NeuroTrailblazers banner

Overview

Electron microscopy is the enabling technology of nanoscale connectomics. To interpret EM images correctly — and to understand why certain artifacts arise and how they propagate into segmentation errors — practitioners need a working understanding of how electrons interact with biological tissue to produce contrast. This document covers the core physics, the two major EM modalities used in connectomics, and the practical resolution and contrast considerations that shape every downstream analysis step.

Instructor script: electron beam fundamentals

Why electrons instead of photons?

The resolution of any imaging system is fundamentally limited by the wavelength of the probe. Visible light microscopy is diffraction-limited to roughly 200 nm (the Abbe limit), which is far too coarse to resolve synaptic membranes (~7 nm thick), synaptic vesicles (~40 nm), or the fine processes of axons that can be thinner than 100 nm. Electrons, by contrast, have wavelengths on the order of picometers at typical accelerating voltages (e.g., ~0.004 nm at 100 kV, via the de Broglie relation λ = h/√(2meV)). In practice, lens aberrations limit EM resolution to ~0.1-1 nm, but even “routine” biological EM achieves 1-5 nm resolution — more than sufficient for connectomics.

Key teaching point: The reason we use EM for connectomics is not that we want atomic resolution, but that we need to reliably resolve membranes, vesicles, and organelles in densely packed neuropil where thousands of processes interleave within each cubic micrometer.

Electron-specimen interactions

When a focused electron beam strikes a biological specimen, several interactions occur:

  1. Elastic scattering: Electrons deflected by atomic nuclei without energy loss. Probability increases with atomic number (Z). This is why heavy-metal staining (osmium, uranium, lead) is essential — biological tissue is mostly C, H, O, N (low Z), which scatter weakly. Heavy metals bind preferentially to membranes and proteins, creating contrast.

  2. Inelastic scattering: Electrons lose energy to the specimen, generating secondary electrons, X-rays, and specimen damage. Inelastic scattering is the primary source of radiation damage and limits total dose.

  3. Secondary electron emission: Low-energy electrons (<50 eV) ejected from the specimen surface. These are the primary signal in scanning EM (SEM) and are sensitive to surface topography and composition.

  4. Backscattered electrons (BSE): Primary beam electrons scattered back out of the specimen. Signal strength depends on local atomic number — heavy-metal-stained structures appear bright (more BSE) against unstained background. BSE imaging is the dominant mode in serial block-face SEM (SBEM) and focused ion beam SEM (FIB-SEM) for connectomics.

  5. Transmitted electrons: Electrons that pass through the specimen. Collected in transmission EM (TEM). Contrast arises because stained structures scatter more electrons out of the beam, appearing dark in bright-field TEM.

Key teaching point on contrast

Contrast in EM is fundamentally about differential scattering. Membranes appear dark in TEM (and bright in BSE-SEM) because osmium tetroxide binds to lipid bilayers, increasing local atomic number. Synaptic vesicle membranes, postsynaptic densities, and mitochondrial membranes all accumulate heavy metals to different degrees, creating the organelle-specific contrast that annotators rely on.

Transmission EM (TEM) for connectomics

How it works

In TEM, a thin section (40-70 nm) of resin-embedded, heavy-metal-stained tissue is placed in the electron beam. Electrons that pass through the section are focused by magnetic lenses onto a detector (camera or phosphor screen). Regions with more heavy metal scatter more electrons out of the beam, appearing darker.

Historical role

TEM was the original method for all EM connectomics, including White et al.’s (1986) C. elegans reconstruction. Sections were cut manually, placed on grids, and imaged one at a time. Serial-section TEM (ssTEM) was the gold standard through the 2000s.

Modern ssTEM at scale

The key innovation for large-volume TEM was automated tape-collecting ultramicrotomy (ATUM — Hayworth et al. 2014) combined with multi-beam SEM or TEM camera arrays. ATUM collects hundreds to thousands of ultrathin sections on a continuous tape, which can then be imaged automatically. The FAFB (Full Adult Fly Brain) dataset (Zheng et al. 2018) and parts of the MICrONS dataset used ATUM-based approaches.

Strengths and limitations

Strength Limitation
Excellent contrast for membranes Requires physical ultrathin sectioning
High resolution (sub-nm possible) Section handling introduces artifacts (compression, folds, tears)
Well-established staining protocols Each section imaged independently — registration needed
Large field of view per image Z-resolution limited by section thickness (~30-50 nm)

Scanning EM (SEM) for connectomics

Serial block-face SEM (SBEM)

Introduced by Denk & Horstmann (2004). A diamond knife inside the SEM chamber shaves a thin layer (~25-30 nm) from the top of a resin-embedded block, and the freshly exposed surface is imaged using BSE detection. The process repeats automatically: cut, image, cut, image — producing an aligned image stack without the need to handle physical sections.

Key advantage: Automatic z-alignment (no section warping, no lost sections). The block face is always flat and in focus.

Key limitation: Destructive — each section is destroyed after imaging. No re-imaging possible. Also, knife chatter artifacts create periodic intensity variations.

Focused ion beam SEM (FIB-SEM)

A gallium ion beam mills a thin layer (~4-8 nm) from the block face, and the exposed surface is imaged by SEM. Offers the finest z-resolution of any volume EM method (isotropic 4-8 nm voxels are achievable). Used when isotropic resolution is critical, such as for resolving fine organelle structure or tracing the thinnest axons.

Key advantage: Isotropic voxels, no mechanical sectioning artifacts.

Key limitation: Very slow acquisition, small field of view, expensive. FIB-SEM volumes are typically limited to 50-100 μm per side (though Xu et al. 2021 at Janelia achieved larger volumes with enhanced FIB-SEM).

Multi-beam SEM

Multiple electron beams (61-91 beams) scan the specimen simultaneously, dramatically increasing throughput. Used in the MICrONS project. Combines the throughput advantages of SBEM with the image quality of TEM-like section imaging.

Resolution vs field of view vs throughput: the connectomics tradeoff triangle

Every EM acquisition must balance three competing demands:

  1. Resolution (voxel size): Finer resolution → better membrane detection → fewer segmentation errors, but more data per unit volume.
  2. Field of view (volume): Larger volumes capture more complete circuits, but generate more data.
  3. Throughput (acquisition speed): Faster imaging → larger volumes feasible, but may sacrifice signal-to-noise ratio (SNR).

Typical connectomics parameters:

Modality XY resolution Z resolution Typical volume Acquisition time
ssTEM (ATUM) 4 nm 30-40 nm mm³ Months
SBEM 8-10 nm 25-30 nm 500 μm³ Weeks-months
FIB-SEM 4-8 nm 4-8 nm (50-100 μm)³ Weeks-months
Multi-beam SEM 4 nm 30-40 nm mm³ Months

Key teaching point: The choice of EM modality is a scientific decision, not just a technical one. The question you want to answer determines the minimum resolution, volume, and completeness required, which in turn constrains the modality.

Contrast mechanisms and staining

The standard heavy-metal staining protocol

Most connectomics EM uses a variant of the following staining sequence:

  1. Osmium tetroxide (OsO₄): Reacts with unsaturated lipids in membranes. Primary source of membrane contrast. Typically 1-2% OsO₄ in buffer for 1-2 hours. Sometimes enhanced with ferrocyanide-reduced osmium (rOTO protocol — Hua et al. 2015) for stronger membrane staining.

  2. Uranyl acetate (UA): Binds to nucleic acids and proteins. Enhances contrast of ribosomes, chromatin, and proteinaceous structures (like PSDs and active zones). Important for synapse identification.

  3. Lead citrate (or lead aspartate): Provides additional general contrast enhancement. Binds to osmicated structures and other heavy-metal-stained components.

The rOTO protocol

The reduced osmium-thiocarbohydrazide-osmium (rOTO) protocol (adapted from Hua et al. 2015) is widely used for volume EM because it provides strong, uniform heavy-metal staining throughout the block — essential for SBEM and FIB-SEM where you cannot post-stain individual sections.

En bloc vs section staining

Practical implication for annotators

Contrast quality directly affects segmentation accuracy. When staining is uneven (e.g., reduced OsO₄ penetration at block edges), membrane detection degrades, merge error rates increase, and proofreading burden rises. This is why acquisition QA must happen before full-volume reconstruction (see acquisition-qa.md).

Worked example: reading an EM image of cortical neuropil

Imagine you are looking at a single 2D EM image (SEM-BSE mode, ~8 nm pixel size) of layer 2/3 mouse cortex:

  1. Dark lines (~7 nm wide, appearing as thin bright lines in BSE): These are membranes. Every distinct process is bounded by its membrane.
  2. Round, dark profiles (~40-50 nm): Synaptic vesicles inside axon terminals.
  3. Electron-dense band on membrane: Postsynaptic density (PSD). If thick (>30 nm), likely excitatory synapse (Gray Type I).
  4. Oval structures with internal folds (0.5-2 μm): Mitochondria. The internal folds are cristae.
  5. Dark parallel lines: Stacked rough ER (if studded with dark dots = ribosomes) or myelin wrapping (if concentric around an axon profile).
  6. Large pale region with dark rim (~10 μm): Neuronal soma with nuclear envelope.
  7. Everything else: Dense neuropil — a tangle of axons, dendrites, spines, glial processes, and extracellular space, all packed tightly together.

Exercise for learners: Given this vocabulary, trace three complete profiles in the image and identify: (a) what compartment type each is, (b) what organelle evidence supports your call, (c) any synapses you can identify between them.

Common misconceptions

Misconception Reality How to verify
“Higher resolution is always better” Resolution must match the question; excess resolution wastes throughput and storage Calculate minimum voxel size needed for target structures
“EM shows you molecules” EM shows heavy-metal staining patterns, not molecules directly Compare with immuno-EM or correlative approaches for molecular identity
“All EM images look the same” Contrast varies dramatically with staining protocol, accelerating voltage, and detector Compare images from different datasets (H01 vs MICrONS vs FlyWire)
“Dark = more biological material” In TEM, dark = more scattering (staining). In BSE-SEM, bright = more scattering Know your modality’s contrast convention

References