Is Illumina Serious About an Alternate Chemistry for the Rapid Amplicon Market?

Back in January, at the end of my post on Illumina's new machine lineup I speculated whether Illumina might see a niche for a lower cost, lower throughput sequencing system that would slot below the MiSeq in their lineup.  Such an instrument, I posited, might go after applications in biosurveilance and diagnostics where relatively small amounts of data are needed quickly. I speculated that perhaps a smaller instrument with less expensive optics could compete in this arena, which is heating up due to Oxford Nanopore and the growing acceptance of DNA-based diagnostics.  As luck would have it, a few days later Molly He, Mostafa Ronaghi and colleagues at Illumina actually published a proof-of-concept paper for just such an instrument.  Unlike many sequencing technology PoC papers, this one demonstrates feasibility of reading actual templates (phiX rides again!). 

The paper is interesting in that it rethinks the Illumina chemistry in a fundamental way.  If I had to make an elevator pitch for it, I might say it borrows a bit from Illumina, Ion Torrent and PacBio concepts. Using Illumina's amplification technology, clusters are formed on a conventional Illumina flowcell.  Pure nucleotides are flowed across the clusters ala Ion Torrent or 454, and imaged with Illumina hardware or variations thereof. The signal is looking at binding kinetics of labeled polymerase during the flows, somewhat reminiscent of PacBio.

The paper builds on a trail of earlier papers by various academics showing that if DNA polymerase has bound the correct nucleotide for incorporation, a closed ternary complex (polymerase, template and nucleotide) is formed which lasts much longer than a more open (the technical term is apparently "ajar") complex if the nucleotide is a mismatch.  Previous studies used techniques such as stopped flow or crystallography to demonstrate this, but the Illumina group shows that this is resolvable in a sequencing instrument if the ionic strength is set appropriately. At low ionic strength.

The authors first go through some enzyme kinetics, explained clearly enough even I can follow them -- alas, as I remarked recently, kinetics theory does not come naturally to me nor did I work hard enough at it as a student any of the three times I had serious coursework touching on it. Boiled down the bare essentials, polymerase binds and then can bind a nucleotide.  If the nucleotide is correct, the polymerase enters the closed conformation, followed by addition of the nucleotide to DNA and release of pyrophosphate.  If no further polymerization is possible given available nucleotides, the polymerase dissociates from the DNA.  By labeling the polymerase, the polymerase-DNA complexes can be detected and these complexes have a longer lifetime if the correct nucleotide for incorporation has been presented. These concepts are illustrated in their Figure 1 (below).

A key aspect of the system is to discourage DNA-polymerase complexes except in the presence of the correct nucleotide, and this is accomplished by running the reaction at high ionic strength so that the DNA off-rate from the polymerase is high. A five-fold increase in discrimination between correct and incorrect nucleotides could be achieved by manipulating ionic strength (Figure 4, below)

After supporting the concept with simulations, they authors go and demonstrate it.  Reagents were delivered to a typical Illumina flowcell (on a GA I!) using a syringe pump, with imaging confined to only the region near the inlet port.  This is a conservative approach based on the fact that nearest the inlet since the farther from the inlet, the less predictable the reagent flow dynamics.  Example signal-level data is seen in Figure 5 (below) from a run with a single synthetic template

Bases could be called using only the maximum amplitude after a flow (with incorrect bases resulting in little fluorescence); the steady state amplitude figured into a simple homopolymer calling algorithm (as seen in Figure 6 below)

The system was tested both on a single synthetic substrate and sheared phiX library.  Reagents were flowed in 44 times in strict G,A,C,T order (Ion Torrent I know uses more complex orderings to reduce dephasing of clusters).  Reads were clipped to 20 bases for analysis purposes.  Raw accuracy is estimated as 95% for non-homopolymeric regions and 90% for homopolymers, with detectable sequence context effects.  The authors then look at some typical human gene amplicon panels and find that they are not particularly rich in homopolymers, suggesting this system might be capable of analyzing such.

It is stressed in the paper that this is a proof-of-concept, with little optimization. The hardware is not ideal; replacing the syringe pump with a pressurized system would lead to more consistent reagent delivery.  Polymerases and flow cell coatings have not been optimized and the base-caller is rudimentary.  They note that secondary structures may interfere with the system, and suggest exploring well-trodden paths such as DMSO, betaine and elevated temperature to address this.  Interestingly, they do go ahead and demonstrate data generation using much less expensive optics, replacing the excitation laser with LEDs (which is the light source used in some Illumina qPCR instruments). 

The huge advantage of flow chemistries is the ability to operate quickly, since the chemistry is simple.  The PoC chemistry had 45 second cycles (20 seconds chemistry, 5 seconds wash and 20 seconds pumping delay).  With optimized hardware, the authors state a belief that cycles could be on the order of 7-15 seconds.  With this and improved enzymes, they believe that complete library to result time could be driven as low as 1.5 hours with 100 basepair reads.  They also float ideas for non-optical detection and microfluidic chemistries, though these would clearly require substantial additional development.  Yet another concept is to use labeled nucleotides and unlabeled polymerase with real-time detection, much like PacBio's chemistry (albeit with clusters rather than single molecules).  More exotic, non-optical detection schemes are also mentioned.

Will Illumina actually develop such an instrument?  They certainly have advantages over a start-up trying a similar strategy.  The entire array of Illumina library preparation methologies and kits would be potentially available upstream, and Illumina has extensive experience with all of the technologies requiring optimization (fluidics, enzyme engineering and bioinformatics). The major risk is that this chemistry will be very different in quality to the existing chemistry, an issue which often faces companies launching re-thinks of existing technologies (a favorite theme of Christensen in The Innovator's (noun) series).

In other words, does Illumina risk marketplace confusion over their brand and the quality represented in exchange for potentially grabbing a bigger share of the growing amplicon sequencing pie?  Would Illumina launch another brand, or go so far as to spin-out a separate company to market such a device?  The rapid amplicon sequencing market has been one area in which Ion Torrent has appeared to gain traction, and now MinION offers extremely inexpensive and rapid data in this area, albeit with very high error rates.  GnuBIO has been quiet since being acquired, but was also gunning for this market, and quite likely Roche/Genia will aim for it as well.  Fighting for a big share of this market might require a bold move to fight possibly lower cost and/or faster entrants.  Will Illumina take the leap?