Fluorescent calcium sensors are widely used to image neural activity. Using structure-based mutagenesis and neuron-based screening, we developed a family of ultrasensitive protein calcium sensors (GCaMP6) that outperformed other sensors in cultured neurons and in zebrafish, flies and mice in vivo. In layer 2/3 pyramidal neurons of the mouse visual cortex, GCaMP6 reliably detected single action potentials in neuronal somata and orientation-tuned synaptic calcium transients in individual dendritic spines. The orientation tuning of structurally persistent spines was largely stable over timescales of weeks. Orientation tuning averaged across spine populations predicted the tuning of their parent cell. Although the somata of GABAergic neurons showed little orientation tuning, their dendrites included highly tuned dendritic segments (5–40-µm long). GCaMP6 sensors thus provide new windows into the organization and dynamics of neural circuits over multiple spatial and temporal scales.

Complete reconstructions of vertebrate neuronal circuits on the synaptic level require new approaches. Here, serial section transmission electron microscopy was automated to densely reconstruct four volumes, totaling 670 μm(3), from the rat hippocampus as proving grounds to determine when axo-dendritic proximities predict synapses. First, in contrast with Peters’ rule, the density of axons within reach of dendritic spines did not predict synaptic density along dendrites because the fraction of axons making synapses was variable. Second, an axo-dendritic touch did not predict a synapse; nevertheless, the density of synapses along a hippocampal dendrite appeared to be a universal fraction, 0.2, of the density of touches. Finally, the largest touch between an axonal bouton and spine indicated the site of actual synapses with about 80% precision but would miss about half of all synapses. Thus, it will be difficult to predict synaptic connectivity using data sets missing ultrastructural details that distinguish between axo-dendritic touches and bona fide synapses.

A primary cilium is a membrane-bound extension from the cell surface that contains receptors for perceiving and transmitting signals that modulate cell state and activity. Primary cilia in the brain are less accessible than cilia on cultured cells or epithelial tissues because in the brain they protrude into a deep, dense network of glial and neuronal processes. Here, we investigated cilia frequency, internal structure, shape, and position in large, high-resolution transmission electron microscopy volumes of mouse primary visual cortex. Cilia extended from the cell bodies of nearly all excitatory and inhibitory neurons, astrocytes, and oligodendrocyte precursor cells (OPCs) but were absent from oligodendrocytes and microglia. Ultrastructural comparisons revealed that the base of the cilium and the microtubule organization differed between neurons and glia. Investigating cilia-proximal features revealed that many cilia were directly adjacent to synapses, suggesting that cilia are poised to encounter locally released signaling molecules. Our analysis indicated that synapse proximity is likely due to random encounters in the neuropil, with no evidence that cilia modulate synapse activity as would be expected in tetrapartite synapses. The observed cell class differences in proximity to synapses were largely due to differences in external cilia length. Many key structural features that differed between neuronal and glial cilia influenced both cilium placement and shape and, thus, exposure to processes and synapses outside the cilium. Together, the ultrastructure both within and around neuronal and glial cilia suggest differences in cilia formation and function across cell types in the brain.

The human genome is extensively folded into 3-dimensional organization. However, the detailed 3D chromatin folding structures have not been fully visualized due to the lack of robust and ultra-resolution imaging capability. Here, we report the development of an electron microscopy method that combines serial block-face scanning electron microscopy with in situ hybridization (3D-EMISH) to visualize 3D chromatin folding at targeted genomic regions with ultra-resolution (5 × 5 × 30 nm in xyz dimensions) that is superior to the current super-resolution by fluorescence light microscopy. We apply 3D-EMISH to human lymphoblastoid cells at a 1.7 Mb segment of the genome and visualize a large number of distinctive 3D chromatin folding structures in ultra-resolution. We further quantitatively characterize the reconstituted chromatin folding structures by identifying sub-domains, and uncover a high level heterogeneity of chromatin folding ultrastructures in individual nuclei, suggestive of extensive dynamic fluidity in 3D chromatin states.

Focused-ion-beam scanning electron microscopy (FIB-SEM) has become an essential tool for studying neural tissue at resolutions below 10 nm × 10 nm × 10 nm, producing data sets optimized for automatic connectome tracing. We present a technical advance, ultrathick sectioning, which reliably subdivides embedded tissue samples into chunks (20 μm thick) optimally sized and mounted for efficient, parallel FIB-SEM imaging. These chunks are imaged separately and then 'volume stitched' back together, producing a final three-dimensional data set suitable for connectome tracing.

Brain networks that mediate motivated behavior in the context of aversive and rewarding experiences involve the prefrontal cortex (PFC) and ventral tegmental area (VTA). Neurons in both regions are activated by stress and reward, and by learned cues that predict aversive or appetitive outcomes. Recent studies have proposed that separate neuronal populations and circuits in these regions encode learned aversive versus appetitive contexts. But how about the actual experience? Do the same or different PFC and VTA neurons encode unanticipated aversive and appetitive experiences? To address this, we recorded unit activity and local field potentials (LFP) in the dorsomedial PFC (dmPFC) and VTA of male rats as they were exposed, in the same recording session, to reward (sucrose) or stress (tail pinch) spaced one hour apart. As expected, experience-specific neuronal responses were observed. About 15-25% of single units in each region responded by excitation or inhibition to either stress or reward, and only stress increased LFP theta oscillation power in both regions and coherence between regions. But the largest number of responses (29% dmPFC and 30% VTA units) involved dual-valence neurons that responded to both stress and reward exposure. Moreover, the temporal profile of neuronal population activity in dmPFC and VTA as assessed by principal component analysis were similar during both types of experiences. These results reveal that aversive and rewarding experiences engage overlapping neuronal populations in the dmPFC and the VTA. These populations may provide a locus of vulnerability for stress related disorders, which are often associated with anhedonia. Animals must recognize unexpected harmful and rewarding events in order to survive. How the brain represents these competing experiences is not fully understood. Two interconnected brain regions implicated in encoding both rewarding and stressful events are the dmPFC and the VTA. In either region, separate neurons and associated circuitry are assumed to respond to events with positive or negative valence. We find, however, that a significant subpopulation of neurons in dmPFC and VTA encode both rewarding and aversive experiences. These dual-valence neurons may provide a computational advantage for flexible planning of behavior when organisms face unexpected rewarding and harmful experiences.

The RNA genome of retroviruses is encased within a protein capsid. To gather insight into the assembly and function of this capsid, we used electron cryotomography to image human immunodeficiency virus (HIV) and equine infectious anemia virus (EIAV) particles. While the majority of viral cores appeared closed, a variety of unclosed structures including rolled sheets, extra flaps, and cores with holes in the tip were also seen. Simulations of nonequilibrium growth of elastic sheets recapitulated each of these aberrations and further predicted the occasional presence of seams, for which tentative evidence was also found within the cryotomograms. To test the integrity of viral capsids in vivo, we observed that  25% of cytoplasmic HIV complexes captured by TRIM5α had holes large enough to allow internal green fluorescent protein (GFP) molecules to escape. Together, these findings suggest that HIV assembly at least sometimes involves the union in space of two edges of a curling sheet and results in a substantial number of unclosed forms.

Modern supervised learning algorithms can learn very accurate and complex discriminating functions. But when these classifiers fail, this complexity can also be a drawback because there is no easy, intuitive way to diagnose why they are failing and remedy the problem. This important question has received little attention. To address this problem, we propose a novel method to analyze and understand a classifier's errors. Our method centers around a measure of how much influence a training example has on the classifier's prediction for a test example. To understand why a classifier is mispredicting the label of a given test example, the user can find and review the most influential training examples that caused this misprediction, allowing them to focus their attention on relevant areas of the data space. This will aid the user in determining if and how the training data is inconsistently labeled or lacking in diversity, or if the feature representation is insufficient. As computing the influence of each training example is computationally impractical, we propose a novel distance metric to approximate influence for boosting classifiers that is fast enough to be used interactively. We also show several novel use paradigms of our distance metric. Through experiments, we show that it can be used to find incorrectly or inconsistently labeled training examples, to find specific areas of the data space that need more training data, and to gain insight into which features are missing from the current representation. 

Code is available at https://github.com/kristinbranson/InfluentialNeighbors.

In practice, understanding the spatial relationships between the surfaces of an object, can significantly improve the performance of object recognition systems. In this paper we propose a novel framework to recognize objects in pictures taken from arbitrary viewpoints. The idea is to maintain the frontal views of the major faces of objects in a global flat map. Then an unfolding warping technique is used to change the pose of the query object in the test view so that all visible surfaces of the object can be observed from a frontal viewpoint, improving the handling of serious occlusions and large viewpoint changes. We demonstrate the effectiveness of our approach through analysis of recognition trials of complex objects with comparison to popular methods.