Parameters to assess performance of dPCR assays

The authors presented post-run evaluation criteria to check if quantification was accurate, evaluated Poisson confidence intervals and gave an alternative to better capture variability in the analytical chain.

In dPCR a reaction is split into a large number of (nanoliter) sub-reactions so that individual target copies are separated by partitioning.

After thermal cycling and read-out this leads to classification of each partition as positive (containing target) or negative (no target present).

The distribution of the target molecules across the partitions can be seen as a Poisson process (the targets end up in partitions independently and with a fixed rate).

Poisson statistics allow calculation of the initial number of targets from the number of positive and negative partitions.

Direct and absolute quantification ​

The direct, absolute quantification makes it attractive for routine analysis of food and feed samples for composition, possible GMO content and compliance with labelling requirements.

Quantitative PCR (qPCR) is based on the proportionality between fluorescence and DNA mass and where quantification is always relative. So calibration curves are needed for absolute quantification.

For categorization of the compartments’ fluorescence readings into positive (containing target DNA, high fluorescence), negative (no target DNA, low fluorescence), and rain (intermediate fluorescence) the team used a procedure based on kernel density estimation.

They defined the ‘rain’ as the compartments whose fluorescence readings are between the maximal negative fluorescence and the minimal positive fluorescence.

One of the mechanisms leading to excessive amounts of ‘rain’ may be sequence dependent (either target or primer sequence).

Re-designing primers for a certain target may be the most effective way to get a robust dPCR assay. Should the choice of target sequence be limited, sonication and/or extended cycling regimes may help in improving an assay.

The researchers said if a PCR assay is specific and efficient, the main source of error in digital quantification is caused by the misclassification of droplets.

So the main goal of the criteria should be to ensure a robust separation of positives and negatives allowing accurate classification of compartments and reliable quantification, they added.

“Three criteria that allow to meet this requirement are (I) single amplification product (there should only be two fluorescence populations) (II) peak resolution (as a measure of the separation between positives and negatives), and (III) the amount of stragglers or ‘rain’ ​(i.e​.​ droplets that have an intermediate fluorescence and do not seem to belong to either the positive or negative population). ​

“Several other criteria were considered (gain, bandwidth, and signal to noise ratio) but were either found to be not very informative or overlapping with the former.”​

They evaluated the usefulness of the criteria and their limits by applying them to 12 validated qPCR assays transferred to digital PCR.

Post-run analysis​

One of the key tasks an operator has in the post-run analysis is to evaluate quality of the data produced by the reaction.

The researchers said in most digital systems, there is only endpoint read-out of the partitions which provides less information as to how the amplification process performed.

“We propose three reaction aspects that can be inspected before deciding to accept or discard it for further analysis: (I) the concentration of the target (is it sufficient to allow accurate quantification?), (II) the degree of compartmentalization (has a sufficient number of partitions been generated to allow accurate quantification?), and (III) the use of repeats to gauge variability in the results (rather than relying on Poisson confidence intervals).”​

A more apt strategy than Poisson confidence to digital PCR confidence intervals involves the original sample being split in two sub-samples which are both extracted for DNA.

Both extracts are analysed in duplicate, and the standard deviation across all four results is used to obtain confidence bounds around the mean.

Droplet and chamber based dPCR systems are subject to variability in the number of compartments generated and/or accepted into the analysis.

Unlike in real time PCR the entire volume of sample loaded into the chamber/droplet generator is not analysed. This corresponds to a form of sub-sampling which may add variation or error to the quantification, especially for reactions with targets at very low abundance, said the team.

Source: PLOS ONE

DOI: 10.1371/journal.pone.0153317​

“Measuring Digital PCR Quality: Performance Parameters and Their Optimization”​

Authors: Antoon Lievens, Sara Jacchia, Dafni Maria Kagkli, Cristian Savini, Maddalena Querci