Thursday, May 30, 2019

The Genius of Plongle

I noted in my roundup of Oxford Nanopore technical announcements that I loved the Plongle concept.  This is really both a new application-specific integrated circuit (ASIC) and using it to build a 96-well sequencer.  Let me expand on what there is to love.

ASICs lie at the heart of ONT's technology -- digitizing the raw signal from the pores -- and have also been a major challenge.  The ASICs in the current instruments are expensive but must be close to the actual sensors.  For PromethION and MinION/GridION flowcells, this means the ASIC is in the flowcell and drives costs high and requires recycling the devices.  Flongle is a clever solution to separate the ASIC from a disposable "wetware" component, but it isn't clear it would scale.

ONT's solution is a new ASIC which they claim is cheap enough to be single use.  Able to monitor 400 channels -- more than Flongle but not quite MinION class -- the component is also inexpensive enough to not be fully utilized.  In the SmidgION cell phone sequencer, it would appear only a fraction of the channels would be in use.

But think about building the other way.  Clive didn't suggest it, but it is obvious that for many applications a new generation MinION flowcell using the new ASIC would be a real winner.  Sure, it might have only 80% of the number of channels and therefore 80% of the data, but MinION has shot past the data requirements of many users, with yields commonly around 20 gigabases and someone reporting 30 gigabases recently.   If flowcells were $200 each and had an expectation of only 15 gigabases per run, who could complain?

A new PromethION flowcell using four of these ASICs would seem likely, though I don't have all the numbers in front of me in terms of interconnects.  But what could also be interesting would be small format flowcells -- or perhaps just a small-format machine that took PromethION flowcells -- that had multiple ASICS and therefore very high simultaneous channel capacity.  MinION uses four-way multiplexing to increase overall throughput while needing only 512 recording channels.  But what if you have an application where speed is at a premium and therefore you would pay a premium?  Such as clinical applications.  In some circumstances, one might even be willing to sacrifice some accuracy for blazing speed such as pores running at 1000 bases a second.  

A cheap ASIC also suggests more opportunities to embed nanopore sensing in other devices and settings.  The PipettION idea I threw out before the conference.  Or perhaps nanopore devices integrated with other microfluidics than VolTRAX.  

I've mentioned the 96-well sequencer to a few colleagues and the response has been the same: where do I sign the PO?  In a high-throughput environment such a device would be welcome.  It gives yet another way to map samples to devices -- we already have time-based multiplexing by washing (or now nuclease flushing) flowcells and barcoding.  Now we could also have spatial.  Depending on your needs and pressures, each may be more-or-less desirable and they can be mixed.

In addition to my current shop, there's an obvious big customer for Plongle: ONT.  Such a device could be used to screen more pore mutants faster.  Or screen lots of sample prep conditions.  The latter is the perfect use case for many devices rather than multiplexing libraries; if one condition poisons the run you lose only that one well.  I could imagine in these situations going to even higher numbers of wells -- perhaps 384 is a stretch with a 400-channel ASIC, but in some cases just having a tiny number of pores could be enough to answer a question and the emphasis would fall on answering many such questions.

Another interesting possibility was raised by a Lightning Talk at London Calling.  Jeff Nivala from University of Washington demonstrated preliminary work showing that proteins can be designed that are mostly folded but have a 17-mer unstructured C-terminal tail.  That tail can be slurped into a pore on the flowcell by the electrical potential, but no further due to the folded protein.  But different tails can be designed to have distinguishable signals and once recorded the current can be reversed to eject the protein -- recycling the pore to look again. 

Approaches like that can measure interesting things with little or no library construction should be the bread-and-butter of this technique.  ONT has talked in the past of strategies using pre-defined probes (somewhat akin to Nanostring) to measure pre-defined analytes.  Reversibly linking nucleic acid tags to antibodies seems to be really taking off. 

Getting the design right will be important; all we have at the moment is an image of a mock-up.  Important questions must be explored with potential customers.  For example, the current mock-up is a large, unitary box that presumably has integrated compute.  Would anyone like to put the device right on a liquid handler deck?  If so, then perhaps a multi-piece device is what the market wants (which is the route Covaris took with their acoustic device).  I'm thinking more likely is that users will want to be able to easily add this to complex setups called "work cells", in which an arm-based robot moves plates between multiple machines.  Then a single box is fine.  But in either case, the ability for a robot to add and remove plate-flowcells and trigger the lid to open and close would be critical.

So many fun ideas to play with!  Will the plates be use-all-at-once, or could you perhaps run them in halves or quarters or a column at a time?  Leaving them out at ambient is the greatest, though the stability of the pores and membranes is remarkable, but for some applications maybe you'd want to launch one column now and the next in an hour and so forth.

Plongle: Loving the idea!  A solid concept with all sorts of directions for growth. It's going to be hard to wait over a year until they start rolling out!

1 comment:

Anonymous said...

Like your thinking. Lack of superfast detection methodologies for DNA barcoded antibodies and other things (ligands, tetramers etc.) is limiting the potential of the approach. Illumina sequencing or Nanostring detection of barcodes are just too slow and expensive. Once this issue is fixed many megaplex assays can take off.