A Process for Digitizing and Simulating Biologically Realistic Oligocellular Networks Demonstrated for the Neuro-Glio-Vascular Ensemble

Abstract

One will not understand the brain without an integrated exploration of structure and function, these attributes being two sides of the same coin: together they form the currency of biological computation. Accordingly, biologically realistic models require the re-creation of the architecture of the cellular components in which biochemical reactions are contained. We describe here a process of reconstructing a functional oligocellular assembly that is responsible for energy supply management in the brain and creating a computational model of the associated biochemical and biophysical processes. The reactions that underwrite thought are both constrained by and take advantage of brain morphologies pertaining to neurons, astrocytes and the blood vessels that deliver oxygen, glucose and other nutrients. Each component of this neuro-glio-vasculature ensemble (NGV) carries-out delegated tasks, as the dynamics of this system provide for each cell-type its own energy requirements while including mechanisms that allow cooperative energy transfers. Our process for recreating the ultrastructure of cellular components and modeling the reactions that describe energy flow uses an amalgam of state-of the-art techniques, including digital reconstructions of electron micrographs, advanced data analysis tools, computational simulations and in silico visualization software. While we demonstrate this process with the NGV, it is equally well adapted to any cellular system for integrating multimodal cellular data in a coherent framework.

Keywords: 3D reconstruction; NGV; electron microscopy; energy metabolism; in silico visualization; simulation.

FIGURE 1

Schematics of the 3DEM pipeline. (A) Sample preparation. Rodents brains are processed following cardiac perfusion (left) and then 100 μm thick coronal section containing somatosensory cortex are prepared for face-block scanning EM (center). The section is then placed on a metal stub and trimmed to expose a squared surface of roughly 0.3 to 0.5 millimeters (right). (B) Images are obtained using a serial-block face electron microscope, such as FIBSEM or a Quanta 600 SEM equipped with a Gatan 3View module (pictured on the left). Hundreds to thousands serial section micrographs, with a relatively high field of view if necessary can be obtained automatically in a few days with minimal human supervision (center) at high resolution (right).

FIGURE 2

3D reconstruction of glial cells and neurons from Layer VI somatosensory cortex brain parenchyma. A digital reconstruction of a micrograph stack from Layer VI of the somatosensory cortex of a P14 rat. Blood vessel in red, two neurons in shades of violet, two microglia in light blue/green, the fiber tract in dark blue, the pericyte in orange at the bottom of the image and two astrocytes, one in yellow and another one in orange, at the top of the image. Glass-like spheres are the nuclei reconstructed from the original image stack.

FIGURE 3

Imaging and quantification of glycogen distribution in EM. (A) Isotropic EM stack volume from CA1 rat hippocampus from Calì et al., 2016. (B) Exploded view of dense reconstruction from (A) Gray, 161 axons; Green, astrocytic process; Blue, 11 dendrites. (C) Magnification of a single micrograph from (A) showing elements of interest in the neuropil of rat hippocampal CA1 region. Synapses (red arrows), glycogen granules (yellow arrows), axons (A), boutons (B), dendrites (D), spines (s) and astrocytic processes (Ast) on EM micrographs. (D) Quantification of the spatial distribution of glycogen granules from sample C; top, number of glycogen granules associated to elements of the neuropil, calculated using a nearest neighbor rationale. Bottom, total number of glycogen granules per bouton type.

FIGURE 4

Hybrid semi-automated large-scale dense reconstruction. Our pipeline combines and extends two public domain software programs for segmentation. (A) TrakEM2 is used for manually specifying exact boundaries and fine details. (B) A binary mask (top) can be extracted from TrakEM2 after a first, detailed segmentation of a structure of interest (a neuron in the example in a, green). (C) Ilastik will use the TrakEM2 binary mask as training for a rough but extensive segmentation of most part of the cell along the z-axis. The segmentation can be then exported again as binary mask (B, bottom), to be imported again in TrakEM2 for further proofreading. The sequence of steps is indicated below each panel and numbered from 1 to 5.

FIGURE 5

Reconstruction time per each cell type. Histogram showing time to complete each reconstruction, including training of naïve reconstructors and proofreading. Outlined bars highlight cells reconstructed using the new pipeline using ilastik as a first pass segmentation.

FIGURE 6

Custom made analysis tools. (A) Example of a 3D reconstructed astrocytic process (green, semi-transparent) enwrapping an axospinous synapse. Presynaptic terminal in purple, postsynaptic terminal blue. (B) Astroproximity Blender addon evaluates the contact surface area between astrocytic processes and synapses. (C) Example of a 3D reconstructed astrocytic process (green, semi-transparent) with its mitochondria (purple) and ER (yellow). (D) ER/mito Blender analysis addon measures minimum distance between ER, mitochondria and their closest synapse, as well as cross sectional area of the organelle. (E) Schematics of Abstractocyte, standalone software for qualitative observations of dense reconstructions of neuropil, specifically designed to analyze three-dimensional relationships between astrocytes and neurites. From left to right, a very occluded view including reconstructed 3D models of astrocytes and neurites can be simplified by showing astrocytes as their skeleton (second panel), to a full abstract view of connecting dots representing contact points between astrocytes and boutons/spines, whose halo highlight glycogen absorption maps (fourth panel), or a hybrid abstract view of astrocytes and contact points with boutons spines from panel one.

FIGURE 7

Glycogen Lactate Absorption Modeling (GLAM). Using as input (A) the accurate 3D reconstructions of cellular processes and the distribution of glycogen granules, a radiance-based model (B) estimates lactate absorption in order to highlight specific absorption patterns (C,D).

FIGURE 8

Noradrenergic modulation of energy supply in the NGV. Schematic compartmental diagram of the NGV model with noradrenergic (NE) locus coeruleus inputs along with neuronal, glial (astrocyte), vascular and extracellular space compartments. Key metabolic and transduction pathways for neuromodulation, glucose metabolism, energy production (ATP), glycogenolysis and lactate (LAC) shuttling from the astrocyte to the neuron are approximated. GLC, glucose; NAD(H) nicotinamide adenine dinucleotide (reduced); G6P, glucose-6-phosphate; PYR, pyruvate, EAAT, excitatory amino acid transporter (for glutamate); Na/K ATPase, ATP-dependent sodium-potassium pump; MCT, monocarboxylate transporter (for LAC); (P)Cr, (phospho)creatinine; GLUT, glucose transporter; AMPA, alpha-amino-3-Hydroxy-5-methyl-4-isoxazole propionic acid - type glutamate ionotropic receptor; O₂, molecular oxygen; H₂O, water; G_p, G-protein; β2R, beta2-adrenergic receptor; AC, adenylate cyclase.

FIGURE 9

Metabolic simulation on the cortical column. (A) Voxelized space containing neurons and astrocytes was mapped onto cubic voxels overlapping the cortical column simulation. More depolarized neuronal processes are shown in red, while hyperpolarized processes are shown in blue. (B) Glutamate release at 10 ms intervals. Glutamate release drives the metabolic simulations. Neurons shown in white, glia in yellow. (C) The concentration of metabolites such as ATP (normalized) was saved as a time series and exhibits location-dependent effects.

FIGURE 10

Visualization of Volumetric Data using Livre. (A) Visualizing the vasculature of a full brain dataset on a (3 × 4) tiled display wall. The visualization uses Livre to render the dataset out-of-core and the frames are streamed to the display wall using Tide. (B) Volume rendering of the ATP concentration change (normalized) in a neocortical column at 10 ms snapshots.

FIGURE 11

Scientific data visualization using open-source tools: ParaView and Voreen. (A) Volume rendering of a volumetric stack, containing an axon (red) surrounded by a glial cell, in green, using the volume rendering plug-in in ParaView. (B) Volume rendering of a light microscopic stack data, showing two astrocytes (in green) and their nuclei (in Blue) using ParaView. (C) Large volume stack of vasculature using the GPU-based volume rendering plug-in in Voreen. Dataset, courtesy of Bruno Weber et al., Institute of Pharmacology and Toxicology –Experimental Imaging and Neuroenergetics, University of Zürich.

FIGURE 12

Scientific data visualization using open-source tools. Reconstructing a complex vasculature dataset using Blender-based Metaball algorithm and rendering the reconstructed model using Cycles, a physically plausible engine integrated in Blender.

FIGURE 13

In silico physically plausible rendering. (A) Physically plausible in silico epi-widefield fluorescence imaging of a single neuron virtually tagged with GFP. The focal plane of the microscope is focused on the soma. (B) Physically plausible in silico epi-widefield fluorescence imaging of a digital slice reconstructed from the somatosensory cortex of a P14 rat.

FIGURE 14

Use of VR to validate the Glycogen Lactate Absorption Modeling (GLAM). (A) GLAM map is used for planning sparse reconstructions in collaborative sessions performed on large-scale VR setups (CAVE), (B) and for intensive visual analysis on HMD-based stereo setups. (C) VR GUI implemented in unity to interact with the 3D models in virtual reality. (D) Loaded model in VR as seen during interactive navigation using HTC VIVE. Written consent from individuals appearing in this figure have been obtained for publication.

FIGURE 15

Infographics of the digitalization, modeling, simulation and visualization pipeline. WF1-WF4 are shown in sequence with example images from each step in the process, along with Output Data (bold lettering) and Software Tools corresponding to each WF step.