Tuesday, February 11, 2020

Sampling Current & Future Directions in PCR Diagnostics

My qPCR explainer seems to have done relatively well, though it took some refinement after readers caught a number of errors.  The most embarrassing of those is that I got my PCR ramp units upside down, so instead of 4 seconds or so per degree C it's degrees C per second so my times were off by a factor of 16! Ouch!  Despite that miscue, I'm here going to explore some of the variants on PCR that are out there, including some that are being employed searching for the newly renamed COVID-19 virus.  Included here are some of my own speculations and musings, so as always remember I'm someone who thinks about these things and sometimes talks other people into running them, but I haven't set up a PCR in 8 years.  Also, the field of PCR variations for diagnostics is enormous and I don't claim to have anything near complete knowledge of it, so this should be seen as a sampler and not a comprehensive review.  Also, the usual reminder I am a paid consultant for a diagnostics company but they are neither aiming at viruses nor using PCR, so I won't discuss them -- but if you feel that shifts your priors on how I treat other companies you have the information to do so.

The Perfect Diagnostic

qPCR is far from the perfect diagnostic scheme, but what would be?  The perfect scheme would require zero specialized equipment, inexpensive to produce (single digit dollars per test), no specialized training, sufficiently sensitive yet selective to provide clinical benefit, could be stored at ambient temperature and can turn results around in minutes after sampling crude biological fluids with no workup.  It would be reasonable to think that this combination is something beyond science fiction, except I've just described a standard home pregnancy test.  But those took decades of refinement, so we can't expect such a test overnight for a new viral outbreak

qPCR Rated

Now we have some meter sticks.  qPCR uses specialized equipment and it costs tens of thousands of dollars for the typical lab machines.  They also require significant power and are quite large; you might mount one in a van (I'm unclear on the vibration sensitivity) but you still need a lot of power.  The skill to operate is modest; certainly any good pathology lab technician could learn to run an assay.  That's a good thing; if testing needs to be scaled up there is a large set of individuals with the correct skillset, and it is easy to train more.  The CDC test does require prior purification of nucleic acids.

There are lyophilized reagents on the market, so eliminating the cold chain is plausible.  A rough guess of the materials cost is around $1 per well, so the CDC qPCR test requires about $4 of materials per patient sample. Remember that many patients or their samples will be retested, so patients x $4 will significantly undershoot the total cost.  There still aren't published estimates on the sensitivity and selectivity; that's particularly a challenge when we are still figuring out what the underlying frequency of infection is and when in the course of a disease a person will shed detectable virus.  In U.S., it was announced today that one patient who had been released from quarantine actually tested positive; whether this was a test failure or an administrative one hasn't been revealed.

So there's definitely a number of ways the test could be improved.  Now, in any process engineering effort you shouldn't be greedy: pick one, possibly two axes to improve on and don't be surprised if you must give something up on others.  So now let's see some options that are out there or could be

Axes for Improving on the CDC Test


I haven't seen a proper sample-to-answer estimate for the CDC test, but it appears to be somewhere near two to three hours.  I'm happy to take corrections on that.  Faster tests can mean releasing uninfected (or recovered to the point of zero virus) individuals to free valuable hospital space.  For screening at borders, long waiting times are highly undesirable because they generate backups.  

For PCR assays, there are a number of variables that affect the overall time.  If you can reliably detect a positive with fewer cycles, that is likely to shave the most off.  Designing your assay to work at higher temperatures enables both less time ramping temperature and shorter extension cycles, as the polymerase works faster at higher temperature.  The CDC's extension times are conservatively long; shortening them if that can be validated could save time.  Having faster ramping thermocyclers is another option.  That could mean more efficient heat transfer, different heating and cooling schemes, or heating and cooling smaller volumes


As noted, the CDC test requires prior DNA isolation, which is likely performed in a biological safety cabinet (CDC guidelines for the virus recommend this if any process that can create aerosols, such as centrifugation, is used).  Having a true single tube assay -- dump the patient sample in a tube, throw some reagents in and seal the tube (or well on a plate) -- that's clearly more attractive from a simplicity (less risk of operator error) and operator efficiency.  


There's a lot of concern that just having potentially sick and not sick individuals queue together at central testing stations.  There's also the potential challenge of screening at every airport or other port-of-entry.  So a portable device could be very useful. 

In The Fray: Cutting-Edge Devices for The Current Crisis

A Luggable from HKUST

Several media stories have reported on a device out of Hong Kong University of Science and Technology for detecting the virus.  In the image in a story, it appears to have a built-in microfuge.  It's a luggable device -- portable but not handheld.  The device is also described as using very fast heaters to reduce cycling time. It appears the device runs a single sample at a time, though the descriptive video is a bit weak on those details. It also appears that the microfluidic chip for the PCR stage must be squeezed in a special vise tool, and just the right amount -- in real operation one wonders how often incorrect execution of this step leads to either downstream failure or a smashed chip.  Detection modality isn't obvious, but is probably qPCR-like.

One curiosity is that some news reports on the HKUST device are showing a very different instrument than the one in the HKUST video.  Different prototypes or really different instruments?

Macau's Virus Hunter

A group at University of Macau has also gotten press for a device named Virus Hunter they have developed in concert with a company called Digifluidic.  This company specializes in digital, electrowetting microfluidics - -the same technology underlying Oxford Nanopore's VolTRAX and Illumina's withdrawn NeoPrep.  Chips for this exist as a grid of squares, with each square's hydrophilicity changed by altering the applied voltage.  If a droplet is on a square that just became hydrophobic it will move to an adjacent hydrophilic square, so drops can be blipped around the device (given my age, the devices remind me of early video games and I hear Pac-Man sounds in my head when I see them in operation)

The small volumes enabled by digital microfluidics appear to allow for very short ramp times.  The device is advertised as running 12 assays per machine for about $7 a test and run times of 20-50 minutes.

Veredus: PCR with Microarray Readout

Veredus is a company in Singapore which has announced an assay that detects the new virus as well as SARS and MERS.  Their VerePLEX Biosystem benchtop device can handle five microfluidic cartridges, each apparently running the assay for a single sample but it appears a separate device images the built-in microarray that serves as the readout mechanism.  Analysis time is advertised as two hours

Biomeme: Handheld qPCR

Biomeme is a U.S. company that has developed a handheld, battery-powered "Franklin" qPCR device along with integrated test strips.  They even have a nucleic acid purification kit for the field.  They also seem unique among U.S. diagnostics startups in developing a kit for the new virus.  For $30/sample one can test for two PCR amplicons.  Bulk reagents are available lyophilized, and the ready-to-go kits do not  require a cold chain.  This is labeled research-use-only on the website; it's unclear whether they are seeking an emergency use authorization from the FDA

Other Options & Ideas

T2 Diagnostics: Integration

T2 Diagnostics is a Boston area company that someone on Twitter pinged me about.  It's a spinout from Robert Langer's lab at MIT.  They have a  instrument that accepts biological samples and uses PCR plus detection with NMR -- probes attached to paramagnetic beads substantially alter the NMR signal if hybridized to PCR product

So the device definitely wins on integration and claims to be very sensitive.  It's a big expensive machine and does not appear to handle multi-well plates of input  but it's hard to get a read on this. Their copy claims same day turnaround for their existing infectious disease panels, so not aiming for high speed.  They do claim very high sensitivity for blood borne pathogenes, 1 CFU per mL.

T2 has not announced a diagnostic for the current outbreak.  In general they are challenged on the business side, with a current share price under $1 (so on the "Pink Sheets" where nobody wants to be) and just under $40M market cap.

Ontera: Competitive PCR 

Another company that hasn't announced an effort for the outbreak but certainly has interesting technology for it is Ontera, formerly Two Pore Guys.  Ontera has proposed assays for other rare nucleic acid analytes by combining competitive PCR with their nanopore based detectors.

In competitive PCR, two different targets are amplified with the same primers.  Typically this means that the primers directed at the real targets are in very limiting amounts but have 5' tails which can be amplified with the common "driver" primers. Alternatively, one could spike in a synthetic standard that amplifies with the diagnostic primers but carries a radically different payload.  When the primers run out, the ratio of the target of interest to the competing target can be used to estimate the original ratio between the two. 

Ontera can use probes to the target (e.g. virus) and competing amplicon to detect it with their handheld nanopore detectors.  Let's emphasize: while Ontera has indicated an ultimate goal of sequencing, these would only be calling out presence of specific analytes.

Their existing device claims to go sample-to-answer in 20 minutes -- there's an integrated thermocycler -- and is a small desktop box with a cartridge to handle the sample. 

If you are an academic and would like to play with their detection technology, there's an access scheme for a device that is just a detector (NanoCounter)

Digital PCR

In standard qPCR the characteristic curve formed during the reaction is used to derive the concentration of the targeted molecule, but without a standard there is ambiguity in the absolute amount.  Digital PCR uses Poisson statistics (which if you're unfamiliar, there is a wonderful explainer that maxes out the pun potential) to actually count molecules. 

The basic idea is that if you take a sufficiently dilute input sample and divide it into discrete vessels, then only some of the vessels will actually receive template.  After running PCR, you can just count the fraction of compartments that generated PCR product and back-calculate the original concentration of the analyte. 

Digital PCR has not really skyrocketed, probably because the field tends to feature expensive instruments and many have multi-stage workflows.  For example, BioRad has a system based on PCR-in-droplets which has separate making, amplifying and readout steps.  So that would be a serious worsening vs. the qPCR assay. QIAGEN has a new instrument, which I believe they are launching at AGBT at the end of this month, which uses special PCR plates in which the plate itself divides the sample up.  Fluidigm at least used to (haven't kept up) have a device based on their flexible microfluidics chips.  And I can imagine ways of repurposing sequencing technology such as bridge PCR.

Digital PCR assays are typically just like qPCR assays with two primers and a probe; otherwise primer-dimers or other amplification artifacts might trigger a compartment to be falsely called positive.

I don't have experience with digital PCR.  What might be tempting is if you can reliably get a signal after many fewer cycles of amplification.  It also might be possible, using a standard primer trick called 5' GC-clamps, to use faster amplification cycles for most of the run.  Remember that after the first few rounds the vast majority of amplifications are copies-of-copies.  So if the PCR primers have GC-rich 5' tails, the annealing temperature for primer-to-copy binding can be reliably higher -- so the cycles can be run much hotter.  With qPCR this is a problem in that you can't use this trick for the probe, though there are modifications such as minor groove binding moieties that might increase the melting temperature some.  But with digital PCR, you only care about endpoint signal -- and half of that in qPCR would be delivered on the final cycle.  So -- and again, this is speculation that would require lab experimentation to show it actually works -- it could be imagined having a PCR program with several early, cooler annealing+extension cycles (55C)  to initiate copying and then shorter, hotter cycles (say 72C) to drive the reaction and finally a cycle or two back at the low temperature to generate the final signal. 


One bit of variation in qPCR thermocyclers is how many dye colors -- and which dye colors -- they can detect.  The CDC assay is set up to use FAM, the most common dye.  But it is possible to run two or three or four PCRs in the same tube.  So one could imagine multiplexing all four amplicons in a single assay to quadruple the volume a single thermocycler can handle.

First drawback is it can be tricky to design four PCR reactions to work in the same tube without interfering with each other.  At a minimum, the amplicons shouldn't overlap.  Second drawback is this limits the number of thermocyclers that can run the assay; some devices can't handle the additional colors.


I've been lukewarm on Oxford Nanopore's VolTRAX digital microfluidics device but now could see a great opportunity if it was a bit further along.  VolTRAX has built in cycling and the possibility of qPCR in a future release; alternatively if the sequencing-on-VolTRAX were working than that would offer another route to a portable device that realistically could take purified nucleic acid and then work unattended until a result were generated. 

Oxford has talked about opening up VolTRAX to developers; if that were already the case I'm sure someone would have done proof-of-concept for virus detection. 

Closing Notes

Isothermal Schemes

There are a number of diagnostics schemes that eliminate the need for a thermocycler by using an isothermal amplification reaction.  These use polymerases other than Taq, as instead of 5'->3' exo activity it is critical to have strand displacement activity to push earlier synthesized strands away so a new strand can be made.  I recently had to skim this field for work reasons, but I'm no expert and there's significant disadvantages versus PCR.  In particular, the temperature cycling requirement of PCR means also being able to tune numerous parameters by changing the temperature.  Isothermal methods can also be challenged with non-specific amplification and specificity issues, particularly if running at relatively low temperatures.  Most require a specified temperature; you still need heating just not the more complicated heating and cooling.  Some schemes require very complex primer schemes involving four primers, or complex loop structures.

Because there's no exo activity, something other than probe disintegration must generate signal.  Some use fancy CRISPR tricks -- there's one CRISPR effector that once it binds its RNA target turns into a beserker which shreds any RNA in the vicinity; artificial RNA can have probe-quencher pairs which release a signal in this way.  Or you can go totally low tech: generating a precipitate from all the pyrophosphate released by the amplification reaction.  Curiously, while there is an isothermal analog of bridge PCR, I haven't found an isothermal analog of digital PCR -- perhaps when trying to get rid of the thermocycler it isn't appealing to add back in digital partitioning equipment.

That's a pretty weak summary of the field, but hopefully better than nothing.

Examples of companies working in this space are Sherlock Biosciences and Mammoth Biosciences; infectious disease diagnostics pioneer Charles Chiu has been working with Mammoth technology for detecting coronaviruses.

Validate, Validate, Validate!

The biggest challenge with any new technology or tweak on an existing technology is proving it really works in the field.  I can imagine mocking up RNA by run-off transcription off of gBlocks or similar fast-and-cheap synthetic DNA products.  There's probably companies out there that will sell qualified researchers sputum -- not the most appealing product line but certainly important in this case.  But ultimately tests need to be fielded against real samples with known truth.  The initial centralization of U.S. testing has been criticized for being slow (the CDC has since expanded the program to many labs), but the one plus is the CDC should have built a useful panel of known positives and negatives. 

But initial validation isn't sufficient; there needs to be regular re-checks.  An example of my thinking here -- I don't know if the authorities concur -- would be to submit to sequencing any samples which repeatedly yield ambiguous results (1-2 but not 3 positives in the CDC panel).  If there were rogue strains of the virus that had mutated their primer or probe binding sites so as to frustrate the assay, you'd want to discover that!  Similarly, negative samples from feverish patients should at least be periodically submitted to more exhaustive search, including metagenomic sequencing.  Again, my thought here is to backstop against the potential for viral forms evading the PCR primers.


Anonymous said...


Anonymous said...

Interesting thoughts and commentary here, but the size and timepoint discussion isn't really true for the state-of-the-art platforms. The CDC assay has to work on validated, approved ABI 7500Dx-style systems which are a bit old; totally understandable from a clinical Dx perspective but the future is already here so to speak. Roche cobas Liat is pretty tiny and runs qPCR tests under 30 minutes (https://diagnostics.roche.com/us/en/products/systems/cobas-liat-system.html), Abbott's isothermal ID NOW is also very small and can do sample-to-answer FluA/B in 13 minutes (https://www.alere.com/en/home/product-details/id-now-influenza-ab-2.html). Plenty more coming along.