Minimal residual disease: the advantages of digital over analog polymerase chain reaction

In this issue of Leukemia and Lymphoma, Goh and colleagues describe the use of a nanofluidic digital polymerase chain reaction (PCR) assay to improve the lower limit of detection of BCR–ABL1 mRNA in patients with chronic myeloid leukemia (CML) [1]. BCR–ABL1 was detectable in a significant proportion of patients with a complete molecular response (CMR) according to a conventional definition based on real-time reverse transcriptase quantitative PCR (RQ-PCR).

In imatinib-treated patients with CML it is known that a CMR is not necessarily equated with eradication of the CML clone, since around 60% of patients with a stable CMR relapse when imatinib is withdrawn [2,3]. More sensitive assays for minimal residual disease (MRD) are required to identify patients who are more likely to remain in remission without therapy. In order to improve on the performance of conventional RQ-PCR it is helpful to consider the factors that limit ‘sensitivity’ (i.e. the lower limit of detection) and specificity (the ability to distinguish between true and false positive signals).

It has been shown that a carefully optimized RQ-PCR assay is capable of detecting a single copy of the target sequence if there is at least a single copy of the target gene in the sample [4]. In practice, the interpretation of RQ-PCR results near the limit of detection is not so simple. In our laboratory, for instance, the lower limit of detection is set at a threshold cycle number (Ct) equivalent to 10 copies of BCR–ABL1, determined from the standard curve. This cut-off was determined empirically as giving the best distinction between true low positive results and the low level of amplification that may be seen in negative control samples. However, by definition we have lost 1-log of sensitivity by adopting this threshold. RQ-PCR is an analog system: the result (Ct) is a continuous variable, but we are forced to apply a somewhat arbitrary cut-off to give a binary (positive or negative) or digital result as the system nears its limits.

In some laboratories the same problem is overcome by switching from RQ-PCR to another binary assay, nested PCR, which gives a positive or negative result. In some laboratories this strategy improves the lower limit of detection, while in other laboratories it does not. The nested assay is non-quantitative, so even if it does confer additional sensitivity it is not an ideal solution to the problem.

The lower limit of detection of an MRD assay is determined primarily by the amount of patient material that is tested. When PCR replicates are performed it is usual to see a mixture of positive and negative results when the number of copies of the target is near the limit of detection. This reflects the stochastic distribution of rare copies of the target gene. The performance of multiple PCR reactions increases sensitivity. If 10 replicates are performed the limit of detection is improved by 1-log compared with a single PCR. However, in order to be confident that a positive result in an MRD assay indicates the presence of the leukemic clone in the patient sample, it is also essential to ensure that the assay incorporates appropriate negative controls. Occasional low level positives may be seen in negative control samples due to non-specificity, contamination between samples, or potentially the presence of rare BCR–ABL1 transcripts, even in normal individuals [5]. One strategy to overcome this limitation in a sensitive replicate PCR assay is to switch from BCR–ABL1 mRNA to genomic BCR–ABL1 DNA. The wide distribution of breaks in the ABL1 gene ensures that the precise BCR–ABL1 breakpoint sequence is more or less unique to each individual patient with CML. The feasibility of this approach has been shown by several groups, including our own [3,6–8].

Digital PCR approaches the problem in a different way. Using nanofluidic technology the nucleic acid sample is divided up into hundreds of individual replicate PCR reactions. If a single copy of BCR–ABL1 cDNA exists among a million copies of other cDNA species and the reaction is divided up into 1000 individual assays in individual wells, then the ratio of BCR–ABL1 to background is now 1:1000 in the single well that contains the single copy of BCR–ABL1. This approach results in more efficient amplification of BCR–ABL, illustrated in the work of Goh et al. by an earlier Ct value [1]. Digital PCR has a second advantage in that it is quantitative by virtue of the high number of replicates, thereby reducing the imprecision of the assay.

Goh and colleagues used a combination of nesting and digital PCR and found that nesting further extended the lower limit of detection by about 1-log [1]. Nesting was achieved by a pre-amplification step of 10 PCR cycles. Given that digital PCR should result in more efficient amplification of a single copy of BCR–ABL1, if a copy is present, what is achieved by the addition of nesting? A substantial part of the effect is altering the amount of nucleic acid that is sampled, i.e. concentrating the target. If we consider a hypothetical sample in which there is a single intact copy of BCR–ABL1 in 40 µL of cDNA, this would result in 1024 copies after 10 cycles of efficient amplification. In the 2.5 µL aliquot of pre-amplified cDNA that is carried over into the digital PCR there would be 11 copies of BCR–ABL1 in a much smaller volume.

BCR–ABL1 measurement by digital PCR without pre-amplification is quantitative. However, the assay as described becomes at best semi-quantitative with nesting, since there is no way of measuring the variability in efficiency of the pre-amplification step. This problem could be addressed by the incorporation of BCR–ABL1 standards in the nested digital PCR. This step would safeguard against PCR failure and would ensure consistent efficiency of the pre-amplification PCR.

One of the disadvantages of BCR–ABL1 mRNA as a target for PCR-based MRD strategies is the potential for contamination and false positive results. Contamination could occur at any stage in the assay: from RNA extraction through cDNA synthesis to the final RQ-PCR. The risk of contamination is particularly problematic when dealing with amplified material. For this reason we routinely include in RQ-PCR a negative control sample (RNA extracted from HeLa cells) that is processed together with the patient samples in the same batch. Some equivalent negative control is essential in any highly sensitive assay. This was incorporated into a digital PCR assay that was used to sensitively measure the T315I kinase domain mutation [9]. The negative control was used to ensure specificity by setting a cycle threshold at which patient data would be considered positive.

Whilst their paper raises some interesting methodological questions, one of the key findings of Goh and colleagues is that BCR–ABL1 mRNA is detectable in patients with CML with a stable CMR [1]. This observation corroborates our own findings in the Australasian Leukaemia and Lymphoma Group CML8 study of imatinib withdrawal in CMR [3], as well as those of Melo and colleagues in imatinib-treated patients [6]. The challenge now is to determine the best way to measure MRD in patients with CMR in order to identify candidates at lower risk of relapse when treatment is stopped, and in order to measure the effect of novel therapies aimed at reducing the currently unmeasurable burden of MRD. Digital PCR is likely to be one of a number of strategies that help us to meet this challenge.

Potential conflict of interest:

Disclosure forms provided by the authors are available with the full text of this article at www.informahealthcare.com/lal.