Tuesday, May 21, 2024

Thoughts on A Decade of Oxford Nanopore Sequencing

I'm writing this the eve of Oxford Nanopore's London Calling conference.  This is a big one, as this summer marks the 10th anniversary of ONT releasing devices into the wild.  It's been a long, interesting journey and I'm much too jet-lagged to try to review old posts or even link to them, but a bunch of thoughts have been in my head the last few days. 

Oxford keeps experimenting with the format of the conference.  Traditionally it was one day of training sessions and two days of talks; this year it is one day of training, two "Innovation Days" and one "Clinical & Biopharma Day" - so three separate tickets.  Much of the "pre-event" training is aimed at novices, but this year there were some courses aimed at experts.  And, apparently, ONT had a "Data After Dark" component announcing software updates that replaces the prior "Data for Breakfast" - and since I use the pre-day for touring / dealing with jet lag, I was in a pub near Parliament instead of in the session.  A bit of ONT's product strategy execution in a nutshell: complex product and not fully documented, but lots of quality material.

Remembering MAP

I was lucky to have gotten Warp Drive into the original MAP (MinION Access Program) cohort, and it was an interesting ride.  There was the stress of awaiting the first shipment, then greater stress seeing the battered overpacking.  There was the tension of watching our expert library preparer make the lambda control library, intial deflation when very little data came out and it was mostly short noise - and then the swooning over a clear full-length 48 kilobase lambda read.  

There was a long wait until the second shipment of reagents and another to the third (and then my memory fails).  There also seemed to be a strong falloff of data yield the farther your lab was from Oxford UK - which turned out to be insufficient sealing of a key reagent to protect it from infiltration by sublimated carbon dioxide.  But many soldiered through, though one group treacherously published a hit piece based on the first data and was stricken from the program.

There were other odd aspects.  The original MinKNOW dumped only FAST5 format; there was no tool to extract FASTQ or FASTQ.  Nor was the FAST5 structure documented - it's a self-describing hierarchical format so that's no too bad when the thing of interest is a long string composed only of A,C,G and T.  Three converters soon showed up in the community implemented in just as many languages - Mick Watson's in R, Nick Loman's in Python and mine in Julia (half the useful Julia progreams I ever wrote; I forget what the other did). Mick's and Nick's soon gained wide acceptance; I was quite comfortable seeing only one other person use mine and it quickly fell into non-maintenance.

But by a year or so, two things were clear.  First, MinION could generate tens to low hundreds of megabases per run of reads that were genuinely long, though very error rich.  The long part made them immediately interesting and useful. Second, ONT's had made an absolutely perfect decision in  targeting MAP to mostly unknown, junior researchers rather than traditional new sequencing testing grounds at big genome centers.  A quirky, low-throughput device would be routinely ignored by nearly all of those big centers, but a self-selected subset of MAP users looked for whatever MinION angle could be interesting and exploited these for preprints and then full papers, all the while pushing out open source tools.  That group became the core evangelists for the technology, and many participated in the PoreCamp training sessions that spread the nanopore message further.

The Lawsuits

ONT also was embroiled for a time with a number of lawsuits with Illumina and Pacific Biosciences.  Illumina's was the most dangerous, targeting the pore used in the intial kits and there was belief at the time that it might be the only viable protein nanopore for sequencing.  ONT pulled a rabbit from the hat with another pore, but without that it was truly an existential crisis.  ONT and PacBio traded suits for methods of error correction from multiple reads, and eventually settled.  That was a contributor to knocking out ONT's original scheme for reading both strands - the company claimed after-the-fact that they were going to abandon it anyways for technical reasons.  

ONT Today

Fast forward to today.  Ten years of access is a double-edged sword: it is clearly a major achievement to be lauded, but ONT is a publicly listed company and investors are likely to start being impatient for the company to break-even and then turn profitable.  ONT's special provisions for blocking takeovers have gone away or go away.  That doesn't mean they will be sold, but the possiblity is now higher.  It's also not uncommon to see experienced ONT folks pop up at startups. For example, Dan Turner often gave very interesting technical updates, but was at AGBT giving a talk for his new Hi-C diagnostics company.  That's not an unusual or worrisome phenomenon at a low rate - if you've hired ambitious, skilled people you can't be surprised that their ambitions may run ahead of what ONT has space to offer in terms of enlarged opportunities.  But it is one more challenge to the company with institutional knowledge periodically walking out the door.

The biggest challenge to ONT is finding a good product strategy; this is a difficult task for every company.  Many of the applications that ONT has demonstrated are very exciting and likely benefit humanity, but aren't the markets that will drive them where their investors want to go.  It's cool to sequence on the International Space Station or in an Antarctic ice field, but those aren't big markets.  

Even markets that sound appealing might be very challenging to launch into.  At the very first Nanopore Community Meeting, I remember two convincing stories of how ONT sequencing might be feasible in the clinic and would yield important gains for patients.  But the challenge, particularly in the US healthcare system beset with insurers and Pharmacy Benefit Managers, is actually showing value to the people who will prescribe it and pay for it.  Nine years later, nanopore sequencing-based diagnostics are still in the future.

ONT, like many companies, responded to the SARS-CoV-2 pandemic in a positive way.  I don't have good statistics anymore, but when I looked several years ago it was clear that ONT was punching above its market share, not just in the UK but in many companies in the Global South.  ONT also released the intriguing LAMPore assay, which combined isothermal LAMP amplification with MinION sequencing.  In theory, this would allow equipment-free amplification feeding into actual sequencing of a variant-specific region of the viral genome.  But the question arises again, who wants that information?  For the most part, just public health trackers, and they preferred whole genome data.  Plus the getting informative mutations was never really show convincingly.  When LAMPore was announced, I thought after the pandemic ONT would release more assays for things like crop diseases - but the stark reality is that users there would probably be satisfied (or even prefer) with a colorimetric LAMP assay, and ONT rightfully has no interest in products unrelated to nanopore sequencing.

Other products have struggled, at least at first.  PromethION represented a significant technological change over MinION, and had a long and complex birth - but now has been a workhorse for many labs for multiple years.  The VolTRAX microfluidic sample prepration device never seems to quite catch fire; it also had a long and painful, multiyear run of missed launches and then problematic consumables.  The end product is a device that doesn't thread the "product-market fit" needle very well - too expensive for many applications such as education, too few samples and too much hands-on manipulation for clinical or industrial use.  

There's been other touted ideas with even less chance of success, particularly since ONT rarely seemed to focus on them.  There was talk of non-enzymatic DNA synthesis, rumored to use VolTRAX technology, that never saw light-of-day.  There was the great theater of Ubiqibopsy, with CTO Clive Brown being sampled on stage and then sequences showing up - very few sequences - by the end of the talk. But what's the use case?  IF point-of-care clinical, what regulatory strategy could ever get such a device to market>?  There was the idea of actually sequencing within a VolTRAX cartridge, supposedly to exhaustively read out the mRNA from a cell.  But if the trend in the field is wanting to read more and more cells, is there a market for this?  There is especially the DNA storage concept; I think Clive will never forgive me for my trashing of that - every time I attempt to revisit I return to the same issues: ill-defined market, bad fit to what market has been suggested and low data content.  Even ideas that might have legs - some sort of adaptation for peptide sequencing - risk being significant distractions.

Could ONT just string together multiple successful niche applications?  For example, with their quality continuing to improve, could ONT challenge PacBio in the rare disease space>  Can their methylation detection be applied to cell-free DNA diagnostics?  That's a crowded field and I was very skeptical that ONT could work with the tiny amount of DNA coming from a standard blood draw, but a recent paper shows proof-of-concept for just that.  And ONT has had some interesting successes - I often carp about their commercialization - the silly splitting of ordering and account management on two websites, the lack of next-day delivery of consumables, past problems with delivering wrong versions of kits, a too complex product lineup with poor documentation and guidance, a confusing user community as an utterly inadequate substitute for professionally written and organized materials - sure, I can lay it on.  But ONT is apparently doing better sales than PacBio and unlike essentially every other company in biotech hasn't announced a major layoff.  I'd still argue all those faults hold them back from better success, but maybe I'm wrong (I know I'm not, but gotta throw that out there!).

Clive's big talk is on the first day, Wednesday, so we will find out early what new evolutions and revolutions the company is planning.  There appear to be a strong mix of talks on all three days; to be honest I don't see anything resembling a sharp division between "Innovation" versus "Clinical & Biopharma" talks.  As a concrete example, there are rare disease talks both times.  The Innovation days will also feature most of the originators - David Deamer, Hagan Bayley, Mark Akeson & Dan Branton (absent is George Church, though admittedly his concept for nanopore sequencing is much less similar to what has emerged than what the others developed).  

So stay tuned - it should be interesting.  The first decade of nanopore sequencing has never been boring; I doubt the second decade will be either.


Anonymous said...

Is there a world where ONT overtakes illumina? If i were to judge based on this blog post i would think that ONT has near 0 chance of being a sucessfull profitable company and is gonna bankrupt in 3 years lol. I think there Is a reason why they are still growing revenue while the comp is not, but only time will tell.

William Gardner said...

Another great post, I love learning about sequencing history from your blog! My job is primarily viral sequencing as a service and I just performed my first ONT sequencing run on a minION last week. I have only used Illumina (MiSeq) in the past and ONT feels like a breathe of fresh air. For my application, Sequence Independent Single Primer Amplification (SISPA), ONT blew Illumina out of the water. It cost about half of what the same run would have cost me on Illumina and yielded hundreds of thousands of reads (albeit not at the same quality as Illumina). The sequencing only took 1 day vs the MiSeq's typical day and a half. Finally the size limit of Illumina short reads requires SISPA samples to be tagmented while ONT reads such a wide range of lengths. Moving forward I would prefer to use ONT so maybe there is hope.

P.S. I have yet to call ONT customer support but Illumina is tough to beat on that front!

Dale Yuzuki said...

Great insight on ONT.

As PacBio and ONT have collectively about only 7% of the market (per some market research I have access to) it is hard to imagine them being able to break out of a 'niche' position.

With the likes of Complete Genomics and Ultima Genomics driving the per-Gigabase cost way below the NovaSeq X (currently at $2.50, while Complete and Ultima are at $1.00) is there any real hope of long reads gaining genuine traction? It seems as if (with PacBio's recent guidance for 2024 downward, and I see ONT projecting 'only' 13% year-over-year growth.

Anonymous said...

For me, the highlight of the three days (innovation + clinical) was Danny Miller’s vision of ONT wgs as a one-and-done diagnostic for rare disease cases. His demonstrations show that Nanopore technology is now up to the task, although uptake will face many challenges other than technological.

Anonymous said...

Watching companies like PacBio throw so much money into marketing and parties while they turn around and layoff 30 percent of the workforce is a testament to their poor leadership. ONT will overtake PacBio in a matter of years. https://x.com/pacbio/status/1797603109050458191?s=46&t=rErEtfScjaA4P01KxITWxw