Sunday, February 22, 2015

Can Ion Torrent Buzz Again?

In my AGBT 2015 Preview / Speculation at one point had a tightly packed (and overly long) paragraph on Ion Torrent, but I realized that this was a symptom of trying to to cram too much in too little a space -- plus I really had a lot more thoughts worth unpacking.  So here's a long form look at Ion Torrent -- with plenty of references to past AGBTs to make writing this now apropos.  One advance bit of excuse making: the historical background that follows is not intended to be a comprehensive history of Ion Torrent technology, but more of an impressionistic sketch (but as always, my worst excesses and omissions are fair game for comments!).

An Impressionistic History

When Ion Torrent was an independent company, CEO Jonathan Rothberg mimicked Steve Jobs in style  The first Ion Torrent talk I went to (in Fall 2009) was all about Moore's Law; I was slow in realizing that his time was going to run out without any actual information on the platform. At AGBT 2010, just a little more information was released, but Ion continued to try to build a reputation as an edgy, not-business-as-usual sequencing platform, with lower cost-of-entry than extant platforms (which at that time was the Illumina GA2, Roche 454, Helicos Helioscope and ABI SOLiD).  Ion would have a community for users to exchange protocols, with prizes for hitting significant milestones.  This would be the machine of garage startups, a platform to support the emergence of a Hewlett-Packard or Apple Computer. A few instruments would be given out to the community (3 in the U.S., 3 in Europe).  Performance upgrades would come from new consumables, not new boxes ("The Chip is the Machine"), radically changing the long-term cost of ownership. Unmodified nucleotides (and the lack of the expensive enzymatic cascade found in 454) would be cheap but high performing. These consumables would be launched at short, regular intervals, enabling semiconductor-like upgrades.  Finally, the platform was based on what Rothberg touted as old semiconductor technology; nothing bleeding edge there.

However, the polish didn't stay long on Ion's carefully constructed street rep. The big surprise was Illumina launching MiSeq, which radically lowered the entry point for the already dominant Illumina platform.   After a lot of initial back-and-forth, MiSeq would end up decisively superior on nearly every performance metric (anyone wishing to argue that 1x400 is superior to 2x300 paired end is welcome to do so in the comments).  Careful analysis of the upfront costs of the Ion Platform showed that MiSeq was still more expensive to buy into, but by only a factor of around 30%.  The initial two follow-on chips would be somewhat delayed, and sequencing performance suffered from a high rate of indel errors.  Further enhancement beyond the 318 chip would seem to require herculean leaps of density, with the fear that cross-talk between wells would become a serious issue. This issue would be resolved -- at the cost of discarding "The Chip is the Machine" -- by launching the Ion Proton with much larger chips.  New equipment started appearing to deal with the unloved emulsion PCR template prep, adding convenience at the cost of further upfront investment.

One of the claims of Ion I have always found counter-intuitive (but I never claim my intuition is infallible) is that the small, but measurable in Ion's hands, superior mismatch error rate for Ion over Illumina is due to presenting the polymerase with only one nucleotide species at a time.  I would think that competition is better for driving precise nucleotide selection.  Alas, a proper comparison is utterly impractical to do.

Proton launched with shorter reads than PGM (which the Ion team has never hidden, but isn't trumpeting either).  Ambitious projections for the PII and PIII follow-on chips initially excited, with the PIII promising HiSeq-like performance at MiSeq prices. Unfortunately, it became increasingly apparent (with apologies to the late, great Cornelius Ryan) that Ion had attempted A Chip Too Far.  PII, just like Nijmegen, turned out to require far more resources to conquer, and far worse completely derailed the operational schedule.  Rothberg's old boasts about semiconductor technology boomeranged badly, as the Ion team had difficulty getting data off the chip sufficiently quickly, which cross-talk between wells raised its spectre again. The follow-on of  PIII , as with Arnhem, went from a spectacular coup de main to a valiant display of failure by overreach.

So Ion has settled in on a few key markets which appreciate its fast chemistry, or some early believers who stuck with the system.  The HiQ chemistry, developed from an extensive screen of mutant polymerases (an important truth about sequencing companies: all of the current ones are backed by impressive protein engineering shops).  This market wants reliable, high-performance chemistry and is unafraid of paying for automation to support it.  Hence, the cleverly-named Ion Chef template prep machine has been a focus. This market isn't into flash or sizzle, and its not surprising that after what felt like a farewell speech at AGBT 2013, Rothberg departed Life Technologies for new ventures. Life would subsequently be bought by Thermo Fisher, so a technology that once defined a company and then became the chance for a second company to regain its sequencing market glory, was now just one more business line in a huge R&D supply conglomerate.  PGMs are now selling at auctions for a substantial fraction of list price (as I write this, a complete PGM kit with OneTouch and server offered on eBay for $14K or $27K; at a recent auction I followed a solo PGM went for around 15% of the list price)

What If?

One of the great What If? questions (in the alternative history sense, not Randall Munroe sense -- though has anyone thought about having him sketch out a sequencer?) for Ion is if someone other than Life Technologies had bought the company.  As I had written at the time, it seemed obvious that some big player in the genomics space would buy the company, as there was little evidence Ion on its own could build the commercial infrastructure to build, supply and market a widely sold sequencer.  Life Tech did embrace Ion nearly whole-heartedly, throwing their best technical group on it and marketing the system heavily, but still retained SOLiD and never tried to bring those platforms together in any way (say, with shared library preparation kits).

Life Tech also seemed to never make acquisitions to support the Ion platform, whereas rival Illumina has consistently made excellent acquisitions.  Imagine if Ion had snagged Nextera technology by acquiring Epicentre; that would have made for an interesting balance of power in the small sequencing market, with Ion having easy library prep vs. Illumina having easy template prep.  Instead, Illumina held the lead in both.  Later, Life Tech let Advanced Liquid Logic fall into Illumina's hands, another pioneer in novel library preparation technology, though this was in the period of Thermo Fisher digesting Life Tech, so perhaps the post-merger chaos snuffed any chance of that opportunity being seen.

Roche/454 would have been an obvious alternate candidate, since that Rothberg-developed platform also had unmodified nucleotides and emulsion PCR.  But, there seemed to be some serious bad blood between Rothberg and 454 which would prevent such a scenario.  Given the genomic community generally jaded view of Roche's handling of new technologies, perhaps that would have gone no better. It is easy to speculate that Roche would have pushed read lengths to the 800bp+ seen with the best 454 chemistries, so as to hold onto customers who built assays around that length.

Ion could have been a great entree for a company without a high-throughput platform -- perhaps GE trying to re-enter the sequence field or Agilent being a player. Maybe Bio-Rad would have jumped in sooner, rather than their later acquisition of GnuBIO. Was it shopped to QIAGEN?  Perhaps then we would have seen a PGM launch in early 2017.

Going Boldly Again?

Since Ion's launch, other companies have stolen pages from the playbook.  For example, Genapsys (which uses a similar unmodified nucleotide chemistry as 454/Ion), had flashy (but data lite) AGBT 2013 presentation, promising a $20K instrument with low running costs and high performance with the form factor of an iPad (alas, said instrument hasn't emerged yet).  Oxford Nanopore really executed on the plan; while Ion promised widespread early access but had very conventional betas, MinIONs have been going to almost anyone but the usual cast of genomics early adopters.  This has misfired sometimes, but mostly been a stunning success -- MinION papers are appearing in the peer-reviewed paper at an increasing pace, well under a year since the launch of MAP. 

Could Ion realistically regain some of the excitement?  I think it could, but it would take some truly bold moves.  First, any such effort would need to be separate from the main company, ideally in a spin-out, so that it can make decisions without any consideration of existing markets.  

My vision for such a product would first try to shrink the instrument.  The form factor for the PGM was a bit bloated when it first launched - one gets the impression the box was designed early before the innards were nailed down, and has an awful lot of empty space.  As I discovered at the first starbase, in a small start-up bench space can quickly become a tight commodity -- the ideal box would be compact (toaster sized?) and capable of operation when on a shelf of reasonable height -- but said box would use existing PGM chips; they're the expensive bit of the supply chain you don't want to mess with. Other consumables could be repackaged as needed. 

Second, the whole ecosystem would need to be bare bones and cheap, but functional.  If you need to trade, performance takes a reasonable hit if it means significantly lower cost-of-entry and/or less bench space. So the isothermal template prep replacement for emPCR, which isn't of high interest to the core markets since (in its current form) means lower sequencing yield, would be critical; an Ion Chef or OneTouch is more upfront money and more space.  The Torrent Server compute requirement needs a serious re-look; we're more than 2 Moore's doublings since the original platform launch so that should already be priced at a quarter what it was.  Can one or more Raspberry Pi sub in? Could any of this processing be moved to the cloud ala Oxford's Metrichor?  

Could such a vision for a cheap (sub $20K for all-in equipment purchase?) platform with low consumable costs (but perhaps an additional revenue stream from the cloud platform?) really work?  I'd like to think so, but the window is closing. While Oxford hasn't announced pricing outside their MAP, they've suggested a few $K per instrument and a few $K per run (perhaps $1K each, the cost of entering the MAP).  Oxford is still working on accuracy, I have the bias of thinking a single molecule technique is going to be inherently nosier than a molecular ensemble approach, but is making astounding strides (such as this fresh Akeson group paper).  But a $10K-$20K machine leveraging reads which don't require extensive correction (if HiQ really delivers; I haven't tested it personally) for a few $100 a run might bite off some new niches. If Roche can get Genia working, or Genapsys really works, or any of a host of other uber-cheap sequencing techs take off, Ion will be a technology in twilight.  Time to make hay while the sun is shining

4 comments:

Unknown said...

Great post!
I used to work at a small company on software development for base alignment and variant calling. SOLiD and Ion Torrent data both were a challenge.
We had invested a lot of time in SOLiD data processing, and it was quite a shock to hear that SOLiD is discontinued.
I'm not surprised to read that the Ion Torrent platform is facing difficulties. I never understood their marketing bullshit about how it would improve with the advances of semiconductor technology. It was enough that my boss was convinced.

James@cancer said...

Nice post Keith. I can't wait to read your look-back on ONT in a few years!

Andy Vierstraete said...

We are a small lab with an Ion Torrent PGM. We are running it as service facility for other labs. We purchased it in 2011 to sequence genomes, transcriptomes, amplicons,... in the biology department. The chip capacity would increase because "the chip is the machine". A few months after it was installed, they announced the proton because the 318 chip seemed to be the limit for the PGM.
To our frustration, there is almost no interest anymore for it because it is to expensive. Only very small projects are done on the system. If one 318 chip is not giving enough data, it is cheaper to run the project on a MySeq or Proton.
With the release of the proton, our hope was that the price of the 318 chip would decrease, but they keep it high. The PI (15Gb) chip is 250€ more expensive than the 318 (2Gb) chip, and is giving 7 times more output... People who want 4 or 6 Gb of data go to an other system because it is cheaper than running 2 or 3 318 chips. That is our frustration that LifeTech does not compete in price setting with other systems or compagnies.

ping said...

thnx a lot