Digital PCR
The reaction is emulsified into roughly 20,000 water-in-oil droplets of about one nanolitre each, cycled in a plate and then read one droplet at a time as they flow single-file past a detector. The partition count is generated afresh for each well rather than fixed by the consumable.
Our ruleSimulations prepare you for the bench but don't replace hands-on practice.
Droplet digital (ddPCR)
The reaction is emulsified into roughly 20,000 water-in-oil droplets of about one nanolitre each, cycled in a plate and then read one droplet at a time as they flow single-file past a detector. The partition count is generated afresh for each well rather than fixed by the consumable.
How it differs from the standard method
- Droplets are made from a 20 µL reaction, so about 1 µL of sample is actually analysed per well: input is what limits sensitivity.
- Droplets can be lost in generation and rejected during reading, so the accepted-droplet count is part of the result.
- Replicate wells can be merged in software to raise the number of molecules counted.
- The emulsion needs its own supermix and gentle handling; vortexing after droplet generation destroys the partitions.
1Counting instead of measuring
Digital PCR takes one ordinary PCR and splits it into thousands of tiny independent reactions before cycling: water-in-oil droplets, or chambers etched into a chip. Each partition gets whatever molecules happen to fall into it: none, one, or a few.
2Why Poisson, and what it does for you
If you simply counted positives as molecules you would undercount, because some partitions receive two or three. The molecules are distributed at random, so the number per partition follows a Poisson distribution with mean λ.
Too dilute
A handful of positives. The estimate is unbiased but the confidence interval is enormous, because counting statistics on 20 positives is ±22%.
The sweet spot
About 80% positive. This is where the relative uncertainty of the Poisson estimate is at its minimum for a given number of partitions.
Running out of room
Over 99% positive. The few remaining negatives carry all the information, so the estimate becomes wildly uncertain.
Saturated
Every partition positive. There is no negative fraction left to take a logarithm of: the result is not 'very high', it is no result at all. Dilute and re-run.
3Dynamic range is set by the partitions, not by the chemistry
The lower end of the range is how few positives you are willing to report: with 20,000 partitions, one positive is about 0.06 copies/µL, and the Poisson confidence interval around a few counts is so wide that the number is only useful as a detection call. The upper end is saturation.
- λ (lambda)
- Mean target molecules per partition, calculated as −ln(1 − fraction positive). The quantity the whole measurement reduces to.
- Partition volume
- The volume of one droplet or chamber, typically 0.3–1 nL. It converts λ into copies per microlitre, so an error in it is a systematic error in every result.
- Accepted partitions
- The partitions the instrument could actually read. Droplets lost in generation or rejected on size do not count, and the total must be reported.
- Rain
- Partitions with amplitudes between the negative and positive clusters, so that no threshold calls them all correctly.
- Absolute quantification
- A concentration obtained without a calibrator, because the unit of measurement is a molecule in a compartment rather than a signal level.
- Duplex assay
- Two targets read in two colours in the same partitions, which is how a mutant and its wild-type allele are measured in the same reaction.
- Allele fraction
- Mutant copies divided by total copies. Measured as a ratio of two counts from the same well, so pipetting and extraction losses cancel.
- False-positive rate
- The fraction of wild-type molecules the assay miscalls as mutant. This, and the number of molecules you load, set the limit of detection for rare-variant work.
4Rain, and where to put the threshold
In a good assay the two clusters are far apart and the threshold between them is obvious. Rain means some partitions ended up part-way: a molecule that started amplifying late, a partially inhibited partition, a probe that binds a sequence variant less well, or an annealing temperature high enough that only some partitions got going.
5Finding one molecule among thousands
Partitioning is what makes rare-variant detection possible: a mutant molecule alone in its droplet is not competing with 10,000 wild-type templates for primers, so it amplifies to full amplitude and is counted. But the limit of detection is not set by the number of partitions.
Key terms
Practise these as flashcards →One of the thousands of independent sub-reactions, a droplet or a chamber. Its volume is the unit that converts a count into a concentration.
λ = −ln(1 − p): the step that accounts for partitions holding more than one molecule. Without it, high concentrations are systematically undercounted.
Every partition is read once, after cycling, as positive or negative. There is no real-time curve and no Cq.
Fluorescence of every partition, drawn as a cloud of negatives, a cloud of positives, and whatever lies between. The raw data worth looking at before the summary number.
The span of concentrations between too few positives to be precise and saturation. Set by partition number and volume, not by the chemistry.
The positive count you get from a sample known to contain no target. Measured, not assumed, and it is what a rare-variant call must exceed.
Combining the partitions of several wells into one measurement, which is how you get more molecules through a fixed-capacity instrument.
When to use it
- You need an absolute concentration with no standard curve: viral load, a reference material, a library for sequencing, a GMO fraction in food.
- You need to resolve a small ratio (a copy-number change of 1.2-fold, or an allele fraction well under a per cent) that is inside the noise of a Cq-based measurement.
- Your sample contains inhibitors: partitioning dilutes each molecule's competition and the endpoint read is insensitive to efficiency loss that would shift a Cq.
- You are measuring a mutant against a wild-type background in the same well, so the ratio of two counts cancels extraction and pipetting losses.
Compared with alternatives
Cheaper, faster and with a far wider dynamic range; prefer it for routine quantification over orders of magnitude, or when a standard curve is already established.
Endpoint PCR with a gel answers present or absent. Digital PCR is the same endpoint chemistry, but partitioned, so the answer becomes a number.
Prefer sequencing when you do not know which variant to look for, or need many loci at once; digital PCR is more sensitive and much cheaper for one known target.
A dye-based measurement gives total DNA mass. Digital PCR counts amplifiable copies of one sequence, which is what matters for a library or a viral load.
Prefer isothermal methods for a yes/no answer in the field; they need no thermocycler but cannot count molecules.
Sub-techniques
Choosing one here switches the whole page, including the protocol and the lab, to it.
The reaction is emulsified into roughly 20,000 water-in-oil droplets of about one nanolitre each, cycled in a plate and then read one droplet at a time as they flow single-file past a detector. The partition count is generated afresh for each well rather than fixed by the consumable.
- Droplets are made from a 20 µL reaction, so about 1 µL of sample is actually analysed per well: input is what limits sensitivity.
- Droplets can be lost in generation and rejected during reading, so the accepted-droplet count is part of the result.
- Replicate wells can be merged in software to raise the number of molecules counted.
- The emulsion needs its own supermix and gentle handling; vortexing after droplet generation destroys the partitions.