The computation of directional selectivity in the Drosophila OFF motion pathway

Connectome-constrained networks predict neural activity across the fly visual system

We can now measure the connectivity of every neuron in a neural circuit1-9, but we cannot measure other biological details, including the dynamical characteristics of each neuron. The degree to which measurements of connectivity alone can inform the understanding of neural computation is an open question10. Here we show that with experimental measurements of only the connectivity of a biological neural network, we can predict the neural activity underlying a specified neural computation. We constructed a model neural network with the experimentally determined connectivity for 64 cell types in the motion pathways of the fruit fly optic lobe1-5 but with unknown parameters for the single-neuron and single-synapse properties. We then optimized the values of these unknown parameters using techniques from deep learning11, to allow the model network to detect visual motion12. Our mechanistic model makes detailed, experimentally testable predictions for each neuron in the connectome. We found that model predictions agreed with experimental measurements of neural activity across 26 studies. Our work demonstrates a strategy for generating detailed hypotheses about the mechanisms of neural circuit function from connectivity measurements. We show that this strategy is more likely to be successful when neurons are sparsely connected-a universally observed feature of biological neural networks across species and brain regions.

Direct comparison reveals algorithmic similarities in fly and mouse visual motion detection

Evolution has equipped vertebrates and invertebrates with neural circuits that selectively encode visual motion. While similarities in the computations performed by these circuits in mouse and fruit fly have been noted, direct experimental comparisons have been lacking. Because molecular mechanisms and neuronal morphology in the two species are distinct, we directly compared motion encoding in these two species at the algorithmic level, using matched stimuli and focusing on a pair of analogous neurons, the mouse ON starburst amacrine cell (ON SAC) and Drosophila T4 neurons. We find that the cells share similar spatiotemporal receptive field structures, sensitivity to spatiotemporal correlations, and tuning to sinusoidal drifting gratings, but differ in their responses to apparent motion stimuli. Both neuron types showed a response to summed sinusoids that deviates from models for motion processing in these cells, underscoring the similarities in their processing and identifying response features that remain to be explained.

A comprehensive neuroanatomical survey of the Drosophila Lobula Plate Tangential Neurons with predictions for their optic flow sensitivity

Flying insects exhibit remarkable navigational abilities controlled by their compact nervous systems. Optic flow, the pattern of changes in the visual scene induced by locomotion, is a crucial sensory cue for robust self-motion estimation, especially during rapid flight. Neurons that respond to specific, large-field optic flow patterns have been studied for decades, primarily in large flies, such as houseflies, blowflies, and hover flies. The best-known optic-flow sensitive neurons are the large tangential cells of the dipteran lobula plate, whose visual-motion responses, and to a lesser extent, their morphology, have been explored using single-neuron neurophysiology. Most of these studies have focused on the large, Horizontal and Vertical System neurons, yet the lobula plate houses a much larger set of 'optic-flow' sensitive neurons, many of which have been challenging to unambiguously identify or to reliably target for functional studies. Here we report the comprehensive reconstruction and identification of the Lobula Plate Tangential Neurons in an Electron Microscopy (EM) volume of a whole Drosophila brain. This catalog of 58 LPT neurons (per brain hemisphere) contains many neurons that are described here for the first time and provides a basis for systematic investigation of the circuitry linking self-motion to locomotion control. Leveraging computational anatomy methods, we estimated the visual motion receptive fields of these neurons and compared their tuning to the visual consequence of body rotations and translational movements. We also matched these neurons, in most cases on a one-for-one basis, to stochastically labeled cells in genetic driver lines, to the mirror-symmetric neurons in the same EM brain volume, and to neurons in an additional EM data set. Using cell matches across data sets, we analyzed the integration of optic flow patterns by neurons downstream of the LPTs and find that most central brain neurons establish sharper selectivity for global optic flow patterns than their input neurons. Furthermore, we found that self-motion information extracted from optic flow is processed in distinct regions of the central brain, pointing to diverse foci for the generation of visual behaviors.

How Flies See Motion

How neurons detect the direction of motion is a prime example of neural computation: Motion vision is found in the visual systems of virtually all sighted animals, it is important for survival, and it requires interesting computations with well-defined linear and nonlinear processing steps-yet the whole process is of moderate complexity. The genetic methods available in the fruit fly Drosophila and the charting of a connectome of its visual system have led to rapid progress and unprecedented detail in our understanding of how neurons compute the direction of motion in this organism. The picture that emerged incorporates not only the identity, morphology, and synaptic connectivity of each neuron involved but also its neurotransmitters, its receptors, and their subcellular localization. Together with the neurons' membrane potential responses to visual stimulation, this information provides the basis for a biophysically realistic model of the circuit that computes the direction of visual motion.

Visual processing in the fly, from photoreceptors to behavior

Originally a genetic model organism, the experimental use of Drosophila melanogaster has grown to include quantitative behavioral analyses, sophisticated perturbations of neuronal function, and detailed sensory physiology. A highlight of these developments can be seen in the context of vision, where pioneering studies have uncovered fundamental and generalizable principles of sensory processing. Here we begin with an overview of vision-guided behaviors and common methods for probing visual circuits. We then outline the anatomy and physiology of brain regions involved in visual processing, beginning at the sensory periphery and ending with descending motor control. Areas of focus include contrast and motion detection in the optic lobe, circuits for visual feature selectivity, computations in support of spatial navigation, and contextual associative learning. Finally, we look to the future of fly visual neuroscience and discuss promising topics for further study.

Disynaptic inhibition shapes tuning of OFF-motion detectors in Drosophila

The circuitry underlying the detection of visual motion in Drosophila melanogaster is one of the best studied networks in neuroscience. Lately, electron microscopy reconstructions, algorithmic models, and functional studies have proposed a common motif for the cellular circuitry of an elementary motion detector based on both supralinear enhancement for preferred direction and sublinear suppression for null-direction motion. In T5 cells, however, all columnar input neurons (Tm1, Tm2, Tm4, and Tm9) are excitatory. So, how is null-direction suppression realized there? Using two-photon calcium imaging in combination with thermogenetics, optogenetics, apoptotics, and pharmacology, we discovered that it is via CT1, the GABAergic large-field amacrine cell, where the different processes have previously been shown to act in an electrically isolated way. Within each column, CT1 receives excitatory input from Tm9 and Tm1 and provides the sign-inverted, now inhibitory input signal onto T5. Ablating CT1 or knocking down GABA-receptor subunit Rdl significantly broadened the directional tuning of T5 cells. It thus appears that the signal of Tm1 and Tm9 is used both as an excitatory input for preferred direction enhancement and, through a sign inversion within the Tm1/Tm9-CT1 microcircuit, as an inhibitory input for null-direction suppression.

Voltage to Calcium Transformation Enhances Direction Selectivity in Drosophila T4 Neurons

An important step in neural information processing is the transformation of membrane voltage into calcium signals leading to transmitter release. However, the effect of voltage to calcium transformation on neural responses to different sensory stimuli is not well understood. Here, we use in vivo two-photon imaging of genetically encoded voltage and calcium indicators, ArcLight and GCaMP6f, respectively, to measure responses in direction-selective T4 neurons of female Drosophila Comparison between ArcLight and GCaMP6f signals reveals calcium signals to have a significantly higher direction selectivity compared with voltage signals. Using these recordings, we build a model which transforms T4 voltage responses into calcium responses. Using a cascade of thresholding, temporal filtering and a stationary nonlinearity, the model reproduces experimentally measured calcium responses across different visual stimuli. These findings provide a mechanistic underpinning of the voltage to calcium transformation and show how this processing step, in addition to synaptic mechanisms on the dendrites of T4 cells, enhances direction selectivity in the output signal of T4 neurons. Measuring the directional tuning of postsynaptic vertical system (VS)-cells with inputs from other cells blocked, we found that, indeed, it matches the one of the calcium signal in presynaptic T4 cells.SIGNIFICANCE STATEMENT The transformation of voltage to calcium influx is an important step in the signaling cascade within a nerve cell. While this process has been intensely studied in the context of transmitter release mechanism, its consequences for information transmission and neural computation are unclear. Here, we measured both membrane voltage and cytosolic calcium levels in direction-selective cells of Drosophila in response to a large set of visual stimuli. We found direction selectivity in the calcium signal to be significantly enhanced compared with membrane voltage through a nonlinear transformation of voltage to calcium. Our findings highlight the importance of an additional step in the signaling cascade for information processing within single nerve cells.

Excitatory and inhibitory neural dynamics jointly tune motion detection

Neurons integrate excitatory and inhibitory signals to produce their outputs, but the role of input timing in this integration remains poorly understood. Motion detection is a paradigmatic example of this integration, since theories of motion detection rely on different delays in visual signals. These delays allow circuits to compare scenes at different times to calculate the direction and speed of motion. Different motion detection circuits have different velocity sensitivity, but it remains untested how the response dynamics of individual cell types drive this tuning. Here, we sped up or slowed down specific neuron types in Drosophila's motion detection circuit by manipulating ion channel expression. Altering the dynamics of individual neuron types upstream of motion detectors increased their sensitivity to fast or slow visual motion, exposing distinct roles for excitatory and inhibitory dynamics in tuning directional signals, including a role for the amacrine cell CT1. A circuit model constrained by functional data and anatomy qualitatively reproduced the observed tuning changes. Overall, these results reveal how excitatory and inhibitory dynamics together tune a canonical circuit computation.

Synaptic counts approximate synaptic contact area in Drosophila

The pattern of synaptic connections among neurons defines the circuit structure, which constrains the computations that a circuit can perform. The strength of synaptic connections is costly to measure yet important for accurate circuit modeling. Synaptic surface area has been shown to correlate with synaptic strength, yet in the emerging field of connectomics, most studies rely instead on the counts of synaptic contacts between two neurons. Here we quantified the relationship between synaptic count and synaptic area as measured from volume electron microscopy of the larval Drosophila central nervous system. We found that the total synaptic surface area, summed across all synaptic contacts from one presynaptic neuron to a postsynaptic one, can be accurately predicted solely from the number of synaptic contacts, for a variety of neurotransmitters. Our findings support the use of synaptic counts for approximating synaptic strength when modeling neural circuits.

Populations of local direction-selective cells encode global motion patterns generated by self-motion

Self-motion generates visual patterns on the eye that are important for navigation. These optic flow patterns are encoded by the population of local direction–selective cells in the mouse retina, whereas in flies, local direction–selective T4/T5 cells are thought to be uniformly tuned. How complex global motion patterns can be computed downstream is unclear. We show that the population of T4/T5 cells in Drosophila encodes global motion patterns. Whereas the mouse retina encodes four types of optic flow, the fly visual system encodes six. This matches the larger number of degrees of freedom and the increased complexity of translational and rotational motion patterns during flight. The four uniformly tuned T4/T5 subtypes described previously represent a local subset of the population. Thus, a population code for global motion patterns appears to be a general coding principle of visual systems that matches local motion responses to modes of the animal’s movement.

Shallow neural networks trained to detect collisions recover features of visual loom-selective neurons

Animals have evolved sophisticated visual circuits to solve a vital inference problem: detecting whether or not a visual signal corresponds to an object on a collision course. Such events are detected by specific circuits sensitive to visual looming, or objects increasing in size. Various computational models have been developed for these circuits, but how the collision-detection inference problem itself shapes the computational structures of these circuits remains unknown. Here, inspired by the distinctive structures of LPLC2 neurons in the visual system of Drosophila, we build anatomically-constrained shallow neural network models and train them to identify visual signals that correspond to impending collisions. Surprisingly, the optimization arrives at two distinct, opposing solutions, only one of which matches the actual dendritic weighting of LPLC2 neurons. Both solutions can solve the inference problem with high accuracy when the population size is large enough. The LPLC2-like solutions reproduces experimentally observed LPLC2 neuron responses for many stimuli, and reproduces canonical tuning of loom sensitive neurons, even though the models are never trained on neural data. Thus, LPLC2 neuron properties and tuning are predicted by optimizing an anatomically-constrained neural network to detect impending collisions. More generally, these results illustrate how optimizing inference tasks that are important for an animal's perceptual goals can reveal and explain computational properties of specific sensory neurons.

Flexible filtering by neural inputs supports motion computation across states and stimuli

Sensory systems flexibly adapt their processing properties across a wide range of environmental and behavioral conditions. Such variable processing complicates attempts to extract a mechanistic understanding of sensory computations. This is evident in the highly constrained, canonical Drosophila motion detection circuit, where the core computation underlying direction selectivity is still debated despite extensive studies. Here we measured the filtering properties of neural inputs to the OFF motion-detecting T5 cell in Drosophila. We report state- and stimulus-dependent changes in the shape of these signals, which become more biphasic under specific conditions. Summing these inputs within the framework of a connectomic-constrained model of the circuit demonstrates that these shapes are sufficient to explain T5 responses to various motion stimuli. Thus, our stimulus- and state-dependent measurements reconcile motion computation with the anatomy of the circuit. These findings provide a clear example of how a basic circuit supports flexible sensory computation.

Motion vision: Pinning down motion computation in an ever-changing circuit

A new electrophysiological study of the Drosophila visual system, recording from columnar inputs to motion-detecting neurons, has provided new insights into the computations that underlie motion vision.

Non-preferred contrast responses in the Drosophila motion pathways reveal a receptive field structure that explains a common visual illusion

Diverse sensory systems, from audition to thermosensation, feature a separation of inputs into ON (increments) and OFF (decrements) signals. In the Drosophila visual system, separate ON and OFF pathways compute the direction of motion, yet anatomical and functional studies have identified some crosstalk between these channels. We used this well-studied circuit to ask whether the motion computation depends on ON-OFF pathway crosstalk. Using whole-cell electrophysiology, we recorded visual responses of T4 (ON) and T5 (OFF) cells, mapped their composite ON-OFF receptive fields, and found that they share a similar spatiotemporal structure. We fit a biophysical model to these receptive fields that accurately predicts directionally selective T4 and T5 responses to both ON and OFF moving stimuli. This model also provides a detailed mechanistic explanation for the directional preference inversion in response to the prominent reverse-phi illusion. Finally, we used the steering responses of tethered flying flies to validate the model's predicted effects of varying stimulus parameters on the behavioral turning inversion.

Predicting individual neuron responses with anatomically constrained task optimization

Artificial neural networks trained to solve sensory tasks can develop statistical representations that match those in biological circuits. However, it remains unclear whether they can reproduce properties of individual neurons. Here, we investigated how artificial networks predict individual neuron properties in the visual motion circuits of the fruit fly Drosophila. We trained anatomically constrained networks to predict movement in natural scenes, solving the same inference problem as fly motion detectors. Units in the artificial networks adopted many properties of analogous individual neurons, even though they were not explicitly trained to match these properties. Among these properties was the split into ON and OFF motion detectors, which is not predicted by classical motion detection models. The match between model and neurons was closest when models were trained to be robust to noise. These results demonstrate how anatomical, task, and noise constraints can explain properties of individual neurons in a small neural network.

The physiological basis for contrast opponency in motion computation in Drosophila

In Drosophila, direction-selective neurons implement a mechanism of motion computation similar to cortical neurons, using contrast-opponent receptive fields with ON and OFF subfields. It is not clear how the presynaptic circuitry of direction-selective neurons in the OFF pathway supports this computation if all major inputs are OFF-rectified neurons. Here, we reveal the biological substrate for motion computation in the OFF pathway. Three interneurons, Tm2, Tm9 and CT1, provide information about ON stimuli to the OFF direction-selective neuron T5 across its receptive field, supporting a contrast-opponent receptive field organization. Consistent with its prominent role in motion detection, variability in Tm9 receptive field properties transfers to T5, and calcium decrements in Tm9 in response to ON stimuli persist across behavioral states, while spatial tuning is sharpened by active behavior. Together, our work shows how a key neuronal computation is implemented by its constituent neuronal circuit elements to ensure direction selectivity.

Conditional protein tagging methods reveal highly specific subcellular distribution of ion channels in motion-sensing neurons

Neurotransmitter receptors and ion channels shape the biophysical properties of neurons, from the sign of the response mediated by neurotransmitter receptors to the dynamics shaped by voltage-gated ion channels. Therefore, knowing the localizations and types of receptors and channels present in neurons is fundamental to our understanding of neural computation. Here, we developed two approaches to visualize the subcellular localization of specific proteins in Drosophila: The flippase-dependent expression of GFP-tagged receptor subunits in single neurons and ‘FlpTag’, a versatile new tool for the conditional labelling of endogenous proteins. Using these methods, we investigated the subcellular distribution of the receptors GluClα, Rdl, and Dα7 and the ion channels para and Ih in motion-sensing T4/T5 neurons of the Drosophila visual system. We discovered a strictly segregated subcellular distribution of these proteins and a sequential spatial arrangement of glutamate, acetylcholine, and GABA receptors along the dendrite that matched the previously reported EM-reconstructed synapse distributions.

Mechanism for analogous illusory motion perception in flies and humans

Visual motion detection is one of the most important computations performed by visual circuits. Yet, we perceive vivid illusory motion in stationary, periodic luminance gradients that contain no true motion. This illusion is shared by diverse vertebrate species, but theories proposed to explain this illusion have remained difficult to test. Here, we demonstrate that in the fruit fly Drosophila, the illusory motion percept is generated by unbalanced contributions of direction-selective neurons' responses to stationary edges. First, we found that flies, like humans, perceive sustained motion in the stationary gradients. The percept was abolished when the elementary motion detector neurons T4 and T5 were silenced. In vivo calcium imaging revealed that T4 and T5 neurons encode the location and polarity of stationary edges. Furthermore, our proposed mechanistic model allowed us to predictably manipulate both the magnitude and direction of the fly's illusory percept by selectively silencing either T4 or T5 neurons. Interestingly, human brains possess the same mechanistic ingredients that drive our model in flies. When we adapted human observers to moving light edges or dark edges, we could manipulate the magnitude and direction of their percepts as well, suggesting that mechanisms similar to the fly's may also underlie this illusion in humans. By taking a comparative approach that exploits Drosophila neurogenetics, our results provide a causal, mechanistic account for a long-known visual illusion. These results argue that this illusion arises from architectures for motion detection that are shared across phyla.

The computation of directional selectivity in the Drosophila OFF motion pathway

In flies, the direction of moving ON and OFF features is computed separately. T4 (ON) and T5 (OFF) are the first neurons in their respective pathways to extract a directionally selective response from their non-selective inputs. Our recent study of T4 found that the integration of offset depolarizing and hyperpolarizing inputs is critical for the generation of directional selectivity. However, T5s lack small-field inhibitory inputs, suggesting they may use a different mechanism. Here we used whole-cell recordings of T5 neurons and found a similar receptive field structure: fast depolarization and persistent, spatially offset hyperpolarization. By assaying pairwise interactions of local stimulation across the receptive field, we found no amplifying responses, only suppressive responses to the non-preferred motion direction. We then evaluated passive, biophysical models and found that a model using direct inhibition, but not the removal of excitation, can accurately predict T5 responses to a range of moving stimuli.