pubmed.ncbi.nlm.nih.gov

Large-scale automated histology in the pursuit of connectomes - PubMed

  • ️Sat Jan 01 2011

Review

Large-scale automated histology in the pursuit of connectomes

David Kleinfeld et al. J Neurosci. 2011.

Abstract

How does the brain compute? Answering this question necessitates neuronal connectomes, annotated graphs of all synaptic connections within defined brain areas. Further, understanding the energetics of the brain's computations requires vascular graphs. The assembly of a connectome requires sensitive hardware tools to measure neuronal and neurovascular features in all three dimensions, as well as software and machine learning for data analysis and visualization. We present the state of the art on the reconstruction of circuits and vasculature that link brain anatomy and function. Analysis at the scale of tens of nanometers yields connections between identified neurons, while analysis at the micrometer scale yields probabilistic rules of connection between neurons and exact vascular connectivity.

PubMed Disclaimer

Figures

Figure 1.
Figure 1.

Motion detection circuit in fruit fly. A, Hassenstein–Reichardt elementary motion detection compares correlations in light intensity signals that are offset in space and time. B, Fruit fly. C, A light microscopic section through fruit fly optic lobe showing the first three stages of visual processing: lamina, medulla, and lobula plate. D, Electron microscopy reconstruction of a medulla column, different colors correspond to different neurons. C is adapted from Takemura et al. (2008).

Figure 2.
Figure 2.

A schematic view of serial block-face scanning electron microscopy technology. A, The electron beam first raster scans the surface of a tissue sample that is embedded in a plastic block. B, After the acquisition of an image, a diamond knife cuts a thin section from the surface of the block. C, A cutaway of the volume of mouse retina acquired for the study of the direction-selectivity circuit in retina. The image shows a reconstructed starburst amacrine cell in black and the locations of its synapses (see inset) onto direction-selective retinal ganglion cells. The synapse locations are color coded by the functionally recorded preferred directions of the direction-selective retinal ganglion cells. The scale bar is shorter near the top as a consequence of perspective. D, Starburst amacrine cell somata (large blue and cyan dots) and the direction of their dendrites (blue and cyan vectors) that form synapses (small blue and cyan dots) onto a direction-selective ganglion cell (gray). Inset, A polar histogram of the vector directions (black) indicate an asymmetric distribution opposite to the functionally recorded preferred direction of the ganglion cell (purple tuning curve). E, Schematic wiring diagrams of starburst amacrine cell (SAC) input to direction-selective ganglion cells (DSGC) for three different locations of the starburst cell soma. The starburst cell dendrite directions are biased toward the null direction (ND). A and B are courtesy of Julia Kuhl. C–E are adapted from Briggman et al. (2011).

Figure 3.
Figure 3.

The process of realizing a connectome from the network anatomy for mouse visual cortex. A, Custom camera array for transmission electron microscopy. Labeled components: 1, motorized positioner knobs; 2, lead shielding; 3, vacuum chamber extension; 4, scintillator; 5, leaded glass; 6, 2 × 2 array of f/2.0 optical lenses coupled to 50 MHz readout of 11 megapixel CCD cameras; 7, camera trigger breakout box; 8, cables to four acquisition computers. B, Acquisition and reconstruction overview. Top row, Left to right, Four cameras with overlapping fields of view of the scintillator; raw camera images; normalized, stitched, and cropped camera images into a single “tile.” Bottom row, Left to right, Mosaic of tiles resulting from sample motion; stitched mosaic, with red square delineating zoom area in the next image; zoom of a single tile showing area tracing. Not shown: imaging of multiple sections, alignment of serial section mosaics, tracing through multiple sections, and validation of tracing. C, Three-dimensional tracing of the 14 physiologically characterized cells and their postsynaptic targets; only those targets receiving convergent synaptic input from two or more physiologically characterized cells are shown. Physiologically characterized orientation preference for the cell given in the key, bottom right. D, Connectivity graph of the anatomical objects shown in C. Magenta targets are excitatory; cyan targets are inhibitory. Circles are postsynaptic dendrites that could be traced back to the cell body; squares are dendrites that left the imaged volume before the cell body was reached. C and D are adapted from Bock et al. (2011).

Figure 4.
Figure 4.

Array tomography allows the resolution, classification and quantification of individual synapses. A, The array tomography proteomic imaging cycle involves multiple rounds of immunostaining, imaging, and antibody elution and allows collection of immunofluorescence data of multiple antibody channels; 24 demonstrated to date. B, Volume rendering of layer 5 of the somatosensory cortex of a Thy-1 YFP-H mouse (Feng et al., 2000), immunolabeled for the general synaptic marker synapsin. Individual synapsin puncta have been segmented digitally and rendered in random color to emphasize that each synapsin punctum is fully resolvable in all three dimensions. Scale bar, 10 μm. C, High dimensional proteomic data are viewed as “synaptograms” with columns representing individual serial sections through a synapse and rows representing each marker. The two synaptograms show examples of a glutamatergic synapse with glutamatergic markers boxed in red and a GABAergic synapse with the respective markers boxed in blue. Each square represents an area of 1 × 1 μm2. D, The density of six synapse subtypes in the somatosensory cortex of a C57BL/6J mouse plotted as a function of depth through the cortical layers, white matter (WM), and part of the striatum (STR).

Figure 5.
Figure 5.

Cortical angiome and its relation with neuronal columns revealed with all optical histology. A, Gelatin-based fluorescent vascular cast of the entire vasculature of the mouse brain. B, Schematics of all optical histology block-face imaging. A sample that contains one or more fluorescently labeled structures is imaged by TPLSM to collect optical sections through the ablated surface. Sections are collected until scattering of the incident light reduces the signal-to-noise ratio below a useful value. The top of the now-imaged region of the tissue is cut away with amplified ultrashort laser pulses to expose a new surface for imaging. The sample is again imaged down to a maximal depth, the new optical sections are added to the previously stored stack, and the process of ablation and imaging is repeated. This process generates thousands of overlapping blocks that are stitched together to generate a contiguous volumes. The position of each block on the global coordinate system is determined by the least-square solution of a minimization problem defined by all cross-correlations between overlapping blocks. C, The raw images are vectorized by fitting the data to tubes. However, sections of low signal-to-noise along the vasculature can result in “gaps,” several to tens of micrometers in length, in the reconstructed microvasculature. These are corrected with an automated technique. Left, A region of low signal-to-noise ratio (arrowhead) and the resulting “gap” (arrow) seen in the max-projections of the grayscale data and its volumetric representation (red isosurface) as well as in the vectorization (red tubes). Center, Automated local threshold relaxation finds a bridging mask (cyan, arrow). Right, A corrected vectorization is computed through the gap (red tube, arrow). D, Localization of cortical columns (“barrels”) using data on all fluorescently labeled neurons. Top, Axially oriented maximal projections across 50 μm of the microvasculature in layer IV. Middle, Cell somata using fluorescent channels for a fluorophore linked to the pan-neuronal antibody αNeuN in the same dataset as the vasculature; columnar boundaries are distinguished in somata channel but not in the vasculature. Bottom, Columns outlines from the cell densities in the middle panel and labeled per convention. E, Vectorized reconstruction across a 1.5 mm3 volume of somatosensory cortex; same dataset as in D.

Figure 6.
Figure 6.

Manual detection of neuronal somata in cortical columns from vibrissa cortex of rat. A, Overlay of a maximum-intensity α-NeuN fluorescence image with markers that were manually placed at the midpoints of α-NeuN-positive somata. B, All NeuN-positive somata in the C2, D2, and D3 columns, defined by extrapolation of the barrel outline in L4 along the vertical axis. C, One-dimensional profiles of α-NeuN density along the respective vertical column axis for the C2, D2, D3, and the mean column. Adapted from Meyer et al. (2010a).

Figure 7.
Figure 7.

Automated methods to reconstruct an entire mouse brain at cellular resolution. A, The brain is cut along the coronal plane into ∼210 consecutive sections. B, Sections are labeled within α-NeuN (red) and α-GAD67 (green) and imaged as a mosaic with high-speed confocal microscopy. C, Enlargement of one box in B. D, Enlargement of one box in C. Excitatory (yellow) and inhibitory (pink) somata are detected and labeled using an automated image-processing pipeline. E, Tilt and/or distortions of individual brain sections are corrected by semiautomated postprocessing. F, The quality of each image stack after postprocessing is sufficient to align consecutive sections with cellular precision. G, The resultant image stack comprises the entire mouse brain and allows segmentation of anatomical structures, e.g., blood vessels or pia (brown), and quantification of soma densities (green).

Similar articles

Cited by

References

    1. Abeles M. Local cortical circuits: an electrophysiological study. Berlin: Springer; 1982.
    1. Adelson EH, Bergen JR. Spatiotemporal energy models for the perception of motion. J Opt Soc Am A. 1985;2:284–299. - PubMed
    1. Amit DJ. Modeling brain function: the world of attractor neural networks. Cambridge: Cambridge UP; 1989.
    1. Blinder P, Shih AY, Rafie C, Kleinfeld D. Topological basis for the robust distribution of blood to rodent neocortex. Proc Natl Acad Sci U S A. 2010;107:12670–12675. - PMC - PubMed
    1. Bock DD, Lee WC, Kerlin AM, Andermann ML, Hood G, Wetzel AW, Yurgenson S, Soucy ER, Kim HS, Reid RC. Network anatomy and in vivo physiology of visual cortical neurons. Nature. 2011;471:177–182. - PMC - PubMed

Publication types

MeSH terms

Grants and funding