Learning speed was variable across mice and laboratories, but once training was complete there were no significant differences in behavior across laboratories. Mice in different laboratories adopted similar reliance on visual stimuli, on past successes and failures, and on estimates of stimulus prior probability to guide their choices. These results reveal that a complex mouse behavior can be reproduced across multiple laboratories. They establish a standard for reproducible rodent behavior, and provide an unprecedented dataset and open-access tools to study decision-making in mice. More generally, they indicate a path toward achieving reproducibility in neuroscience through collaborative open-science approaches., eLife digest In science, it is of vital importance that multiple studies corroborate the same result. Researchers therefore need to know all the details of previous experiments in order to implement the procedures as exactly as possible. However, this is becoming a major problem in neuroscience, as animal studies of behavior have proven to be hard to reproduce, and most experiments are never replicated by other laboratories. Mice are increasingly being used to study the neural mechanisms of decision making, taking advantage of the genetic, imaging and physiological tools that are available for mouse brains. Yet, the lack of standardized behavioral assays is leading to inconsistent results between laboratories. This makes it challenging to carry out large-scale collaborations which have led to massive breakthroughs in other fields such as physics and genetics. To help make these studies more reproducible, the International Brain Laboratory (a collaborative research group) et al. developed a standardized approach for investigating decision making in mice that incorporates every step of the process; from the training protocol to the software used to analyze the data. In the experiment, mice were shown images with different contrast and had to indicate, using a steering wheel, whether it appeared on their right or left. The mice then received a drop of sugar water for every correction decision. When the image contrast was high, mice could rely on their vision. However, when the image contrast was very low or zero, they needed to consider the information of previous trials and choose the side that had recently appeared more frequently. This method was used to train 140 mice in seven laboratories from three different countries. The results showed that learning speed was different across mice and laboratories, but once training was complete the mice behaved consistently, relying on visual stimuli or experiences to guide their choices in a similar way. These results show that complex behaviors in mice can be reproduced across multiple laboratories, providing an unprecedented dataset and open-access tools for studying decision making. This work could serve as a foundation for other groups, paving the way to a more collaborative approach in the field of neuroscience that could help to tackle complex research challenges.

A major role of vision is to guide navigation, and navigation is strongly driven by vision1–4. Indeed, the brain’s visual and navigational systems are known to interact5,6, and signals related to position in the environment have been suggested to appear as early as in the visual cortex6,7. Here, to establish the nature of these signals, we recorded in the primary visual cortex (V1) and hippocampal area CA1 while mice traversed a corridor in virtual reality. The corridor contained identical visual landmarks in two positions, so that a purely visual neuron would respond similarly at those positions. Most V1 neurons, however, responded solely or more strongly to the landmarks in one position rather than the other. This modulation of visual responses by spatial location was not explained by factors such as running speed. To assess whether the modulation is related to navigational signals and to the animal’s subjective estimate of position, we trained the mice to lick for a water reward upon reaching a reward zone in the corridor. Neuronal populations in both CA1 and V1 encoded the animal’s position along the corridor, and the errors in their representations were correlated. Moreover, both representations reflected the animal’s subjective estimate of position, inferred from the animal’s licks, better than its actual position. When animals licked in a given location—whether correctly or incorrectly—neural populations in both V1 and CA1 placed the animal in the reward zone. We conclude that visual responses in V1 are controlled by navigational signals, which are coherent with those encoded in hippocampus and reflect the animal’s subjective position. The presence of such navigational signals as early as a primary sensory area suggests that they permeate sensory processing in the cortex. When running through a virtual reality corridor, a mouse’s position is represented in both the hippocampus (as expected) and the primary visual cortex, for places that are visually identical.

The cortex projects to the dorsal striatum topographically1,2 to regulate behaviour3-5, but spiking activity in the two structures has previously been reported to have markedly different relations to sensorimotor events6-9. Here we show that the relationship between activity in the cortex and striatum is spatiotemporally precise, topographic, causal and invariant to behaviour. We simultaneously recorded activity across large regions of the cortex and across the width of the dorsal striatum in mice that performed a visually guided task. Striatal activity followed a mediolateral gradient in which behavioural correlates progressed from visual cue to response movement to reward licking. The summed activity in each part of the striatum closely and specifically mirrored activity in topographically associated cortical regions, regardless of task engagement. This relationship held for medium spiny neurons and fast-spiking interneurons, whereas the activity of tonically active neurons differed from cortical activity with stereotypical responses to sensory or reward events. Inactivation of the visual cortex abolished striatal responses to visual stimuli, supporting a causal role of cortical inputs in driving the striatum. Striatal visual responses were larger in trained mice than untrained mice, with no corresponding change in overall activity in the visual cortex. Striatal activity therefore reflects a consistent, causal and scalable topographical mapping of cortical activity.

Focal epilepsy involves excessive cortical activity that propagates both locally and distally. Does this propagation follow the same routes as normal cortical activity? We pharmacologically induced focal seizures in primary visual cortex (V1) of awake mice, and compared their propagation to the retinotopic organization of V1 and higher visual areas. We used simultaneous local field potential recordings and widefield imaging of a genetically encoded calcium indicator to measure prolonged seizures (ictal events) and brief interictal events. Both types of event are orders of magnitude larger than normal visual responses, and both start as standing waves: synchronous elevated activity in the V1 focus and in homotopic locations in higher areas, i.e. locations with matching retinotopic preference. Following this common beginning, however, seizures persist and propagate both locally and into homotopic distal regions, and eventually invade all of visual cortex and beyond. We conclude that seizure initiation resembles the initiation of interictal events, and seizure propagation respects the connectivity underlying normal visual processing.Focal cortical seizures result from local and widespread propagation of excitatory activity. Here the authors employ widefield calcium imaging in mouse visual areas to demonstrate that these seizures start as local synchronous activation and then propagate along the connectivity that underlies normal sensory processing.

During navigation, the visual responses of neurons in primary visual cortex (V1) are modulated by the animal’s spatial position. Here we show that this spatial modulation is similarly present across multiple higher visual areas but largely absent in the main thalamic pathway into V1. Similar to hippocampus, spatial modulation in visual cortex strengthens with experience and requires engagement in active behavior. Active navigation in a familiar environment, therefore, determines spatial modulation of visual signals starting in the cortex.

Setting up an experiment in behavioral neuroscience is a complex process that is often managed with ad hoc solutions. To streamline this process, we developed Rigbox, a high-performance, open-source software toolbox that facilitates a modular approach to designing experiments (https://github.com/cortex-lab/Rigbox). Rigbox simplifies hardware input-output, time aligns datastreams from multiple sources, communicates with remote databases, and implements visual and auditory stimuli presentation. Its main submodule, Signals, allows intuitive programming of behavioral tasks. Here we illustrate its function with the following two interactive examples: a human psychophysics experiment, and the game of Pong. We give an overview of running experiments in Rigbox, provide benchmarks, and conclude with a discussion on the extensibility of the software and comparisons with similar toolboxes. Rigbox runs in MATLAB, with Java components to handle network communication, and a C library to boost performance.

The dorsal striatum is organized into domains that drive characteristic behaviors1–7, and receive inputs from different parts of the cortex8,9 which modulate similar behaviors10–12. Striatal responses to cortical inputs, however, can be affected by changes in connection strength13–15, local striatal circuitry16,17, and thalamic inputs18,19. Therefore, it is unclear whether the pattern of activity across striatal domains mirrors that across the cortex20–23 or differs from it24–28. Here we use simultaneous large-scale recordings in the cortex and the striatum to show that striatal activity can be accurately predicted by spatiotemporal activity patterns in the cortex. The relationship between activity in the cortex and the striatum was spatially consistent with corticostriatal anatomy, and temporally consistent with a feedforward drive. Each striatal domain exhibited specific sensorimotor responses that predictably followed activity in the associated cortical regions, and the corticostriatal relationship remained unvaried during passive states or performance of a task probing visually guided behavior. However, the task’s visual stimuli and corresponding behavioral responses evoked relatively more activity in the striatum than in associated cortical regions. This increased striatal activity involved an additive offset in firing rate, which was independent of task engagement but only present in animals that had learned the task. Thus, striatal activity largely reflects patterns of cortical activity, deviating from them in a simple additive fashion for learned stimuli or actions.

What are neuroscientists actually interested in? To find out, I studied their itineraries at the annual meeting of the Society for Neuroscience. I obtained a chart where topics that were commonly in the same itineraries form clusters. The empty spaces between these clusters might constitute opportunities for future efforts.

The operating mode of the visual system depends on behavioural states such as arousal1,2. This dependence is seen both in primary visual cortex3–7 (V1) and in subcortical brain structures receiving direct retinal input4,8. Here we show that this effect arises as early as in the output of the retina. We first measured activity in a region that receives retinal projections9, the superficial superior colliculus (sSC), and found that this activity strongly depended on behavioural state. This modulation was not mediated by feedback inputs from V1 as it was immune to V1 inactivation. We then used Neuropixels probes10 to record activity in the optic tract, and we found some retinal axons whose activity significantly varied with arousal, even in darkness. To characterize these effects on a larger sample of retinal outputs, we imaged the activity of retinal boutons11,12 in sSC during behaviour using a calcium indicator. The activity of these boutons correlated with arousal as strongly as that of sSC neurons, and this correlation persisted also during darkness. These results reveal a novel property of retinal function in mice, which could be observed only during behaviour: retinal outputs are modulated by behavioural state before they reach the rest of the brain.

Deciding between stimuli requires combining their learned value with one's sensory confidence. We trained mice in a visual task that probes this combination. Mouse choices reflected not only present confidence and past rewards but also past confidence. Their behavior conformed to a model that combines signal detection with reinforcement learning. In the model, the predicted value of the chosen option is the product of sensory confidence and learned value. We found precise correlates of this variable in the pre-outcome activity of midbrain dopamine neurons and of medial prefrontal cortical neurons. However, only the latter played a causal role: inactivating medial prefrontal cortex before outcome strengthened learning from the outcome. Dopamine neurons played a causal role only after outcome, when they encoded reward prediction errors graded by confidence, influencing subsequent choices. These results reveal neural signals that combine reward value with sensory confidence and guide subsequent learning.