A microscope shows you a field of cells and asks you to judge what you see. A flow cytometer does something structurally different: it examines cells one at a time, records a set of measurements for each, and hands you a table with thousands or millions of rows. Nobody looks at that table. What people look at is a plot, and the plot is where interpretation begins.
That shift, from images to per-cell measurements, is what makes the technique powerful. It also means the answer depends on decisions made after the instrument has finished. Where you draw the boundary around a cluster determines what percentage you report, and two competent analysts drawing boundaries on the same data will not always produce the same number.
Understanding flow cytometry properly therefore means understanding two things: the physics that turns a cell suspension into measurements, and the judgement that turns measurements into populations. The first is elegant and fairly rigid. The second is where most of the interesting arguments happen.
Key takeaways
- Hydrodynamic focusing forces cells into single file so each one crosses the laser alone.
- Forward and side scatter give a rough size and internal-complexity reading with no staining required.
- Fluorophore emission spectra overlap, and compensation is the arithmetic that corrects for it.
- Gating is a human decision, and the sequence of gates changes the reported result.
- Percentages depend entirely on the denominator population, so a percentage without its parent gate is uninterpretable.
Hydrodynamic Focusing Into Single File
Everything the instrument does depends on cells arriving at the interrogation point one at a time, evenly spaced and in the same position relative to the laser. A tube of cell suspension does not naturally behave that way, so the fluidics system arranges it.
The sample stream is injected into the centre of a much faster-flowing stream of buffer, called sheath fluid, inside a nozzle that narrows. Because the flow is laminar rather than turbulent, the two streams do not mix. The outer sheath drags the inner sample stream along and, as the channel narrows, stretches it into a thin thread. Cells suspended in that thread are pulled into a line and separated from one another along it. The physical principle is the same one that lets you draw a thread of syrup out of a pool without it breaking up.
The width of the core stream can be adjusted by changing the pressure differential. A narrow core, produced by running the sample slowly relative to the sheath, keeps cells tightly positioned in the laser’s brightest region and gives the best measurement precision. A wider core lets more cells through per second but allows them to wander across the beam, which broadens the measurement distribution and blurs populations that sit close together. Running fast when the question requires resolution is one of the most common self-inflicted problems in cytometry.
Two failures follow directly from this arrangement. Coincidence occurs when two cells cross the beam close enough together to be recorded as one event, producing a false signal with roughly double the expected intensity. Clumping produces the same effect on a larger scale and can block the nozzle entirely. Both are managed by filtering the sample, keeping the concentration reasonable and, in analysis, excluding doublets using the shape of the pulse each event produces.
Forward and Side Scatter as Size and Complexity
When a cell crosses the laser, it deflects light. Two detectors capture that deflection from different angles, and together they provide a description of the cell that requires no reagent at all.
Forward scatter is measured close to the axis of the beam, behind an obscuration bar that blocks the direct laser light. It is dominated by diffraction around the cell, and it correlates broadly with cell size. The correlation is real but loose: refractive index and shape also contribute, so forward scatter should be read as roughly proportional to size rather than as a measurement of diameter.
Side scatter is collected at right angles and reflects light refracted and reflected by internal structures. Granules, complex nuclei and dense organelles all increase it. This is why granulocytes, packed with granules, sit high on a side scatter axis while lymphocytes, which are small and internally smooth, sit low.
Plotting the two axes against each other produces the familiar scatter plot in which the major leukocyte populations occupy distinct regions before any antibody has been applied. This plot does more work than its simplicity suggests. It is used to exclude debris, which falls near the origin with low signal on both axes, to identify dead cells, which typically shift in scatter as their membranes fail, and to define the starting population for everything that follows.
Scatter also provides the first quality check on a sample. A scatter plot with populations in unexpected positions, an unusually large debris cloud or a smear between clusters usually indicates a preparation problem, and no amount of downstream analysis will rescue it.
Fluorophores and Antibody Panels
Scatter distinguishes broad cell types. Answering finer questions requires labelling specific molecules, and this is done with antibodies carrying fluorescent tags.
A fluorophore absorbs light at one range of wavelengths and re-emits it at longer ones. The instrument excites it with a laser of appropriate wavelength and collects the emitted light through optical filters that pass a defined band, each band feeding its own detector. By conjugating different fluorophores to antibodies against different surface markers, several properties of each cell can be recorded simultaneously.
Panel design is the intellectual work. Several constraints interact at once. Each fluorophore must be excitable by one of the instrument’s available lasers. Its emission must fall into a detector band that is not already committed. Brighter fluorophores should be assigned to markers expressed at low density, since a dim tag on a scarce target may not separate from background, while abundant markers can tolerate dimmer tags. Markers expressed on the same cells should ideally use fluorophores whose spectra overlap least, because overlap between co-expressed markers is where compensation errors do the most damage.
Tandem dyes, which pair two fluorophores so that one absorbs and transfers energy to the other, extend the usable range but are chemically fragile. They degrade with light exposure, fixation and storage, and degradation shifts their emission in ways that break previously validated settings. Lot-to-lot variation in tandem conjugates is a known source of drift between experiments.
Controls are part of the panel rather than an addition to it. Unstained samples establish autofluorescence. Single-stained controls provide the data for compensation. Controls in which one marker is omitted from the full panel define where the boundary between positive and negative should sit for that marker in the actual staining context.
Compensation and Spectral Overlap
Fluorophore emission spectra are broad and asymmetric, with long tails. A filter selecting a band for one fluorophore will therefore also admit some light emitted by others. The detector cannot tell which molecule the photons came from; it reports total light in its band.
Compensation is the arithmetic that removes the predictable contribution of each fluorophore to detectors other than its own. The measurement is made empirically using single-stained controls: a sample stained with only one fluorophore is run, and the signal appearing in every other detector is recorded as a fixed proportion of the signal in its primary detector. Those proportions form a matrix, and applying its inverse to the raw data subtracts the spillover.
| Concept | What it means | Where it goes wrong |
|---|---|---|
| Spillover | Light from one fluorophore reaching another detector | Increases as panels grow and spectra crowd |
| Compensation | Subtracting the measured spillover contribution | Wrong controls give wrong coefficients |
| Spreading error | Increased variance left behind after subtraction | Cannot be removed; limits panel design |
| Autofluorescence | Natural cell emission unrelated to staining | Varies by cell type; needs an unstained control |
| Spectral unmixing | Fitting full spectra rather than band subtraction | Requires accurate reference spectra per fluorophore |
Two rules govern the controls and both are routinely broken. The single-stained control must use the same fluorophore conjugate as the experiment, since spillover is a property of the specific dye rather than the antibody. And the control must be bright enough and contain a clear negative population, because the coefficient is derived from the difference between positive and negative.
Compensation corrects the average but not the spread. When a large signal is subtracted, the statistical uncertainty in that signal remains, appearing as increased variance in the corrected channel. This spreading error is why a dim marker measured in a detector receiving heavy spillover may be impossible to resolve however carefully the compensation is set. Spectral cytometers address the problem differently, collecting the full emission spectrum of each event and fitting it as a combination of known reference spectra, which handles crowded panels better but depends heavily on the quality of those references.
Gating Strategies and Their Subjectivity
Gating is the process of selecting subsets of events for further analysis by drawing regions on plots. It is where the data becomes an answer, and it is unavoidably a matter of judgement.
A typical sequence starts broad and narrows. Exclude debris on a scatter plot. Exclude doublets using pulse geometry. Exclude dead cells using a viability dye. Select the broad population of interest, then apply marker-based gates to identify subsets within it. Each gate defines the denominator for everything downstream, and the sequence is not commutative: applying the same gates in a different order can produce different results when populations overlap.
The subjectivity is concentrated at boundaries. Marker expression is frequently continuous rather than bimodal, so the line between positive and negative is a decision rather than an observation. Where a clear negative population exists, the boundary can be anchored to it. Where it does not, controls that omit a single marker from the full panel provide the most defensible reference, since they account for spreading error from the other channels in the actual experimental context.
Rare populations magnify every one of these issues. Identifying a subset present at a very low frequency requires collecting enough events for the count to be statistically meaningful, and it requires the gating to exclude background reliably, because at low frequencies a small proportion of misclassified abundant cells can outnumber the genuine target.
The honest response to all of this is transparency. A gating strategy shown as a sequence of plots, with the controls used to set each boundary, allows a reader to evaluate the analysis. A final percentage presented alone does not.
Cell Sorting and Physical Separation
An analyser measures and discards. A sorter measures and then physically separates cells matching defined criteria, which turns cytometry from a descriptive technique into a preparative one.
The most common mechanism operates in a stream in air. After passing the interrogation point, the fluid stream is vibrated at high frequency by a piezoelectric element, breaking it into uniform droplets at a fixed distance from the nozzle. The instrument calculates how long a cell takes to travel from the laser to the point of droplet formation, and charges the stream at exactly the moment the droplet containing a target cell separates. Charged droplets pass between deflection plates and are steered into collection vessels; uncharged ones go to waste.
The timing is the delicate part. That drop delay must be calibrated precisely, and it drifts if pressure, nozzle condition or temperature change. An incorrect delay charges the wrong droplet and the sort purity collapses.
Sorting introduces trade-offs that analysis does not. Speed, purity and yield cannot all be maximised at once: strict modes that abort any droplet containing a non-target cell give high purity at the cost of recovery, while permissive modes recover more cells and accept contamination. Cells experience pressure, shear and electrical charge, all of which stress them, so viability after sorting is a real consideration for anything intended for culture or transplantation. Aerosol generation also makes sorting of infectious material a containment issue requiring appropriate engineering controls.
Alternative approaches avoid droplets entirely, diverting cells within a closed microfluidic channel. These are gentler and safer but generally slower, and they cannot deposit single cells into wells the way a droplet sorter can.
Reporting Populations as Percentages
Cytometry results are usually reported as percentages, and a percentage is meaningless without its denominator. The same subset can be legitimately described as a proportion of all events, of all live cells, of all lymphocytes, or of a specific parent subset, and those four numbers can differ by an order of magnitude.
This is not pedantry. A reported shift in a subset percentage can reflect a genuine change in that subset, or no change in it at all while the parent population moved. If a treatment expands one lineage substantially, every other subset expressed as a proportion of the total falls without any of them changing in absolute terms. The remedy is to report absolute counts wherever the sample allows it, using a known volume or a defined quantity of counting beads added to the tube, and to state the parent gate explicitly whenever a percentage is used.
Statistical confidence also deserves attention. The precision of a proportion depends on the number of events collected in the numerator, not on the total events acquired. Collecting a large number of events overall does not help if the population of interest contributed only a handful of them, and reporting a precise-looking percentage derived from very few events overstates what the data supports.
The broader point is that flow cytometry produces measurements of exceptional quality and then hands them to a process that is only as good as its documentation. The instrument’s contribution is reproducible; the analyst’s is not, unless the gating strategy, the controls and the denominators are all recorded alongside the result. Papers and reports that show that chain can be evaluated. Ones that show a single number cannot, and the difference between the two has more effect on whether a finding replicates than any specification on the instrument.
Frequently asked questions
Why do my results change when someone else analyses the same data file?
Because the raw file contains per-cell measurements, not populations, and populations are created by gating decisions. Two analysts will place boundaries slightly differently, may choose a different gating sequence, and may use different controls to define positivity. On well-separated populations the disagreement is small. On continuous markers, rare subsets or data with substantial spreading error it can be large. This is why laboratories running cytometry for clinical purposes use documented, standardised gating templates rather than leaving each analysis to individual judgement.
What does it mean when a marker looks positive on cells that should not express it?
Several possibilities need excluding before believing it. Inadequate compensation causes signal from another channel to appear in this one. Antibodies can bind non-specifically through Fc receptors, particularly on monocytes and other myeloid cells, which is why blocking reagents are used. Dead cells bind antibody indiscriminately and must be excluded with a viability dye. Autofluorescence varies substantially between cell types and can mimic a dim positive signal. Only after ruling those out does an unexpected positive become interesting biology.
How many events do I need to collect?
It depends entirely on the frequency of the population you care about, since the statistical uncertainty of a proportion is governed by the count in the target gate. A population making up a large fraction of the sample needs relatively few total events for a stable estimate. A population present at very low frequency requires enormous acquisition to accumulate enough target events, and the practical limit is usually sample volume rather than instrument time. Deciding the required target-event count before acquisition, rather than acquiring a round number of total events, is the more rigorous approach.
Is a spectral cytometer simply better than a conventional one?
It is better suited to some problems and unnecessary for others. Spectral instruments collect the full emission profile of each event and unmix it computationally, which allows fluorophores with similar peak emissions to be distinguished and makes large panels more tractable. They also handle autofluorescence more elegantly by treating it as another spectral component. The trade-offs are cost, a dependence on accurate reference spectra for every fluorophore and cell type used, and greater complexity in troubleshooting when unmixing goes wrong. For a modest panel with well-separated dyes, a conventional instrument is entirely adequate.
Why does the sample need to be a single-cell suspension?
Because the entire measurement model assumes one cell per event. Anything that presents to the laser as a single object is recorded as a single object, so a pair of stuck-together cells produces one event with combined scatter and combined fluorescence, which can place it in a gate neither cell belongs to. This is particularly damaging when looking for cells expressing two markers, since a doublet of two single-positive cells mimics a genuine double-positive cell. Tissue samples therefore require enzymatic or mechanical dissociation, filtering before acquisition, and doublet exclusion during analysis.
