Particle, signal, number.
A flow cytometer does little more than push cells single file past a laser and count flashes of light. So how does it tell one kind of cell from another, how small a particle can it see, and when can its count be trusted as a test result?
Before we start.
Most explanations of flow cytometry fit on one slide. Cells flow in single file through a laser beam, each cell scatters light and glows where fluorescent antibodies have bound to it, detectors measure the light, and software sorts the cells into populations on a plot. The slide is correct, and it leaves out almost everything an engineer, a founder or a newcomer to the field needs. It does not say why forward scatter is not a measure of size, why two dyes that look different to the eye are hard to tell apart in a detector, why a vesicle a tenth of a micrometer across is nearly invisible to an instrument that sees cells easily, why two laboratories can report different percentages from the same blood sample, or what a cytometer and its reagents must show before a hospital may use the result to diagnose a leukemia.
This guide is built around one question instead: a flow cytometer does little more than push cells single file past a laser and count flashes of light, so how does it tell one kind of cell from another, how small a particle can it see, and when can its count be trusted as a test result?
Here is the short answer, which the rest of the guide exists to justify. The instrument never touches a cell and never sees one. A stream of fluid lines particles up so that they cross a focused laser one at a time, each in a few microseconds. Every crossing sends out two kinds of light: light scattered by the particle itself, and fluorescence from dye molecules that antibodies have carried to particular proteins on its surface. Cells are told apart because different kinds of cell carry different proteins, so they glow in different combinations of color, and because their size and inner structure change how they scatter. How small a particle can be seen is set by photons: for particles much smaller than the wavelength of light, scattered light falls roughly with the sixth power of the diameter, so halving the diameter of a vesicle cuts its scatter about fiftyfold, and below a size that depends on the instrument its signal sinks into the background. A count can be trusted as a test result only when every step from particle to number has been made reproducible and checked: the instrument set up and calibrated in physical units, the reagents and the analysis fixed, the whole assay validated for its purpose, and, for a product, the evidence shown to a regulator.
One fact runs through every chapter. Every number in a cytometry file is a measurement of light, not of a particle. The particle, the signal it produced and the number written into the file are three different things. Most errors in cytometry, from a misplaced gate to a wrong count of vesicles, come from treating one of them as if it were another, and most of the engineering in a cytometer exists to keep the chain between them known.
How to read this guide
Chapters are numbered straight through, and each one opens with the question a careful reader would ask after the previous chapter. Plates are numbered separately so that any one can be cited on its own. Four kinds of box recur:
Blue edge. Carries the structural point of a section, or a worked calculation.
Orange edge. Names a common misreading, a trap, or the limit of a claim.
Green edge. Maps the idea onto the reader's own work: in instrument and assay design, at the bench, in verification, or in a regulatory file.
- Each chapter closes with what it established, in four lines.
The mathematics stays at percentages, ratios, powers of ten and squares. One relationship, the steep fall of scattered light with size, appears as a ratio worked out with real numbers, and a few statistical ideas appear as counts and spreads rather than formulas. Where a number is derived, the arithmetic is shown once with real values and the assumptions are stated. Where a plate is schematic rather than data, it says so. The guide uses US spelling.
Inside the plates, color is a legend and never decoration. Particles, the fluid that carries them, the laser light that goes in, the fluorescence that comes back and the light that particles scatter each have one color throughout the guide. The measured readout, the signal and every number computed from it, is always blue, and one orange mark in each plate points at the detail that matters most. A key strip under every plate lists only the colors that plate uses.
This is a reference for understanding, not a laboratory protocol, an engineering specification, a regulatory opinion or clinical advice. Regulatory status, standards editions and product details are stated as of October 2026 and will date. Where a figure is a manufacturer's claim, or comes from a study written by a manufacturer's staff, the text says so. Named instruments, reagents and software are examples of a principle, not recommendations. The letters FACS, used for fluorescence-activated cell sorting, also form part of the product names of Becton, Dickinson and Company; this guide writes "cell sorting" instead.
+ Part I · The particle, the signal, the number
What a cytometer records, and why it records particles one at a time.
A cytometry result arrives as a percentage or a count, and it is easy to read it as a direct observation of cells. These two chapters take it apart. They separate the particle that crossed the laser, the light it sent to a detector and the number written into a file, and they show what measuring every particle separately reveals that measuring a whole sample at once hides. Nothing here assumes more than school biology and physics.
What a flow cytometer actually measures.
+ The questionA report says that 12 % of the cells are CD4-positive. What did the instrument record to produce that number, and what did it never see?
A number with nothing visible behind it
A clinical report or a research figure states that 12 % of the cells in a blood sample carry a protein called CD4. The sentence sounds like an observation, as if someone had looked at the cells and counted. Nothing of the kind happened. The cells were stained with an antibody that carries a fluorescent dye, suspended in fluid and drawn into an instrument that pushed them, one at a time, through a focused laser beam. Each cell crossed the beam in a few microseconds. While it was there, it deflected some of the laser light and its dye molecules gave some light back at a longer wavelength. Detectors behind lenses and filters turned those flashes into electrical pulses, electronics turned each pulse into a few numbers, and software decided which of those numbers belonged to the cells the report is about.
The instrument repeated that sequence for every particle it could detect, typically several thousand per second. An analyzer of the kind found in clinical laboratories is specified by one manufacturer for up to 10,000 events per second, and a stream moving at a few meters per second carries each particle through a beam about 5 to 20 micrometers deep along the flow in a microsecond or a few. No image of any cell was taken, no cell was touched, and none was identified as a cell by the instrument. The 12 % is a ratio of two counts of numbers that passed through a series of decisions.
Three things that are easy to confuse
That sequence contains three different things, and this guide keeps them apart in every chapter.
The particle is the physical object: a lymphocyte, a red cell, a clump of two cells, a fragment of a dead cell, a calibration bead, a bubble, or a vesicle far smaller than any cell. The instrument has no direct access to it.
The signal is the light the particle sent toward a detector during its few microseconds in the beam, and the electrical pulse the detector made from that light. It depends on the particle, but also on the laser's power, the angle and area of the light collected, the filters chosen, the detector's settings and the background light that arrives whatever the particle does.
The number is what the electronics computed from the pulse and wrote into a data file: the height of the pulse, its area or its width, in arbitrary units that mean nothing until something converts them. Software then transforms these numbers for display, subtracts the overlap between dyes and groups them into populations by drawing boundaries called gates.
A cytometry result is trustworthy only to the extent that each step from particle to signal and from signal to number is known. The rest of the guide follows that chain: Part II covers how particles produce light, Part III how the instrument turns light into numbers, Part IV how numbers become populations, and Part V how the whole chain is engineered, calibrated and verified.
An event is not a cell
The data file does not contain cells. It contains events. The international data standard for cytometry files, FCS 3.1, defines an event as "an ordered list of the cytometric measurements for one particle." The minimum reporting standard for cytometry experiments, MIFlowCyt, calls it "a unit of data representing one particle (e.g., a cell) as detected by flow cytometer." Both say particle, and both give a cell only as an example.
The difference is not pedantic. Two cells that cross the beam stuck together, or one just behind the other, produce one pulse and one event. A guide to cell sorting written by two experienced operators notes that even with modern electronics "an event can still be 2 cells." Debris from dead cells, bits of membrane and particles in the reagents produce events too. In one public example file of stained, freeze-dried control blood cells, distributed with an open-source analysis package, a "not debris" gate drawn by a human analyst removed 23 % of all recorded events before any analysis began, and a second gate for single cells removed a further 4 %.
At the small end the difference becomes larger still. When vesicles far below the instrument's detection limit are present at high concentration, hundreds of them can be inside the beam at the same moment, and their combined light can cross the threshold as a single event. A 2012 study showed that one such "swarm" event could represent of the order of a hundred vesicles or more. Chapter 19 returns to it, because it is one of the ways a cytometer can report a confident number about particles it cannot see.
What becomes an event at all is decided by a setting. The operator chooses one detector, often forward scatter, as the trigger, and a level, the threshold, that its pulse must exceed. Every pulse below the threshold is discarded and leaves no trace in the file. FCS 3.1 provides an optional keyword, $TR, to record the trigger and threshold: "When the threshold is exceeded, an event is declared." Files from some instruments record the threshold in a manufacturer's own keyword instead, or not at all. A count of events is therefore always a count of pulses that were bright enough in one chosen channel, which is a property of the sample and the settings together.
An event is a pulse that crossed a threshold. It may be one cell, two cells, a fragment, or many vesicles at once, and a particle whose signal stayed below the threshold is not in the file at all. Any statement of the form "the sample contained N cells per microliter" depends on knowing which particles produced events, which did not, and what volume of sample was analyzed.
What the file does not contain
The numbers in an FCS file carry no physical units. A pulse area of 52,000 in a detector labeled for a green dye is not 52,000 of anything: it depends on the detector's voltage, the laser power and the optical path on that day and on that instrument. The standard provides an optional keyword, $PnCALIBRATION, for converting arbitrary units into a defined unit such as molecules of equivalent soluble fluorochrome, but nothing requires it, and most files do not carry one. Several properties that change the meaning of the numbers have no place in the standard at all: there is no keyword for the angle over which forward scatter was collected, and MIFlowCyt asks for it separately.
The file also leaves open what has already been done to the numbers. FCS 3.1 states that "the stored data is always uncompensated", meaning that the overlap between dyes has not yet been subtracted, while files written by some instruments mark their data as compensated in their own keywords. A number on a plot axis has usually been transformed again for display. The value stored in the file, the value after compensation and the value displayed are three different numbers, and a report must say which it used.
A percentage of positive cells is the number of events inside one gate divided by the number inside another. MIFlowCyt requires the denominator to be stated: "percentage of events within the gate shall be provided specifically stating the denominator." For the CD4 example, the numerator is events above a chosen boundary in the CD4 detector, and the denominator might be all lymphocytes, all white cells, or all events that passed the debris and single-cell gates. Each choice gives a different percentage from the same file. The boundary itself is a decision, made with the help of control samples (Chapter 11).
The vesicle problem
Most of the methods in this guide were developed for cells, which are typically 5 to 20 micrometers across and give bright, easily separated signals. This guide keeps returning to a much harder measurement. Extracellular vesicles are particles that cells release, delimited by a lipid membrane and unable to replicate; the MISEV2023 guidelines of the International Society for Extracellular Vesicles, published in 2024, describe small vesicles as "often" below 200 nanometers, while noting that measured size depends on the method. They are present in blood and other body fluids, and they are studied as possible markers of disease.
A vesicle of 100 nanometers is about a hundred times smaller in diameter than a lymphocyte. It scatters so little light that many instruments cannot distinguish it from the background, it carries only a few copies of any protein for an antibody to label, and it can arrive in such numbers that several are in the beam together. Measuring vesicles pushes every part of the chain from particle to number to its limit, which is why it serves as this guide's running case. Leukemia diagnosis, where the cells are large and bright but the consequences of a wrong number are serious, is the second case, in Part VII.
A cytometry result that will be compared, published or acted on should travel with the facts that make it interpretable: which pulse measurement was used (height or area), the trigger channel and threshold, the gates and the denominator of every percentage, whether values were compensated and transformed, and the instrument and settings. MIFlowCyt lists these as minimum reporting requirements. In a regulated assay they become part of the fixed procedure, and in an instrument they become requirements on what the software must record.
- A cytometer pushes particles one at a time through a laser and records the light each one sends to its detectors, typically thousands per second, each in a few microseconds.
- The particle, the signal and the number are different things; every result depends on the chain that links them.
- An event is a pulse that crossed an operator-set threshold: it may be a cell, two cells, debris or a swarm of vesicles, and particles below the threshold leave no record.
- The file holds arbitrary units without physical meaning unless calibrated, and every percentage depends on its gates and its stated denominator.
Why measure one particle at a time.
+ The questionA spectrophotometer measures a whole tube of cells at once. What does measuring every particle separately reveal that an average hides?
What an average cannot say
The simplest way to measure how much of a fluorescent label a sample of cells has taken up is to put the whole suspension in a cuvette, shine light through it and measure the total glow. A fluorometer does exactly that, and it is fast, cheap and precise. It is also blind to the question cytometry exists to answer, because it reports one number for millions of cells.
Suppose that 12 % of the white cells in a sample carry about 50,000 molecules of a dye each and the rest carry none. The average over all cells is 6,000 molecules per cell. A second sample in which every cell carries exactly 6,000 molecules gives the same total glow, the same average and the same reading on the fluorometer. Biologically the two samples could not be more different: one contains a distinct population of strongly labeled cells, the other a uniform population of weakly labeled ones. Only a measurement of each cell separately tells them apart.
Measuring each particle separately preserves three kinds of information that a bulk measurement averages away. The first is the distribution: whether a population is one group or several, and how wide each is. The second is rare events: a population that makes up 0.001 % of a sample, one cell in 100,000, changes a bulk signal by less than its noise but is countable if enough cells are measured one by one. The third is correlation: whether the same particle that is bright in one color is also bright in another, which is how cytometry separates helper T cells, which carry both CD3 and CD4, from monocytes, which carry some CD4 but no CD3.
Those three properties are why cytometry became one of the standard tools of immunology and hematology. Blood and bone marrow are mixtures of dozens of cell types that look alike under a light microscope and differ in the proteins on their surfaces. A disease such as leukemia shows up as an unusual population, sometimes a large one at diagnosis and sometimes, after treatment, a very small one hidden among normal cells. A method that measures every cell, and several properties of each, is the natural instrument for both.
Counting particles in a stream
Measuring one particle at a time began with counting. In 1949 Wallace Coulter filed a patent, granted in 1953, for a device that counted particles suspended in a conducting fluid as they passed one at a time through a small opening carrying an electric current. Each particle displaced some of the fluid and briefly changed the resistance of the opening, so each produced an electrical pulse whose size grew with the particle's volume. The Coulter principle is electrical rather than optical, and it is still used in the hematology analyzers that count red and white cells in routine blood tests.
Optical counting of cells in a flowing stream was explored from the 1930s onward, and in 1953 a paper described a way of confining the cells to the center of a flowing stream by surrounding them with a faster-moving sheath of clean fluid. This hydrodynamic focusing is the arrangement that almost every flow cytometer still uses to put particles in single file at a fixed point in a light beam (Chapter 6).
From absorption to fluorescence
In 1965 Louis Kamentsky and colleagues described an instrument that measured the absorption of ultraviolet light by individual cells, as an estimate of their nucleic acid content, at rates above 500 cells per second. In the same year Mack Fulwyler at Los Alamos described the first instrument that could separate cells. It measured each cell's volume by the Coulter principle. It then broke the stream into droplets by a method that Richard Sweet had just published for ink-jet recording, charged the droplets that contained the cells it wanted, and deflected them into a separate container. Neither instrument measured fluorescence.
Fluorescence arrived at the end of the decade. Wolfgang Dittrich and Wolfgang Göhde in Germany filed a patent in December 1968 and published in 1969 a method of measuring the fluorescence of single cells in a flowing suspension, and an instrument based on it was sold commercially soon afterward. Also in 1969, a Los Alamos group reported measuring fluorescence from cells excited by a laser, and Leonard Herzenberg's group at Stanford published the first sorting of cells according to their fluorescence. The Stanford group's 1972 paper, titled "Fluorescence Activated Cell Sorting," described a laser-based instrument, and Becton Dickinson brought commercial sorters to market in the early 1970s. The name "flow cytometry" came into general use in the second half of the decade.
The decisive change came from biology rather than engineering. In 1975 Georges Köhler and César Milstein described how to make monoclonal antibodies, identical antibodies produced without limit by one cloned line of cells. A monoclonal antibody binds to one site on one protein, and when a fluorescent dye is attached to it, it labels every cell that carries that protein. From the late 1970s onward, panels of such antibodies turned the cytometer from an instrument that measured size and DNA content into one that could identify cell types by the proteins they display.
More colors at once
Each additional color measured on the same cell adds a dimension in which populations can be separated. The Stanford group recalled in 2002 that they had started with two colors, fluorescein and rhodamine, and by then had a twelve-color instrument. A 2004 review described seventeen-color cytometry. In 2020 a published panel, two of whose three authors worked for the instrument's manufacturer, used forty colors on a full-spectrum instrument. In 2024 another described fifty; it was developed on instruments from two makers, and two of its five authors worked for one of them.
The growth did not come from adding more of the same detectors. Fluorescent dyes emit over broad, overlapping ranges of wavelength, so beyond a handful of colors the light of one dye inevitably reaches the detectors meant for others, and the instrument must separate them mathematically (Chapter 10). Full-spectrum or spectral instruments take this to its conclusion: instead of one detector behind one filter for each dye, they spread the light from each cell across an array of dozens of detectors and work out the contribution of every dye from the shape of the whole spectrum. The first commercial full-spectrum cytometer was announced by its maker in 2012 and launched in 2013, and Chapter 14 compares the two architectures as a systems engineer would.
Every added color brings more light from other dyes into each detector and more uncertainty into every value calculated from it. A forty-color panel is possible because of careful choice of dyes, detectors and controls, not simply because the instrument has enough detectors. A panel with fewer colors, chosen well, often resolves a given question better than a larger one chosen badly.
What stayed the same
Across seventy years the principle has not changed. A cytometer brings particles one at a time to a fixed point, measures what each one does to a beam of light or a current, and records numbers for each. Every generation has added detail to what is measured about each particle: volume, then absorption, then scatter and fluorescence, then many colors and whole spectra. None has changed the fact that the measurement is of a signal from a particle, not of the particle.
Sorting also stayed with the same idea. A sorter makes the same measurements as an analyzer and then acts on them within about a millisecond, charging a droplet that contains a chosen particle so that an electric field deflects it into a collection tube. This guide treats sorting in Chapter 24, as an extension of the analyzer rather than as a separate subject.
A bulk measurement is cheaper, faster and more precise for a quantity that is uniform across a sample, such as the total concentration of a soluble protein. Single-particle measurement earns its cost when the answer depends on a distribution, a rare population or a combination of markers on the same particle. Stating which of the three a measurement needs is the first step in specifying an instrument or designing an assay, because each one sets different demands on throughput, sensitivity and the number of colors.
- A bulk measurement reports one average and cannot distinguish a small bright population from a uniformly dim one; measuring each particle can.
- Single-particle data preserve distributions, rare events and correlations between markers on the same particle.
- Flow measurement grew from Coulter's electrical counting (patent filed 1949) through absorption and volume sorting (1965) to fluorescence (1968–1972), monoclonal antibodies (1975) and full-spectrum instruments (2013).
- Every generation added detail to the signal; none changed the principle that the cytometer measures light from a particle, one particle at a time.
+ Part II · Light meets matter
How a particle produces the light a cytometer measures.
Everything a cytometer reports begins as light leaving a particle. These three chapters explain the two ways that happens. A dye absorbs light and gives some of it back at a longer wavelength, which is fluorescence; the particle itself deflects light, which is scatter. A third chapter shows how antibodies and dyes are used to make particular proteins on a cell, or on a vesicle, into sources of fluorescence, and what decides how bright they look.
Why a molecule glows a different color.
+ The questionWhy does a dye give back light of a longer wavelength than the light it absorbed, and why does that difference make multicolor cytometry possible?
Light in, different light out
A fluorescent molecule absorbs a photon and, a few billionths of a second later, emits one. The emitted photon almost always has less energy than the absorbed one, which means a longer wavelength: a dye excited by blue light at 488 nanometers glows green, one excited by red light at 640 nanometers glows in the deep red.
The reason is what happens in between. Absorbing a photon lifts one of the molecule's electrons into a higher energy state, and usually into a vibrating version of that state with some energy to spare. The molecule sheds that spare energy as heat to the surrounding water within a tiny fraction of the time it spends excited. It then falls back to its lowest state by emitting a photon, from the bottom of the excited state, carrying less energy than the photon that arrived. A rule named for the chemist Michael Kasha states the regularity: molecules of this kind emit light with appreciable yield only from their lowest excited state, whichever higher state the absorbed photon reached.
The gap between the peak of absorption and the peak of emission for the same transition is the Stokes shift. The International Union of Pure and Applied Chemistry defines it between the band maxima of absorption and fluorescence "arising from the same electronic transition." It is usually expressed in frequency units rather than in nanometers, because a gap of 10 nanometers in the red corresponds to less energy than a gap of 10 nanometers in the blue.
The Stokes shift is what makes fluorescence measurable at all in a cytometer. The laser delivers far more light to the particle than any dye gives back, and much of it is scattered toward the detectors at the laser's own wavelength. Because the fluorescence comes out at longer wavelengths, an optical filter can block the laser light and pass the dye's glow. Without that difference in color, the faint emission would be lost in the glare.
How bright is a dye
Two properties of a molecule decide how much light it can give back. The molar absorption coefficient, often still called the extinction coefficient, measures how strongly the molecule absorbs light of a given wavelength. The quantum yield is the fraction of absorbed photons that come back out as fluorescence; the rest of the energy is lost as heat or through other processes. Fluorescein, the dye of the earliest two-color instruments, has a quantum yield of about 0.93 in alkaline solution, according to a technical report of the International Union of Pure and Applied Chemistry, so nine of every ten photons it absorbs return as light.
The product of the two is the usual measure of a dye's intrinsic brightness. A 2005 review of fluorescent proteins defined brightness in its tables as the "product of extinction coefficient and quantum yield," and the convention is common throughout fluorescence work. For the two protein dyes most used in cytometry, a 1982 paper and one manufacturer's handbook agree on R-phycoerythrin, or PE: an absorption coefficient of 1,960,000 and a quantum yield of 0.82. For allophycocyanin, or APC, the handbook gives 700,000 and 0.68 for the whole molecule of about 104,000 daltons; the 1982 paper gives the same quantum yield and 232,000 for a unit of 33,000 daltons, one of the molecule's three identical units, and three times that coefficient is about 700,000.
PE: 1,960,000 × 0.82 ≈ 1.6 million. APC: 700,000 × 0.68 ≈ 476,000. On intrinsic brightness, comparing whole molecules as they are attached to antibodies, PE is about 3.4 times brighter than APC. The units, liters per mole per centimeter, are the same for both and cancel in the ratio. The comparison assumes each dye is excited at its own absorption peak and that every emitted photon is counted, which no cytometer achieves.
Why the ranking does not survive the instrument
The same review that defined intrinsic brightness warned that perceived brightness depends on much more: the wavelength and intensity of the illumination, the spectra of the filters and mirrors, and the sensitivity of the detector to the emitted color. A cytometer imposes all three. A dye is excited by whichever laser line the instrument has, which may sit well away from its absorption peak. The filter in front of its detector passes only part of its emission. The detector converts photons to electrons with an efficiency that varies with wavelength, and background light from other sources sets the floor against which the signal must be seen.
The effect can be large. In one manufacturer's application note, measured on its own instruments, the separation of stained from unstained cells measured with the same antibody against CD4, once labeled with PE and once with APC, differed between the two dyes by only about 10 %, against the threefold difference in intrinsic brightness. The comparison mixes many factors, from laser power to the number of dye molecules on each antibody, and that is the point. The current community guidelines for cytometry in immunology state it plainly in a figure caption: "fluorochrome brightness can be instrument-specific."
A table of dyes ranked by intrinsic brightness is a starting point for panel design, not an answer. What matters in a cytometer is how well a dye separates labeled from unlabeled particles on that instrument, with its lasers, filters and detectors, and that has to be measured on the instrument.
A nanosecond inside a microsecond
A dye molecule typically stays excited for a few nanoseconds before it emits, while a cell spends about a microsecond or a few in the laser beam. In principle each molecule could therefore be excited and emit hundreds of times during one pass, and more laser power would give proportionally more light.
In practice it does not. At the high intensities in a focused laser spot, some molecules cross into a long-lived "dark" state from which they cannot be excited again for a while, and some are destroyed altogether, which is called photobleaching. A 1997 study measured fluorescein, PE and APC on cells and found them "considerably saturated and bleached in standard flow cytometric conditions." For APC excited with red light, raising the laser power about fourteenfold, from 35 to 510 milliwatts, raised the maximum fluorescence only about elevenfold, and long illumination at high power caused serious bleaching.
Saturation matters to an engineer for two reasons. Once a dye saturates, doubling laser power does not double the signal, so a more powerful laser buys less than its datasheet suggests. And because the shortfall depends on the dye, the intensity of the beam and how long the particle spends in it, two instruments with different lasers or flow speeds can rank the same dyes differently.
Broad colors that overlap
Fluorescence spectra are broad. A typical dye emits over a range of many tens of nanometers, with a long tail toward longer wavelengths, because the emitting molecule can land in many vibrational levels of its ground state. The Stokes shift of the protein dyes is small, about 13 nanometers for PE and 10 for APC, so cytometers excite them away from their absorption peaks and collect their emission with filters placed to one side of the laser line.
Breadth has a consequence that shapes the rest of the guide. When several dyes are used on the same particle, the tail of one dye's emission falls inside the filter meant for another. That overlap, called spillover, cannot be removed by better filters without throwing away most of each dye's light, and it is the reason a cytometer measuring many colors must separate them mathematically (Chapter 10).
When a panel is designed or an instrument specified, measure each candidate dye's separation of stained from unstained particles on the instrument itself, with the antibody it will actually carry, rather than relying on tables of intrinsic brightness or on another instrument's results. For an instrument specification, state sensitivity per detector with the laser line and filter, because a single figure for "sensitivity" cannot describe how each dye will look.
- A fluorescent molecule absorbs a photon, loses some energy as heat, and emits a photon of longer wavelength; that Stokes shift lets filters separate the glow from the laser light.
- Intrinsic brightness is absorption coefficient times quantum yield: PE is about 3.4 times APC by that measure.
- Measured brightness depends on the laser line, filters, detector and background, and on one manufacturer's analyzers PE and APC differed by only about 10 %; brightness is instrument-specific.
- Dyes saturate and bleach at cytometer intensities, and their broad, overlapping spectra cause spillover between detectors.
What scattered light can and cannot tell you.
+ The questionForward scatter is often read as a measure of size. Why is that wrong, and why does a vesicle of 100 nanometers scatter almost no light at all?
Light the particle itself redirects
Fluorescence needs a dye. Scatter needs nothing but the particle. Any object whose refractive index differs from that of the surrounding fluid deflects some of the light that strikes it, at the wavelength of the light that arrived, and a cytometer collects that deflected light at two places.
Forward scatter is the light deflected through small angles, from about one degree to roughly fifteen depending on the instrument, and collected by a lens or detector placed opposite the laser. The direct laser beam would swamp it, so a small opaque strip called an obscuration bar blocks the beam itself and lets only the slightly deflected light through. Side scatter is light deflected roughly at right angles to the beam, collected by the same high-aperture lens that gathers fluorescence and sent to its own detector through a filter at the laser's wavelength.
Because every particle scatters, scatter is the usual trigger that decides what becomes an event (Chapter 1), and the familiar first plot of almost every analysis is forward scatter against side scatter. On that plot, blood separates into three clouds that correspond, more or less, to lymphocytes, monocytes and granulocytes. That picture is the origin of two habits of speech: forward scatter as "size" and side scatter as "granularity."
What forward scatter is called, and what it is
Neither habit is a measurement. A 2011 study of how well optical signals track the volume of cells found that forward scatter area "performed poorly" as an indicator of volume. The authors listed why: forward scatter is influenced by the difference in refractive index between particle and fluid, by light-absorbing substances inside the particle, and by the optical design of the forward-scatter path. In some cell lines side scatter tracked volume better than forward scatter did. For particles comparable in size to the wavelength of light, the relation is not even monotonic: scattered light can rise and fall as size increases, because light waves scattered from different parts of the particle interfere.
A specification sheet that quotes a forward-scatter "resolution" of 1 micrometer describes how well the instrument separates beads of known size from one another. It is not a claim that forward scatter measures diameter, and the size it implies holds only for particles made of the same material as the beads.
Forward scatter depends on size, refractive index, shape, absorption and the instrument's collection angles. Side scatter depends on internal structure and on size as well. Both are useful for separating populations whose differences are already known, and neither measures a physical dimension unless the instrument has been calibrated with a model of the particle's optical properties (Chapter 17).
Refractive index decides as much as size
The refractive index of a material is how much it slows light compared with a vacuum, and scatter depends on the contrast between particle and medium. Water at 488 nanometers has an index of about 1.33. Polystyrene, the material of most calibration beads, has about 1.61 at that wavelength, and silica about 1.44. Living cells are close to water: a 2016 review gives about 1.35 for most normal cells, with higher values for some cancer cells and for red cells. Extracellular vesicles are closer still. Published estimates run from about 1.37 to 1.45, with values near 1.39 to 1.40 most often used, while lipoprotein particles in plasma, which outnumber vesicles, have higher indices, near 1.47 in one study.
The contrast matters enormously. At 100 nanometers, a sphere of polystyrene scatters about seventeen times as much light as a vesicle of the same diameter, and about ten times at 1 micrometer. A 2010 review noted that a gold particle of 200 nanometers scatters 27 times more than a polystyrene sphere of similar size, which in turn scatters 15 times more than a vesicle. A small change in the assumed index of a vesicle, from 1.40 to 1.36, cuts its expected scatter about sixfold.
The sixth power
For particles much smaller than the wavelength of light, scattering follows a simple law worked out by Lord Rayleigh: the scattered light grows with the sixth power of the diameter and falls with the fourth power of the wavelength, multiplied by a factor set by the refractive-index contrast. A 2010 review of vesicle detection put the consequence plainly: "If a microvesicle is only 10-fold smaller than another microvesicle, the scattering cross-section and thus the scattered amount of light decreases 10⁶-fold."
Under Rayleigh's law, halving the diameter divides scattered light by 2⁶ = 64. Vesicles of 100 to 200 nanometers are at the edge of that law in blue light, where the full theory of scattering by spheres, Mie theory, applies instead. A Mie calculation made for this guide, for vesicles of index 1.40 in water at 488 nanometers, gives a factor of 53 between 50 and 100 nanometers rather than 64. The same calculation gives a 1-micrometer polystyrene bead about 300,000 times the total scatter of a 100-nanometer vesicle, and a sphere the size of a lymphocyte about ten million times. A 2014 study reported the same order of magnitude: the smallest and largest vesicles in one sample typically differ 25-fold in size and 10,000,000-fold in scattered light.
That is why a cytometer that sees every white cell with ease can miss most vesicles entirely. A cell sends thousands of photons into the detectors in one pass. A 100-nanometer vesicle of low refractive index sends so few that the signal is indistinguishable from the light that the optics, the fluid and the detector produce with no particle present. A 2010 survey put the lower detection limit of commercial cytometers for polystyrene beads at 300 to 500 nanometers, and because vesicles scatter far less than polystyrene, the smallest vesicle detectable on the same instruments was larger still.
Seeing small particles with side scatter
The same physics points to the remedies. A particle much smaller than the wavelength scatters light over a wide range of angles, so the side-scatter path, with its large collecting lens, gathers more of it than the narrow forward path, which must also contend with glare around the obscuration bar. In the Mie calculation above, taking side scatter as 45 to 135 degrees and forward scatter as 1.5 to 15 degrees, a 100-nanometer vesicle sends about nineteen times more light to the side than forward. A shorter wavelength helps through the fourth-power term: moving from blue light at 488 nanometers to violet at 405 roughly doubles the scatter of a small particle, which is why some instruments offer violet side scatter for small particles.
How far an instrument can be pushed was shown in 2020, when a group rebuilt a common clinical analyzer step by step: a laser ten times more powerful, a photomultiplier in place of the forward-scatter photodiode, a redesigned obscuration bar, a pinhole to reject stray light and a narrower sample stream. The changes improved separation of particles from noise about 30-fold on side scatter, and gave estimated detection limits for vesicles of index 1.40 of 246 nanometers on forward scatter and 91 nanometers on side scatter.
Beads cannot substitute for that calculation. Because polystyrene scatters so much more than a vesicle, a 100-nanometer polystyrene bead gives about the same side scatter as a vesicle of about 175 nanometers. In 2012 a widely used bead mixture for setting size gates was shown, on one instrument, to select single vesicles between 800 and 2,400 nanometers across. The reporting framework for vesicle cytometry, published in 2020, concludes that "a gating strategy based on polystyrene beads alone is not a sound standardization methodology."
An instrument meant for small particles should be specified by its scatter detection limit for a stated material and refractive index, at a stated wavelength and collection angle, rather than by a bead size alone. A method that reports particle size from scatter should state the optical model it used, the refractive index it assumed and the instrument's collection angles, as Chapter 17 describes.
- Scatter is laser light deflected by the particle itself: forward scatter at small angles, side scatter near right angles; it needs no label and usually triggers events.
- Forward scatter is not size and side scatter is not granularity: both depend on refractive index, shape, absorption and the instrument's optics.
- Refractive index matters as much as diameter: at 100 nanometers, polystyrene scatters about 17 times more than a vesicle of the same size.
- Below a few hundred nanometers scatter falls roughly with the sixth power of diameter, so small vesicles vanish into the noise unless the optics are built and calibrated for them.
Turning a molecule into a light source.
+ The questionA cell carries a few thousand copies of a protein. How do antibodies and dyes make each copy give off light, and what decides how bright the result looks?
Making a protein visible
Scatter says little about what kind of cell a particle is. The proteins on its surface say a great deal, because different kinds of cell display different combinations of them. Helper T cells carry CD3 and CD4; B cells carry CD19; immature blood cells carry CD34. The CD numbers come from an international scheme, the clusters of differentiation, that names surface proteins recognized by antibodies.
A cytometer cannot see a protein, so the protein is made into a source of light. An antibody that binds to one site on it, its epitope, is chemically joined to a fluorescent dye, and the conjugate is mixed with the cells. Each antibody that binds brings its dye molecules with it, so the fluorescence of a stained cell grows with the number of antibodies bound, which in turn depends on how many copies of the protein the cell carries. That number, the antigen density, differs widely between proteins and between cell types.
Calibrated measurements give a sense of the range. A 2024 study at the US National Institute of Standards and Technology treats about 40,000 antibodies bound per cell as the known value for CD4 on the CD4-positive T cells of fixed whole blood with one widely used reagent, and it measured roughly 9,000 to 11,000 for CD19 and 4,000 to 5,000 for CD22 on control cells. Some proteins of interest are present at a few hundred copies; others at hundreds of thousands. An antibody against a protein with a few thousand copies must carry a bright enough dye, or the stained cells will not separate from unstained ones.
Three families of dye
The dyes attached to antibodies fall into three families, and each brings different strengths and weaknesses.
Small-molecule dyes, such as fluorescein and its modern successors, are compact and stable, and several can be attached to one antibody. They are moderately bright. Phycobiliproteins are light-harvesting proteins from algae and cyanobacteria, introduced as labels in 1982. They are extraordinarily bright and enormous: R-phycoerythrin weighs about 240,000 daltons and allophycocyanin about 104,000, so a single PE molecule is heavier than the antibody it is attached to. Polymer dyes, introduced for cytometry in the 2010s, are long chains of light-absorbing units that act as one large chromophore; they made bright labels available for the violet and ultraviolet lasers, where few had existed.
Tandem dyes extend the range of colors that one laser can excite. A tandem joins two dyes so closely that the first, the donor, passes its absorbed energy to the second, the acceptor, without emitting, and the acceptor emits at its own longer wavelength. A PE–cyanine tandem is excited by the same laser as PE but glows in the far red. The gap between the donor's absorption and the acceptor's emission is large, but it is not a Stokes shift in the strict sense, which applies to a single transition (Chapter 3).
Tandems have a weakness that matters in practice: the link between donor and acceptor can fail. Light, extremes of temperature, prolonged fixation and processes in some living cells cause the acceptor to stop receiving energy, so the tandem loses its far-red emission and the donor glows in its own color instead. The extent varies from one manufacturing lot to another. A 2006 guideline on controls, written by staff of one manufacturer, warned that, with tandems, substituting a different antibody for compensation than is present in the test samples "is a risky practice" (Chapter 10).
Brightness meets antigen density
What matters in a measurement is not a dye's brightness but how well it separates labeled cells from unlabeled ones. The common measure is the stain index: the difference between the median fluorescence of the positive and negative populations, divided by twice the standard deviation of the negative population. A large stain index means the positive cells sit far above the spread of the negatives.
Suppose the negative population has a median of 200 and a standard deviation of 100 in a detector, and the stained population a median of 6,200. The difference is 6,000, twice the spread is 200, and the stain index is 6,000 ÷ 200 = 30. Lowering the stained median to 3,200, which halves its distance above the negatives, halves the index to 15. Doubling the negatives' spread to 200 also halves it, even though no dye molecule changed: a wider negative population, from background or from other dyes' light, is as damaging as a dimmer stain. Software sometimes uses a robust version of the standard deviation, so a stain index should state which it used.
A manufacturer published stain indices measured on two of its analyzers for fifteen dyes, all attached to the same antibody clone against CD4. The brightest, a PE–cyanine tandem, gave 353, PE 302 and APC 278, while fluorescein gave 56 and two dyes gave 25 or less. The same note warned that, given the variety of instrument configurations, it is impossible to define the best combinations of dyes for six, eight or more colors.
Panel design follows from these numbers. The general practice is to pair the brightest dyes with the proteins present at the fewest copies, and the dimmer dyes with abundant proteins that separate easily anyway. The practice is a design rule rather than a law of physics: it has to be weighed against how much light each dye throws into the other detectors, which Chapter 10 shows can make a bright dye costly to its neighbors.
Labels that are not about proteins
Some dyes report the state of the cell rather than a protein. Viability dyes are kept out of living cells by an intact membrane and enter dead ones, either binding DNA or reacting with the proteins inside. Dead and dying cells matter because they bind many antibodies nonspecifically, so in a stained sample they appear as false positives in almost every color. Excluding them with a viability dye is one of the simplest ways to make positive populations real. DNA dyes measure the DNA content of each cell, which is how cytometers measure the cell cycle and some abnormalities of chromosome number.
Cells can also bind antibodies through Fc receptors, proteins on some white cells that hold antibodies by their constant tail rather than by their binding site. That binding is unrelated to the protein the antibody was meant to detect, and blocking it, with excess unlabeled antibody or a blocking reagent, is part of many staining procedures.
A vesicle carries a thousandth of the label
For vesicles the label is the hard part, not just the scatter. A vesicle of 100 nanometers has a surface area of about 0.031 square micrometers, roughly a nine-hundredth of the surface of a platelet 3 micrometers across. If a protein were spread at the same density on a vesicle's membrane as on its parent cell, the vesicle would carry about a thousandth of the parent cell's copies and give about a thousandth of its fluorescence. The estimate is geometry only: proteins are not sorted into vesicles evenly, and many vesicles may carry none of a given protein.
A signal a thousand times dimmer than a cell's lies close to the background of most detectors. The most sensitive instruments in a 2025 comparison across many laboratories detected vesicle labels of the order of 10 to 100 molecules of equivalent soluble fluorochrome, a calibrated unit explained in Chapter 17, and many instruments could not reach that. A vesicle that carries only a handful of antibodies is therefore near the limit of what a well-built cytometer can register at all.
Degraded tandems glow partly in the donor's color, so a tandem-stained population can appear falsely positive in the donor's detector and dimmer in its own. Controls for compensation must use the same reagent, from the same lot, handled the same way as the stained sample. A control made with a different lot or an older vial can introduce errors that look like biology.
In an assay specification, treat each antibody–dye conjugate as a component with its own identity: clone, dye, lot, storage and handling limits, and the instrument channel it is assigned to. Record antibodies bound per cell where calibration allows, so that the brightness a panel needs can be stated in physical terms. A change of clone, dye or lot is a change to the assay and must be checked as one.
- Antibodies conjugated to dyes make particular surface proteins into sources of light; fluorescence grows with the number of antibodies bound, about 40,000 per CD4-positive T cell for CD4 with one reagent.
- Small-molecule, phycobiliprotein and polymer dyes differ in brightness and size; tandems extend colors through energy transfer but degrade, and vary by lot.
- The stain index measures separation, not brightness: on one manufacturer's analyzers PE scored 302 and fluorescein 56 on the same antibody.
- A 100-nanometer vesicle carries roughly a thousandth of the label of a 3-micrometer platelet at the same protein density, close to the detection limit of most cytometers.
+ Part III · The instrument, piece by piece
How light from a particle becomes a number.
Part II described the light a particle can give. These four chapters follow that light through the instrument: the fluidics that bring each particle to one point in the beam, the lasers that illuminate it, the optics and detectors that collect what comes back, and the electronics that turn a flash lasting a few microseconds into numbers in a file. Each part has a job that can be stated simply and done well only with care, and each sets a limit on what the instrument can see.
One particle at a time past a fixed point.
+ The questionHow does a stream of fluid line particles up in single file at one point in a laser beam, and what happens when two of them arrive together?
Bringing particles to a point
A cytometer measures how much light each particle sends to its detectors. That comparison is fair only if every particle receives the same illumination, and a focused laser beam is not uniform: its intensity is highest at the center and falls away toward the edges, usually in the bell-shaped profile called Gaussian. A particle that passes through the edge of the beam glows less than an identical particle that passes through its center. The fluidics therefore have two jobs. They must bring particles to the beam one at a time, and they must bring every particle through nearly the same spot.
Most commercial cytometers do this by hydrodynamic focusing. The sample is injected through a fine tube into the middle of a faster-flowing stream of clean fluid, the sheath, inside a channel that narrows. The narrowing accelerates the whole flow and squeezes the sample into a thin thread along the axis, the core stream, without any wall touching it. The particles in the core are carried in single file through the laser beam, which crosses the stream at a fixed point called the interrogation point.
A guide to high-speed cell sorting summarizes the trade-off in two sentences: the diameter of the core stream "is dependent on the differential pressure between the sample and the sheath fluids," and "as the sample pressure is increased to raise the cell analysis rate, the ability to resolve single cells will decrease."
How wide the core is
The width of the core follows from conservation of volume. The sample that enters each second must pass through the core each second, so for a core moving at a given speed, its cross-sectional area grows in proportion to the sample flow rate and its diameter with the square root of the rate.
One manufacturer gives the particle velocity in the flow cell of an older, once widely used analyzer as about 6 meters per second. At 12 microliters per minute, a core moving at that speed must be about 6.5 micrometers across; at 60 microliters per minute, about 14.6 micrometers. Quadrupling the sample rate doubles the core's diameter. Now compare the beam, taking it as Gaussian and 66 micrometers wide between the points where its intensity falls to 13.5 % of the center, the 1/e² width. A particle at the outer edge of the 6.5-micrometer core receives about 98 % of the light at the center, while one at the edge of the 14.6-micrometer core receives about 91 %. These figures are derived for this guide from the stated velocity and beam size; they assume the core moves as a uniform plug and that the 66 micrometers quoted is the 1/e² width.
The consequence is that running a sample faster does not just count it faster. Particles spread across a wider core sample a wider range of beam intensities, so a population that is physically uniform produces a broader spread of signals. The community guidelines for cytometry in immunology note that increasing the flow rate on instruments with hydrodynamic focusing increases coincidence "and thus higher CV," the coefficient of variation that measures the width of a population's signal. Instruments built for precise measurement of DNA content, where a narrow peak is the whole point, are run at low sample rates for this reason.
Speed and time in the beam
The speed of the stream sets how long each particle is illuminated. The same manufacturer's 6 meters per second, with a beam about 22 micrometers high along the direction of flow, gives a lymphocyte of 7 micrometers a pulse lasting about 5 microseconds: the time it takes to travel the beam's height plus its own diameter. In a jet-in-air sorter, where the stream leaves the nozzle as a free jet, speeds are higher. The sorting guide gives about 28 meters per second at a typical pressure, so that the cell is in the laser beam for about 900 nanoseconds.
Where the measurement happens also matters. In a flow cell analyzer the laser meets the stream inside a small quartz cuvette, and a lens with a high numerical aperture, around 1.2, can be coupled closely to the glass to collect light over a wide cone. In a jet-in-air sorter the beam strikes the free stream, which acts as a small cylindrical lens and scatters a bright line of laser light that must be blocked, taking some fluorescence with it. The sorting guide notes that cuvette systems "can be more sensitive than jet-in-air sorters (particularly notable at certain wavelengths, e.g., long red)," and a 2018 comparison of 23 instruments found the jet-in-air sorters among the instruments with the lowest detection efficiency (Chapter 8).
Acoustic focusing
A second way to line particles up uses sound. In a channel driven by ultrasound at a frequency chosen so that a standing pressure wave forms across it, particles are pushed by the acoustic radiation force toward the pressure node at the center. A cytometer can then position particles without relying only on a large flow of sheath. The method was developed at Los Alamos in the 2000s, and the first commercial acoustic-focusing cytometer was announced in December 2009, according to the manufacturer's press release. Because the particles are positioned before they reach the beam, the sample can move more slowly through it and be illuminated for longer, or be run at higher volumetric rates without the core widening in the same way. The current manufacturer's claims include sample rates up to 1,000 microliters per minute.
When two arrive together
Particles in a suspension do not arrive at evenly spaced intervals. They arrive at random, independently of one another, so the number arriving in any short interval follows the statistics named for the mathematician Siméon Denis Poisson, the same statistics that describe raindrops falling on a paving stone. Some particles arrive so close behind the one before that the electronics cannot separate their pulses.
If events arrive at an average rate R and the electronics need a window of time τ to process each one, the expected number of other arrivals within a window is R × τ, and the chance that a given event has a neighbor inside its window is 1 − e^(−Rτ). With a window of 3 microseconds: at 1,000 events per second, about 0.3 %; at 10,000, about 3 %; at 30,000, about 9 %; at 100,000, about 26 %. With a 10-microsecond window the figures roughly triple at low rates. These are illustrative figures derived for this guide; real windows are instrument settings.
When two pulses overlap, the instrument either records them as one event, which may later look like a single bright or large cell, or discards them as an electronic abort. FCS 3.1 provides optional keywords for two kinds of loss: $ABRT for events lost to coincidence in the electronics, and $LOST for events lost because the computer was busy. Neither records pulses merged into one event. A count of particles in a measured volume, as in volumetric counting and the vesicle concentrations of Chapter 17, is low by the fraction lost unless the instrument corrects for it; the bead-ratio method of Chapter 21 cancels losses that strike beads and cells alike.
For particles much smaller than the beam, coincidence takes a different form. Several of them can be inside the illuminated volume at the same moment. A 2012 study calculated the illuminated volume of the same analyzer at its highest sample rate as about 54 picoliters, taking the core to be 56 micrometers wide; the 6-meters-per-second figure above gives a core of about 15 micrometers at that rate, and about 4 picoliters. The two published figures cannot both hold. At a concentration of ten billion particles per milliliter, of the order reported for vesicles in plasma, even the smaller volume holds dozens of particles at any instant, and the larger one hundreds. Individually each is too faint to cross the threshold; together their light can. Chapter 19 calls this swarm detection and shows how dilution reveals it.
Raising the sample rate widens the core, broadens the spread of signals and raises the fraction of coincident events. A result obtained at one flow rate is not automatically comparable with one obtained at another, and an assay that has been validated at a stated rate should be run at that rate.
In an instrument specification, state the measurement precision, as the coefficient of variation of a reference bead population, as a function of sample flow rate, and state the maximum event rate together with the coincidence or abort fraction at that rate. In an assay procedure, fix the sample flow rate, the sample concentration and the dilution, and record them with the data. For small particles, include a dilution series to show that counts fall in proportion to dilution.
- Hydrodynamic focusing squeezes the sample into a thin core inside a sheath so that particles cross the laser one at a time at nearly the same point of the beam.
- The core's diameter grows with the square root of the sample flow rate; a wider core means particles see more varied illumination and populations look broader.
- Particles cross the beam in about a microsecond in a jet-in-air sorter and a few microseconds in an analyzer or cuvette sorter; cuvettes collect light better than jets in air.
- Arrivals are random, so coincidence rises with event rate, from about 3 % at 10,000 events per second in a 3-microsecond window, and small particles at high concentration can sit in the beam by the dozen, or by the hundred if the wider published core is right.
Why the lasers cross the stream at different places.
+ The questionSeveral lasers cross one stream of particles. How does the instrument choose them, shape them, and know which flash of light belongs to which particle?
Why lasers
A cytometer needs a source of light that can be focused to a spot between about 5 and 22 micrometers high, delivers enough power to make a dim particle glow measurably in a microsecond, and has a single, narrow color, so that a filter can block it completely and pass the fluorescence beside it. A laser does all three. Early instruments used mercury arc lamps, and some still use light-emitting diodes for particular purposes, but the laser has been the standard source since the 1970s.
The colors in use come in families set by the dyes they excite. A 2015 review of near-infrared lasers in cytometry lists ultraviolet at 355 nanometers, violet diodes at 405, blue at 488, which was originally produced by argon-ion lasers, green and yellow lines at 532 to 561 that suit phycoerythrin and the red fluorescent proteins, red at 633 to 640, and near-infrared diodes from 660 to 730 nanometers. Some current instruments go further into the infrared. A current five-laser analyzer, according to its manufacturer, carries lines at 355, 405, 488, 561 and 637 nanometers at powers between 60 and 200 milliwatts.
Matching lines to dyes
Each laser excites the dyes whose absorption bands it falls within, and each dye can be paired with only the lasers it absorbs. Phycoerythrin absorbs strongly in the green and yellow, so it is excited well by a 561-nanometer laser and less well, though still usefully, by the blue line at 488. Allophycocyanin needs a red laser. The polymer dyes of Chapter 5 were developed largely to make the violet and ultraviolet lasers useful for bright labels.
Adding lasers is the most effective way to add colors. Dyes excited by different lasers can emit at similar wavelengths and still be told apart, because their light arrives at the detectors at a different moment and through a different collection path, as the rest of this chapter shows. A cytometer with five lasers can therefore use many more dyes, with less overlap between them, than one with a single laser and the same number of detectors.
Shaping the spot
The beam is focused to an ellipse rather than a circle. Its short dimension lies along the direction of flow and sets how long each particle is illuminated, and therefore how close behind one another two particles can pass and still be resolved as two (Chapter 6). Its long dimension lies across the stream and gives tolerance: a particle slightly off the axis of the core still receives nearly the full intensity. Manufacturers quote spots of 22 by 66 micrometers on an older analyzer, 9 by 65 on a newer one and 5 by 80 on another. Some instruments shape the beam so that its intensity is flat across the center rather than Gaussian, to make the illumination less sensitive to where a particle rides; that is the design intent, and this guide found no published comparison that measured the benefit.
The mean intensity of a focused beam is its power divided by the area of the spot. Twenty milliwatts focused into an ellipse 9 by 65 micrometers gives about 4 × 10⁷ watts per square meter, roughly forty thousand times the intensity of sunlight at the Earth's surface. A 1997 study described fluorescein, PE and APC as considerably saturated and bleached in standard cytometer conditions (Chapter 3). Its clearest case, APC, gave only about eleven times more light for fourteen times more power, at a mean intensity of about 10⁹ watts per square meter. On these figures, typical analyzers run one to two orders of magnitude below that intensity, a 50-milliwatt source in a 5-by-80-micrometer spot comes within a factor of ten of it, and sources of 100 to 200 milliwatts focused to a similar spot approach it. Fluorescein and PE lost signal within microseconds in the same study, so a lower intensity does not guarantee a linear response. The powers quoted by manufacturers are at the laser, so the intensity at the stream is lower still. The comparison is derived for this guide.
The intensity explains a choice that otherwise seems puzzling. Jet-in-air sorters move particles faster than analyzers, so each particle spends less time in the beam, and the sorting guide of Chapter 6 notes that "higher laser powers are generally required to illuminate the cell with the same number of photons." For dyes that already saturate, extra power buys little extra signal, and it can bleach a dye partway through the beam. The best power for a measurement depends on the dye, the flow speed and the spot.
One stream, several beams
An instrument with several lasers could focus them all on the same spot. Most do not. They place the beams at separate points a short distance apart along the stream, so that each particle passes through one laser, then the next. The light produced at each point is collected by its own optics and sent to its own set of detectors. A dye excited only by the red laser then produces a signal only at the red point, even if its emission overlaps with that of a dye excited at the blue point.
Separating the beams in space separates them in time. A particle crosses the first beam, which usually triggers the event, and reaches the next one microseconds or tens of microseconds later. The electronics must assign the second pulse to the same event by waiting a fixed interval, the laser delay, before looking for it. The delay depends on the distance between the beams and the speed of the stream, and the speed depends on the sheath pressure, so the delay must be measured with beads and checked as part of routine setup.
A wrong delay does not stop the instrument working. It makes the second laser's detectors integrate the wrong part of the pulse, or the tail of the previous particle. A note from one core facility shows a delay corrected from −93 to −91 microseconds: before the correction, a population that should have been almost entirely positive for two markers read 89.4 % double-positive; after it, 97.6 %. Two microseconds is a large part of a pulse at analyzer speeds; in the same note the affected channel's fluorescence intensity read about 11 % lower before the correction, enough to push part of the population out of the positive gate.
Whether a fluorescence value belongs to the right particle depends on the delay setting, yet FCS 3.1 has no keyword for laser delays, windows or sheath pressure. Some instruments write them into their own keywords, in units the standard does not define. A data set exchanged between laboratories can therefore lose the one setting that decides whether its multi-laser values are valid.
For an instrument specification, state each laser's wavelength, its power at the interrogation point rather than at the laser, the spot dimensions and profile, and the spacing between beams. For daily operation, include laser-delay verification with beads in the setup procedure, record the delays with the data, and repeat the check after any change of sheath pressure, nozzle or flow cell. MIFlowCyt asks for the power measured "at the intersection of the light source beam with particles" where possible.
- Lasers give a narrow, intense, focusable color; current instruments use lines from about 355 to beyond 730 nanometers at tens to a couple of hundred milliwatts.
- Each laser excites only the dyes that absorb at its line, and adding lasers adds colors with less overlap than adding detectors to one laser.
- The spot is an ellipse, short along the flow and long across it; typical analyzers run one to two orders of magnitude below the intensity at which APC clearly saturated in a 1997 study, though some dyes lose signal at lower intensities.
- Separate beams along the stream are separated in time, and a laser delay assigns each pulse to its event; a delay off by two microseconds can turn 97.6 % into 89.4 %.
Whether a photon becomes a count.
+ The questionOf the photons a particle sends out, how many reach a detector, and what sets the dimmest signal the instrument can see?
Most of the light never arrives
A fluorescent particle emits light in every direction. A cytometer collects it with a lens placed close to the stream, and the fraction it can gather is set by the lens's numerical aperture, a measure of the width of the cone of light it accepts. Analyzers with a quartz flow cell use lenses with numerical apertures of about 1.2 coupled to the cuvette with optical gel, according to their manufacturers, and some claim more than 1.3.
Even so, most of the light is lost. For light emitted equally in all directions into water, a lens of numerical aperture 1.2 collects about 28 % and one of 1.3 about 39 %; a lens of 0.5 in air would collect less than 7 %. The collected light then passes mirrors and filters that let through only part of each dye's emission band, and finally reaches a detector that converts only a fraction of the photons that strike it into electrons.
Take a dye that emits 10,000 photons while a particle crosses the beam. A lens of numerical aperture 1.2 collects about 2,800. A filter that passes half of the dye's emission band leaves about 1,400. A photomultiplier with a quantum efficiency of 25 % at that color converts about 350 of them into photoelectrons. The figures are illustrative, derived for this guide; real budgets depend on each component, and losses at every surface and mirror reduce them further. The pattern holds generally: of every hundred photons emitted, a few become the signal.
Filters and mirrors sort the colors
The collected light contains every color the particle produced. A series of dichroic mirrors divides it by wavelength: a long-pass mirror reflects light shorter than a stated wavelength and transmits longer light, and a short-pass mirror does the reverse. Each branch then passes through a bandpass filter in front of its detector, which defines exactly which colors that detector sees. The notation 530/30 means a filter centered on 530 nanometers with a band 30 nanometers wide, passing light from 515 to 545 nanometers. The order of mirrors along the path matters, because each mirror decides which colors continue, and MIFlowCyt asks that the components be listed in that order.
Filters age. Their coatings degrade with time, humidity and intense light, and MIFlowCyt asks for each filter's installation date because "filter performance needs to be monitored at intervals." A filter that has drifted changes what fraction of each dye reaches each detector, which changes the overlap that Chapter 10 subtracts.
Three kinds of detector
The detector converts photons into an electrical signal, and three types are in use. The photomultiplier tube has been the standard for decades. A photon striking its photocathode frees an electron, and a chain of electrodes multiplies that electron by a factor of up to about ten million, depending on the voltage applied, into a measurable pulse. A manufacturer's datasheet for one common tube gives a quantum efficiency of about 30 % at its best wavelength, 260 nanometers in the ultraviolet, lower across the visible and falling toward the red, and a gain of about ten million at 1,000 volts.
The avalanche photodiode is a semiconductor that converts photons more efficiently, up to about 80 % in the red and near-infrared according to one manufacturer's comparison, but with much less internal gain, so its own electronic noise matters more. The silicon photomultiplier is an array of tiny photodiodes operated so that each one fires on a single photon, with high gain and, in the same manufacturer's comparison, a photon detection efficiency of up to about 50 %, at the cost of more noise. Instruments now use all three: photomultiplier tubes in many analyzers and spectral instruments, arrays of avalanche photodiodes in others, and silicon photomultipliers in some.
No type is best everywhere. The choice trades efficiency at each color against noise, dynamic range and cost, and it changes how dim a signal can be detected in each channel, which is why the rest of this chapter measures detectors by what they deliver rather than by type.
Counting photoelectrons
The signal from a dim particle is ultimately a count of photoelectrons, and counts of independent random events fluctuate. A 2006 paper put the rule for cytometry precisely: the spread of repeated measurements "scales with the square root of the number of photoelectrons detected." A population of identical particles that each give 100 photoelectrons on average shows a spread of about 10 % from counting alone; at 1,000 photoelectrons the spread falls to about 3 %, and at 10,000 to 1 %.
This is why dim populations always look broad, however uniform the particles, and why a dim population cannot be separated from an unlabeled one by better software. The width is physics, set by how many photoelectrons the instrument extracts from each particle.
Q and B decide the dimmest signal
Two numbers describe what a detection channel can do. Q, the detection efficiency, is the number of photoelectrons the channel produces per unit of dye on the particle. B, the background, is the light and electronic noise that the channel registers even when no labeled particle is present, from the sheath fluid, stray laser light, the optics and the electronics. A dim signal is lost either because it produces too few photoelectrons or because it sinks into the background.
A 2018 study measured Q and B on 23 instruments in nine laboratories, using light pulses from a stable LED source and multi-level beads; its authors included staff of a cytometer manufacturer and of an analysis-software company. Among eight analyzers of the same model, Q ranged over not much more than a factor of two in some channels and over an order of magnitude in others, and the method identified defective channels in need of service. The three jet-in-air sorters had Q values far below the median, and the one instrument that used avalanche photodiodes had the highest Q in every channel.
The same study defines the dimmest resolvable signal as the level at which a positive population, two of its own standard deviations below its mean, still sits two standard deviations above the negatives. Counting statistics plus background give that level as 4 × (1 ÷ Q + the background's standard deviation), both in units of dye. One manufacturer published example values for one channel: Q = 0.306 photoelectrons per unit of dye and B = 326 units of dye, the background light expressed as an equivalent dye signal; its counting spread is √(B ÷ Q) = √(326 ÷ 0.306), about 32.6 units. The counting term 1 ÷ Q is then about 3.3 units and the background term 32.6, so the dimmest resolvable signal is about 4 × 36 ≈ 144 units. Background, not the detector's efficiency, dominates. The arithmetic is derived for this guide from the published formula and values.
The rule of thumb that follows is that background decides how dim a signal can be seen, and Q decides how precisely a bright one is measured. For vesicles, which sit at the bottom of the scale, background is everything. Their compendium of methods notes that, unlike cells, "signals originating from EVs partly overlap with the background noise." A 2014 study gave lower size limits for vesicles of 270 to 600 nanometers on a conventional cytometer and 150 to 190 nanometers on one built for small particles, and the 2020 rebuild described in Chapter 4 gained a fivefold improvement in side-scatter separation from a pinhole that rejected stray light alone.
Instrument specifications often state sensitivity as a number of fluorescent molecules equivalent to the faintest detectable signal. A study coordinated by the US National Institute of Standards and Technology and the International Society for Advancement of Cytometry found that values assigned to the same reference beads by different groups "varied by orders of magnitude in most cases." A figure measured with one maker's beads and method cannot be compared with another maker's figure, and it describes one channel, not the instrument.
State detection performance channel by channel, as Q and B measured by a stated method, or as a calibrated detection threshold with its method and bead lot, rather than as one sensitivity figure for the instrument. Track Q and B over time as part of performance qualification, because a falling Q or a rising B reveals an aging detector, a degraded filter or a misaligned laser before it shows up in patient or experimental data.
- A lens of numerical aperture 1.2 collects about 28 % of the light; filters and detector efficiency leave only a few percent of emitted photons as signal.
- Dichroic mirrors and bandpass filters such as 530/30 decide which colors reach which detector; filters age and must be monitored.
- Photomultipliers, avalanche photodiodes and silicon photomultipliers trade efficiency, gain and noise; counting statistics make dim populations broad, with a spread of about 1 ÷ √N.
- The dimmest resolvable signal is 4 × (1 ÷ Q + background spread); background dominates in the published example, and it is what hides vesicles.
How a flash of light becomes a number.
+ The questionA particle crosses the laser in a few microseconds. How does that pulse become a height, an area and a width, and who decides which pulses count as events at all?
The shape of a pulse
As a particle moves through the laser spot, the light it sends to a detector rises, peaks and falls. The detector turns that light into a current that follows the same course, and the electronics see a pulse: a bump in the signal lasting from under a microsecond in a fast sorter to a few microseconds in an analyzer. Its shape is set by the profile of the beam along the direction of flow and by the size of the particle, because a large cell takes longer to enter and leave the beam than a small one.
Three numbers are taken from each pulse. A manufacturer's technical bulletin defines them simply: the height is the peak intensity of the pulse, the width is the time the particle takes to pass through the interrogation point, and the area is "calculated by integration and represents the whole pulse." Area grows with the total light collected during the passage, height with the brightest moment, and width with the time spent in the beam. FCS 3.1 does not define the suffixes -A, -H and -W that most software attaches to these values; they are naming habits, and MIFlowCyt requires a report to say which pulse measurement it used.
Telling one particle from two
The three measurements together reveal clumps. Two cells stuck together, or passing one directly behind the other, produce a pulse that is longer than one cell's and carries about twice its area, while its height rises much less, so area and height no longer keep the ratio single cells show. Plotting area against height separates single cells, which fall along a diagonal line, from pairs, which fall above it; plotting width against height does the same, with pairs at greater width. Gating on that plot, doublet discrimination, is a routine first step of analysis.
It has limits. The sorting guide of Chapter 6 notes that pulse processing reduces doublets but "can never completely eliminate" them, and the method works only when a particle is comparable to or larger than the height of the beam. Two small particles side by side across the stream, or several vesicles inside the beam at once, produce a pulse that looks like one particle's. A doublet of two cells that is not removed can look like a single cell positive for both their markers, which is exactly the kind of population an immunophenotyping assay is looking for.
The threshold decides what exists
Most of the time, nothing is passing the beam, and the detectors register only background. The electronics need a rule for deciding when a particle has arrived, and the rule is the trigger and threshold of Chapter 1: one chosen detector, and a level its signal must exceed. When it does, the electronics open a window, measure the pulse in every detector, and record an event. Everything below the threshold is discarded and leaves no record.
The choice of trigger shapes the data as much as any other setting. A threshold set too high discards real but dim particles; set too low, it fills the file with noise events and increases the overlap of pulses, which leads to more electronic aborts. FCS 3.1 can record one trigger detector and its threshold, in channel units, through its $TR keyword. More elaborate triggers, such as an event declared when either of two detectors crosses its level, have no standard representation.
For vesicles, the choice of trigger is decisive. Their scatter is so weak that a scatter trigger set above the noise misses most of them. Triggering on fluorescence instead, so that an event is declared when a labeled particle glows, changes the count dramatically: a 2016 study of plasma vesicles found that fluorescence triggering raised the number detected 15-fold for vesicles from red cells, 40-fold for those labeled with annexin V and 75-fold for those from platelets. MIFlowCyt asks for the threshold to be stated, and its extension for vesicle cytometry asks for the trigger channel and threshold, preferably in calibrated units, such as a scatter cross-section in square nanometers or a fluorescence level in calibrated molecules, rather than in channels.
From current to digits
Modern cytometers are described as digital, but the sorting guide is blunt about it: "A fully digital flow cytometer does not exist. All cytometers, including sorters, are hybrid analog–digital systems." The detector's current is converted to a voltage and amplified by analog electronics, and then an analog-to-digital converter samples the voltage tens of millions of times a second, typically at 10 to 100 megahertz according to the same guide. Software computes height, area and width from the samples.
A converter with n bits distinguishes 2ⁿ levels: 16 bits give 65,536. The span from the smallest step to the largest value, expressed as powers of ten, is the number of decades: 16 bits span about 4.8 decades, 18 bits about 5.4 and 22 bits about 6.6. One manufacturer quotes "22-bit 6.5 log decades," consistent with that arithmetic. But a linear converter has very few steps at the bottom of its range: in an 18-bit range, the first decade, from 1 to 10, contains only nine integer steps. A pulse lasting 3 microseconds sampled at 10 megahertz yields about 30 samples, at 60 megahertz about 180. The figures are derived for this guide.
The bit depth printed in a specification does not always describe the converter. Some instruments combine two 16-bit converters per channel to produce data with an effective 23-bit range, and an area stored as a 32-bit number is a sum of many samples, not a 32-bit conversion. What matters to the measurement is the noise and resolution at the bottom of the scale, where dim particles live.
The ghost of the log amplifier
Older instruments did not digitize a linear signal. They passed it through logarithmic amplifiers that compressed four decades of signal into 1,024 channels, which made bright and dim populations visible on the same axis. Analog log amplifiers were never perfect: a manufacturer's patent notes that their transfer function "can contain 'ripples' that may distort the input signal," and the 2018 study of 23 instruments found its measurements less precise on instruments with log amplifiers than on linear ones.
Linear digitization removed the hardware but not the habit of viewing data on a logarithmic axis, and the habit collided with compensation. Subtracting one dye's light from another detector (Chapter 10) produces values near zero and below zero for cells that carry none of a dye, and a logarithmic axis cannot show zero or negative numbers. A 2006 paper described how log displays "often show a peak above the actual mean or median of the population with a pileup of events on the baseline," a distortion known as the log artifact. Its remedy, the logicle or biexponential display, is linear near zero and logarithmic far from it, and Chapter 12 shows that it is a choice of display that can change how a plot looks without changing the data.
Height and area of the same pulse can rank particles differently, the threshold decides which particles exist at all, and the transform decides how populations look. A number from a cytometer without these three facts cannot be reproduced on another instrument or in another laboratory.
In an instrument specification, state the sampling rate, the converter resolution, the electronic noise and the baseline stability in units of the detector signal, and test linearity down to the dimmest populations the instrument is meant to measure. In an assay, fix the trigger channel, threshold and pulse measurements, and record them; for small particles, express the threshold in calibrated units so that it means the same on another instrument.
- Each pulse yields a height, an area and a width; the suffixes are conventions, and a report must say which measurement it used.
- Area against height separates single particles from doublets when particles are comparable to the beam height, but never perfectly.
- The trigger and threshold decide what becomes an event; for plasma vesicles, fluorescence triggering found 15 to 75 times more particles than scatter triggering.
- Cytometers are hybrid analog–digital systems; linear converters, which have few steps at the bottom of the scale, replaced log amplifiers, and logicle displays replaced logarithmic axes to show compensated values near and below zero.
+ Part IV · From signals to populations
How rows of numbers become kinds of cell.
Part III ended with every particle reduced to a row of numbers, one for each detector and pulse measurement. These three chapters turn the rows into populations. Chapter 10 separates dyes whose light lands in the same detectors, Chapter 11 sets the controls that decide where a positive signal begins, and Chapter 12 draws the boundaries that group events into the populations a result reports. Each step is a decision taken by people and software after the instrument has finished, and each can change the answer.
When two dyes land in one detector.
+ The questionCompensation subtracts one dye's light from another dye's detector. What does it fix, what can it never fix, and why is spectral unmixing the same problem?
Every detector sees more than one dye
The emission band of a dye is broad, tens of nanometers wide, with a long tail toward the red (Chapter 3). A bandpass filter passes a slice of the spectrum to each detector, and that slice usually contains the tail of a neighboring dye as well as the peak of its own. Fluorescein glows most strongly near 520 nanometers, but part of its light reaches a detector set for phycoerythrin near 575. The fraction of a dye's signal that appears in another dye's detector is called spillover. It is a property of the dye, the filters and the detector settings, not of the cell, so it is the same for every cell measured under the same conditions, except where a tandem dye degrades on some cells (Chapter 5).
Spillover is also linear. A cell carrying twice as much fluorescein sends twice as much light into the phycoerythrin detector, so each detector's reading is a sum of contributions, one from each dye, each proportional to the amount of that dye on the cell. That linearity is what makes the problem solvable. If the fraction each dye contributes to each detector is known, the readings can be converted back into the amount of each dye.
Compensation undoes the mixing
Compensation is that conversion. A 2001 review by Mario Roederer called it "a simple application of linear algebra, with the goal to correct for spillovers of all dyes into all detectors." The fractions are measured with single-stain controls, samples each stained with only one of the panel's dyes, so that any light in the other detectors must be spillover. With two dyes there are two fractions to measure; with n dyes there are n² − n, one for each dye in each detector other than its own. An eight-color panel has 56 spillover values and a twenty-color panel 380.
Suppose 15 % of fluorescein's signal spills into the phycoerythrin detector and 2 % of phycoerythrin's into the fluorescein detector. A cell carrying fluorescein worth 10,000 and phycoerythrin worth 4,000 reads 10,080 in the fluorescein detector and 5,500 in the phycoerythrin detector. Subtracting 15 % of the first reading from the second leaves 3,988; dividing by 1 − 0.15 × 0.02, which is 0.997, restores 4,000. The same steps restore 10,000 for fluorescein. The correction is exact for the averages, and the small division matters only when two dyes spill heavily into each other. The values are illustrative, for this guide.
The calculation is done by software, not by the instrument. The FCS 3.1 file stores the detector readings as measured and can carry the spillover values alongside them, and the standard states that "the stored data is always uncompensated." A laboratory can therefore recompute compensation with a better matrix long after acquisition, which older instruments that compensated in their electronics did not allow. Chapter 18 shows what happens when software drops the matrix.
What compensation can never remove
Compensation restores the average amount of each dye. It cannot restore the precision that spillover took away. The light that spills into a detector arrives as photons, and the photoelectrons they produce fluctuate by the square root of their number (Chapter 8). Subtracting the average spillover leaves the fluctuation behind, so a population that carries none of the second dye ends up centered on zero but spread more widely than cells that carried neither. The 2001 review shows that the spread rises with the square root of the spilled signal and that it cannot be removed by subtracting more, "because compensation is a linear process."
A cell bright in fluorescein spills light worth 400 photoelectrons, on average, into the phycoerythrin detector. Counting statistics give that light a spread of about 20 photoelectrons, the square root of 400. After compensation the cell's phycoerythrin value is centered on zero, with that spread around it. A cell four times brighter spills 1,600 photoelectrons and leaves a spread of 40: four times the light, twice the spread. A dim phycoerythrin label that adds 30 photoelectrons to a positive cell is lost within the spread of the bright cells, though it would stand clear of unstained ones. The figures are illustrative, derived for this guide.
On a plot, the effect looks like a fan: the compensated negatives widen as the first dye grows brighter, and the widened edge can look like a dim positive population. A 2013 study named it spillover spreading and defined a value for each dye and each detector it spills into that measures it independently of brightness, collected into a spillover spreading matrix for each instrument. A 2021 study showed that the value changes with detector voltage, so it describes an instrument at its settings, not the dyes alone, and proposed an index that does not.
The matrix completes the panel-design rule of Chapter 5. Pairing the brightest dyes with the scarcest proteins is half of it; the other half is to avoid reading a dim marker in a detector that receives heavy spread from a dye that is bright on the same cells. For vesicles, whose labels sit within tens of molecules of the background, every source of spread raises the floor a label must clear, which argues for few colors and well-separated dyes.
Unmixing is the same arithmetic with more detectors
A full-spectrum or spectral cytometer replaces one detector per dye with many detectors per laser, so that each dye is recorded as a signature: the pattern of its light across all of them. A cell's readings are then the sum of the signatures of the dyes it carries, each weighted by the amount of dye. With more detectors than dyes and noisy readings, there is in general no exact solution. Unmixing finds the amounts whose combined signatures fit the readings best, usually by the method of least squares. A 2022 comparison by the founder of an analysis-software company describes conventional compensation as a special case of the same mixture model: the square case, where the number of detectors equals the number of dyes.
Unmixing inherits both limits of compensation. It restores the averages and leaves the counting noise, and it makes common a third limit that compensation meets only when two dyes spill heavily into each other, the small division of the worked example: when two signatures are similar, small fluctuations in the readings can be assigned to either dye, which amplifies the spread. A 2026 study, led by an engineer at one manufacturer with academic co-authors, measured this unmixing-dependent spreading on four spectral instruments from three manufacturers and found that it is not new noise but a magnification of uncertainty already present in the raw data. Plain least squares also assumes that every detector is equally noisy, which photon counts are not, and a 2013 analysis by the same software developer and academic colleagues argued that photon-count data violate the assumptions of plain least squares, linking the violation to spreading and negative values, and proposed a model built on photon statistics. Autofluorescence can be treated as one more signature and subtracted the same way. The similarity indices that software reports for pairs of dyes have no peer-reviewed formula, and Chapter 14 weighs when a spectral instrument is the better choice.
The matrix is only as good as its controls
Every spillover value comes from a single-stain control, so the controls decide whether compensation is right. A 2006 guideline by staff of one manufacturer asked that each control be stained with one of the panel's own reagents and be at least as bright as the test samples. A later guideline for spectral cytometry, with a manufacturer's staff among its authors, added that controls use "the same antibody lot" and be treated like the samples, because fixation and pH can change a dye's spectrum. Tandem dyes need controls from the same vial and lot (Chapter 5), because their spillover changes as they degrade.
The EuroFlow consortium, whose leukemia panels return in Chapter 21, turned these rules into a schedule. Its protocol sets compensation with single-antibody-stained samples, by default once a month, with values specific to each conjugate for its two tandem dyes. In a test of eight laboratories, the compensation matrices of the seven instruments compared were similar, which is part of what lets a pattern read in one laboratory mean the same in another.
Too little compensation leaves a diagonal population that looks positive for two dyes; too much pushes negatives below zero and can hide dim positives. Spillover spread can look like a dim population. Editing the matrix by eye until a plot looks clean changes the averages to hide the spread, which compensation cannot remove, and makes the result wrong in a way that no later reader can detect.
In an assay specification, define the single-stain controls (reagent, lot, cell or bead type, brightness relative to samples), how often the matrix is recalculated and its acceptance limits. Keep the uncompensated data and the matrix used for each result. Design panels against a spillover spreading matrix measured on the instrument at its settings, and recalculate after a change of reagent lot, detector voltage, filter or laser.
- Dye emission is broad, so each detector receives light from several dyes; spillover is linear and depends on dyes, filters and settings, not on the cell.
- Compensation inverts the mixing with values measured on single-stain controls, n² − n of them for n dyes, and FCS files store data uncompensated.
- Compensation restores averages but not precision: spread grows with the square root of the spilled light and limits dim markers in receiving detectors.
- Spectral unmixing is the same model with more detectors than dyes; it adds spread when signatures are similar and depends just as much on its controls.
Where a positive signal begins.
+ The questionEvery cell gives some light in every detector, stained or not. Which comparison decides that a cell carries a marker, and why do some widely used controls answer the wrong question?
Positive is a comparison, not a value
No detector reads zero for an unlabeled cell. Cells glow faintly by themselves, from flavins and other molecules inside them, which is called autofluorescence. Light from the panel's other dyes spills into the detector and leaves its spread behind after compensation (Chapter 10). Antibodies stick weakly to cells that lack their target, through Fc receptors, dead cells and simple stickiness (Chapter 5). A 2009 review grouped these sources of background into three: autofluorescence, spectral overlap and undesirable antibody binding.
A cell is therefore never positive in an absolute sense. It is positive when its signal in one detector is higher than that of comparable cells which differ only in lacking the specific binding being tested. A control is the sample, or the population within a sample, that supplies that comparison. Each kind of control holds back a different source of background, so the choice of control decides what the boundary between positive and negative actually measures.
The boundary matters most where a marker is dim. A bright marker separates from its negatives on almost any reasonable control. A dim one lies within a few multiples of the background, and moving the boundary by the width of the negative population can change the fraction of positive cells severalfold. That is the region in which leukemic cells, with their abnormally dim markers, and vesicles, whose labels sit close to the background (Chapter 5), have to be read.
The unstained cell is the wrong reference
The simplest control is a sample of the same cells with no stain at all. It measures autofluorescence and nothing else. Mario Roederer's 2001 review explained why that is not enough for a multicolor panel: on compensated data, "any gate based on a completely unstained sample" would give wrong results, because at high intensities the negatives spread upward into the positive region. An unstained sample has no spillover, so it has no spillover spread, and a boundary set on it is too low for the stained tube.
The same review gave the alternative a name. A fluorescence-minus-one control, or FMO, contains every antibody of the panel except the one whose positive boundary is being set. It carries all the spillover into that detector, with all its spread, and none of the specific signal. The 2006 guideline of Chapter 10, written by one manufacturer's staff, defined it as a sample with all the conjugates "except one" and noted its limit: an FMO does not measure the background binding of the missing antibody itself, because that antibody is absent.
Suppose 99 % of unstained cells fall below 300 in the phycoerythrin detector, and a boundary is set there. In the FMO tube, where the other dyes are present but phycoerythrin is not, spillover spread pushes the upper edge of the negatives higher, and 8 % of cells now lie above 300. Those cells carry no phycoerythrin, so in the fully stained tube the same 8 % appear positive for a marker they lack. A boundary set on the FMO, where 99 % of its cells fall below 1,200, removes them. The values are illustrative, for this guide.
FMO controls are needed most for dim markers, markers expressed on a continuum, and detectors that receive heavy spread. They are needed least where positive and negative cells form clearly separated groups, which is why laboratories choose them channel by channel rather than for every marker of every panel.
Why isotype controls usually fail
An isotype control is an antibody of the same class and subclass as the test antibody, carrying the same dye, but raised against a target the cells do not have. The idea is that it binds nonspecifically as the test antibody does, so its signal shows how much of the test signal is background. The 2006 guideline stated the intent: such controls "are meant to account for nonspecific staining" of a given isotype conjugated to a given dye.
In practice the match is poor. Two antibodies of the same subclass differ in structure, in the number of dye molecules attached and in how they were purified, and each conjugate has its own background. A 1998 editorial on CD34 and lymphocyte counting, titled "time to let go!", reported that isotype controls do not stain the same number of events as the test antibody stains nonspecifically, and argued that their widespread use should be questioned. The 2006 guideline added that, on their own, isotype controls "do not account for fluorescence spillover from other channels."
The evidence against them is a guideline and an editorial with observations, not a controlled study, and some reagent suppliers still recommend isotype controls from the same supplier. Their remaining use is narrow. The 2006 guideline describes one case, an isotype in the channel of interest together with all the panel's other antibodies, which takes care of spillover but still does not match the test antibody's own background. It does not set a positive boundary by itself.
Biology often gives the best negative
The most faithful control is often inside the tube. Cells of a type known not to carry a marker, measured in the same sample, have experienced the same staining, the same spillover, the same antibody and the same instrument settings. T cells do not carry CD19, so in a tube that stains B cells for CD19 the T cells give the negative directly. The 2006 guideline made the same point for stimulation assays, where the unstimulated sample "usually provides the best means" of telling positive from negative.
Clinical immunophenotyping reads markers this way. The ICSH and ICCS recommendations of 2013 describe abnormal intensity, bright or dim, relative to the normal counterpart of the abnormal cells. The 2018 international guidelines for chronic lymphocytic leukemia describe surface immunoglobulin, CD20 and CD79b as characteristically dim compared with normal B cells, a judgment that is easiest when normal B cells are present in the same sample. Chapter 21 shows how a report states such patterns. Dead cells, which bind many antibodies nonspecifically, are excluded beforehand with a viability dye (Chapter 12).
For vesicles, the first question is whether an event is a vesicle at all, and the controls change accordingly. The reporting framework for vesicle cytometry lists controls for this: buffer alone and buffer with the labeling reagents, because antibodies and dyes form small aggregates that cross the threshold like particles, and a detergent step that dissolves membranes, so that events which survive it cannot be vesicles. Chapter 20 takes these controls further.
Titration sets how far positive sits from negative
How much antibody to add is itself a control on the boundary. Too little leaves binding sites empty, so positive cells are dim and their brightness depends on how many cells are in the tube. Too much raises nonspecific binding, so the negatives brighten and spread. Between the two lies an amount at which positives are well separated from negatives, and titration finds it by staining the same cells with a series of antibody amounts and measuring separation at each.
The measure of separation is the stain index of Chapter 5, which rewards a narrow negative population as much as a bright positive one. The best concentration is therefore not always the one that makes positive cells brightest. A titration holds only for the reagent lot, cell type, staining volume and instrument it was done on, and a spectral guideline with a manufacturer's staff among its authors advised that new lots of tandem dyes be titrated again.
A result such as "12 % CD38-positive" is a property of the cells, the reagent and the boundary together. If the boundary was set on an unstained or isotype sample, the figure includes spillover spread and nonspecific binding of unknown size. Without the control named, a percentage of dim positive cells cannot be compared with another laboratory's.
In an assay specification, state for each marker which control sets its positive boundary (FMO, internal negative population, unstimulated sample or, for bright separated markers, the stained tube itself), with acceptance limits for the control. Record titration data for each reagent and repeat it after a change of lot, staining volume or cell type. For vesicles, include buffer-only, reagent-only and detergent controls in every run.
- Background comes from autofluorescence, spillover spread and nonspecific binding, so a positive signal is always a comparison with a control.
- An unstained sample sets the boundary too low for a stained tube; a fluorescence-minus-one control includes the spillover spread and sets it for that, though not for the missing antibody's own background.
- Isotype controls rarely match the test antibody's background and do not account for spillover; cells known to lack the marker, in the same tube, are often the best negative.
- Titration chooses the antibody amount that maximizes separation, not brightness, and must be repeated for new lots, especially of tandems.
How a population is drawn from a cloud of events.
+ The questionA file holds a million events and a report gives a handful of percentages. Who draws the boundaries between them, how much does the display decide, and how many events make a result?
A gate is a decision about a cloud
After compensation, each event is a point in a space with one axis for every detector and pulse measurement. Cells of one kind form a cloud in that space, because they carry similar amounts of each marker and scatter light similarly, and the job of analysis is to say which events belong to which cloud. The traditional tool is the gate: a boundary drawn on a plot of two parameters, which keeps the events inside it and passes them to the next plot.
Gates are applied in sequence, a gating hierarchy. A typical analysis first plots time against a scatter or fluorescence signal and removes stretches where the flow was disturbed; a 2021 set of guidelines describes removing "possible clogs" this way. It then removes doublets with area against height (Chapter 9), removes dead cells with a viability dye, selects a lineage such as T cells, and finally divides that lineage into subsets. Every percentage in a report is a fraction of some parent gate, so "8 % of CD4 T cells" and "8 % of lymphocytes" describe very different numbers of cells.
The early gates are not formalities. The doublet gate works only for particles comparable to or larger than the height of the beam; a teaching guide on sorting names bacteria, yeast and platelets as exceptions. For vesicles, which are all far smaller than the beam, it is unavailable, and the problem it solves for cells reappears as the swarm detection of Chapter 19.
The display can move the boundary
A gate is drawn on a picture, and the picture depends on how the axis is scaled. Compensated data contain values near zero and below it (Chapter 10), which a logarithmic axis cannot show. A 2006 paper described how log displays "often show a peak above the actual mean or median of the population with a pileup of events on the baseline," and how they make compact populations look split in two. Its remedy, the logicle transform, is linear around zero and logarithmic at high values, with a parameter that sets the width of the linear region. The arcsinh transform used in mass cytometry and many analysis packages does the same with a parameter called the cofactor, conventionally 150 for fluorescence and 5 for mass cytometry.
In one exercise done for the material behind this guide, on a public data set of stained control blood cells distributed with open-source analysis software, a single file of 61,832 events was displayed with two logicle transforms that differed only in the width of the linear region. A gate was placed at the same position on the screen in each. Above it lay 2.413 % of events in one display and 0.996 % in the other: a 2.4-fold difference in a reported percentage, with no change to a single measurement. The figures come from one file; the size of the effect depends on where a population sits relative to zero.
The transform does not change the data, but it changes where a person's eye places a boundary, and a boundary placed in display coordinates moves with the display. A 2010 study found that transforms with parameters fitted to the data reduced the misclassification of events compared with default settings. The practical consequence is that the transform and its parameters are part of the analysis and must be recorded with it, as MIFlowCyt asks.
People gate differently
Manual gating is skilled judgment, and judgments differ. A 2005 study led by staff of one manufacturer, which also supplied the reagents, sent the same samples to many laboratories and found inter-laboratory coefficients of variation of 17 to 44 % for a measurement of responding T cells, concluding that analysis, "particularly gating," is a significant source of variability. Analyzing all the files centrally reduced the inter-laboratory variation by 5 to 20 %. The scatter depended on how rare the population was: 18 to 24 % for samples with more than 0.5 % of cells responding, 57 to 82 % for samples with fewer than 0.1 %.
Standardized panels and central analysis narrow the spread. A 2016 study across nine laboratories, with one manufacturer's staff among its authors, using five standardized eight-color panels, found variation below 10 % across sites for large, easily identified subsets such as CD4 T cells, and higher variation for dim or deeply gated ones. Software itself is a variable as well: in the exercise described above for this guide, two programs applied to the same gated workspace gave different counts for 44 of 48 population and sample combinations, by up to 3.8 %, for a reason not yet traced.
Algorithms, clusters and maps
Automated methods replace hand-drawn boundaries with rules applied identically to every file. In a 2013 community benchmark, 14 groups submitted 36 sets of results, and many algorithms agreed with expert gating at a level the organizers judged sufficient for reliable use; combining several algorithms scored better than any single one. In its sample-classification part, 12 submissions classified all 359 samples of an acute myeloid leukemia data set correctly, half of which had been supplied with labels for training. The 2016 nine-laboratory study found that automated gating matched central manual analysis "for all tested panels." Benchmarks of this kind score algorithms against expert manual gates, so they measure agreement with people rather than truth.
Clustering goes further and groups events without predefined gates. A 2016 comparison of 18 clustering methods found FlowSOM best or near best on every data set and, given its speed, recommended it as a first choice. Clusters are often shown on two-dimensional maps made by t-SNE or UMAP, which place similar events near one another. Such maps distort: a 2019 analysis found that the relative position of clusters on a t-SNE plot is "almost arbitrary" and depends mostly on random initialization, and a 2023 study showed that reducing many dimensions to two inevitably distorts distances. Both studies come from single-cell genomics, and the point carries over by analogy: a map shows which events are similar, not how far apart populations are.
What n is
A tube may contain a million events, but they come from one sample of one donor, processed once. The events measure that sample precisely; they say nothing about how a second donor, or a second preparation of the same donor, would differ. Treating events as independent replicates is pseudoreplication, defined in a 1984 paper as using statistics to test for effects when the replicates are not independent. That paper found it in about a quarter of 176 ecological studies. In cytometry the replicate is the donor, the patient or the independent preparation, and the events within a tube decide only how precisely each sample's frequency is known.
That within-tube precision follows the counting statistics of Chapter 8, applied to events rather than photoelectrons. What matters is the number of events in the population of interest, not the total.
A frequency estimated from r events of interest has a counting spread of about 1 ÷ √r. One hundred events give about 10 %, four hundred about 5 %. For a population at 1 in 10,000 cells, collecting 100 of them takes 1,000,000 events; at 1 in 100,000, it takes 10,000,000. This is why tests for residual disease after treatment acquire millions of cells per sample (Chapter 22). The arithmetic is derived for this guide and assumes events arrive independently.
The 2005 study's figures make the same point from the other side: the rarest populations were the least reproducible between laboratories. Raw intensities are no easier. A 2024 study at the US National Institute of Standards and Technology found that raw fluorescence intensities of the same cells varied widely between four instruments, while values converted to antibodies bound per cell agreed closely, which is the case for calibration made in Chapter 17.
A gate set by eye on one display, at one transform setting, on one instrument's scale moves when any of the three changes. Copying a gate template from one file to another without checking that the populations sit in the same place can cut through a population or include part of its neighbor, and the error appears in the report as a change in biology.
For an assay, document the gating hierarchy with example plots, the control that sets each boundary (Chapter 11), the transform and its parameters, the minimum number of events in the reported population and the software and version. Validate the software as part of the assay, and treat a change of template, transform or version as a change to be verified. In a study design, count donors or independent preparations as n, never events.
- Gates select events in sequence, from time and doublets through viability and lineage to subsets, and every percentage is a fraction of a parent gate.
- The transform changes how populations look and where a boundary drawn by eye falls: one file gave 2.413 % or 0.996 % with no change in the data.
- Gating is a major source of variation between laboratories; standardized panels, central analysis and automated gating reduce it, measured against expert gates.
- Events are not replicates: n is the number of donors or preparations, and a rare population's precision depends on how many of its own events were counted.
+ Part V · The cytometer as a system
Engineering the whole.
Parts III and IV took the measurement apart: the instrument piece by piece, then the arithmetic and judgment that turn signals into populations. These six chapters put it back together the way a systems engineer would. They show where the parts meet and fail, how two architectures trade against each other, how a stated need becomes a specification that can be verified, how an instrument is kept the same from day to day and made to agree with others, how its scales are tied to physical units, and what the data file keeps of all this. Together they give the engineering half of the answer to when a count can be trusted.
Where the parts of a cytometer meet.
+ The questionEach part of a cytometer has one job and can be built to do it well. Why do so many failures happen between the parts rather than inside them, and what does each one look like in the data?
No part knows what a cell is
A flow cytometer can be described as a set of subsystems, each with one job. Sample handling delivers a suspension of the right concentration, free of clumps, at a steady rate. The fluidics of Chapter 6 bring each particle to the same point of the beam. The excitation optics of Chapter 7 deliver light of known color, power and shape to that point at the right place along the stream. The collection optics and detectors of Chapter 8 turn the light that comes back into currents, and the electronics of Chapter 9 turn currents into numbers. Software applies compensation, transforms and gates and writes the file.
None of these parts knows anything about cells. The fluidics move water; the detectors count photons. Whether the instrument reports the right fraction of leukemic blasts depends on how the parts work together, and the most consequential engineering decisions concern the boundaries between them.
The community has published a decomposition of the instrument, though not for design. The reporting standard MIFlowCyt lists the details an experiment must report as a set of subsystems: the flow cell and fluidics, the light sources, the excitation optics, the filters, the detectors and the optical paths, with a final item for "other relevant instrument details" such as agitation and temperature control. Its preamble states that the system and its configurations "have significant impact on experimental results." This guide found no peer-reviewed treatment of a cytometer in systems-engineering terms, with an architecture, interface definitions and a failure analysis; the closest published decomposition is a list of what to report.
The interfaces that are clocks
Several of the most important interfaces in a cytometer are timings. The duration of a pulse is set by the speed of the stream, the height of the beam and the size of the particle (Chapter 9). The laser delay is set by the distance between beams and the speed of the stream (Chapter 7). The processing window decides how close two particles can pass before one is aborted (Chapter 6). The converter's sampling rate decides how many samples describe each pulse, and in a sorter the drop delay decides which droplet carries a chosen cell (Chapter 24).
Each of these clocks belongs to one subsystem and is set by another. The stream speed belongs to the fluidics, but the laser delay, which depends on it, belongs to the electronics. A change in one side that is not matched on the other produces an error that no single subsystem detects.
Every delay between beams is a distance divided by a speed. If the stream slows by 5 %, because the sheath pressure dropped or the sheath filter began to clog, every delay grows by about 5 % (1 ÷ 0.95). A delay of 40 microseconds becomes about 42, an error of about 2 microseconds, which is the size of the error that turned 97.6 % double-positive cells into 89.4 % in Chapter 7. The figures are illustrative, derived for this guide; real delays and speeds are properties of each instrument.
These timings are set when data are acquired, and the standard file has nowhere to keep most of them. FCS 3.1, the version most instruments write, has keywords for each detector's voltage, gain and filter, for the excitation wavelength and power, for the trigger and for the spillover matrix. It has none for sheath pressure, laser delay, window extension, drop delay, baseline restoration or detector temperature (Chapter 18).
Failures live at the boundaries
Many failures of a cytometer in use can be traced to one interface, and each, like the failures inside a single part, leaves its own signature in the data. A partial clog, at the boundary between sample preparation and fluidics, appears as steps in event rate and signal over time. A surge of flow or a bubble, between fluidics and data, appears as a shift in the median signal for a few seconds. A core widened by a high sample rate, between fluidics and optics, appears as broader populations (Chapter 6). A wrong or drifting laser delay, between fluidics and electronics, appears as lower values on the downstream lasers only.
On the optical side, a laser that has drifted in alignment or power appears as a wider and dimmer bead peak in daily quality control. The EuroFlow protocol accepts a coefficient of variation of the brightest bead peak below 4 % on its blue and violet channels and below 6 % on its red channels and its PE–cyanine 7 channel (Chapter 16). A filter whose coating has aged appears as a slow loss of detection efficiency in one color band, and a defective detector as low Q and high B on one channel (Chapter 8). The 2018 study of 23 instruments found channels in need of service this way. In the electronics, a baseline that is restored wrongly shifts every dim value, as one manufacturer's patent describes, and pulses that overlap appear as counts that fall short of the true number as the event rate rises.
The value of this list is that each signature points to a test. A failure that shows up in event rate over time is checked by plotting time; one that shows up in bead peaks is checked by daily quality control; one that shows up only at high rates is checked by running a dilution series.
Time is a measurement too
Every cytometry file records when each event occurred, and the time parameter is the simplest tool for finding interface failures. A plot of time against a scatter or fluorescence signal shows clogs, bubbles and drifts as steps, gaps and slopes in what should be a steady band. The 2021 guidelines for cytometry in immunology remove possible clogs this way, and the analysis package flowAI, released in 2016, automates the check by looking for surges in flow rate, shifts in median signal and events piled at the limits of the scale.
The same plot carries the vesicle thread. When particles are far smaller than the beam, a count that rises or falls during a run can mean that the sample is settling or sticking to the tubing, not that the particles changed. Chapter 20 adds the dilution series, which tests whether counts fall in proportion to dilution, as the check that each event is one particle.
Sample preparation is part of the instrument
Most failures at the first interface begin before the sample reaches the instrument. Clumped cells block the flow cell and form doublets, and dead cells release DNA that makes them stick. A teaching guide on cell sorting recommends filtering through a 30-micrometer mesh "at each step" and adding DNase and EDTA to reduce clumping. The age of a blood sample and its anticoagulant change some markers more than others, which the clinical recommendations of Chapter 22 turn into stability limits.
The EuroFlow consortium standardized preparation together with the instrument settings for exactly this reason. Its protocol fixes the lysing reagent, the volume of blood per tube and the stain–lyse–wash sequence, because instrument settings agreed to the last volt would not make results agree if each laboratory prepared its samples differently.
A drifting laser delay lowers the fraction of double-positive cells; a partial clog dims or shifts signals for part of the run; overlapping pulses at high rate inflate apparent doublets and brighten populations. None of these produces an error message. The data look like a plausible sample with a different composition, which is why interface checks must be part of every run rather than a response to suspicious results.
Keep an interface definition for each boundary: what crosses it, in what units, within what limits, and who owns each side. For a cytometer the list includes sample concentration and filtering, sheath pressure and stream speed, beam spacing and laser delays, processing window, trigger and threshold, detector voltages, compensation matrix, transform and gating template, and software version. For each, name the check that detects a failure, such as a time plot, daily beads, or a dilution series, and record the settings that the file cannot hold.
- A cytometer is sample handling, fluidics, excitation, collection, detection, electronics and software; none knows what a cell is, and results emerge from how they work together.
- Several key interfaces are clocks, such as pulse duration, laser delay, processing window and drop delay; a 5 % change in stream speed shifts a 40-microsecond delay by about 2 microseconds.
- Each interface failure leaves a signature in the data, from steps over time to wider bead peaks, and most look like biology rather than like errors.
- Sample preparation is part of the system; EuroFlow standardized preparation together with instrument settings so that results would agree between laboratories.
One detector per dye, or the whole spectrum.
+ The questionA conventional cytometer gives each dye its own detector behind its own filter, while a spectral one spreads every particle's light across dozens of detectors. What does each architecture buy, what does it cost, and when is the older design the right one?
Two ways to divide the light
A conventional cytometer divides the collected light with a cascade of dichroic mirrors and gives each dye a detector behind a bandpass filter chosen to pass its peak (Chapter 8). The number of detectors roughly equals the number of dyes, and the overlap between them is corrected by compensation with a square spillover matrix (Chapter 10). The FCS file stores that matrix as a single keyword, with each value defined as the ratio of a dye's signal in another detector to its signal in its own.
A full-spectrum cytometer spreads the light from each laser over many detectors instead, with a prism, a grating or a long cascade of narrow filters, and records the whole shape of each particle's emission. Each dye is identified by its signature across all the detectors, and the amounts are found by unmixing. Manufacturers state 48 fluorescence channels on one instrument, 64 on another and up to 186 detectors on a third. In 1979 a group reported what it called the first complete emission spectra of cells in a flow cytometer, spread by a grating across the 500 channels of a vidicon detector. A 32-channel instrument was described in 2012, and the first commercial spectral cytometer was launched by its maker in 2013 (Chapter 2).
The two designs share everything upstream of the collection optics: the same fluidics, lasers and beams, and the same physics of detection. They differ in how the colors are separated, in hardware or in arithmetic, and that difference decides what each one is good at.
What the whole spectrum buys
The first gain is the number of dyes. Two dyes whose emission peaks are too close to be split by filters can still differ in the shape of their spectra, or in how strongly each laser excites them, and a spectral instrument can tell them apart by the whole pattern. A forty-color panel was published in 2020 by authors mostly from one manufacturer, and a 2026 study led by an engineer at one manufacturer noted that fifty fluorescent analytes measured at once had been reported. Changing a panel also needs no change of filters, because every detector records every particle.
The second gain is autofluorescence. A conventional instrument sees it only as background in each detector. A spectral instrument can record its signature and treat it as one more component to unmix, which removes it from the other dyes and can even use it as a parameter, since different cell types glow differently. A 2026 study of liver cells showed both the gain and its conditions: several autofluorescence signatures pooled across conditions worked, while a single default signature silently underestimated the number of macrophages.
What it costs
The cost is paid in spread and in controls. Chapter 10 showed that unmixing adds spread when two signatures are similar, because small fluctuations in the readings can be assigned to either dye. The similarity of two signatures can be measured as a number between 0 for signatures with no overlap and 1 for identical ones, and the closer it is to 1, the more noise unmixing amplifies.
Take the simplest case: two dyes measured in detectors that all carry equal, independent noise, with s the cosine similarity of their signatures (0 for no overlap, 1 for the same shape). Compared with two signatures that do not overlap, least-squares unmixing multiplies the spread of each dye's estimated amount by 1 ÷ √(1 − s²). At a similarity of 0.5 the factor is about 1.15; at 0.9, about 2.3; at 0.98, about 5. A pair of dyes that looks acceptably distinct on a spectrum plot can still multiply the spread fivefold. The arithmetic is derived for this guide; real instruments have unequal noise between detectors and more than two dyes.
Pairs are not the whole story. The 2026 study that measured unmixing-dependent spreading on four spectral instruments found that pairwise similarity is only loosely predictive, because spread can come from combinations of three or more dyes, and it proposed a matrix that shows where in a panel the spread concentrates. The similarity indices that instrument software reports have no peer-reviewed formula; the only explicit formulas this guide found are in a patent assigned to one manufacturer.
The second cost is the reference library. Every signature used for unmixing comes from a single-stain control, and a spectral panel of thirty dyes needs thirty controls, plus unstained cells for each tissue, each meeting the rules of Chapter 10. One clinical review, not peer-reviewed, described unmixing as dependent on "rigorously maintained" single-color and autofluorescence controls. The third cost is dependence on the algorithm: ordinary least squares, weighted least squares and models built on photon-counting statistics can give different results from the same raw data, and the algorithm is part of the instrument's software, not a choice the user always sees.
When the conventional design is right
The trade study does not always favor more detectors. A panel small enough that filters separate its dyes cleanly gains nothing from unmixing, and it avoids the spread that collinear signatures add. A laboratory with panels already validated on conventional instruments, a stable high-throughput workflow, or no capacity to manage a reference library may see no net benefit, as the same clinical review put it.
Standardization across laboratories is the strongest case. The EuroFlow consortium built its leukemia panels on instruments with identical filter configurations and three lasers at about 405, 488 and 633 nanometers, and its quality program covered 14 instruments in 11 laboratories. A spectral instrument could reproduce the panels, but it could not join that system without re-deriving its targets and checks. Regulatory status also differs by instrument and country: a 2023 review stated that spectral cytometry had been approved for diagnostic use in China and Europe, while the US clearances found for this guide are for conventional analyzers, and this guide did not confirm a US clearance of a spectral analyzer as of October 2026 (Chapter 23).
For vesicles the choice turns on a different question. A particle whose label is equivalent to a few tens of dye molecules can rarely support many colors at once, so what matters is how dim a signal each channel can see, measured by Q and B (Chapter 8), rather than how many channels the instrument has. Either architecture can be good or poor at that.
The same raw data unmixed with a different set of reference signatures gives different amounts and different spreads, and the 2026 study defined unmixing-dependent spreading by exactly that comparison. A changed reference control, a missing autofluorescence signature or an updated algorithm can change a result without any change in the sample or the instrument.
Compare candidate instruments against the panels they must run: number of dyes, the spread each panel will carry, autofluorescence of the target tissues, control burden, throughput, standardization with existing panels and regulatory status in the target markets. For a spectral instrument, specify the reference library, how often it is renewed, the unmixing algorithm and its version, and a spread check for every panel. For a conventional one, fix the filter configuration and treat any change of it as a change to every assay that uses it.
- Conventional instruments separate dyes with filters and one detector per dye; spectral instruments record each particle's whole emission and separate dyes by unmixing.
- The spectrum buys more dyes, panel changes without new filters, and autofluorescence handled as a component of its own.
- It costs spread when signatures are similar, at about fivefold for a similarity of 0.98, a larger library of reference controls, and dependence on the unmixing algorithm.
- Small validated panels, cross-laboratory standardization such as EuroFlow's and regulatory status can make the conventional design the right choice.
Turning performance words into numbers.
+ The questionSensitive, precise and linear are words on a datasheet. How does each become a number that can be measured, what does a test of it look like, and which standards decide whether the instrument is safe to sell?
From a need to a number
A systems engineer starts from what the instrument must do for its user, stated as a need: count blasts at one in ten thousand cells, measure CD4 T cells in blood, detect vesicles of 150 nanometers. Each need is then broken into requirements for the instrument, written so that a test can show whether they are met. A requirement that says "highly sensitive" cannot be tested; one that names a quantity, a value, a method and a condition can.
The performance of a cytometer comes down to a short list of such quantities. Each has a definition, a way of measuring it and, on a datasheet, a value claimed by the manufacturer under the manufacturer's own method. The rest of this chapter takes them in turn and then turns to the standards that govern safety rather than performance.
Sensitivity and precision
Sensitivity is how dim a signal the instrument can separate from nothing. Chapter 8 gave the measurable form: the detection efficiency Q and the background B of each channel, which together set the dimmest resolvable signal. Datasheets more often quote a single figure in molecules of equivalent soluble fluorochrome, and the interlaboratory study cited there found that such values, assigned to the same beads by different groups, varied by orders of magnitude in most cases. A sensitivity requirement is therefore best written per channel, as Q and B measured by a stated method.
Precision is how narrow a population of identical particles appears. It is measured as a coefficient of variation, often in a robust form that ignores outliers. One manufacturer's quality-control system defines the robust CV as 100 × 0.7413 × the distance between the 25th and 75th percentiles, divided by the median; the factor converts that distance into the equivalent of a standard deviation for a bell-shaped population. A peak whose quartiles lie at 9,500 and 10,500 around a median of 10,000 has a robust CV of about 7.4 %. For DNA measurements, manufacturers state precision as the CV of nuclei from chicken red blood cells stained with a DNA dye, typically below 3 %.
Linearity, range and carryover
Linearity is the proportionality of the output to the input: twice the light should give twice the number. One quality-control system reports the range over which the deviation stays within ±2 %; another manufacturer claims a fit with R² of at least 0.995 for two dyes. The two are not equivalent. A straight-line fit over five decades can reach a very high R² while the bottom decade, where dim populations lie, deviates by far more than 2 %, so a useful requirement states the permitted deviation and the range over which it holds. The dynamic range is the span from the dimmest to the brightest signal measured within that deviation, which is not the same as the converter's bit depth (Chapter 9).
Carryover is the fraction of one sample that appears in the next. It is measured by running a sample and then particle-free buffer and dividing the concentration found in the buffer by the concentration in the sample. Manufacturers claim values from below 0.01 % to below 1 % on different instruments, and in clinical use the figure matters most when a heavily diseased sample is followed by one being tested for residual disease.
Suppose an instrument carries over 0.1 % of each sample, and a marrow sample in which half the cells are leukemic blasts is followed by a sample tested for residual disease. If the two have similar cell concentrations, the second sample receives blasts amounting to 0.05 % of its cells, five times a residual-disease threshold of 0.01 %. A carryover of 0.01 % would still contribute 0.005 %, half the threshold. The figures are illustrative, derived for this guide; the cure is a low-carryover specification, rinses between samples and a run order that does not place high-burden samples before residual-disease tests.
Throughput is specified as a maximum event rate, which manufacturers state from 10,000 events per second with eight parameters on an older analyzer to 35,000 or 40,000 on spectral instruments. A rate is only half a specification: Chapter 6 showed that coincidence rises with rate, so a useful requirement states the abort or coincidence fraction at that rate as well.
Verification is a test with a method
Verification shows that the instrument meets its specification. It is different from the validation of Chapter 22, which shows that an assay run on the instrument serves its clinical or scientific purpose. A 2006 protocol described a three-part program to optimize, calibrate and monitor cytometers for panels of five or more colors, using beads with eight levels of brightness measured over a series of detector voltages; its 2012 successor stated that the complete process takes 3 to 4 hours, with a subset repeated routinely.
Every verification result carries its method. The same quantity measured with different beads, lots, voltages or software gives different numbers, and a claim without its method cannot be repeated by the buyer. The trap is sharpest for small particles, because scatter sensitivity is almost always specified with polystyrene beads. One manufacturer states that its instrument separates 80-nanometer polystyrene beads from noise, another states sensitivity to 0.2-micrometer beads, and a third gives below 200 nanometers. Chapter 4 showed why none of these is a vesicle size.
With the scatter arithmetic of Chapter 4, an 80-nanometer polystyrene bead sends about as much side scatter at 488 nanometers as a vesicle of about 135 nanometers with a refractive index of 1.40. A 100-nanometer bead matches a vesicle of about 175 nanometers, as Chapter 4 showed, and a 200-nanometer bead one of several hundred. The figures are derived for this guide for side scatter collected between 45 and 135 degrees in water; real instruments collect over different angles, and vesicles vary in refractive index from about 1.37 to 1.45.
Safety and electromagnetic compatibility
Performance is a matter between maker and buyer; safety is a matter of law. A cytometer sold as a laboratory or diagnostic instrument is designed and tested to international standards, and their editions change. As of October 2026, the general safety standard for laboratory equipment, IEC 61010-1, is in its third edition of 2010 with a first amendment of 2016 and a second close to publication. Its particular part for in vitro diagnostic equipment, IEC 61010-2-101, is in its third edition of 2018. Electromagnetic compatibility is covered by IEC 61326-1, third edition of 2020, and for diagnostic equipment by IEC 61326-2-6, whose fourth edition of 18 June 2025 replaced the 2020 edition with changes to test levels and documentation.
Lasers have their own standard, IEC 60825-1, in its third edition of 2014. Manufacturers declare enclosed cytometers Class 1 laser products, the least hazardous class, even though the lasers inside deliver tens to hundreds of milliwatts. The class describes what a user can reach with covers and interlocks in place, which is why servicing an instrument with its covers removed is a different hazard. In the United States, an FDA notice of 8 May 2019 stated that the agency does not intend to enforce its own comparable laser rules for manufacturers that comply with the IEC standard's third edition.
A sensitivity figure from one maker's beads, a linearity R² over five decades, a scatter limit in polystyrene and a maximum event rate with no coincidence fraction each describe a test that the buyer cannot repeat or compare. Accepting such figures as requirements moves the risk from the specification to the first assay that fails.
For each performance quantity, write the value, the test method, the reference material and lot, the settings and the acceptance criterion. The list covers Q and B per channel, robust CV of a named bead, linearity deviation over a stated range, carryover with the rinse procedure, event rate with its coincidence fraction, and scatter sensitivity in both bead size and the estimated vesicle size it implies. List the safety, laser and electromagnetic standards by edition, and recheck the editions before each design review.
- A need becomes a requirement only when it names a quantity, a value, a method and a condition; performance words on a datasheet do not.
- Sensitivity is best specified per channel as Q and B; precision as a robust CV; linearity as a permitted deviation over a range rather than an R².
- Carryover of 0.1 % from a sample half full of blasts would put blasts at five times a 0.01 % residual-disease threshold into the next sample.
- Safety rests on IEC 61010-1 and its diagnostic part, laser safety on IEC 60825-1 and electromagnetic compatibility on IEC 61326, with editions that change.
Keeping one instrument the same, and two alike.
+ The questionA cytometer's lasers age, its filters fade and its detectors drift. How is one instrument kept the same from day to day, and how far can laboratories go toward making different instruments give the same numbers?
Every instrument drifts
A cytometer that met its specification on the day it was installed will not meet it unchanged a year later. Laser output and alignment drift, filter coatings age (Chapter 8), detectors lose efficiency, and the fluidics wear. Each change shifts the numbers a sample produces, and Chapter 13 showed that most shifts look like biology rather than like faults. Stability therefore has to be measured, every day the instrument is used, with something that does not change.
That something is a set of quality-control beads: plastic particles dyed with fixed amounts of fluorescent dye, run before samples to check that each channel reads where it should, with the spread it should have. One manufacturer's system uses beads at three levels and explains the choice. Its bright beads have negligible counting contribution to their spread, its mid beads a large one, and its dim beads a significant contribution from background. Because the three levels expose the three sources of spread of Chapter 8 separately, a system of this kind can report relative values of Q and B for each channel, along with detector voltages and laser delays.
Each day's results are plotted against time, often on a Levey–Jennings chart with limits drawn around the expected value, so that a slow trend becomes visible before it crosses a limit. This guide found no peer-reviewed data set following many instruments over years; the published evidence on drift is mostly the daily logs of standardization programs such as the one described below.
Choosing the detector voltage
Before an instrument can be kept the same, its detector settings have to be chosen. The voltage on a photomultiplier sets its gain. Too low, and dim signals are lost in the electronic noise of the amplifier and converter. Too high, and bright signals run off the top of the scale. Between the two lies a range in which raising the gain no longer improves the spread of a dim population, because counting statistics, not electronics, now dominate.
The EuroFlow consortium found that range with beads of eight brightness levels. On a reference instrument, the second-dimmest bead population was measured at voltages from 300 to 999 in steps of 50, and its robust CV calculated at each; the chosen voltage was "at the beginning of the plateau phase of the curve." Manufacturers' systems use related rules, such as placing a dim bead at ten times the spread of the electronic noise, and one core-facility protocol aims for electronic noise to contribute only 10 to 20 % of the measured variance. Above the plateau, extra gain buys nothing at the bottom of the scale and loses room at the top.
Target values make instruments alike
Keeping one instrument stable is not enough when results from many laboratories must mean the same thing. EuroFlow's 2012 description of its instrument settings, the product of six years of joint experiments, addressed that problem for the leukemia panels of Chapter 21. Its laboratories used instruments with three lasers at about 405, 488 and 633 nanometers and identical filter configurations, with the exceptions listed, and a fixed set of eight dyes.
The method turned one instrument's settings into targets for all. After the voltages were chosen on the reference instrument, the brightest bead peak was measured in every channel, and those values became the target values that every other instrument had to reach, adjusting its own voltages. One master lot of beads was used throughout. Each day, an instrument passed if its bead values lay within 15 % of target and the spread of the brightest peak was below 4 % on the blue and violet channels and below 6 % on the red channels and the PE–cyanine 7 channel.
The approach worked on beads. Across the eight instruments of the 2012 evaluation, the variation of the brightest peak's value was "systematically lower than 5.5%." The instruments were not identical, but they had been made to read the same reference material the same way, which is a property that can be checked every day.
How far standardization reaches
Beads are the easy part. Between 2010 and 2013, EuroFlow ran four annual rounds of quality assessment on 123 blood samples from healthy donors, measured on 14 instruments in 11 laboratories in nine countries with uniform gating. The average median fluorescence in each round ranged from 86 to 125 % of the overall median, and for 7 of 11 markers the variation stayed below 30 %. The authors described the scheme as a complement to established proficiency tests, not a replacement.
A 2019 study transferred the method to 10 clinical laboratories in Switzerland, six with instruments of one manufacturer and four with another's. Over three rounds, the average variation across 11 markers was 30.3 %, 32.9 % and 27.8 %, comparable with EuroFlow's own laboratories, and median fluorescence did not differ significantly between the two makes. By the third round, 89 % of participants had at least nine in ten of their measurements within the acceptance criteria.
A bead is the same particle in every laboratory, measured with the same dye, so its value tests only the instrument. A cell population measured in a quality round carries everything else as well: blood from different donors, reagents from different vials and lots, staining by different hands, and gating by different analysts. The bead figure is the floor set by the instruments; the cell figure, near 30 % even after full standardization, is the realistic spread for intensities compared between sites. The comparison of the two published figures is made for this guide.
Without a shared scheme the spread is larger. A study across 133 cytometers found that cross-calibration with hard-dyed beads varied by typically 20 % or more, with variation between instrument models often greater than within one. The 2018 study of Chapter 8 found detection efficiency varying between analyzers of the same model by about a factor of two in some channels and by more than an order of magnitude in others. For vesicles, where no shared settings exist, Chapters 17 and 20 show how calibration in physical units narrows a spread that is far larger still.
What standardization buys is that populations land in the same place. A pattern of markers read in one laboratory can be compared with another's, and with reference databases of normal and malignant cells, which is the basis of the automated comparison that EuroFlow's panels were designed for.
Bead checks show that the instrument reads beads as it did yesterday. They do not detect a degraded tandem in a reagent vial, a wrong compensation matrix, a sample prepared differently, or a gate moved by hand. Quality control of the instrument is necessary for a trustworthy result and is not sufficient for one.
For an instrument used in a regulated or multi-site setting, define how detector voltages are chosen, the bead material and lot that carries the targets, the daily acceptance limits for value and spread, and what happens when a limit is crossed. Keep the charts as records, review trends at fixed intervals, and repeat the voltage and target procedure after any service that touches lasers, filters or detectors. Join an external quality assessment for each panel in routine use.
- Lasers, filters and detectors drift, so stability is measured daily with dyed beads and tracked on charts, with Q and B where the bead system allows.
- Detector voltages are chosen at the start of the plateau where a dim population's spread stops improving with gain.
- EuroFlow made instruments alike by setting each to target bead values with one bead lot, reaching variation below 5.5 % on beads across eight instruments.
- Cell intensities still varied by about 30 % between standardized laboratories, enough for populations to land in the same place, not at identical values.
Turning channels into units.
+ The questionA cytometer reports light in channels whose meaning differs from one instrument to the next. How can a channel be tied to a number of molecules, a diameter or a volume, and what does each conversion assume?
Channels have no units
The number a cytometer records for a particle depends on the detector voltage, the gain of the electronics, the efficiency of the optics and the converter's scale. Two instruments, or one instrument on two days with different settings, can give the same particle values that differ tenfold. Chapter 16 showed how laboratories make instruments read the same reference material alike. That is standardization: agreement on a scale, without saying what the scale means.
Calibration goes further. A 2023 review of vesicle cytometry defined it as "a procedure to relate the arbitrary units of measured data to comparable standard units, preferably SI units." A calibrated value says how many dye molecules' worth of light a particle gave, how large it is, or how many particles were in a milliliter. The conversion needs a reference material whose value is known with a stated uncertainty, and it needs a model of how the reference and the sample behave alike. Every calibration is only as good as that model.
Fluorescence in molecules
The oldest fluorescence unit is MESF, molecules of equivalent soluble fluorochrome: the number of free dye molecules in solution that would give the same fluorescence as the particle. The US National Institute of Standards and Technology issues a fluorescein solution, Standard Reference Material 1932, certified at 60.97 ± 0.40 micromoles per kilogram, for establishing that scale. Bead manufacturers assign MESF values to beads of several brightness levels, and running those beads gives a line that converts each channel of the instrument into molecules.
Suppose beads assigned 1,000, 10,000 and 100,000 MESF read 210, 2,100 and 21,000 in a channel. The line through them gives about 4.8 MESF for every channel unit, so a stained population reading 630 carries about 3,000 MESF. Real bead sets rarely fall on a line through zero: the dimmest beads often sit above it because of background, and the fit then has an intercept as well as a slope. The values are illustrative, for this guide.
MESF assumes that the bead and the stained cell carry the same dye in equivalent surroundings. Beads dyed through their bulk are generally not spectrally matched to the dyes used on cells, and each instrument's filters respond differently to the mismatch. The 2012 study of 133 cytometers concluded that any spectrally unmatched bead used as a general calibrator "must be verified and characterized for every particular instrument model." The response, a unit the Institute had defined in 2011 and the study applied channel by channel, was ERF, equivalent reference fluorophore. It is the number of molecules of a stated reference dye in solution that give the same signal as one bead, with fluorescein for the green channel, Nile Red for phycoerythrin's channel, allophycocyanin for its own and coumarin 30 for the violet channel. Assigning ERF values to beads carries an uncertainty of about 12 to 15 %.
Antibodies bound per cell
For immunophenotyping, the quantity of interest is often not dye but protein. A calibration in antibodies bound per cell uses a biological standard instead of a dye: a protein whose number of bound antibodies is known on a well-characterized cell, with CD4 on T cells, taken as about 40,000 antibodies bound per cell with one reagent, the usual choice (Chapter 5). The 2024 study at the Institute found that raw fluorescence of the same cells varied widely between four instruments, while values in antibodies bound per cell agreed closely, with coefficients of variation below 2 %; when two laboratories prepared their own antibody cocktails, the variation rose to about 5 to 11 %.
Antibodies bound are not proteins present. A 2018 study at the same institute estimated about 132,000 CD4 molecules per lymphocyte, and found that the number of antibodies bound differed about twofold between two labeled forms of anti-CD4, because an antibody can bind one or two copies of its target. The unit is therefore tied to the reagent and its clone, and a calibrated value should name them.
Scatter in nanometers
Scatter calibration is the hardest, because scatter depends on refractive index and collection angle as well as size (Chapter 4). The method published in 2012 measures beads of known size and refractive index, uses the physics of scattering by spheres, Mie theory, to compute how much light each bead sends into the instrument's detector, and derives the factor that converts channels into a physical quantity of scattered light, a cross-section in square nanometers. A software tool published in 2020 does the same from bead data, wavelength and collection angles, and fitted the collection angles of several commercial instruments to within 0.05 to 2.6 degrees of their specifications.
The cross-section is a property of the measurement; the diameter is an inference from it. Converting scatter into a vesicle diameter requires assuming the vesicle's refractive index, which varies from about 1.37 to 1.45 (Chapter 4), and that assumption dominates the uncertainty.
A side-scatter signal that a model assigns to a 150-nanometer vesicle of refractive index 1.40 fits a vesicle of about 120 nanometers if the index is 1.45, and about 200 nanometers if it is 1.37. For a signal assigned to 200 nanometers, the range is about 150 to 300. The figures are derived for this guide for side scatter collected between 45 and 135 degrees at 488 nanometers in water.
The reporting framework for vesicle cytometry draws the consequence: a gating strategy based on polystyrene beads alone is not a sound basis for standardization, because a bead of a given size and a vesicle of the same size scatter very differently. Gates and thresholds should be stated in calibrated scatter, with the model used to convert them into diameters.
Volume, the forgotten calibration
A concentration is a count divided by a volume, and the volume is easy to take for granted. Many cytometers infer the analyzed volume from the set flow rate and the run time. The 2018 interlaboratory study of vesicle cytometry found flow rates that differed sixfold between instruments set to the same value, so concentrations reported from set rates could differ by the same factor before any optical difference was considered. FCS 3.1 has a keyword for the analyzed volume, in nanoliters, but it is optional.
Clinical absolute counts solve the problem a different way: a known number of counting beads is added to a known volume of sample, and cells are counted relative to the beads, so the instrument's flow rate drops out (Chapter 21). For vesicles no such reference exists yet; the 2023 review noted that submicrometer reference materials with a traceably determined concentration are not commercially available. Chapter 20 shows how much calibration of flow rate, fluorescence and scatter together narrowed the spread of vesicle counts between laboratories.
An MESF value assumes spectral matching between bead and dye; antibodies bound per cell assume a reagent and its binding; a vesicle diameter assumes a refractive index. A calibrated number is more comparable than a channel value, but it is not model-free, and the model and the reference material used must travel with it.
For each reported quantity, state the unit, the reference material and lot, its assigned values and uncertainty, the fitting model, and for scatter the refractive indices and collection angles assumed. Calibrate the analyzed volume directly rather than trusting the set flow rate. Record calibration factors with the data; note that FCS 3.1 stores one multiplying factor per parameter, so a fit with an intercept must be documented outside the standard keyword.
- Channel values have no units; standardization agrees on a scale, while calibration converts it into molecules, diameters or volumes with a reference material and a model.
- MESF ties fluorescence to a NIST fluorescein standard but assumes spectral matching; ERF names a reference dye per channel, with uncertainties of about 12 to 15 %.
- Antibodies bound per cell varied by less than 2 % across four instruments where raw intensities did not, but they count bound antibodies, not proteins.
- Scatter calibrates to a cross-section; the diameter depends on an assumed refractive index, so a signal read as 150 nanometers can mean 120 to 200.
What the file keeps, and what it drops.
+ The questionEvery event ends up as a row in a data file. What does the standard file record about how its numbers were made, what is lost on the way to a report, and when does the file itself become the record that matters?
Three segments and a list of keywords
Cytometry data have had a standard format for four decades. The first Flow Cytometry Standard, FCS 1.0, was adopted in 1984 and revised in 1990 and 1997; version 3.1, published in 2010, is the one most instruments write, and a version 3.2 followed in 2021. Its stability is a large part of why files from different instruments and decades can still be opened by the same software.
An FCS file has three main segments. A short header says where the others begin. The text segment holds keywords, each a name and a value, such as the number of events, the number of parameters, and each parameter's name, range and detector settings. The data segment holds the measurements in list mode: one row per event and one column per parameter. Each parameter carries a short name for the detector, such as FL1-A, and may carry a long name for the axis label, such as "CD45 FITC Fluorescence - Area," so that the file can tell the detector from the marker it measured.
A file stores each measurement in 4 bytes when it uses 32-bit numbers. One million events with 24 parameters therefore need about 96 million bytes of data. A test for residual disease that acquires 10 million events produces a file of nearly a gigabyte. The standard's header keeps its offsets in eight digits, so once a segment ends beyond byte 99,999,999, its positions move into keywords in the text segment. The sizes are derived for this guide.
The data can be stored as unsigned integers, as 32-bit floating-point numbers or as 64-bit ones. The choice matters wherever values near and below zero occur, as they do after compensation (Chapter 10) in files that software exports or re-saves: they survive in floating point, while unsigned integers cannot hold negative numbers at all.
What the file records
FCS 3.1 has keywords for much of the acquisition. For each parameter it can record the detector type, voltage, gain and filter, and the wavelength and power of the light that excited it. It can record one trigger channel and its threshold (Chapter 9), the volume analyzed, and the spillover matrix, which the standard calls "the only standardized way to specify compensation." It states that "the stored data is always uncompensated," so compensation is applied by software from the matrix, and only one matrix can be stored for each data set.
Three keywords added in version 3.1 speak to the later chapters. One carries a calibration factor per parameter, converting channels into units such as MESF (Chapter 17); it is a single multiplier and is usually absent. One recommends a display scale, but recognizes only linear and logarithmic, so a logicle or arcsinh display cannot be recorded there. And one records the file's originality, from Original to DataModified, with the standard's advice that modifying data sets is bad practice and that "the original file should always be kept." The file can end with a checksum over the whole data set, but the checksum is optional, and a writer that skips it fills the space with zeros.
What the file does not record
The list of what FCS 3.1 has no keyword for is as instructive as what it has. It does not record the name or version of the acquisition software, only the computer and operating system. It has no keyword for laser delays, processing window, sheath pressure, baseline restoration or detector temperature (Chapter 13), none for the daily quality-control results of Chapter 16, and none for any trigger more elaborate than one channel above one level. Manufacturers write some of these into their own keywords, in units the standard does not define.
The consequence is that a file alone cannot reproduce an acquisition. It holds the measurements and much of the optical configuration, but several settings that decide whether those measurements are valid live only in the instrument, its log or its software. A data set exchanged between laboratories arrives without them unless they are documented separately.
What export and analysis drop
Further losses happen after acquisition. Exporting software often converts data to a linear scale, which is why the display keyword exists. A spreadsheet export, which the reporting standard MIFlowCyt permits for list-mode data, keeps the numbers and loses every keyword: the detector settings, the matrix and the provenance. Gates and transforms with their parameters are not part of FCS 3.1 at all, apart from crude legacy region keywords; they live in the analysis software's workspace or in a separate standard, Gating-ML 2.0, published in 2015, which can describe logicle and arcsinh transforms and gate boundaries exactly. A compensation matrix edited at analysis time also lives outside the file.
MIFlowCyt, published in 2008 and still the community's reference, asks for what the file cannot hold. A report "shall" state the denominator of every gate percentage, give the compensation matrix and say whether it is a spillover or a compensation matrix, and describe transforms by an open specification; the software name and version are only recommended. Chapter 20 shows how its extension for vesicles, MIFlowCyt-EV, adds the trigger channel and threshold, the flow rate, and calibrations in standard units. As of 6 October 2026, the public repository FlowRepository had temporarily stopped accepting new experiments for lack of storage, while downloads continued.
The file as a record
In regulated settings the file becomes evidence. The US rule on electronic records, 21 CFR Part 11, applies to records kept under the FDA's own record requirements, such as those of a device manufacturer, and requires validated systems, accurate and complete copies, protected retrieval and "secure, computer-generated, time-stamped audit trails." Clinical laboratories in the United States keep their records under CLIA instead, which requires quality-control and patient test records, "including instrument printouts," to be kept for at least two years and pathology reports for ten.
The file is also becoming the test object. EuroFlow's external quality assessment for residual myeloma sends participants FCS files rather than samples, so that laboratories are compared on analysis alone, with the variation of the instrument and the sample removed. For that comparison to mean anything, the files must arrive with their keywords intact.
A file exported to a spreadsheet, converted to integers, re-saved after editing or passed through software that rewrites keywords can lose the spillover matrix, the scale, negative values and its originality status, while still opening normally. The numbers look the same; the information needed to check them is gone.
Archive the original FCS files unchanged, with a checksum computed on arrival. Store the analysis workspace or Gating-ML file, the compensation matrix used, the transform parameters and the software version with each result. Record laser delays, sheath pressure, threshold and quality-control results in an instrument log linked to the files. Use MIFlowCyt as the checklist for reports, and set retention by the rules that apply: CLIA for a US clinical laboratory, and for records kept under FDA requirements the retention period of the governing FDA rule, with Part 11 controls on the electronic records.
- FCS files hold a header, keyword text and list-mode data; version 3.1 is the common standard, and data are always stored uncompensated with one spillover matrix.
- The file records detector settings, excitation, one trigger, volume and an optional calibration factor, but not laser delays, sheath pressure, software version or quality control.
- Export and analysis can drop keywords, scales, negatives and gates; Gating-ML and MIFlowCyt cover what the file cannot hold.
- In regulated use the file is a record kept under Part 11 or CLIA, and EuroFlow already compares laboratories on supplied files.
+ Part VI · Below the wavelength of light
The thread, followed to the limit.
Every chapter so far has met the guide's thread, extracellular vesicles, at the edge of what its subject could do. These two chapters follow the thread to the end. The first shows which limits fail first as particles shrink below about 200 nanometers, and how a count can go wrong without any warning. The second asks how many vesicles a sample contains and how large they are, and what calibration, reporting and comparison between laboratories have made of those questions. Together they answer the second part of the central question: how small a particle a cytometer can see, and what it takes to know.
What breaks below 200 nanometers.
+ The questionEvery limit described so far tightens as particles shrink. Below about 200 nanometers, which limits fail first, and how does a vesicle count go wrong without any sign of error?
Where most vesicles are
Extracellular vesicles are defined by the international guidelines MISEV2023, published in 2024, as particles released from cells, delimited by a lipid bilayer, that cannot replicate on their own. The guidelines note that vesicles below 200 nanometers are often described as "small" and those above it as "large," and add that measured diameter depends on the method used to measure it. The word exosome is reserved for vesicles shown to come from a particular compartment inside the cell, which a cytometer cannot show, so this guide speaks of vesicles.
Their sizes are not spread evenly. Measured by electron microscopy and by sensitive cytometers, the number of vesicles in blood plasma rises steeply as diameter falls, and a 2014 study of urine vesicles fitted their distribution with a power law because that form was least affected by each method's smallest detectable size. Most vesicles, by number, are therefore at the small end, exactly where Chapter 4 showed scatter vanishing.
How many there are depends on what is counted. A 2019 review found that reported concentrations of vesicles in plasma spanned more than seven orders of magnitude, with a geometric mean of about 10¹⁰ per milliliter. Lipoproteins, particles of fat and protein that carry cholesterol in blood, reach about 10¹⁶ per milliliter, which made the vesicle, in the review's words, "the one in a million." A cytometer measuring vesicles in plasma is looking for rare particles among far more numerous ones of similar size.
Four limits fail together
Four limits that are separate for cells converge for vesicles. Scatter falls roughly with the sixth power of diameter (Chapter 4), so a vesicle of 100 nanometers sends almost nothing to the detectors. Its label is about a thousandth of its parent cell's (Chapter 5), so fluorescence is close to the background (Chapter 8). And the particles are so numerous and so small that many can sit in the beam at once (Chapter 6). An instrument built for cells meets all four at the same time.
The result is that the smallest detectable size depends on the instrument more than on anything else. A 2014 study measured one sample of urine vesicles by several methods, two of them cytometers. Their lower limits of detection were about 70 to 90 nanometers for particle tracking, 70 to 100 for resistive pulse sensing, 150 to 190 for a cytometer built for small particles and 270 to 600 for a conventional cytometer, which found about 300 times fewer vesicles than the methods reaching smallest. In a 2018 comparison of 46 instruments, 24 % could not detect polystyrene beads of 400 nanometers, which scatter far more than vesicles of that size.
Better instruments go further. In the largest interlaboratory study so far, published in 2025, the more sensitive half of the instruments reached vesicle diameters of roughly 110 to 135 nanometers and labels of the order of 10 to 100 molecules of equivalent fluorochrome, while the 95th-percentile instrument reached about 240 nanometers. Even the best cytometers see only the larger, brighter fraction of the vesicles in a sample, a bias the lead author's thesis stated explicitly.
Swarm detection
The fourth limit produces the strangest failure. A 2012 study showed that when vesicles below the detection limit are present at high concentration, many of them can be illuminated at once and counted as a single event. It named the effect swarm detection: "multiple vesicles are simultaneously illuminated by the laser beam and counted as a single event signal." No single vesicle crosses the threshold; together, they do.
The experiment that demonstrated it was simple. Silica beads of 89 nanometers, too small to detect on the instrument used, were mixed with silica beads of 610 nanometers at increasing ratios. At 100 small beads for each large one, the measured count stayed similar. At 10,000 for each large one, it rose by 40 %, and at 100,000, by 350 %, although the number of large beads had not changed. On that instrument the smallest single beads it could detect were about 200 nanometers, and the smallest detectable single vesicle was estimated at 300 to 700 nanometers.
Swarm detection is not a gain in sensitivity. Each event reports the combined light of an unknown number of particles, so its scatter suggests a single larger particle and the count is wrong in both directions: too many events, because particles that are individually invisible are counted, and too few particles, because each event stands for many. A later review described it as a special form of coincidence that can be reduced by dilution.
Dilution reveals it
Dilution is the test because single particles and swarms respond to it differently. When a sample of single, separately detected particles is diluted twofold, the event rate halves and the scatter and fluorescence of each event stay the same. When events are swarms, dilution reduces the number of particles in the beam at any moment, the combined signals shrink, and many fall below the threshold: the event rate falls faster than the dilution, and the median signal per event falls with it.
Suppose a plasma sample gives 8,000 events per second at a dilution of 1 in 10. At 1 in 20, single particles would give 4,000, with the same median scatter. If the instrument instead records 2,000 events per second and the median scatter has dropped by half, most events at the first dilution were swarms. The series is repeated until further dilution halves the rate and leaves the medians unchanged; counts are trusted only in that range. The values are illustrative, for this guide.
The same test protects cell counts at high event rates, where coincidence rather than swarming adds and merges events (Chapter 6), but for vesicles it is not optional. The reporting framework for vesicle cytometry lists a serial dilution among its controls for this reason.
Not everything is a vesicle
Even single, separately detected events are not all vesicles. Lipoproteins outnumber vesicles in plasma, and the 2019 review found that preparations of isolated vesicles contained at least ten times as many low-density lipoprotein particles as vesicles. Antibodies and dyes form aggregates of vesicle size that are labeled and scatter (Chapter 11). Cell fragments and protein complexes add more.
The controls of Chapter 11 answer this. Buffer alone shows the instrument's own events; buffer with the labeling reagents shows aggregates; and a detergent that dissolves lipid membranes removes vesicles while leaving aggregates of antibody and dye behind. In one completed report under the vesicle framework, the labeled events fell by more than 90 % after detergent. An event that survives detergent is not a vesicle, whatever its size and brightness.
At high concentration, a cytometer can report more events than there are detectable particles, by counting swarms of invisible ones, and each such event looks like one larger particle. Without a dilution series, a higher count from one sample than another may reflect concentration, not biology, and an apparent size distribution may be an artifact of the swarm.
For every sample type, run a serial dilution and use only the range in which event rate is proportional to dilution and medians are constant. Include buffer-only, reagent-only and detergent controls in each run. Trigger on fluorescence where the label allows (Chapter 9), and state the detection limits in calibrated scatter and fluorescence units (Chapter 17) with every count.
- Most vesicles by number are small, and plasma holds far more lipoproteins of similar size, so a vesicle count is a search for rare particles near every limit at once.
- Scatter, label, background and coincidence limits converge below 200 nanometers; a conventional cytometer detected vesicles only above about 270 to 600 nanometers in one comparison.
- Swarm detection counts many invisible vesicles as one event: adding 100,000 sub-threshold beads per real one raised counts by 350 %.
- A dilution series in which rate falls with dilution and medians stay constant shows events are single, and detergent shows they are vesicles.
How many vesicles, and how big.
+ The questionA vesicle count means something only with its size window, its brightness window and its volume. What do calibrated cytometers report today, how far do laboratories agree, and how do their numbers compare with other methods?
A count is a window
The question "how many vesicles are in this plasma?" has no single answer. Chapter 19 showed reported concentrations spanning more than seven orders of magnitude. Methods that count every particle above about 70 to 100 nanometers give values near 10¹⁰ per milliliter, and include lipoproteins. A calibrated cytometer counting vesicles that carry a cell-type marker, between 150 and 1,000 nanometers and above a stated brightness, reports about 10⁷ per milliliter for each cell type. Both can be right, because they count different things.
Every vesicle count therefore needs three limits stated with it: the smallest and largest diameter included, in calibrated scatter with the refractive index assumed (Chapter 17); the dimmest label included, in calibrated fluorescence units; and the analyzed volume, measured rather than assumed. Without them, the count describes the instrument as much as the sample.
Suppose vesicle numbers rise steeply toward small sizes, so that each halving of the lower size limit multiplies the count by about eight. A method that counts from 75 nanometers would then find about eight times as many vesicles as one that counts from 150, and about sixty-four times as many as one that starts at 300, from the same sample. The steepness is illustrative, chosen for this guide; real distributions differ by sample and are poorly known at the smallest sizes. The point holds whatever the exponent: a change in the lower limit changes the count more than most biological differences do.
Reporting what was measured
The community's answer is a reporting framework, MIFlowCyt-EV, published in 2020 by a working group of the societies for vesicles, cytometry and thrombosis. It extends MIFlowCyt (Chapter 18) with what vesicle measurements need and, in its own words, "does not prescribe specific protocols"; it asks that whatever was done be reported in a form others can compare. The international guidelines MISEV2023, published in 2024, point to it for flow cytometry.
Its components cover the sample before analysis, its preparation, the controls of Chapters 11 and 19 (buffer, reagent, unstained, isotype, single-stained, procedural, serial dilution and detergent), and the instrument's calibration: the trigger channel and threshold, the flow rate, and fluorescence and scatter calibration. Results are reported as a calibrated detection range, a concentration in the starting material and brightness, preferably in standard units. The framework also states that a gating strategy based on polystyrene beads alone is not a sound basis for standardization, the conclusion of Chapter 4's history of bead mixtures.
How far laboratories agree
The history of interlaboratory comparisons shows calibration closing a very wide gap. Workshops of the international thrombosis society in 2009 and 2010, as summarized in a 2023 review, used a bead mixture to set size gates, and the counts proved comparable only among instruments of the same type. A 2017 workshop of 44 laboratories found that counts of platelet vesicles in its two samples varied by 37 and 28 % between laboratories; the maker of the beads it used supplied them and employed two of its authors.
A 2018 study by the same society sent samples to 33 participants with 46 instruments. With gates set by bead diameter, counts varied by 139 %; with gates set in vesicle diameter, derived from scatter models, by 81 %, for a gate of 1,200 to 3,000 nanometers that 31 of the 46 instruments could detect. The largest comparison so far, published in 2025 with 25 cytometers from 18 laboratories, calibrated flow rate, fluorescence and scatter on every instrument. The median absolute deviation of counts between laboratories fell from 67 % before calibration to 25 to 31 % after it.
The 2025 figure is a median absolute deviation of counts, not a coefficient of variation of intensities, so it is not directly comparable with the variation of about 30 % that fully standardized clinical laboratories show for cell intensities (Chapter 16), although the two are of the same order. It applies, however, only to vesicles inside the gates that the compared instruments shared: above about 113 to 133 nanometers when the more sensitive half of the instruments were compared, and above about 240 nanometers, with higher fluorescence thresholds, when 18 to 19 of the 25 were compared. Calibration has made vesicle counts comparable between laboratories for the larger, brighter fraction of vesicles, not for vesicles as a whole.
The first reference ranges
Comparable counts made reference ranges possible. The 2025 study measured plasma from 224 healthy donors on one calibrated instrument, in a window of 150 to 1,000 nanometers and above stated fluorescence thresholds, and described the results as the first reference ranges of cell-type-specific vesicles in human plasma. Median concentrations were about 20 million per milliliter for vesicles from red cells, 18 million for those from white cells and 34 million for those from platelets.
The same group had described a plasma vesicle test sample in 2023, a ready-to-use material resembling plasma particles and stable for 12 months at −80 °C, so that laboratories can check their whole procedure on a shared sample. Reference materials with traceable size and concentration remain the missing piece: a European project set targets of 50 to 1,000 nanometers, refractive indices of 1.37 to 1.42 and concentrations of 10⁹ to 10¹² per milliliter, with candidates ranging from hollow organosilica beads and liposomes to recombinant vesicles. As of October 2026 this guide found no cleared or approved diagnostic test based on vesicle cytometry; the measurement is a research tool on its way to becoming a clinical one.
Other methods see other vesicles
Comparisons with other methods belong here, because they are comparisons of windows. A 2014 study measured one pooled sample of urine vesicles by electron microscopy, resistive pulse sensing, particle tracking and two cytometers. In its words, each technique "gave a different size distribution and a different concentration for the same vesicle sample," with concentrations from 10⁴ to 10¹² per milliliter, and the differences were "primarily caused by differences between the minimum detectable vesicle sizes." The size distribution peaked at 45 nanometers by electron microscopy, 75 by resistive pulse sensing and 95 by particle tracking; for the last two, the peak lay just above the method's own lower limit.
The same study judged the dedicated cytometer the most accurate at sizing reference beads but expected it to be less accurate on vesicles, because their refractive index varies, and named its strengths as speed and the ability to measure several fluorescent labels on each particle. Particle tracking, the method behind most high concentration estimates, counts lipoproteins and other particles along with vesicles. No method sees every vesicle; each sees those above its own threshold, and the cytometer is the one that can say which cell they came from.
A count from particle tracking, a count from a conventional cytometer and a count from a calibrated small-particle cytometer describe different subsets of the same sample. Comparing them, or comparing a published value with a laboratory's own without matching the size window, label threshold and volume calibration, compares methods rather than samples.
Follow MIFlowCyt-EV for every report: state the size window in calibrated scatter with the refractive index assumed, the fluorescence threshold in calibrated units, the measured analyzed volume, the dilution range in which counts were proportional, and the results of buffer, reagent and detergent controls. Run a shared test sample where one exists, and compare counts with other laboratories or methods only within a matched window.
- A vesicle count is defined by its size window, brightness threshold and measured volume; without them it describes the instrument as much as the sample.
- MIFlowCyt-EV asks for controls, trigger, flow rate and calibrations in standard units, and rules out bead gates as a basis for standardization.
- Calibration of flow rate, fluorescence and scatter cut the spread between laboratories from 67 % to 25 to 31 %, for the larger and brighter vesicles.
- The first reference ranges for plasma vesicles by cell type appeared in 2025, and other methods report different counts because they see different windows.
+ Part VII · Cytometry as a test
From a measurement to a medical result.
A research measurement has to convince its author; a clinical test has to be right for a patient who will never see the data. These three chapters follow cytometry into the clinic through the guide's second case, leukemia immunophenotyping. The first shows how marker patterns become part of a diagnosis and what a report contains. The second shows how a laboratory proves that an assay performs as claimed, down to the rarest cells it reports. The third shows how the instrument and its reagents are regulated in the United States, the European Union and India. Together they answer the last part of the central question: when a count can be trusted as a test result.
When a result becomes a diagnosis.
+ The questionA clinical laboratory turns a tube of blood or marrow into a sentence in a patient's record. How do patterns of markers point to a disease, what does a report contain, and what can the cytometer not decide on its own?
What clinical cytometry is asked
The US classification for flow cytometric leukemia and lymphoma panels lists the conditions they help to diagnose: acute and chronic leukemias, non-Hodgkin lymphomas, myeloma, myelodysplastic syndromes and myeloproliferative neoplasms, in blood, bone marrow and lymph node. Beyond these, clinical laboratories use cytometry to count CD4 T cells in people with HIV and CD34 stem cells before transplantation. They also use it to detect the missing surface proteins of paroxysmal nocturnal hemoglobinuria, to screen for inherited immune deficiencies and to look for leukemia left after treatment.
An international consensus meeting in 2006 set out when diagnostic cytometry is indicated. A 2025 educational module from the College of American Pathologists restates the list for blood: abnormal cells seen under the microscope, an increase in lymphocytes that may be reactive or malignant, increased blasts, suspected relapse and several rarer situations. It also says when cytometry is not indicated. Normal mature neutrophils, red cells and platelets look the same whether or not they come from a clone, so an isolated rise in neutrophils, red cells or platelets is not a reason to order it. Routine diagnostic panels characterize abnormal populations above about 1 % of cells; detecting residual disease below 0.01 % is a separate, specialized test (Chapter 22).
A pattern, not a marker
No single marker identifies a leukemia. Normal blood and marrow contain cells at every stage of maturation in every lineage, and almost every marker found on leukemic cells is also found on some normal cell. What identifies a population is a combination: which markers it carries, how brightly, and in what pattern relative to its normal counterpart (Chapter 11).
The EuroFlow panels encode this in their design. Each eight-color tube combines backbone markers, which identify the main populations in the sample, with characterization markers, which describe them further. In multi-tube panels the backbone sits on the same dye in every tube, so the population of interest lands in the same place in each tube and the tubes can be analyzed together (Chapter 16).
The panel's entry point for suspected acute leukemia is a single tube, the acute leukemia orientation tube, which asks one question: which lineage? It pairs markers found inside the cell, myeloperoxidase for the myeloid lineage, CD3 for T cells and CD79a for B cells, with surface markers of immaturity and lineage, including CD34, CD45, CD19, CD7 and surface CD3. It was refined over five design cycles on 385 acute leukemias and then tested prospectively on 483 further cases in eight laboratories. In the textbook reading, myeloperoxidase points to a myeloid leukemia, cytoplasmic CD3 with CD7 to a T-lineage one, and CD19 with cytoplasmic CD79a to a B-lineage one, while CD34 and dim CD45 mark the cells as immature.
A worked case: chronic lymphocytic leukemia
Chronic lymphocytic leukemia shows how a pattern becomes a criterion. The 2018 international guidelines require at least 5 × 10⁹ clonal B lymphocytes per liter of blood, persisting for at least three months, with clonality shown by cytometry. The cells carry CD5, normally a T-cell marker, together with the B-cell markers CD19, CD20 and CD23, and their surface immunoglobulin, CD20 and CD79b are characteristically dim compared with normal B cells.
Clonality is shown through the light chains of the antibodies B cells carry on their surface. Every B cell uses either a kappa or a lambda light chain, and a normal population contains both. A clone, descended from one cell, uses only one. A population of CD5-positive B cells that is all kappa or all lambda is therefore clonal, which is the measurement the diagnosis rests on.
The thresholds draw the boundaries of the diagnosis. The same clonal cells below 5 × 10⁹ per liter, without enlarged lymph nodes or organs, low blood counts or symptoms, define monoclonal B lymphocytosis; with enlarged nodes they suggest small lymphocytic lymphoma, confirmed by biopsy. A European harmonization project in 2018 agreed on six required markers, CD19, CD5, CD20, CD23, kappa and lambda, plus recommended markers for borderline cases, and validated the approach retrospectively on 14,643 cases from 13 centers with more than 97 % concordance.
Counting cells in absolute numbers
Some clinical results are counts per microliter rather than percentages. The US Centers for Disease Control and Prevention's 2003 guidelines for CD4 counts favored single-platform methods, in which a known number of fluorescent counting beads is added to a measured volume of blood. The alternative, dual-platform methods, multiply a cytometry percentage by a count from a separate hematology analyzer and compound the errors of both.
Suppose a tube receives 50,000 counting beads and 50 microliters of blood, so there are 1,000 beads for each microliter. After staining, the cytometer records 40,000 bead events and 20,000 CD4 T-cell events. Recording 40,000 of the 50,000 beads means that 40 microliters of blood were analyzed, so the count is 20,000 ÷ 40,000 × 1,000, or 500 CD4 T cells per microliter. The guidelines ask for at least 2,500 gated lymphocytes and accurate pipetting, because errors in the bead count and the blood volume pass straight into the result. The values are illustrative, for this guide.
The same principle counts CD34 stem cells, where the US Pharmacopeia's method asks for at least 75,000 CD45-positive events and at least 100 CD34-positive cells. Because both the cells and the beads are counted in the same run, the instrument's flow rate drops out of the result, the solution to the volume problem that Chapter 17 found still open for vesicles.
The cytometer is one witness
A cytometry result is never a diagnosis on its own. The 2022 World Health Organization classification of myeloid neoplasms describes itself as predicated on integrating morphology, immunophenotype and molecular and cytogenetic data, and it dropped the requirement of 20 % blasts for most acute myeloid leukemias that carry defining genetic abnormalities. A parallel International Consensus Classification, also of 2022, takes the same multiparameter approach. The US regulation for leukemia and lymphoma panels requires their labeling to say that results "should be interpreted by a pathologist or equivalent professional" together with other clinical and laboratory findings.
The 2013 recommendations of the international hematology and clinical cytometry societies describe the report that results. Interpretation is reserved for people who know malignant hematology, instrumentation, software and data analysis. Abnormal populations are quantified as percentages and described by their intensity, bright or dim, relative to normal counterparts, and the findings are correlated with morphology. Counts such as lymphocyte subsets carry a comment against the reference range. An Italian society's format for acute leukemia lists 13 elements, from specimen type and sample quality, including viability, to the gating procedure, the blast phenotype and the panel to use later for residual disease.
Blasts can be reported as a percentage of all events, of CD45-positive cells, of nucleated cells or of a gated population, and the numbers can differ widely in a marrow with many red-cell precursors or much debris. A report that does not state the denominator, and the gating behind it, cannot be compared with a microscope count or with the next report on the same patient.
For a panel developed for clinical use, state the intended use (conditions, specimen types, the population reported and its denominator), the limit of detection for an abnormal population, and the interpretation statement the labeling will carry. For a laboratory, fix a report template that names the specimen, sample quality and viability, the gating strategy, each population with its denominator and intensity relative to normal, absolute counts where relevant, and the interpreter's comment.
- Clinical cytometry helps diagnose leukemias, lymphomas and related neoplasms, counts CD4 and CD34 cells, and detects residual disease, but it is not indicated for isolated rises in mature blood cells.
- Lineage is read from patterns: EuroFlow's orientation tube pairs internal lineage markers with surface markers, refined on 385 leukemias and tested on 483.
- Chronic lymphocytic leukemia rests on CD5-positive B cells with a single light chain at 5 × 10⁹ per liter or more for at least three months.
- Results are integrated with morphology and genetics by a qualified interpreter, and every reported percentage needs its denominator.
Proving an assay does what it claims.
+ The questionAn instrument can be verified with beads, but a clinical result depends on cells that have no certified reference. How does a laboratory show that a flow assay is precise, sensitive and stable enough, and how many cells does a rare-event result need?
Validation without a reference cell
Most clinical tests are anchored to reference materials: a solution of known glucose, a serum of known hormone concentration. Cytometry has almost none for the quantities it reports. The 2013 recommendations of the international hematology and clinical cytometry societies, published in five parts, state the problem plainly: stable reference preparations "are unavailable for the cell-based measurements," and tracing a cell measurement back to a defined molecular quantity is not yet feasible. A laboratory must therefore prove its assay with patient samples, stabilized cells and careful design rather than with certified standards.
The most complete guide to doing so is the Clinical and Laboratory Standards Institute's H62, Validation of Assays Performed by Flow Cytometry, a 234-page first edition published on 27 October 2021. Its governing idea is fit for purpose: "the level of validation should be tailored to the intended use of the data," which, it adds, is "not an endorsement for inadequate validation." The institute noted that before H62 there were no official guidance documents for validating flow assays, and this guide found no FDA guidance specific to them.
In the United States the legal floor is set by the CLIA regulations. A laboratory using an unmodified cleared test verifies its accuracy, precision and reportable range and confirms the reference intervals for its own patients. For a test it developed or modified, it must establish accuracy, precision, analytical sensitivity and specificity, reportable range and reference intervals itself.
What kind of result
What has to be validated depends on what the assay reports. H62 and the 2013 recommendations sort flow assays into categories. A quantitative assay with true reference standards reports absolute values, which cytometry rarely achieves. A quasi-quantitative assay reports numbers that are proportional to the quantity but have no calibration standard; a CD4 count is the usual example. A qualitative assay reports a category, such as the presence of an abnormal population with a given phenotype, and leukemia immunophenotyping belongs here. The US authorization of the first leukemia panels described them as qualitative assays.
The category decides the validation menu. A quantitative or quasi-quantitative assay needs precision across its range, linearity, limits of detection and quantitation, and reference intervals. A qualitative assay needs agreement with a reference diagnosis, precision of the call, and the smallest abnormal population it can reliably detect, which the US special controls express as limits of blank, detection and quantitation. Both need specimen and reagent stability, carryover and the effect of reagent lots.
Precision and agreement
Precision is measured with replicates, but each replicate is already the average of thousands of cells. The 2013 analytical recommendations therefore limited replicates to no more than six, "since each 'measurement' actually is derived from measurements of tens of thousands of cells." Targets from the series' fifth part, on assay performance, as summarized in a 2024 teaching workshop, are a coefficient of variation below 10 % for most results and below 20 % for populations of about one in a thousand or rarer. The US special controls for leukemia and lymphoma panels go further for manufacturers, requiring precision on clinical samples at a minimum of three sites, at least two of them external, and reproducibility across at least three reagent lots.
Agreement is where the category matters most. In the manufacturer's study behind the first authorized leukemia panels, reviewed by the FDA, 279 specimens from four sites were read by two blinded experts and compared with each site's final diagnosis. Their sensitivity was 82 and 86 %, their specificity 94 and 93 %. A later submission by the same manufacturer, for a newer instrument, reported 100 % agreement with its predicate instrument across 512 specimens stained with the ten-color successor panels. Both figures are correct, and they answer different questions: the first measures how often the panel's reading matches the patient's eventual diagnosis, the second how often two instruments give the same reading.
Rare events and the limits of detection
Residual-disease testing pushes validation to its limit. The 2018 guidelines define undetectable disease in chronic lymphocytic leukemia as fewer than one leukemic cell in 10,000 leukocytes. For myeloma, a 2017 study by the EuroFlow consortium described next-generation flow: two eight-color tubes, bulk lysis to acquire at least 10 million cells per sample, and automated gating against a database of normal marrows. It defined the limit of detection as 20 abnormal plasma cells among at least 10⁷ events, 2 in a million or lower, and the lower limit of quantitation as 50, 5 in a million or lower. In 110 marrows from patients in deep response, 47 % were positive by the new method against 34 % by conventional eight-color cytometry.
The limits are counts, so they depend on how many cells were actually acquired. In the same study the median sample yielded 10.4 million cells, and 7 of 110 fell below 7 million. For B-cell precursor leukemia, a 2017 study found flow cytometry at least as sensitive as the molecular method, reaching 1 in 100,000, "if sufficient cells are measured," more than 4 million. In a 2022 analysis of acute myeloid leukemia, the target of 500,000 CD45-positive events was reached in 158 of 261 patients.
Counts of rare cells follow Poisson statistics (Chapter 12). If a sample truly contains enough abnormal cells to give 20 events on average, a single measurement shows 20 or more only about 53 % of the time; a 95 % chance of reaching 20 needs an expected count of about 28. Seeing at least one abnormal cell with 95 % probability needs an expected count of 3. The 20- and 50-event conventions are rules for how many events make a population, not statistical detection limits in the sense used for chemistry tests. The probabilities are derived for this guide.
Regulators have taken note of residual disease as evidence. A 2020 FDA guidance on its use in drug development asked that an assay detect at least tenfold below the clinical decision threshold, and an FDA advisory committee voted 12 to 0 in April 2024 that residual disease could support accelerated approval of myeloma drugs. As of October 2026, the only residual-disease device this guide found authorized in the United States was a sequencing test; flow cytometric residual-disease assays run as laboratory-developed tests.
Stability, lots and quality over time
A validated assay stays valid only within its tested conditions. The 2013 preanalytical recommendations describe blood in heparin as optimally stable for 48 to 72 hours and in EDTA for up to 48 hours at room temperature, and note that some markers degrade faster with time than others, CD138 and CD16 more than CD45 or CD64. CD138 marks plasma cells, so in a myeloma residual-disease test the age of the marrow sample bears directly on the result. Stability is validated by testing at least five specimens of each type and anticoagulant over time, up to the oldest sample the laboratory will accept.
Reagent lots are a recurring change. A new lot, or any major hardware change, calls for a full compensation procedure under the analytical recommendations, and tandem dyes need lot-specific values (Chapter 10). Over time, quality is held by daily instrument checks (Chapter 16), by external quality assessment, which the 2013 recommendations make mandatory at least three times a year, and by laboratory accreditation, increasingly to ISO 15189:2022. One UK scheme runs 11 accredited and 6 pilot programs for cytometry, from stem-cell counts to residual disease. List-mode files and analysis files are kept for at least two years (Chapter 18).
A residual-disease result reported as negative at 1 in 100,000 is negative at that level only if enough cells were acquired, the sample was fresh enough and the controls passed. A sample that yields a tenth of the target cells has a detection limit ten times higher, and a report that omits the number of cells analyzed hides that change.
State the assay category, then list the parameters it needs, with acceptance criteria: precision with replicates and sites, agreement with a stated reference, limits of detection and quantitation as event counts and cell numbers, specimen and reagent stability, carryover, lot-to-lot reproducibility and reference intervals. Use H62 and the 2013 recommendations as the checklist. For rare-event assays, report the number of cells acquired and the resulting detection limit with every result.
- Cytometry lacks certified reference cells, so assays are validated with patient samples, stabilized cells and design; CLSI H62 of 2021 sets a fit-for-purpose approach.
- The assay category, quantitative, quasi-quantitative or qualitative, decides what must be shown; agreement with a diagnosis differs from agreement between instruments.
- Residual-disease limits are counts: 20 abnormal cells in 10⁷ for myeloma, which only samples that actually reach 10⁷ cells can claim.
- Stability, reagent lots, external quality assessment and accreditation keep a validated assay valid over time.
From panel to authorized product.
+ The questionWhat must a cytometer, its reagents and a laboratory's own panel show before they may be used on patients in the US, the EU and India?
Three objects, three questions
Clinical cytometry puts three kinds of object in front of regulators. The instrument measures light. The reagents, single antibodies or premixed panels, with the software and templates that come with them, decide what is measured. And many laboratories build their own assays from single reagents, validate them themselves (Chapter 22) and report the results. Each jurisdiction treats the three differently.
The evidence asked for is more uniform. The European regulation for diagnostic devices names three questions. Scientific validity: is the marker pattern associated with the condition? Analytical performance: does the complete system, instrument included, detect and measure it correctly? Clinical performance: do its results correspond to the condition in the intended patients? The status of everything below is stated as of October 2026.
United States
In the United States the FDA classifies devices by risk and type, and the Clinical Laboratory Improvement Amendments, CLIA, regulate the laboratories that use them.
Instruments and most cleared reagents. Cytometers are cleared under the regulation for automated differential cell counters, 21 CFR 864.5220, whose text covers devices using "a flow cytometric method utilizing monoclonal CD (cluster designation) markers," under a product code for flow cytometric reagents and accessories. These are class II devices that need a premarket notification, a 510(k). Clearances include analyzers from two manufacturers between 2013 and 2023. One, cleared on 3 July 2017, was given an in vitro diagnostic claim for six fluorescence channels only, although its larger configurations have more. Another, cleared on 22 November 2023, replaced the photomultipliers of its predicate with avalanche photodiodes (Chapter 8) and was cleared on a method comparison. No full-spectrum analyzer was found cleared in the United States; the clearances found are for conventional instruments, although spectral analyzers have been approved for clinical use elsewhere, including China.
Leukemia and lymphoma panels. The first panels were authorized through the De Novo route: a request for five-color reagents was received on 3 October 2016 and granted on 29 June 2017, and the FDA called it the first authorization of a flow cytometry test to aid in detecting these blood cancers. It created a new regulation, 21 CFR 864.7010, codified on 27 December 2017 and amended in December 2025, without change of substance, to refer to the new Quality Management System Regulation. Its special controls ask for the neoplasms validated, the number of events to collect, precision at three or more sites, reproducibility across three reagent lots, carryover, stability, limits of blank, detection and quantitation, and "clear examples of all expected phenotypic patterns and gating strategies." A ten-color successor cleared by 510(k) in March 2019 carries CD45 and CD34 on the same dyes in all four of its tubes, the backbone design of Chapter 21 in a commercial product.
Single reagents and laboratory tests. A single antibody sold for laboratories to build their own tests is an analyte-specific reagent, class I and exempt by default. It may be sold only to manufacturers, high-complexity laboratories and non-diagnostic users, without performance claims, and a laboratory reporting a test built from such reagents must state: "This test was developed and its performance characteristics determined by (Laboratory Name). It has not been cleared or approved by the U.S. Food and Drug Administration." Under the FDA's 2007 guidance, reagents bundled as a preconfigured mixture are not analyte-specific reagents, so a premixed clinical cocktail needs its own clearance. Reagents labeled for research use only may not be promoted for diagnosis.
Laboratory-developed tests. The FDA's 2024 rule bringing such tests under device regulation was vacated by a federal court on 31 March 2025, and the agency removed its language from the regulations on 19 September 2025. Flow cytometry tests developed in laboratories therefore remain under CLIA, as high-complexity tests that must establish their own performance. No flow cytometric residual-disease assay was found authorized by the FDA; the one authorized residual-disease device found is a sequencing test, granted in September 2018.
European Union
The In Vitro Diagnostic Medical Devices Regulation, the IVDR, has applied since 26 May 2022. It sorts devices into classes A to D by rules in its Annex VIII, and the class decides whether the manufacturer self-declares conformity or a notified body assesses it.
Instruments. Rule 5 places instruments that the manufacturer specifically intends for diagnostic procedures in class A, while the reagents they run are classed in their own right. A cytometer sold for use with antibody reagents falls there by this reading, and a non-sterile class A device is self-declared.
Reagents. The classification guidance of the Medical Device Coordination Group, MDCG 2020-16, gave no flow cytometry example in its September 2026 revision. It does place CD4 T-cell counting in people with HIV in class C, under the rule for managing life-threatening disease, and it places FISH panels for lymphoma, myeloma and leukemia in class C under the rule for cancer diagnosis. Leukemia immunophenotyping panels are therefore class C by close analogy rather than by a named example.
Transition. Reagents of this kind were generally self-certified under the old directive and now need a notified body for the first time. Under Regulation (EU) 2024/1860, a legacy class C device may stay on the market until 31 December 2028 only if its manufacturer had an IVDR quality management system in place by 26 May 2025, applied to a notified body by 26 May 2026 and signed an agreement by 26 September 2026, and the device has not changed significantly. Both dates have passed, so a legacy panel that missed them cannot rely on the extension.
In-house devices. A hospital laboratory that assembles a panel from research-use reagents and uses it for diagnosis makes an in-house device under Article 5(5), which the coordination group's 2023 guidance confirmed for research-use components. The laboratory must work under an appropriate quality system, comply with EN ISO 15189 or national provisions and publish a declaration. From 26 May 2028 it must also justify why no equivalent CE-marked device meets its patients' needs. A Commission proposal of December 2025 would drop that condition; as of October 2026 it was still at committee stage in the European Parliament.
India
The Medical Devices Rules, 2017, in force since 1 January 2018, classify diagnostic devices into classes A to D, and the Central Drugs Standard Control Organisation publishes the classification of device types. Its list updated on 25 October 2023 names the "Flow cytometry analyser," intended to count, examine or sort cells or microscopic particles in a clinical specimen, as class A, beside hematology and blood-cell-count analyzers. Neither that list nor the 2017 list names a flow cytometry reagent; by analogy with tumor-marker tests, which the 2017 list mostly places in class C, leukemia panels are probably class C, but this guide could not confirm it.
Licensing has been mandatory for classes A and B since 1 October 2022 and for classes C and D since 1 October 2023, with manufacturing licenses issued by the state authority for classes A and B and by the central authority for classes C and D. New devices in classes B to D need a performance evaluation on three batches by an accredited or government-recognized laboratory. Products for research use that are not intended for diagnosis fall outside the rules. This guide found no Indian pathway corresponding to CLIA's laboratory-developed tests or the IVDR's in-house devices.
One evidence core, three presentations
The three systems classify the same objects differently. The instrument is class A in the EU and India and class II with premarket notification in the United States. A leukemia panel is class II under a dedicated regulation in the United States, class C by analogy in the EU and probably class C in India. A laboratory's own panel is a CLIA test, an in-house device, or an unaddressed case, depending on the country.
The evidence underneath barely changes. Intended use, precision across sites and lots, analytical sensitivity as events and cells, agreement with diagnosis, stability, carryover and an interpretation statement answer all three, and the laboratory guidance of Chapter 22 supplies the study designs. A manufacturer's file and a laboratory's validation therefore look alike, and a developer who builds the evidence once, organized by the system's requirements, can present it three ways.
A reagent labeled for research use only carries no claim of clinical performance. When a laboratory uses it to diagnose, the responsibility moves to the laboratory: in the United States the test becomes a laboratory-developed test under CLIA, in the EU an in-house device under Article 5(5), and in India the rules give no clear route. In none of the three does the label make clinical use lawful by itself.
For a cytometer, a reagent panel or a laboratory's own assay, write the intended use once, determine its class in each target market against current regulations and guidance, and record the route, the assessor and the evidence required. Keep a dated register of the deadlines that apply, including the EU's 26 May 2028 condition for in-house devices, and recheck it whenever guidance is revised.
- In the US, cytometers and cleared reagents other than leukemia panels are class II under 21 CFR 864.5220, leukemia panels fall under 21 CFR 864.7010 since a 2017 De Novo, and laboratory-developed tests remain under CLIA.
- In the EU, instruments are class A, leukemia panels class C by analogy, legacy class C deadlines passed in 2026, and in-house panels face a no-equivalent-device test from 26 May 2028.
- In India, the flow cytometry analyzer is class A, reagent classes are not listed, and new class B to D devices need a three-batch performance evaluation.
- One evidence core of intended use, precision, sensitivity, agreement, stability and interpretation answers all three regulators.
+ Part VIII · Beyond the analyzer
The same principles, stretched.
The analyzer measures each particle once and lets it go. This last chapter looks at instruments that do more with the same principles: sorters that keep chosen cells alive, imaging cytometers that take a picture of each one, mass cytometers that count metal atoms instead of photons, compact counters built for clinics without a laboratory, and algorithms that read the data in place of people. It asks what each gains and gives up, and sets out what remains unsolved.
Sorting, imaging, mass and compact cytometers.
+ The questionThe analyzer measures and discards. What changes when a cytometer must also sort cells, photograph them, weigh their labels, fit in a clinic, or be read by an algorithm?
Sorting: a decision before the drop breaks
A droplet sorter measures each cell as an analyzer does, decides within a fraction of a millisecond whether to keep it, and then acts on the stream. The jet leaving the nozzle is shaken by a vibrating crystal and breaks into drops at a regular rate; a teaching guide on sorting gives the optimum spacing as about 4.5 times the jet's diameter. With a 70-micrometer nozzle at 60 pounds per square inch, most sorters produce 90,000 to 100,000 drops a second and accept up to about 70,000 cells a second. When a wanted cell reaches the point where drops form, its drop is charged and deflected into a tube by an electric field. The drop delay, the time from the laser to that point, must be set exactly; if it is wrong, the charged drop is not the one carrying the cell.
Because cells arrive at random (Chapter 6), some drops hold two. The sorter applies a mask: a purity mask aborts a sort when an unwanted cell falls too close to a wanted one, giving purity at the cost of yield, while an enrichment mode keeps every candidate drop and accepts impurity. The arithmetic decides how fast a sort can run.
At 30,000 cells a second and 93,000 drops a second, each drop holds on average 0.32 cells. By Poisson statistics, about 72 % of drops are empty, 23 % hold one cell and 4 % hold two or more. At 70,000 cells and 100,000 drops a second, the average rises to 0.70 and the share with two or more to about 16 %. The guide on sorting tabulates the consequence for a target making up a fifth of the cells: with a one-drop envelope at 93,000 drops a second, 96 % of wanted cells are sorted at 5,000 cells a second and 77 % at 30,000. The occupancy figures are derived for this guide.
Sorting also changes the cells. A 2018 study found that nearly half of the metabolic features measured in sorted astrocytes changed by at least 1.5-fold and that reactive oxygen species rose by 50 %, while a 2020 study of human blood cells found no significant differences in viability and proliferation between sort conditions. The endpoint decides whether sorting harms the experiment. Sorters also make aerosols from samples that may be infectious, which is why biosafety rules built on the international cytometry society's 2014 standard require aerosol management and regular containment testing.
Pictures and ions
An imaging flow cytometer records a picture of each cell as it flows, combining, in the words of a 2007 review by staff of its first manufacturer, the statistical power of flow cytometry with the spatial resolution of microscopy. The cost is speed. One manufacturer states 5,000 objects a second at 20-fold magnification, 2,000 at 40-fold and 1,200 at 60-fold, against tens of thousands for an analyzer. Image-activated sorting closed the loop: a 2018 system imaged, classified with deep learning and sorted about 100 cells a second, and its 2020 successor about 2,000 with a sensitivity near 50 molecules of equivalent fluorochrome. In 2022 a manufacturer's developers, with an academic group, reported an instrument that sorted on fluorescence images at up to about 15,000 events a second.
Mass cytometry replaces dyes with metal isotopes. Each antibody carries a different metal, each cell is vaporized in an argon plasma, and a time-of-flight mass spectrometer counts the metal ions. A 2011 study whose authors included staff of the instrument's maker measured 34 parameters per cell at up to 1,000 cells a second, without forward or side scatter, and a 2015 primer reported that only about 30 % of the cells introduced were recorded, rising to 50 % in a later model. In mass cytometry the number in the file is a count of ions rather than of photons; the guide's invariant holds in its general form, because the number still measures a signal, not a particle.
Compact counters and the clinic
Some of cytometry's most important clinical uses are simple counts in places without a laboratory. By 2012 the World Health Organization had prequalified a point-of-care CD4 analyzer that reports results in under twenty minutes, according to its maker, and in 2014 it prequalified a second. A 2015 meta-analysis of the first, covering 22 studies and 11,803 paired measurements, found a mean bias of 23 cells per microliter below the laboratory reference at a median of 383, and a sensitivity of 93 % at a threshold of 350 cells per microliter. In 2020 the organization prequalified a CD4 test that is a strip read by eye, reporting only whether the count is above or below 200 cells per microliter. Point-of-care CD4 testing is a clinical role rather than an instrument class. The strip is not a cytometer at all, and neither analyzer is a flow cytometer, although the organization lists the first under that heading. Each labels cells with fluorescent antibodies in a disposable cartridge and counts them in camera images: the organization's public report on the first describes static image analysis, and the US clearance of the second describes an imaging cytometer whose sample is still while it is imaged.
At the other extreme, dedicated small-particle analyzers are built for the vesicle thread. One such instrument is described, in figures from its maker, as detecting biological particles best between 40 and 200 nanometers, the range that Chapter 19 found beyond conventional cytometers. Claims of this kind need the calibration and controls of Chapters 17 and 20 before their counts can be compared.
Algorithms as readers
Software increasingly does the reading. A 2026 study from one clinical laboratory described a pipeline for residual chronic lymphocytic leukemia that cleaned the data, clustered events, mapped them and classified them with a neural network. It agreed closely with expert review in 227 research samples and 192 validation samples, detected every case at or above 0.01 % and cut analysis time from 9.0 to 0.9 minutes per case. The study did not state a regulatory status, and in October 2026 this guide found no flow cytometry analysis software on the FDA's list of AI-enabled devices, though the list was only partly searchable. Other methods go further: ghost cytometry, published in 2018 by a group whose members founded a company to commercialize it, classifies cells from structured light without forming an image, at more than 10,000 cells a second.
An algorithm inherits everything upstream. It reads compensated, transformed, gated data from an instrument at its settings (Chapters 10 to 13), so it is part of the system to be specified, validated and regulated (Chapters 22 and 23), not an independent judge.
What remains unsolved
Several problems run through every instrument in this chapter. Reference materials for cells and for vesicles, with values traceable to physical units, remain scarce, so most results are standardized rather than calibrated (Chapter 17). Sensitivity below 100 nanometers remains out of reach for nearly all cytometers, and most vesicles lie there (Chapter 19). The file still cannot hold several settings that decide whether its numbers are valid (Chapter 18). And comparability of intensities between laboratories stalls near 30 % even under full standardization (Chapter 16).
The central question asked how a machine that pushes cells single file past a laser and counts flashes of light can tell one kind of cell from another, how small a particle it can see, and when its count can be trusted as a test result. Patterns of light from labeled proteins answer the first part. Physics answers the second, at roughly 100 to 250 nanometers for well-calibrated instruments today. The third depends on the whole system, specified, calibrated, controlled, validated and recorded, which is where most of the remaining work lies.
A sorter, an imaging cytometer, a mass cytometer and a compact counter each measure differently from the analyzer they resemble. Their limits of detection, their precision and the populations they report must be established on their own, not carried over from the analyzer whose results they are compared with.
Before choosing beyond an analyzer, name what the result needs: living cells for further work, morphology, more parameters than fluorescence allows, a count without a laboratory, or faster reading. Each instrument class buys one of these and gives up speed, recovery, scatter or breadth, and the right choice follows from the intended use, as every other decision in this guide does.
- Droplet sorters run at 90,000 to 100,000 drops a second with a 70-micrometer nozzle, trading purity against yield, and sorting alters metabolism more than viability.
- Imaging cytometers trade speed for pictures, and mass cytometers count metal ions for more parameters at lower rates and without scatter.
- Point-of-care CD4 counting is a clinical role filled by compact analyzers and even strips, while small-particle analyzers target the vesicle range.
- Algorithms can read data faster than experts but inherit every upstream setting; reference materials, sub-100-nanometer sensitivity, the file and comparability remain unsolved.
Lessons.
The chapters reduce to a short set of working rules for anyone who designs, verifies, buys, regulates or writes about flow cytometers and the tests run on them. Each one traces back to a mechanism explained earlier.
- Treat every number as a measurement of light, not of a particle. An event is a pulse that crossed a threshold: it may be one cell, two cells, debris or a swarm of vesicles, and a particle that stayed below the threshold leaves no record at all (Chapters 1, 6, 9 and 19).
- Specify detection channel by channel, as Q and B. Background, not detector efficiency, usually sets the dimmest resolvable signal, and single sensitivity figures on datasheets cannot be compared between manufacturers (Chapters 8 and 15).
- Never read forward scatter as size, or bead size as vesicle size. Scatter depends on refractive index and collection angle as much as on diameter; a 100-nanometer polystyrene bead scatters like a vesicle of about 175 nanometers (Chapters 4, 15 and 17).
- Fix the flow rate, the trigger and the threshold, and record them. A faster run widens the core and raises coincidence, and the threshold decides which particles exist in the data (Chapters 6 and 9).
- Expect compensation to restore averages, never precision. Spillover spread grows with the square root of the spilled light, so dim markers belong in detectors that bright dyes on the same cells do not spread into (Chapters 10 and 14).
- Set each positive boundary on a control that contains everything except the specific binding. Fluorescence-minus-one controls and cells known to lack the marker do that; unstained and isotype samples do not (Chapter 11).
- Keep the original file, and record what it cannot hold. Transforms, gates, matrices, denominators, laser delays and software versions decide what a number means, and the standard file keeps only some of them (Chapters 12, 13 and 18).
- Count donors as replicates, and count enough events of the population reported. A million events from one sample are one observation, and a rare population's precision depends on its own events: one hundred give about 10 % (Chapters 12 and 22).
- Own the interfaces, and check them in every run. Stream speed, laser delays, sample preparation and settings fail between subsystems, and their failures look like biology rather than like errors (Chapters 7, 13 and 16).
- Standardize to make instruments alike; calibrate to make numbers mean something. Shared targets made beads agree within 5.5 % and cells within about 30 %; calibrated units carry the assumptions of their models (Chapters 16 and 17).
- Prove that each vesicle event is one particle, and a vesicle. A dilution series, buffer and reagent controls and a detergent step come before any count, and every count needs its size window, brightness threshold and measured volume (Chapters 19 and 20).
- Validate for the intended use, and report the cells acquired with every rare-event result. A limit of 20 abnormal cells in ten million applies only to a sample that reached ten million cells (Chapters 21 and 22).
- Build one evidence core and present it to each regulator. The instrument, the reagents and a laboratory's own panel are classed differently in the US, the EU and India, and a research-use label never makes clinical use lawful by itself (Chapter 23).
Glossary.
Terms are defined as they are used in this guide.
- 510(k)
- A US premarket notification showing that a device is substantially equivalent to a legally marketed device; a cleared 510(k) allows marketing.
- Abort (electronic)
- An event discarded by the electronics because a second pulse arrived while the first was still being processed.
- Analog-to-digital converter
- The circuit that samples a detector's voltage many millions of times a second and turns it into numbers.
- Analyte-specific reagent
- In the US, a single reagent such as an antibody sold for laboratories to build their own tests, without claims of performance.
- Antibodies bound per cell
- A calibrated unit giving the number of labeled antibodies bound to a cell, based on a biological standard such as CD4 on T cells.
- Antigen density
- The number of copies of a protein on a cell, which sets how many antibodies, and how much dye, a cell can carry.
- Area, height and width
- The three numbers taken from each pulse: the integrated signal, its peak and its duration.
- Autofluorescence
- The faint fluorescence of unlabeled cells, from molecules such as flavins inside them.
- Avalanche photodiode
- A semiconductor detector with high efficiency in the red and near-infrared and modest internal gain.
- Backbone markers
- Markers placed on the same dyes in every tube of a panel so that the population of interest lands in the same place in each tube.
- Bandpass filter
- A filter that passes a stated band of wavelengths; 530/30 passes 515 to 545 nanometers.
- Calibration
- Relating an instrument's arbitrary units to physical units, such as molecules, square nanometers or volume, through a reference material and a model.
- Carryover
- The fraction of one sample that appears in the next sample run on the same instrument.
- CD (cluster of differentiation)
- The international numbering of cell-surface proteins recognized by antibodies, such as CD4 or CD19.
- CLIA
- The US Clinical Laboratory Improvement Amendments, which regulate clinical laboratories, including the tests they develop themselves.
- Clonality
- Descent of a population of cells from one cell; for B cells, shown by a single light chain, kappa or lambda.
- Clustering
- Grouping events by similarity with an algorithm instead of hand-drawn gates.
- Coefficient of variation (CV)
- The standard deviation of a population divided by its mean, as a percentage; a robust CV uses percentiles instead.
- Coincidence
- Two or more particles in the laser beam within one processing window, recorded as one event or aborted.
- Compensation
- Subtracting each dye's spillover from the other detectors, using values measured on single-stain controls.
- Control
- A sample, or a population within a sample, that supplies the comparison on which a positive boundary is set.
- Core stream
- The thin thread of sample at the center of the sheath flow that carries particles in single file through the beam.
- De Novo
- The US route for a new type of low-to-moderate-risk device with no predicate; a granted request creates a new classification.
- Dichroic mirror
- A mirror that reflects light on one side of a stated wavelength and transmits the rest, dividing collected light by color.
- Dimmest resolvable signal
- The smallest signal whose population still separates from the negatives by two standard deviations each; 4 × (1 ÷ Q + background spread).
- Doublet
- Two cells passing together and recorded as one event; often removed by comparing pulse area with height.
- Drop delay
- In a sorter, the time between a cell crossing the laser and its arrival at the point where the jet breaks into drops.
- Dynamic range
- The span from the dimmest to the brightest signal measured within a stated deviation from linearity.
- ERF
- Equivalent reference fluorophore: the number of molecules of a stated reference dye in solution giving the same signal as one bead.
- Event
- A pulse that crossed the threshold and was recorded as one row in the data file.
- Extracellular vesicle
- A particle released from cells, delimited by a lipid bilayer, that cannot replicate on its own; most are smaller than 200 nanometers.
- Fc receptor
- A protein on some white cells that holds antibodies by their constant tail, causing binding unrelated to the target.
- FCS
- Flow Cytometry Standard, the file format for cytometry data; version 3.1 stores keywords and list-mode data, always uncompensated.
- Fit for purpose
- Validation tailored to the intended use of the data, as set out in CLSI H62.
- Flow cell
- The quartz cuvette in which the laser meets the stream in most analyzers.
- Fluorescence-minus-one (FMO) control
- A sample stained with every antibody of a panel except one, used to set that marker's positive boundary.
- Forward scatter
- Light scattered by a particle at small angles around the laser's direction.
- Full-spectrum (spectral) cytometer
- An instrument that records each particle's emission across many detectors and separates dyes by unmixing.
- Gate
- A boundary drawn on a plot that selects events for the next step of analysis.
- Gating hierarchy
- The sequence of gates, from time and doublets through viability and lineage to subsets, behind every reported percentage.
- Gating-ML
- A standard for describing gates and transforms exactly, outside the FCS file.
- Hydrodynamic focusing
- Narrowing the sample into a thin core inside a faster sheath flow so that particles cross the beam one at a time.
- Imaging flow cytometer
- An instrument that records a picture of each cell as it flows.
- In-house device
- In the EU, a test made and used within a health institution under Article 5(5) of the IVDR.
- Interface
- A boundary between two parts of a system, with parameters both sides must agree on.
- Interrogation point
- The fixed point at which the laser crosses the stream and each particle is measured.
- Intrinsic brightness
- A dye's molar absorption coefficient multiplied by its quantum yield.
- Isotype control
- An antibody of the same class and dye as the test antibody but against an absent target; rarely a good match for its background.
- IVDR
- Regulation (EU) 2017/746 on in vitro diagnostic medical devices, applicable since 26 May 2022.
- Jet-in-air
- A sorter design in which the laser meets the free stream after it leaves the nozzle.
- Laboratory-developed test
- A test made and used within one laboratory; in the US regulated under CLIA.
- Laser delay
- The interval the electronics wait between a particle's pulse at one laser and its pulse at the next.
- Limit of detection
- The smallest amount an assay detects with stated reliability; in rare-event cytometry set by convention as 20 events in the final gate, a cluster-size rule rather than a statistical limit.
- Linearity
- The proportionality of an instrument's output to its input over a stated range.
- List mode
- Data stored as one row per event and one column per parameter.
- Logicle
- A display transform that is linear near zero and logarithmic at high values, so compensated data near and below zero can be shown.
- Lower limit of quantitation
- The smallest amount measured with acceptable precision; in rare-event cytometry often set as 50 events.
- Mass cytometry
- Cytometry in which antibodies carry metal isotopes that are counted by a mass spectrometer after the cell is vaporized.
- MESF
- Molecules of equivalent soluble fluorochrome: the number of free dye molecules in solution giving the same fluorescence as a particle.
- Mie theory
- The physics of light scattering by spheres of any size, used to relate scatter to diameter and refractive index.
- MIFlowCyt
- The minimum information standard for reporting a flow cytometry experiment, published in 2008.
- MIFlowCyt-EV
- The extension of MIFlowCyt for vesicle measurements, published in 2020.
- Molar absorption coefficient
- How strongly a molecule absorbs light of a given wavelength; also called the extinction coefficient.
- Numerical aperture
- A measure of the width of the cone of light a lens accepts, which sets the fraction of emitted light collected.
- Photomultiplier tube
- A detector in which one photoelectron is multiplied about ten million times by a chain of electrodes.
- Pseudoreplication
- Treating measurements that are not independent, such as events from one sample, as replicates.
- Pulse
- The rise and fall of a detector's signal as a particle crosses the beam.
- Q and B
- A channel's detection efficiency, in photoelectrons per unit of dye, and its background; together they set the dimmest resolvable signal.
- Qualitative, quasi-quantitative and quantitative assays
- Assays that report a category, a number without a calibration standard, or an absolute value traceable to standards.
- Quantum yield
- The fraction of absorbed photons that a molecule emits as fluorescence.
- Reference material
- A material whose value is known with a stated uncertainty, used for calibration.
- Refractive index
- How much a material slows light compared with a vacuum; scatter depends on the contrast between particle and medium.
- Residual disease (MRD)
- Leukemic or myeloma cells remaining after treatment, measured by flow cytometry at frequencies down to about 2 in a million.
- Sheath
- The clean fluid that surrounds and narrows the sample stream.
- Side scatter
- Light scattered by a particle at wide angles, collected by the same lens as fluorescence.
- Signature
- The pattern of one dye's light across all the detectors of a spectral instrument.
- Silicon photomultiplier
- An array of tiny photodiodes, each firing on a single photon, with high gain.
- Single-platform count
- An absolute count made in one tube on one instrument relative to a known number of counting beads.
- Single-stain control
- A sample stained with one dye of a panel, used to measure that dye's spillover or spectral signature.
- Spillover
- The fraction of a dye's signal in its own detector that appears in another dye's detector.
- Spillover spread
- The widening of compensated populations caused by the counting noise of spilled light.
- Stain index
- The difference between positive and negative medians divided by twice the spread of the negatives.
- Standardization
- Making instruments read the same reference material alike, without assigning physical units.
- Stokes shift
- The difference between the wavelengths at which a dye absorbs and emits most strongly.
- Swarm detection
- Many particles below the detection limit, illuminated together, recorded as one event.
- Tandem dye
- Two dyes joined so that the first passes its energy to the second, which emits at a longer wavelength.
- Target values
- Bead intensities that every instrument in a standardized network is adjusted to reach.
- Threshold
- The level a chosen detector's signal must exceed for the electronics to record an event.
- Titration
- Staining with a series of antibody amounts to find the one that best separates positive from negative cells.
- Trigger
- The detector whose signal is compared with the threshold to decide that a particle has arrived.
- Unmixing
- Finding the amount of each dye whose combined signatures best fit a particle's readings across many detectors.
- Validation
- Showing that an assay serves its intended use, with precision, sensitivity, agreement and stability established.
- Verification
- Showing that an instrument or design meets its specification.
- Viability dye
- A dye excluded by intact membranes that marks dead cells so they can be removed from analysis.
Sources.
Sources are listed by the chapter in which they are first used. Figures from manufacturers, and from studies written by manufacturers' staff, are identified as such in the text. Regulatory status and standards editions are as of October 2026.
- Ch. 1 — Wikipedia. Flow cytometry (History section), accessed 6 Oct 2026 (web page; secondary). en.wikipedia.org/wiki/Flow_cytometry
- Ch. 1 — BD Biosciences. BD FACSCanto II technical specifications, X23-1091801EUROPE (2011). Manufacturer.
- Ch. 1 — Spidlen J., Moore W., Parks D., et al. Data File Standard for Flow Cytometry, Version FCS 3.1 — Normative Reference. ISAC (2008–2009); and Data file standard for flow cytometry, version FCS 3.1. Cytometry A 77, 97–100 (2010). doi:10.1002/cyto.a.20825
- Ch. 1 — Lee J.A., Spidlen J., Boyce K., et al. MIFlowCyt: the minimum information about a flow cytometry experiment. Cytometry A 73A, 926–930 (2008). doi:10.1002/cyto.a.20623; and MIFlowCyt 1.0 standard document, ISAC Recommendation (21 Feb 2008).
- Ch. 1 — Arnold L.W., Lannigan J. Practical issues in high-speed cell sorting. Curr Protoc Cytom 51, 1.24.1–1.24.30 (2010). doi:10.1002/0471142956.cy0124s51
- Ch. 1 — flowWorkspaceData (Bioconductor experiment data package): FlowJo workspaces with FCS files of CytoTrol freeze-dried control cells, including CytoTrol_1; gate counts, transform and software comparisons rerun in R 4.6.1 (1 Sep 2026).
- Ch. 1 — van der Pol E., van Gemert M.J.C., Sturk A., Nieuwland R., van Leeuwen T.G. Single vs. swarm detection of microparticles and exosomes by flow cytometry. J Thromb Haemost 10, 919–930 (2012). doi:10.1111/j.1538-7836.2012.04683.x
- Ch. 1 — Welsh J.A., Goberdhan D.C.I., O'Driscoll L., et al. Minimal information for studies of extracellular vesicles (MISEV2023): from basic to advanced approaches. J Extracell Vesicles 13, e12404 (2024). doi:10.1002/jev2.12404; definitions as reproduced in ISEV, MISEV2023 slide set (web document). isev.org
- Ch. 2 — Coulter W.H. Means for counting particles suspended in a fluid. US Patent 2,656,508 (filed 27 Aug 1949; granted 20 Oct 1953).
- Ch. 2 — Picot J., Guerin C.L., Le Van Kim C., Boulanger C.M. Flow cytometry: retrospective, fundamentals and recent instrumentation. Cytotechnology 64, 109–130 (2012). doi:10.1007/s10616-011-9415-0
- Ch. 2 — Crosland-Taylor P.J. A device for counting small particles suspended in a fluid through a tube. Nature 171, 37 (1953). doi:10.1038/171037b0 (not opened; Van Dilla et al. 1969 credit their laminar flow chamber to this design)
- Ch. 2 — Kamentsky L.A., Melamed M.R., Derman H. Spectrophotometer: new instrument for ultrarapid cell analysis. Science 150, 630–631 (1965).
- Ch. 2 — Fulwyler M.J. Electronic separation of biological cells by volume. Science 150, 910–911 (1965). doi:10.1126/science.150.3698.910
- Ch. 2 — Sweet R.G. High frequency recording with electrostatically deflected ink jets. Rev Sci Instrum 36, 131–136 (1965). doi:10.1063/1.1719502
- Ch. 2 — Dittrich W., Göhde W. Automatisches Meß- und Zählgerät für die Teilchen einer Dispersion. German patent DE1815352 (priority 18 Dec 1968; published 14 Jan 1971); and Impulsfluorometrie bei Einzelzellen in Suspensionen. Z Naturforsch (1969).
- Ch. 2 — Van Dilla M.A., Trujillo T.T., Mullaney P.F., Coulter J.R. Cell microfluorometry: a method for rapid fluorescence measurement. Science 163, 1213–1214 (1969). doi:10.1126/science.163.3872.1213 (Los Alamos Scientific Laboratory; argon-ion laser at 488 nm)
- Ch. 2 — Hulett H.R., Bonner W.A., Barrett J., Herzenberg L.A. Cell sorting: automated separation of mammalian cells as a function of intracellular fluorescence. Science 166, 747–749 (1969).
- Ch. 2 — Köhler G., Milstein C. Continuous cultures of fused cells secreting antibody of predefined specificity. Nature 256, 495–497 (1975). doi:10.1038/256495a0
- Ch. 2 — Bonner W.A., Hulett H.R., Sweet R.G., Herzenberg L.A. Fluorescence activated cell sorting. Rev Sci Instrum 43, 404–409 (1972). doi:10.1063/1.1685647
- Ch. 2 — Herzenberg L.A., Parks D., Sahaf B., et al. The history and future of the fluorescence activated cell sorter and flow cytometry: a view from Stanford. Clin Chem 48, 1819–1827 (2002).
- Ch. 2 — Perfetto S.P., Chattopadhyay P.K., Roederer M. Seventeen-colour flow cytometry: unravelling the immune system. Nat Rev Immunol 4, 648–655 (2004). doi:10.1038/nri1416
- Ch. 2 — Park L.M., Lannigan J., Jaimes M.C. OMIP-069: forty-color full spectrum flow cytometry panel for deep immunophenotyping of major cell subsets in human peripheral blood. Cytometry A 97, 1044–1051 (2020). doi:10.1002/cyto.a.24213 (two of three authors at Cytek Biosciences) [authors include manufacturer staff]
- Ch. 2 — Konecny A.J., Mage P.L., Tyznik A.J., Prlic M., Mair F. OMIP-102: 50-color phenotyping of the human immune system with in-depth assessment of T cells and dendritic cells. Cytometry A 105, 430–436 (2024). doi:10.1002/cyto.a.24841 (developed on a Sony ID7000 and a BD FACSDiscover S8; two authors are BD Biosciences employees, and BD supplied 11 antibodies) [authors include manufacturer staff]
- Ch. 2 — Nolan J.P. The evolution of spectral flow cytometry. Cytometry A 101, 812–817 (2022). doi:10.1002/cyto.a.24566 (read as reprinted in Wiley, Spectral Flow Cytometry Article Collection, 2024)
- Ch. 2 — Sony Biotechnology. Press release: launch of the SP6800 Spectral Analyzer, 17 May 2013; and announcement of a spectral analyzer, 19 June 2012. Manufacturer.
- Ch. 3 — IUPAC Gold Book. Stokes shift (S06031), from Glossary of terms used in photochemistry, Pure Appl Chem 68, 2223 (1996), p. 2278; and Kasha rule (K03370) (web pages). goldbook.iupac.org
- Ch. 3 — Brouwer A.M. Standards for photoluminescence quantum yield measurements in solution (IUPAC Technical Report). Pure Appl Chem 83, 2213–2228 (2011).
- Ch. 3 — Shaner N.C., Steinbach P.A., Tsien R.Y. A guide to choosing fluorescent proteins. Nat Methods 2, 905–909 (2005).
- Ch. 3 — Oi V.T., Glazer A.N., Stryer L. Fluorescent phycobiliprotein conjugates for analyses of cells and molecules. J Cell Biol 93, 981–986 (1982). doi:10.1083/jcb.93.3.981 (Table 1: absorption coefficients and quantum yields; the allophycocyanin value is per 33,000-dalton unit)
- Ch. 3 — Thermo Fisher Scientific. Molecular Probes Handbook, Table 6.2: Spectral data for B-PE, R-PE and APC (web page). Manufacturer.
- Ch. 3 — Maecker H., Trotter J. Selecting reagents for multicolor flow cytometry with BD LSR II and BD FACSCanto systems. Nat Methods application note (Dec 2008). Manufacturer.
- Ch. 3 — Cossarizza A., Chang H.-D., Radbruch A., et al. Guidelines for the use of flow cytometry and cell sorting in immunological studies (third edition). Eur J Immunol 51, 2708–3145 (2021). doi:10.1002/eji.202170126
- Ch. 3 — Doornbos R.M.P., de Grooth B.G., Greve J. Experimental and model investigations of bleaching and saturation of fluorescence in flow cytometry. Cytometry 29, 204–214 (1997).
- Ch. 4 — Tzur A., Moore J.K., Jorgensen P., Shapiro H.M., Kirschner M.W. Optimizing optical flow cytometry for cell volume-based sorting and analysis. PLoS One 6, e16053 (2011). doi:10.1371/journal.pone.0016053
- Ch. 4 — van der Pol E., Sturk A., van Leeuwen T., et al. Standardization of extracellular vesicle measurements by flow cytometry through vesicle diameter approximation. J Thromb Haemost 16, 1236–1245 (2018). doi:10.1111/jth.14009
- Ch. 4 — Liu P.Y., et al. Cell refractive index for cell biology and disease diagnosis: past, present and future. Lab Chip 16, 634–644 (2016). doi:10.1039/C5LC01445J
- Ch. 4 — de Rond L., et al. Refractive index to evaluate staining specificity of extracellular vesicles by flow cytometry. J Extracell Vesicles 8, 1643671 (2019).
- Ch. 4 — Johnsen K.B., Gudbergsson J.M., Andresen T.L., Simonsen J.B. What is the blood concentration of extracellular vesicles? Implications for the use of extracellular vesicles as blood-borne biomarkers of cancer. Biochim Biophys Acta Rev Cancer 1871, 109–116 (2019). doi:10.1016/j.bbcan.2018.11.006
- Ch. 4 — van der Pol E., Hoekstra A.G., Sturk A., et al. Optical and non-optical methods for detection and characterization of microparticles and exosomes. J Thromb Haemost 8, 2596–2607 (2010). doi:10.1111/j.1538-7836.2010.04074.x
- Ch. 4 — van der Pol E., Coumans F.A.W., Grootemaat A.E., et al. Particle size distribution of exosomes and microvesicles determined by transmission electron microscopy, flow cytometry, nanoparticle tracking analysis, and resistive pulse sensing. J Thromb Haemost 12, 1182–1192 (2014). doi:10.1111/jth.12602
- Ch. 4 — de Rond L., van der Pol E., Bloemen P.R., et al. A systematic approach to improve scatter sensitivity of a flow cytometer for detection of extracellular vesicles. Cytometry A 97A, 582–591 (2020). doi:10.1002/cyto.a.23974
- Ch. 4 — Welsh J.A., van der Pol E., Arkesteijn G.J.A., et al. MIFlowCyt-EV: a framework for standardized reporting of extracellular vesicle flow cytometry experiments. J Extracell Vesicles 9, 1713526 (2020). doi:10.1080/20013078.2020.1713526 (working group of ISEV, ISAC and ISTH; Table 1 and the light-scatter calibration section read)
- Ch. 5 — Tian L., Nelson A.R., Lowe T., et al. Standardization of flow cytometric detection of antigen expression. Cytometry B (2024). doi:10.1002/cyto.b.22155
- Ch. 5 — Chattopadhyay P.K., Gaylord B., Palmer A., et al. Brilliant Violet fluorophores: a new class of ultrabright fluorescent compounds for immunofluorescence experiments. Cytometry A 81A, 456–466 (2012). doi:10.1002/cyto.a.22043 (two authors employed by Sirigen, which developed the dyes, and one by BioLegend, which sells conjugates) [authors include manufacturer staff]
- Ch. 5 — BioLegend. Tandem dyes (technical web page). Manufacturer.
- Ch. 5 — Maecker H.T., Trotter J. Flow cytometry controls, instrument setup, and the determination of positivity. Cytometry A 69A, 1037–1042 (2006). doi:10.1002/cyto.a.20333 (BD Biosciences) [authors include manufacturer staff]
- Ch. 5 — Ferrer-Font L., Pellefigues C., Mayer J.U., et al. Panel design and optimization for high-dimensional immunophenotyping assays using spectral flow cytometry. Curr Protoc Cytom 92, e70 (2020). doi:10.1002/cpcy.70 (read as bioRxiv preprint 784884; Cytek Biosciences co-author) [authors include manufacturer staff]
- Ch. 5 — Thermo Fisher Scientific. Eliminate dead cells for increased accuracy. BioProbes 64 (April 2011). Manufacturer.
- Ch. 5 — Bettin B.A., Li B., Falkena K., et al. Calibration of flow cytometers enables reproducible measurements of extracellular vesicle concentrations and reference range establishment. J Extracell Vesicles 14, e70189 (2025). doi:10.1002/jev2.70189
- Ch. 6 — BD Biosciences. BD FACSCalibur technical specifications. Manufacturer.
- Ch. 6 — Parks D.R., Moore W.A., Brinkman R.R., et al. Methodology for evaluating and comparing flow cytometers: a multisite study of 23 instruments. Cytometry A 93, 1087–1091 (2018). doi:10.1002/cyto.a.23605 [authors include manufacturer and software-company staff]
- Ch. 6 — Kaduchak G., Goddard G., et al. Ultrasonic analyte concentration and application in flow cytometry. US Patent 7,340,957 B2 (priority 29 Jul 2004; Los Alamos National Laboratory, assignee now Triad National Security).
- Ch. 6 — Goddard G., Martin J.C., Graves S.W., Kaduchak G. Ultrasonic particle-concentration for sheathless focusing of particles for analysis in a flow cytometer. Cytometry A 69A, 66–74 (2006).
- Ch. 6 — Applied Biosystems. Applied Biosystems debuts industry's first acoustic flow cytometer (press release, 7 Dec 2009). Manufacturer.
- Ch. 6 — Thermo Fisher Scientific. The Attune NxT acoustic focusing cytometer, BioProbes 71; and Hornstein J., Attune blog (31 Oct 2012). Manufacturer.
- Ch. 7 — Telford W. Near infrared lasers in flow cytometry. Methods (2015). PMID 25814439
- Ch. 7 — BD Biosciences. BD FACSymphony A5 SE User's Guide. Manufacturer.
- Ch. 7 — Beckman Coulter. CytoFLEX LX specification sheet FLOW-2429SPEC03.17 (2017). Manufacturer.
- Ch. 7 — Cytek Biosciences. Cytek Aurora brochure N9-20001. Manufacturer.
- Ch. 7 — BD Biosciences. CS&T webinar (April 2008). Manufacturer.
- Ch. 7 — Whitehead Institute flow cytometry core. Flow post-it: laser delays (core facility note, Jan 2022; author unlisted). wi.mit.edu
- Ch. 8 — Hamamatsu Photonics. R3896/R12896 photomultiplier tube datasheet (Feb 2014). Manufacturer.
- Ch. 8 — Schiffman A., Finan E. APD vs PMT and beyond: flow cytometry light detectors. Thermo Fisher Scientific blog (30 Sep 2025). Manufacturer.
- Ch. 8 — Ladd B. SiPM vs PMT: a photon finish. University of Chicago Cytometry and Antibody Technology facility blog (22 Nov 2021).
- Ch. 8 — Parks D.R., Roederer M., Moore W.A. A new “Logicle” display method avoids deceptive effects of logarithmic scaling for low signals and compensated data. Cytometry A 69A, 541–551 (2006).
- Ch. 8 — BD Biosciences. Qr and Br, Application Note 23-10516-01 (June 2012). Manufacturer.
- Ch. 8 — Welsh J.A., Arkesteijn G.J.A., Bremer M., et al. A compendium of single extracellular vesicle flow cytometry. J Extracell Vesicles 12, e12299 (2023). doi:10.1002/jev2.12299
- Ch. 8 — Hoffman R.A., Wang L., Bigos M., Nolan J.P. NIST/ISAC standardization study: variability in assignment of intensity values to fluorescence standard beads and in cross calibration of standard beads to hard dyed beads. Cytometry A 81A, 785–796 (2012). doi:10.1002/cyto.a.22086
- Ch. 9 — BD Biosciences. Window extension, Technical Bulletin 23-11552-00 (Feb 2010). Manufacturer.
- Ch. 9 — Arraud N., Gounou C., Turpin D., Brisson A. Fluorescence triggering: a general strategy for enumerating and phenotyping extracellular vesicles by flow cytometry. Cytometry A 89, 184–195 (2016). doi:10.1002/cyto.a.22669
- Ch. 9 — exRNA consortium (ERCC). Completed MIFlowCyt-EV report, ERCC mAb database (17 Oct 2022) (web document). exrna.org
- Ch. 9 — Sony. ID7000 technical specifications. Manufacturer.
- Ch. 9 — Beckman Coulter. System for high dynamic range analysis in flow cytometry. US Patent 7,590,500 B2 (filed 28 Feb 2005). Manufacturer.
- Ch. 10 — Roederer M. Spectral compensation for flow cytometry: visualization artifacts, limitations, and caveats. Cytometry 45, 194–205 (2001). doi:10.1002/1097-0320(20011101)45:3<194::AID-CYTO1163>3.0.CO;2-C
- Ch. 10 — Shapiro H.M. Compensation: an introduction. Boston Users' Group for Cytometry (slides, 10 Sep 2003).
- Ch. 10 — Nguyen R., Perfetto S., Mahnke Y.D., Chattopadhyay P., Roederer M. Quantifying spillover spreading for comparing instrument performance and aiding in multicolor panel design. Cytometry A 83A, 306–315 (2013). doi:10.1002/cyto.a.22251
- Ch. 10 — Bhowmick D., van Diepen F., Pfauth A., Tissier R., Ratliff M.L. A gain and dynamic range independent index to quantify spillover spread to aid panel design in flow cytometry. Sci Rep (2021). doi:10.1038/s41598-021-99831-7
- Ch. 10 — Novo D. A comparison of spectral unmixing to conventional compensation for the calculation of fluorochrome abundances from flow cytometric data. Cytometry A 101, 885–891 (2022). doi:10.1002/cyto.a.24669 [author founded an analysis-software company]
- Ch. 10 — Mage P.L., Konecny A.J., Mair F. Measurement and prediction of unmixing-dependent spreading due to collinearity in spectral flow cytometry. Cytometry A 109, 436–455 (2026). doi:10.1002/cyto.a.70044 [first author is manufacturer staff]
- Ch. 10 — Novo D., Grégori G., Rajwa B. Generalized unmixing model for multispectral flow cytometry utilizing nonsquare compensation matrices. Cytometry A 83A, 508–520 (2013). doi:10.1002/cyto.a.22272 [first author founded an analysis-software company]
- Ch. 10 — Kalina T., Flores-Montero J., van der Velden V.H.J., et al. EuroFlow standardization of flow cytometer instrument settings and immunophenotyping protocols. Leukemia 26, 1986–2010 (2012). doi:10.1038/leu.2012.122
- Ch. 11 — Hulspas R., O'Gorman M., Wood B., Gratama J.-W., Sutherland R. Considerations for the control of background fluorescence in clinical flow cytometry. Cytometry B 76, 355–364 (2009). doi:10.1002/cyto.b.20485
- Ch. 11 — Keeney M., Gratama J.W., Chin-Yee I.H., Sutherland D.R. Isotype controls in the analysis of lymphocytes and CD34+ stem and progenitor cells by flow cytometry—time to let go! Cytometry 34, 280–283 (1998).
- Ch. 11 — Barnett D., Louzao R., Gambell P., et al. Validation of cell-based fluorescence assays: practice guidelines from the ICSH and ICCS – part IV – postanalytic considerations. Cytometry B 84B, 309–314 (2013). doi:10.1002/cyto.b.21107
- Ch. 11 — Hallek M., Cheson B.D., Catovsky D., et al. iwCLL guidelines for diagnosis, indications for treatment, response assessment, and supportive management of CLL. Blood 131, 2745–2760 (2018).
- Ch. 12 — diffcyt (Bioconductor package). transformData documentation (web page), citing Bendall S.C., et al., Science (2011), Suppl. Fig. S2. rdrr.io/bioc/diffcyt
- Ch. 12 — Finak G., Perez J.-M., Weng A., Gottardo R. Optimizing transformations for automated, high throughput analysis of flow cytometry data. BMC Bioinformatics 11, 546 (2010).
- Ch. 12 — Maecker H.T., Rinfret A., D'Souza P., et al. Standardization of cytokine flow cytometry assays. BMC Immunol 6, 13 (2005). doi:10.1186/1471-2172-6-13 (first and corresponding author and two others at BD Biosciences, which supplied the reagents) [authors include manufacturer staff]
- Ch. 12 — Finak G., Langweiler M., Jaimes M., et al. Standardizing flow cytometry immunophenotyping analysis from the Human ImmunoPhenotyping Consortium. Sci Rep 6, 20686 (2016). [authors include manufacturer staff]
- Ch. 12 — Aghaeepour N., Finak G., FlowCAP Consortium, DREAM Consortium, et al. Critical assessment of automated flow cytometry data analysis techniques. Nat Methods 10, 228–238 (2013). doi:10.1038/nmeth.2365
- Ch. 12 — Weber L.M., Robinson M.D. Comparison of clustering methods for high-dimensional single-cell flow and mass cytometry data. Cytometry A 89, 1084–1096 (2016). doi:10.1002/cyto.a.23030
- Ch. 12 — Van Gassen S., Callebaut B., Van Helden M.J., et al. FlowSOM: using self-organizing maps for visualization and interpretation of cytometry data. Cytometry A 87, 636–645 (2015). doi:10.1002/cyto.a.22625
- Ch. 12 — Kobak D., Berens P. The art of using t-SNE for single-cell transcriptomics. Nat Commun 10, 5416 (2019). doi:10.1038/s41467-019-13056-x
- Ch. 12 — Chari T., Pachter L. The specious art of single-cell genomics. PLoS Comput Biol 19, e1011288 (2023). doi:10.1371/journal.pcbi.1011288
- Ch. 12 — Hurlbert S.H. Pseudoreplication and the design of ecological field experiments. Ecol Monogr 54, 187–211 (1984).
- Ch. 13 — flowAI: automatic and interactive anomaly discerning tools for flow cytometry data (2016), PMID 27153628 (volume and pages not verified); and Monaco G., flowAI vignette, Bioconductor (3 Aug 2016).
- Ch. 14 — Wade C.G., Rhyne R.H. Jr., Woodruff W.H., Bloch D.P., Bartholomew J.C. Spectra of cells in flow cytometry using a vidicon detector. J Histochem Cytochem 27, 1049–1052 (1979). doi:10.1177/27.6.110874
- Ch. 14 — Grégori G., Patsekin V., Rajwa B., et al. Hyperspectral cytometry at the single-cell level using a 32-channel photodetector. Cytometry A 81A, 35–44 (2012). doi:10.1002/cyto.a.21120
- Ch. 14 — Daemen S., Adelaar T.I., Barlampas P., et al. Managing autofluorescence in spectral flow cytometry for macrophage identification in the liver. Eur J Immunol 56, e70217 (2026). doi:10.1002/eji.70217
- Ch. 14 — Jaimes M., Zhang Z., Zhou J., et al. Methods of forming multi-color fluorescence-based flow cytometry panel. US Patent Application US 2022/0082488 A1 (published 17 Mar 2022), granted as US 12,209,948 B2 (28 Jan 2025); assignee Cytek Biosciences. Manufacturer.
- Ch. 14 — Lucas, André, Snyder. All the colors, none of the drama: a practical primer on spectral flow cytometry in clinical settings. College of American Pathologists member article (2026; not peer-reviewed). cap.org
- Ch. 14 — Beckman Coulter Life Sciences. Unlocking insights: the vital role of unmixing algorithms in spectral flow cytometry (application note, undated). Manufacturer.
- Ch. 14 — Kalina T., Flores-Montero J., Lecrevisse Q., et al. Quality assessment program for EuroFlow protocols: summary results of four-year (2010–2013) quality assurance rounds. Cytometry A 87, 145–156 (2015). doi:10.1002/cyto.a.22581
- Ch. 14 — Brestoff J.R. Full spectrum flow cytometry in the clinical laboratory. Int J Lab Hematol 45 (Suppl 2), 44–49 (2023). doi:10.1111/ijlh.14098
- Ch. 14 — FDA. 510(k) database records K130373 (Navios Flow Cytometer System, decision 18 Sep 2013), K141468 (BD FACSCanto II, 27 Feb 2015) and K162897 (Navios EX, 23 Jun 2017).
- Ch. 14 — FDA. 510(k) K170974: BD FACSLyric, FACSuite Clinical, Multitest assays (decision 3 Jul 2017) and 510(k) summary (performance data submitted by the manufacturer).
- Ch. 14 — FDA. 510(k) K232600: DxFLEX Flow Cytometer; ClearLLab 10C Panels (decision 22 Nov 2023) and 510(k) summary (performance data submitted by the manufacturer).
- Ch. 15 — BD Biosciences. CS&T webinar (June 2012). Manufacturer.
- Ch. 15 — BD Biosciences. BD FACSDiscover A8 cell analyzer technical specifications, BD-133318 v6.0. Manufacturer.
- Ch. 15 — Thermo Fisher Scientific. Accurate carryover measurements, Attune NxT (application note COL02200 0716, July 2016). Manufacturer.
- Ch. 15 — Perfetto S.P., Ambrozak D., Nguyen R., Chattopadhyay P., Roederer M. Quality assurance for polychromatic flow cytometry. Nat Protoc 1, 1522–1530 (2006). doi:10.1038/nprot.2006.250
- Ch. 15 — Perfetto S.P., Ambrozak D., Nguyen R., Chattopadhyay P.K., Roederer M. Quality assurance for polychromatic flow cytometry using a suite of calibration beads. Nat Protoc 7, 2067–2079 (2012). doi:10.1038/nprot.2012.126
- Ch. 15 — IEC 61010-1:2010 (Ed. 3.0) with AMD1:2016 (IEC Webstore); IEC 61010-1/AMD2 ED3 project status, IEC project 107234 (as of Sept 2026); and SIQ, IEC 61010-1 Amendment 2: upcoming changes to safety requirements (3 Apr 2025).
- Ch. 15 — IEC 61010-2-101:2018 (Ed. 3.0). Particular requirements for in vitro diagnostic (IVD) medical equipment (IEC Webstore).
- Ch. 15 — IEC 61326-1:2020 (Ed. 3.0) and IEC 61326-2-6:2025 (Ed. 4.0, replacing the 2020 edition, withdrawn 18 Jun 2025). Electrical equipment for measurement, control and laboratory use — EMC requirements (IEC Webstore).
- Ch. 15 — IEC 60825-1:2014 (Ed. 3.0). Safety of laser products (IEC Webstore); Ed. 4 status from a third-party standards tracker (genorma.com).
- Ch. 15 — FDA. Laser products — conformance with IEC 60825-1 Ed. 3 and IEC 60601-2-22 Ed. 3.1 (Laser Notice No. 56). Federal Register (8 May 2019), FR Doc. 2019-09381.
- Ch. 16 — Yuan J. Methods for optimizing PMT voltages. SickKids-UHN flow cytometry facility (2010).
- Ch. 16 — Glier H., et al. Standardization of 8-color flow cytometry across different flow cytometer instruments: a feasibility study in clinical laboratories in Switzerland. J Immunol Methods 475, 112348 (2019). doi:10.1016/j.jim.2017.07.013
- Ch. 17 — Bettin B.A., Varga Z., Nieuwland R., van der Pol E. Standardization of extracellular vesicle concentration measurements by flow cytometry: the past, present, and future. J Thromb Haemost 21, 2032–2044 (2023). doi:10.1016/j.jtha.2023.04.042
- Ch. 17 — NIST. Certificate of Analysis, Standard Reference Material 1932, Fluorescein Solution (issued 1 Dec 2022). tsapps.nist.gov/srmext/certificates/1932.pdf
- Ch. 17 — Wang L., Gaigalas A.K. Development of multicolor flow cytometry calibration standards: assignment of equivalent reference fluorophores (ERF) unit. J Res Natl Inst Stand Technol 116, 671 (2011). doi:10.6028/jres.116.012
- Ch. 17 — Wang L., DeRose P., Gaigalas A.K. Assignment of the number of equivalent reference fluorophores to dyed microspheres. J Res Natl Inst Stand Technol 121, 269 (2016). doi:10.6028/jres.121.012
- Ch. 17 — Wang L., Gaigalas A., Marti G., Abbasi F., Hoffman R. Toward quantitative fluorescence measurements with multicolor flow cytometry. Cytometry A 73A, 279–288 (2008). doi:10.1002/cyto.a.20507
- Ch. 17 — Wang L., Gaigalas A.K., DeRose P.C. A model for the binding of fluorescently labeled anti-human CD4 monoclonal antibodies to CD4 receptors on human lymphocytes. J Res Natl Inst Stand Technol 123, 123022 (2018). doi:10.6028/jres.123.022
- Ch. 17 — Welsh J.A., Horak P., Wilkinson J.S., et al. FCMPASS software aids extracellular vesicle light scatter standardization. Cytometry A 97, 569–581 (2020). doi:10.1002/cyto.a.23782
- Ch. 18 — Spidlen J., Moore W., Parks D., et al. Data file standard for flow cytometry, version FCS 3.2. Cytometry A 99, 100–102 (2021). doi:10.1002/cyto.a.24225
- Ch. 18 — Spidlen J., Moore W., ISAC Data Standards Task Force, Brinkman R.R. ISAC's Gating-ML 2.0 data exchange standard for gating description. Cytometry A 87, 683–687 (2015). doi:10.1002/cyto.a.22690; and flowCore reference manual (Bioconductor), Gating-ML 2.0 transformation classes.
- Ch. 18 — ISAC. Data standards (web page, accessed 6 Oct 2026). isac-net.org/data-standards
- Ch. 18 — FlowRepository. Quick start guide, with notice of temporary suspension of new experiment creation (web page, accessed 6 Oct 2026). flowrepository.org
- Ch. 18 — 21 CFR Part 11 (eCFR, current as of 2 Oct 2026).
- Ch. 18 — 42 CFR 493.1105 (eCFR, current as of 2 Oct 2026).
- Ch. 18 — EuroFlow. External Quality Assessment Program (web page). euroflow.org/qa
- Ch. 19 — Bettin B.A. From beads to bedside: standardizing extracellular vesicle concentration measurements. PhD thesis, University of Amsterdam (2024). dare.uva.nl
- Ch. 20 — Bettin B.A., Varga Z., Nieuwland R., van der Pol E. Standardization of extracellular vesicle concentration measurements by flow cytometry: the past, present, and future. J Thromb Haemost 21, 2032–2044 (2023). doi:10.1016/j.jtha.2023.04.042 (source of the 2009–2010 workshop history)
- Ch. 20 — Cointe S., Judicone C., Robert S., et al. Standardization of microparticle enumeration across different flow cytometry platforms: results of a multicenter collaborative workshop. J Thromb Haemost 15, 187–193 (2017). doi:10.1111/jth.13514 (two authors employed by BioCytex, which supplied the beads) [authors include manufacturer staff]
- Ch. 20 — Welsh J.A., van der Pol E., Arkesteijn G.J.A., et al. MIFlowCyt-EV: a framework for standardized reporting of extracellular vesicle flow cytometry experiments. J Extracell Vesicles 9, 1713526 (2020). doi:10.1080/20013078.2020.1713526
- Ch. 20 — Bettin B.A., van der Pol E., Nieuwland R. Plasma extracellular vesicle test sample to standardize flow cytometry measurements. Res Pract Thromb Haemost 7, 100181 (2023). doi:10.1016/j.rpth.2023.100181
- Ch. 20 — Welsh J.A., van der Pol E., Bettin B.A., et al. Towards defining reference materials for measuring extracellular vesicle refractive index, epitope abundance, size and concentration. J Extracell Vesicles 9, 1816641 (2020). doi:10.1080/20013078.2020.1816641
- Ch. 21 — FDA. Product classification database, product code PWD (21 CFR 864.7010). accessdata.fda.gov
- Ch. 21 — Borowitz M.J., et al. Guidelines for the diagnosis and monitoring of paroxysmal nocturnal hemoglobinuria and related disorders by flow cytometry. Cytometry B 78, 211–230 (2010).
- Ch. 21 — van Dongen J.J.M., et al. EuroFlow-based flowcytometric diagnostic screening and classification of primary immunodeficiencies of the lymphoid system. Front Immunol 10, 1271 (2019). doi:10.3389/fimmu.2019.01271
- Ch. 21 — Davis B.H., Holden J.T., Bene M.C., et al. 2006 Bethesda International Consensus recommendations on the flow cytometric immunophenotypic analysis of hematolymphoid neoplasia: medical indications. Cytometry B 72 (Suppl 1), S5–S13 (2007).
- Ch. 21 — Nakashima M.O., Kelley J.T. Peripheral blood flow cytometry testing (educational module), v1.0. College of American Pathologists (27 Aug 2025).
- Ch. 21 — van Dongen J.J.M., Lhermitte L., Böttcher S., et al. EuroFlow antibody panels for standardized n-dimensional flow cytometric immunophenotyping of normal, reactive and malignant leukocytes. Leukemia 26, 1908–1975 (2012). doi:10.1038/leu.2012.120
- Ch. 21 — Rawstron A.C., et al. Reproducible diagnosis of chronic lymphocytic leukemia by flow cytometry: an ERIC & ESCCA harmonisation project. Cytometry B 94, 121–128 (2018). doi:10.1002/cyto.b.21595
- Ch. 21 — Mandy F.F., Nicholson J.K.A., McDougal J.S. Guidelines for performing single-platform absolute CD4+ T-cell determinations with CD45 gating for persons infected with human immunodeficiency virus. MMWR 52 (RR-2), 1–14 (2003).
- Ch. 21 — USP General Chapter <127> Flow cytometric enumeration of CD34+ cells (USP 40).
- Ch. 21 — Khoury J.D., et al. The 5th edition of the WHO Classification of Haematolymphoid Tumours: myeloid and histiocytic/dendritic neoplasms. Leukemia 36, 1703–1719 (2022). doi:10.1038/s41375-022-01613-1
- Ch. 21 — Arber D.A., et al. International Consensus Classification of myeloid neoplasms and acute leukemias. Blood 140, 1200–1228 (2022). doi:10.1182/blood.2022015850
- Ch. 21 — 21 CFR 864.7010, Flow cytometric test system for hematopoietic neoplasms (eCFR; 82 FR 61165, 27 Dec 2017; amended 90 FR 55981, 4 Dec 2025).
- Ch. 21 — Del Vecchio L., Brando B., Lanza F., et al. Recommended reporting format for flow cytometry diagnosis of acute leukemia. Haematologica 89, 594–598 (2004).
- Ch. 22 — Davis B.H., Wood B., Oldaker T., Barnett D. Validation of cell-based fluorescence assays: practice guidelines from the ICSH and ICCS – part I – rationale and aims. Cytometry B 84B, 282–285 (2013). doi:10.1002/cyto.b.21104
- Ch. 22 — CLSI. H62: Validation of assays performed by flow cytometry, 1st ed. (H62Ed1E, 27 Oct 2021); product page and sample pages.
- Ch. 22 — 42 CFR 493.1253, Establishment and verification of performance specifications (eCFR).
- Ch. 22 — Tanqri S., Vall H., Kaplan D., et al. Validation of cell-based fluorescence assays: practice guidelines from the ICSH and ICCS – part III – analytical issues. Cytometry B 84B, 291–308 (2013). doi:10.1002/cyto.b.21106
- Ch. 22 — Litwin V., Salkowitz-Bokal J., Steele P. Flow cytometry: new considerations for validating laboratory-developed tests. Clinical Laboratory News (1 Dec 2013).
- Ch. 22 — FDA. Evaluation of automatic class III designation for ClearLLab Reagents: decision summary, DEN160047; and De Novo database entry DEN160047 (received 3 Oct 2016; granted 29 Jun 2017) (performance data submitted by the manufacturer).
- Ch. 22 — Wood B., Jevremovic D., Béné M.C., et al. Validation of cell-based fluorescence assays: practice guidelines from the ICSH and ICCS – part V – assay performance criteria. Cytometry B 84, 315–323 (2013). doi:10.1002/cyto.b.21108 (abstract only)
- Ch. 22 — Gothot A. Quality assurance in flow cytometry (workshop slides, UCLL, 9 Feb 2024), summarizing ICSH/ICCS 2013 and CLSI H62.
- Ch. 22 — Flores-Montero J., Sanoja-Flores L., Paiva B., et al. Next Generation Flow for highly sensitive and standardized detection of minimal residual disease in multiple myeloma. Leukemia 31, 2094–2103 (2017). doi:10.1038/leu.2017.29
- Ch. 22 — Theunissen P., Mejstrikova E., Sedek L., et al. Standardized flow cytometry for highly sensitive MRD measurements in B-cell acute lymphoblastic leukemia. Blood 129, 347–357 (2017). doi:10.1182/blood-2016-07-726307
- Ch. 22 — Buccisano F., Palmieri R., Piciocchi A., et al. Clinical relevance of an objective flow cytometry approach based on limit of detection and limit of quantification for measurable residual disease assessment in acute myeloid leukemia. Haematologica 107, 2823–2833 (2022). doi:10.3324/haematol.2021.279777
- Ch. 22 — FDA. Hematologic Malignancies: Regulatory Considerations for Use of Minimal Residual Disease in Development of Drug and Biological Products for Treatment (final guidance, January 2020; docket FDA-2018-D-3090). fda.gov/media/134605/download
- Ch. 22 — OncLive. FDA's ODAC recognizes MRD as an accepted end point for accelerated approval in multiple myeloma (12 Apr 2024).
- Ch. 22 — FDA. De Novo database entry DEN170080, clonoSEQ (granted 28 Sep 2018; 21 CFR 866.6100).
- Ch. 22 — Davis B.H., Dasgupta A., Kussick S., Han J.Y., Estrellado A. Validation of cell-based fluorescence assays: practice guidelines from the ICSH and ICCS – part II – preanalytical issues. Cytometry B 84B, 286–290 (2013). doi:10.1002/cyto.b.21105
- Ch. 22 — ILAC. ISO 15189:2022 for medical labs published (news, 8 Dec 2022). ilac.org
- Ch. 22 — UK NEQAS for Leucocyte Immunophenotyping. EQA/PT programmes (web page, accessed 6 Oct 2026). ukneqasli.co.uk
- Ch. 23 — Regulation (EU) 2017/746 on in vitro diagnostic medical devices (IVDR), consolidated version of 10 Jan 2025, Art. 5(5), Art. 56 and Annex XIII. EUR-Lex 02017R0746-20250110
- Ch. 23 — 21 CFR 864.5220, Automated differential cell counter (eCFR; 67 FR 1607, 14 Jan 2002).
- Ch. 23 — FDA. Product classification database, product code OYE, Flow cytometric reagents and accessories. accessdata.fda.gov
- Ch. 23 — FDA. FDA allows marketing of test to aid in the detection of certain leukemias and lymphomas (news release, 29 Jun 2017).
- Ch. 23 — FDA. Medical Devices; Quality Management System Regulation Technical Amendments. 90 FR 55981 (4 Dec 2025), FR Doc. 2025-21955; effective 2 Feb 2026.
- Ch. 23 — FDA. 510(k) K183592: ClearLLab 10C Panels, Navios, Navios EX (decision 21 Mar 2019) and 510(k) summary (performance data submitted by the manufacturer).
- Ch. 23 — 21 CFR 864.4020, Analyte specific reagents (eCFR; 62 FR 62260, 21 Nov 1997).
- Ch. 23 — 21 CFR 809.30, Restrictions on the sale, distribution and use of analyte specific reagents (eCFR; 62 FR 62259, 21 Nov 1997).
- Ch. 23 — FDA. Commercially Distributed Analyte Specific Reagents (ASRs): Frequently Asked Questions (final guidance, 14 Sep 2007; docket FDA-2006-D-0357).
- Ch. 23 — FDA. Distribution of In Vitro Diagnostic Products Labeled for Research Use Only or Investigational Use Only (final guidance, November 2013; docket FDA-2011-D-0305).
- Ch. 23 — FDA. Medical Devices; Laboratory Developed Tests. Final rule, 89 FR 37286 (6 May 2024); American Clinical Laboratory Association v. FDA, No. 4:24-CV-479-SDJ (E.D. Tex., 31 Mar 2025), consolidated with AMP v. FDA; and Medical Devices; Laboratory Developed Tests; Implementation of Vacatur. Final rule, 90 FR 45134 (19 Sep 2025).
- Ch. 23 — 42 CFR 493.17, Test categorization (eCFR).
- Ch. 23 — MDCG 2020-16 rev.5. Guidance on classification rules for in vitro diagnostic medical devices under Regulation (EU) 2017/746 (September 2026).
- Ch. 23 — Regulation (EU) 2024/1860 of 13 Jun 2024 amending Regulations (EU) 2017/745 and (EU) 2017/746 (OJ 9 Jul 2024).
- Ch. 23 — MDCG 2023-1. Guidance on the health institution exemption under Article 5(5) of Regulation (EU) 2017/745 and Regulation (EU) 2017/746 (January 2023).
- Ch. 23 — Regulation (EU) 2022/112 of 25 Jan 2022 amending Regulation (EU) 2017/746 (OJ 28 Jan 2022).
- Ch. 23 — European Commission. COM(2025) 1023 final (16 Dec 2025); European Parliament Legislative Observatory, procedure 2025/0404(COD) (accessed 6 Oct 2026).
- Ch. 23 — Government of India. Medical Devices Rules, 2017, G.S.R. 78(E) (31 Jan 2017).
- Ch. 23 — CDSCO. List of in-vitro diagnostic medical devices under Medical Devices Rules, 2017, File IVD/Misc/196/2020 (updated 25 Oct 2023).
- Ch. 23 — DCG(I). Classification of medical devices and in vitro diagnostic medical devices under the provisions of the Medical Devices Rules, 2017 (1 Nov 2017).
- Ch. 23 — Lok Sabha. Unstarred Question No. 1669 (16 Dec 2022).
- Ch. 23 — CDSCO. In-vitro diagnostic (IVD) medical devices: frequently asked questions, CDSCO/IVD/FAQ/03/2022.
- Ch. 24 — Llufrio E.M., Wang L., Naser F.J., Patti G.J. Sorting cells alters their redox state and cellular metabolome. Redox Biol 16, 381–387 (2018). doi:10.1016/j.redox.2018.03.004
- Ch. 24 — Pfister G., Toor S.M., Sasidharan Nair V., Elkord E. An evaluation of sorter induced cell stress (SICS) on peripheral blood mononuclear cells (PBMCs) after different sort conditions — are your sorted cells getting SICS? J Immunol Methods 487, 112902 (2020). doi:10.1016/j.jim.2020.112902
- Ch. 24 — Holmes K.L., Fontes B., Hogarth P.J., et al. International Society for the Advancement of Cytometry cell sorter biosafety standards. Cytometry A 85, 434–453 (2014). doi:10.1002/cyto.a.22454
- Ch. 24 — NIH Division of Occupational Health and Safety. NIH Manual Chapter 3038: Installation and Operation of Cell Sorters (26 Sep 2017). policymanual.nih.gov/3038
- Ch. 24 — Stanford Environmental Health & Safety. Cell Sorting (FACS) Guidance (web document). ehs.stanford.edu
- Ch. 24 — Basiji D.A., Ortyn W.E., Liang L., Venkatachalam V., Morrissey P. Cellular image analysis and imaging by flow cytometry. Clin Lab Med 27, 653–670 (2007). doi:10.1016/j.cll.2007.05.008 [authors are manufacturer staff, Amnis]
- Ch. 24 — Cytek Biosciences. Amnis ImageStreamX Mk II (product page). Manufacturer.
- Ch. 24 — Nitta N., Sugimura T., Isozaki A., et al. Intelligent image-activated cell sorting. Cell 175, 266–276.e13 (2018). doi:10.1016/j.cell.2018.08.028; and Kyoto University press release (5 Sep 2018).
- Ch. 24 — Isozaki A., Mikami H., Tezuka H., et al. Intelligent image-activated cell sorting 2.0. Lab Chip 20, 2263–2273 (2020). doi:10.1039/d0lc00080a
- Ch. 24 — Schraivogel D., Kuhn T.M., Rauscher B., et al. High-speed fluorescence image-enabled cell sorting. Science 375, 315–320 (2022). doi:10.1126/science.abj3013 (developed at BD with EMBL) [authors include manufacturer staff]
- Ch. 24 — Bendall S.C., Simonds E.F., Qiu P., et al. Single-cell mass cytometry of differential immune and drug responses across a human hematopoietic continuum. Science 332, 687–696 (2011). [authors include manufacturer staff]
- Ch. 24 — Immune monitoring technology primer: flow and mass cytometry. J Immunother Cancer 3, 44 (2015). doi:10.1186/s40425-015-0085-x (authors not confirmed)
- Ch. 24 — Alere Inc. Pima CD4 Test receives WHO prequalification (press release, Dec 2011). Manufacturer.
- Ch. 24 — WHO Prequalification of In Vitro Diagnostics. Pima CD4 Test, product 0099-032-00: product page (prequalified 25 November 2011; assay format listed as flow cytometry) and public report PQDx 0099-032-00 v6.0 (9 May 2022; static image analysis with an on-board camera; gives the listing date as 20 November 2012). extranet.who.int/prequal
- Ch. 24 — BD. WHO issues prequalification of BD FACSPresto near-patient CD4 counter system (press release, 23 Sep 2014). Manufacturer.
- Ch. 24 — FDA. 510(k) K150815, BD FACSPresto System and CD4/Hb Cartridge: decision summary (substantially equivalent, 17 Dec 2015; product code PMG). accessdata.fda.gov
- Ch. 24 — Manabe Y.C., Wang Y., Elbireer A., Auerbach B., Castelnuovo B. Evaluation of portable point-of-care CD4 counter with high sensitivity for detecting patients eligible for antiretroviral therapy. PLoS One 7, e34319 (2012). doi:10.1371/journal.pone.0034319
- Ch. 24 — Scott L.E., Campbell J., Westerman L., et al. A meta-analysis of the performance of the Pima CD4 for point of care testing. BMC Med 13, 168 (2015). doi:10.1186/s12916-015-0396-2
- Ch. 24 — WHO Prequalification of In Vitro Diagnostics. Public report: VISITECT CD4 Advanced Disease, PQDx 0384-077-00, v5.0.
- Ch. 24 — Imperial College London, Hammersmith flow cytometry facility. NanoFCM Flow NanoAnalyzer (equipment page restating the manufacturer's figures). Manufacturer figures.
- Ch. 24 — Chiu A., et al. Artificial intelligence-enhancement of flow cytometry data accelerates the identification of measurable residual chronic lymphocytic leukemia. Leukemia 40, 1516–1526 (2026). doi:10.1038/s41375-026-02986-3
- Ch. 24 — FDA. Artificial Intelligence-Enabled Medical Devices (web page, accessed 6 Oct 2026).
- Ch. 24 — Ota S., Horisaki R., Kawamura Y., et al. Ghost cytometry. Science 360, 1246–1251 (2018). doi:10.1126/science.aan0096 (competing interests: three authors are founders and shareholders of ThinkCyte, which develops the method).
Take the PDF with you.
Reading online needs no sign-up. For the PDF edition, laid out for print and offline reading, tell us where to send it and we will email you.
Check your inbox.
The download link is on its way to your email. If it has not arrived in a few minutes, check your spam folder or write to info@unplex.tech.
Preview only — nothing was sent