Structured learning provides a powerful framework for empirical risk minimization on the predictions of structured models. It allows end-to-end learning of model parameters to minimize an application specific loss function. This framework is particularly well suited for discrete optimization models that are used for neuron reconstruction from anisotropic electron microscopy (EM) volumes. However, current methods are still learning unary potentials by training a classifier that is agnostic about the model it is used in. We believe the reason for that lies in the difficulties of (1) finding a representative training sample, and (2) designing an application specific loss function that captures the quality of a proposed solution. In this paper, we show how to find a representative training sample from human generated ground truth, and propose a loss function that is suitable to minimize topological errors in the reconstruction. We compare different training methods on two challenging EM-datasets. Our structured learning approach shows consistently higher reconstruction accuracy than other current learning methods.

Anatomical, molecular, and physiological interactions between astrocytes and neuronal synapses regulate information processing in the brain. The fruit fly Drosophila melanogaster has become a valuable experimental system for genetic manipulation of the nervous system and has enormous potential for elucidating mechanisms that mediate neuron-glia interactions. Here, we show the first electrophysiological recordings from Drosophila astrocytes and characterize their spatial and physiological relationship with particular synapses. Astrocyte intrinsic properties were found to be strongly analogous to those of vertebrate astrocytes, including a passive current-voltage relationship, low membrane resistance, high capacitance, and dye-coupling to local astrocytes. Responses to optogenetic stimulation of glutamatergic pre-motor neurons were correlated directly with anatomy using serial electron microscopy reconstructions of homologous identified neurons and surrounding astrocytic processes. Robust bidirectional communication was present: neuronal activation triggered astrocytic glutamate transport via Eaat1, and blocking Eaat1 extended glutamatergic interneuron-evoked inhibitory post-synaptic currents in motor neurons. The neuronal synapses were always located within a micron of an astrocytic process, but none were ensheathed by those processes. Thus, fly astrocytes can modulate fast synaptic transmission via neurotransmitter transport within these anatomical parameters. This article is protected by copyright. All rights reserved.

Neuronal circuit mapping using electron microscopy demands laborious proofreading or reconciliation of multiple independent reconstructions. Here, we describe new methods to apply quantitative arbor and network context to iteratively proofread and reconstruct circuits and create anatomically enriched wiring diagrams. We measured the morphological underpinnings of connectivity in new and existing reconstructions of Drosophila sensorimotor (larva) and visual (adult) systems. Synaptic inputs were preferentially located on numerous small, microtubule-free 'twigs' which branch off a single microtubule-containing 'backbone'. Omission of individual twigs accounted for 96% of errors. However, the synapses of highly connected neurons were distributed across multiple twigs. Thus, the robustness of a strong connection to detailed twig anatomy was associated with robustness to reconstruction error. By comparing iterative reconstruction to the consensus of multiple reconstructions, we show that our method overcomes the need for redundant effort through the discovery and application of relationships between cellular neuroanatomy and synaptic connectivity.

Animals move by adaptively coordinating the sequential activation of muscles. The circuit mechanisms underlying coordinated locomotion are poorly understood. Here, we report on a novel circuit for propagation of waves of muscle contraction, using the peristaltic locomotion of Drosophila larvae as a model system. We found an intersegmental chain of synaptically connected neurons, alternating excitatory and inhibitory, necessary for wave propagation and active in phase with the wave. The excitatory neurons (A27h) are premotor and necessary only for forward locomotion, and are modulated by stretch receptors and descending inputs. The inhibitory neurons (GDL) are necessary for both forward and backward locomotion, suggestive of different yet coupled central pattern generators, and its inhibition is necessary for wave propagation. The circuit structure and functional imaging indicated that the commands to contract one segment promote the relaxation of the next segment, revealing a mechanism for wave propagation in peristaltic locomotion.

Automated reconstruction of neural connectivity graphs from electron microscopy image stacks is an essential step towards large-scale neural circuit mapping. While significant progress has recently been made in automated segmentation of neurons and detection of synapses, the problem of synaptic partner assignment for polyadic (one-to-many) synapses, prevalent in the Drosophila brain, remains unsolved. In this contribution, we propose a method which automatically assigns pre- and postsynaptic roles to neurites adjacent to a synaptic site. The method constructs a probabilistic graphical model over potential synaptic partner pairs which includes factors to account for a high rate of one-to-many connections, as well as the possibility of the same neuron to be pre-synaptic in one synapse and post-synaptic in another. The algorithm has been validated on a publicly available stack of ssTEM images of Drosophila neural tissue and has been shown to reconstruct most of the synaptic relations correctly.

To stimulate progress in automating the reconstruction of neural circuits, we organized the first international challenge on 2D segmentation of electron microscopic (EM) images of the brain. Participants submitted boundary maps predicted for a test set of images, and were scored based on their agreement with a consensus of human expert annotations. The winning team had no prior experience with EM images, and employed a convolutional network. This “deep learning” approach has since become accepted as a standard for segmentation of EM images. The challenge has continued to accept submissions, and the best so far has resulted from cooperation between two teams. The challenge has probably saturated, as algorithms cannot progress beyond limits set by ambiguities inherent in 2D scoring and the size of the test dataset. Retrospective evaluation of the challenge scoring system reveals that it was not sufficiently robust to variations in the widths of neurite borders. We propose a solution to this problem, which should be useful for a future 3D segmentation challenge.

Bilaterally symmetric motor patterns-those in which left-right pairs of muscles contract synchronously and with equal amplitude (such as breathing, smiling, whisking, and locomotion)-are widespread throughout the animal kingdom. Yet, surprisingly little is known about the underlying neural circuits. We performed a thermogenetic screen to identify neurons required for bilaterally symmetric locomotion in Drosophila larvae and identified the evolutionarily conserved Even-skipped(+) interneurons (Eve/Evx). Activation or ablation of Eve(+) interneurons disrupted bilaterally symmetric muscle contraction amplitude, without affecting the timing of motor output. Eve(+) interneurons are not rhythmically active and thus function independently of the locomotor CPG. GCaMP6 calcium imaging of Eve(+) interneurons in freely moving larvae showed left-right asymmetric activation that correlated with larval behavior. TEM reconstruction of Eve(+) interneuron inputs and outputs showed that the Eve(+) interneurons are at the core of a sensorimotor circuit capable of detecting and modifying body wall muscle contraction.

Natural events present multiple types of sensory cues, each detected by a specialized sensory modality. Combining information from several modalities is essential for the selection of appropriate actions. Key to understanding multimodal computations is determining the structural patterns of multimodal convergence and how these patterns contribute to behaviour. Modalities could converge early, late or at multiple levels in the sensory processing hierarchy. Here we show that combining mechanosensory and nociceptive cues synergistically enhances the selection of the fastest mode of escape locomotion in Drosophila larvae. In an electron microscopy volume that spans the entire insect nervous system, we reconstructed the multisensory circuit supporting the synergy, spanning multiple levels of the sensory processing hierarchy. The wiring diagram revealed a complex multilevel multimodal convergence architecture. Using behavioural and physiological studies, we identified functionally connected circuit nodes that trigger the fastest locomotor mode, and others that facilitate it, and we provide evidence that multiple levels of multimodal integration contribute to escape mode selection. We propose that the multilevel multimodal convergence architecture may be a general feature of multisensory circuits enabling complex input–output functions and selective tuning to ecologically relevant combinations of cues.

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.

The generation of four-dimensional (4D) confocal datasets; consisting of 3D image sequences over time; provides an excellent methodology to capture cellular behaviors involved in developmental processes.  The ability to track and follow cell movements is limited by sample movements that occur due to drift of the sample or, in some cases, growth during image acquisition. Tracking cells in datasets affected by drift and/or growth will incorporate these movements into any analysis of cell position. This may result in the apparent movement of static structures within the sample. Therefore prior to cell tracking, any sample drift should be corrected. Using the open source Fiji distribution (1)  of ImageJ (2,3) and the incorporated LOCI tools (4), we developed the Correct 3D drift plug-in to remove erroneous sample movement in confocal datasets. This protocol effectively compensates for sample translation or alterations in focal position by utilizing phase correlation to register each time-point of a four-dimensional confocal datasets while maintaining the ability to visualize and measure cell movements over extended time-lapse experiments.