In this month’s installment of The Primer, we’re going to take a more in-depth look at a topic briefly mentioned a few months back: digital PCR. This is a technology that is increasing in popularity and may be coming to your lab soon, so it is deserving of a more detailed look.
By analogy: counting the beans
In keeping with this series’ stated goal of avoiding heavily mathematical explanations, we are going to start off with a simple, intuitive analogy for how this process works. Imagine we have a sack of jelly beans, and a bunch of small containers. We don’t know how many jelly beans are in the sack, but we do know how many containers we have (let’s say 1,000, for sake of argument). We’d like to know how many jelly beans there are. One small twist: we can tell if one of our containers is empty, or if it contains a jelly bean or beans, but we can’t distinguish how many beans are in a container that is not empty. Our analysis is thus “digital,” in that it can report only one of two states for each container—empty or non-empty.
Our approach is to start by randomly distributing jelly beans from the sack into our 1,000 containers, until the sack is empty. Now, we look at each of the containers and score whether it’s empty or contains bean(s). If we count the full 1,000 containers and get 900 empty and 100 containing bean(s), what can we conclude? An immediately obvious interpretation is that the sack contained at least 100 jelly beans. Notice, however, that most of the containers are empty. Since we distributed beans to containers randomly, this tells us that there are many more containers than there are beans; and if we consider this, it’s logical to assume that as a result, it’s unlikely (not impossible, but unlikely) that we put more than one bean in any container. If we assume that each non-empty container thus only has one bean, we’re left with the conclusion that the sack contained exactly 100 jelly beans.
Notice that an important part of this logical process was the observation that most of our containers were empty. If only a small number of the containers were empty, then it’s intuitively obvious that some of the non-empty containers likely contain more than one bean, and our “one bean per non-empty container” assumption is no longer accurate. (If this isn’t intuitively obvious, consider an extreme case where we have fewer containers than beans. Empty containers would be rare or non-existent, and non-empty containers must contain on average more than one bean.)
Now let’s go even further with our analogy. Did we need to count all 1,000 of our containers to get our 100-bean answer? Suppose we counted 100 of the containers at random. On average, we’d have counted 90 empty and 10 “bean-positive” containers. We’d know from this smaller sample that empty containers outnumber non-empty containers (our one bean per non-empty container assumption is valid), and we’d know that 10% of the containers are non-empty. Since we know there were 1,000 total containers, 10% of 1,000 gives us our answer of 100 again—but this time with 90% fewer measurements.
There are two important questions in assessing the validity of our approach:
- What fraction of containers need to be empty to allow us to reasonably assume each non-empty container has one bean?
- As that fraction changes, what’s the likelihood of a non-empty container having more than one bean?
Conveniently, all of the above has been studied in detail by Siméon Poisson, a French statistician who not only predated digital PCR, but probably also jelly beans (1781-1840). His work, most familiar in the context of the Poisson Distribution, addresses precisely these issues. It even gives us an adjustment factor to apply as the ratio of non-empty to empty containers rises; that is, it allows us to account for the impact of the uncommon cases where a non-empty container has more than one jelly bean.
Once you’ve grasped the preceding, you’ve got the theory behind digital PCR; what’s left is a method to apply this approach to selected DNA sequences. These sequences are akin to the jelly beans of the analogy; they’re the discrete items within the sample which we would like to quantify. Our method of detecting these sequences will be PCR, using sequence-specific primers and examining for evidence of successful primer amplification as defining a “non-empty” container. As for the containers of our analogy, we can use techniques from microfluidics to portion a larger input sample (plus target appropriate PCR master mix, with primers, polymerase, nucleotides, and other components) into tiny separate droplets. Different physical approaches to this are used by differing instruments, ranging from creating suspended emulsions of tiny droplets in an oil medium, to surface or capillary-based microchannel systems. Regardless of the exact physical approach, each of these methods creates a very large population of a known number of tiny individual PCR reactions. The digital PCR instrument now thermocycles these micro-reactions (all of them, or a known fraction of the total; in either case, a very large number) and examines each micro-reaction in parallel for evidence of PCR amplification.
Digital PCR and its advantages
This detection is done in current systems by fluorescent methods, in the same way classical qPCR “real-time” devices work (covered in the June and July installments of The Primer). As with traditional qPCR, either dsDNA binding dye reporters such as SYBR Green or probe-based methods can work here. Although either endpoint or real-time data can be collected in a digital PCR instrument, in this case we’re looking only at the final result for each micro-reaction as “PCR positive” or “PCR negative.” This is because CT doesn’t have meaningful significance here, as we’re assuming each micro- reaction to have just one template in any case. If this leaves you wondering, why bother collecting real-time amplification data at all in this context, it’s because the amplification curve shape can still yield information on how well optimized the reaction is, or how a known reaction is performing on a specific sample.
By having very careful control over fluid volumes, our digital PCR system “knows” how many total micro-reactions were made, how many were analyzed, and what fraction of those analyzed were PCR positive—that is, non-empty of template, in the terms of our jelly bean analogy. Software algorithms now take this data and apply a Poisson correction based on the ratio of non-empty (PCR positive) samples to empty (PCR negative) samples, and provide a final result in terms of absolute number of PCR target templates present from the original sample. Knowing what this sample volume was, we also know our target template concentration in the input sample.
Digital PCR has numerous advantages over traditional qPCR. A highly significant one is that no external standard needed to be run in order to get our quantitative result; digital PCR provides an absolute quantity, directly. Strictly speaking, use of at least one known concentration calibrator or standard is still suggested, particularly when trying to compare results from different laboratories, as factors such as polymerase activity can influence whether all template-containing micro-reactions successfully amplify. Another advantage of digital PCR over traditional qPCR is that it is less susceptible to measurement errors arising from partial sample inhibition. Samples which amplify slowly due to such inhibition give late CT values in qPCR, appearing as lower than true concentration; however, as the reactions proceed to positivity, in digital PCR, this endpoint is still captured and true concentration is determined.
Probe-based digital PCR can be multiplexed in the same manner as traditional real-time qPCR, through the use of different spectrally resolvable fluorophores for each target. Variations on this approach are commercially available with the capacity of multiplexing up to ten targets per reaction. Compared to mulitplexed traditional qPCR, digital PCR may perform better when we are trying to quantify very low levels of one target alongside much higher levels of another target material. This arises because in traditional PCR squelching (sequestration of reagents) by the stronger reaction can suppress the weaker reaction in the same tube; digital PCR, by contrast, works at limiting dilutions where each micro-reaction is generally only positive for one assay template by itself and is therefore free of reagent competition.
Of course, as the saying goes, “there’s no free lunch,” and compared to traditional qPCR, digital PCR has some downsides as well. The instrumentation required, with precise microfluidics, is at present more expensive than many real-time fluorescent qPCR systems. As covered in the preceding discussion, it requires sample dilution to a particular range so that a low but not insignificantly low fraction of micro-reactions are PCR positive; even with application of Poisson adjustments, this occurs across a relatively narrow concentration range. Thus, application of digital PCR requires either an initial concentration estimate, or else being repeated across a range of sample input dilutions with data analysis from the most statistically viable data set. In contrast, real-time qPCR reactions can work well across as many as nine logs of concentration range and thus don’t require pre-titration to a functional range in most cases. (Note that this has implications for digital PCR multiplexing, in that it will work best in cases where the multiple targets do not have wildly dissimilar concentrations. If they do, then different dilutions have to be evaluated for different targets for optimal accuracy.)
When traditional qPCR methods arose, that was followed by the development and general acceptance of MIQE Guidelines (“Minimum Information for Quantitative PCR Experiments) outlining what experimental parameters should be reported in a qPCR method to assure its validity. Coming from this background, digital PCR has been quick to have a technology-specific version suggested in the form of the digital MIQE guidelines. (See Huggett JF, et al. Clin Chem. 2013;59(6):892-902.)
With its many favorable characteristics, digital PCR is already in clinical use in some applications, such as minimal residual disease monitoring for leukemias, and it will likely become more widespread in the near future. With user-friendly instrumentation handling the microfluidics and statistical analysis components of the technique, it provides a simple yet highly robust tool for the clinical setting, which is increasingly looking to replace purely qualitative MDx results with quantitative measures of higher informational value.
John Brunstein, PhD, a member of the MLO Editorial Advisory Board, is President and CSO of British Columbia-based PathoID, Inc.