Notes on Nanopores
I’ve recently been looking at novel approaches to DNA and RNA sequencing that might be disruptive in the next 10 years, in particular disruptively low cost approaches for use in pandemic response.
Nanopores could be a good fit here. So, I thought I’d start by reviewing the only commercially released DNA sequencing approach: Ionic current, biological nanopore sequencing. And thinking a little about how things might develop over the short to medium term.
What is Nanopore Sequencing?
Oxford Nanopore is one of only two nanopore sequencing platforms that are commercially available. But nanopore sequencing as a general approach has been under development for at least 25 years, with nano-scale imaging of DNA going back more than 30 years.
In this short series of posts we’re going to broaden the definition somewhat, not just considering “pores” but any kind of nano-scale direct detection/control method (including nano-gaps, nano-probes etc.).
The most obvious question is why do we need “nano” scale sensors at all? Well, DNA is nanoscale polymer, single molecule direct detection of that polymer is likely going to require features on that scale. In fact the base stacking distance in DNA is 0.34nm, so sub-nanometer features are attractive:
Ionic Current Sensing using Biological Nanopores
Nanopore approaches vary in terms of the sensing approach used, and the methods used to construct the pore. To date only two companies have commercially released nanopore sequencers: Oxford Nanopore and Qitan Technology.
Both these companies use the same basic biological ionic current sensing approach.
Ionic current sensing detects analytes through the blockage of an ionic current. Schematically this is shown below:
An ionic buffer solution is placed either side of an aperture under a bias voltage. Current flows between the electrodes transported by the flow of charged atoms (ions) through the aperture. In a nanopore, this current is on the order of a few hundred picoamps. We can get some sense of the magnitude of a picoamp by comparing it to the number of electrons required to carry this much current, roughly 6 million electrons per second.
Noise
This seems like a large number, until we consider how fast DNA translocates (moves) though the pore, and practical considerations around sensing small currents.
As each nucleotide blocks the pore and restricts current flow to a different degree. The difference between the highest and lowest blockage currents tends to be ~15pA to 40pA.
The best current sense amplifiers have a noise level of 0.145pA RMS at 10Khz bandwidth, this isn’t too fair off the thermal noise limit for such amplifiers… so it’s unlikely we can do much better than this.
Biological nanopores sense multiple nucleotides at the same time. Let’s assume that in most pores that will be ~5 nucleotides. That’s 1024 different possible states, if they were evenly distributed in the ~40pA range this would be a 0.04pA spacing, below our noise floor. Of course, we don’t need to accurately identify every possible state to accurately reconstruct the original sequence. But, this is still not an easy problem.
Current biological nanopore sequencing platforms sense DNA translocation at <1000 bases/s. Pushing past this would require higher currents, or larger current deflections and options seem limited here…
Due to the fundamental sensing limitations described above, it’s unlikely that this approach will be able to significantly exceed 1000bases/s using a biological nanopore/current sensing approach.
Slowing things down
Natively, DNA will translocate through a biological nanopore at millions of bases per second. As discussed above, this is far too fast to sense in a biological, current sensing nanopore. In order to slow down translocation, biological nanopore sequencing approaches have incorporated enzymatic motion control strategies.
The use of Phi29 for enzymatic motion control was published 10 years ago but the basic idea of using an enzyme for motion control was patented back in 1995 and has therefore long expired. It’s possible therefore that Qitan is sidestepping this IP through the use of a different enzyme (or it’s entirely possibly they just don’t care).
Apart from this enzymatic motion control strategy others have investigated different approaches. The most obvious example being Genia with their “nanopore-SBS” approach. Here they incorporate nucleotides with long labels which enter the pore:
The approach neatly solves two problems at the same time, the first is the use of a polymerase to control the detection speed and the second is the use of bulkier, easier to detect tags. However, while this is a neat solution in a world where direct nanopore sequencing exists I suspect it’s not a very compelling one, with no clear advantage.
There are other similar biological nanopore approaches too, where an enzyme is coupled to the top of the pore are conformational changes are detected during the incorporation process. Again, technically interesting, but likely very challenging and of limited benefit.
Arrays
For biological nanopore sequencing to be of practical utility you generally need 100s to 1000s of pores.
To break this down further. You need a surface, with well and electrodes on which you can build bilayers membranes. You then need to insert nanopores into these bilayers. If you randomly insert pores some bilayers will have no pore, some one and other multiple. Only those bilayers with a single pore will generate usable data (~36% of pores). It’s possible to compensate for this by adding additional wells (and multiplexing), or you can use methods to prevent multiple pore insertion.
A ONT Minion sensor array contains 2048 sensing regions. Not having a Minion to hand, I had to go by publicly posted images, but the active sensing area seems to be roughly ~5x13mm:
This is bonded directly to a sensing ASIC in a disposable flowcell. The chip size and pore count suggest wells are on the order of 150 micron. At volume, we would expect these dies, on a standard CMOS process to cost $5 to $18. Now… the Minion is no doubt not a CMOS standard process, but this gives some sense of the costs involved.
Oxford Nanopore list margins in the 50% range. Minion flowcells are listed at $475, suggesting a COGS (cost-of-goods-sold) of at least $200. ONTs financial reports indicate that they are more motivated by their higher end customers. This makes sense as nearly half their revenue in 2021 seemed to come from 56 PromethION (S3) customers.
The report generally describes Minion users as helping to develop the market for higher range instruments:
“Typically, they have a MinION, Mk1B or Mk1C and tend to have smaller scale and varied frequency projects across a broad range of diverse use cases. S1 customers are innovators who develop new uses for our products and publish their research and finding in scientific journals” … “MinION would benefit these users, in turn resulting in greater familiarity with the platform and opportunities to later develop into S2 and S3 customers.”
All this is a long way of saying I wouldn’t be surprised if the margins on the MinION were worse than average…
So, something needs to explain the difference in cost between a basic CMOS process and the MinION COGS. I imagine this is most likely a combination of a few factors relating to novel manufacturing processes, relatively low volume and poor yield on both the fabrication process and bilayer formation/pore insertion.
Oxford Nanopore have suggested that they can decrease well size to 40 microns, how this might translate into a COGS reduction is less clear. But I imagine, it would more likely be used to increase density, pushing the Minion throughput ~x4.
Conclusion
As stated up front, I’m mostly interested in thinking about how biological current sensing nanopore platforms might evolve over the next 10 years. It’s clear that Oxford Nanopore lead here, with Qitan only recently entering the market, seemingly only in China, and with a product that is inferior to Oxford’s.
But Qitan’s entry has made it clear that a nanopore sequencing company can be spun up relatively quickly. The review above suggests that key parts of the stranded sequencing approach are relatively old, and much of the IP has already expired. It seems at least possible that we could see innovation coming not just from ONT but other players too.
It seems unlikely that we’ll see much improvement in the basic sensing speed of these biological nanopore platforms which is likely limited to <1000 bases/s. Throughput improvements would likely come from larger arrays.
The aspect that is of most interest to me is cost optimization. So far ONT have shown only modest improvements in margins. Their cheapest flow cell (the Flongle) is perhaps the most interesting here, being essentially a single die which sits in a flow cell adapter. But the Flongle flowcells still cost $90 (suggesting at least $45 COGS).
Any route to a super-low COGS here could be compelling, and address the infectious disease testing applications I’ve previously talked about.
I’m still a little skeptical that a nanopore platform can be delivered with a COGS (per-run) lower than $45 and ideally more in the $1 range. But I suspect cost optimization is where we’re likely to see the most disruption. As always, comments are additional data point are always welcome, so please get in touch on twitter or via email (new@sgenomics.org).
Later this week I plan to look at the difficulties in developing solid state approaches to nanopore sequencing, so if that might be of interest consider subscribing!