Neural Networks for Navigation: From Connections to Computations.

Many animals can navigate toward a goal they cannot see based on an internal representation of that goal in the brain's spatial maps. These maps are organized around networks with stable fixed-point dynamics (attractors), anchored to landmarks, and reciprocally connected to motor control. This review summarizes recent progress in understanding these networks, focusing on studies in arthropods. One factor driving recent progress is the availability of the Drosophila connectome; however, it is increasingly clear that navigation depends on ongoing synaptic plasticity in these networks. Functional synapses appear to be continually reselected from the set of anatomical potential synapses based on the interaction of Hebbian learning rules, sensory feedback, attractor dynamics, and neuromodulation. This can explain how the brain's maps of space are rapidly updated; it may also explain how the brain can initialize goals as stable fixed points for navigation.

Parallel Transformation of Tactile Signals in Central Circuits of Drosophila.

To distinguish between complex somatosensory stimuli, central circuits must combine signals from multiple peripheral mechanoreceptor types, as well as mechanoreceptors at different sites in the body. Here, we investigate the first stages of somatosensory integration in Drosophila using in vivo recordings from genetically labeled central neurons in combination with mechanical and optogenetic stimulation of specific mechanoreceptor types. We identify three classes of central neurons that process touch: one compares touch signals on different parts of the same limb, one compares touch signals on right and left limbs, and the third compares touch and proprioceptive signals. Each class encodes distinct features of somatosensory stimuli. The axon of an individual touch receptor neuron can diverge to synapse onto all three classes, meaning that these computations occur in parallel, not hierarchically. Representing a stimulus as a set of parallel comparisons is a fast and efficient way to deliver somatosensory signals to motor circuits.

Transforming a head direction signal into a goal-oriented steering command.

To navigate, we must continuously estimate the direction we are headed in, and we must correct deviations from our goal1. Direction estimation is accomplished by ring attractor networks in the head direction system2,3. However, we do not fully understand how the sense of direction is used to guide action. Drosophila connectome analyses4,5 reveal three cell populations (PFL3R, PFL3L and PFL2) that connect the head direction system to the locomotor system. Here we use imaging, electrophysiology and chemogenetic stimulation during navigation to show how these populations function. Each population receives a shifted copy of the head direction vector, such that their three reference frames are shifted approximately 120° relative to each other. Each cell type then compares its own head direction vector with a common goal vector; specifically, it evaluates the congruence of these vectors via a nonlinear transformation. The output of all three cell populations is then combined to generate locomotor commands. PFL3R cells are recruited when the fly is oriented to the left of its goal, and their activity drives rightward turning; the reverse is true for PFL3L. Meanwhile, PFL2 cells increase steering speed, and are recruited when the fly is oriented far from its goal. PFL2 cells adaptively increase the strength of steering as directional error increases, effectively managing the tradeoff between speed and accuracy. Together, our results show how a map of space in the brain can be combined with an internal goal to generate action commands, via a transformation from world-centric coordinates to body-centric coordinates.

Dopamine promotes head direction plasticity during orienting movements.

In neural networks that store information in their connection weights, there is a tradeoff between sensitivity and stability1,2. Connections must be plastic to incorporate new information, but if they are too plastic, stored information can be corrupted. A potential solution is to allow plasticity only during epochs when task-specific information is rich, on the basis of a 'when-to-learn' signal3. We reasoned that dopamine provides a when-to-learn signal that allows the brain's spatial maps to update when new spatial information is available-that is, when an animal is moving. Here we show that the dopamine neurons innervating the Drosophila head direction network are specifically active when the fly turns to change its head direction. Moreover, their activity scales with moment-to-moment fluctuations in rotational speed. Pairing dopamine release with a visual cue persistently strengthens the cue's influence on head direction cells. Conversely, inhibiting these dopamine neurons decreases the influence of the cue. This mechanism should accelerate learning during moments when orienting movements are providing a rich stream of head direction information, allowing learning rates to be low at other times to protect stored information. Our results show how spatial learning in the brain can be compressed into discrete epochs in which high learning rates are matched to high rates of information intake.

Transforming representations of movement from body- to world-centric space.

When an animal moves through the world, its brain receives a stream of information about the body's translational velocity from motor commands and sensory feedback signals. These incoming signals are referenced to the body, but ultimately, they must be transformed into world-centric coordinates for navigation1,2. Here we show that this computation occurs in the fan-shaped body in the brain of Drosophila melanogaster. We identify two cell types, PFNd and PFNv3-5, that conjunctively encode translational velocity and heading as a fly walks. In these cells, velocity signals are acquired from locomotor brain regions6 and are multiplied with heading signals from the compass system. PFNd neurons prefer forward-ipsilateral movement, whereas PFNv neurons prefer backward-contralateral movement, and perturbing PFNd neurons disrupts idiothetic path integration in walking flies7. Downstream, PFNd and PFNv neurons converge onto hΔB neurons, with a connectivity pattern that pools together heading and translation direction combinations corresponding to the same movement in world-centric space. This network motif effectively performs a rotation of the brain's representation of body-centric translational velocity according to the current heading direction. Consistent with our predictions, we observe that hΔB neurons form a representation of translational velocity in world-centric coordinates. By integrating this representation over time, it should be possible for the brain to form a working memory of the path travelled through the environment8-10.

Bayesian inference in ring attractor networks.

Working memories are thought to be held in attractor networks in the brain. These attractors should keep track of the uncertainty associated with each memory, so as to weigh it properly against conflicting new evidence. However, conventional attractors do not represent uncertainty. Here, we show how uncertainty could be incorporated into an attractor, specifically a ring attractor that encodes head direction. First, we introduce a rigorous normative framework (the circular Kalman filter) for benchmarking the performance of a ring attractor under conditions of uncertainty. Next, we show that the recurrent connections within a conventional ring attractor can be retuned to match this benchmark. This allows the amplitude of network activity to grow in response to confirmatory evidence, while shrinking in response to poor-quality or strongly conflicting evidence. This "Bayesian ring attractor" performs near-optimal angular path integration and evidence accumulation. Indeed, we show that a Bayesian ring attractor is consistently more accurate than a conventional ring attractor. Moreover, near-optimal performance can be achieved without exact tuning of the network connections. Finally, we use large-scale connectome data to show that the network can achieve near-optimal performance even after we incorporate biological constraints. Our work demonstrates how attractors can implement a dynamic Bayesian inference algorithm in a biologically plausible manner, and it makes testable predictions with direct relevance to the head direction system as well as any neural system that tracks direction, orientation, or periodic rhythms.

Sensorimotor experience remaps visual input to a heading-direction network.

In the Drosophila brain, 'compass' neurons track the orientation of the body and head (the fly's heading) during navigation 1,2. In the absence of visual cues, the compass neuron network estimates heading by integrating self-movement signals over time3,4. When a visual cue is present, the estimate of the network is more accurate1,3. Visual inputs to compass neurons are thought to originate from inhibitory neurons called R neurons (also known as ring neurons); the receptive fields of R neurons tile visual space5. The axon of each R neuron overlaps with the dendrites of every compass neuron6, raising the question of how visual cues are integrated into the compass. Here, using in vivo whole-cell recordings, we show that a visual cue can evoke synaptic inhibition in compass neurons and that R neurons mediate this inhibition. Each compass neuron is inhibited only by specific visual cue positions, indicating that many potential connections from R neurons onto compass neurons are actually weak or silent. We also show that the pattern of visually evoked inhibition can reorganize over minutes as the fly explores an altered virtual-reality environment. Using ensemble calcium imaging, we demonstrate that this reorganization causes persistent changes in the compass coordinate frame. Taken together, our data suggest a model in which correlated pre- and postsynaptic activity triggers associative long-term synaptic depression of visually evoked inhibition in compass neurons. Our findings provide evidence for the theoretical proposal that associative plasticity of sensory inputs, when combined with attractor dynamics, can reconcile self-movement information with changing external cues to generate a coherent sense of direction7-12.

Early olfactory processing in Drosophila: mechanisms and principles.

In the olfactory system of Drosophila melanogaster, it is relatively straightforward to target in vivo measurements of neural activity to specific processing channels. This, together with the numerical simplicity of the Drosophila olfactory system, has produced rapid gains in our understanding of Drosophila olfaction. This review summarizes the neurophysiology of the first two layers of this system: the peripheral olfactory receptor neurons and their postsynaptic targets in the antennal lobe. We now understand in some detail the cellular and synaptic mechanisms that shape odor representations in these neurons. Together, these mechanisms imply that interesting neural adaptations to environmental statistics have occurred. These mechanisms also place some fundamental constraints on early sensory processing that pose challenges for higher brain regions. These findings suggest some general principles with broad relevance to early sensory processing in other modalities.

Thermosensory processing in the Drosophila brain.

In Drosophila, just as in vertebrates, changes in external temperature are encoded by bidirectional opponent thermoreceptor cells: some cells are excited by warming and inhibited by cooling, whereas others are excited by cooling and inhibited by warming. The central circuits that process these signals are not understood. In Drosophila, a specific brain region receives input from thermoreceptor cells. Here we show that distinct genetically identified projection neurons (PNs) in this brain region are excited by cooling, warming, or both. The PNs excited by cooling receive mainly feed-forward excitation from cool thermoreceptors. In contrast, the PNs excited by warming ('warm-PNs') receive both excitation from warm thermoreceptors and crossover inhibition from cool thermoreceptors through inhibitory interneurons. Notably, this crossover inhibition elicits warming-evoked excitation, because warming suppresses tonic activity in cool thermoreceptors. This in turn disinhibits warm-PNs and sums with feed-forward excitation evoked by warming. Crossover inhibition could cancel non-thermal activity (noise) that is positively correlated among warm and cool thermoreceptor cells, while reinforcing thermal activity which is anti-correlated. Our results show how central circuits can combine signals from bidirectional opponent neurons to construct sensitive and robust neural codes.

Sound localization behavior in Drosophilamelanogaster depends on inter-antenna vibration amplitude comparisons.

Drosophila melanogaster hear with their antennae: sound evokes vibration of the distal antennal segment, and this vibration is transduced by specialized mechanoreceptor cells. The left and right antennae vibrate preferentially in response to sounds arising from different azimuthal angles. Therefore, by comparing signals from the two antennae, it should be possible to obtain information about the azimuthal angle of a sound source. However, behavioral evidence of sound localization has not been reported in Drosophila Here, we show that walking D. melanogaster do indeed turn in response to lateralized sounds. We confirm that this behavior is evoked by vibrations of the distal antennal segment. The rule for turning is different for sounds arriving from different locations: flies turn toward sounds in their front hemifield, but they turn away from sounds in their rear hemifield, and they do not turn at all in response to sounds from 90 or -90 deg. All of these findings can be explained by a simple rule: the fly steers away from the antenna with the larger vibration amplitude. Finally, we show that these behaviors generalize to sound stimuli with diverse spectro-temporal features, and that these behaviors are found in both sexes. Our findings demonstrate the behavioral relevance of the antenna's directional tuning properties. They also pave the way for investigating the neural implementation of sound localization, as well as the potential roles of sound-guided steering in courtship and exploration.

Asymmetric neurotransmitter release enables rapid odour lateralization in Drosophila.

In Drosophila, most individual olfactory receptor neurons (ORNs) project bilaterally to both sides of the brain. Having bilateral rather than unilateral projections may represent a useful redundancy. However, bilateral ORN projections to the brain should also compromise the ability to lateralize odours. Nevertheless, walking or flying Drosophila reportedly turn towards the antenna that is more strongly stimulated by odour. Here we show that each ORN spike releases approximately 40% more neurotransmitter from the axon branch ipsilateral to the soma than from the contralateral branch. As a result, when an odour activates the antennae asymmetrically, ipsilateral central neurons begin to spike a few milliseconds before contralateral neurons, and at a 30 to 50% higher rate than contralateral neurons. We show that a walking fly can detect a 5% asymmetry in total ORN input to its left and right antennal lobes, and can turn towards the odour in less time than it requires the fly to complete a stride. These results demonstrate that neurotransmitter release properties can be tuned independently at output synapses formed by a single axon onto two target cells with identical functions and morphologies. Our data also show that small differences in spike timing and spike rate can produce reliable differences in olfactory behaviour.

Mechanisms Underlying Population Response Dynamics in Inhibitory Interneurons of the Drosophila Antennal Lobe.

Local inhibitory neurons control the timing of neural activity in many circuits. To understand how inhibition controls timing, it is important to understand the dynamics of activity in populations of local inhibitory interneurons, as well as the mechanisms that underlie these dynamics. Here we describe the in vivo response dynamics of a large population of inhibitory local neurons (LNs) in the Drosophila melanogaster antennal lobe, the analog of the vertebrate olfactory bulb, and we dissect the network and intrinsic mechanisms that give rise to these dynamics. Some LNs respond to odor onsets ("ON" cells) and others to offsets ("OFF" cells), whereas still others respond at both times. Moreover, different LNs signal odor concentration fluctuations on different timescales. Some respond rapidly, and can track rapid concentration fluctuations. Others respond slowly, and are best at tracking slow fluctuations. We found a continuous spectrum of preferred stimulation timescales among LNs, as well as a continuum of ON-OFF behavior. Using in vivo whole-cell recordings, we show that the timing of an LN's response (ON vs OFF) can be predicted from the interplay of excitatory and inhibitory synaptic currents that it receives. Meanwhile, the preferred timescale of an LN is related to its intrinsic properties. These results illustrate how a population of inhibitory interneurons can collectively encode bidirectional changes in stimulus intensity on multiple timescales, and how this can arise via an interaction between synaptic and intrinsic mechanisms. SIGNIFICANCE STATEMENT: Most neural circuits contain diverse populations of inhibitory interneurons. The way inhibition shapes network activity will depend on the spiking dynamics of the interneuron population. Here we describe the dynamics of activity in a large population of inhibitory interneurons in the first brain relay of the fruit fly olfactory system. Because odor plumes fluctuate on multiple timescales, the drive to this circuit can vary over a range of frequencies. We show how synaptic and cellular mechanisms interact to recruit different interneurons at different times, and in response to different temporal features of odor stimuli. As a result, inhibition is recruited over a range of conditions, and there is the potential to tune the timing of inhibition as the environment changes.

Glutamate is an inhibitory neurotransmitter in the Drosophila olfactory system.

Glutamatergic neurons are abundant in the Drosophila central nervous system, but their physiological effects are largely unknown. In this study, we investigated the effects of glutamate in the Drosophila antennal lobe, the first relay in the olfactory system and a model circuit for understanding olfactory processing. In the antennal lobe, one-third of local neurons are glutamatergic. Using in vivo whole-cell patch clamp recordings, we found that many glutamatergic local neurons are broadly tuned to odors. Iontophoresed glutamate hyperpolarizes all major cell types in the antennal lobe, and this effect is blocked by picrotoxin or by transgenic RNAi-mediated knockdown of the GluClα gene, which encodes a glutamate-gated chloride channel. Moreover, antennal lobe neurons are inhibited by selective activation of glutamatergic local neurons using a nonnative genetically encoded cation channel. Finally, transgenic knockdown of GluClα in principal neurons disinhibits the odor responses of these neurons. Thus, glutamate acts as an inhibitory neurotransmitter in the antennal lobe, broadly similar to the role of GABA in this circuit. However, because glutamate release is concentrated between glomeruli, whereas GABA release is concentrated within glomeruli, these neurotransmitters may act on different spatial and temporal scales. Thus, the existence of two parallel inhibitory transmitter systems may increase the range and flexibility of synaptic inhibition.

Lateral presynaptic inhibition mediates gain control in an olfactory circuit.

Olfactory signals are transduced by a large family of odorant receptor proteins, each of which corresponds to a unique glomerulus in the first olfactory relay of the brain. Crosstalk between glomeruli has been proposed to be important in olfactory processing, but it is not clear how these interactions shape the odour responses of second-order neurons. In the Drosophila antennal lobe (a region analogous to the vertebrate olfactory bulb), we selectively removed most interglomerular input to genetically identified second-order olfactory neurons. Here we show that this broadens the odour tuning of these neurons, implying that interglomerular inhibition dominates over interglomerular excitation. The strength of this inhibitory signal scales with total feedforward input to the entire antennal lobe, and has similar tuning in different glomeruli. A substantial portion of this interglomerular inhibition acts at a presynaptic locus, and our results imply that this is mediated by both ionotropic and metabotropic receptors on the same nerve terminal.

Interactions between specialized gain control mechanisms in olfactory processing.

Gain control is a process that adjusts a system's sensitivity when input levels change. Neural systems contain multiple mechanisms of gain control, but we do not understand why so many mechanisms are needed or how they interact. Here, we investigate these questions in the Drosophila antennal lobe, where we identify several types of inhibitory interneurons with specialized gain control functions. We find that some interneurons are nonspiking, with compartmentalized calcium signals, and they specialize in intra-glomerular gain control. Conversely, we find that other interneurons are recruited by strong and widespread network input; they specialize in global presynaptic gain control. Using computational modeling and optogenetic perturbations, we show how these mechanisms can work together to improve stimulus discrimination while also minimizing temporal distortions in network activity. Our results demonstrate how the robustness of neural network function can be increased by interactions among diverse and specialized mechanisms of gain control.

Fine-grained descending control of steering in walking Drosophila.

Locomotion involves rhythmic limb movement patterns that originate in circuits outside the brain. Purposeful locomotion requires descending commands from the brain, but we do not understand how these commands are structured. Here we investigate this issue, focusing on the control of steering in walking Drosophila. First, we describe different limb "gestures" associated with different steering maneuvers. Next, we identify a set of descending neurons whose activity predicts steering. Focusing on two descending cell types downstream from distinct brain networks, we show that they evoke specific limb gestures: one lengthens strides on the outside of a turn, while the other attenuates strides on the inside of a turn. Notably, a single descending neuron can have opposite effects during different locomotor rhythm phases, and we identify networks positioned to implement this phase-specific gating. Together, our results show how purposeful locomotion emerges from brain cells that drive specific, coordinated modulations of low-level patterns.

Cell death triggers olfactory circuit plasticity via glial signaling in Drosophila.

The Drosophila antennal lobe is organized into glomerular compartments, where olfactory receptor neurons synapse onto projection neurons. Projection neuron dendrites also receive input from local neurons, which interconnect glomeruli. In this study, we investigated how activity in this circuit changes over time when sensory afferents are chronically removed in vivo. In the normal circuit, excitatory connections between glomeruli are weak. However, after we chronically severed receptor neuron axons projecting to a subset of glomeruli, we found that odor-evoked lateral excitatory input to deafferented projection neurons was potentiated severalfold. This was caused, at least in part, by strengthened electrical coupling from excitatory local neurons onto projection neurons, as well as increased activity in excitatory local neurons. Merely silencing receptor neurons was not sufficient to elicit these changes, implying that severing receptor neuron axons is the relevant signal. When we expressed the neuroprotective gene Wallerian degeneration slow (Wld(S)) in receptor neurons before severing their axons, this blocked the induction of plasticity. Because expressing Wld(S) prevents severed axons from recruiting glia, this result suggests a role for glia. Consistent with this, we found that blocking endocytosis in ensheathing glia blocked the induction of plasticity. In sum, these results reveal a novel injury response whereby severed sensory axons recruit glia, which in turn signal to central neurons to upregulate their activity. By strengthening excitatory interactions between neurons in a deafferented brain region, this mechanism might help boost activity to compensate for lost sensory input.

Transduction in Drosophila olfactory receptor neurons is invariant to air speed.

In the vertebrate nose, increasing air speed tends to increase the magnitude of odor-evoked activity in olfactory receptor neurons (ORNs), given constant odor concentration and duration. It is often assumed that the same is true of insect olfactory organs, but this has not been directly tested. In this study, we examined the effect of air speed on ORN responses in Drosophila melanogaster. We constructed an odor delivery device that allowed us to independently vary concentration and air speed, and we used a fast photoionization detector to precisely measure the actual odor concentration at the antenna while simultaneously recording spikes from ORNs in vivo. Our results demonstrate that Drosophila ORN odor responses are invariant to air speed, as long as odor concentration is kept constant. This finding was true across a >100-fold range of air speeds. Because odor hydrophobicity has been proposed to affect the air speed dependence of olfactory transduction, we tested a >1,000-fold range of hydrophobicity values and found that ORN responses are invariant to air speed across this full range. These results have implications for the mechanisms of odor delivery to Drosophila ORNs. Our findings are also significant because flies have a limited ability to control air flow across their antennae, unlike terrestrial vertebrates, which can control air flow within their nasal cavity. Thus, for the fly, invariance to air speed may be adaptive because it confers robustness to changing wind conditions.

Locomotor and olfactory responses in dopamine neurons of the Drosophila superior-lateral brain.

The Drosophila brain contains about 50 distinct morphological types of dopamine neurons.1,2,3,4 Physiological studies of Drosophila dopamine neurons have been largely limited to one brain region, the mushroom body,5,6,7,8,9,10,11,12,13 where they are implicated in learning.14,15,16,17,18 By comparison, we know little about the physiology of other Drosophila dopamine neurons. Interestingly, a recent whole-brain imaging study found that dopamine neuron activity in several fly brain regions is correlated with locomotion.19 This is notable because many dopamine neurons in the rodent brain are also correlated with locomotion or other movements20,21,22,23,24,25,26,27,28,29,30; however, most rodent studies have focused on learned and rewarded behaviors, and few have investigated dopamine neuron activity during spontaneous (self-timed) movements. In this study, we monitored dopamine neurons in the Drosophila brain during self-timed locomotor movements, focusing on several previously uncharacterized cell types that arborize in the superior-lateral brain, specifically the lateral horn and superior-lateral protocerebrum. We found that activity of all of these dopamine neurons correlated with spontaneous fluctuations in walking speed, with different cell types showing different speed correlations. Some dopamine neurons also responded to odors, but these responses were suppressed by repeated odor encounters. Finally, we found that the same identifiable dopamine neuron can encode different combinations of locomotion and odor in different individuals. If these dopamine neurons promote synaptic plasticity-like the dopamine neurons of the mushroom body-then, their tuning profiles would imply that plasticity depends on a flexible integration of sensory signals, motor signals, and recent experience.