Search

State-of-the-art algorithms for visual place recognition, and related visual navigation systems, can be broadly split into two categories: computer-science-oriented models including deep learning or image retrieval-based techniques with minimal biological plausibility, and neuroscience-oriented dynamical networks that model temporal properties underlying spatial navigation in the brain. In this letter, we propose a new compact and high-performing place recognition model that bridges this divide for the first time. Our approach comprises two key neural models of these categories: (1) FlyNet, a compact, sparse two-layer neural network inspired by brain architectures of fruit flies, Drosophila melanogaster, and (2) a one-dimensional continuous attractor neural network (CANN). The resulting FlyNet+CANN network incorporates the compact pattern recognition capabilities of our FlyNet model with the powerful temporal filtering capabilities of an equally compact CANN, replicating entirely in a hybrid neural implementation the functionality that yields high performance in algorithmic localization approaches like SeqSLAM. We evaluate our model, and compare it to three state-of-the-art methods, on two benchmark real-world datasets with small viewpoint variations and extreme environmental changes - achieving 87% AUC results under day to night transitions compared to 60% for Multi-Process Fusion, 46% for LoST-X and 1% for SeqSLAM, while being 6.5, 310, and 1.5 times faster, respectively., Preprint version of article published in IEEE Robotics and Automation Letters

The brain evolved to produce behaviors that help an animal inhabit the natural world. During natural behaviors, the brain is engaged in many levels of activity from the detection of sensory inputs to decision-making to motor planning and execution. To date, most brain studies have focused on small numbers of neurons that interact in limited circuits. This allows analyzing individual computations or steps of neural processing. During behavior, however, brain activity must integrate multiple circuits in different brain regions. The activities of different brain regions are not isolated, but may be contingent on one another. Coordinated and concurrent activity within and across brain areas is organized by (1) sensory information from the environment, (2) the animal’s internal behavioral state, and (3) recurrent networks of synaptic and non-synaptic connectivity. Whole-brain recording with cellular resolution provides a new opportunity to dissect the neural basis of behavior, but whole-brain activity is also mutually contingent on behavior itself. This is especially true for natural behaviors like navigation, mating, or hunting, which require dynamic interaction between the animal, its environment, and other animals. In such behaviors, the sensory experience of an unrestrained animal is actively shaped by its movements and decisions. Many of the signaling and feedback pathways that an animal uses to guide behavior only occur in freely moving animals. Recent technological advances have enabled whole-brain recording in small behaving animals including nematodes, flies, and zebrafish. These whole-brain experiments capture neural activity with cellular resolution spanning sensory, decision-making, and motor circuits, and thereby demand new theoretical approaches that integrate brain dynamics with behavioral dynamics. Here, we review the experimental and theoretical methods that are being employed to understand animal behavior and whole-brain activity, and the opportunities for physics to contribute to this emerging field of systems neuroscience.

Flexible integration of warming and cooling pathways underlies thermal homeostasis in larval Drosophila., Body temperature homeostasis is essential and reliant upon the integration of outputs from multiple classes of cooling- and warming-responsive cells. The computations that integrate these outputs are not understood. Here, we discover a set of warming cells (WCs) and show that the outputs of these WCs combine with previously described cooling cells (CCs) in a cross-inhibition computation to drive thermal homeostasis in larval Drosophila. WCs and CCs detect temperature changes using overlapping combinations of ionotropic receptors: Ir68a, Ir93a, and Ir25a for WCs and Ir21a, Ir93a, and Ir25a for CCs. WCs mediate avoidance to warming while cross-inhibiting avoidance to cooling, and CCs mediate avoidance to cooling while cross-inhibiting avoidance to warming. Ambient temperature–dependent regulation of the strength of WC- and CC-mediated cross-inhibition keeps larvae near their homeostatic set point. Using neurophysiology, quantitative behavioral analysis, and connectomics, we demonstrate how flexible integration between warming and cooling pathways can orchestrate homeostatic thermoregulation.

Animals use their olfactory systems to avoid predators, forage for food, and identify mates. Olfactory systems detect and distinguish odors by responding to the concentration, temporal dynamics, and identities of odorant molecules. Studying the temporal neural processing of odors carried in air has been difficult because of the inherent challenge in precisely controlling odorized airflows over time. Odorized airflows interact with surfaces and other air currents, leading to a complex transformation from the odorized airflow that is desired to the olfactory stimulus that is delivered. Here, we present a method that achieves precise and automated control of the amplitude, baseline, and temporal structure of olfactory stimuli. We use this technique to analyze the temporal processing of olfactory stimuli in the early olfactory circuits and navigational behavior of larval Drosophila. Precise odor control and calcium measurements in the axon terminal of an Olfactory Receptor Neuron (ORN-Or42b) revealed dynamic adaptation properties: as in vertebrate photoreceptor neurons, Or42b-ORNs display simultaneous gain-suppression and speedup of their neural response. Furthermore, we found that ORN sensitivity to changes in odor concentration decreases with odor background, but the sensitivity to odor contrast is invariant – this causes odor-evoked ORN activity to follow the Weber-Fechner Law. Using precise olfactory stimulus control with freely-moving animals, we uncovered correlations between the temporal dynamics of larval navigation motor programs and the neural response dynamics of second-order olfactory neurons. The correspondence between neural and behavioral dynamics highlights the potential of precise odor temporal dynamics control in dissecting the sensorimotor circuits for olfactory behaviors.

The first olfactory processing center in the larval Drosophila brain uses information about feeding state to shape behavior., Animals exhibit different behavioral responses to the same sensory cue depending on their internal state at a given moment. How and where in the brain are sensory inputs combined with state information to select an appropriate behavior? Here, we investigate how food deprivation affects olfactory behavior in Drosophila larvae. We find that certain odors repel well-fed animals but attract food-deprived animals and that feeding state flexibly alters neural processing in the first olfactory center, the antennal lobe. Hunger differentially modulates two output pathways required for opposing behavioral responses. Upon food deprivation, attraction-mediating uniglomerular projection neurons show elevated odor-evoked activity, whereas an aversion-mediating multiglomerular projection neuron receives odor-evoked inhibition. The switch between these two pathways is regulated by the lone serotonergic neuron in the antennal lobe, CSD. Our findings demonstrate how flexible behaviors can arise from state-dependent circuit dynamics in an early sensory processing center.

Body temperature homeostasis is an essential function that relies upon the integration of the outputs from multiple classes of cooling- and warming-responsive cells. The computations that integrate these diverse outputs to control body temperature are not understood. Here we discover a new set of Warming Cells (WCs), and show that the outputs of these WCs and previously described Cooling Cells (CCs1) are combined in a cross-inhibition computation to drive thermal homeostasis in larval Drosophila. We find that WCs and CCs are opponent sensors that operate in synchrony above, below, and near the homeostatic set-point, with WCs consistently activated by warming and inhibited by cooling, and CCs the converse. Molecularly, these opponent sensors rely on overlapping combinations of Ionotropic Receptors to detect temperature changes: Ir68a, Ir93a, and Ir25a for WCs; Ir21a, Ir93a, and Ir25a for CCs. Using a combination of optogenetics, sensory receptor mutants, and quantitative behavioral analysis, we find that the larva uses flexible cross-inhibition of WC and CC outputs to locate and stay near the homeostatic set-point. Balanced cross-inhibition near the set-point suppresses any directed movement along temperature gradients. Above the set-point, WCs mediate avoidance to warming while cross-inhibiting avoidance to cooling. Below the set-point, CCs mediate avoidance to cooling while cross-inhibiting avoidance to warming. Our results demonstrate how flexible cross-inhibition between warming and cooling pathways can orchestrate homeostatic thermoregulation.

The sense of smell enables animals to react to long-distance cues according to learned and innate valences. Here, we have mapped with electron microscopy the complete wiring diagram of the Drosophila larval antennal lobe, an olfactory neuropil similar to the vertebrate olfactory bulb. We found a canonical circuit with uniglomerular projection neurons (uPNs) relaying gain-controlled ORN activity to the mushroom body and the lateral horn. A second, parallel circuit with multiglomerular projection neurons (mPNs) and hierarchically connected local neurons (LNs) selectively integrates multiple ORN signals already at the first synapse. LN-LN synaptic connections putatively implement a bistable gain control mechanism that either computes odor saliency through panglomerular inhibition, or allows some glomeruli to respond to faint aversive odors in the presence of strong appetitive odors. This complete wiring diagram will support experimental and theoretical studies towards bridging the gap between circuits and behavior. DOI: http://dx.doi.org/10.7554/eLife.14859.001, eLife digest Our sense of smell can tell us about bread being baked faraway in the kitchen, or whether a leftover piece finally went bad. Similarly to the eyes, the nose enables us to make up a mental image of what lies at a distance. In mammals, the surface of the nose hosts a huge number of olfactory sensory cells, each of which is tuned to respond to a small set of scent molecules. The olfactory sensory cells communicate with a region of the brain called the olfactory bulb. Olfactory sensory cells of the same type converge onto the same small pocket of the olfactory bulb, forming a structure called a glomerulus. Similarly to how the retina generates an image, the combined activity of multiple glomeruli defines an odor. A particular smell is the combination of many volatile compounds, the odorants. Therefore the interactions between different olfactory glomeruli are important for defining the nature of the perceived odor. Although the types of neurons involved in these interactions were known in insects, fish and mice, a precise wiring diagram of a complete set of glomeruli had not been described. In particular, the points of contact through which neurons communicate with each other – known as synapses – among all the neurons participating in an olfactory system were not known. Berck, Khandelwal et al. have now taken advantage of the small size of the olfactory system of the larvae of Drosophila fruit flies to fully describe, using high-resolution imaging, all its neurons and their synapses. The results define the complete wiring diagram of the neural circuit that processes the signals sent by olfactory sensory neurons in the larva’s olfactory circuits. In addition to the neurons that read out the activity of a single glomerulus and send it to higher areas of the brain for further processing, there are also numerous neurons that read out activity from multiple glomeruli. These neurons represent a system, encoded in the genome, for quickly extracting valuable olfactory information and then relaying it to other areas of the brain. An essential aspect of sensation is the ability to stop noticing a stimulus if it doesn't change. This allows an animal to, for example, find food by moving in a direction that increases the intensity of an odor. Inhibition mediates some aspects of this capability. The discovery of structure in the inhibitory connections among glomeruli, together with prior findings on the inner workings of the olfactory system, enabled Berck, Khandelwal et al. to hypothesize how the olfactory circuits enable odor gradients to be navigated. Further investigation revealed more about how the circuits could detect slight changes in odor concentration regardless of whether the overall odor intensity is strong or faint. And, crucially, it revealed how the worst odors – which can signal danger – can still be perceived in the presence of very strong pleasant odors. With the wiring diagram, theories about the sense of smell can now be tested using the genetic tools available for Drosophila, leading to an understanding of how neural circuits work. DOI: http://dx.doi.org/10.7554/eLife.14859.002

Neural circuits for behavior transform sensory inputs into motor outputs in patterns with strategic value. Determining how neurons along a sensorimotor circuit contribute to this transformation is central to understanding behavior. To do this, a quantitative framework to describe behavioral dynamics is needed. In this study, we built a high-throughput optogenetic system for Drosophila larva to quantify the sensorimotor transformations underlying navigational behavior. We express CsChrimson, a red-shifted variant of channelrhodopsin, in specific chemosensory neurons and expose large numbers of freely moving animals to random optogenetic activation patterns. We quantify their behavioral responses and use reverse-correlation analysis to uncover the linear and static nonlinear components of navigation dynamics as functions of optogenetic activation patterns of specific sensory neurons. We find that linear–nonlinear models accurately predict navigational decision-making for different optogenetic activation waveforms. We use our method to establish the valence and dynamics of navigation driven by optogenetic activation of different combinations of bitter-sensing gustatory neurons. Our method captures the dynamics of optogenetically induced behavior in compact, quantitative transformations that can be used to characterize circuits for sensorimotor processing and their contribution to navigational decision making. DOI: http://dx.doi.org/10.7554/eLife.06225.001, eLife digest Living organisms can sense their surroundings and respond in appropriate ways. For example, animals will often move towards the smell of food or away from potential threats, such as predators. However, it is not fully understood how an animal's nervous system is setup to allow sensory information to control how the animal navigates its environment. Optogenetics is a technique that allows neuroscientists to control the activities of individual nerve cells in freely moving animals, simply by shining light on to them. Here, Hernandez-Nunez et al. have used optogenetics in fruit fly larvae to activate nerve cells that normally respond to smells and tastes, while the larvae's movements were tracked. Fruit fly larvae were chosen because they have a simple, but well-studied, nervous system. These larvae also move in two distinct ways: ‘runs’, in which a larva moves forward; and ‘turns’, during which a larva sweeps its head back and forth until it selects the direction of a new run. The data from these experiments were quantified using a specific type of statistical analysis called ‘reverse correlation’ and used to build mathematical models that predict navigational behavior. This analysis of the experiments allowed Hernandez-Nunez et al. to reveal how specific sensory nerve cells can contribute to pathways that control an animal's navigation—and an independent study by Gepner, Mihovilovic Skanata et al. revealed similar results. The approach of using optogenetics in combination with quantitative analysis, as used in these two independent studies, is now opening the door to a more complete understanding of the connections between the activity of sensory nerve cells and perception and behavior. DOI: http://dx.doi.org/10.7554/eLife.06225.002

Complex animal behaviors are built from dynamical relationships between sensory inputs, neuronal activity, and motor outputs in patterns with strategic value. Connecting these patterns illuminates how nervous systems compute behavior. Here, we study Drosophila larva navigation up temperature gradients toward preferred temperatures (positive thermotaxis). By tracking the movements of animals responding to fixed spatial temperature gradients or random temperature fluctuations, we calculate the sensitivity and dynamics of the conversion of thermosensory inputs into motor responses. We discover three thermosensory neurons in each dorsal organ ganglion (DOG) that are required for positive thermotaxis. Random optogenetic stimulation of the DOG thermosensory neurons evokes behavioral patterns that mimic the response to temperature variations. In vivo calcium and voltage imaging reveals that the DOG thermosensory neurons exhibit activity patterns with sensitivity and dynamics matched to the behavioral response. Temporal processing of temperature variations carried out by the DOG thermosensory neurons emerges in distinct motor responses during thermotaxis.

Bacterial chemotaxis systems are as diverse as the environments that bacteria inhabit, but how much environmental variation can cells tolerate with a single system? Diversification of a single chemotaxis system could serve as an alternative, or even evolutionary stepping-stone, to switching between multiple systems. We hypothesized that mutations in gene regulation could lead to heritable control of chemotactic diversity. By simulating foraging and colonization of E. coli using a single-cell chemotaxis model, we found that different environments selected for different behaviors. The resulting trade-offs show that populations facing diverse environments would ideally diversify behaviors when time for navigation is limited. We show that advantageous diversity can arise from changes in the distribution of protein levels among individuals, which could occur through mutations in gene regulation. We propose experiments to test our prediction that chemotactic diversity in a clonal population could be a selectable trait that enables adaptation to environmental variability. DOI: http://dx.doi.org/10.7554/eLife.03526.001, eLife digest Bacterial colonies are generally made up of genetically identical cells. Despite this, a closer look at the members of a bacterial colony shows that these cells can have very different behaviors. For example, some cells may grow more quickly than others, or be more resistant to antibiotics. The mechanisms driving this diversity are only beginning to be identified and understood. Escherichia coli bacteria can move towards, or away from, certain chemicals in their surrounding environment to help them navigate toward favorable conditions. This behavior is known as chemotaxis. The signals from all of these chemicals are processed in E. coli by just one set of proteins, which control the different behaviors that are needed for the bacteria to follow them. Different numbers of these proteins are found in different—but genetically identical—bacteria, and the number of proteins is linked to how the bacteria perform these behaviors. It has been suggested that diversity can be beneficial to the overall bacterial population, as it helps the population survive environmental changes. This suggests that the level of diversity in the population should adapt to the level of diversity in the environment. However, it remains unknown how this adaptation occurs. Frankel et al. developed and combined several models and simulations to investigate whether differences in chemotaxis protein production help an E. coli colony to survive. The models show that in different environments, it can be beneficial for the population as a whole if different cells have different responses to the chemicals present. For example, if a lot of a useful chemical is present, bacteria are more likely to survive by heading straight to the source. If not much chemical is detected, the bacteria may need to move in a more exploratory manner. Frankel et al. find that different amounts of chemotaxis proteins produce these different behaviors. To survive in a changing environment, it is therefore best for the E. coli colony to contain cells that have different amounts of these proteins. Frankel et al. propose that the variability of chemotaxis protein levels between genetically identical cells can change through mutations in the genes that control how many of the proteins are produced, and predict that such mutations allow populations to adapt to environmental changes. The environments simulated in the model were much simpler than would be found in the real world, and Frankel et al. describe experiments that are now being performed to confirm and expand on their results. The model could be used in the future to shed light on the behavior of other cells that are genetically identical but exhibit diverse behaviors, from other bacterial species to more complex cancer cells. DOI: http://dx.doi.org/10.7554/eLife.03526.002

Inputs to signaling pathways can have complex statistics that depend on the environment and on the behavioral response to previous stimuli. Such behavioral feedback is particularly important in navigation. Successful navigation relies on proper coupling between sensors, which gather information during motion, and actuators, which control behavior. Because reorientation conditions future inputs, behavioral feedback can place sensors and actuators in an operational regime different from the resting state. How then can organisms maintain proper information transfer through the pathway while navigating diverse environments? In bacterial chemotaxis, robust performance is often attributed to the zero integral feedback control of the sensor, which guarantees that activity returns to resting state when the input remains constant. While this property provides sensitivity over a wide range of signal intensities, it remains unclear how other parameters such as adaptation rate and adapted activity affect chemotactic performance, especially when considering that the swimming behavior of the cell determines the input signal. We examine this issue using analytical models and simulations that incorporate recent experimental evidences about behavioral feedback and flagellar motor adaptation. By focusing on how sensory information carried by the response regulator is best utilized by the motor, we identify an operational regime that maximizes drift velocity along chemical concentration gradients for a wide range of environments and sensor adaptation rates. This optimal regime is outside the dynamic range of the motor response, but maximizes the contrast between run duration up and down gradients. In steep gradients, the feedback from chemotactic drift can push the system through a bifurcation. This creates a non-chemotactic state that traps cells unless the motor is allowed to adapt. Although motor adaptation helps, we find that as the strength of the feedback increases individual phenotypes cannot maintain the optimal operational regime in all environments, suggesting that diversity could be beneficial., Author Summary The biased random walk is a fundamental strategy used by many organisms to navigate their environment. Drift along the desired direction is achieved by reducing the probability to reorient whenever conditions improve. In the chemotaxis system of Escherichia coli, this is accomplished with a sensory module that implements negative integral feedback control, the output of which is relayed to the flagellar motors (the actuators) by a response regulator to control the probability to change direction. The proper dynamical coupling between sensor and actuator is critical for the performance of the random walker. Here, we identify an optimal regime for this coupling that maximizes drift velocity in the direction of the gradient in multiple environments. Our analysis reveals that feedback of the behavior onto the system in steep gradients can constrain individual cell performance, by causing bi-stable behavior that can trap cells in non-chemotactic states. These limitations are inherent in the biased random walk strategy with integral feedback control, but can be alleviated if the output of the pathway adapts, as recently characterized for the flagellar motors in Escherichia coli.