Automated synapse-level reconstruction of neural circuits in the larval zebrafish brain.

Dense reconstruction of synaptic connectivity requires high-resolution electron microscopy images of entire brains and tools to efficiently trace neuronal wires across the volume. To generate such a resource, we sectioned and imaged a larval zebrafish brain by serial block-face electron microscopy at a voxel size of 14 × 14 × 25 nm3. We segmented the resulting dataset with the flood-filling network algorithm, automated the detection of chemical synapses and validated the results by comparisons to transmission electron microscopic images and light-microscopic reconstructions. Neurons and their connections are stored in the form of a queryable and expandable digital address book. We reconstructed a network of 208 neurons involved in visual motion processing, most of them located in the pretectum, which had been functionally characterized in the same specimen by two-photon calcium imaging. Moreover, we mapped all 407 presynaptic and postsynaptic partners of two superficial interneurons in the tectum. The resource developed here serves as a foundation for synaptic-resolution circuit analyses in the zebrafish nervous system.

The Mind of a Mouse.

Large scientific projects in genomics and astronomy are influential not because they answer any single question but because they enable investigation of continuously arising new questions from the same data-rich sources. Advances in automated mapping of the brain's synaptic connections (connectomics) suggest that the complicated circuits underlying brain function are ripe for analysis. We discuss benefits of mapping a mouse brain at the level of synapses.

Full reconstruction of large lobula plate tangential cells in Drosophila from a 3D EM dataset.

With the advent of neurogenetic methods, the neural basis of behavior is presently being analyzed in more and more detail. This is particularly true for visually driven behavior of Drosophila melanogaster where cell-specific driver lines exist that, depending on the combination with appropriate effector genes, allow for targeted recording, silencing and optogenetic stimulation of individual cell-types. Together with detailed connectomic data of large parts of the fly optic lobe, this has recently led to much progress in our understanding of the neural circuits underlying local motion detection. However, how such local information is combined by optic flow sensitive large-field neurons is still incompletely understood. Here, we aim to fill this gap by a dense reconstruction of lobula plate tangential cells of the fly lobula plate. These neurons collect input from many hundreds of local motion-sensing T4/T5 neurons and connect them to descending neurons or central brain areas. We confirm all basic features of HS and VS cells as published previously from light microscopy. In addition, we identified the dorsal and the ventral centrifugal horizontal, dCH and vCH cell, as well as three VSlike cells, including their distinct dendritic and axonal projection area.

High-precision automated reconstruction of neurons with flood-filling networks.

Reconstruction of neural circuits from volume electron microscopy data requires the tracing of cells in their entirety, including all their neurites. Automated approaches have been developed for tracing, but their error rates are too high to generate reliable circuit diagrams without extensive human proofreading. We present flood-filling networks, a method for automated segmentation that, similar to most previous efforts, uses convolutional neural networks, but contains in addition a recurrent pathway that allows the iterative optimization and extension of individual neuronal processes. We used flood-filling networks to trace neurons in a dataset obtained by serial block-face electron microscopy of a zebra finch brain. Using our method, we achieved a mean error-free neurite path length of 1.1 mm, and we observed only four mergers in a test set with a path length of 97 mm. The performance of flood-filling networks was an order of magnitude better than that of previous approaches applied to this dataset, although with substantially increased computational costs.

Volume EM Reconstruction of Spinal Cord Reveals Wiring Specificity in Speed-Related Motor Circuits.

Spinal interneurons coordinate the activity of motoneurons to generate the spatiotemporal patterns of muscle contractions required for vertebrate locomotion. It is controversial to what degree the orderly, gradual recruitment of motoneurons is determined by biophysical differences among them rather than by specific connections from presynaptic interneurons to subsets of motoneurons. To answer this question, we mapped all connections from two types of interneurons onto all motoneurons in a larval zebrafish spinal cord hemisegment, using serial block-face electron microscopy (SBEM). We found specific synaptic connectivity from dorsal but not from ventral excitatory ipsilateral interneurons, with large motoneurons, active only when strong force is required, receiving specific inputs from dorsally located interneurons, active only during fast swims. By contrast, the connectivity between inhibitory commissural interneurons and motoneurons lacks any discernible pattern. The wiring pattern is consistent with a recruitment mechanism that depends to a considerable extent on specific connectivity.

Progress and remaining challenges in high-throughput volume electron microscopy.

Recent advances in the effectiveness of the automatic extraction of neural circuits from volume electron microscopy data have made us more optimistic that the goal of reconstructing the nervous system of an entire adult mammal (or bird) brain can be achieved in the next decade. The progress on the data analysis side-based mostly on variants of convolutional neural networks-has been particularly impressive, but improvements in the quality and spatial extent of published VEM datasets are substantial. Methodologically, the combination of hot-knife sample partitioning and ion milling stands out as a conceptual advance while the multi-beam scanning electron microscope promises to remove the data-acquisition bottleneck.

Single-protein detection in crowded molecular environments in cryo-EM images.

We present an approach to study macromolecular assemblies by detecting component proteins' characteristic high-resolution projection patterns, calculated from their known 3D structures, in single electron cryo-micrographs. Our method detects single apoferritin molecules in vitreous ice with high specificity and determines their orientation and location precisely. Simulations show that high spatial-frequency information and-in the presence of protein background-a whitening filter are essential for optimal detection, in particular for images taken far from focus. Experimentally, we could detect small viral RNA polymerase molecules, distributed randomly among binding locations, inside rotavirus particles. Based on the currently attainable image quality, we estimate a threshold for detection that is 150 kDa in ice and 300 kDa in 100 nm thick samples of dense biological material.

EM connectomics reveals axonal target variation in a sequence-generating network.

The sequential activation of neurons has been observed in various areas of the brain, but in no case is the underlying network structure well understood. Here we examined the circuit anatomy of zebra finch HVC, a cortical region that generates sequences underlying the temporal progression of the song. We combined serial block-face electron microscopy with light microscopy to determine the cell types targeted by HVC(RA) neurons, which control song timing. Close to their soma, axons almost exclusively targeted inhibitory interneurons, consistent with what had been found with electrical recordings from pairs of cells. Conversely, far from the soma the targets were mostly other excitatory neurons, about half of these being other HVC(RA) cells. Both observations are consistent with the notion that the neural sequences that pace the song are generated by global synaptic chains in HVC embedded within local inhibitory networks.

High-resolution whole-brain staining for electron microscopic circuit reconstruction.

Currently only electron microscopy provides the resolution necessary to reconstruct neuronal circuits completely and with single-synapse resolution. Because almost all behaviors rely on neural computations widely distributed throughout the brain, a reconstruction of brain-wide circuits-and, ultimately, the entire brain-is highly desirable. However, these reconstructions require the undivided brain to be prepared for electron microscopic observation. Here we describe a preparation, BROPA (brain-wide reduced-osmium staining with pyrogallol-mediated amplification), that results in the preservation and staining of ultrastructural details throughout the brain at a resolution necessary for tracing neuronal processes and identifying synaptic contacts between them. Using serial block-face electron microscopy (SBEM), we tested human annotator ability to follow neural 'wires' reliably and over long distances as well as the ability to detect synaptic contacts. Our results suggest that the BROPA method can produce a preparation suitable for the reconstruction of neural circuits spanning an entire mouse brain.

Space-time wiring specificity supports direction selectivity in the retina.

How does the mammalian retina detect motion? This classic problem in visual neuroscience has remained unsolved for 50 years. In search of clues, here we reconstruct Off-type starburst amacrine cells (SACs) and bipolar cells (BCs) in serial electron microscopic images with help from EyeWire, an online community of 'citizen neuroscientists'. On the basis of quantitative analyses of contact area and branch depth in the retina, we find evidence that one BC type prefers to wire with a SAC dendrite near the SAC soma, whereas another BC type prefers to wire far from the soma. The near type is known to lag the far type in time of visual response. A mathematical model shows how such 'space-time wiring specificity' could endow SAC dendrites with receptive fields that are oriented in space-time and therefore respond selectively to stimuli that move in the outward direction from the soma.

Miniaturization of two-photon microscopy for imaging in freely moving animals.

This article describes the development and application of miniaturized two-photon-excited fluorescence microscopes ("two-photon fiberscopes"). Two-photon fiberscopes have been developed with the aim of enabling high-resolution imaging of neural activity in freely behaving animals. They use fiber optics to deliver laser light for two-photon excitation. Their small front piece typically contains a miniature scanning mechanism and imaging optics. Two-photon fiberscopes can be made sufficiently small and lightweight to be carried by rats and mice and to allow virtually unrestricted movement within a behavioral arena. Typically mounted to the animal's skull above a cranial window, two-photon fiberscopes permit imaging of cells down to at least 250 µm below the brain surface (e.g., in rat neocortex). In freely exploring animals, action-potential-evoked calcium transients can be imaged in individual somata of visual cortex neurons bulk-labeled with a calcium indicator. Two-photon fiberscopes thus enable high-resolution optical recording of neural activity with cellular resolution during natural behaviors.

Connectomic reconstruction of the inner plexiform layer in the mouse retina.

Comprehensive high-resolution structural maps are central to functional exploration and understanding in biology. For the nervous system, in which high resolution and large spatial extent are both needed, such maps are scarce as they challenge data acquisition and analysis capabilities. Here we present for the mouse inner plexiform layer--the main computational neuropil region in the mammalian retina--the dense reconstruction of 950 neurons and their mutual contacts. This was achieved by applying a combination of crowd-sourced manual annotation and machine-learning-based volume segmentation to serial block-face electron microscopy data. We characterize a new type of retinal bipolar interneuron and show that we can subdivide a known type based on connectivity. Circuit motifs that emerge from our data indicate a functional mechanism for a known cellular response in a ganglion cell that detects localized motion, and predict that another ganglion cell is motion sensitive.

Low-dosage Maximum-A-Posteriori Focusing and Stigmation.

Radiation damage is often an issue during high-resolution imaging, making low-dose focusing and stigmation essential, in particular when no part of the sample can be "sacrificed" for this. An example is serial block-face electron microscopy, where the imaging resolution must be kept optimal during automated acquisition that can last months. Here, we present an algorithm, which we call "Maximum-A-Posteriori Focusing and Stigmation (MAPFoSt)," that was designed to make optimal use of the available signal. We show that MAPFoSt outperforms the built-in focusing algorithm of a commercial scanning electron microscope even at a tenfold reduced total dose. MAPFoSt estimates multiple aberration modes (focus and the two astigmatism coefficients) using just two test images taken at different focus settings. Using an incident electron dose density of 2,500 electrons/pixel and a signal-to-noise ratio of about one, all three coefficients could be estimated to within <7% of the depth of focus, using 19 detected secondary electrons per pixel. A generalization to higher-order aberrations and to other forms of imaging in both two and three dimensions appears possible.

Learning to segment neurons with non-local quality measures.

Segmentation schemes such as hierarchical region merging or correllation clustering rely on edge weights between adjacent (super-)voxels. The quality of these edge weights directly affects the quality of the resulting segmentations. Unstructured learning methods seek to minimize the classification error on individual edges. This ignores that a few local mistakes (tiny boundary gaps) can cause catastrophic global segmentation errors. Boundary evidence learning should therefore optimize structured quality criteria such as Rand Error or Variation of Information. We present the first structured learning scheme using a structured loss function; and we introduce a new hierarchical scheme that allows to approximately solve the NP hard prediction problem even for huge volume images. The value of these contributions is demonstrated on two challenging neural circuit reconstruction problems in serial sectioning electron microscopic images with billions of voxels. Our contributions lead to a partitioning quality that improves over the current state of the art.

Staining and embedding the whole mouse brain for electron microscopy.

The development of methods for imaging large contiguous volumes with the electron microscope could allow the complete mapping of a whole mouse brain at the single-axon level. We developed a method based on prolonged immersion that enables staining and embedding of the entire mouse brain with uniform myelin staining and a moderate preservation of the tissue's ultrastructure. We tested the ability to follow myelinated axons using serial block-face electron microscopy.

3D segmentation of SBFSEM images of neuropil by a graphical model over supervoxel boundaries.

The segmentation of large volume images of neuropil acquired by serial sectioning electron microscopy is an important step toward the 3D reconstruction of neural circuits. The only cue provided by the data at hand is boundaries between otherwise indistinguishable objects. This indistinguishability, combined with the boundaries becoming very thin or faint in places, makes the large body of work on region-based segmentation methods inapplicable. On the other hand, boundary-based methods that exploit purely local evidence do not reach the extremely high accuracy required by the application domain that cannot tolerate the global topological errors arising from false local decisions. As a consequence, we propose a supervoxel merging method that arrives at its decisions in a non-local fashion, by posing and approximately solving a joint combinatorial optimization problem over all faces between supervoxels. The use of supervoxels allows the extraction of expressive geometric features. These are used by the higher-order potentials in a graphical model that assimilate knowledge about the geometry of neural surfaces by automated training on a gold standard. The scope of this improvement is demonstrated on the benchmark dataset E1088 (Helmstaedter et al., 2011) of 7.5billionvoxels from the inner plexiform layer of rabbit retina. We provide C++ source code for annotation, geometry extraction, training and inference.

Structural neurobiology: missing link to a mechanistic understanding of neural computation.

High-resolution, comprehensive structural information is often the final arbiter between competing mechanistic models of biological processes, and can serve as inspiration for new hypotheses. In molecular biology, definitive structural data at atomic resolution are available for many macromolecules; however, information about the structure of the brain is much less complete, both in scope and resolution. Several technical developments over the past decade, such as serial block-face electron microscopy and trans-synaptic viral tracing, have made the structural biology of neural circuits conceivable: we may be able to obtain the structural information needed to reconstruct the network of cellular connections for large parts of, or even an entire, mouse brain within a decade or so. Given that the brain's algorithms are ultimately encoded by this network, knowing where all of these connections are should, at the very least, provide the data needed to distinguish between models of neural computation.