NeuroTrailblazers banner

Overview

Glia outnumber neurons in many brain regions and occupy a substantial fraction of the neuropil volume. In connectomics, glia are not “background” — they are critical structures that must be correctly identified and segmented. Glia-neuron boundary errors are a major source of false connections and morphological distortion. Misidentifying a glial process as part of a neuron inflates that neuron’s arbor and may create false synapses. This document provides the identification cues and decision protocols for the three major glial types encountered in EM.

Instructor script: why glia matter for connectomics

The scale of the problem

In mammalian cortex, approximately 20-40% of the tissue volume is occupied by glial cells and their processes (depending on region and species). In a densely segmented EM volume:

If automated segmentation incorrectly merges a glial process with a nearby neuron, the consequences are:

Teaching point: “Correct glia segmentation is not a side project — it directly affects the quality of your neuronal connectome.”

Astrocytes

EM identification cues

Astrocytes are the most abundant glial type in cortex and the most frequently confused with neuronal processes.

Soma features:

Process features:

Key distinction from neurons: The most common confusion is between fine astrocytic processes and thin dendritic or axonal segments. The discriminating cues:

Feature Astrocyte process Neuronal process
Glycogen granules Present Absent
Synaptic participation No vesicles, no PSD Vesicles (axon) or PSD (dendrite)
Microtubules Absent or very rare Usually present
Shape Irregular, sheet-like, fills gaps Cylindrical, continuous trajectory
Cytoplasm appearance Pale, “watery,” few organelles Denser, with visible organelles
GFAP filaments Present (fine bundles) Absent

Oligodendrocytes

EM identification cues

Oligodendrocytes produce and maintain myelin sheaths in the CNS. They are the easiest glial type to identify because of their distinctive nuclear morphology and their intimate association with myelinated axons.

Soma features:

Process features:

Key distinction from neurons: Oligodendrocyte soma are rarely confused with neuronal soma because of the strikingly dark nucleus. The main confusion arises with oligodendrocyte precursor cells (OPCs), which have paler nuclei and can resemble small neurons or astrocytes.

Microglia

EM identification cues

Microglia are the resident immune cells of the CNS. They are the least abundant and the most morphologically variable of the three glial types.

Soma features:

Process features:

Key distinction from neurons and other glia: The bean-shaped nucleus and lysosome-rich cytoplasm are the primary cues. In activated microglia (responding to injury or pathology), the soma swells and processes retract, making them easier to identify but also more likely to be confused with macrophages or other immune cells.

Decision protocol: glial classification in practice

Step 1: Is this a neuronal process or a glial process?

Start with the most common question — “is this process neuronal or glial?”

Quick tests:

  1. Does the process contain synaptic vesicles or a PSD? → Neuronal (high confidence)
  2. Does the process contain glycogen granules? → Astrocyte (high confidence)
  3. Does the process form or contact a myelin sheath? → Oligodendrocyte (high confidence)
  4. Does the process contain microtubules? → Leans neuronal (but not conclusive — oligodendrocytes have some)
  5. Is the cytoplasm pale, “watery,” and organelle-poor? → Leans astrocyte

If none of the quick tests is conclusive, proceed to Step 2.

Step 2: Follow the process to its soma

If you can trace the process back to a soma, nuclear morphology is the most reliable discriminator:

Nucleus Identity
Large, pale, round, prominent nucleolus Neuron
Pale, irregular contour, dispersed chromatin Astrocyte
Small, very dark, round Oligodendrocyte
Bean-shaped, heterochromatin clumps Microglia

Step 3: Use neighborhood context

If you cannot trace to a soma and local cues are inconclusive:

Step 4: Assign confidence and escalate if needed

If evidence is still ambiguous, assign uncertain-glia or uncertain-neuron/glia label and flag for secondary review. Forcing a classification with insufficient evidence creates errors that are harder to find later.

Worked example: astrocyte vs thin dendrite

Scenario: A thin process (~150 nm) runs between two synaptic boutons. It contains no visible organelles except two small dense granules (~25 nm). It does not participate in any synapses (no vesicles, no PSD). Its cytoplasm is noticeably paler than surrounding neuronal processes.

Analysis:

  1. No synaptic participation → not clearly neuronal
  2. Two small dense granules consistent with glycogen → suggestive of astrocyte
  3. Pale cytoplasm with no microtubules → consistent with astrocyte
  4. Position between synaptic boutons → consistent with perisynaptic astrocyte wrapping
  5. Can trace process 5 sections in either direction — it forms thin sheet-like expansions rather than a cylindrical tube

Classification: Astrocyte perisynaptic process. Confidence: high (glycogen + pale cytoplasm + sheet morphology + perisynaptic position).

Worked example: oligodendrocyte precursor cell (OPC) vs small neuron

Scenario: A small cell body (~7 μm) with a moderately dark nucleus and several thin processes extending into the neuropil. The nucleus is darker than surrounding neuronal nuclei but lighter than a classic oligodendrocyte. The cytoplasm contains some RER.

Analysis:

  1. Nucleus is intermediate — darker than astrocyte, lighter than mature oligodendrocyte
  2. Cell body is small but has several processes (could be either)
  3. No obvious synaptic connections seen on the processes
  4. No myelin sheath connections seen
  5. Contains RER (present in both neurons and oligodendrocyte lineage)
  6. No glycogen granules (argues against astrocyte)

Classification: Oligodendrocyte precursor cell (OPC), also known as NG2 cell. Confidence: medium. Reasoning: intermediate nuclear darkness + small soma + no synaptic features + no myelin association. OPCs are notoriously difficult to classify in EM without molecular markers.

Impact on connectomics analysis

Quantifying glia-neuron boundary errors

In the MICrONS dataset, analysis of proofreading corrections showed that glia-neuron merge errors account for approximately 5-10% of all merge errors (Turner et al. 2022). These are not the most frequent error type, but they are among the most impactful because:

  1. A glia-neuron merge can add false “branches” to a neuron that are actually astrocytic processes, distorting morphological measurements
  2. Synapses near the merge boundary may be attributed to the wrong cell
  3. If the glial process contacts a blood vessel, the neuron appears to have a vascular contact (biologically false)

Best practice: include glia in proofreading queues

Many proofreading campaigns focus exclusively on neurons. However, correcting glia-neuron boundaries should be part of the high-priority QC queue, especially for:

Common misconceptions

Misconception Reality How to verify
“Glia are just background” Glia-neuron boundary errors directly corrupt the neuronal connectome Quantify false connections attributed to glia-neuron merges
“Dark nucleus = neuron” Oligodendrocyte nuclei are the darkest in the neuropil Compare with known cell types; check for myelin association
“All thin processes are axons” Many thin processes in neuropil are astrocytic Check for glycogen, synaptic role, and cytoplasm appearance
“OPCs are easy to identify” OPCs have intermediate features and are challenging even for experts Flag uncertain cases rather than forcing a classification

References