Building a better mousetrap for accurate and sensitive polymerase chain reaction

The natural history of chronic myeloid leukaemia (CML) is forever changed with the advent of tyrosine kinase inhibitor (TKI) therapy. All patients with CML have the unique BCR–ABL1 fusion gene, and the protein product of that gene is the target of the TKI, while the mRNA fusion provides the target for disease monitoring in the peripheral blood. Quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) of BCR–ABL1 mRNA is a sensitive method for monitoring CML, yet qRT-PCR assay variation from lab to lab can be considerable.^{1, 2} A large international effort of CML researchers developed and validated the International Standard scale (IS) for BCR–ABL1 as a way to facilitate lab to lab comparisons.³ Still the IS is just a correction factor, and no magical powers. Applying it to a less than accurate BCR–ABL1 value just gives you a different less than accurate BCR–ABL1 value.

The measurement of disease burden via BCR–ABL1 transcript levels has substantial clinical utility. The decline of disease after the start of TKI therapy is associated with a very favourable long-term response. Conversely, patients who have BCR–ABL1 >10% IS after three months of treatment have a worse long-term response compared to cases with a BCR–ABL1 of less than 10% IS. Secondly, patients who achieve a BCR–ABL1 level of <0·1% IS (MR3) enjoy a ‘safe harbour’ where relapse or progression is quite rare. Finally, patients who achieve a stable deep molecular response (MR4 or below) can potentially discontinue TKI therapy. Many trials have shown that ~40–60% of patients who discontinue their TKI after a prolonged deep response can enjoy years of treatment-free remission (TFR).^{4, 5} Naturally, discontinuation must be accompanied by frequently monitoring for disease recurrence. So important is BCR–ABL1 testing to the management of CML that the European Leukemia Network (ELN) and the National Comprehensive Cancer Network and developed guidelines for BCR–ABL1 monitoring over the entire course of CML therapy and discontinuation.^{6, 7}

A good qRT-PCR assay can detect the BCR–ABL1 transcript in a background of ~100 000 total mRNA transcripts.⁸ The fact that this can be done with some precision and accuracy is rather amazing. Consider: the problem of detection of any target by PCR is the issue of signal (BCR–ABL1) to noise (all other mRNA). This is often likened to finding the needle in the haystack. Conventional qRT-PCR requires the exponential amplification of the target and a separate exponential amplification of the control gene. If the amplification of these two targets differs, the calculations will not be inaccurate. ‘Digital’ PCR (dPCR) is a methodological approach to improve the signal/noise limitation.⁹ The concept is essentially to divide the sample into an excess of partitions, so that each partition houses only a few copies of either the target (signal), or the background (noise). Digitalization can be done by distributing samples into an array of thousands of tubes and wells (Fluidigm), or into droplets (first Raindance, and now Bio-Rad). Going back to the needle/haystack analogy, each well/droplet could either be empty, contain a piece of hay, or a needle. Amplification yields a positive signal with target amplification (digital code = 1), and no amplification in partitions without the target (digital code = 0). The Poisson distribution is then used to estimate the initial input target BCR–ABL1 copy numbers.

There are several fundamental characteristics of dPCR which makes it attractive on a technical level. First, one measures actual molecules (positive droplets) rather than estimating amount of starting template by amplification slope and standard curve samples.^{10, 11} Secondly, since it uses end-point determination of signal (yes or no), it is potentially less sensitive to elements that can influence amplification efficiency, such as sample contaminants, poor-quality mRNA, etc. Third, dPCR allows the sampling of a large number of wells/droplets, which should improve the precision and sensitivity of the assay. These qualities suggest the dPCR assay should be sensitive and reproducible from assay to assay and lab to lab.^{12, 13}

In this issue, Scott et al.¹⁴ report on the test characteristics of a digital droplet (dd)PCR assay for BCR–ABL1 (Bio-Rad), and demonstrate that the assay is accurate, sensitive, and robust across a large panel of international labs. Contrived samples of K562 (which have multiple copies of BCR–ABL1) were diluted into HL60 cells (BCR–ABL1-negative) and sent to 26 participating labs. The dilution series range from a 1:10 dilution (MR1) to 1:100 000 (MR%). The most important findings were: (i) all labs detected a MR4 level dilution. This is important as this is the level of disease burden required in several TFR studies. (ii) Most labs detected even lower amounts of BCR–ABL1, as 96% of labs detected MR4·5, 78% MR4·7, and 87% MR5. (iii) Linearity over the dilution series was excellent, at an average r² of 0·99. (iv) Eleven labs performed both ddPCR and their standard RT-PCR technique, and while the ddPCR did not seem to have an advantage at detection of BCR–ABL1 at low levels, linearity across labs was better than for qRT-PCR, suggesting that at low levels of BCR–ABL1, comparisons across labs would be better. This has obvious advantages for multicentre clinical trials, as well as for the occasional nomadic CML patient.

The paper does have some potential limitations. First, only 23 of 26 centres returned data, and of those, only 11 did both ddPCR and their conventional assay (one wonders about selection bias — why didn’t the other labs report their standard assay?). Second, the analysis is done on contrived samples rather than patient samples, which is potentially problematic since K562 has multiple BCR–ABL1 fusions. That being said, ddPCR was used in the US LAST trial, a large study of 174 CML patients in deep molecular response who underwent subsequent discontinuation.¹⁵ Patients had conventional RT-PCR performed as the study standard, but also had ddPCR done prior to discontinuation. Of the patients with undetectable BCR–ABL1 by RT-PCR, 56 (39%) had BCR–ABL1 detected by ddPCR, suggesting that in real patients, ddPCR might be more sensitive than conventional qRT-PCR.

There are now several BCR–ABL1 tests approved by regulatory agencies, and a myriad of ‘home brew’ assays. How does one pick which assay to employ (full disclosure: our lab generated data for both Cepheid’s and Bio-Rad’s US FDA application, and we use both instruments)? Chief considerations are test performance, cost, speed, and scale. Standard PCR is usually the cheapest, will likely perform well in its home lab, but will not travel as well as a standardized reaction. Cartridge systems (Cepheid) offer the greatest ease of use, give very rapid results, and are an excellent solution for point-of-care testing. The downside is cost, particularly if one wants to do very large-scale studies (e.g. a retrospective analysis of hundreds of samples). Digital PCR systems give very sensitive and reproducible results, but require more technician time, and may not be cost-effective if a lab is doing a low volume of samples and cannot have the luxury of batching samples.

Perhaps the most important message in this paper is not the BCR–ABL1 assay per se, but rather the proof that ddPCR is an excellent solution for accurate and precise molecular monitoring. The use of measurable residual disease (MRD) as a measure of drug efficacy and prediction of outcome is important in all the leukaemias, and ddPCR stands to become a major platform for MRD assessment. Further, adaptations of current digital PCR systems could potentially be used for point-of-care use in low-resource settings to make sample preparation, partitioning, and analysis simple and automated and with devices that are low-cost and portable.^{16, 17}

Digital PCR is an elegant solution to divorce signal from noise. Too bad it is not so simple with social media.