Bioimaging Capabilities to Enable Mapping of the Neural Connections in a Complex Brain

Rick Stevens, Gayla Orr2020 · DOI

TL;DR

Mapping the neural connections in a complex brain (the connectome) is a grand challenge requiring unprecedented advancements in bioimaging, computation, and data handling. This report summarizes a DOE-convened roundtable that outlines the necessary capabilities and identifies unique opportunities for the Department of Energy to lead in this endeavor, leveraging its expertise in large-scale scientific infrastructure and high-performance computing.

What Problem Does This Paper Address?

The paper addresses the immense challenge of mapping the complete neural circuitry of complex brains, from mice to humans. This "connectome" is crucial for understanding brain function, disease, learning, and evolution. Current technologies fall short in terms of resolution, speed, and data processing capabilities required for such a massive undertaking. The report outlines the requirements for a full connectome, the limitations of existing bioimaging technologies, and the computational and data infrastructure needed to achieve this goal, highlighting the role the Department of Energy (DOE) can play.

Methods

This report is a summary of a virtual roundtable, not a research paper detailing specific experimental methods. The "methods" discussed are the existing and proposed technologies for:

  • Bioimaging: Optical microscopy (including expansion microscopy, lattice light sheet microscopy, and Brainbow), electron microscopy (EM) and tomography, and X-ray microscopy and tomography. The report discusses the strengths and weaknesses of each for connectomics.
  • Computation and Data: The need for exascale and beyond computing, massive data storage, advanced algorithms for segmentation, registration, and reconstruction, and the development of automated workflows.
  • Connectome Requirements: Defining the necessary neuronal properties (morphology, location, synapse anatomy, molecular identity, function) and the required resolution (microscale ~10 nm, mesoscale ~100 nm) for mapping.

Datasets

Not discussed in the paper. The report focuses on the conceptual framework and technological requirements for future data acquisition and processing, rather than specific datasets used for training or evaluation.

Key Insights and Ideas

  • Multimodal Imaging is Essential: No single imaging modality can capture all the necessary information. A combination of optical, electron, and X-ray techniques will be required, often used correlatively.
  • Scale is the Primary Hurdle: The sheer volume of data and the required resolution for a whole brain connectome are orders of magnitude beyond current capabilities. Mapping a mouse brain is an exabyte-scale problem, while a human brain is zettabyte-scale.
  • DOE's Unique Position: The DOE's combination of state-of-the-art experimental facilities (synchrotrons, electron microscopes) and leadership-class supercomputing infrastructure at the same sites provides a unique advantage for tackling this challenge.
  • Automation is Non-Negotiable: Current workflows rely heavily on human intervention, which is not scalable. Full end-to-end automation of data acquisition, processing, and reconstruction is critical.
  • Functional Information is Key: Registering structural connectome data with neural activity imaging will greatly enhance scientific discovery and understanding of brain function.

Key Results

The report does not present quantitative experimental results but rather outlines the estimated scale and timelines for achieving connectomes:

  • Mouse Connectome: Estimated to require exascale computing and storage, potentially achievable in about a year with planned 2025 DOE machines.
  • Human Connectome: Requires zettascale storage and computational power. Achieving a full human connectome in a year would necessitate 10^4 exascale computing systems, with current projections suggesting it might be feasible in 5-10 years for mesoscale and potentially longer for microscale.
  • Mesoscale Human Connectome: Could potentially be achieved in 1 year with ~100 lattice light sheet microscopes.
  • EM-based Human Connectome: Estimated to require 3000 MultiSEMs operating in parallel over 5 years.

Limitations and Future Work

  • Resolution vs. Throughput Trade-off: Many imaging modalities face a trade-off between achieving the necessary resolution (especially for synapses) and the speed required to image large volumes.
  • Sample Preparation Challenges: Staining and sectioning large brain volumes for EM and X-ray imaging remain significant technical hurdles.
  • Synapse Detection in X-ray: X-ray techniques, while promising for large volumes, have yet to demonstrate the ability to reliably resolve synapses.
  • Data Storage and Throughput: Storage capacity and throughput are growing slower than compute power, posing a bottleneck for handling massive connectome datasets.
  • Future Work: The report emphasizes the need for continued R&D in:
  • Developing novel bioimaging technologies and correlative imaging systems.
  • Creating fully automated pipelines for data processing and reconstruction.
  • Advancing AI and machine learning for segmentation, proofreading, and data analysis.
  • Developing specialized hardware accelerators for connectome computing.
  • Establishing distributed "facilities" for connectomics research.

Summary updated

Abstract

DOE’s Office of Science (SC), through its BRAIN Initiative Working Group, convened a virtual roundtable on November 2-4, 2020, to develop a conceptual framework for the capabilities needed to enable an interdisciplinary community of researchers to fully map the neural connections initially in the brain of a mouse and ultimately in the brain of a human. Over 30 researchers and programmatic observers from diverse scientific backgrounds, including those working in both the basic and applied sciences, participated in the roundtable and conceptualized forward-leaning ideas. In developing the framework, participants were asked to address questions under the three themes: connectome requirements, bioimaging technology, and computing and data.
BibTeX
@misc{stevens2020bioimaging,
  author = {Rick Stevens and Gayla Orr},
  title = {{Bioimaging Capabilities to Enable Mapping of the Neural Connections in a Complex Brain}},
  year = {2020},
  doi = {10.2172/1821173},
}
Figure