+ Guide · PCR

Copy and count.

A first-principles guide to PCR: how heat, two primers and a heat-stable enzyme copy one stretch of DNA a billion times, how the instrument heats, reads and counts, how it is engineered as a system, and what a PCR test must prove before its result is trusted.

+ The central question

A PCR instrument does little more than heat and cool a tube on a schedule. So how does it find one sequence among billions, copy it a billion times, and tell you how much was there?

Starts at:school biology: DNA is a double helix of paired bases. No chemistry, statistics, optics or regulatory background is assumed.Ends at:reading an amplification curve, an instrument datasheet and a test's instructions for use critically, and knowing what an instrument and a test must show before a laboratory may report a result. Built around:one real test: the CDC's 2019-nCoV real-time RT-PCR panel, from its primers to its emergency authorization.

Most explanations of PCR fit on one slide. Heat separates the two strands of DNA, cooling lets two short primers stick, an enzyme copies each strand, and every cycle doubles the target, so thirty cycles make a billion copies. The slide is correct, and it leaves out almost everything an engineer or a founder needs. It does not say how two primers twenty letters long find their place among three billion, why a real test runs forty-five cycles rather than thirty, why the instrument's datasheet quotes temperature uniformity to a fraction of a degree, how a glow in a sealed plastic well becomes a number on a report, or why the same sample can give results eight cycles apart on two different tests.

This guide is built around one question instead: a PCR instrument does little more than heat and cool a tube on a schedule, so how does it find one sequence among billions, copy it a billion times, and tell you how much was there?

Here is the short answer, which the rest of the guide exists to justify. The instrument is not clever; the chemistry in the tube is. DNA's two strands are held together by forces that heat can overcome and cooling restores, while each strand's backbone is far stronger, so the same molecule can be pulled apart and copied again and again without breaking. Two primers, short pieces of designed DNA, bind only where their sequence matches, and only the stretch between two primers facing each other is copied over and over; that is how one sequence is found. A heat-stable enzyme builds each new strand, and because every copy becomes a template in the next cycle, the target roughly doubles each time; that is how it is copied a billion times. A dye or a probe glows in proportion to the copies, and the cycle at which the glow crosses a fixed line says how much target there was at the start; that is how it is counted. The instrument's real work is to make the temperature exact inside every well and to read the light at the same moment in every cycle, and for an authorized test the instrument, the reagents and the software are designed, verified and authorized together as one system.

One fact runs through every chapter. A cycle can at most double the number of target molecules, never more. Copying and counting are the same fact read in two directions. Read forward, thirty doublings turn one molecule into about a billion. Read backward, the number of cycles a sample needs to reach a fixed amount of product tells you how many doublings it was short of that amount, and therefore how much it started with.

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:

The mathematics stays at doubling, powers of two and of ten, percentages and simple ratios. Logarithms appear only as "how many doublings" or "how many tenfold steps", and the plates show them. 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. Template DNA from the sample, the copies made in the tube, primers and probes, RNA, and heat each have one color throughout the guide. The measured readout, the fluorescence and everything 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.

The color key

+ Part I · The molecule

What PCR copies, and why copying is needed.

Every PCR test begins with a problem of scale: the information is in the sample, but in far too few molecules for any instrument to read. These three chapters set out that gap in numbers, describe the physical structure of DNA that heat can exploit, and take stock of what a cell uses to copy its DNA and how little of that a tube needs. Nothing here assumes more than school biology.

+ The questionA swab already holds the virus's genetic material. Why can no instrument read it until it has been copied a billion times?

The information is there, in too few molecules

A nasal swab from someone with a respiratory infection carries the answer to a diagnostic question in a very specific form: a sequence of letters in the genome of the virus, written in a molecule a few nanometers wide. The sequence identifies the virus as surely as a fingerprint, and it is present in the sample from the moment the swab is taken. The difficulty is not that the information is missing. It is that it is carried by a number of molecules so small that no ordinary instrument can register it.

The scale is easiest to see by weighing a target. One base pair of double-stranded DNA has an average mass of about 650 daltons. A typical PCR target, a stretch of 200 base pairs, therefore weighs about 2 × 10⁻¹⁹ grams. Even an entire human genome, about 3.05 billion base pairs in the first complete sequence published in 2022, weighs only about 3.3 picograms, which is why a nanogram of human DNA, a quantity too small to see, contains only about 300 copies of any one gene.

A diagnostic test rarely works with even that much. The CDC's real-time test for SARS-CoV-2, the case this guide returns to, adds 5 microliters of purified extract to each 20-microliter reaction, and it was shown to detect about five copies of the viral RNA per reaction. Five molecules of a few hundred base pairs weigh about a millionth of a millionth of a microgram.

01 — From swab to reaction

Template DNARNAFocal detail

What an instrument needs to see

Compare that with what a detector needs. The classic way to see DNA after PCR is to run it through a gel and stain it with a fluorescent dye such as ethidium bromide. A visible band needs roughly half a nanogram to five nanograms of DNA, which for a 200-base-pair product is between about 2 × 10⁹ and 2 × 10¹⁰ copies. A real-time instrument, which watches fluorescence build up during the reaction, sets its detection line at a similar order of magnitude: working backward from published figures, the line sits at around 10¹⁰ to 10¹¹ copies of product per reaction.

So the gap between what is in the tube and what an instrument can register is about a billionfold, and often more. Five copies are not slightly below the detection limit; they are nine or ten orders of magnitude below it. No refinement of the detector closes a gap of that size at a price a clinical laboratory can pay, and concentrating the sample cannot close it either: putting more of the extract into the reaction gains a few-fold, nowhere near a billionfold, and no concentration step creates molecules that were never collected.

02 — The billionfold gap

Template DNACopiesReadoutFocal detail

The way across is to make more of the molecule itself. If the target can be copied, and the copies copied, the sequence that was present in five molecules can be made present in ten billion, and an ordinary fluorescence detector or a stained gel can then see it. Copying has a second advantage that no signal amplifier has: each copy carries the same sequence as the original, so the product can be checked for size, read by a probe or sequenced, and the identity of the target is preserved through the whole process.

Two demands on any copier

Copying sounds simple until two demands are written down.

The first is selectivity. A swab contains human DNA from shed cells, DNA and RNA from the bacteria that live in the nose, and possibly the virus. A human cell alone contains about six billion base pairs of DNA. The copier must reproduce one stretch of a few hundred letters belonging to the target and leave everything else alone, or the product will be dominated by whatever was most abundant, which is almost never the target.

The second is speed of multiplication. Suppose a process could make one new copy of each original molecule per round. Starting from five targets, thirty rounds would give 155 molecules: a linear increase that never approaches a billion. To cross nine orders of magnitude in a practical number of rounds, the copies must themselves become templates, so that the count multiplies rather than adds. A process in which each round doubles the count reaches a billion from one molecule in thirty rounds, and from five molecules in twenty-eight.

The paradox

Polymerase chain reaction, PCR, meets both demands, and it does so with an instrument that looks almost trivially simple. The instrument at its heart is a thermal cycler, often called a PCR machine: a metal block that holds the reaction tubes and changes their temperature on a programmed schedule. A real-time PCR instrument adds a light source and a detector so that it can watch the reaction as it runs. Nothing in either instrument touches the DNA. There is no mechanism that grips a molecule, no sensor that finds a sequence, no step in which the instrument decides what to copy.

That is the paradox this guide resolves. The selectivity comes from two short pieces of designed DNA, the primers, that bind only where their sequence matches (Chapter 5). The multiplication comes from an enzyme that copies DNA and a cycle of temperatures that lets every copy serve as a template in the next round (Chapters 4, 6 and 7). The counting comes from watching how many cycles it takes for the copies to become visible (Chapters 11 to 14). The instrument supplies exact temperatures and a precisely timed reading of light, and Chapters 9, 10 and 15 to 19 show that doing those two things well is a serious engineering problem.

The CDC test appears in almost every chapter. It is a well-documented example of a real-time RT-PCR test: its primers and probes were published, its instructions for use and its authorization letters are public, it ran on a named instrument, and its first weeks in February 2020 show what goes wrong when design, manufacture and quality control fall short under time pressure. Chapter 22 tells that story in full.

+ The questionWhat is it about DNA's structure that lets heat pull the two strands apart without breaking either one?

Two kinds of bond

DNA is two long chains wound around each other. Each chain, or strand, is a backbone of sugar and phosphate groups joined end to end by covalent bonds, and from every sugar a base sticks out sideways: adenine, thymine, guanine or cytosine, the A, T, G and C of the genetic code. The sequence of bases along one strand is the information. The two strands run in opposite directions, and they face each other base to base, so that the double helix looks like a ladder whose rungs are pairs of bases.

The two kinds of connection in this structure differ enormously in strength, and PCR depends on that difference. Along each strand, the links between one sugar and the next phosphate are covalent bonds, shared electrons that need a chemical reaction to break. Between the strands there are no covalent bonds at all. The strands are held together by many weak interactions, each of which ordinary thermal motion can disrupt.

03 — Two kinds of bond

Template DNAHeatFocal detail

Two of those weak interactions matter. The first is base pairing: A sits opposite T, held by two hydrogen bonds, and G sits opposite C, held by three. The shapes and hydrogen-bonding patterns of the bases fit only in these combinations, which is why one strand determines the sequence of the other. The second is stacking. The flat bases lie on top of one another along the helix like a pile of coins, and the attraction between neighboring pairs in the stack is what actually holds the duplex together most firmly.

Melting

Heat is random molecular motion. As a DNA solution is warmed, the motion grows until it overwhelms the weak forces between the strands, while the covalent backbone, far stronger, is untouched. The duplex comes apart into two intact single strands. This is called melting or denaturation, and its midpoint, the temperature at which half the DNA is single-stranded, is the melting temperature, written Tm.

Melting is not gradual across all temperatures. Because each base pair's stability depends on its neighbors in the stack, a region that starts to open makes the next region easier to open, and long DNA comes apart over a narrow range of a few degrees. The Tm itself depends on composition and on the solution. For long genomic DNA measured in a fairly salty buffer, 0.2 molar sodium, a classic study found that Tm rises in a straight line with G–C content: about 69.3 °C plus 0.41 °C for every percentage point of G and C.

The solution matters as much as the sequence. Positive ions shield the negative charges of the two backbones, which repel each other; more salt means a more stable duplex and a higher Tm. Raising sodium from the tens of millimolar to one molar can shift the Tm of a short DNA strand by as much as 20 °C, and magnesium, used in every PCR, has an even larger effect per unit of concentration. The concentration of the strands themselves also moves the Tm of short pieces of DNA by several degrees. A melting temperature is therefore a property of a sequence in a particular solution, not of the sequence alone.

04 — Melting curves

Template DNAHeatFocal detail

This is why PCR denatures at 94 to 98 °C. The temperature is chosen to be safely above the Tm of the target in the reaction buffer, so that every duplex opens in a few seconds, and low enough that the enzyme in the tube survives as long as possible, a balance Chapter 7 quantifies.

Cooling puts the strands back

Melting is reversible. When the solution cools, single strands collide, and when two complementary sequences meet, the base pairs form, the stack reassembles, and a duplex reappears. This re-pairing is called annealing or renaturation, and its speed depends on how often the right partners meet, which depends on their concentration.

PCR uses annealing deliberately. When the reaction cools after denaturation, two kinds of pairing compete. The long original strands can find each other again, and the primers, short designed pieces of DNA added in vast excess, can find their matching sites on the single strands. Chapter 3 shows that the primers outnumber the template strands by about two hundred million to one, so the primers almost always win.

Temperature also decides how perfect a match must be. A duplex with a mismatched base pair in it is less stable than a perfect one and melts at a lower temperature. At a carefully chosen annealing temperature, a primer that matches its target perfectly stays bound while one sitting on a near-match falls off. This is called stringency, and it is the first of PCR's defenses against copying the wrong sequence. Chapter 5 shows how it is used.

What heat costs

The backbone survives 95 °C, but not without a price. Heat slowly damages DNA chemically, most often by converting cytosine into uracil, which the enzyme then reads as thymine. A single-molecule sequencing study by two scientists at New England Biolabs, a polymerase supplier, measured this damage at about 1.4 × 10⁻⁶ changes per base per cycle. For a 100-base target over 40 cycles that is about 100 × 40 × 1.4 × 10⁻⁶ ≈ 0.006, so fewer than one molecule in a hundred carries a heat-induced change. It is irrelevant to detecting a virus and relevant to anyone using PCR to prepare DNA for sequencing.

The larger cost falls on the enzyme. Every second spent at denaturing temperatures shortens its working life, and the loss accelerates sharply above 95 °C. That is the reason the top temperature of the cycle is not simply set as high as possible, and the reason an instrument that overshoots its set point is not harmless.

+ The questionA cell uses more than a dozen proteins to copy its DNA. How few can a tube manage with?

How a cell copies its DNA

Every time a cell divides, it copies its entire genome, and it does so with a team of proteins working at a moving structure called the replication fork. A helicase travels along the double helix and unwinds it, separating the two strands. Proteins that bind single strands keep them from snapping back together. A primase lays down a short RNA primer on each exposed strand. A DNA polymerase then extends each primer, adding nucleotides one at a time, and a sliding clamp keeps the polymerase attached. Other enzymes relieve the twisting strain ahead of the fork, replace the RNA primers with DNA, and a ligase seals the gaps.

05 — The cell's copying team

Template DNACopiesRNAFocal detail

At the center of this machinery is the polymerase, and its rules are the rules PCR must live with. A DNA polymerase cannot start a new strand from nothing. It can only add to the free end of an existing strand that is paired to a template, the end chemists call the 3′ end. It reads the template one base at a time and adds the nucleotide that pairs with it: A opposite T, G opposite C. And it works in one direction only, extending each new strand from its 5′ end toward its 3′ end. The building blocks are deoxynucleoside triphosphates, dNTPs, one for each base; each one added releases a small pyrophosphate molecule, and the reaction needs magnesium ions to proceed.

What the tube keeps and what it replaces

PCR keeps the polymerase and replaces almost everything else with something simpler.

Heat replaces the helicase. Chapter 2 showed that raising the temperature above the melting point separates the strands without breaking them, so no unwinding enzyme is needed, and no proteins are needed to hold the strands apart: at 95 °C they stay apart on their own.

Synthetic DNA replaces the primase. Instead of an enzyme laying down RNA primers wherever replication starts, PCR adds two short pieces of DNA, about 18 to 30 bases long, designed to match the two ends of the stretch to be copied. These primers do two jobs at once. They satisfy the polymerase's need for a starting point, and, because they bind only where their sequence matches, they decide which stretch is copied.

The rest of the cell's team is unnecessary. A PCR target is short, typically tens to a few thousand bases, so there are no long strands to untangle and no fragments to join. The polymerase works on its own, and in most diagnostic PCR without the proofreading that cells use to correct errors, a choice Chapter 7 explains.

06 — What goes in the tube

Template DNAPrimers and probesFocal detail

What remains is a short list of ingredients. In a typical 50-microliter reaction with the enzyme Taq polymerase, a manufacturer's protocol calls for each of the four dNTPs at about 200 micromolar, magnesium chloride at 1.5 to 2 millimolar, each primer at 0.1 to 0.5 micromolar, about 1.25 units of polymerase, a buffer to hold the pH and salt, and the template: anything from a nanogram to a microgram of human DNA, or between 1 picogram and 10 nanograms of a small viral or plasmid genome.

Magnesium deserves a note because it is easy to upset. The polymerase is inactive without enough free magnesium, and too much makes it less accurate and encourages wrong products. The dNTPs bind magnesium, and so do chelators carried in with a sample, such as EDTA from a blood tube or citrate. What matters is the free magnesium left over, which is one reason a reaction that works with clean DNA can fail with a crude sample.

The primers win by numbers

When a reaction cools after denaturation, each single template strand can either pair with its original partner or bind a primer. The primers win because there are so many of them.

At those odds the template strands almost never find each other before a primer finds them. The primer excess also means that the primers are not used up in the course of a normal reaction, a point that matters when Chapter 8 asks why a reaction eventually stops.

The idea that made it a chain reaction

None of the ingredients was new in the early 1980s. Heating DNA to separate its strands had been studied for decades, and chemists had used DNA polymerases to extend short synthetic primers on templates before. What Kary Mullis, a chemist making synthetic DNA at the Cetus Corporation in California, conceived in 1983 according to most accounts was an arrangement: two primers, one on each strand, pointing toward each other across the target, and the steps of separating, priming and extending repeated over and over. In that arrangement the new strand made from one primer contains the binding site for the other primer, so every product becomes a template in the next round. The process feeds on its own output, which is why it is called a chain reaction, and why it multiplies rather than adds.

The first published application, by Randall Saiki and colleagues at Cetus, Mullis among them, in Science in 1985, amplified a 110-base-pair stretch of the human β-globin gene in order to tell the normal gene from the sickle-cell variant. It used a polymerase from the bacterium E. coli, ran 20 cycles with extension at 30 °C, and, because the enzyme was destroyed at the denaturing temperature, needed fresh enzyme added by hand at every cycle. Each cycle copied about 85 % of the target, and twenty of them amplified it about 220,000-fold. Mullis received half of the 1993 Nobel Prize in Chemistry for the invention; the other half went to Michael Smith for unrelated work.

+ Part II · The cycle

How heat, primers and an enzyme copy one stretch a billion times.

With the ingredients in the tube, the instrument has one control: temperature. These five chapters show how a repeating sequence of three temperatures runs three different processes, how two primers select one stretch of DNA, how doubling reaches a billion, why the enzyme had to survive repeated heating to 95 °C before the process could be automated, and why every reaction eventually stops.

+ The questionIf nothing can reach into the sealed tube, how can temperature alone run three different steps?

Three processes, three temperature dependences

Once a PCR tube is sealed, nothing more is added to it and nothing is taken out. The instrument cannot stir it, cannot add enzyme at the right moment, cannot remove a product. It can only change the temperature of the block the tube sits in. Yet a PCR needs three distinct things to happen, in order, in every cycle: the strands must separate, the primers must bind, and the polymerase must extend them.

The trick is that each of the three processes responds to temperature differently. Strand separation happens only above the melting temperature of the target, so it needs the tube hot, near 95 °C. Primer binding needs the opposite: the short primers form stable duplexes with their targets only below their own melting temperature, typically somewhere around 55 to 65 °C. Extension needs the polymerase to be active, and a polymerase from a heat-loving bacterium works fastest at around 75 to 80 °C, slower at lower temperatures and hardly at all near room temperature. Because the three processes are switched on and off by different temperature ranges, a single variable, changed in the right sequence, can run them one after another. Temperature is a switch with three positions.

07 — One knob, three steps

Template DNACopiesPrimers and probesHeatReadoutFocal detail

Denature

The cycle starts hot. At 94 to 98 °C the target duplexes open into single strands (Chapter 2). Once the liquid in the tube is at temperature, a short target separates within seconds, so the denaturing step of each cycle in a modern diagnostic test is brief: the CDC test holds 95 °C for 3 seconds. The first denaturation of a run is longer, typically 2 minutes, for two reasons. Long genomic DNA, and DNA still packed with proteins from the sample, takes longer to open completely the first time, and many enzymes are supplied in a blocked form that is released only by an initial period of heat, a feature Chapter 5 explains.

Every second spent hot costs enzyme activity, so the denaturing step is kept as short as the target allows. That is why the instrument's ability to bring the liquid, not just the block, to temperature quickly matters: a step of 3 seconds is meaningless if the sample needs 10 seconds to arrive.

Anneal

Next the instrument cools the tube to the annealing temperature, where the primers bind. The temperature is chosen relative to the primers' own melting temperature: a manufacturer's guideline for its Taq polymerase is about 5 °C below the lower of the two primers' melting temperatures. Primer binding is fast, because the primers are present at such enormous excess (Chapter 3), so a few seconds are enough.

The choice of temperature is a trade-off between yield and specificity. Too low, and primers stay bound to partial matches elsewhere in the sample, giving products that are not the target. Too high, and even the correct sites are occupied only part of the time, so each cycle copies fewer targets and the reaction's efficiency falls. Chapter 5 shows what goes wrong at the low end.

Extend

Then the polymerase extends each bound primer along its template. The rate of extension depends strongly on temperature, and the numbers explain the shape of modern PCR protocols. Measurements on Taq polymerase, published by its developers in 1988, found it added more than 60 nucleotides per second at 70 °C, about 24 per second at 55 °C, 1.5 per second at 37 °C and a quarter of a nucleotide per second at 22 °C. A later characterization of the purified enzyme, by scientists at Roche Molecular Systems, which sold it, measured about 150 nucleotides per second at 75 °C, close to its optimum.

08 — How fast Taq copies

CopiesHeatFocal detail

For long products, a separate extension step near 72 °C is still used: a three-step cycle of denature, anneal and extend. For the short targets of diagnostic tests, the anneal and extend steps are usually merged into one. The CDC test runs 45 cycles of just two steps: 95 °C for 3 seconds, then 55 °C for 30 seconds, during which the primers bind, the polymerase extends them, and the instrument reads the fluorescence. The two-step cycle saves the time of heating to a third temperature and cooling back down, about 45 times per run.

The program is the only instruction

Put together, a PCR run is a temperature program: an initial hold, then a fixed sequence of two or three temperatures repeated 25 to 45 times, sometimes followed by a slow heating to check the product (Chapter 12). Older endpoint protocols, written for slower instruments, held each step for 15 to 60 seconds and took one to two hours. A diagnostic real-time test on a fast instrument runs 40 to 45 short cycles in well under an hour, and research systems have run a complete PCR in less than a minute (Chapter 24).

Nothing in the tube tells the instrument how the reaction is going, and in an ordinary thermal cycler nothing in the instrument checks. The instrument follows its program, and the program assumes that each set point is actually reached by the liquid in every well and held for the stated time. Whether that assumption holds is the subject of Chapters 9 and 10.

+ The questionHow do two short primers pick out one target from billions of DNA letters, and what makes them pick the wrong one?

How rare is a twenty-letter word?

A primer twenty bases long is a word of twenty letters written in an alphabet of four. There are 4²⁰ such words, about 1.1 × 10¹². A human genome of about 3.05 billion base pairs, read along both strands, offers about 6.1 × 10⁹ places where a twenty-letter word could sit. If genomes were random strings of letters, the expected number of chance matches to any given twenty-letter primer would be 6.1 × 10⁹ ÷ 1.1 × 10¹² ≈ 0.006. A primer designed for a viral sequence would, on this arithmetic, be very unlikely to find a perfect match anywhere in human DNA.

That arithmetic is the reason primers work at all, and it is also too optimistic, for three reasons. Real genomes are not random: they contain repeated sequences, some, such as the Alu elements, present in about a million copies, and a primer that happens to resemble one of them has many places to bind. A primer does not need a perfect match to bind at a low enough temperature: a near-match with one or two mismatches can hold, especially if the mismatches are far from the end the polymerase extends. And the sample contains more than human DNA. So the specificity of PCR does not rest on any single primer being unique. It rests on the primers working as a pair, on the annealing temperature, and on the way the polymerase treats the end of a primer.

The pair is what selects

Recall the polymerase's rules from Chapter 3: it extends only from a primer, only in one direction. PCR uses two primers designed to bind to opposite strands, at the two ends of the target, each pointing toward the other. Each primer, extended, produces a new strand that runs across the target and through the binding site of the other primer.

09 — Two primers facing each other

Template DNACopiesPrimers and probesFocal detail

This geometry is the real filter. A primer that binds by chance at an unrelated site produces a single extended strand, but that strand is copied exponentially only if the second primer also binds to it, on the opposite strand, facing back, within the distance the polymerase can cover in one extension step. The chance that both primers find mismatched sites arranged like that, close enough together, is far smaller than the chance that either finds one alone. A stray binding event usually produces a few strands that grow by one per cycle and never become visible; the true target, bracketed by both primers, grows exponentially and swamps them.

Where the defined product first appears

The geometry also explains a detail that surprises most people when they first follow the cycles one by one. In the first cycle, each primer is extended along a long original strand, and the polymerase simply keeps going past the far end of the target until the step ends. The new strand has one defined end, where the primer started, and one ragged end. These are called long products. In the second cycle, the other primer binds to one of those long products and is extended toward its defined end. Because that template strand ends exactly where the first primer began, the copy stops there: the first strand of exactly the target's length. In the third cycle, two such exact strands pair into the first double-stranded product with both ends defined by the primers.

10 — Where the exact-length product appears

Template DNACopiesPrimers and probesFocal detail

From then on, the exact-length product doubles every cycle, while the long products increase by only two per cycle from each original duplex, because only the original strands make them.

Mispriming, primer-dimers and the hot start

Two kinds of wrong product are common, and both start when the temperature is too low for the primers to be choosy.

The first is mispriming: a primer binds to a site that only partly matches and is extended. Whether that matters depends on the end of the primer. The polymerase extends only from a 3′ end that is paired to the template, so a mismatch at or very near the 3′ end greatly reduces extension, while a mismatch near the other end of the primer is often tolerated. Primer designers use this deliberately, placing the most discriminating bases near the 3′ end.

The second is the primer-dimer: the two primers, present at enormous concentration, bind briefly to each other through a few complementary bases at their 3′ ends, and the polymerase extends both across each other. The result is a short double-stranded product that contains both primer sequences and therefore amplifies as efficiently as a real target, competing with it for the reaction's resources and, with some detection chemistries, producing signal of its own.

11 — A primer-dimer, and the hot start

CopiesPrimers and probesHeatFocal detail

Both errors are most likely before cycling starts. When a reaction is assembled at room temperature, the primers bind loosely to anything partly complementary, and a polymerase that keeps a little activity at room temperature extends those mistakes. A 1992 study by scientists at Cetus, which held the PCR patents, prevented any activity before the first heating step, using a layer of wax that kept the reagents apart until it melted, and found large improvements for targets below about a thousand copies, enough to amplify a single target molecule of HIV in a background of human DNA. Hot-start methods now do the same job in other ways: an antibody that binds and blocks the polymerase until heat destroys the antibody, a chemical modification of the enzyme reversed by a long initial heat step, or a short DNA molecule that binds the enzyme and lets go when heated. With the TaqPath mix, the two-minute hold at 95 °C at the start of the CDC test's cycling serves this purpose.

Designing for the right target

The design rules that follow from all this are conventions rather than laws, and suppliers state them a little differently. Primers are usually 18 to 30 bases long, with 40 to 60 % G and C, a G or C at the 3′ end to anchor it, and the two primers' melting temperatures within about 5 °C of each other so that one annealing temperature suits both. Runs of four or more identical bases are avoided, and so are sequences that can fold on themselves or pair with the other primer. For diagnostic tests the target itself is chosen with care: a region that is conserved across all known variants of the organism, so that no strain escapes detection, and absent from its relatives and from human DNA, so that nothing else is detected.

Most of these checks are done by computer before any reagent is made. The designer aligns the primers against every available sequence of the target organism to check that all are matched, and against the sequences of related organisms and human DNA to check that none is. During the COVID-19 pandemic, the FDA's template for emergency authorization requests asked developers to show by such alignments that their primers matched all published SARS-CoV-2 sequences, and to treat any similarity above 80 % between a primer and another organism as a possible cross-reaction to be tested in the laboratory.

+ The questionDoubling thirty times makes a billion. Why do real runs need closer to forty cycles?

Products become templates

In each cycle, every target strand in the tube can bind a primer and be copied, and every new strand contains the binding site for the other primer, so in the next cycle it is a template too. If every target is copied in every cycle, the count doubles: 1, 2, 4, 8, 16. Ten cycles give 2¹⁰ = 1,024, twenty give about a million, thirty give 2³⁰ ≈ 1.07 × 10⁹, and forty give about 1.1 × 10¹². This is the arithmetic behind the claim that PCR makes a billion copies in thirty cycles, and it is the forward reading of the fact stated at the start of this guide: a cycle can at most double the number of targets.

12 — Doubling on two scales

CopiesFocal detail

Doubling is the ceiling, and real reactions sit below it. The fraction of targets actually copied in a cycle is called the efficiency, E. At 100 % efficiency every target is copied and the count doubles; at 90 % efficiency, nine targets in ten are copied and the count is multiplied by 1.9. After n cycles, starting from N₀ targets, the reaction holds N₀ × (1 + E)ⁿ.

Efficiency compounds

A shortfall of a few percent per cycle sounds harmless. Compounded over thirty cycles it is not.

Doubling has a shape that is worth fixing in mind now, because Part IV depends on it. Plotted on an ordinary scale, an exponential curve hugs zero for most of its length and then shoots up, so the early cycles look like nothing is happening. Plotted on a logarithmic scale, where each step up the axis is a factor of ten, the same curve is a straight line. Perfect doubling climbs one tenfold step every 3.32 cycles, because 3.32 doublings make a factor of ten; at 90 % efficiency it climbs one step every 3.59 cycles. The steepness of that line is the efficiency, and a sample that starts with ten times more target runs along a parallel line, 3.32 cycles ahead.

Efficiency falls short of 100 % for several reasons that recur through this guide: some strands are not fully separated in a short denaturing step, some primer sites are not occupied at the annealing temperature, some extensions do not finish, and substances carried in from the sample slow the enzyme (Chapter 20). Because the count is a power of the per-cycle factor, any difference in efficiency between two reactions grows with every cycle. That matters for counting, as Chapter 13 shows: two samples with the same starting amount but different efficiencies reach the detection line at different cycles.

Where the forty comes from

The number of cycles a test needs is set by three things: how few molecules it must detect, how many molecules the detector needs to see, and the efficiency in between.

Chapter 1 put the detection line of a real-time instrument at around 10¹⁰ to 10¹¹ copies of product per reaction. From a single copy, perfect doubling would reach 10¹⁰ in about 33 cycles. Real reactions do not double perfectly, and they rarely start from exactly one copy. The 2025 revision of the MIQE guidelines, the field's reference for reporting real-time PCR, gives worked figures: at 90 % efficiency, 10 starting copies cross a typical detection line at about cycle 35 and 3 copies at about cycle 37; at 70 % efficiency the same inputs cross at about cycles 42 and 45. A run of 40 to 45 cycles therefore covers the handful of molecules a sensitive test must detect, even with an efficiency below perfect, and little is gained by going further. The CDC test runs 45 cycles and counts a target as detected only if it crosses the line before cycle 40.

The floor no instrument can lower

There is a limit below which no number of cycles helps, and it is set not by chemistry but by sampling. When a few microliters of extract are pipetted into a reaction, the number of target molecules that land in the tube is a matter of chance. If the extract contains, on average, one target molecule per 5 microliters, some reactions receive none, some one, some two or three.

The chance that a reaction receives zero molecules follows a simple rule, the Poisson distribution, which describes items scattered at random. With an average of one molecule per reaction, 37 % of reactions get none and must be negative however good the chemistry. Each additional molecule in the average multiplies that chance by 0.37 again: about 14 % for an average of two, about 5 % for three.

13 — The sampling floor

Template DNAFocal detail

The same chance governs reactions that do receive molecules. A reaction that receives two molecules instead of four starts one doubling behind and crosses the detection line about one cycle later, so at very low concentrations replicate reactions scatter by a cycle or more for reasons that have nothing to do with the instrument or the chemistry. The MIQE guidelines put the lower limit of reliable quantification of a good assay at about ten copies, above the limit of detection, for this reason.

So a test that is "as sensitive as possible" is one that reliably detects a few copies per reaction, and no more cycles, brighter dyes or better optics can change that. The only way to detect a rarer target is to put more of the sample into the reaction, by extracting from a larger volume or concentrating the extract.

+ The questionWhy couldn't PCR be automated until an enzyme was found in a hot spring?

The Klenow years

The first PCR experiments used a fragment of DNA polymerase from the bacterium E. coli, called the Klenow fragment. It copied DNA well at moderate temperatures, and it was destroyed by the first heating to separate the strands. So in the 1985 experiment that first used PCR, the tube was opened after every denaturation and fresh enzyme was added by hand: twenty times in a twenty-cycle run. Extension ran at 30 °C, the temperature the enzyme tolerated, and annealing at the same low temperature.

Both features limited the method. Adding enzyme every cycle was laborious and opened the tube twenty times to contamination. And annealing at 30 °C let primers bind to partial matches all over the sample, so the reaction made many products besides the target. The first experiments compensated by detecting the product with a labeled probe, which picked the right product out of the mixture. Cetus engineers built automated prototypes to do the pipetting, nicknamed "Mr. Cycle" after a coffee machine, but the real solution was an enzyme that did not need replacing.

14 — Before and after Taq

CopiesHeatFocal detail

A bacterium from Yellowstone

That enzyme came from a microbe isolated twenty years earlier for an unrelated reason. In 1966 the microbiologist Thomas Brock and his undergraduate student Hudson Freeze took samples from microbial mats in Mushroom Spring, in the Lower Geyser Basin of Yellowstone National Park, where the water was about 71.5 °C. From one sample they grew a bacterium that thrived in heat, and in 1969 they described it as a new species, Thermus aquaticus. It grows between about 40 and 79 °C, best at about 70 °C.

An organism that lives at 70 °C needs proteins that work at 70 °C. In 1976 the DNA polymerase of T. aquaticus was purified and found to be most active near 80 °C. In 1988 the Cetus group published PCR with this enzyme, now called Taq polymerase, in place of Klenow. Because Taq survived the denaturing step, it could be added once at the start, the tube could stay sealed, and the entire run could be automated with a programmable heater. Because it worked at high temperature, annealing and extension could be done hot, where primers bind only to good matches. The paper reported higher specificity and yield, products up to about 2,000 bases long, and amplification of single-copy human genes more than ten-million-fold. A commercial thermal cycler from Perkin-Elmer Cetus had reached the market in 1987, and PCR spread through laboratories within a few years.

Heat stability has a half-life

Taq is thermostable, not indestructible. A careful characterization of the purified enzyme, published in 1993, measured how long it kept half its activity at denaturing temperatures: about 45 to 50 minutes at 95 °C, but only 9 minutes at 97.5 °C. Two and a half degrees cut its life by a factor of five.

15 — Enzyme life at temperature

HeatFocal detail

The arithmetic makes the point of Chapter 2 concrete. An instrument that lets the liquid in its wells overshoot 95 °C by a couple of degrees, for a few seconds every cycle, consumes enzyme life the assay was not designed to spend, and one whose wells fall short risks leaving strands unseparated. The block itself is often driven past the set point on purpose (Chapter 10); what matters is the temperature of the liquid. Both errors are invisible unless someone measures them.

No proofreading, and a useful side activity

Cells copy DNA with proofreading polymerases that check each base and remove mistakes. Taq has no such proofreading activity, so it makes more errors. How many depends on how errors are measured, and different methods give results more than tenfold apart for the same enzyme, so comparisons are only meaningful within one study. A 2017 study that counted errors by sequencing single molecules, by scientists at a company that sells high-fidelity enzymes, measured about 1.5 × 10⁻⁴ errors per base per doubling for Taq, against about 5 × 10⁻⁶ for the proofreading enzyme Pfu and 5.3 × 10⁻⁷ for the company's own engineered enzyme, Q5.

For detection, Taq's error rate hardly matters. For a 100-base target copied through 30 doublings, the average number of errors per final molecule is about 100 × 30 × 1.5 × 10⁻⁴ ≈ 0.45, but the errors fall at random positions in different molecules, the population as a whole still carries the correct sequence, and a probe or a primer binds regardless of an occasional changed base elsewhere. Where one molecule is picked out of the mixture, to be cloned or to define a sequence, the errors matter, and a proofreading enzyme is used.

Taq has one more activity that turned out to be valuable. As it extends a strand, it degrades any short piece of DNA bound to the template in its path, an activity called 5′ to 3′ exonuclease. In 1991 its developers showed that this could be used to cut a labeled probe sitting inside the target, generating a signal as the target is copied. Chapter 12 shows how this became the most common detection chemistry in diagnostic PCR.

+ The questionIf every cycle doubles the product, why does every run flatten out?

The shape of a real reaction

If doubling continued, forty cycles would turn a single molecule into about a trillion, and fifty into a quadrillion. No reaction does this. Plotted on an ordinary scale, the amount of product in a PCR stays invisibly small for many cycles, then rises steeply, then bends over and levels off. The level at which it stops is called the plateau, and for a standard reaction it corresponds to roughly 0.3 to 1 picomole of product: about 2 × 10¹¹ to 6 × 10¹¹ molecules.

16 — The S-shaped curve

CopiesFocal detail

The plateau is not a sudden wall. Efficiency declines gradually as product accumulates, so the curve bends rather than breaks: a phase of near-constant doubling, a phase in which each cycle adds less than the last, and finally a phase in which cycles add almost nothing. That gradual bend is what gives an amplification curve its familiar S shape.

The usual explanation, and what experiments found

The usual explanation is that the reaction runs out of something, primers or building blocks. The numbers from Chapter 3 make that hard to sustain for a normal reaction.

Experiments that looked directly for the cause point elsewhere. A 1994 study ruled out exhaustion of the enzyme, the dNTPs or the primers, the accumulation of pyrophosphate, and enzyme degradation of the product, each as a sufficient explanation, and proposed that the product strands themselves were responsible. Once product is abundant, the two strands of each product find each other quickly when the reaction cools, and re-pairing with a long, perfectly matched partner competes with binding a short primer. Every strand that re-pairs before a primer reaches it is not copied in that cycle.

The competition shifts as the reaction proceeds, and the reason is a matter of who meets whom. The rate at which product strands re-pair depends on how often a product strand meets its complement, so it grows with the square of the product concentration: twice the product, four times the re-pairing. The rate at which primers bind depends on how often a product strand meets a primer, and the primer concentration hardly changes during the reaction. So the fraction of product strands lost to re-pairing in each cycle grows in step with the product itself. While product is scarce, in the early cycles, almost every strand is primed and copied; once it is abundant, a growing share re-pairs first, and the efficiency falls cycle by cycle.

A study published in 2000 found a second effect of abundant product. The polymerase binds double-stranded DNA in general, not only at primer ends, and as product accumulates the enzyme is increasingly held on product duplexes where it does nothing useful. In that study, short double-stranded DNA fragments in about thirtyfold molar excess over the polymerase inhibited it completely, while the depletion of dNTPs and the loss of enzyme to heat were not the main limits.

17 — What stops the reaction

CopiesPrimers and probesFocal detail

Other factors contribute in particular reactions: nonspecific products and primer-dimers competing for enzyme and primers, the gradual loss of enzyme activity to heat, and incomplete separation of strands when the product is very concentrated. The common thread is that the reaction is limited by its own success. The more product there is, the more the product gets in the way of making more.

The plateau forgets where the reaction started

The most important consequence of the plateau concerns measurement. The 1994 study also observed that reactions of the same target started from different amounts reached the same maximum. A sample with a thousand target molecules and a sample with a million both end in roughly the same amount of product, because both are stopped by the same self-limiting effects. The plateau carries almost no information about how much target there was at the start.

The same logic explains why the plateau is reached at a similar level whatever the starting amount. The effects that stop the reaction depend on the amount of product in the tube, not on how many cycles it took to get there, so a reaction that starts with more target simply reaches the limiting amount sooner. A reaction with a million starting copies is about ten cycles ahead of one with a thousand all the way up the exponential phase, and then both stall at roughly the same height.

That is why measuring the product at the end of a run, the way the first generation of PCR did by running it on a gel, can say whether the target was present and whether the product is the right size, but cannot reliably say how much target was present. The information about the starting amount lives earlier, in the cycles where the reaction is still doubling. Two samples that differ tenfold in their starting amount are, in those cycles, separated by a steady gap of a little over three cycles: the sample with more target reaches any given amount of product sooner. Part IV builds on exactly that gap.

+ Part III · The thermal cycler

The box that changes temperature.

The chemistry assumes that every well reaches each temperature, holds it for the stated time and leaves it on schedule. These two chapters open the instrument that is supposed to make that true: the path heat takes from the liquid to the room and back, the thermoelectric modules that pump it, and the engineering of speed, evenness and accuracy, including how each is specified, measured and sometimes missed.

+ The questionWhat has to be inside a box whose job is to change temperature on a schedule?

The path heat takes

A thermal cycler moves heat into a small volume of liquid and then out again, about forty times per run, as quickly and evenly as possible. Everything inside the box serves that path, and it helps to follow it from the sample outward.

The reaction, typically 10 to 50 microliters, sits in a thin-walled plastic tube or in one well of a plastic plate. The tube sits in a shaped hole in a metal block, which holds 96 or 384 of them. Beneath the block are thermoelectric modules, usually called Peltier modules, which pump heat into the block or out of it depending on the direction of the current through them. Beneath the modules is a heat sink, a large finned piece of metal, and a fan that blows room air across the fins. Above the tubes is a heated lid. Somewhere in the block is a temperature sensor, and a controller reads it and sets the current in the modules.

18 — The heat path

HeatFocal detail

Each layer resists the flow of heat and stores some of it, and the instrument's behavior is the sum of those resistances and stores. The plastic wall is a poor conductor, so the liquid always lags behind the block. The block must be massive enough to spread heat evenly across all its wells and light enough to change temperature quickly. The modules must pump enough heat to move the block several degrees per second. The heat sink must dump not only the heat taken from the block but also the electrical power the modules consume. A design that improves one of these usually worsens another, and Chapter 16 shows how such trade-offs are budgeted.

The Peltier module

A Peltier module is a flat array of small blocks of two kinds of semiconductor, connected electrically in series and sandwiched between two thin ceramic plates. When a current flows, it carries heat from one plate to the other. Reverse the current and the heat flows the other way. The same component therefore heats the block when the current runs one way and cools it when the current runs the other, with no compressor, no liquid and no moving parts, which is why it suits a compact instrument that must switch direction every few seconds.

19 — The Peltier module and its life

HeatFocal detail

The physics comes with losses. The current that pumps heat also generates resistive heat in the semiconductor, and some heat leaks back from the hot plate to the cold one. A manufacturer's technical reference writes the net heat pumped at the cold face as the pumping term minus half the resistive heat minus the leakage. The resistive heat works against the module when it cools the block and adds to the heat delivered when it warms it. The leakage runs from the hotter plate to the cooler one, and because a PCR block sits above the temperature of the heat sink, it slightly aids cooling, but the resistive loss dominates. Cooling is therefore slower than heating in many instruments: one manufacturer specifies its 96-well block at 4.4 °C per second heating and 2.2 °C per second cooling. And because the heat sink must reject both the heat pumped out of the block and all the electrical power put in, the fan and fins are sized for more heat than the block ever holds.

A Peltier module has no moving parts, but it does wear out. Each swing in temperature expands and contracts its layers by different amounts, because the semiconductor, the solder and the ceramic expand differently with heat, and the repeated stress cracks joints. A module maker's reliability guide calls thermal cycling the operating mode with the highest failure rates and notes that failures are often abrupt. In its own test of cycling-rated modules swung between 30 and 100 °C, every module survived at least 25,000 cycles and the group's mean time between failures was calculated at 68,000 cycles. A university study that cycled a commercial module to destruction found its performance down by a fifth after 40,000 cycles and failure at 45,000.

The block, the lid and the sensor

The block's job is to make every well the same temperature at the same moment. Aluminium is common: one early manufacturer's patent describes a block machined from a solid piece of a corrosion-resistant aluminium alloy, chosen for thermal uniformity. Silver conducts heat almost twice as well as pure aluminium (429 against 237 watts per meter per kelvin), and some instruments use silver or gold-plated silver blocks; one manufacturer lists the maximum ramp rate of the same instrument at 2.6 °C per second with a silver block and 2.3 with an aluminium one. Conductivity is only one limit, beside the power of the modules and the heat the block must store. Some designs spread heat with a sealed chamber in which a fluid evaporates at the hot side and condenses at the cold, the principle of a heat pipe.

The heated lid sits on the tube caps or the plate seal at about 104 to 105 °C. Without it, water would evaporate from the hot liquid and condense on the cooler cap, changing the volume and concentration of the reaction a little more in every cycle. The lid also presses the tubes down into the block, because a tube that sits loosely in its hole has an air gap around it and heats more slowly than its neighbors.

The temperature sensor is in the block, not in the liquid. The patent mentioned above placed an electronic sensor in a hole drilled into the block next to a column of wells. The controller reads it many times a second and adjusts the current in the modules so that the block follows the programmed profile. The liquid, behind its plastic wall, is a few seconds behind, and Chapter 10 shows that handling that lag is the central problem of temperature control.

Not every block instrument uses Peltier modules. The same early patent heated its block with a thin electrical heater and cooled it with chilled liquid pumped through channels in the block, supplied by a refrigeration unit. The Peltier module won in most later designs for its compactness and its ability to both heat and cool, not because nothing else works.

Other ways to change temperature

Moving a metal block up and down in temperature is one solution, and it carries the block's mass with it. Other architectures avoid the block.

In 1989 Carl Wittwer and colleagues described a cycler that put the reaction in thin glass capillaries and heated them with a stream of hot air, venting room air in to cool them. Glass capillaries hold little liquid and have a large surface for their volume, and air has little mass to heat, so the sample changes temperature quickly. Roche's original LightCycler instruments used this design; the later of them specified heating a 20-microliter capillary from 40 to 95 °C in under 15 seconds. Another design spins its tubes in a rotor inside a chamber of moving air; its maker claims a temperature uniformity, measured as a standard deviation, of ±0.02 °C, because every tube passes through the same air.

20 — Ways to change temperature

HeatFocal detail

Microfluidic designs can move the sample instead of the temperature. In a 1998 demonstration, a glass chip carried the reaction mixture in a channel that wound back and forth across three zones held at fixed temperatures, so that each pass through the zones was one cycle; twenty cycles took between 90 seconds and about 19 minutes depending on the flow rate. Other demonstrations have let natural convection carry the liquid between a hot bottom and a cooler top, or heated a thin gold film under the wells with blue light, completing 30 cycles in 5 minutes. A review of microfluidic PCR groups these into designs in which the sample stays put and the temperature cycles, where the profile can be changed in software, and designs in which the sample moves through fixed zones, where the cycle is built into the channel.

+ The questionWhy is changing a block's temperature quickly and evenly, without overshoot, the hard part of the engineering?

The sample is behind the block

The controller measures the block. The chemistry happens in the liquid. Between them is a plastic wall a fraction of a millimeter thick and the liquid's own bulk, and heat takes time to cross both. When the block jumps to a new temperature, the liquid follows it smoothly, covering a fixed fraction of the remaining difference in each interval of time. Engineers describe this as a first-order lag and characterize it by a time constant: the time the liquid takes to cover about 63 % of a step. An early manufacturer's patent measured a time constant of about 9 seconds for 100 microliters in a standard tube.

Two engineering responses solve this, and modern instruments use both. The first is to calculate the sample's temperature instead of measuring it. The controller knows the block's temperature history, the volume of liquid the user has entered and the time constant of the tube, and from these it computes, moment by moment, where the liquid's temperature must be. The hold timer starts when the calculated sample temperature reaches the set point, not when the block does. The second response is to overshoot. To make the liquid arrive sooner, the controller drives the block deliberately past the set point for a moment and then brings it back, so that the liquid approaches its target quickly without passing it. The patent describes exactly this: a model of the sample as a single time constant, a block driven past its set point, and a control loop, of the proportional-integral-derivative kind used throughout industry, running on the instrument's processor.

21 — The sample lags the block

HeatFocal detail

Current instruments expose the choice to the user. One manufacturer's cyclers offer a "calculated" mode, which uses the sample model, and a "block" mode, which simply holds the block. Sample-temperature control of this kind is what makes a 3-second hold meaningful. It also means that the reaction volume entered in the software is part of the temperature control: enter 50 microliters for a 20-microliter reaction and the model is wrong.

How fast is fast

Speed is quoted as a ramp rate, in degrees per second, and the number on a datasheet is easy to misread because manufacturers define it differently. Some quote the maximum rate the block reaches for a moment; some the average rate of the block over a temperature change; some the average rate of the calculated sample. For the instrument named in the CDC test, the manufacturer's brochure gives a peak block heating rate of 5.5 °C per second and a sample ramp rate of 3.5 °C per second in its fast mode and 1.6 °C per second in its standard mode. Another manufacturer gives a maximum of 5 °C per second and an average of 3.3, without saying whether either refers to the block or the sample; a third gives a maximum block rate of 6.5 and an average sample rate of 3.66.

22 — What a ramp rate means

HeatFocal detail

These numbers are not comparable unless their definitions match, and even then a maximum rate may be held only briefly. An instrument maker's own educational article makes the practical point that the honest test of speed is to run the same protocol on two instruments and time it. For the user, what matters is the time the liquid spends getting from one temperature to the next, multiplied by twice the number of cycles, since every cycle has a rise and a fall.

Every well the same

A 96-well block is a slab of metal several centimeters across, and its edges lose heat to the surrounding air and to the parts that hold it. Left alone, the corner and edge wells would run cooler than the center when the block is hot and warmer when it is cold. Designers fight this with geometry and with extra heat. The early patent cut a groove around the perimeter of the block, a guard band that limited heat flow from the central wells to the edge, and added separately controlled heaters along the edges to replace the heat lost there.

The result is specified as temperature uniformity, and here too definitions differ. One manufacturer states ±0.4 °C well to well, measured within 10 seconds of the block arriving at 90 °C; its later model states ±0.3 °C, with the hottest and coldest wells no more than 0.6 °C apart, measured 10 seconds after the block reaches its target. The brochure for the CDC's instrument states a uniformity of ±1 °C and an accuracy of ±0.5 °C without stating the conditions. Uniformity and accuracy are different properties: accuracy says how close the average well is to the set point, uniformity how far apart the wells are from one another. A block can be perfectly uniform and two degrees off, or accurate on average with a cold corner.

23 — Uniform is not the same as accurate

HeatFocal detail

Some instruments turn non-uniformity into a feature. A gradient block deliberately holds a range of temperatures across the block, up to 24 °C wide in one design, so that a developer can test several annealing temperatures in one run. Another design divides its block into six independently controlled zones that can differ by up to 5 °C.

When temperature goes wrong

Temperature errors are not hypothetical. A 2005 study published in a clinical microbiology journal tested six new 96-well Peltier cyclers with a measuring system certified to ±0.3 °C, at thirteen positions in each block, and then with a diagnostic PCR. The errors were largest in the first 15 seconds of the denaturing step. When the hold timer started, the liquid in the accurate instruments was 0.5 to 1.5 °C from the set point, while in two less accurate instruments it was still 13 to 20 °C short. With those two the diagnostic test gave no positive results until the denaturation temperature and time were increased. The authors named the critical parameters: holds timed too early, undershoot and overshoot, and differences in heat transfer across the block.

A cool well at the annealing step lets primers bind to near-matches, inviting wrong products (Chapter 5). A cool well at the denaturing step leaves strands paired, lowering the efficiency, and every cycle of lower efficiency delays the curve (Chapter 6). A hot well spends enzyme life (Chapter 7). Because the effects compound over forty cycles, a well that is consistently a degree or two off can give a result that differs from its neighbors', and a well that is grossly off can give no result.

Measuring what the wells see

Since the controller only reads the block, the wells must be measured separately. Temperature verification kits place calibrated probes in selected wells and record what they read through a test program. One manufacturer used such a kit, with probes calibrated traceably to national standards, to measure eleven cycler models from six manufacturers, and found steady-state errors between −0.33 and +0.39 °C and measured performance that did not always match the published specifications. That study was run by a competitor of most of the instruments tested, which is reason to read it as an illustration of method rather than a ranking. Accredited calibration services offer the same measurements under ISO/IEC 17025: steady accuracy, well-to-well spread, overshoot and undershoot, and stability during holds. One such service states an overall accuracy better than 0.1 °C and recommends calibration once a year or every 250 runs.

A metrology paper presented in 2017 by staff of CYCLERtest, a company that sells cycler calibration systems, argued that methods which judge a cycler by running a PCR on it cannot be traced to physical units and carry large uncertainties, and that acceptance limits should be derived from the needs of the user's own PCR method rather than taken from the manufacturer's specification. The point generalizes: how much error a cycler may have is a property of the test it runs.

+ Part IV · Counting

From copies to a number.

Copying answers whether a target is there; the central question also asks how much. These four chapters keep the chemistry fixed and change only when and how the product is read: once at the end, in every cycle, or one partition at a time. They show how light reports the copies, how the cycle at which the light crosses a line becomes a quantity, what that number can and cannot be trusted for, and how splitting a sample into thousands of partitions counts molecules directly. By the end of this part the central question is answered.

+ The questionOnce PCR could copy, what did each generation add to answering "how much"?

One chemistry, three ways to read it

The cycle described in Part II has not changed in its essentials since Taq polymerase made it automatic in 1988: denature, anneal, extend, repeat. What has changed, and changed the questions PCR can answer, is when and how the product is observed. The history is best told by holding the chemistry constant and asking, for each generation, what it reads, when it reads it, and therefore what question it can answer. Each answers a sharper question than the last.

Endpoint: is it there, and is it the right size?

The first generation ran the reaction to completion and then looked at the product. The standard way to look was gel electrophoresis. DNA's backbone carries a negative charge on every phosphate group, so in an electric field DNA moves toward the positive electrode. Pushed through a slab of gel made from agarose, a seaweed polysaccharide that forms a mesh of pores, short pieces of DNA move faster than long ones. After half an hour or so, the products of each reaction have spread out by size, a lane of DNA fragments of known lengths run alongside acts as a ruler, and a fluorescent stain makes the DNA visible under ultraviolet or blue light.

24 — Reading a gel

CopiesFocal detail

A band at the expected position says that the reaction made a product of the expected size, which is good evidence that the primers found their target. A band stained with ethidium bromide needs roughly half a nanogram to five nanograms of DNA, according to a stain supplier; more sensitive stains need less, but a band is still visible only well into the reaction, near or at the plateau. That is why the endpoint answers the question "is it there?" well and the question "how much?" badly: Chapter 8 showed that reactions started from very different amounts end at similar plateaus.

Endpoint detection has two other weaknesses. An empty lane can mean that the target was below the limit or that the reaction failed, and only a positive control run alongside distinguishes the two. And opening the tube after amplification releases billions of copies of product into the laboratory, where a trace can reach the next reaction and give a false positive; Chapter 20 describes the chemical defense against that.

Real-time: how much, compared with known amounts?

The second generation kept the tube closed and watched the reaction as it ran. In 1992, Russell Higuchi and colleagues, in the PCR group that passed from Cetus to Roche in 1991, added ethidium bromide, a dye that fluoresces much more strongly when bound to double-stranded DNA, to the PCR itself, and followed the fluorescence through the wall of the tube as the product accumulated. In 1993 they used a video camera to watch many reactions at once, cycle by cycle, and observed the relationship that real-time PCR has used ever since: the fewer cycles a reaction needs to reach detectable fluorescence, the more target it started with.

Reading in every cycle changes what can be measured. The plateau forgets the starting amount, but the exponential phase remembers it: a tenfold difference in starting target appears as a steady gap of about 3.3 cycles (Chapter 6). By recording the cycle at which each reaction's fluorescence crosses a fixed line, the instrument turns that gap into a number, and by running samples of known amount alongside, the number can be converted into a quantity. Real-time PCR answers "how much, relative to known standards?" A 1996 description of real-time PCR with fluorescent probes reported a working range of at least five orders of magnitude, with no handling of the product after amplification. Chapters 12 and 13 show how the light is produced and how the number is read.

Digital: exactly how many molecules?

The third generation abandoned the curve altogether. If a sample is diluted and split into many separate reactions so small that most contain one target molecule or none, each little reaction either amplifies or does not, and counting the reactions that light up counts the molecules. A 1992 study used this limiting-dilution principle with statistics to count the targets of a leukemic cell clone, detecting about two such targets among 160,000 normal genomes. In 1999, Bert Vogelstein and Kenneth Kinzler named the approach digital PCR, describing it as turning the exponential, analog output of PCR into a linear, digital count, and used it to detect a cancer-causing mutation in DNA shed into the stool of patients with colorectal cancer.

25 — Three ways to read one chemistry

CopiesHeatReadoutFocal detail

Digital PCR answers "exactly how many target molecules were in this volume?" without a curve and without standards of known amount, although, as Chapter 14 shows, it depends on knowing the volume of each partition and on a statistical correction for partitions that receive more than one molecule.

The evolution at a glance

The chemistry in all three is the same primers, polymerase and cycle. What changed is when the light is read and how many separate reactions the sample is divided into.

+ The questionHow does an instrument watch copies form inside a sealed tube, and tell several targets apart?

Light in, different light out

A real-time instrument cannot weigh the product or look at it under a microscope. It can only shine light into the well and measure the light that comes back. The method that makes this work is fluorescence. Certain molecules, fluorophores, absorb light of one color and, a few billionths of a second later, give off light of a longer wavelength, a slightly redder color. The instrument shines light of the absorbed color, the excitation, through a filter that passes only that band, and measures the returning light through a second filter that passes only the emitted band. Because the two bands differ, the detector sees the fluorophore's glow and very little of the much brighter excitation light.

What the instrument measures, then, is not DNA but light from a dye, and the whole art of real-time detection is to make that light rise in proportion to the copies of the target. Two families of chemistry do this.

Dyes that light up on any double-stranded DNA

The simplest approach adds a dye whose fluorescence increases strongly when it binds double-stranded DNA. Higuchi used ethidium bromide; most laboratories now use dyes such as SYBR Green I. As product accumulates, there is more double-stranded DNA for the dye to bind, and the well glows more brightly. SYBR Green I is often called an intercalating dye, one that slides between base pairs, but measurements show that it intercalates only at low ratios of dye to DNA and gives most of its signal from binding the surface of the helix.

The weakness of such a dye is that it reports any double-stranded DNA, including primer-dimers and other wrong products. The standard check is a melt curve. After the last cycle, the instrument heats the product slowly while watching the fluorescence. As each kind of product reaches its melting temperature, its strands separate and the dye lets go, so the fluorescence drops. Plotting the rate of that drop against temperature gives a peak at the melting temperature of each product: one sharp peak suggests one product; a second, lower peak often reveals primer-dimers. A supplier's technical note warns that a single product with regions of different G–C content can give two peaks, so a melt curve is an indicator, not proof, and a gel or sequencing confirms.

Probes that light up only on the target

Diagnostic tests mostly use a third short oligonucleotide, a probe, that binds inside the target between the two primers and carries a fluorophore at one end and a quencher at the other. While the probe is intact, the quencher sits close enough to absorb the fluorophore's energy, and the probe is dark.

The most widely used design exploits the activity of Taq described in Chapter 7. As the polymerase extends a primer along the template, it meets the probe bound in its path and cuts it apart with its 5′ to 3′ exonuclease activity. The fluorophore, now separated from the quencher, glows. Each cycle cuts probes on every newly copied target, so the fluorescence rises with the product, but only with the product the probe recognizes. The method was shown with a cleavable probe in 1991, made fluorescent with a reporter and a quencher by 1993, and described as real-time quantitative PCR with these probes in 1996. It is often called a TaqMan or hydrolysis probe. Other designs, such as molecular beacons, introduced in 1996, are hairpins that open and glow when they bind a matching target.

26 — How a hydrolysis probe reports

Template DNACopiesPrimers and probesReadoutFocal detail

Because a probe must bind as well as both primers, a probe test has a third check on identity, which makes it far less likely than a DNA-binding dye to report primer-dimers or wrong products, although a poorly designed probe can still produce signal without target (Chapter 22). The CDC test used hydrolysis probes carrying the fluorophore FAM at one end and a quencher called Black Hole Quencher 1 at the other, for each of its two viral targets and for a human control gene. All three probes used the same fluorophore, so each specimen was run in three separate wells, one per target, and the instrument read the fluorescence at the 55 °C step of each cycle, when the probes are being cut. In March 2020 the FDA also accepted probes with an additional internal quencher. The test's instructions told users to set the quencher to "none" in the instrument's software; that is a software setting for a quencher that emits no light of its own, not a probe without a quencher.

The hardware of a reading

An optical system needs a light source, filters and a detector, and real instruments combine them in several ways. The instrument named in the CDC test uses a 75-watt halogen lamp, a set of five excitation filters and five emission filters, and a CCD camera that images the whole plate at once. Another instrument uses a 100-watt xenon lamp and a cooled camera. Newer designs use light-emitting diodes: one uses a white LED and a CMOS camera with six excitation and six emission filters; another uses six colored LEDs and six filtered photodiodes mounted on a small carriage that moves over the plate and reads each well in turn, the same light path for every well, taking about 12 seconds to read all channels of a 96-well plate, according to its maker. A rotor instrument reads each tube as it spins past a photomultiplier tube, a very sensitive detector.

27 — Three optical architectures

ReadoutFocal detail

Each architecture trades something. A camera reads all wells at once but views each well from a slightly different position and angle, so it needs calibrations that locate every well on the image and correct for where it sits. A scanning carriage reads every well identically but takes time, which must fit inside the cycle step during which the reading is valid. Lamps give broad, bright light and age; LEDs are stable and long-lived, and colored LEDs give one narrow band each.

Several targets in one well

A filter pair defines a channel, and an instrument with five or six channels can, in principle, read five or six fluorophores in the same well. One instrument's channels give the idea: the first excites at 450 to 490 nanometers and reads 510 to 530, for FAM; the second excites at 515 to 535 and reads 560 to 580, for a dye called HEX; and so on up to the near infrared. The CDC's later multiplex test detected four targets in one well, influenza A, influenza B, SARS-CoV-2 and the human control, with four different fluorophores.

28 — Channels and crosstalk

ReadoutFocal detail

Multiplexing has a cost: fluorophores emit over broad bands, so a bright dye leaks some light into a neighboring channel. Instruments correct this with calibration. The maintenance guide for one family of instruments lists the calibrations its owner must run: a region-of-interest calibration that maps where each well falls on the camera image, every six months or after a lamp change; a background calibration that measures the instrument's own fluorescence, monthly; and a pure-dye calibration that records each dye's spectrum on that instrument, every six months. During a run the software uses the pure-dye spectra to work out how much of the measured light came from each dye, a step called multicomponent analysis. Many master mixes also contain an inert reference dye, ROX, and some instruments divide the reporter signal by it, which corrects for differences in volume, bubbles and condensation. The CDC test's instructions switched that correction off, setting the passive reference to "none", although its original master mix contained ROX.

+ The questionWhat does a Ct value measure, and how far can it be trusted as a quantity?

From light to a curve

In every cycle, at the step the test specifies, the instrument records the fluorescence of every well in every channel. Plotted against cycle number, the readings for a positive well trace the S shape of Chapter 8: flat for many cycles, then a steep rise, then a plateau. The flat early part is not zero. It is the background: light from the dye, the plastic and the optics that is there before any product accumulates. The analysis software estimates this baseline from early cycles, typically somewhere between cycles 3 and 15, and subtracts it, so that each curve starts from a common floor.

Then a horizontal line is drawn, the threshold, above the background noise and well below the plateau, in the region where the curves are rising steadily. The cycle at which a well's curve crosses the threshold, interpolated between readings so that it can be a fraction such as 27.4, is the number the instrument reports. Applied Biosystems called it the threshold cycle, Ct; Roche called it the crossing point, Cp; the 2009 MIQE guidelines proposed the neutral name quantification cycle, Cq, which this guide uses alongside the familiar Ct.

29 — Baseline, threshold and Cq

ReadoutFocal detail

The CDC test's instructions show how much of this is in the user's hands. They told laboratories to leave the baseline at the software's default and to set the threshold by hand, dragging it into the exponential part of the curves above the background. The threshold is a setting, and the Ct depends on it.

Why the crossing cycle measures the start

The threshold corresponds to a fixed amount of product: every well that crosses it holds roughly the same number of copies at that moment. A well that started with more target gets there in fewer doublings. This is the backward reading of the fact that a cycle can at most double the target. If one sample starts with ten times more target than another, it reaches any given amount of product as many cycles earlier as it takes to multiply by ten. At perfect doubling that is 3.32 cycles, because 3.32 doublings make a factor of ten.

The arithmetic contains an assumption that is easy to forget: the efficiency is known and the same for the samples being compared. Two samples with the same starting amount but different efficiencies, perhaps because one carried an inhibitor from the specimen, cross at different cycles, and the difference will be misread as a difference in amount.

The standard curve

To turn Cq into a number of copies, a quantitative test runs standards: a dilution series of known amounts of target, for example tenfold steps from ten copies to a million. Plotting each standard's Cq against the logarithm of its amount gives a straight line, and an unknown sample's Cq can be read off the line as an amount.

30 — The standard curve

ReadoutFocal detail

The slope of the line reports the efficiency. Perfect doubling gives a slope of −3.32 cycles per tenfold step; 90 % efficiency gives −3.59, because more cycles are needed for each tenfold increase; a slope of −3.10 corresponds to an apparent efficiency of 110 %, more than doubling, which is physically impossible and signals that something other than clean amplification is shaping the curves. A range of 90 to 110 % is widely used as an acceptance criterion, together with a straight-line fit with R² of at least 0.98. These figures come from laboratory practice, for example the criteria of the European network of laboratories that test for genetically modified food, rather than from the MIQE guidelines, which set no fixed range.

The 2025 revision of MIQE asks for more than a slope. It recommends converting Cq values into efficiency-corrected quantities reported with intervals that show their uncertainty, exporting the raw fluorescence data in vendor-neutral formats, and stating both a limit of detection and a lower limit of quantification, the smallest amount that can be measured with useful precision. It notes that the limit of detection cannot be below about three copies (Chapter 6) and that a good assay quantifies from about ten.

Comparing samples without standards

Many research studies want to know whether a gene's activity changes, not how many copies there are. The widely used shortcut, published by Kenneth Livak and Thomas Schmittgen in 2001 and known as the ΔΔCt method, compares each sample's target with a reference gene in the same sample, and then compares those differences between samples, converting the result with powers of two. It works only if the target and the reference amplify with nearly the same, nearly perfect efficiency. The 2025 MIQE revision strongly recommends against relying on it and favors efficiency-corrected quantities instead.

Reading a qualitative result

Most infectious-disease tests do not report a quantity at all. They compare each Cq with a cut-off and apply rules that combine the targets and the controls. The CDC test counted a target as detected if its curve crossed the threshold before cycle 40.00 of its 45, and combined three wells per specimen with the run's controls. The table shows how its rules read a set of wells. The Cq values are illustrative, chosen to show each rule; they are not from a real run.

The last row surprises many readers. The human control gene, RNase P, is there to show that a negative result is believable: that the swab collected human cells and that extraction and amplification worked. A specimen in which both viral targets are detected has already shown that amplification worked, so under the CDC rules it stands whether or not the control is detected.

What a Ct cannot tell you

A Cq is a property of one test on one instrument with one threshold. The same sample run on different tests can give very different values: Canada's public health agency reports that identical samples tested in several provinces gave Ct values up to 8 cycles apart, for example 22 on one test and 30 on another, which is a difference of more than a hundredfold if read naively. A statement by the American Association for Clinical Chemistry notes that there is no international unit for Ct, and a joint statement by two professional societies lists the reasons: the collection device, the specimen type, the extraction, the genetic target and the chemistry all shift it.

31 — Same sample, different Ct

ReadoutFocal detail

During the COVID-19 pandemic, many people hoped Ct would show whether a patient was infectious. Studies that tried to grow live virus from positive samples illustrate the difficulty. A 2020 Canadian study cultured 90 samples and grew no virus from any with a Ct above 24. A 2020 English study of 324 samples found that the chance of growing virus fell to 8 % at Ct values above 35. The two studies used different tests, culture methods and laboratories, and the highest Ct at which they still grew virus differs by more than ten cycles. That gap cannot be assigned to the tests alone, but it shows that a Ct cut-off found in one setting cannot be carried to another. The professional bodies advised laboratories not to report Ct for routine patient management, or to report it only with an explanatory comment.

+ The questionHow can splitting one sample into thousands of partitions count molecules without a standard curve?

Split, amplify, count

Real-time PCR infers the starting amount from how long the reaction takes to become visible, and that inference depends on efficiency and standards. Digital PCR asks a simpler question of many small reactions instead of a hard question of one large one. The reaction mix, with the sample in it, is divided into thousands of tiny partitions, droplets of oil-separated liquid or chambers etched in a chip, each holding a fraction of a nanoliter. The whole set is cycled. A partition that received at least one target molecule amplifies it to a bright signal; a partition that received none stays dark. The instrument then counts the bright partitions.

The count does not depend on efficiency in the way a Cq does. A partition that amplifies a little slowly still ends up bright after forty cycles; it is counted as positive just the same. And no standards are needed, because the answer is a count of molecules in a known volume, not a comparison with samples of known amount.

32 — Split, amplify, count

Template DNAReadoutFocal detail

Molecules arrive at random

The catch is that molecules do not distribute themselves one to a partition. They land at random, so at any useful concentration some partitions receive two or three molecules and light up only once. Counting bright partitions therefore undercounts the molecules, and the undercount grows with concentration. Digital PCR corrects for this with the same statistics of random scattering that set the sampling floor in Chapter 6.

The correction works from the dark partitions, because those are unambiguous: a dark partition received exactly zero molecules. Chapter 6 showed that if the average is one molecule per partition, 37 % of partitions receive none, and each additional molecule in the average multiplies that fraction by 0.37 again. Run backward, the fraction of dark partitions gives the average number of molecules per partition.

The undercount grows quickly. At an average of 0.1 molecules per partition, about 5 % of the bright partitions hold more than one molecule; at an average of one, 42 % do; at an average of 1.6, about 60 % do. The popular picture of digital PCR, in which each partition holds one molecule or none, is true only at very low concentrations; at the concentrations that give the best precision, most bright partitions hold more than one molecule, and the correction does real work.

33 — How much the correction does

Template DNAReadoutFocal detail

Precision and range

Because the answer comes from counting, its precision follows the statistics of counting. A modeling study by scientists at a digital PCR manufacturer found that precision is best at an average of about 1.6 molecules per partition, when about 20 % of partitions are dark, and that it improves with the square root of the number of partitions: four times as many partitions halve the relative uncertainty.

The range has hard edges at both ends. At the top, when nearly every partition is bright, the few dark ones carry all the information: with 20,000 partitions, a single remaining dark partition corresponds to an average of about 9.9 molecules per partition, and beyond that the measurement collapses. A run with every partition bright, or every partition dark, carries no quantitative information at all. At the bottom, the limit is the volume actually analyzed, because no molecule can be counted that was not loaded: one manufacturer's plates analyze about 3 microliters per well in one format and about 24 microliters in another, and the larger volume can detect rarer targets.

The partition volume is the calibration

The phrase "absolute quantification without a standard curve" needs one qualification. Digital PCR counts molecules per partition, and converting that into molecules per microliter requires the volume of a partition. If the volume is wrong, every result is wrong by the same factor. A 2015 study by the European Commission's Joint Research Centre measured the droplets of a widely used system under a microscope and found a mean volume of 0.834 nanoliters, about 8 % smaller than the value then built into the instrument's software, enough to explain a systematic disagreement with another digital platform. The software value was later changed to 0.85 nanoliters. Digital PCR needs no standard curve, but it needs its partitions to be the volume the software assumes, uniformly.

The instruments

The systems described here divide a reaction into between about 8,500 and 100,000 partitions, depending on the platform and the format, and they read them in one of two ways. The droplet systems that made the method widespread make about 20,000 droplets from a 20-microliter sample in a separate droplet generator; the droplets are cycled in an ordinary thermal cycler and then streamed single file past a detector in a reader, one instrument after another, with plates moved between them by hand. The original description of this system, in 2011, reported droplets of about a nanoliter, generated and read at about a thousand per second. Later droplet readers detect six colors, and a series launched in 2025 detects seven.

34 — Two ways to build a digital PCR system

HeatReadoutFocal detail

Other systems fix the partitions in place and image them. Some integrate everything: the user loads a plate and the instrument partitions, cycles and images it. One offers plates with about 8,500 or 26,000 partitions per well; another uses 20,480 microchambers per sample, with results in under 90 minutes according to its maker. A third partitions plates of 20,000, 28,000 or 100,000 nanowells in a separate partitioning step before its analyzer cycles and images them. Another approach packs droplets into a flat single layer in a chip and images it. The choice between them is an engineering trade between throughput, hands-on time, the volume analyzed and the cost per sample, and the market is still moving: in July 2025 one of the large suppliers announced it had acquired a smaller company whose droplet-imaging technology it now sells.

Digital PCR earns its cost where counting matters more than speed: measuring a rare mutation among many normal copies of the same gene, for example a tumor mutation in DNA from a blood sample, measuring copy numbers absolutely without standards, and in situations where substances from the sample slow the reaction, which digital PCR is reported to tolerate better than real-time PCR. It is slower and more expensive per sample than real-time PCR and covers a narrower range of concentrations in one run, so it complements rather than replaces it.

The central question, answered

The guide's central question can now be answered in full. A PCR instrument does little more than heat and cool a tube on a schedule. It finds one sequence among billions because two primers bind only where their sequences match and only the stretch between two primers facing each other is copied exponentially (Chapter 5). It copies that stretch a billion times because each cycle of separating, priming and extending lets every copy act as a template, so the count nearly doubles each cycle, and a heat-stable enzyme survives the whole run (Chapters 4, 6 and 7). It tells you how much was there because a probe or dye makes light in proportion to the copies, and the cycle at which the light crosses a line, or the fraction of partitions that never light, reports the starting amount (Chapters 12 to 14). The rest of the guide deepens the second half of the answer: how the instrument is engineered so that its temperatures and its readings can be trusted, and how a test built on it earns that trust.

+ Part V · The instrument as a system

Engineering the whole.

Earlier chapters took the instrument apart: a thermal system, an optical system, the chemistry they serve. These five chapters put it back together the way a systems engineer would. They show how the parts are orchestrated and how a number is produced, how a stated need becomes a specification for every part, why the instrument and the assay cannot be specified apart, what it takes to integrate a whole laboratory into a cartridge, and how every part and the whole are verified.

+ The questionHow do the heater, the optics and the software act as one instrument, and how do photons become a number?

No part is clever

A real-time PCR instrument can be described as a set of subsystems, each with one job. None of them knows anything about DNA.

The thermal subsystem, the block, its Peltier modules, the heat sink, the lid and the temperature sensors of Chapters 9 and 10, delivers the temperature profile to the liquid in every well. The optical subsystem, the light source, filters and detector of Chapter 12, measures the fluorescence of every well in every channel. A motion subsystem moves what must move: the tray that carries the plate into the instrument, a filter wheel, or a carriage that scans the optics across the plate. The electronics supply power, drive the modules with currents of several amperes, condition the small signals from the temperature sensors and the light detector, and convert them into numbers. Firmware on the instrument's processors runs the control loops and the sequence of events in real time. Software, usually on an attached computer, lets the user set up a run, analyzes the data into curves, Cq values and calls, and stores and exports the results. Two more parts belong to the system although they are not in the box: the consumables, plates, tubes and seals, and the assay, the chemistry in the tube.

35 — The instrument as a system

Primers and probesHeatReadoutFocal detail

The instrument's behavior is not a property of any one of these. It emerges from how they work together, which is why the most consequential engineering decisions are about the connections between them.

The interfaces are where the system lives

Each pair of subsystems that touch shares an interface, and each interface has parameters that both sides must agree on. The thermal interface between block and plate depends on the shape of the wells, the wall thickness of the plastic and the pressure of the lid. The optical interface depends on the seal or cap the light passes through, its clarity and its own faint fluorescence. The data interface between the instrument and the computer, and between the computer and the laboratory's information system, carries run settings one way and results the other. The electrical interface to the mains and to the electromagnetic environment of a laboratory is governed by safety and compatibility standards (Chapter 19). And the human interface, the software screens and the physical loading of plates, is where a user can enter the wrong reaction volume and quietly change the temperature model of Chapter 10.

Choreography

The firmware's central task is timing. Within each cycle, it must bring the calculated sample temperature to each set point, hold it for the programmed time, and, at the step the assay specifies, trigger the optical subsystem to read every well in every channel the assay uses. In the CDC test, the read happened during the 55 °C step, which lasted 30 seconds. The read has to fit inside that step after the liquid has reached temperature, and it has to happen at the same point in every cycle, because a reading taken a few seconds earlier or later in the step, while probes are still being cut, would not be comparable with its neighbors.

36 — Choreography of one cycle

HeatReadoutFocal detail

How long a read takes depends on the optical architecture. One manufacturer states that its scanning carriage reads all channels of a 96-well plate in about 12 seconds, or a single channel in about 3; a camera instrument takes an exposure for each filter pair, changing filters between them. Either way, the read time is part of the cycle time and constrains how short the reading step can be. Meanwhile the thermal control loop runs continuously, reading the block sensor many times a second and adjusting the current in the modules. The instrument is one closed loop, the thermal control, and one measurement, the optical read, placed within it by a schedule. No single part is clever; the coordination is.

From copies to a result

The most useful way to understand what the instrument does is to follow one fact, the number of target copies in a well, through every form it takes before it reaches a report. The instrument never measures that number directly. It measures something that depends on it, and converts.

Copies of target are made in the well. As copies are made, the polymerase cuts probes bound to them, and each cut probe releases a fluorophore that emits light when excited. The light passes through the seal and an emission filter and strikes a detector, which converts photons into electrons, an analog signal that is amplified and converted into digital counts. Those counts are then corrected: the instrument locates each well in its field of view, subtracts its own background light, corrects for differences in sensitivity across the plate, and separates the contributions of different dyes using their calibrated spectra. Optionally it divides by a reference dye. What remains is one number per well, per dye, per cycle. The software subtracts each well's baseline, draws the threshold, finds the fractional cycle at which each curve crosses it, and compares that Cq with a cut-off or a standard curve to give a call or a quantity, which is written into a report or sent to a laboratory information system.

37 — From copies to a result

CopiesPrimers and probesReadoutFocal detail

Every one of those conversions can add error, and the final number is only as good as the weakest of them. This is why the calibrations of Chapter 12 exist, and why instruments are verified as a whole. The maintenance guide for one instrument family describes a verification run with a reference plate of a human gene, in which the instrument must distinguish wells of 5,000 genome copies from wells of 10,000 with 99.7 % confidence.

Software and data are part of the device

Because the result is computed, the software that computes it is part of the measuring instrument. For a diagnostic test, the version matters: the CDC test was authorized for one instrument running one version of its software, version 1.4. Changing the algorithm that sets the baseline or finds the crossing changes the Cq values, and therefore potentially the calls. The FDA's 2023 guidance on the content of premarket submissions for device software functions, applies to device software functions including firmware and software that controls a device, which covers the software of a diagnostic instrument (Chapter 19).

The data matter beyond the run that produced them. The 2025 MIQE revision asks that raw fluorescence data be exported in vendor-neutral formats, so that a result can be re-analyzed with different software and checked. An instrument that can only report Cq values, and keeps its raw data in a closed format, makes that impossible.

+ The questionHow does a need such as "detect a few copies in about an hour" become numbers for a heater, a lens and a line of code?

The intended use comes first

Every instrument and test begins with a statement of what it is for, and in a regulated product that statement, the intended use, is written down before anything is designed. It names what is measured, in what kind of specimen, by whom, in what setting, to inform what decision, and what kind of result is given. The CDC test's intended use, in its instructions, was the qualitative detection of nucleic acid from SARS-CoV-2 in upper and lower respiratory specimens, in laboratories certified for high-complexity testing.

That one sentence decides much of the system. Qualitative means a cut-off and rules, not standards and a quantity. Nucleic acid from SARS-CoV-2, an RNA virus, means a reverse transcription step (Chapter 20) and RNA-friendly handling. Respiratory specimens means a sample type with its own inhibitors and variability, and therefore an internal control. High-complexity laboratories means trained staff, a separate extraction step and an instrument that can be fed plates by hand. A test intended for a clinic, run by a nurse in twenty minutes, would share the chemistry and almost nothing else.

The V

Systems engineers draw the path from a need to a working product as a V, a picture credited to Kevin Forsberg and Harold Mooz in 1991. Down the left side, the need is decomposed: user needs into system requirements, system requirements into requirements for each subsystem, those into detailed designs. At the bottom the parts are built. Up the right side they are assembled and checked: each part against its own requirements, the assembled subsystems against theirs, the whole system against the system requirements, and finally the finished product against the original user needs.

38 — The V

Focal detail

The two checks on the right side have names that regulators use precisely. The FDA's 1997 guidance on design controls defines verification as confirming that the design's outputs meet its inputs, that the product was built as specified, and validation as confirming that the device meets the user's needs and intended uses, that the right product was built. The same guidance asks that design inputs be developed to an engineering level of detail, and notes that real development loops back between phases. In the United States these design controls are now required through the Quality Management System Regulation, which took effect on 2 February 2026 and incorporates the international quality standard ISO 13485:2016, including its clause on design and development.

Requirements flow down

The left side of the V is where a need such as "detect a few copies in about an hour" becomes numbers. Each system requirement splits into requirements on several subsystems, and most subsystems answer to several system requirements. The example below is illustrative, built from figures in earlier chapters.

39 — Requirements flow down

Focal detail

Sensitivity. A limit of detection of about five copies per reaction cannot be lower than the sampling floor of about three (Chapter 6), so the assay must put enough of the specimen into each reaction: in the CDC test, 5 microliters of extract in 20. The extraction must recover the RNA efficiently and remove inhibitors. The optics must distinguish the light of 10¹⁰ to 10¹¹ copies of product from background reliably enough that the threshold can sit low, and the run must have enough cycles, with a cut-off at cycle 40 of 45 in the CDC test, for a handful of copies at realistic efficiency to cross it.

Time to result. The run time is the number of cycles times the cycle time, plus the initial steps. The cycle time is the hold times plus the time the liquid takes to move between temperatures twice, plus any time the read adds. So the time requirement becomes a requirement on the sample ramp rate, on the sample-temperature algorithm that makes short holds real, and on the read time.

Throughput. A requirement for a number of specimens per run becomes a plate format and a layout. If each specimen takes three wells, as in the CDC test, and three controls take three wells each, a 96-well plate holds 29 specimens. A multiplex test that detects all targets in one well with different dyes would hold 93, about three times as many, at the cost of more optical channels, calibration and unmixing.

Specificity and robustness. The assay's tolerance to temperature error becomes a requirement on accuracy and uniformity at the annealing and denaturing steps, and the instrument's guaranteed accuracy and uniformity become a requirement on the assay's robustness.

Error budgets

Where a requirement depends on several contributions, engineers allocate it among them in a budget. Temperature at the annealing step is a clear case. The actual temperature of a given well differs from the set point by the instrument's accuracy, plus that well's departure from the average, plus any residual error from the sample-temperature model, plus the uncertainty of the sensor's calibration. One manufacturer's specification of ±0.2 °C accuracy and ±0.4 °C uniformity within 10 seconds of arrival, both stated at 90 °C, means a well could, in the worst case, be about 0.6 °C from its set point at that temperature, and the specification does not say what holds at the annealing step; the brochure figures for the CDC's instrument, ±0.5 °C accuracy and ±1 °C uniformity, allow about 1.5 °C. Engineers often combine independent errors by adding their squares and taking the square root, which gives a smaller, more likely figure, but a diagnostic test must work in the worst well.

40 — Budgets for time and temperature

HeatFocal detail

The budget closes on the assay side. A test developer establishes how far the annealing temperature can move before the test loses sensitivity or specificity, typically by running it deliberately at temperatures above and below the set point, and that tolerance must exceed the instrument's worst-case error with margin. The optical budget closes the same way: background, noise and calibration error set the lowest signal the instrument can distinguish, which sets where the threshold can sit, which sets how precisely Cq is determined, which the verification plate of Chapter 15 checks.

Trade-offs

Requirements pull against each other, and the systems engineer's job is to make the trade explicit rather than let it happen by accident. Speed pulls against uniformity, because a lighter block changes temperature faster and spreads heat less evenly. More optical channels allow more targets per well and cost more, demand more calibration and leak more light between channels. More cycles reach lower copy numbers and lengthen the run. A sealed cartridge prevents contamination and simplifies the user's work, and fixes the assay and the volume. Every one of these is a decision about which requirement matters more for the intended use, and every one shows up later as a specification someone must verify.

+ The questionWhy can't an instrument be specified without its assay, or an assay validated without its instrument?

The tube is an interface

The plastic tube or plate looks like packaging. It is an engineered part of both the instrument and the test, and it sits on two interfaces at once.

On the thermal side, the wall of the tube is the resistance that makes the liquid lag the block (Chapter 10). A thinner wall, a smaller volume or a lower-profile well shortens the lag, which is why fast instruments use their own plates: the instrument named in the CDC test used fast 96-well plates or 8-tube strips with 100-microliter wells, and reactions of 10 to 30 microliters. The tube's fit in the block and the lid's pressure on it decide how well heat crosses from metal to plastic. The volume in the tube is an input to the instrument's sample-temperature model. Change the plate, and the temperature the liquid actually reaches can change without any change to the instrument's settings.

On the optical side, light enters and leaves through the cap or seal. Its clarity, its own faint fluorescence and whether condensation forms on it all affect the signal, which is one reason for the heated lid. The reaction volume sets the height of the liquid in the well and therefore how much of it the optics see.

41 — The tube sits on two interfaces

HeatReadoutFocal detail

The chemistry is matched to the machine

The reagents are matched to the instrument as closely as the plastic is. A master mix's hot-start enzyme needs an activation step of a certain length at a certain temperature, and the instrument's profile must provide it: the CDC test's four authorized mixes used three different initial steps: reverse transcription for 15 minutes at 50 °C, 10 minutes at 50 °C or 15 minutes at 45 °C, followed by 2 or 3 minutes at 95 °C. A mix formulated for slow cycling may not finish its extension in the short holds of a fast instrument. The fluorophores on the probes must be ones the instrument has channels for and has been calibrated with; the brochure of the CDC's instrument lists the dyes it was calibrated for, beginning with FAM, the dye all three of the test's probes carried.

The software is part of the method

Then there are the settings, which look like configuration and are in fact part of the test. The CDC's instructions specified that the instrument run in its standard mode, not its faster mode; that the passive reference dye be set to none; that each detector be set to FAM with the quencher set to none; that the baseline be left at the default and the threshold set by hand in the exponential region; and that a target count as detected below a Ct of 40.00. Each setting changes what the instrument does or how its output is read. Running the same chemistry in fast mode would change the time the liquid spends at each temperature; changing the threshold would change every Ct; changing the cut-off would change the calls.

The test system

Regulators draw the consequence explicitly: what is evaluated is a test system, the assay, the instrument and the software together. The FDA's guidance on assay migration, finalized in April 2013, covers moving an assay to a new instrument system whose performance the agency has not evaluated, and defines the system as the assay, the instrument and the software. It is written for approved (class III) assays, licensed donor-screening tests and cleared assays whose migration raises particular concerns. For a quantitative test it recommends, among other studies, showing that the limits of blank and detection are equivalent, testing precision at several sites and comparing results on at least 180 samples across the measuring range; for a qualitative one, panels of low-level samples and comparison panels of at least 100 positive and 100 negative samples; and for every nucleic acid test, studies of carry-over between samples.

A later FDA policy, finalized in August 2022, starts from the premise that an assay is cleared for use on specific instruments. It describes an instrument family, instruments from the same manufacturer with the same general architecture, design, tolerances and capabilities, and the conditions under which a cleared assay may be extended to another member of the family without a new submission. It notes that if the software must be modified to run the assay on the added instrument, a new submission is likely to be needed.

42 — The test system

Primers and probesReadoutFocal detail

The CDC test shows the test system in practice. Its instructions named one instrument, with one software version. They named the master mixes that could be used: initially one, with three more added on 30 March 2020. They named the extraction methods: a list of kits and automated systems, extended in June and July 2020 as new platforms were shown to work, plus a heat-treatment alternative with a warning that it had been tested on few clinical specimens and might lose sensitivity. Each addition came with data and a revised authorization. A laboratory that ran the test on another instrument, or with another extraction kit, was no longer running the authorized test; it was running a modified test whose performance it had to establish itself.

What this means for developers

For a company developing an assay, the choice of instrument platform is therefore a long-term commitment: the assay's performance will be established on that platform, and moving it later is a project with its own studies. For a company developing an instrument, the assays that will run on it are part of its design inputs: the plates, volumes, dyes, profiles and analysis settings they need, and the stability of those parameters over the instrument's life. Software updates, a new lamp or LED supplier, a changed block design or a different plate supplier all reach the tests running on the instrument, and must be controlled with that in mind.

Some instruments are closed: they run only the manufacturer's own assays, in its own consumables, with fixed settings, so that the test system is defined once. Others are open: they run any assay a laboratory develops, which offers flexibility and moves the burden of establishing the test system's performance onto the laboratory.

+ The questionWhat does it take to put extraction, PCR and reading into one closed cartridge?

Why put a laboratory in a cartridge

A test like the CDC's runs in a molecular laboratory. A trained technician extracts nucleic acid from each specimen with a kit or a separate automated system, pipettes the extract and the reagents into a plate, seals it, loads it into the instrument and, after the run, interprets the curves. CDC estimated in February 2020 that the process took about four hours from the start of sample processing to a result, and the test was authorized only for laboratories certified for high-complexity testing. Every manual step needs skill, takes time, and opens a route for error or contamination, and every opened tube of specimen or extract is a route for the cross-contamination that causes false positives; the real-time plate itself stays sealed after cycling (Chapter 11).

A sample-to-answer system puts all of those steps into one sealed disposable cartridge and one instrument. The user adds the specimen, closes the cartridge, puts it in the instrument and receives a result. Times to result reported by makers or in regulatory summaries are about 45 minutes for one SARS-CoV-2 test cartridge authorized in March 2020, about 45 minutes, according to its maker, for a system that runs a panel of 22 respiratory targets at once, and 15 to 20 minutes for another. The engineering that makes this possible is integration: every function of the laboratory, built into a molded part that costs little enough to throw away.

43 — Steps and time to result

Focal detail

One cartridge, one valve

The GeneXpert system, described in a 2005 paper and in its maker's technical documents, shows one way to do it. Its cartridge is a plastic body with chambers for the sample, the reagents and the waste, arranged around a central rotary valve. The instrument turns the valve to connect any chamber to a central channel, and drives a plunger that draws liquid in and pushes it out, so that one moving part routes every liquid where it is needed in sequence. An ultrasonic horn in the instrument presses against the cartridge to break open cells and viruses. The reagents are freeze-dried beads that dissolve when liquid reaches them: an enzyme bead, a bead of primers and probes for the target, and a bead containing a sample-processing control. The final reaction is pushed into a thin reaction tube that sits in a thermal module with ceramic heater plates and a fan for cooling, and in optical blocks that read six or more colors. No part of the instrument touches the sample, so nothing carries from one cartridge to the next.

44 — One valve routes everything

Primers and probesHeatReadoutFocal detail

The 2005 paper reported that the system extracted RNA in just over six minutes with about 70 % of the yield of a conventional laboratory kit, and completed extraction, reverse transcription and real-time PCR in about 35 minutes, with 95 % of replicate Ct values expected within about 1.5 cycles either side of the average. Its SARS-CoV-2 test, authorized in March 2020, detected two targets and carried two built-in checks: a sample-processing control that showed extraction and amplification worked, and a probe-check control that verified the dried reagents had dissolved, the reaction tube had filled and the probes were intact before cycling began. It could also stop early and report a strongly positive sample before the full run finished.

Two stages in a pouch

The FilmArray system, described in 2011, takes a different route and runs many tests at once. Its pouch has a rigid plastic fitting with reservoirs for reagents and a flexible film with blisters in which the work happens. The instrument moves liquid between blisters by pressing on them with pneumatic actuators. Ceramic beads beaten against the sample break open the organisms; magnetic beads coated with silica capture the nucleic acid, are washed, and release it. A first stage of PCR amplifies all the targets together in one blister for 26 cycles of 4 seconds at 94 °C and 19 seconds at 60 °C. The product is then diluted into an array of 102 wells of about one microliter each, every well pre-loaded with primers for one target, and a second stage of 30 cycles runs inside each well with primers that bind within the first-stage product, a nested PCR that adds specificity. A dye and a melt curve in each well identify the products. Two Peltier devices, the pneumatic actuators, a blue LED and a camera are the instrument's essential hardware. The whole process takes about an hour.

45 — Two stages in a pouch

CopiesPrimers and probesHeatReadoutFocal detail

What integration costs

Putting a laboratory into a cartridge moves the difficulty rather than removing it. Liquids must be moved reliably in small volumes without bubbles, without leaks and without leaving residue that carries over between steps. Reagents must survive storage at room temperature, which is why they are freeze-dried, and dissolve completely and uniformly at the right moment. The thermal design must suit a small volume in a thin chamber, and the optics must read through molded plastic. Controls must be built into every cartridge, because no technician watches each step. And the cartridge must be manufactured in large numbers at low cost with every chamber, valve and seal within tolerance, because one bad cartridge is one wrong result.

A 2012 review of point-of-care microfluidic devices observed how few of the many technologies demonstrated in research had reached the market, and argued for designing integrated, real-world systems rather than optimizing single components. Integration is where these products succeed or fail.

Not every fast molecular test is PCR. One widely used point-of-care COVID-19 test, which reported results in 13 minutes or less, amplifies its target with an isothermal method, at a single temperature, with no thermal cycling at all. Chapter 24 explains what such methods give up in exchange for removing the cycle.

+ The questionHow do you show that every part, and the whole, does what it was specified to do?

Two questions on the right side of the V

The right side of the V in Chapter 16 asks two different questions. Verification asks whether the product was built as specified: does each part, and the assembled whole, meet its written requirements? Validation asks whether the specified product is the right one: does it meet the user's needs in the intended use? For a PCR instrument and its test, verification runs from the smallest part up to the integrated system; validation shows that the test gives correct results on real specimens. The two meet in the analytical studies of the assay on the finished system, which belong to both.

46 — What is verified, and how

Focal detail

Subsystems first

Each subsystem is verified against its own requirements, by methods that measure the property directly.

The thermal subsystem is measured in the wells with traceable probes (Chapter 10): accuracy and uniformity at each set point, overshoot and undershoot, stability during holds, and ramp times under the conditions the specification states, with the plates and volumes the assays will use. The optical subsystem is verified for its ability to locate each well, its background, its uniformity across the plate, its response to known amounts of each calibrated dye, and the separation of dyes in neighboring channels. Motion is verified for positioning and repeatability, electronics for noise and for the drive currents the modules need, firmware for the timing of events within each cycle, so that the read falls where it should every time.

Software is verified as carefully as hardware, because it computes the result. The FDA's guidance on the content of premarket submissions for device software functions, finalized in June 2023, applies to device software including firmware and software that controls a device. It asks for documentation at one of two levels, a basic level, or an enhanced level where a software failure could present a probable risk of death or serious injury, and it allows a manufacturer to declare conformity with the recognized software life-cycle standard, IEC 62304, as part of that documentation. The current edition of IEC 62304 dates from 2006 with an amendment in 2015; a second edition is in development and was not expected before 2028. In the European Union, the IVD Regulation's general requirements include requirements for software, and following IEC 62304 is a common way to meet them.

Then the system

Assembled subsystems can pass every test and still fail together, which is why the integrated instrument is verified as a whole. The reference-plate test of Chapter 15, an installation and periodic check, shows the kind of whole-system test needed: distinguishing 5,000 from 10,000 copies of a human gene with 99.7 % confidence, exercises the thermal uniformity, the optics, the calibrations and the analysis software in one run. Beyond it, the assay's analytical studies are run on the finished system: the limit of detection, precision across runs, days, operators, reagent lots and instruments, the ability to tell targets from near relatives, and carry-over, whether a strongly positive sample can contaminate the negative sample run after it. The FDA's guidance on moving assays between instruments asks for carry-over studies in every migration of a nucleic acid test for exactly this reason (Chapter 17).

Risk: how a wrong number becomes harm

A diagnostic instrument rarely hurts anyone directly. It can harm through its result: a false negative that sends an infected person home untreated, a false positive that leads to an unnecessary treatment or isolation, a wrong quantity, a result that arrives too late to matter. Risk management for a diagnostic device therefore traces each possible fault forward to a wrong result and on to its consequence. The international standard for medical device risk management, ISO 14971:2019, applies, and its companion guidance, ISO/TR 24971:2020, includes an annex on applying it to diagnostic devices; that guidance has been marked for revision.

47 — From a fault to a harm

HeatFocal detail

A worked chain shows the method. A fault: the block runs cold in one corner at the denaturing step. In those wells some target strands do not separate and amplification is delayed or fails, so a specimen with little virus crosses the threshold too late or not at all: a false-negative result. A hazardous situation: that result is reported and acted on. A harm: a missed diagnosis. Controls can act at each step. The design adds edge heaters and guard zones. Verification measures the corners at installation and at intervals. Detection catches gross failures: the positive control fails if it sits in an affected well, and a specimen whose internal control, the human gene amplified in every specimen, is not detected before the cut-off is reported invalid rather than negative; a modest delay that leaves the internal control inside the cut-off passes unnoticed. And information tells the user the limits of a negative result. The FDA's special controls for multi-target tests that include SARS-CoV-2 require labeling stating that a negative result does not rule out infection and that sequence variants can cause incorrect results, and continuous monitoring for new variants: results suggesting a variant affects the stated performance must be sent to the agency immediately, and the evaluations submitted within 48 hours if the FDA asks for them.

Safety, electromagnetic compatibility and the user

A laboratory instrument must also be safe to use and live peacefully with other equipment. The IEC 61010 family covers the safety of electrical equipment for measurement, control and laboratory use, against shock, fire, heat and mechanical hazards: Part 1 gives the general requirements, in an edition of 2010 amended in 2016 and consolidated in 2017, and a second amendment had reached its final stage in 2026. Particular parts add requirements for equipment that heats materials (Part 2-010), for automatic laboratory analyzers (Part 2-081) and for in vitro diagnostic equipment (Part 2-101, 2018). Electromagnetic compatibility, the instrument's ability to work without disturbing other equipment or being disturbed by it, is covered by IEC 61326-1, with a particular part for diagnostic equipment, IEC 61326-2-6, published in a new edition in 2025. Which parts apply to a given instrument is a judgement made from its functions; one manufacturer states that its air-heated cycler conforms to Parts 1 and 2-101 of IEC 61010. Usability engineering, IEC 62366-1, addresses the errors users can make, such as entering the wrong volume or loading a plate the wrong way round, and the labeling standard ISO 18113-3 covers the information supplied with diagnostic instruments for professional use.

48 — The standards around one instrument

Focal detail

Production and the field

Verification does not end when the design is approved. Every unit built is tested before it ships, its thermal and optical performance checked against limits; it is calibrated at installation; and it is checked at intervals in use. One accredited calibration service recommends calibration once a year or every 250 runs. Parts with finite lives, the Peltier modules, the fan, the lamp, are inspected or replaced on a schedule, and complaints and service records are collected and analyzed for trends. All of this happens inside a quality management system. ISO 13485:2016 was confirmed without change after review in October 2025, and in the United States the Quality Management System Regulation, in force since February 2026, incorporates it; on the same date the FDA retired its long-standing inspection technique and began inspecting under a new compliance program.

+ Part VI · The test

From a specimen to a trusted result.

An instrument and an assay that work in the development laboratory are not yet a test. These four chapters follow the steps that make them one: reading RNA with a method that copies DNA, getting nucleic acid out of a specimen intact and free of inhibitors, the controls and limits that make a result believable, the CDC's test as it was designed, made and shipped in the first weeks of 2020, and what regulators in the US, the EU and India require before a laboratory may report results to patients.

+ The questionMost of the viruses that matter carry RNA. How does a method that copies DNA read them?

A step before the first cycle

Taq polymerase copies DNA. It needs a DNA template, and it does not copy RNA usefully. Yet many of the targets that matter most in diagnostics are RNA: the genomes of SARS-CoV-2, influenza, respiratory syncytial virus, HIV and hepatitis C virus are all made of RNA, and the activity of a gene is measured through its messenger RNA. Reading them with PCR needs one extra step at the start: copying the RNA into DNA.

The enzyme that does this was discovered in 1970, independently by Howard Temin and Satoshi Mizutani and by David Baltimore, in viruses whose RNA genomes must be copied into DNA to infect a cell. It is called reverse transcriptase, because it runs the usual flow of genetic information backward, from RNA to DNA. Given an RNA template and a primer, it builds a strand of complementary DNA, called cDNA, which PCR can then amplify. The enzymes used in laboratories come from two other retroviruses, avian myeloblastosis virus and Moloney murine leukemia virus, and their engineered descendants. They work at moderate temperatures, around 40 to 50 °C for the original enzymes, and higher temperatures help them read RNA that folds back on itself.

49 — From RNA to cDNA

CopiesPrimers and probesRNAHeatFocal detail

One tube or two

The reverse transcription and the PCR can be run in separate tubes or in one. In a two-step method, the RNA is first copied into cDNA in its own reaction, with its own buffer, and portions of that cDNA are then used in separate PCRs. Each step can be optimized, and the cDNA can be stored and used for several tests. In a one-step method, both enzymes and both sets of conditions share one tube: the reaction is held at the reverse-transcription temperature for some minutes, then heated, and the PCR cycles begin. Fewer steps mean less handling and less risk of contamination, which is why diagnostic tests are almost always one-step.

The reverse transcriptase needs a primer too. In a one-step diagnostic test, the PCR's own reverse primer usually serves: it binds the viral RNA and is extended into cDNA, which the forward primer then copies in the first PCR cycle. The CDC test was one-step. With the TaqPath mix, its profile began with 2 minutes at 25 °C, a step for the contamination control described at the end of this chapter, then 15 minutes at 50 °C for reverse transcription, then 2 minutes at 95 °C to activate the polymerase, before its 45 cycles.

The test's two viral targets lay in the gene for the nucleocapsid protein, the N gene. The CDC's paper describing the test explained one reason for the choice: during infection, the virus makes many extra RNA copies of its N gene region, which may make the target more abundant than other regions and the test more sensitive.

RNA is fragile, and enzymes that destroy it are everywhere

RNA is easier to destroy than DNA, and the enzymes that destroy it, ribonucleases or RNases, are everywhere: in tears, saliva, mucus and sweat, on skin and hair, and made by bacteria and fungi. A supplier's technical note on RNase control describes them as small, compact proteins held together by several internal cross-links, so robust that they refold after being heated and keep substantial activity even after autoclaving. A specimen of viral RNA in a swab, a tube touched with a bare hand or a pipette tip that was not certified free of RNase can lose its target before the test begins. Hence gloves, certified plastics, inhibitors of RNase in the reaction mix, and keeping specimens cold.

Getting the nucleic acid out

Between the swab and the reaction comes extraction. Its jobs are to break open the virus and the cells, to separate the nucleic acid from everything else in the specimen, including RNases and substances that would slow the polymerase, and often to concentrate it. The most common method uses a strong salt solution of guanidinium to break open the particles and denature proteins, binds the nucleic acid to silica, on a small column or on magnetic beads, washes the silica with buffers containing ethanol, and releases the purified nucleic acid in a small volume of water or buffer.

50 — Extraction on silica

RNAFocal detail

Extraction is part of the test's performance, not a preliminary to it. The CDC test's instructions listed the extraction kits and automated systems it was validated with, and its limit of detection differed between them: about 3 RNA copies per microliter of the spiked sample with one automated system and 1 copy per microliter with a manual kit. The instructions also allowed a simple heat treatment in place of extraction, with a warning that it had been tested on few clinical specimens and that some loss of sensitivity could not be ruled out.

Extraction can also add a problem of its own. A 2020 study showed that traces of the extraction kits' buffers carried into the reaction delay amplification: about 10 % of the ethanol wash buffer, or as little as 0.3 to 1 % of the guanidinium lysis buffer, delayed the Cq by at least half a cycle.

Inhibitors

Many substances in clinical and environmental samples slow or stop the reaction. A review classifies them by where they act: on the nucleic acid, on the polymerase, on primer binding, or on the fluorescence itself.

An inhibitor lowers the efficiency, which delays the curve and, if strong enough, stops it. In a test without an internal control, that looks exactly like a negative result. Additives can help: in a study of blood inhibitors, bovine serum albumin was the most effective at relieving them.

Keeping old product out of new reactions

Every positive PCR makes billions of copies of its target, and a laboratory that runs thousands of reactions accumulates them on benches, pipettes and gloves. A trace of old product in a new reaction is a perfect template and gives a false positive. Laboratories separate the rooms where reactions are prepared from the rooms where amplified product is handled, and closed real-time tubes that are never opened help. There is also a chemical defense, described in 1990: run every PCR with dUTP, a building block carrying uracil, in place of dTTP, so that every product contains uracil instead of thymine. Before each new PCR, an enzyme called uracil-DNA glycosylase, UNG, is allowed to act: it removes uracil from any DNA, which damages carried-over product so that it can no longer be copied, while the specimen's own DNA, which contains thymine, is untouched. The enzyme is then inactivated by heat, before the new uracil-containing copies are made. The 2 minutes at 25 °C at the start of the CDC test's TaqPath profile is that step.

51 — Destroying old product

Template DNACopiesHeatFocal detail

+ The questionWhat turns a reaction that works into a test whose result can be trusted?

Every result carries its own evidence

A single PCR that lights up says only that something amplified. A test result must say more: that the right target amplified, that nothing else did, that a negative is a real absence and not a failure, and that the test would have detected the target at the concentrations that matter. Each of those claims is supported by a specific piece of evidence, some of it built into every run as controls, some of it established once, before the test is used, as its performance characteristics.

Controls: one question each

Controls are reactions, or targets within reactions, whose expected result is known. Each answers one question about the run.

A positive control contains the target at a known, modest level. If it fails, the reagents or the instrument failed, and no negative in that run can be believed. A no-template control contains everything but the sample. If it amplifies, something in the run is contaminated, and no positive can be believed. An extraction control is a specimen-like sample carried through extraction alongside the patients' specimens; in the CDC test it was a human specimen control, which had to be negative for the viral targets and positive for the human gene, showing that extraction worked and introduced no viral target. An internal control is amplified in every patient's reaction: in the CDC test, the human gene RNase P, present in the cells any good swab collects. It shows that the specimen contained human material, that extraction recovered it and that amplification worked in that very well.

52 — One question per control

ReadoutFocal detail

The CDC's rules combined them. A run was valid only if the positive control was positive for all three targets, the no-template control showed no amplification at all and the human specimen control behaved as expected. Within a valid run, a specimen with both viral targets detected was positive, whether or not RNase P was detected; a specimen with neither viral target and RNase P detected was negative; a specimen with only one viral target was inconclusive and retested; and a specimen with nothing detected, not even RNase P, was invalid, to be extracted again or collected again.

The limit of detection

The most important performance characteristic of a detection test is how little target it can detect. Laboratory medicine defines it carefully. A guideline from the Clinical and Laboratory Standards Institute, EP17-A2, whose second edition was published in 2012 and reaffirmed in 2017, distinguishes the limit of blank, the highest result expected from a sample with no analyte, from the limit of detection, the lowest amount that can be reliably told apart from it. For qualitative molecular tests the limit of detection is usually defined by hit rate: the lowest concentration at which at least 95 % of replicates are positive. In practice, developers test 20 replicates and require at least 19 positive.

53 — The limit of detection

Template DNAReadoutFocal detail

The CDC test established its limit by spiking a synthetic copy of the N gene's RNA into a suspension of human cells in transport medium, extracting it and testing replicates at a series of concentrations. Its instructions give the limit as about 3 RNA copies per microliter of the spiked sample with one extraction system and 1 copy per microliter with another; the CDC's paper describes it as about 5 RNA copies per reaction, which corresponds to the lower figure. That figure sits just above the sampling floor of Chapter 6, which is as good as any PCR can be. Below the limit, detection does not stop abruptly; it becomes hit-or-miss, and that is where inconclusive results, with one target detected and not the other, come from.

Inclusivity, exclusivity and interference

A test must detect every version of its target and nothing else. Inclusivity is the first half: the primers and probe must match all known variants of the organism. During the pandemic, the FDA's template for emergency submissions expected developers to show, by aligning their oligonucleotides against published sequences, that all published SARS-CoV-2 sequences would be detected, and to keep checking as new variants appeared. Exclusivity, or cross-reactivity, is the second half: the test must not react with related organisms or with others likely to be in the same specimen. The same template flagged any similarity above 80 % between an oligonucleotide and another organism for laboratory testing, and listed organisms to test at high concentrations. The CDC's paper reported no cross-reaction with the four common human coronaviruses or with MERS coronavirus. Its third target, N3, had been designed to detect a broader group of related coronaviruses and did react with the 2003 SARS virus, as intended. Interference testing, guided by another CLSI document, EP07, checks that substances likely to be in specimens, blood or mucus, do not change results.

Clinical performance, and why prevalence changes everything

Analytical performance describes the test on prepared samples. Clinical performance describes it on patients. Against a reference standard that establishes who truly has the condition, a test's sensitivity is the fraction of true cases it calls positive, and its specificity the fraction of non-cases it calls negative. Often no perfect reference exists, and the new test is compared with another test instead. The FDA's statistical guidance of 2007 insists that such comparisons be reported as positive percent agreement and negative percent agreement, not as sensitivity and specificity, because agreement with an imperfect comparator is not accuracy, and that each be reported with a 95 % confidence interval. The FDA's emergency template for SARS-CoV-2 molecular tests asked for at least 30 positive and 30 negative clinical specimens and agreement of at least 95 % in each.

What a clinician needs, though, is the probability that a positive result is right, the positive predictive value, and that depends on how common the condition is among the people tested.

54 — The same test in two populations

ReadoutFocal detail

Where false results come from

Every false result has a mechanism, and each mechanism has a defense described somewhere in this guide.

False positives come from contamination, by old product (Chapter 20) or by synthetic material made in the same place as the reagents; from cross-reaction with a related organism (Chapter 5); from oligonucleotides that react with each other in every tube, as the CDC's N3 did; and from mix-ups of specimens or wells. The no-template control catches some of these, and design and laboratory practice must prevent the rest.

False negatives come from a poor specimen, a swab that collected little or was taken too early or too late in the infection; from losses in extraction, degraded RNA and inhibitors (Chapter 20); from a target concentration below the limit of detection; from a change in the organism's sequence under a primer or probe, which is why the FDA's special controls for SARS-CoV-2 multi-target tests require ongoing monitoring of variants; and from instrument faults that leave wells too cool or misread (Chapters 10 and 19). The internal control catches several of these by turning them into invalid results.

Laboratories that run such tests work to their own standards: the international standard for medical laboratories, ISO 15189:2022, a quality practice guide specifically for nucleic acid amplification tests for microbial pathogens, ISO 17822:2020, and CLSI documents on validating multiplex molecular tests and on setting up molecular testing environments.

+ The questionWhat happened when one PCR test had to be designed, made and shipped across a country in weeks?

Weeks, not years

On 10 or 11 January 2020, depending on the time zone, the first genome sequence of the new coronavirus was posted publicly on behalf of a consortium led by Yong-Zhen Zhang of Fudan University. Everything in this guide follows from what happened next: a sequence is all a PCR test needs to begin. Within days, a group at the Charité hospital in Berlin circulated a protocol, and on 23 January it published a workflow in the journal Eurosurveillance with targets in three viral genes, designed from the sequence and from related coronaviruses because no samples of the new virus were yet available to them.

The US Centers for Disease Control and Prevention designed its own test. It chose two targets in the N gene, called N1 and N2, designed to be specific to the new virus, and a third, N3, designed to detect a broader group of related coronaviruses. A fourth reaction detected the human gene RNase P as a control. On 4 February 2020 the Secretary of Health and Human Services declared that circumstances justified the emergency authorization of diagnostic tests, and the same day the FDA issued an Emergency Use Authorization to the CDC for its panel. Shipping began the next day. About 200 kits were to go to US laboratories at first and a similar number to selected laboratories abroad, each enough for roughly 700 to 800 specimens; 115 US and 191 international laboratories were qualified to order them.

55 — Weeks, not years

RNAFocal detail

The test as designed

The panel used everything earlier chapters describe. It was a one-step real-time RT-PCR (Chapter 20) in 20-microliter reactions, 15 microliters of mix and 5 of extract. Each target had a hydrolysis probe carrying the dye FAM and a quencher (Chapter 12), and because all the probes used the same dye, each specimen needed a separate well for each target. The profile was a reverse-transcription step, a polymerase activation step and 45 cycles of 3 seconds at 95 °C and 30 seconds at 55 °C, with fluorescence read at 55 °C (Chapter 4). It ran on one named instrument, with one version of its software, in its standard mode, without a reference dye (Chapter 17). A target was detected if it crossed the threshold before cycle 40.00 (Chapter 13). Controls in every run and the internal control in every specimen governed the calls (Chapter 21). Its limit of detection was about five copies per reaction. It was authorized for laboratories certified for high-complexity testing, and the CDC estimated the process took about four hours from the start of sample processing to a result.

56 — The CDC panel on a plate

RNAFocal detail

What went wrong

Within days of the first shipments, public health laboratories reported problems: no-template controls were giving signals, mainly for N3 and to a lesser extent N1, and runs had to be called inconclusive. The CDC's own analysis of the first lot of kits, published in the journal PLoS ONE in December 2021, found two separate faults.

The first was contamination. About 2 % of the no-template controls for N1 from the initial lot showed false reactivity. Sequencing traced it to a synthetic template that the CDC had made at about the same time as the kits, which had reached the bulk reagent during quality assessment. The problem did not recur in later lots.

The second was design. The 3′ end of the N3 probe could pair with the 3′ end of the N3 reverse primer. In a tube with no target, the two paired, and the CDC's sequencing found duplex and larger products containing the N3 probe sequence, which gave fluorescence with no template (Chapter 5). False signals appeared in about 97 % of the N3 no-template controls made with the authorized kits' reagents, against 0.5 to 2 % with oligonucleotides from earlier batches and commercial suppliers, and they became more frequent as the reagents aged. The flaw was present in all three sources of the oligonucleotides; how often it showed depended on the batch. The authors concluded that flaws in both the assay's design and the handling of bulk material caused the first lot's problems.

57 — The N3 flaw

CopiesPrimers and probesReadoutFocal detail

A report by the Inspector General of the Department of Health and Human Services, published in October 2023, found the circumstances that allowed this: inadequate control of documents, about four different versions of quality-control procedures in use, a lack of separation of duties, viral material and test reagents made in the same core laboratory, and weak oversight. It made six recommendations, all of which had been closed as implemented by January 2025. According to NPR's November 2020 report on a CDC internal review, a final quality-control test had indicated that the kit could fail about a third of the time, and it was released anyway. The same Inspector General's report also found that the CDC had a viable test within about two months of the genome's publication.

The fix

The repair came quickly once the causes were understood. On 26 February the CDC validated a protocol that left out the third target and used only N1 and N2. By 28 February it was distributing a working kit to public health laboratories. On 29 February the FDA issued a policy allowing laboratories certified for high-complexity testing to use tests they had validated themselves while their requests for authorization were reviewed, which let other laboratories bring their own tests into service. On 15 March the FDA reissued the CDC's authorization without N3, accepting probes with an additional internal quencher and reagents made by commercial suppliers. The CDC's paper describing the test reported that testing with N1 and N2 alone agreed completely with the original three-target interpretation.

Further revisions followed as the test was used: three more master mixes on 30 March 2020, more extraction systems in June and July, pooling of up to four specimens in December 2020, and later home-collected nasal swabs. On 2 July 2020 the CDC received authorization for a multiplex test that detected influenza A, influenza B and SARS-CoV-2, with the human control, in one well with four dyes. For that test, its 2021 analysis noted, the CDC used a stricter review process, with outside experts and a pilot in three public health laboratories.

Afterward

In July 2021 the CDC told laboratories that it would ask the FDA to withdraw the authorization of its original panel after 31 December 2021 and encouraged them to move to a multiplex test that could also detect influenza; in August it clarified that there were no concerns about the panel's performance and that its primer and probe designs would remain public. The FDA's list of authorized tests still showed the panel in August 2026, with its most recent letter dated 7 March 2023. On 29 June 2026 the Secretary of Health and Human Services terminated the 2020 declaration that justified emergency authorization of COVID-19 diagnostic tests, effective 26 December 2026; from that date, the remaining COVID-19 test authorizations cease to be in effect.

What the case shows

Nothing in the first weeks went wrong with PCR's chemistry. The cycle copied, the probes reported, the instrument heated and read. What failed was the system around the chemistry: a design check of how the oligonucleotides interacted with each other (Chapter 5), separation of the work that made synthetic targets from the work that made reagents (Chapter 20), and quality control at release strong enough to stop a faulty lot. Even the conditions of the emergency authorization pointed at those layers: the FDA waived most of its manufacturing requirements for the kit but kept those for acceptance activities, for nonconforming product and for statistical techniques, the parts that decide whether a lot is fit to ship.

The case also shows the reach of the test system. Because the authorization named one instrument and particular reagents and extraction methods, laboratories with other equipment could not simply run it, and each extension of the list took data and a revised letter (Chapter 17). Speed came from the emergency route's flexibility; safety had to come from the same design controls, risk management and quality system that apply to any test (Chapters 16 and 19), applied under pressure.

+ The questionWhat must a PCR test and its instrument show before laboratories may use them on patients in the US, the EU and India?

Three questions every regulator asks

The rules differ between jurisdictions, but the evidence they ask for answers three questions, in order. The European Union's regulation for diagnostic devices names them precisely. Scientific validity: is the thing measured, the analyte, associated with the clinical condition? For a SARS-CoV-2 test, is the virus's RNA in a respiratory specimen associated with infection? Analytical performance: does the device correctly detect or measure the analyte? This is the limit of detection, the specificity, the precision and the rest of Chapter 21, established on the complete test system, instrument included. Clinical performance: do the device's results correspond to the clinical condition in the people it is meant for, as sensitivity, specificity and predictive values in the intended population? A further question, whether using the test improves outcomes for patients, its clinical utility, matters to doctors and to those who pay for tests, and is generally not what market authorization requires.

58 — Three questions regulators ask

Focal detail

The status of everything below is stated as of October 2026.

United States

In the United States the FDA regulates diagnostic devices by risk class and product type, and a separate law, the Clinical Laboratory Improvement Amendments (CLIA), regulates the laboratories that run them.

Instruments. Real-time nucleic acid amplification systems are classified under the regulation for instrumentation for clinical multiplex test systems, 21 CFR 862.2570, as class II devices with special controls, and they are now exempt from premarket notification, subject to the general limits on exemptions. The instrument named in the CDC test was cleared through a premarket notification, a 510(k), in September 2008, before that exemption.

Tests. A test is authorized for its intended use with specified instruments. For tests that detect SARS-CoV-2 together with other respiratory pathogens, the FDA created a classification, 21 CFR 866.3981, through the De Novo route: it granted a De Novo request for a respiratory panel from BioFire Diagnostics on 17 March 2021, the first traditional marketing authorization of a SARS-CoV-2 test, and the classification was codified in August 2024. Its special controls require, among other things, limit-of-detection, cross-reactivity and clinical studies, labeling stating that a negative result does not rule out infection and that sequence variants can cause incorrect results, and continuous monitoring of viral variants, with results suggesting an effect on performance sent to the FDA immediately and evaluations submitted within 48 hours if the agency asks. Later tests of the same type reach the market through 510(k)s that cite earlier ones.

Emergency use. Under section 564 of the Federal Food, Drug, and Cosmetic Act, the FDA may authorize an unapproved test during a declared emergency if the disease is serious, the test may be effective, its known and potential benefits outweigh its risks, and no adequate, approved and available alternative exists. An authorization lasts until the declaration ends or the FDA revokes it. The COVID-19 declaration for diagnostic devices, made on 4 February 2020, was terminated by the Secretary of Health and Human Services on 29 June 2026, effective 26 December 2026; the notice observed that more than fifty COVID-19 tests now hold traditional marketing authorization. The FDA's transition guidance of 2023 expects manufacturers who want to keep supplying an emergency-authorized test to have a marketing submission accepted for review before the termination takes effect.

Laboratories. CLIA sorts tests into waived, moderate-complexity and high-complexity categories. Waived tests meet separate criteria of simplicity and low risk of error; the rest are scored on seven criteria, such as the knowledge, training and judgment a test demands; a total of 12 or less is moderate complexity, above 12 high. The FDA categorizes commercially marketed tests. The CDC panel was authorized only for high-complexity laboratories.

Laboratory-developed tests. In May 2024 the FDA issued a rule stating that tests made and used within a single laboratory are devices, and phasing out its long-standing practice of not enforcing device requirements on them. On 31 March 2025 a federal district court in Texas vacated the rule, holding that such tests are professional services rather than the products the device provisions of the law cover. On 19 September 2025 the FDA removed the rule's language from its regulations. Laboratory-developed tests remain subject to CLIA.

Quality system. Since 2 February 2026, the FDA's Quality Management System Regulation has incorporated ISO 13485:2016 by reference, so design controls, risk management and production controls for diagnostic manufacturers follow the international standard, with some FDA-specific additions.

European Union

In the EU, the In Vitro Diagnostic Medical Devices Regulation (EU) 2017/746, the IVDR, has applied since 26 May 2022. It classifies devices into four classes, A to D, by seven rules in its Annex VIII, and the class decides who assesses conformity: class A devices that are not sterile are self-declared by the manufacturer, while classes B, C and D need a notified body.

Instruments. Rule 5 puts instruments that the manufacturer specifically intends for diagnostic procedures in class A. The classification guidance of the Medical Device Coordination Group, MDCG 2020-16, gives a PCR thermal cycler, a sequencer and an instrument for automated nucleic acid purification and PCR set-up as examples.

Tests. Rule 1 puts in class D, among others, devices that detect a transmissible agent causing a life-threatening disease with a high or suspected high risk of spreading, and SARS-CoV-2 tests were long classified there. In March 2025 a revision of MDCG 2020-16, based on expert-panel advice that SARS-CoV-2 no longer posed a life-threatening risk with significant mortality to the general EU population, removed it from those examples: SARS-CoV-2 tests for professional use are now treated as class B under the general Rule 6, and self-tests as class C. The guidance is not legally binding, and notified bodies published a position in December 2025 on how to handle devices already certified as class D. Common specifications, Implementing Regulation (EU) 2022/1107, set detailed performance requirements for certain class D devices, including an annex on SARS-CoV-2; whether that annex still applies to tests now treated as class B had not been clarified. EU reference laboratories, the first of which began work in October 2024, verify the performance of class D devices and test batches.

Transition. Under Regulation (EU) 2024/1860, devices holding a certificate under the previous directive may stay on the market until 31 December 2027. Devices that needed no notified body before but do now may stay until 31 December 2027 for class D, 31 December 2028 for class C and 31 December 2029 for class B and sterile class A devices. Both depend on the manufacturer having an IVDR quality system and having applied to, and signed an agreement with, a notified body by set dates. A Commission proposal of December 2025 to simplify the medical device and diagnostic regulations had not been adopted as of October 2026; it was still at committee stage in the European Parliament.

India

In 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 (CDSCO) publishes the classification of individual device types. Its list, updated on 25 October 2023, places a nucleic acid amplification (PCR) analyzer in class A and COVID-19 RT-PCR and isothermal kits in class C, as it does RNA and DNA extraction kits. Licensing became mandatory for classes A and B from 1 October 2022 and for classes C and D from 1 October 2023. Manufacturers of class A and B devices apply to the state licensing authority (Form MD-3, licence on Form MD-5); manufacturers of class C and D devices to the central licensing authority (Form MD-7, licence on Form MD-9); importers of any class on Form MD-14, with the licence on Form MD-15.

India adds a step the other two jurisdictions do not have in the same form. New diagnostic devices in classes B, C and D need a performance evaluation by a testing laboratory set up or designated by the central government, or accredited by the national accreditation board, on three batches made from different lots of key raw materials. During the pandemic, guidelines from the Indian Council of Medical Research of 4 June 2020 required the first batch of most RT-PCR kits to be validated at one of 24 designated centers before approval; kits authorized by the US FDA were exempt (the guideline's words were "US-FDA approved"; in 2020 these were emergency use authorizations). The council and the CDSCO have since begun publishing standard evaluation protocols, the first, for a real-time PCR test, released for comment in January 2025.

59 — One test, three regulators

Focal detail

One evidence core, three presentations

Set side by side, the three systems classify the same objects differently. The instrument is class A in the EU and in India and class II but exempt from premarket review in the United States. A SARS-CoV-2 laboratory test is now class B in the EU, class C in India and, as part of a multi-target respiratory test, class II in the United States. The routes, the documents and the assessors differ.

The evidence underneath them barely does. Scientific validity, analytical performance on the complete test system, clinical performance with confidence intervals, a risk file, software documentation, safety and compatibility test reports for the instrument, and a quality management system to ISO 13485 serve all three. A developer who builds that core once, organized by the system's requirements (Chapter 19), presents it three ways; one who builds it three times, for three regulators, pays three times and risks three inconsistent answers.

60 — EU deadlines

Focal detail

+ Part VII · What comes next

The cost of the cycle, and what removes it.

PCR's costs are time, a laboratory and an instrument that can move heat quickly and evenly. This last chapter looks at methods that cut those costs, faster cycling, amplification at a single temperature and enzymes that detect sequences directly, and asks what each gives up, before setting out what remains unsolved.

+ The questionIf PCR's costs are time, a laboratory and an instrument, what do faster and simpler methods give up to remove them?

Faster cycling

The cycle takes time because heat takes time to move (Chapters 9 and 10). Shrink the volume and the distance heat must travel, raise the concentrations so that each step needs less time, and the cycle shortens. A 2015 study by Carl Wittwer's group, called extreme PCR, raised primer and polymerase concentrations 10 to 20 times above normal, used samples of 1 to 5 microliters and cycles of 0.4 to 2 seconds, and amplified short human DNA targets of 45 to 102 base pairs through 35 cycles in 15 to 60 seconds, at efficiencies of about 92 to 96 %. A 2015 demonstration of photonic PCR heated a thin gold film under its wells with blue light and completed 30 cycles in 5 minutes. A 2019 commentary described the pairing of extreme PCR with rapid melting analysis as a step toward molecular diagnosis while the patient waits.

What faster cycling gives up is visible in the figures: short targets, small volumes and much more reagent per reaction. A smaller volume holds fewer target molecules at a given specimen concentration, which brings the sampling floor of Chapter 6 closer, and high reagent concentrations raise the cost and the risk of side reactions. Fast cycling is also only as good as the temperature control that delivers it; a three-second hold means nothing if the liquid never arrives.

Amplifying at one temperature

The other route removes the cycle altogether. Isothermal amplification methods copy nucleic acid at a single temperature, using enzymes that separate strands as they copy instead of relying on heat to separate them.

Loop-mediated isothermal amplification, LAMP, published in 2000 by Tsugunori Notomi and colleagues, runs at about 65 °C with a polymerase that displaces the strand ahead of it as it copies, and with four primers that recognize six distinct regions of the target. The primers are designed so that the products fold into loops that serve as new starting points, and the reaction accumulated about a billion copies in under an hour. Recombinase polymerase amplification, RPA, published in 2006, uses proteins borrowed from the machinery cells use to repair and recombine DNA to push primers into double-stranded DNA at 37 °C, and detected fewer than ten copies of genomic DNA in under 30 minutes. Other isothermal schemes use enzymes that nick one strand to start each round of copying. One widely used point-of-care COVID-19 test amplified its target isothermally and reported results in 13 minutes or less.

61 — Removing the cycle

Template DNACopiesPrimers and probesHeatFocal detail

Removing the cycle removes the instrument's hardest job: the box needs only to hold one temperature, which a simple heater can do, so isothermal tests suit small, cheap, battery-powered readers. The price is paid in the chemistry. Many primers in one tube give many chances for primers to interact, and isothermal reactions are prone to amplifying something even without a target. Without cycles, there is no Cq built on doubling per cycle, so quantification is harder. And the design of four to six primers per target constrains where on a genome a test can be placed.

Enzymes that recognize a sequence

A third approach adds sequence recognition by the bacterial immune enzymes known as CRISPR systems. Some of them, guided by a short RNA that matches a target, become indiscriminate cutters once they bind their target and cut any nearby single-stranded nucleic acid. Mixed with a reporter molecule that fluoresces when cut, they turn recognition of one sequence into a signal. The method called SHERLOCK, published in 2017, used the enzyme Cas13 after an RPA pre-amplification and reported sensitivity down to attomolar concentrations with the ability to distinguish single-base differences; DETECTR, published in 2018, used Cas12a and distinguished two closely related types of human papillomavirus. In May 2020 the FDA issued an emergency authorization for a CRISPR-based SARS-CoV-2 test, the first authorized use of the technology.

62 — What each method gives up

RNAHeatFocal detail

CRISPR detection is recognition, not amplification: on its own it is not sensitive enough for a few copies, so it is usually paired with an isothermal amplification step, and it inherits that step's strengths and weaknesses. What it adds is a second check of sequence identity, like a probe's, in a format that can be read by a simple fluorescence reader or a paper strip.

Where PCR stays

None of this has displaced PCR, and the reasons are the subject of this guide. PCR's chemistry is understood in detail, its primers are simple to design, its exponential phase gives a Cq that can be turned into a quantity, its digital form counts molecules, and its instruments, assays, controls and regulatory routes are mature. Real-time PCR runs most laboratory molecular diagnostics; sample-to-answer cartridges have brought it close to the patient; digital PCR measures rare variants in blood; and PCR prepares the DNA for most sequencing. The newer methods find their places where time, cost or portability matter more than quantification and the breadth of established tests.

What remains unsolved

Several problems run through every method in this chapter, and none is solved. Sample preparation remains the slowest, least standardized and least integrated step outside the cartridges of Chapter 18: amplification takes minutes, but getting clean nucleic acid out of a specimen takes skill or an expensive disposable. Comparability remains elusive: a Ct from one test cannot be compared with a Ct from another (Chapter 13), and the field's move toward vendor-neutral raw data and efficiency-corrected quantities is recent. Evolving targets never stop: every test that detects a virus must be checked against each new variant for as long as it is used. And cost per result still separates what is technically possible from what a laboratory, a clinic or a health system can provide at scale.

The central question asked how an instrument that does little more than heat and cool a tube on a schedule can find one sequence among billions, copy it a billion times and tell you how much was there. The chemistry answers the first two parts. The last part, and the trust a laboratory places in the answer, depends on the engineering of the instrument as a system and on the evidence built around the test, which is where most of the remaining work lies.

The chapters reduce to a short set of working rules for anyone who designs, verifies, buys, regulates or writes about PCR instruments and tests. Each one traces back to a mechanism explained earlier.

1. Specify sensitivity in copies per reaction. The chemistry, the optics and the statistics of sampling act on the molecules that reach the well, not on the concentration in the specimen (Chapters 1, 6 and 16).
2. Add sample, not cycles, to detect rarer targets. Below about three copies per reaction more than one reaction in twenty receives none, and no number of cycles or brightness of dye changes that (Chapters 6 and 21).
3. Check every oligonucleotide against every other, at the 3′ ends above all. Any pair that can pair and be extended makes a product in every tube, including those with no target (Chapters 5 and 22).
4. Read quantity in the exponential phase, never at the plateau. The plateau forgets the starting amount; the cycle at which a curve rises remembers it (Chapters 8, 11 and 13).
5. State efficiency with every quantity. A tenfold difference is 3.32 cycles only at perfect doubling, and unequal efficiencies masquerade as unequal amounts (Chapters 6 and 13).
6. Never compare Ct values across tests, instruments or thresholds. A Ct is a property of one validated test system, not a unit of viral load (Chapter 13).
7. Measure temperature in the wells, and time holds from the sample's arrival. The controller reads the block; specifications need their definitions and conditions, and real instruments can miss their set points (Chapters 9 and 10).
8. Trace every reported number back through its conversions. Copies become light, electrons, counts, corrected values, a curve, a Cq and a call, and each step needs its own calibration and check (Chapters 12 and 15).
9. Own the interfaces. Plate, seal, volume, dyes, read timing, data formats and software version are where instruments and assays fail together (Chapters 15 and 17).
10. Freeze the test system and treat any change as a change to the test. Instrument, software, reagents, extraction, settings and rules were validated together and must be re-shown together (Chapters 17 and 22).
11. Put an internal control in every specimen. It turns a poor swab, a failed extraction, an inhibitor or a cold well from a false negative into an invalid result (Chapters 19, 20 and 21).
12. Read predictive value with the prevalence of the people tested. The same test can be right about two positives in three in one population and about forty-nine in fifty in another (Chapter 21).
13. Build one evidence core and present it to each regulator. Analytical and clinical studies, risk, software and safety evidence serve the US, the EU and India; keep filed, cleared, granted, authorized, approved and certified apart (Chapters 19 and 23).

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.

Accuracy (temperature)

How close the average temperature of the wells is to the set point; distinct from uniformity.

Amplicon

The stretch of DNA between and including the two primers, which PCR copies exponentially; also the product itself.

Analytical performance

A test's ability to detect or measure its analyte correctly: limit of detection, specificity, precision and related characteristics.

Annealing

The re-pairing of complementary strands on cooling; in PCR, the step in which primers bind their targets.

Baseline

The background fluorescence of the early cycles, subtracted from each curve before a threshold is applied.

cDNA

Complementary DNA: DNA copied from an RNA template by reverse transcriptase.

Clinical performance

How well a test's results correspond to the clinical condition in the intended population: sensitivity, specificity and predictive values.

CLIA

The US Clinical Laboratory Improvement Amendments, which regulate clinical laboratories and sort tests into waived, moderate-complexity and high-complexity categories.

Common specifications

EU requirements set by implementing regulation for the performance of certain class D diagnostic devices.

Cq (Ct)

The fractional cycle at which a well's fluorescence curve crosses the threshold; MIQE's quantification cycle, called threshold cycle (Ct) or crossing point (Cp) by some manufacturers.

CRISPR detection

Detection using a guided bacterial enzyme that, on binding its target sequence, cuts nearby reporter molecules to produce a signal.

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.

Denaturation

Separating the two strands of DNA, in PCR by heating to 94 to 98 °C.

Digital PCR

PCR in which the sample is divided into thousands of partitions that are each read as positive or negative, giving an absolute count of molecules.

dNTP

Deoxynucleoside triphosphate: one of the four building blocks the polymerase adds to a growing DNA strand.

dUTP and UNG

A contamination defense: products are made with uracil, and uracil-DNA glycosylase destroys any carried-over product before the next PCR.

Efficiency

The fraction of target molecules copied in one cycle; at 100 % the target doubles each cycle.

Emergency Use Authorization (EUA)

US authorization of an unapproved product during a declared emergency, lasting until the declaration ends or the authorization is revoked.

Endpoint PCR

PCR in which the product is examined only after the last cycle, typically on a gel.

Error budget

The allocation of a requirement, such as a temperature tolerance, among the separate sources of error that contribute to it.

Exclusivity

A test's failure to react with organisms other than its target; tested as cross-reactivity.

Extension

The step in which the polymerase lengthens a bound primer along its template.

Extraction

Releasing and purifying nucleic acid from a specimen, removing inhibitors and often concentrating it.

Fluorophore

A molecule that absorbs light of one color and emits light of a longer wavelength.

Heated lid

The plate above the tubes held at about 105 °C to prevent condensation and to press the tubes into the block.

Hot start

Any method that keeps the polymerase inactive until the first heating step, to prevent wrong products forming during set-up.

Hydrolysis probe

A probe carrying a fluorophore and a quencher that the polymerase cuts as it copies the target, releasing fluorescence; often called a TaqMan probe.

Inclusivity

A test's ability to detect all variants of its target organism.

Inhibitor

A substance carried in with a specimen or from extraction that slows or stops amplification.

Instrument family

Instruments from one manufacturer with the same general architecture, design, tolerances and capabilities, for the purpose of extending a cleared assay.

Intended use

The written statement of what a device measures, in what specimen, for whom, by whom, where and for what decision.

Interface

A boundary between two parts of a system, with parameters both sides must agree on.

Internal control

A target amplified in every specimen's reaction to show that the specimen, extraction and amplification were adequate; in the CDC test, RNase P.

Isothermal amplification

Copying nucleic acid at a single temperature, using enzymes that separate strands as they copy.

IVDR

Regulation (EU) 2017/746 on in vitro diagnostic medical devices, applicable since 26 May 2022.

LAMP

Loop-mediated isothermal amplification: amplification at about 65 °C with a strand-displacing polymerase and four to six primers.

Laboratory-developed test

A test made and used within one laboratory.

Limit of blank

The highest result expected from a sample containing no analyte.

Limit of detection

For a qualitative molecular test, the lowest concentration detected in at least 95 % of replicates.

Long product

A strand made in the early cycles from an original template, with one end defined by a primer and the other ragged.

Lower limit of quantification

The smallest amount a quantitative test can measure with acceptable precision.

Master mix

The ready-made mixture of polymerase, dNTPs, magnesium and buffer to which primers, probes and sample are added.

Melt curve

A plot of how fluorescence falls as the product is heated slowly; peaks mark the melting temperatures of the products present.

Melting temperature (Tm)

The temperature at which half of a DNA duplex has separated into single strands; it depends on sequence, salt, magnesium and strand concentration.

MIQE

Minimum Information for Publication of Quantitative Real-Time PCR Experiments: reporting guidelines published in 2009 and revised in 2025.

Mispriming

A primer binding to and being extended from a partly matching site.

Multicomponent analysis

Separating the contributions of several dyes to measured light using each dye's calibrated spectrum.

Multiplex

Detecting several targets in one reaction, usually with a different fluorophore for each.

No-template control

A reaction with every reagent but no sample, which should show no amplification; any signal reveals contamination.

Notified body

An organization designated in the EU to assess the conformity of devices in classes that require third-party assessment.

Partition

One of the thousands of small separate reactions into which digital PCR divides a sample.

Passive reference

An inert dye such as ROX added to every well, by which the reporter signal is divided to correct for volume and optical differences.

Peltier module

A thermoelectric device that pumps heat from one face to the other in the direction set by its current.

Plateau

The final phase of a PCR, in which product stops increasing; reached at similar levels from very different starting amounts.

Poisson distribution

The statistics of items scattered at random, which govern how many molecules land in a reaction or partition.

Polymerase

The enzyme that builds a new DNA strand by adding nucleotides to a primer, copying a template.

Positive control

A reaction containing the target at a known level, which must be positive for the run to be valid.

Positive and negative percent agreement

Agreement of a new test with a comparator that is not a reference standard; not to be reported as sensitivity and specificity.

Predictive value

The probability that a positive (or negative) result is correct; it depends on the prevalence in the people tested.

Primer

A short designed piece of DNA that binds the template and gives the polymerase a starting point; PCR uses two, facing each other.

Primer-dimer

A short product formed when two primers pair with each other and are extended.

Probe

A short labeled oligonucleotide that binds inside the target and reports its presence by fluorescence.

QMSR

The US Quality Management System Regulation, 21 CFR 820, in force since 2 February 2026, incorporating ISO 13485:2016.

Quencher

A molecule that absorbs a nearby fluorophore's energy and keeps it dark until they are separated.

Ramp rate

The rate of temperature change, in degrees per second; defined variously for the block or the sample, as a maximum or an average.

Real-time PCR

PCR in which fluorescence is read in every cycle, so the cycle at which the signal rises can be used to infer the starting amount.

Reverse transcriptase

An enzyme that copies RNA into complementary DNA.

RNase

An enzyme that degrades RNA; robust and widespread.

RNase P

A human gene amplified as the CDC test's internal control.

RPA

Recombinase polymerase amplification: isothermal amplification at about 37 °C using recombination proteins to insert primers.

Sample-temperature model

The controller's calculation of the liquid's temperature from the block's history, the volume and the tube's time constant.

Sensitivity (clinical)

The fraction of people with the condition whom a test calls positive, measured against a reference standard.

Specificity (clinical)

The fraction of people without the condition whom a test calls negative, measured against a reference standard.

Stacking

The attraction between neighboring base pairs along the helix, which provides most of a duplex's stability.

Standard curve

A plot of Cq against the logarithm of known starting amounts, used to convert an unknown's Cq into an amount and to estimate efficiency.

Taq polymerase

The heat-stable DNA polymerase of the bacterium Thermus aquaticus, used in most PCR since 1988.

Test system

The assay, instrument and software together, evaluated and authorized as one.

Thermal cycler

The instrument that heats and cools PCR reactions on a programmed schedule; often called a PCR machine.

Threshold

The fluorescence level, set above background in the exponential region, at which Cq is read.

Time constant

For a first-order lag, the time the sample takes to cover about 63 % of a step in temperature.

Uniformity (temperature)

How closely the wells of a block agree with one another at a set point; distinct from accuracy.

V model

The systems-engineering picture of decomposing needs into requirements and designs, then verifying and validating upward.

Validation

Showing that a device meets its users' needs and intended use.

Verification

Showing that a design's outputs meet its inputs: that the product was built as specified.

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.

1. Ch. 1 — Nurk S., Koren S., Rhie A., et al. The complete sequence of a human genome. Science 376, 44–53 (2022). doi:10.1126/science.abj6987
2. Ch. 1 — New England Biolabs. Nucleic acid data (web page). neb.com/en/tools-and-resources/usage-guidelines/nucleic-acid-data
3. Ch. 1 — Thermo Fisher Scientific. Creating standard curves with genomic DNA or plasmid DNA templates for use in quantitative PCR (application note). Manufacturer.
4. Ch. 1 — Thermo Fisher Scientific. Ethidium bromide: detection limit (product page). Manufacturer.
5. Ch. 1 — Bustin S.A., Ruijter J.M., van den Hoff M.J.B., et al. MIQE 2.0: revision of the Minimum Information for Publication of Quantitative Real-Time PCR Experiments guidelines. Clin Chem 71, 634–651 (2025). doi:10.1093/clinchem/hvaf043
6. Ch. 1 — CDC. CDC 2019-Novel Coronavirus (2019-nCoV) Real-Time RT-PCR Diagnostic Panel: Instructions for Use. CDC-006-00019, Rev. 08, effective 7 Mar 2023. fda.gov/media/134922/download
7. Ch. 1 — Lu X., Wang L., Sakthivel S.K., et al. US CDC real-time reverse transcription PCR panel for detection of severe acute respiratory syndrome coronavirus 2. Emerg Infect Dis 26(8) (2020). doi:10.3201/eid2608.201246
8. Ch. 2 — Yakovchuk P., Protozanova E., Frank-Kamenetskii M.D. Base-stacking and base-pairing contributions into thermal stability of the DNA double helix. Nucleic Acids Res 34, 564–574 (2006). doi:10.1093/nar/gkj454
9. Ch. 2 — Marmur J., Doty P. Determination of the base composition of deoxyribonucleic acid from its thermal denaturation temperature. J Mol Biol 5, 109–118 (1962). doi:10.1016/S0022-2836(62)80066-7
10. Ch. 2 — Integrated DNA Technologies. Understanding melting temperature (Tm) (web article). Manufacturer.
11. Ch. 2 — Potapov V., Ong J.L. Examining sources of error in PCR by single-molecule sequencing. PLoS One 12, e0169774 (2017). doi:10.1371/journal.pone.0169774 (authors employed by New England Biolabs)
12. Ch. 3 — New England Biolabs. Guidelines for PCR optimization with Taq DNA polymerase (web page). Manufacturer.
13. Ch. 3 — Promega. GoTaq PCR Core Systems, Technical Bulletin TB254 (rev. 5/19). Manufacturer.
14. Ch. 3 — Saiki R.K., Scharf S., Faloona F., et al. Enzymatic amplification of β-globin genomic sequences and restriction site analysis for diagnosis of sickle cell anemia. Science 230, 1350–1354 (1985). doi:10.1126/science.2999980
15. Ch. 3 — Arizona State University Embryo Project Encyclopedia. Polymerase chain reaction (PCR). embryo.asu.edu
16. Ch. 3 — Spahich N. A not-so-simple idea. The Scientist (1 June 2023).
17. Ch. 3 — Nobel Prize Outreach. The Nobel Prize in Chemistry 1993 (summary) and Mullis K.B., The polymerase chain reaction, Nobel lecture (8 Dec 1993). nobelprize.org
18. Ch. 4 — Innis M.A., Myambo K.B., Gelfand D.H., Brow M.A.D. DNA sequencing with Thermus aquaticus DNA polymerase and direct sequencing of polymerase chain reaction-amplified DNA. PNAS 85, 9436–9440 (1988). doi:10.1073/pnas.85.24.9436 (rates as given in the companion US Patent 5,075,216)
19. Ch. 4 — Lawyer F.C., Stoffel S., Saiki R.K., et al. High-level expression, purification, and enzymatic characterization of full-length Thermus aquaticus DNA polymerase and a truncated form deficient in 5′ to 3′ exonuclease activity. PCR Methods Appl 2, 275–287 (1993). doi:10.1101/gr.2.4.275 (authors at Roche Molecular Systems)
20. Ch. 4 — Eckert K.A., Kunkel T.A. High fidelity DNA synthesis by the Thermus aquaticus DNA polymerase. Nucleic Acids Res 18, 3739–3744 (1990). doi:10.1093/nar/18.13.3739
21. Ch. 5 — Chou Q., Russell M., Birch D.E., Raymond J., Bloch W. Prevention of pre-PCR mis-priming and primer dimerization improves low-copy-number amplifications. Nucleic Acids Res 20, 1717–1723 (1992). doi:10.1093/nar/20.7.1717 (authors at Cetus)
22. Ch. 5 — Kellogg D.E., Rybalkin I., Chen S., et al. TaqStart antibody: hot start PCR facilitated by a neutralizing monoclonal antibody directed against Taq DNA polymerase. BioTechniques 16, 1134–1137 (1994).
23. Ch. 5 — Birch D.E., Kolmodin L., Wong J., et al. Simplified hot start PCR. Nature 381, 445 (1996). doi:10.1038/381445a0
24. Ch. 5 — Dang C., Jayasena S.D. Oligonucleotide inhibitors of Taq DNA polymerase facilitate detection of low copy number targets by PCR. J Mol Biol 264, 268–278 (1996). doi:10.1006/jmbi.1996.0640
25. Ch. 5 — Carr S.M. The polymerase chain reaction (in detail). Memorial University of Newfoundland (teaching page). mun.ca/biology/scarr/PCR_in_detail.html
26. Ch. 5 — Thermo Fisher Scientific. PCR primer design tips. Behind the Bench (25 Sep 2019). Manufacturer.
27. Ch. 5 — FDA. Molecular diagnostic template for SARS-CoV-2 emergency use authorization requests (version of 22 Sep 2020), as reproduced by PRoDTEC. fda.gov/media/135900/download
28. Ch. 5 — Lee J.S., Goldstein J.M., Moon J.L., et al. Analysis of the initial lot of the CDC 2019-Novel Coronavirus (2019-nCoV) real-time RT-PCR diagnostic panel. PLoS One 16, e0260487 (2021). doi:10.1371/journal.pone.0260487
29. Ch. 6 — Bustin S.A., Benes V., Garson J.A., et al. The MIQE guidelines: minimum information for publication of quantitative real-time PCR experiments. Clin Chem 55, 611–622 (2009). doi:10.1373/clinchem.2008.112797
30. Ch. 7 — Brock T.D., Freeze H. Thermus aquaticus gen. n. and sp. n., a nonsporulating extreme thermophile. J Bacteriol 98, 289–297 (1969). doi:10.1128/jb.98.1.289-297.1969
31. Ch. 7 — US Geological Survey, Yellowstone Volcano Observatory. How a thermophilic bacterium from a Yellowstone hot spring is helping fight against COVID-19 (web article). usgs.gov
32. Ch. 7 — Chien A., Edgar D.B., Trela J.M. Deoxyribonucleic acid polymerase from the extreme thermophile Thermus aquaticus. J Bacteriol 127, 1550–1557 (1976). doi:10.1128/jb.127.3.1550-1557.1976
33. Ch. 7 — Saiki R.K., Gelfand D.H., Stoffel S., et al. Primer-directed enzymatic amplification of DNA with a thermostable DNA polymerase. Science 239, 487–491 (1988). doi:10.1126/science.2448875
34. Ch. 7 — Smithsonian Institution Archives. History of the Polymerase Chain Reaction Videohistory Collection (RU 9577). si.edu
35. Ch. 7 — Thermo Fisher Scientific. 30 years of thermal cycler innovations #1: the world's first thermal cycler. Behind the Bench (blog). Manufacturer.
36. Ch. 7 — Cline J., Braman J.C., Hogrefe H.H. PCR fidelity of Pfu DNA polymerase and other thermostable DNA polymerases. Nucleic Acids Res 24, 3546–3551 (1996).
37. Ch. 7 — Holland P.M., Abramson R.D., Watson R., Gelfand D.H. Detection of specific polymerase chain reaction product by utilizing the 5′→3′ exonuclease activity of Thermus aquaticus DNA polymerase. PNAS 88, 7276–7280 (1991). doi:10.1073/pnas.88.16.7276
38. Ch. 7 — Sigma-Aldrich (Merck). Antibody/enzyme-mediated hot start PCR (technical page). Manufacturer.
39. Ch. 8 — Morrison C., Gannon F. The impact of the PCR plateau phase on quantitative PCR. Biochim Biophys Acta 1219, 493–498 (1994). doi:10.1016/0167-4781(94)90076-0
40. Ch. 8 — Kainz P. The PCR plateau phase – towards an understanding of its limitations. Biochim Biophys Acta 1494, 23–27 (2000). doi:10.1016/S0167-4781(00)00200-1
41. Ch. 8 — US National Institute of Justice. DNA amplification online training: plateau effect. nij.ojp.gov
42. Ch. 9 — Ferrotec. Thermoelectric technical reference: mathematical modeling of thermoelectric cooling modules (section 11) and reliability of thermoelectric cooling modules (section 10). thermal.ferrotec.com. Manufacturer.
43. Ch. 9 — Barako M.T., Park W., Marconnet A.M., Asheghi M., Goodson K.E. Thermal cycling, mechanical degradation, and the effective figure of merit of a thermoelectric module. J Electron Mater 42, 372–381 (2013). doi:10.1007/s11664-012-2366-1
44. Ch. 9 — Roche Diagnostics. LightCycler 480 Instrument Operator's Manual, software 1.5, rev. 3.0 (2008). Manufacturer.
45. Ch. 9 — Atwood J.G., Mossa A.C., Goven L.M., et al. Thermal cycler for automatic performance of the polymerase chain reaction with close temperature control. US Patent 5,475,610 (filed 20 Apr 1992; issued 12 Dec 1995).
46. Ch. 9 — National Institute of Advanced Industrial Science and Technology (AIST). Thermophysical property database: thermal conductivity of metals. tpds.db.aist.go.jp
47. Ch. 9 — Thermo Fisher Scientific. What is the max block ramp rate of the GeneAmp PCR System 9700? (product FAQ). Manufacturer.
48. Ch. 9 — Bio-Rad. CFX96 Touch Real-Time PCR Detection System, Bulletin 6093 Ver F (2020). Manufacturer.
49. Ch. 9 — Wittwer C.T., Fillmore G.C., Hillyard D.R. Automated polymerase chain reaction in capillary tubes with hot air. Nucleic Acids Res 17, 4353–4357 (1989). doi:10.1093/nar/17.11.4353
50. Ch. 9 — Roche Diagnostics. LightCycler 2.0 Instrument (product page). Manufacturer.
51. Ch. 9 — QIAGEN. Rotor-Gene Q MDx (US) (product page). Manufacturer.
52. Ch. 9 — Kopp M.U., de Mello A.J., Manz A. Chemical amplification: continuous-flow PCR on a chip. Science 280, 1046–1048 (1998). doi:10.1126/science.280.5366.1046
53. Ch. 9 — Krishnan M., Ugaz V.M., Burns M.A. PCR in a Rayleigh–Bénard convection cell. Science 298, 793 (2002).
54. Ch. 9 — Son J.H., Cho B., Hong S., et al. Ultrafast photonic PCR. Light Sci Appl 4, e280 (2015). doi:10.1038/lsa.2015.53
55. Ch. 9 — Ahrberg C.D., Manz A., Chung B.G. Polymerase chain reaction in microfluidic devices. Lab Chip 16, 3866–3884 (2016). doi:10.1039/C6LC00984K
56. Ch. 10 — Thermo Fisher Scientific. Applied Biosystems 7500 Fast Dx Real-Time PCR Instrument (brochure CO016365, 2015). Manufacturer.
57. Ch. 10 — Bio-Rad. C1000 Touch Thermal Cycler, Bulletin 6095 Ver C (2019). Manufacturer.
58. Ch. 10 — Bio-Rad. CFX Opus 96 Real-Time PCR System, Bulletin 7299 Ver D (2023). Manufacturer.
59. Ch. 10 — Thermo Fisher Scientific. QuantStudio 5 Real-Time PCR System, specification sheet CO014799 (2015). Manufacturer.
60. Ch. 10 — Eppendorf. Mine is the fastest cycler ever – really? Lab Academy (18 Sep 2018). Manufacturer.
61. Ch. 10 — Schoder D., Schmalwieser A., Schauberger G., et al. Novel approach for assessing performance of PCR cyclers used for diagnostic testing. J Clin Microbiol 43, 2724–2728 (2005). doi:10.1128/JCM.43.6.2724-2728.2005
62. Ch. 10 — Thermo Fisher Scientific. Thermal cycler temperature accuracy: a comparison of several models (application note APN-11799200). Manufacturer.
63. Ch. 10 — METTLER TOLEDO. PCR thermal cycler calibration (service page). mt.com
64. Ch. 10 — Span M., Verblakt M., Hendrikx T. Measurement uncertainty in calibration and compliancy testing of PCR and qPCR thermal cyclers. 18th International Congress of Metrology, 05003 (2017). doi:10.1051/metrology/201705003 (authors employed by CYCLERtest)
65. Ch. 11 — Higuchi R., Dollinger G., Walsh P.S., Griffith R. Simultaneous amplification and detection of specific DNA sequences. Bio/Technology 10, 413–417 (1992). doi:10.1038/nbt0492-413
66. Ch. 11 — Higuchi R., Fockler C., Dollinger G., Watson R. Kinetic PCR analysis: real-time monitoring of DNA amplification reactions. Bio/Technology 11, 1026–1030 (1993). doi:10.1038/nbt0993-1026
67. Ch. 11 — Heid C.A., Stevens J., Livak K.J., Williams P.M. Real time quantitative PCR. Genome Res 6, 986–994 (1996). doi:10.1101/gr.6.10.986
68. Ch. 11 — Sykes P.J., Neoh S.H., Brisco M.J., et al. Quantitation of targets for PCR by use of limiting dilution. BioTechniques 13, 444–449 (1992).
69. Ch. 11 — Vogelstein B., Kinzler K.W. Digital PCR. PNAS 96, 9236–9241 (1999). doi:10.1073/pnas.96.16.9236
70. Ch. 12 — Zipper H., Brunner H., Bernhagen J., Vitzthum F. Investigations on DNA intercalation and surface binding by SYBR Green I, its structure determination and methodological implications. Nucleic Acids Res 32, e103 (2004). doi:10.1093/nar/gnh101
71. Ch. 12 — Integrated DNA Technologies. Interpreting melt curves: an indicator, not a diagnosis (web article). Manufacturer.
72. Ch. 12 — Lee L.G., Connell C.R., Bloch W. Allelic discrimination by nick-translation PCR with fluorogenic probes. Nucleic Acids Res 21, 3761–3766 (1993). doi:10.1093/nar/21.16.3761
73. Ch. 12 — Tyagi S., Kramer F.R. Molecular beacons: probes that fluoresce upon hybridization. Nat Biotechnol 14, 303–308 (1996). doi:10.1038/nbt0396-303
74. Ch. 12 — FDA. Emergency Use Authorization letter for the CDC 2019-nCoV Real-Time RT-PCR Diagnostic Panel, reissued 1 Dec 2020. fda.gov/media/134919/download
75. Ch. 12 — Applied Biosystems. 7300/7500/7500 Fast Real-Time PCR System Installation and Maintenance Guide, P/N 4378657 Rev. A (2006). Manufacturer.
76. Ch. 12 — Thermo Fisher Scientific. QuantStudio 5 Dx Real-Time PCR System, specification sheet COL25645 (2022). Manufacturer.
77. Ch. 12 — Shu B., Kirby M.K., Davis W.G., et al. Multiplex real-time reverse transcription PCR for influenza A virus, influenza B virus, and severe acute respiratory syndrome coronavirus 2. Emerg Infect Dis 27(7) (2021). doi:10.3201/eid2707.210462
78. Ch. 12 — Thermo Fisher Scientific. ROX passive reference dye for troubleshooting real-time PCR (application note CO016741, 2015). Manufacturer.
79. Ch. 13 — QIAGEN. Commonly used terms in PCR (web page). Manufacturer.
80. Ch. 13 — Verification of real-time PCR methods for qualitative and quantitative testing of genetically modified organisms (2012), quoting the minimum performance requirements of the European Network of GMO Laboratories. Secondary source.
81. Ch. 13 — Livak K.J., Schmittgen T.D. Analysis of relative gene expression data using real-time quantitative PCR and the 2^−ΔΔCT method. Methods 25, 402–408 (2001). doi:10.1006/meth.2001.1262
82. Ch. 13 — Public Health Agency of Canada. COVID-19 testing: polymerase chain reaction (PCR) and cycle threshold (Ct) values (modified 11 Sep 2026). canada.ca
83. Ch. 13 — AACC. AACC recommendation for reporting SARS-CoV-2 cycle threshold (Ct) values (17 Mar 2021). aacc.org
84. Ch. 13 — Association for Molecular Pathology and Infectious Diseases Society of America. Important issues to consider before interpreting and applying Ct values in clinical practice (2021). amp.org
85. Ch. 13 — Bullard J., Dust K., Funk D., et al. Predicting infectious severe acute respiratory syndrome coronavirus 2 from diagnostic samples. Clin Infect Dis 71, 2663–2666 (2020). doi:10.1093/cid/ciaa638
86. Ch. 13 — Singanayagam A., Patel M., Charlett A., et al. Duration of infectiousness and correlation with RT-PCR cycle threshold values in cases of COVID-19, England, January to May 2020. Euro Surveill 25, 2001483 (2020). doi:10.2807/1560-7917.ES.2020.25.32.2001483
87. Ch. 14 — Majumdar N., Wessel T., Marks J. Digital PCR modeling for maximal sensitivity, dynamic range and measurement precision. PLoS One 10, e0118833 (2015). doi:10.1371/journal.pone.0118833 (authors employed by Thermo Fisher Scientific)
88. Ch. 14 — Corbisier P., Pinheiro L., Mazoua S., et al. DNA copy number concentration measured by digital and droplet digital quantitative PCR using certified reference materials. Anal Bioanal Chem 407, 1831–1840 (2015). doi:10.1007/s00216-015-8458-z
89. Ch. 14 — Hindson B.J., Ness K.D., Masquelier D.A., et al. High-throughput droplet digital PCR system for absolute quantitation of DNA copy number. Anal Chem 83, 8604–8610 (2011). doi:10.1021/ac202028g
90. Ch. 14 — Bio-Rad. QX200 Droplet Digital PCR System (product page) and QX600 Droplet Digital PCR System, Bulletin 3557 Ver A (2023). Manufacturer.
91. Ch. 14 — QIAGEN. QIAcuity Digital PCR System and QIAcuity Nanoplates (product pages). Manufacturer.
92. Ch. 14 — Thermo Fisher Scientific. QuantStudio Absolute Q Digital PCR System, specification sheet COL25472 (2021). Manufacturer.
93. Ch. 14 — Roche. Roche Digital LightCycler System: a technological guide (2022). Manufacturer.
94. Ch. 14 — Bio-IT World. Bio-Rad expands Droplet Digital PCR offering through strategic acquisition and platform rollout (9 Jul 2025).
95. Ch. 14 — Sidstedt M., Rådström P., Hedman J. PCR inhibition in qPCR, dPCR and MPS — mechanisms and solutions. Anal Bioanal Chem (2020). doi:10.1007/s00216-020-02490-2
96. Ch. 15 — FDA. Content of Premarket Submissions for Device Software Functions (final guidance, 14 Jun 2023). fda.gov/media/153781/download
97. Ch. 16 — Mooz H., Forsberg K. The relationship of system engineering to the project cycle. Joint conference of NCOSE and the American Society for Engineering Management (1991); INCOSE International Symposium 1(1). doi:10.1002/j.2334-5837.1991.tb01484.x
98. Ch. 16 — FDA. Design Control Guidance for Medical Device Manufacturers (11 Mar 1997).
99. Ch. 16 — FDA. Medical Devices; Quality System Regulation Amendments. Final rule, 89 FR 7496 (2 Feb 2024), effective 2 Feb 2026; and FDA, Quality Management System Regulation (QMSR) (web page).
100. Ch. 16 — ISO 13485:2016. Medical devices — Quality management systems — Requirements for regulatory purposes (confirmed 2025).
101. Ch. 17 — FDA. Assay Migration Studies for In Vitro Diagnostic Devices (final guidance, April 2013). fda.gov/media/73669/download
102. Ch. 17 — FDA. Replacement Reagent and Instrument Family Policy for In Vitro Diagnostic Devices (final guidance, 17 Aug 2022). fda.gov/media/111186/download
103. Ch. 18 — CDC. Shipping of CDC 2019 novel coronavirus diagnostic test kits begins (press release, 6 Feb 2020).
104. Ch. 18 — Raja S., Ching J., Xi L., et al. Technology for automated, rapid, and quantitative PCR or reverse transcription-PCR clinical testing. Clin Chem 51, 882–890 (2005). doi:10.1373/clinchem.2004.046474
105. Ch. 18 — Cepheid. Technology of the GeneXpert System, training document 301-7932 Rev B.1 (2020). Manufacturer.
106. Ch. 18 — FDA. Emergency Use Authorization letter for Xpert Xpress SARS-CoV-2 (first issued 20 Mar 2020); Cepheid, Xpert Xpress SARS-CoV-2 Instructions for Use, Rev. G (2022); Cepheid press release (21 Mar 2020). Manufacturer for time to result.
107. Ch. 18 — Poritz M.A., Blaschke A.J., Byington C.L., et al. FilmArray, an automated nested multiplex PCR system for multi-pathogen detection: development and application to respiratory tract infection. PLoS One 6, e26047 (2011). doi:10.1371/journal.pone.0026047
108. Ch. 18 — FDA. 510(k) decision summary K243406: cobas liat SARS-CoV-2, Influenza A/B & RSV (2025).
109. Ch. 18 — bioMérieux. BioFire Respiratory 2.1 (RP2.1) Panel (product page). Manufacturer.
110. Ch. 18 — Chin C.D., Linder V., Sia S.K. Commercialization of microfluidic point-of-care diagnostic devices. Lab Chip 12, 2118–2134 (2012). doi:10.1039/C2LC21204H
111. Ch. 18 — Abbott. ID NOW COVID-19 Emergency Use Authorization Instructions for Use, Rev. 8 (Aug 2021). fda.gov/media/136525/download
112. Ch. 19 — IEC 62304:2006+AMD1:2015. Medical device software — Software life cycle processes; and Johner Institute, IEC 62304 second edition (8 Jul 2026).
113. Ch. 19 — ISO 14971:2019. Medical devices — Application of risk management to medical devices (confirmed 2025); ISO/TR 24971:2020, Guidance on the application of ISO 14971, Annex H.
114. Ch. 19 — 21 CFR 866.3981. Device to detect and identify nucleic acid targets in respiratory specimens from microbial agents that cause the SARS-CoV-2 respiratory infection and other microbial agents (eCFR).
115. Ch. 19 — IEC 61010-1:2010+AMD1:2016 CSV; IEC 61010-2-010:2019; IEC 61010-2-081:2019; IEC 61010-2-101:2018 (IEC Webstore); and SIQ, IEC 61010-1 Amendment 2: upcoming changes (3 Apr 2025).
116. Ch. 19 — IEC 61326-1:2020 and IEC 61326-2-6:2025. Electrical equipment for measurement, control and laboratory use — EMC requirements (IEC Webstore).
117. Ch. 19 — IEC 62366-1:2015+AMD1:2020. Medical devices — Application of usability engineering to medical devices; ISO 18113-3:2022, IVD instruments for professional use (labeling).
118. Ch. 20 — Temin H.M., Mizutani S. RNA-dependent DNA polymerase in virions of Rous sarcoma virus. Nature 226, 1211–1213 (1970). doi:10.1038/2261211a0
119. Ch. 20 — Baltimore D. RNA-dependent DNA polymerase in virions of RNA tumour viruses. Nature 226, 1209–1211 (1970). doi:10.1038/2261209a0
120. Ch. 20 — Roche Life Science. Reverse transcription technology (web article). Manufacturer.
121. Ch. 20 — Thermo Fisher Scientific. What is RT-qPCR? Basic principles (web article). Manufacturer.
122. Ch. 20 — Thermo Fisher Scientific (Ambion). The basics: RNase control (technical article). Manufacturer.
123. Ch. 20 — Jue E., Witters D., Ismagilov R.F. Two-phase wash to solve the ubiquitous contaminant-carryover problem in commercial nucleic-acid extraction kits. Sci Rep 10, 1940 (2020). doi:10.1038/s41598-020-58586-3
124. Ch. 20 — Schrader C., Schielke A., Ellerbroek L., Johne R. PCR inhibitors – occurrence, properties and removal. J Appl Microbiol 113, 1014–1026 (2012). doi:10.1111/j.1365-2672.2012.05384.x
125. Ch. 20 — Al-Soud W.A., Rådström P. Purification and characterization of PCR-inhibitory components in blood cells. J Clin Microbiol 39, 485–493 (2001). doi:10.1128/jcm.39.2.485-493.2001
126. Ch. 20 — Longo M.C., Berninger M.S., Hartley J.L. Use of uracil DNA glycosylase to control carry-over contamination in polymerase chain reactions. Gene 93, 125–128 (1990). doi:10.1016/0378-1119(90)90145-H
127. Ch. 21 — CLSI. EP17-A2: Evaluation of detection capability for clinical laboratory measurement procedures, 2nd ed. (2012; reaffirmed 2017).
128. Ch. 21 — CLSI. EP07: Interference testing in clinical chemistry (current edition).
129. Ch. 21 — FDA. Statistical Guidance on Reporting Results from Studies Evaluating Diagnostic Tests (final guidance, 13 Mar 2007).
130. Ch. 21 — ISO 15189:2022. Medical laboratories — Requirements for quality and competence; ISO 17822:2020, IVD test systems — Nucleic acid amplification-based examination procedures for detection and identification of microbial pathogens — Laboratory quality practice guide.
131. Ch. 21 — CLSI. MM17: Validation and verification of multiplex nucleic acid assays, 2nd ed. (2018); MM19: Establishing molecular testing in medical laboratory environments, 2nd ed. (2025).
132. Ch. 22 — Holmes E.C., on behalf of the consortium led by Zhang Y.-Z. Novel 2019 coronavirus genome. virological.org (posted 10/11 Jan 2020).
133. Ch. 22 — Corman V.M., Landt O., Kaiser M., et al. Detection of 2019 novel coronavirus (2019-nCoV) by real-time RT-PCR. Euro Surveill 25, 2000045 (2020). doi:10.2807/1560-7917.ES.2020.25.3.2000045
134. Ch. 22 — HHS. Determination of public health emergency. 85 FR 7316 (7 Feb 2020).
135. Ch. 22 — HHS Office of Inspector General. CDC's internal control weaknesses led to its initial COVID-19 test kit failure, but CDC ultimately created a working test kit. A-04-20-02027 (24 Oct 2023).
136. Ch. 22 — NPR. CDC report: officials knew coronavirus test was flawed but released it anyway (6 Nov 2020).
137. Ch. 22 — Sarata A.K. Early development and regulation of diagnostic testing for COVID-19: frequently asked questions. Congressional Research Service R46261 (updated 17 Mar 2021).
138. Ch. 22 — CDC. Lab alerts: Changes to CDC RT-PCR for SARS-CoV-2 testing (21 Jul 2021); Clarifications about retirement of the CDC 2019-nCoV Real-Time RT-PCR Diagnostic Panel (2 Aug 2021).
139. Ch. 22 — FDA. In vitro diagnostics EUAs — molecular diagnostic tests for SARS-CoV-2 (web page, content current as of 11 Aug 2026).
140. Ch. 22 — HHS. Termination of three declarations authorizing emergency use of medical devices during the COVID-19 pandemic. 91 FR 40544 (2 Jul 2026), effective 26 Dec 2026.
141. Ch. 23 — Regulation (EU) 2017/746 on in vitro diagnostic medical devices (IVDR), consolidated version of 10 Jan 2025. EUR-Lex 32017R0746
142. Ch. 23 — 21 CFR 862.2570, Instrumentation for clinical multiplex test systems (eCFR); FDA product classification database, product code OOI.
143. Ch. 23 — FDA. 510(k) K082562, Applied Biosystems 7500 Fast Dx (decision 30 Sep 2008).
144. Ch. 23 — FDA. Revocation of authorization of emergency use of an in vitro diagnostic device for detection and/or diagnosis of COVID-19; availability. 86 FR 28841 (28 May 2021), recording De Novo DEN200031, granted 17 Mar 2021; classification codified at 89 FR 66554 (16 Aug 2024).
145. Ch. 23 — 21 U.S.C. 360bbb-3 (Federal Food, Drug, and Cosmetic Act section 564).
146. Ch. 23 — FDA. Transition Plan for Medical Devices Issued Emergency Use Authorizations Related to COVID-19 (final guidance, March 2023).
147. Ch. 23 — 42 CFR 493.17, Test categorization (eCFR); FDA, CLIA categorizations (web page).
148. Ch. 23 — FDA. Medical Devices; Laboratory Developed Tests. Final rule, 89 FR 37286 (6 May 2024); Laboratory Developed Tests; Implementation of Vacatur, 90 FR 45134 (19 Sep 2025); Covington & Burling, FDA's LDT rule struck down by federal district court (3 Apr 2025).
149. Ch. 23 — MDCG 2020-16. Guidance on classification rules for in vitro diagnostic medical devices under Regulation (EU) 2017/746, rev. 4 (March 2025) and rev. 5 (September 2026).
150. Ch. 23 — Team-NB. Considerations for conformity assessments done by (IVD) notified bodies in the case of down classification of SARS-CoV-2 tests (12 Dec 2025).
151. Ch. 23 — Commission Implementing Regulation (EU) 2022/1107 laying down common specifications for certain class D in vitro diagnostic medical devices. OJ L 178, 5.7.2022.
152. Ch. 23 — European Commission. EU reference laboratories (in vitro diagnostics web page).
153. Ch. 23 — Regulation (EU) 2024/1860 amending Regulations (EU) 2017/745 and (EU) 2017/746 (13 Jun 2024).
154. Ch. 23 — European Commission. COM(2025) 1023 final (16 Dec 2025); European Parliament Legislative Observatory, procedure 2025/0404(COD); Council progress report ST 9801/26 (5 Jun 2026).
155. Ch. 23 — Government of India. Medical Devices Rules, 2017, G.S.R. 78(E) (31 Jan 2017).
156. Ch. 23 — CDSCO. List of in-vitro diagnostic medical devices under Medical Devices Rules, 2017 (updated 25 Oct 2023).
157. Ch. 23 — Lok Sabha. Unstarred Question No. 1669, answered 16 Dec 2022 (Ministry of Health and Family Welfare).
158. Ch. 23 — CDSCO. In-vitro diagnostic medical devices: frequently asked questions, CDSCO/IVD/FAQ/03/2022.
159. Ch. 23 — ICMR. Guidelines for validation and batch testing of COVID-19 diagnostic kits (4 Jun 2020).
160. Ch. 23 — ICMR and CDSCO. Inviting comments on standard IVD evaluation protocol (28 Jan 2025).
161. Ch. 24 — Farrar J.S., Wittwer C.T. Extreme PCR: efficient and specific DNA amplification in 15–60 seconds. Clin Chem 61, 145–153 (2015). doi:10.1373/clinchem.2014.228304
162. Ch. 24 — Makrigiorgos G.M. Extreme PCR meets high-speed melting: a step closer to molecular diagnostics 'while you wait'. Clin Chem 65, 217–219 (2019). doi:10.1373/clinchem.2018.298356
163. Ch. 24 — Notomi T., Okayama H., Masubuchi H., et al. Loop-mediated isothermal amplification of DNA. Nucleic Acids Res 28, e63 (2000). doi:10.1093/nar/28.12.e63
164. Ch. 24 — Piepenburg O., Williams C.H., Stemple D.L., Armes N.A. DNA detection using recombination proteins. PLoS Biol 4, e204 (2006). doi:10.1371/journal.pbio.0040204
165. Ch. 24 — Gootenberg J.S., Abudayyeh O.O., Lee J.W., et al. Nucleic acid detection with CRISPR-Cas13a/C2c2. Science 356, 438–442 (2017). doi:10.1126/science.aam9321
166. Ch. 24 — Chen J.S., Ma E., Harrington L.B., et al. CRISPR-Cas12a target binding unleashes indiscriminate single-stranded DNase activity. Science 360, 436–439 (2018). doi:10.1126/science.aar6245
167. Ch. 24 — Contagion Live. FDA authorizes first CRISPR test kit for COVID-19 (May 2020).