Bench philosophy: All or None
by Steven Buckingham, Labtimes 06/2013
Tiny droplets are used as reaction chambers in droplet digital PCR.
Everybody knows digital is (most often) better than analogue. This holds true not only for computers but also for some biological applications.
It didn’t take long after PCR burst onto the scene for it to dawn on biologists that quantifying the PCR technique would open up a whole host of new possibilities. After all, plain old vanilla PCR just tells you what sequence is present in your sample, but not how much. At best, the only clue as to how much is there comes from looking at the size of the smear on the gel – a crude measure at best. Then along came quantitative (real-time) PCR. Dilute your sample to an appropriate level, run PCR as per the instructions on the packet but stop the reaction every so often and see if you get any detectable product. This tells you when the product reached a threshold, providing in turn a relative measure of how much product you started off with.
This is fine and works well in many situations. But it falls down when the amount of the target sequence is very low. And as for being quantitative, well, it only gives you a relative measure, not an absolute one, raising the need for some sort of external calibration. But there is another problem, which arises from the very power of the PCR reaction itself: because it is a chain reaction, it not only amplifies your target but also low levels of near-miss sequences as well. That doesn’t happen if you have a lot of your target to start off with, because the target amplification simply out-competes the junk. But when target and junk start from a level playing field, you are just as likely to amplify primer dimers or pseudogenes, for example.
So, if your target is rare, what you would ideally want is to find some way of increasing its relative concentration. That sounds like putting the cart before the horse but a little bit of lateral thinking shows how it can be done: by decreasing the volume of the reaction. If you take the reaction volume right down, you might, just by chance, get a reaction mix with one, perhaps two, molecules of your target, along with much less of the junk.
This, in essence, is the idea behind digital PCR (dPCR). Split the reaction into thousands of tiny reactions (partitions) until stochastic statistics kick in. You dilute the reaction mix to start with and if you get the dilution just right you will, by the laws of statistics, end up with one molecule of target in some, but not all, reactions. Then you do your PCR amplification and just count the partitions that have a product. The logic is simple: if at least one molecule of target is present, there will be a reaction product, otherwise there will be none. All or none; one or zero. Hence digital.
Sometimes, of course, there will be more than one molecule of target in a reaction – won’t that skew the results? At such low levels, assuming you have got the dilution more or less right, the number of molecules in individual partitions follows a Poisson distribution. There is a simple statistical trick called the “Poisson correction” that can take account of this.
There are several advantages of digital PCR over “traditional” qPCR. Some are typical advantages of digital over analogue: the relative insensitivity to noise, for example. Digital is clean: you just decide whether something is there or not, yes or no. The PCR chain reaction, like a bi-stable switch, lends itself to just this sort of approach. dPCR also gives you the absolute levels of the target – the proportion of droplets with product, divided by the total number of reactions, gives you the absolute concentration of target molecules. So, no need for calibration curves.
It is also less liable to off-target false positives, because you are dealing with such small volumes, which mean lower concentrations of pseudogenes etc. dPCR is also robust against another bug-bear of qPCR – the destructive effects of inhibitors.
So what is the best way of getting the small volumes? Forget buying in thousands of 386-well plates, you need to deal with thousands, if not millions, of micro- or even pico-litre reactions here. In practice, there are two ways you can achieve this: chips or droplets. Chips physically divide up the reactions into tiny chambers, which are then analysed in parallel. But there are only so many miniature chambers you can fit onto a reasonably-sized chip and this effectively reduces the dynamic range of the assay.
Enter microfluidics: use oil to separate your reactions into an emulsion of tiny droplets. The advantages of droplets over chips are obvious – you can do more, and faster. The microfluidics cupboard is stocked full of tricks and techniques to make uniformly-sized aqueous droplets and move them about, merge them, split them, heat them and measure them. You also run less risk of contamination – a major issue when you are dealing with single molecules – because your reaction mix never sees the same chamber twice!
So much for the theory – is there actually any evidence that dPCR out-performs good ol’ qPCR? Surprisingly, this question has been given scant empirical evidence, until recently. This year, Chris Hindson and his colleagues gave an answer to the question in a Nature Methods paper (doi:10.1038/nmeth.2633). The group did the obvious test: take some RNA sequences and prepare your own gold standard material, and see who does best at detecting and measuring it – a dPCR/qPCR head-to-head.
The team took six human microRNAs and performed nested, serial dilutions, which were then put through droplet dPCR and a standard qPCR. They used synthetic oligos, so they knew what they were probing for and ran the experiment in two conditions: one in which the oligos were diluted in pure water and another where they were diluted in plasma RNA from a healthy human donor.
They found that while dPCR was no more sensitive than qPCR, it was very much cleaner – the coefficient of variation between the reported concentrations of target and the actual values was some 80% less for digital. This advantage held, regardless of whether plasma or water was used as the diluent. They then ran the two methods head-to-head on sera from people suffering advanced prostate cancer compared to healthy controls. This is because advanced prostate cancer is signalled by the presence of one of the microRNAs in the blood. Digital PCR achieved a p-value of less than 0.005, compared to more than 0.1 for qPCR.
Admittedly, Hindson works for BioRad, a manufacturer of a droplet dPCR device. But the general consensus is that dPCR is, indeed, cleaner and more reliable than real-time PCR when you are trying to detect small levels, where it is emerging as the method of choice. Keith Jerome, for example, of The Vaccine and Infectious Disease Institute at the Fred Hutchinson Cancer Research Center, Seattle, WA, reported that when they tested samples they had deliberately seeded with inhibitors, they found droplet dPCR much more robust than qPCR (www.clinchem.org/content/early/2013/08/28/clinchem.2013.211045.full.pdf).
And there is no lack of applications in this area crying out for better methods of detecting rare copies. Take HIV, for one example. Earlier this year, Matthew Strain and others reported in PLoS ONE that the very features we have talked about make it an attractive approach for measuring HIV in clinical DNA samples (PLoS One 2013; 8(4):e55943). In another application, Dany Morriset and colleagues showed how the selling points of dPCR can help to detect something sniffy in our food – in this case, markers for genetically-modified organisms (GMOs). They claim that it was cheaper and more reliable in detecting GMOs in the complex, dirty matrices we find in tins on supermarket shelves. Any horsemeat in there, I wonder?
But as with all new techniques, several notes of caution need to be sounded. dPCR may solve a lot of qPCR’s problems but it doesn’t solve them all. For instance, there is nothing to stop sequences similar to the target occasionally still being amplified, such as pseudogenes, resulting in false positives. There is also a lower limit to the rarity of the sequence that can be detected: if a sequence is only present at one molecule per 100 microlitres, you may not pick it up unless you run your dPCR on at least that volume. You also have to get the dilution factor right – too much or too little and you lose accuracy.
And of course there are the usual dangers when you work on quantal numbers – for instance, if the target molecule “drops out” of your reactions, you will critically underestimate its abundance. Finally, things can also get a bit complicated if it is DNA you are after rather than RNA, because complementary strands of denatured DNA might partition into more than one reaction. Indeed, the dangers of overconfidence in the technique have provoked the recent publication of a set of guidelines for the minimum information for publication of dPCR papers (Hugget et al., 2013, Clinical Chemistry 59:6).
If you are unsure of whether to leave the qPCR comfort zone and launch out into the uncharted waters of dPCR, fear not. Manufacturers of dPCR devices are aware of this anxiety and their trend so far has been to market devices capable of both qPCR and dPCR. The four main players in dPCR technology are Life Technologies, Fluidigm, BioRad and RainDance. The first three offer chip-based devices, whereas BioRad and RainDance take the droplet approach.
So, should I go for chips or droplets? You may be tempted by the price: any dPCR machine will set you back about €22.000 to €150.000 but chip-based devices are around 50 % more expensive. Droplet devices have considerably higher throughput, while manufacturers of chip devices claim their approach has higher accuracy.
Last Changed: 10.10.2013