The optical fractionator estimates the total number of cells in a defined tissue region by counting cells in a known fraction of that region. It combines optical disector counting with a sampling design that selects sections, positions within sections, and a depth through each sampled position. The count is then scaled by the reciprocal of the sampling fractions.
Its practical advantage is that total cell number does not require a separate estimate of tissue volume. The original optical fractionator study by West, Slomianka and Gundersen established this approach for hippocampal neurons, avoiding assumptions about cell size and shape. However, “unbiased” describes a correctly implemented sampling and counting procedure, not a guarantee attached to a software button.
What the Optical Fractionator Measures
The result is an estimated total, usually written as N̂, for a stated population within a stated reference region. That might be neurons in one hippocampal subdivision or cells meeting a defined staining criterion in a tissue compartment. The anatomical boundary and cell identification rule belong in the definition of the measurement, not just in the laboratory notes.
Consider a hypothetical comparison. Region A contains 100,000 cells in 10 mm3; region B contains 80,000 cells in 8 mm3. Both have a density of 10,000 cells per mm3, despite a 20% difference in total number. Density answers a different question. Neither result should stand in for the other.
The optical disector provides the counting rule; the fractionator provides the route from sampled counts to a population estimate. For the focal-plane and boundary rules themselves, see the guide to optical disector counting. Here, the focus is how those counts become a defensible total.
The Optical Fractionator Formula
For a conventional design with constant sampling fractions, the estimator is:
N̂ = ΣQ− × (1/ssf) × (1/asf) × (1/hsf)
ΣQ− is the sum of accepted counting events across the sampled region. The other terms describe the three sampling stages. This structure is used in the optical fractionator protocol for developing human forebrain.
| Sampling fraction | Meaning | Calculation in a simple design |
|---|---|---|
| Section sampling fraction, ssf | The fraction of the serial sections selected | 1/k when every kth section is sampled after a random start |
| Area sampling fraction, asf | The fraction of section area sampled by counting frames | Frame area divided by the area of one sampling grid tile |
| Height sampling fraction, hsf | The fraction of section thickness sampled optically | Disector height h divided by measured section thickness t, when thickness is constant |
The height sampling fraction is also called the thickness sampling fraction, or tsf. These names refer to the same sampling stage. Keep the terminology consistent within a study.
A Worked Calculation
Suppose a hypothetical study samples every tenth section. Each counting frame measures 50 × 50 µm, and the sampling grid has steps of 200 µm in both directions. The disector height is 10 µm within sections that are uniformly 20 µm thick after processing. Across the complete sample, 250 cells meet the counting rules.
The fractions are ssf = 1/10, asf = 2,500/40,000 = 1/16, and hsf = 10/20 = 1/2. Therefore:
N̂ = 250 × 10 × 16 × 2 = 80,000 cells.
The combined sampling fraction is 1/320, so each accepted count represents 320 cells in the estimate. These values illustrate the arithmetic; they are not recommended settings for every tissue.
Notice that the area fraction uses the area of a grid tile, not the entire microscope field. Also, section thickness means the thickness being examined, not automatically the microtome setting. Substituting 40 µm for the measured 20 µm in this example would double the estimate to 160,000 without adding a single counted cell.
Build the Sampling Design Before Counting
Define the Reference Region
Write an operational boundary rule before beginning the main study. Specify the anatomical landmarks, treatment of transitional areas, and whether the estimate covers one side or both. Store representative annotated images so that boundary decisions can be reviewed.
A useful planning question is: could another observer identify the same region without knowing the expected result? If not, resolve the ambiguity before increasing the counting workload. More frames cannot resolve an inconsistent definition.
Select Sections and Fields Without Preference
For every tenth section, select the starting section randomly from the first ten positions, then retain that interval through the region. Within each selected section, apply a sampling grid with a random origin. The procedure should determine where counting occurs, rather than the appearance of a promising field. The guide to systematic uniform random sampling covers the design in more detail.
For example, a random starting position of seven produces the sequence 7, 17, 27, 37, and so on. Choosing section seven because it has the clearest staining is a different procedure, even if the remaining interval is perfectly regular.
Keep a section inventory that records serial position, selection, staining, damage, and availability. Do not silently substitute a neighboring section for one that was lost. Record the problem and assess its consequences for the sampling design before continuing.
Count a Defined Event, Not Every Visible Profile
An optical fractionator count needs an identifiable event as focus passes through the disector. For nuclear counting, a common choice is the first appearance of the nucleus under a predefined focal criterion. A nucleus already visible at the exclusion plane is not a new event inside the counting depth. Apply the frame’s inclusion and exclusion boundaries consistently as well.
Distinguish the counted object from the biological label. Counting nuclei estimates cell number only when the relationship between nuclei and cells supports that interpretation. Do not treat nuclei, nucleoli, and cells as interchangeable counting units.
Identification also needs validation. A fluorescent signal must distinguish the target population throughout the sampled depth, rather than only near the section surface. Multiple labels can help separate a cell class from a broader marker-positive population; the protocol combining optical fractionator sampling with multiple immunofluorescence demonstrates this approach and addresses staining penetration and photobleaching.
Before production counting, prepare a short decision record with accepted, rejected, and ambiguous examples. Have observers classify these without group labels, then resolve disagreements. A practical rule should survive ordinary variation in staining, not work only on the best image in the folder.
Section Thickness and Guard Zones Need Separate Checks
The fractionator avoids multiplying numerical density by a separately measured reference volume. That does not make thickness irrelevant. The height sampling fraction still depends on how much of the processed section is examined.
Measure thickness across the sampled material rather than assuming that the cutting setting survives processing unchanged. When thickness varies, use an estimator that accounts for that variation. For a constant disector height, number-weighted mean thickness is an established approach; it weights thickness measurements by their associated counts rather than treating every measurement equally. The Dorph-Petersen, Nyengaard and Gundersen paper on tissue shrinkage develops estimators for these conditions.
Ask which thickness calculation the analysis software actually uses. “Average thickness” is not a sufficient description. Retain the measurements and associated counts so the calculation can be checked independently.
Guard zones leave tissue near the cut surfaces outside the counting depth. Their purpose is to avoid problematic surface regions, but choosing a larger guard zone is not automatically safer. Differential compression can change the distribution of particles through the section. Experimental work on z-axis distortion across sectioning methods shows why counting-box placement must be evaluated against the tissue preparation.
During a pilot, examine where accepted events occur through section depth and whether staining remains detectable. Check that the intended disector and guard zones fit within the thinner sampled locations. If they do not, revise the preparation or sampling protocol instead of improvising at individual fields.
The detailed checks belong in a documented section thickness, guard zone and shrinkage assessment. Treat thickness variation between locations and distortion within a section as separate problems: correcting one does not automatically correct the other.
Choose Sampling Effort Through a Pilot Study
Use the pilot to test the whole measurement, not just the counting speed. Include material that spans the expected range of tissue quality and population abundance. Record counts by section, sampled locations, thickness measurements, ambiguous objects, and time spent.
Then identify what needs improvement. If large stretches of the region receive little coverage, reconsider section spacing. If counts are concentrated in a few fields, reconsider grid spacing. If observers disagree about cell identity, increasing the number of frames is not the first repair.
Assess sampling precision with a coefficient of error, or CE, appropriate to the design. The Gundersen and colleagues study of systematic sampling variance shows that variance estimation depends on the sampling conditions. State the estimator used rather than reporting an unexplained CE value.
Do not make a universal cell-count target the sole acceptance criterion. Two specimens can produce the same total count with very different distributions across sections and fields. Nor should a low CE be presented as proof that cell identification, regional boundaries, or thickness measurements are correct.
For planning purposes, separate three questions: Is the measurement biased? Is sampling within each specimen sufficiently precise? Are there enough independent specimens to address the biological question? Counting more cells in one specimen addresses only part of that task.
Report Enough Detail to Reconstruct the Estimate
A useful methods section should allow a reader to follow the path from selected tissue to counted events to estimated total. The guide to reporting stereological methods and results provides the broader reporting framework.
For an optical fractionator study, retain and report the following:
- The reference region, laterality, target population, staining criteria, and counting event.
- The section interval and random-start procedure, frame dimensions, grid steps, disector height, and guard zones.
- The thickness measurement procedure, observed variation, and estimator used to handle it.
- The sampled sections and locations, accepted counts, estimated totals, and CE calculation for each specimen.
- Missing tissue, uncountable locations, protocol departures, blinding, and observer checks.
Include enough detail to distinguish a planned exclusion from an unnoticed omission. If different regions use different sampling fractions, keep their counts and estimates separate until the correct scaling has been applied. A pooled count alone is not an audit trail.
Finally, name the result accurately. If the protocol identifies marker-positive nuclei in one anatomical compartment, report that population rather than a broader claim about all cells in an organ. The strongest optical fractionator result is not the one with the most decimal places. It is the one whose boundaries, sampling probabilities, counting decisions, and calculation can all be checked.