considerations for digital pcr as an accurate molecular diagnostic

Considerations for Digital PCR as an AccurateMolecular Diagnostic Tool

Jim F. Huggett,1,2* Simon Cowen,1 and Carole A. Foy1

BACKGROUND: Digital PCR (dPCR) is an increasinglypopular manifestation of PCR that offers a number ofunique advantages when applied to preclinical research,particularly when used to detect rare mutations and inthe precise quantification of nucleic acids. As is commonwith many new research methods, the application ofdPCR to potential clinical scenarios is also being increas-ingly described.

CONTENT: This review addresses some of the factors thatneed to be considered in the application of dPCR. Com-pared to real-time quantitative PCR (qPCR), dPCRclearly has the potential to offer more sensitive and con-siderably more reproducible clinical methods that couldlend themselves to diagnostic, prognostic, and predictivetests. But for this to be realized the technology will needto be further developed to reduce cost and simplify ap-plication. Concomitantly the preclinical research willneed be reported with a comprehensive understanding ofthe associated errors. dPCR benefits from a far more pre-dictable variance than qPCR but is as susceptible to up-stream errors associated with factors like sampling andextraction. dPCR can also suffer systematic bias, partic-ularly leading to underestimation, and internal positivecontrols are likely to be as important for dPCR as they arefor qPCR, especially when reporting the absence of asequence.

SUMMARY: In this review we highlight some of the con-siderations that may be needed when applying dPCR anddiscuss sources of error. The factors discussed here aim toassist in the translation of dPCR to diagnostic, predictive,or prognostic applications.© 2014 American Association for Clinical Chemistry

Digital PCR (dPCR)3 is an increasingly applied tech-nique that offers a number of advantages for both detect-

ing and quantifying nucleic acids. The method, which isperformed by diluting target nucleic acid across a largenumber of reactions (termed partitions) so that some ofthe reactions contain no template, is gaining considerabletraction. The concept behind dPCR has been long estab-lished (1 ), but it has been made practical in recent yearsby the development of microfluidics and/or emulsionchemistries to simplify and automate the process. Thefrequently cited benefits of improved precision affordingfiner fold-change measurements and deeper rare variantdetection have been highlighted and increasingly appliedto clinical challenges, including oncology (2–5 ) and in-fectious diseases (6–10) as well as fetal genetic screening(11, 12 ), and in predicting transplant rejection (13 ).

Despite positive press, dPCR is far from infallible. Itdepends on PCR, which, although it is possibly the mostquintessential molecular method yet developed, requiresfairly complex and accurate thermocycling instrumenta-tion to perform. Real-time quantitative PCR (qPCR)revolutionized clinical application of PCR partly becauseit automated analysis by removing the need for postreac-tion manipulation. Most dPCR instruments have taken astep backward in this sense, requiring downstream ma-nipulation to analyze the results, increasing time, andcomplicating the process. dPCR is also developing in thesame primordial sea as next generation sequencing,which continues to advance with both hardware andchemistry to increase throughput as well as sequence readnumber and length. Advanced sequencing methods arealso likely to eventually replace most molecular methodsthat are used to identify and quantify nucleic acids, in-cluding PCR in its many guises. Yet dPCR is here nowand has much to offer. This review focuses on some of thekey technical considerations in the performance dPCRfor a clinical application.

Routine Adoption of Quantitative ClinicalMolecular Measurement

One undeniable truth about the clinical adoption of newdiagnostic, predictive, or prognostic methods is it takes avery long time. Any clinical measurement that also re-quires quantification of a given analyte poses major ad-

1 LGC, Teddington, UK; 2 Research Department of Infection, Division of Infection and Im-munity, UCL, London, UK.

* Address correspondence to this author at: LGC, Queens Rd., Teddington, TW11 0LY, UK.Fax +44-(0)-20-8943-2767; e-mail [email protected].

Received June 26, 2014; accepted September 11, 2014.Previously published online at DOI: 10.1373/clinchem.2014.221366© 2014 American Association for Clinical Chemistry3 Nonstandard abbreviations: dPCR, digital polymerase chain reaction; qPCR, real-time

quantitative PCR; Cq, quantification cycle; SNP, single-nucleotide polymorphism;cfDNA, cell-free DNA; LAMP, loop-mediated isothermal amplification; RPA, recombinase

polymerase amplification; dMIQE, Minimum Information for the Publication of Quanti-tative dPCR Experiments; HER2, human epidermal growth factor-2; cdPCR, chip digitalPCR; ddPCR, droplet digital PCR.

Clinical Chemistry 61:179–88 (2015) Reviews

79

ditional challenges which are dependent upon both bio-logical and technical factors. qPCR is over 20 years old(14 ), but it has only really been applied clinically in areasin which alternatives are not practically possible, such asmonitoring treatment in diseases like chronic-phasechronic myeloid leukemia (15 ) or for some key blood-borne viruses (16 ). Yet despite this there are thousands ofpublications reporting the use of qPCR for clinical mea-surement, with little translation of their findings to pa-tient care (17 ).

Quantitative clinical measurements of a candidateDNA (or RNA) molecule using qPCR, or any molecularmethod, have 2 major problems to resolve. First, themeasurement is at the mercy of the suitability of theanalyte as a clinically useful target. A given experimentmay report a compelling finding, but this may be difficultto reproduce owing to biological differences associatedwith factors like population demographics. Second, stan-dardization of qPCR is highly challenging. qPCR pro-vides a very precise measurement over a vast dynamicrange; but the quantification cycle (Cq) is fairly arbitrary.Depending on factors such as the instrument, reporter,primers, and mastermix, a given Cq value can differ inassociated copy number by a factor of over a thousand.

Calibration curves are frequently employed to re-duce the error associated with qPCR, but they in turn arechallenging to select, value assign, and apply in a mannerthat will be reproducible; their application also containsinherent error that is almost never considered. Arguably,a key problem with applying qPCR to areas such as thediscovery of biomarkers that will eventually be translatedto clinical care is understanding whether poor reproduc-ibility is biological or due to issues around the fact thetechnique is difficult to perform reproducibly.

dPCR will be no better than qPCR at resolving thefact that DNA and RNA analytes can be biologicallyhighly variable and that a particular experiment may yielda compelling but irreproducible result [discussed in moredetail by Sanders et al. (18 )]. Where we predict dPCRwill have the edge over qPCR will be with technical re-producibility. This is because the digital output derivedfrom diluting the sample essentially counts the numberof molecules, which is far more reproducible than themore analog Cq output offered by qPCR. This has thepotential to improve both quantitative and qualitativemolecular measurements. But for preclinical dPCR re-search to be most valuable it must be performed with acomprehensive understanding of the potential inherentfailings, and conclusions must be made with an awarenessof sources of bias that may produce an incorrect result.

Advantages

dPCR principally offers advantages when considering theidentification of the presence and abundance of rare se-

quence mutations (19 ) and in the quantification ofnucleic acids (20 ), while also enabling the cis/trans rela-tionships (21, 22 ) to be determined. Quantitative andqualitative measurements offer different challenges whenconsidering error, and many of the challenges that befalllegacy PCR and qPCR when designing and optimizingan analytically sensitive and specific method directly ap-ply to dPCR; i.e., a nonspecific and/or poorly optimizedset of primers may bind the wrong target, leading to afalse-positive signal. Consequently, when dPCR is per-formed, it is vital to also perform control experimentsusing templates that are likely to complicate a result ifthey are present within a sample.

Because dPCR performs by identifying single parti-tions as positive or negative, it offers the potential toperform precise low-level quantification (23 ). However,studies that take advantage of this capability to investi-gate trace or low-abundance sequences should also mea-sure sources of false-positive signal due to low-level con-tamination and/or nontarget amplification (caused bymis-priming and/or primer dimers). If large numbers oftrace experiments are performed, the chances of false pos-itives increase, and these false-positive results must becaptured when assessing limits of detection (24 ). Conse-quently, sufficient replication of appropriate controls isrequired to confidently attribute any low-level signal tothe sample of interest and not to low-frequency falsepositives.

Rare Variant Detection

The term “digital” PCR was coined in a study that fo-cused upon the detection of rare variant mutants of theras oncogene in a sample that predominantly consistedof wild-type sequences (19 ). Designing a qPCR (orarray)–based assay to a given single-nucleotide polymor-phism (SNP) is not usually highly challenging because, aslong as the target sequence is suitable, it is fairly simple toselect a primer or probe to the desired SNP. The un-wanted sequence will usually also be detected, but at amuch lower (approximately 5%) detection efficiency(25 ); i.e., if the mutant and wild-type sequence are mea-sured at the same concentration by qPCR, the wild-typesequence will amplify approximately 4 cycles later thanthe mutant.

Where wild-type and mutant sequences are mixedthere will be a proportion (peq) of the wild type for whichan equivalent amount of signal comes from the wild-typeand mutant sequence following equation 1:

peq �1

1 � f(1)

where f is the efficiency (or frequency) by which themutant probe binds the wild-type signal. So for a typ-

Reviews

80 Clinical Chemistry 61:1 (2015)

ical qPCR assay for which wild-type sequences aredetected at a frequency of 0.05 compared to the mu-tant, the wild-type proportion at which 50% of thesignal comes from each genotype would be 0.952. Thismeans that when there are 20 wild-type sequences and1 mutant sequence, 50% of the amplification plot willcomprise wild-type sequence. The problem apparentfrom this equation is that the wild-type sequence doesnot have to be very abundant for it to be the predom-inant sequence detected, thus compromising the spec-ificity of the assay. There have been several developmentsto further reduce the efficiency by which the wild-type se-quence is detected using blocker probes (26, 27) that fur-ther reduce f. However, because the need to detect even rarermutations is demanded, the specificity of the assay contin-ues to be a potential problem.

dPCR sidesteps this problem because the parti-tioning that is its hallmark also reduces the wild-type/mutant ratio and the likelihood that peq will bereached. This process can be taken to the extreme bypartitioning the template at such a low concentrationthat each partition contains only a single molecule,either SNP or wild type (28 ). However, SNP detec-tion by any PCR method is a balance between assayspecificity and sensitivity in which design and optimi-zation are paramount.

dPCR can be more sensitive than conventional ap-proaches but remains susceptible to poor assay design,leading to cross-reactivity and false positives. Partitioningmay not prevent assay cross-reactivity, however, and “wild-type only” partitions can still lead to a signal that may com-plicate analysis (Fig. 1). Consequently, robust application ofsuch assays may require optimization and potentially rede-sign to determine the assay performance using both the de-sired and undesired genotypes to determine if the differentgenotypes can be discriminated by applying an amplitudethreshold (Fig. 1C) or not (Fig. 1D).

Another unsung fact about dPCR is that, although thetechnique may be able to measure 1 SNP in a sea of 10000wild-type sequences, the practical fact is that this wouldrequire the measurement of approximately 50000 gene cop-ies to reach a 99% probability that the sample contains atleast 1 mutant. This affords 2 challenges.

First, 50000 gene copies of wild-type sequence will betoo concentrated for accurate measurement with the use ofall but one of the instruments that are currently availablecommercially. Consequently, if an accurate estimation ofthe abundance of mutant to wild type is needed, then thewild type must be measured at a concentration differentfrom that of the mutant, rendering multiplexing useless andfurther complicating the measurement.

Second, depending on the biological sample,50 000 copies of human genomic DNA could require alarge amount of tissue. Although many sample speci-mens readily provide large amounts of DNA, this is

not always the case. Cell-free nucleic acid measure-ment in plasma is an increasingly popular source ofsample for rare-variant detection because it is less in-vasive than a tissue biopsy. To this end this sample hasalso been the focus of several studies using dPCR(5, 13 ). However, because cell-free DNA (cfDNA) isat a concentration of approximately 1000 copies permL of plasma in healthy individuals (29 ) it wouldrequire about 50 mL of blood to measure 50 000cfDNA copies. This is a very large sample for a poten-tial routine clinical test and, in this case, 1 mutation in10 000 wild types may not be possible for practicalrather than technical reasons.

Absolute Quantification

Quantification of nucleic acids is the other major strength ofdPCR because it does not routinely require a calibrationcurve to provide a numeric value. This is further strength-ened by the random nature of the distribution of the DNAmolecules across the partitions that makes the precision ofmeasurement both predictable and precise compared toqPCR (20), a fact that may also be applicable to RNA mea-surement (30) as well as other nucleic acid amplificationmethods like loop-mediated isothermal amplification(LAMP) (31) and recombinase polymerase amplification(RPA) (32).

We describe in the Minimum Information for thePublication of Quantitative dPCR Experiments(dMIQE) guidelines that dPCR is uniquely able todetect and provide an “absolute” quantitative value ofsmall amounts of specific target in a complex back-ground without the need for a calibration curve (33 ).However, this accolade makes it very difficult to assessaccuracy (if the results are actually correct) becausethere are few methods that are able to verify dPCRresults. The vast majority of other molecular methods,including qPCR, are “relative” methods that must becalibrated using a similar template that must, in turn,be given a quantity (or value assigned) using anotherabsolute quantification method.

Most value assignment for molecular standards islikely to be performed using spectrophotometry at anabsorbance of A260. Comparison of spectrophotomet-ric methods for DNA measurements with dPCR hasgenerally shown good agreement, differing by �2-fold(23, 34, 35 ). To begin with this is very promising be-cause when molecular quantification is routinely ap-plied in the clinic, decisions are seldom made onchanges of �10-fold. However, it must be remem-bered that although these studies show promise, be-cause they suggest that the agreement between dPCRand A260 is fairly accurate, they could also be equallybiased. This is particularly pertinent when we considerthe fact that usually nucleic analysis of a clinical sam-

Molecular Diagnostics Using dPCR Reviews

Clinical Chemistry 61:1 (2015) 81

ple requires processing to purify and concentrate thenucleic acids.

Error: Bias

To better understand the accuracy of dPCR, which isinfluenced by bias and variance, we will initially con-sider sources of bias (with which the given result isincorrect). As with any PCR assay, poor design or

optimization may lead to overestimation due to non-specific amplification, as will probe bleed-throughduring multiplexing (through which the signal fromone fluorophore is detected during the reading of an-other). Accepting this, a well-designed and optimized,specific dPCR assay is limited in the ability for over-estimation of the target sequence. This is because, un-like qPCR (which is dependent on factors like valueassignment of the calibrator and PCR efficiency), the-

Fig. 1. Example of assay cross-reactivity using the QX100 (Bio-Rad).(A and C), Probes designed to specify the mutant sequences can cross-react, detecting the wild-type sequence with a reduced ampli-tude. When a clear distinction can be made, a threshold can be drawn to discriminate the mutant sequence from the wild type (B). Thiscan be challenging if the difference in amplitude between mutant and wild type is less pronounced (D). Solutions to reduce the impactof sequence cross-reactivity can include optimizing annealing temperature, PCR cycles, concentration of probe and/or primers or byredesigning the assay.

Continued on page 83

Reviews


oretically, overestimation by dPCR can occur only ifDNA molecules become denatured during partition-ing and the resulting single strands frequent 2 differentpartitions. Consequently the worst case scenario willbe a 2-fold overestimation.

Underestimation is a far greater potential sourceof bias when performing dPCR and is rooted in thefact that dPCR is only able to count positive parti-tions. A negative partition is routinely assumed to con-tain no DNA; however, it is known that when usingmethods like digital LAMP (31 ) and when detectingRNA by reverse transcriptase dPCR (30 ) it is possiblefor a large number of partitions to contain templatethat is not amplified. This “molecular dropout” islikely to be low in a dPCR reaction with a pure DNAtemplate (36 ). However, template must be amplifiable

and factors that damage DNA or that make DNA lessaccessible, like plasmid supercoiling (23 ), are likely toalso lead to molecular dropout.

Inhibition of dPCR appears to cause less negativebias than with qPCR (31, 37, 38 ); however, inhibitorsare able to reduce the quantitative value measured bydPCR while maintaining good precision (31 ), and westrongly recommend including internal positive con-trols when measuring clinical samples, especially ifnegative results need to be reported. The popular rou-tine use of internal positive controls when applyingqPCR clinically is equally important for dPCR.

Other sources of underestimation will occur for whichsome of the assumptions that underpin dPCR are not met.The current method used for quantification by dPCR istotally dependent upon the random distribution of tem-

Fig. 1. Continued.



plate. Situations in which this condition cannot be met,owing to factors like template linkage and/or sample inho-mogeneity, will lead to an underestimation. A pertinent ex-ample is the measurement of genome instabilities like hu-man epidermal growth factor-2 (HER2) amplification,which manifests as cis amplification of the HER2 oncogeneon chromosome 17 (39). Unless the individual targets areseparated they cannot frequent different partitions and willnot be independently counted by dPCR. This effect, whichdoes not have an impact on qPCR, may lead to underesti-mation of the number of copies when measured by dPCR(20).

Underestimation may also occur because the under-lying model that is used to estimate the number of copies(Eq. 2)(33 ) describes a situation which is physicallyimpossible:

� � �ln�1 �k

n� (2)

This is because with Eq. 2, the number of positive parti-tions k is divided by the total number of partitions n,assuming that there is no partition volume variation. Be-cause the partition volume must have an associated error(34, 40 ), if this is not factored into the equation thenthere is potential for underestimation (Fig. 2). This has agreater effect in situations for which � is high and will nothave an impact on the results if experiments are per-formed at a � that is low enough to ensure single-molecule occupancy of partitions. What is not clear is

how this might really affect quantification by dPCRwhen the other associated errors are considered and howthis variance could be factored into the equation (onepossible solution is offered in Box 1 and the Supplemen-tary Material that accompanies the online version ofthis article at http://www.clinchem.org/content/vol61/issue1). It is possible that the partition volume variance of agiven instrument may be negligible compared to the errorassociated with other factors like pipetting, sample process-ing, and so on. However, what is needed is for this informa-tion to be made available for the different instruments so thiscan be tested, enabling users to be more confident in the biasof their measurements. Ideally, where run-to-run differencesare substantial, instruments would provide these types ofquality metrics for each run to enable these variations to bemeasured on a day to day basis.

Error: Variance

dPCR measurement is well documented as being basedon the Poisson distribution (20, 41 ) which, for a givenmean, has a constant variance. Although this may be truefor the measurement of a given pure DNA extract, inhib-itory substances copurified or introduced during the ex-traction may or may not affect the overall measurementvariability (31 ). The other major factor that effects allmolecular methods is the source of variance (and bias)contributed by the upstream steps involved in sampling,storing, and extracting the nucleic acid (42 ) to whichdPCR is equally susceptible.

Consequently, when designing a dPCR experimentthat aims to make quantitative clinical measurement, anappropriate estimation of the variance (and therefore po-tential confidence) must be captured by replicating up-stream steps, including extraction procedure, to estimateall the potential sources of technical variance. This willalso reduce the likelihood of the high technical precisionof the dPCR step leading to significant but erroneousbiased results.

Measuring Linkage

The problem for quantification by dPCR if sequences arelinked and not randomly distributed can also be usedadvantageously. We used this in combination with mul-tiplex analysis to estimate molecular dropout and inves-tigate the impact of different template types on error(36 ). This application also has the potential to offer clin-ically valuable tools to investigate haplotypes in prenatalscreening for diseases like �-thalassemia (43 ) and poly-morphic gene complex arrangements encoding immuno-logical cell receptors (21 ).

Fig. 2. Schematic diagram showing the effect of partitionvolume variation on the distribution of sample among thepartitions.The left (fixed) and right (variable) panels both contain the samenumber of copies, but in the latter case, the volume variabilityleads to a smaller number of partitions with a positive signal.Applying the standard equation for � would produce a negativebias on the estimate.

Reviews


http://www.clinchem.org/content/vol61/issue1

http://www.clinchem.org/content/vol61/issue1

Box 1. A New Method for Improving the Accuracy of Quantification by dPCR.dPCR is performed by distributing DNA across a large number of reactions (partitions) at such a concentra-tion that a proportion of the partitions will contain no template at all, resulting in a negative signal. The pro-portion of positive signals is then used to estimate the average number of copies per partition (�) using thePoisson distribution.

One of the key assumptions is that the DNA distribution is random and equal for each partition. The lattercannot be met if the volume varies among partitions; if partition volume variability is not considered whencalculating �, the distribution of molecules among partitions may lead to a bias in the estimated number ofcopies per reaction (Fig. 2). This has implications for any research area in which accurate measurement is par-ticularly important.

There is currently a range of commercially available instruments employing either chip or emulsion partition-ing, with additional formats also being developed. With continuing developments meeting the challenges ofreaction partitioning, differences in the associated volume variation are likely to lead to the possibility that thiscould have an impact on instrument comparability. To raise awareness of the existence of uncertainty in parti-tion volume, the dMIQE guidelines included partition volume variance as one of the list of desirable pieces ofinformation (“Desirables”) to include in a publication when available.

The problem with this Desirable is that instrument volume variability is not readily available, so it cannot betaken into account when estimating �. In particular, what has not been addressed is how such variation mayhave an impact on dPCR results if partition volume variation is significant, and � is calculated using the con-ventional equation.

We have developed a potentially more accurate estimation of �, which considers volume variation for a givenobserved count. By removing the assumption that the partition volume is fixed (and � constant), we use amodel with a �-distributed volume and consider � as a mean over the volume variation. The resulting expres-sions for � and its uncertainty remove the bias and provide a more accurate estimate. A detailed derivation isgiven in the online Supplementary Material, together with some example calculations.

When the volume variability is small, or the sample concentration is low, the effects described here are negligi-ble, and so in some operating ranges, it is not relevant. However, the effect becomes more pronounced withincreasing sample concentration and partition volume variability (see online Supplementary Fig. 1). In addi-tion, there is associated deterioration in precision (see online Supplementary Fig. 2). These findings also high-light a difference in the value of � for which dPCR is most precise (reportedly approximately 1.6 using theconventional equation and assuming no volume variation) between instruments of differing partition volumevariability.

The volume effect we describe will initially be most relevant to measuring fine differences in fold changes orto assigning values to reference materials intended to support measurements in areas such as testing for geneticmodification. However, as dPCR becomes more established with more instruments available, it may havebroader implications when we consider interinstrument comparisons, for which the partition volume variabil-ity differs. It is important that manufacturers of dPCR instruments provide information on partition volumevariation if absolute quantification is to be offered. Further, incorporating volume variability in their instru-ment analysis software using an approach such as the one we propose would be desirable for the most accuratedPCR measurement.

Another factor that is raised by our findings is the potential importance of knowing how such volume variabil-ity may differ between batches of chips, because this will affect the degree of bias present. In situations inwhich instruments generate partitions (as with the current droplet formats), it is important that they are rou-tinely checked to define volume variability if run-to-run differences are to be avoided.

The impact of this phenomenon will ultimately increase as instruments capable of processing larger numbers ofpartitions are developed, enabling ever-increasing range in �. Our findings demonstrate that when � is low andvariation is small (less than about 10%), volumetric variability is unlikely to have a major impact. As � increases, thebias becomes more severe, impacting measurement accuracy. We encourage instrument manufacturers to providepartition volume uncertainty information, and our model provides a mechanism for more accurate measurementwhen using instruments with differing partition volume variation.



dPCR experiments that measure linkage depend oncolocalization of targets within the partitions that do notfollow the random distribution. As with quantification,other factors that disrupt random distribution of tem-plate, like sample inhomogeneity, will also cause colocal-ization and may cause artifacts. Conversely, negative as-sociations will also be incorrectly measured if the linkedmolecules are separated. Consequently template qualityand the physical distance of the target sequences of inter-est must be considered in data analysis.

Translation of dPCR to Routine MolecularDiagnostics

The earliest impact dPCR is likely to have on patient carewill be through the value assigning of reference materials(44, 45 ) to support existing molecular diagnostics. Nextwill be the potential to use dPCR for routine calibrationof the standards used in qPCR. Whether these are simplein-house standards or tied to other reference materials,for many of the reasons discussed above using dPCR islikely to improve the accuracy of assigning calibrators andthus the reproducibility of qPCR.

Many of the considerations discussed above willneed to be addressed for dPCR to be routinely used as adiagnostic tool. There are, however, also some rudimen-tary issues that must be considered if the impact of dPCRon patient care is to be maximized. One of the key ad-vantages of dPCR is that, unlike its main competitor nextgeneration sequencing, most of the reagents have beendeveloped by the demands of qPCR users and the re-sponse of some industrious biotechnology companies.qPCR has set the path and is increasingly accepted as adiagnostic tool; however, there are some key factors thatwe also predict will be necessary to implement for dPCRto maximize its clinical impact.

Greater Reaction Volumes and a HigherDynamic Range

qPCR has started to define what is possible for moleculardiagnostics, and this is particularly pertinent for clinicalmolecular quantification because it is versatile in manyways; one key advantage is that qPCR is readily scalable.Consequently, although dPCR has the potential to bemore sensitive than qPCR when sample volumes arematched, qPCR will always have the edge if sensitivitycan be improved by performing a larger-volume reaction(46, 47 ). With the use of most currently available instru-ments, dPCR is not readily scalable; a sample can bepartitioned across multiple reactions to increase sensitiv-ity (and dynamic range), but this adds complexity andcost.

qPCR is also capable of measuring a very wide dy-namic range (�9 orders of magnitude), which meansthat essentially the same PCR setup can be used to quan-tify analytes that are present clinically at high, medium,and low abundance. This is not currently possible withthe majority of the commercially available dPCR instru-ments, because a different dilution strategy is required forhigh and low analytes to ensure the instruments are notsaturated, which complicates the protocol. To simplifythis, what is needed are dPCR instruments with at least100000 partitions, which will begin to compete withwhat qPCR has to offer.

Cheaper, Faster, Simpler Instruments withAutomated Analysis of Results

Although this requirement can be seen as stating the ob-vious, a specific point is noteworthy when dPCR is com-pared with qPCR. The automated reading associatedwith qPCR is a major advantage when applying thismethod clinically, such that it is often used for nonquan-titative molecular analysis. Because dPCR does not needto be measured in real time, to measure the amplificationcurve the fluorescence reading need not be as complex.Consequently this offers the potential for dPCR instru-ments that offer automated reading while also being sim-pler than those required for qPCR. Ideally this will leadto cheaper formats which will concomitantly reduce theturnaround time by reducing the time for instrumentsetup and analysis.

High Throughput vs Near Patient

qPCR has been applied in both high-throughput (sup-ported by robotic automation and/or miniaturization)and lower-throughput “near-patient” formats. Althoughthese developments arguably represent the pinnacle ofquantitative clinical molecular measurement, when usingqPCR to quantify, both are at the mercy of the amplifi-cation plot and Cq value. Furthermore, because moreautomation (for high throughput or near patient) resultsin the need for less expert input, the increased depen-dence on computer algorithms required to determinewhether a PCR amplification curve is suitable increasesthe challenge. The fact that dPCR is inherently simpler tomeasure, coupled with improved reproducibility, willlend itself far better to the analysis associated with auto-mation than qPCR.

Terminology

One of the factors that is likely to hinder translation ofany promising technology is the popular use of specificterms and/or acronyms among the early protagonists.One of the purposes of the dMIQE guidelines (33 ) was

Reviews


to use specific terminology to facilitate a common lan-guage and reduce confusion. The use of acronyms has itsplace, but they must be used sparingly and with a specificpurpose. What is increasingly unclear is why there is aneed to specify the fact that some instruments apply a“chip digital PCR” format (termed cdPCR) and others a“droplet digital PCR” format (termed ddPCR) when oneof the increasingly key findings is the impressive concor-dance between the 2 formats.

When qPCR was developed it was not felt necessaryto highlight whether the instrument used glass capillariesor a 96-well plate, which begs the question why this isnecessary for dPCR. The preparation required to set updifferent dPCR formats can be very distinct, but this isnot limited to chip and droplet formats and we advocatecomprehensive description of experimental proceduresin the material and methods or supplemental materialssections of any publication reporting the use of dPCR.To remove this confusion we would advocate using justthe term “dPCR” (rather than cdPCR or ddPCR), withthe format being apparent from the description of themanufacturer and make of the instrument. This can onlyassist in increasing the understanding and acceptance,and thus the likelihood, of eventual translation of suchmethods to the clinic.

Final Thoughts

If dPCR is developed into clinical tools it is likely to beinitially most successful in the lower-throughput for-mats, but if the procedure is simplified to improve auto-mation then there is no reason why high-throughputdPCR could not also be applied both in specialist areaswhere dPCR excels, like SNP detection, as well as routinequantification, like with viral load monitoring. Whencompared to qPCR the advantages of dPCR are many,whereas the only real disadvantages are those associated

with any young technology. This is an interesting timefor molecular diagnostics; if those who are developingsequencing technology rapidly succeed then it is likelythat PCR-based methods will be superseded for clinicalapplications. We predict, however, that routine clinicalapplication of sequencing-based technologies is likely totake at least 10 years to realize. If this is correct, then PCRstill has much to offer the clinic and dPCR is in a timelyplace to have a major impact on improving robust repro-ducible molecular diagnostics.

Author Contributions: All authors confirmed they have contributed tothe intellectual content of this paper and have met the following 3 require-ments: (a) significant contributions to the conception and design, acquisi-tion of data, or analysis and interpretation of data; (b) drafting or revisingthe article for intellectual content; and (c) final approval of the publishedarticle.

Authors’ Disclosures or Potential Conflicts of Interest: Upon man-uscript submission, all authors completed the author disclosure form. Dis-closures and/or potential conflicts of interest:

Employment or Leadership: J.F. Huggett, LGC; S. Cowen, LGC;C.A. Foy, LGC.Consultant or Advisory Role: None declared.Stock Ownership: None declared.Honoraria: None declared.Research Funding: UK National Measurement System and the Euro-pean Metrology Research Programme (EMRP), which is jointly fundedby the EMRP participating countries within The European Associationof National Metrology Institutes (EURAMET) and the EuropeanUnion.Expert Testimony: None declared.Patents: None declared.

Acknowledgments: We are grateful to Alison Devonshire and PhilipWilson for their critical review of the manuscript before submission.We would also like the thank Dr. Ion Gutierrez, and colleagues, at theNational Institute of Biology (Department of Biotechnology and Sys-tems Biology, Vecna pot 111, SI-1000 Ljubljana, Slovenia) for provid-ing the experimental data for use in Fig. 1.

References

1. Morley AA. Digital PCR: a brief history. Biomol DetectionQuant. 2014;1:1–2.

2. Cochran RL, Cravero K, Chu D, Erlanger B, Toro PV, Bea-ver JA, et al. Analysis of BRCA2 loss of heterozygosity intumor tissue using droplet digital polymerase chain re-action. Human Pathol 2014;45:1546 –50.

3. Gevensleben H, Garcia-Murillas I, Graeser MK, SchiavonG, Osin P, Parton M, et al. Noninvasive detection ofHER2 amplification with plasma DNA digital PCR. ClinCancer Res 2013;19:3276 – 84.

4. Jennings LJ, George D, Czech J, Yu M, Joseph L. Detec-tion and quantification of BCR-ABL1 fusion transcriptsby droplet digital PCR. J Mol Diagn 2014;16:174 –9.

5. Taly V, Pekin D, Benhaim L, Kotsopoulos SK, Le CorreD, Li X, et al. Multiplex picodroplet digital PCR todetect KRAS mutations in circulating DNA from theplasma of colorectal cancer patients. Clin Chem2013;59 1722–31.

6. Kelley K, Cosman A, Belgrader P, Chapman B, Sullivan

DC. Detection of methicillin-resistant Staphylococcusaureus by a duplex droplet digital PCR assay. J Clin Mi-crobiol 2013;51:2033–9.

7. Last AR, Roberts C, Cassama E, Nabicassa M, Molina-Gonzalez S, Burr SE, et al. Plasmid copy number anddisease severity in naturally occurring ocular Chlamydiatrachomatis infection. J Clin Microbiol 2014;52:324 –7.

8. Pholwat S, Stroup S, Foongladda S, Houpt E. DigitalPCR to detect and quantify heteroresistance in drugresistant Mycobacterium tuberculosis. PLoS One2013;8:e57238.

9. Sedlak RH, Cook L, Cheng A, Magaret A, Jerome KR.Clinical utility of droplet digital PCR for human cyto-megalovirus. J Clin Microbiol 2014;52:2844 – 8.

10. Sedlak RH, Cook L, Huang ML, Magaret A, Zerr DM,Boeckh M, Jerome KR. Identification of chromosomallyintegrated human herpesvirus 6 by droplet digital PCR.Clin Chem 2014;60:765–72.

11. Barrett AN, McDonnell TC, Chan KC, Chitty LS. Digital PCR

analysis of maternal plasma for noninvasive detection ofsickle cell anemia. Clin Chem 2012;58:1026 –32.

12. Lo YM, Lun FM, Chan KC, Tsui NB, Chong KC, Lau TK,et al. Digital PCR for the molecular detection of fetalchromosomal aneuploidy. Proc Natl Acad Sci U S A2007;104:13116 –21.

13. Beck J, Bierau S, Balzer S, Andag R, Kanzow P, SchmitzJ, et al. Digital droplet PCR for rapid quantification ofdonor DNA in the circulation of transplant recipients asa potential universal biomarker of graft injury. ClinChem 2013;59:1732– 41.

14. Higuchi R, Dollinger G, Walsh PS, Griffith R. Simultane-ous amplification and detection of specific DNA se-quences. Biotechnology 1992;10:413–7.

15. Cross NC, White HE, Muller MC, Saglio G, Hochhaus A.Standardized definitions of molecular response in chronicmyeloid leukemia. Leukemia 2012;26:2172–5.

16. Fryer JF, Baylis SA, Gottlieb AL, Ferguson M, VinciniGA, Bevan VM, et al. Development of working refer-



ence materials for clinical virology. J Clin Virol2008;43:367–71.

17. Bustin SA, Benes V, Garson J, Hellemans J, Huggett J,Kubista M, et al. The need for transparency and goodpractices in the qPCR literature. Nat Methods 2013;10:1063–7.

18. Sanders R, Mason DJ, Foy CA, Huggett JF. Consider-ations for accurate gene expression measurementby reverse transcription quantitative PCR whenanalysing clinical samples. Anal Bioanal Chem2014;406:6471– 83.

19. Vogelstein B, Kinzler KW. Digital PCR. Proc Natl AcadSci U S A 1999;96:9236 – 41.

20. Whale AS, Huggett JF, Cowen S, Speirs V, Shaw J, Elli-son S, et al. Comparison of microfluidic digital PCR andconventional quantitative PCR for measuring copynumber variation. Nucleic Acids Res 2012;40:e82.

21. Roberts CH, Jiang W, Jayaraman J, Trowsdale J, Hol-land MJ, Traherne JA. Killer-cell immunoglobulin-likereceptor gene linkage and copy number variation anal-ysis by droplet digital PCR. Genome Med 2014;6:20.

22. Tadmor AD, Ottesen EA, Leadbetter JR, Phillips R.Probing individual environmental bacteria for vi-ruses by using microfluidic digital PCR. Science2011;333:58 – 62.

23. Sanders R, Huggett JF, Bushell CA, Cowen S, Scott DJ,Foy CA. Evaluation of digital PCR for absolute DNAquantification. Anal Chem 2011;83:6474 – 84.

24. Strain MC, Lada SM, Luong T, Rought SE, Gianella S,Terry VH, et al. Highly precise measurement of HIV DNAby droplet digital PCR. PLoS One 2013;8:e55943.

25. Roma C, Esposito C, Rachiglio AM, Pasquale R, Iannac-cone A, Chicchinelli N, et al. Detection of EGFR muta-tions by TaqMan mutation detection assays powered bycompetitive allele-specific TaqMan PCR technology.Biomed Res Int 2013;2013:385087.

26. Didelot A, Le Corre D, Luscan A, Cazes A, Pallier K, EmileJF, et al. Competitive allele specific TaqMan PCR forKRAS, BRAF and EGFR mutation detection in clinical for-malin fixed paraffin embedded samples. Exp MolPathol 2012;92:275– 80.

27. Little S. Amplification-refractory mutation system

(ARMS) analysis of point mutations. Curr Protoc HumGenet 2001;Chapter 9:Unit 9.8.

28. Zhong Q, Bhattacharya S, Kotsopoulos S, Olson J, TalyV, Griffiths AD, et al. Multiplex digital PCR: breaking theone target per color barrier of quantitative PCR. LabChip 2011;11:2167–74.

29. Devonshire AS, Whale AS, Gutteridge A, Jones G, Co-wen S, Foy CA, Huggett JF. Towards standardisation ofcell-free DNA measurement in plasma: controls for ex-traction efficiency, fragment size bias and quantifica-tion. Anal Bioanal Chem 2014;406:6499 –512.

30. Sanders R, Mason DJ, Foy CA, Huggett JF. Evaluation ofdigital PCR for absolute RNA quantification. PLoS One2013;8:e75296.

31. Nixon G, Garson JA, Grant P, Nastouli E, Foy CA, Hug-gett JF. Comparative study of sensitivity, linearity, andresistance to inhibition of digital and nondigital poly-merase chain reaction and loop mediated isothermalamplification assays for quantification of human cyto-megalovirus. Anal Chem 2014;86:4387–94.

32. Shen F, Davydova EK, Du W, Kreutz JE, Piepenburg O, Is-magilov RF. Digital isothermal quantification of nucleic ac-ids via simultaneous chemical initiation of recombinasepolymerase amplification reactions on SlipChip. AnalChem 2011;83:3533– 40.

33. Huggett JF, Foy CA, Benes V, Emslie K, Garson JA, HaynesR, et al. The digital MIQE guidelines: Minimum Informa-tion for Publication Of Quantitative Digital PCR Experi-ments. Clin Chem 2013;59:892–902.

34. Pinheiro LB, Coleman VA, Hindson CM, Herrmann J,Hindson BJ, Bhat S, Emslie KR. Evaluation of a dropletdigital polymerase chain reaction format for DNA copynumber quantification. Anal Chem 2012;84:1003–11.

35. Bhat S, Curach N, Mostyn T, Bains GS, Griffiths KR,Emslie KR. Comparison of methods for accuratequantification of DNA mass concentration with trace-ability to the international system of units. Anal Chem2010;82:7185–92.

36. Whale AS, Cowen S, Foy CA, Huggett JF. Methods forapplying accurate digital PCR analysis on low copy DNAsamples. PLoS One 2013;8:e58177.

37. Dingle TC, Sedlak RH, Cook L, Jerome KR. Tolerance of

droplet-digital PCR vs real-time quantitative PCR to in-hibitory substances. Clin Chem 2013;59:1670 –2.

38. Hoshino T, Inagaki F. Molecular quantification of envi-ronmental DNA using microfluidics and digital PCR.Syst Appl Microbiol 2012;35:390 –5.

39. Kallioniemi OP, Kallioniemi A, Kurisu W, Thor A, ChenLC, Smith HS, et al. ERBB2 amplification in breast can-cer analyzed by fluorescence in situ hybridization. ProcNatl Acad Sci U S A 1992;89:5321–5.

40. Bhat S, Herrmann J, Armishaw P, Corbisier P, EmslieKR. Single molecule detection in nanofluidic digital ar-ray enables accurate measurement of DNA copy num-ber. Anal Bioanal Chem 2009;394:457– 67.

41. Dube S, Qin J, Ramakrishnan R. Mathematical analysisof copy number variation in a DNA sample using digitalPCR on a nanofluidic device. PLoS One 2008;3:e2876.

42. Tichopad A, Kitchen R, Riedmaier I, Becker C, StahlbergA, Kubista M. Design and optimization of reverse-transcription quantitative PCR experiments. Clin Chem2009;55:1816 –23.

43. Lam KW, Jiang P, Liao GJ, Chan KC, Leung TY, Chiu RW,Lo YM. Noninvasive prenatal diagnosis of monogenicdiseases by targeted massively parallel sequencing ofmaternal plasma: application to �-thalassemia. ClinChem 2012;58:1467–75.

44. Haynes RJ, Kline MC, Toman B, Scott C, Wallace P, But-ler JM, Holden MJ. Standard reference material 2366for measurement of human cytomegalovirus DNA. JMol Diagn 2013;15:177– 85.

45. White H, Deprez L, Corbisier P, Hall V, Lin F, Mazoua S,et al. A certified plasmid reference material for the stan-dardisation of BCR-ABL1 mRNA quantification by real-time quantitative PCR. Leukemia [Epub ahead of print2014 July 18].

46. Hayden RT, Gu Z, Ingersoll J, Abdul-Ali D, Shi L, PoundsS, Caliendo AM. Comparison of droplet digital PCR toreal-time PCR for quantitative detection of cytomegalo-virus. J Clin Microbiol 2013;51:540 – 6.

47. Henrich TJ, Gallien S, Li JZ, Pereyra F, Kuritzkes DR.Low-level detection and quantitation of cellular HIV-1DNA and 2-LTR circles using droplet digital PCR. J VirolMethods 2012;186:68 –72.

Reviews


considerations for digital pcr as an accurate molecular diagnostic

Documents