Comparing the representation of a simple visual stimulus across the cerebellar network.

The cerebellum is a conserved structure of the vertebrate brain involved in the timing and calibration of movements. Its function is supported by the convergence of fibers from granule cells (GCs) and inferior olive neurons (IONs) onto Purkinje cells (PCs). Theories of cerebellar function postulate that IONs convey error signals to PCs that, paired with the contextual information provided by GCs, can instruct motor learning.Here, we use the larval zebrafish to investigate (i) how sensory representations of the same stimulus vary across GCs and IONs and (ii) how PC activity reflects these two different input streams. We use population calcium imaging to measure IONs and GCs responses to flashes of diverse luminance and duration. First, we observe that GCs show tonic and graded responses, as opposed to IONs, whose activity peaks mostly at luminance transitions, consistently with the notion that GCs and IONs encode context and error information, respectively. Secondly, we show that GC activity is patterned over time: some neurons exhibit sustained responses for the entire duration of the stimulus, while in others activity ramps up with slow time constants. This activity could provide a substrate for time representation in the cerebellum. Together, our observations give support to the notion of an error signal coming from IONs and provide the first experimental evidence for a temporal patterning of GC activity over many seconds.Significance statement The cerebellum is an important brain structure shared by all vertebrates, playing a crucial role in sensorimotor behavior. By using calcium imaging in larval zebrafish, we investigated sensory responses across granule cells (GCs), Purkinje cells. (PCs), and inferior olive neurons (IONs). We describe how GCs show tonic and graded responses, as opposed to IONs, more active at stimulus transition times, confirming the notion that GCs and IONs encode context and error information, respectively. We also show how time could be represented in the cerebellum by patterned activity in GCs. Together, our observations support to the notion of an error signal coming from IONs and provide the first experimental evidence for a temporal patterning of GC activity over many seconds.

Neural dynamics and architecture of the heading direction circuit in zebrafish.

Animals generate neural representations of their heading direction. Notably, in insects, heading direction is topographically represented by the activity of neurons in the central complex. Although head direction cells have been found in vertebrates, the connectivity that endows them with their properties is unknown. Using volumetric lightsheet imaging, we find a topographical representation of heading direction in a neuronal network in the zebrafish anterior hindbrain, where a sinusoidal bump of activity rotates following directional swims of the fish and is otherwise stable over many seconds. Electron microscopy reconstructions show that, although the cell bodies are located in a dorsal region, these neurons arborize in the interpeduncular nucleus, where reciprocal inhibitory connectivity stabilizes the ring attractor network that encodes heading. These neurons resemble those found in the fly central complex, showing that similar circuit architecture principles may underlie the representation of heading direction across the animal kingdom and paving the way to an unprecedented mechanistic understanding of these networks in vertebrates.

Do Neural Networks for Segmentation Understand Insideness?

The insideness problem is an aspect of image segmentation that consists of determining which pixels are inside and outside a region. Deep neural networks (DNNs) excel in segmentation benchmarks, but it is unclear if they have the ability to solve the insideness problem as it requires evaluating long-range spatial dependencies. In this letter, we analyze the insideness problem in isolation, without texture or semantic cues, such that other aspects of segmentation do not interfere in the analysis. We demonstrate that DNNs for segmentation with few units have sufficient complexity to solve the insideness for any curve. Yet such DNNs have severe problems with learning general solutions. Only recurrent networks trained with small images learn solutions that generalize well to almost any curve. Recurrent networks can decompose the evaluation of long-range dependencies into a sequence of local operations, and learning with small images alleviates the common difficulties of training recurrent networks with a large number of unrolling steps.

Valence and State-Dependent Population Coding in Dopaminergic Neurons in the Fly Mushroom Body.

Neuromodulation permits flexibility of synapses, neural circuits, and ultimately behavior. One neuromodulator, dopamine, has been studied extensively in its role as a reward signal during learning and memory across animal species. Newer evidence suggests that dopaminergic neurons (DANs) can modulate sensory perception acutely, thereby allowing an animal to adapt its behavior and decision making to its internal and behavioral state. In addition, some data indicate that DANs are not homogeneous but rather convey different types of information as a heterogeneous population. We have investigated DAN population activity and how it could encode relevant information about sensory stimuli and state by taking advantage of the confined anatomy of DANs innervating the mushroom body (MB) of the fly Drosophila melanogaster. Using in vivo calcium imaging and a custom 3D image registration method, we found that the activity of the population of MB DANs encodes innate valence information of an odor or taste as well as the physiological state of the animal. Furthermore, DAN population activity is strongly correlated with movement, consistent with a role of dopamine in conveying behavioral state to the MB. Altogether, our data and analysis suggest that DAN population activities encode innate odor and taste valence, movement, and physiological state in a MB-compartment-specific manner. We propose that dopamine shapes innate perception through combinatorial population coding of sensory valence, physiological, and behavioral context.

Evidence accumulation during a sensorimotor decision task revealed by whole-brain imaging.

Although animals can accumulate sensory evidence over considerable time scales to appropriately select behavior, little is known about how the vertebrate brain as a whole accomplishes this. In this study, we developed a new sensorimotor decision-making assay in larval zebrafish based on whole-field visual motion. Fish responded by swimming in the direction of perceived motion, such that the latency to initiate swimming and the fraction of correct turns were modulated by motion strength. Using whole-brain functional imaging, we identified neural activity relevant to different stages of the decision-making process, including the momentary evaluation and accumulation of sensory evidence. This activity is distributed in functional clusters across different brain regions and is characterized by a wide range of time constants. In addition, we found that the caudal interpeduncular nucleus (IPN), a circular structure located ventrally on the midline of the brain, reliably encodes the left and right turning rates.

Stytra: An open-source, integrated system for stimulation, tracking and closed-loop behavioral experiments.

We present Stytra, a flexible, open-source software package, written in Python and designed to cover all the general requirements involved in larval zebrafish behavioral experiments. It provides timed stimulus presentation, interfacing with external devices and simultaneous real-time tracking of behavioral parameters such as position, orientation, tail and eye motion in both freely-swimming and head-restrained preparations. Stytra logs all recorded quantities, metadata, and code version in standardized formats to allow full provenance tracking, from data acquisition through analysis to publication. The package is modular and expandable for different experimental protocols and setups. Current releases can be found at https://github.com/portugueslab/stytra. We also provide complete documentation with examples for extending the package to new stimuli and hardware, as well as a schema and parts list for behavioral setups. We showcase Stytra by reproducing previously published behavioral protocols in both head-restrained and freely-swimming larvae. We also demonstrate the use of the software in the context of a calcium imaging experiment, where it interfaces with other acquisition devices. Our aims are to enable more laboratories to easily implement behavioral experiments, as well as to provide a platform for sharing stimulus protocols that permits easy reproduction of experiments and straightforward validation. Finally, we demonstrate how Stytra can serve as a platform to design behavioral experiments involving tracking or visual stimulation with other animals and provide an example integration with the DeepLabCut neural network-based tracking method.

Sensorimotor Representations in Cerebellar Granule Cells in Larval Zebrafish Are Dense, Spatially Organized, and Non-temporally Patterned.

A fundamental question in neurobiology is how animals integrate external sensory information from their environment with self-generated motor and sensory signals in order to guide motor behavior and adaptation. The cerebellum is a vertebrate hindbrain region where all of these signals converge and that has been implicated in the acquisition, coordination, and calibration of motor activity. Theories of cerebellar function postulate that granule cells encode a variety of sensorimotor signals in the cerebellar input layer. These models suggest that representations should be high-dimensional, sparse, and temporally patterned. However, in vivo physiological recordings addressing these points have been limited and in particular have been unable to measure the spatiotemporal dynamics of population-wide activity. In this study, we use both calcium imaging and electrophysiology in the awake larval zebrafish to investigate how cerebellar granule cells encode three types of sensory stimuli as well as stimulus-evoked motor behaviors. We find that a large fraction of all granule cells are active in response to these stimuli, such that representations are not sparse at the population level. We find instead that most responses belong to only one of a small number of distinct activity profiles, which are temporally homogeneous and anatomically clustered. We furthermore identify granule cells that are active during swimming behaviors and others that are multimodal for sensory and motor variables. When we pharmacologically change the threshold of a stimulus-evoked behavior, we observe correlated changes in these representations. Finally, electrophysiological data show no evidence for temporal patterning in the coding of different stimulus durations.