Current sequencing-by-synthesis systems rely on a flowcell, which is basically a fancy microscope slide with the ability to flow liquid through. Flowcells for continuous sequencing systems like PacBio and Oxford Nanopore don't need much flowing other than to start the reaction (or to flush it with ONT), but for cyclic sequencing-by-synthesis schemes from Illumina, Genapsys, Ion Torrent, and SeqLL (ex-Helicos) (and historically 454 and sequencing-by-ligation SOLiD and so forth), these flowcells see many flows per sequencing run.
As noted in the BGI preprint, this has a number of inherent challenges -- delivering liquid evenly across the flowcell. That must be performed via pumping mechanisms (pumps or pressure), valves and tubing. Flowcell shapes must consider the potential for eddies, dead zones and so forth -- and must be manufactured to not leak! Driving liquid through a tiny gap above the flowcell surface and its top cover requires increasing pressure as flowcell sizes increase. Increasing flowcell size is a tempting way to increase the number of bases per run, as detection limits -- density of sensors for electrical detection or optical resolution limits -- make packing clusters/beads/polonies/rolonies/DNBs tighter less and less appealing.
So BGI first went to dip tanks. No pumping, no pressure. The system was more factory / assembly line than a conventional instrument; an automation workcell devoted to sequence data acquisition. But those tanks represent huge amounts of dead volume to be filled with expensive reagents.
So in the new work they spread reagents very thinly across a polymer (PET; the stuff of soda bottles) tape. Amusingly, in this work the edges were delineated with a thin bit of Scotch tape, which also served as the spacer to control the contact with the sequencing slide. "Slot die" devices deliver a thin film of reagent onto the polymer tape.
There's a lot of details which could use better elaboration. It appears that each step in the chemistry gets its own zone on the PET tape, and it wasn't immediately apparent if these are reused or not -- but then they talk about how many meters of tape come on a roll and that this can support 300 cycles, so clearly each zone is used once. There's a video in the supplementary materials, but I didn't find it as helpful as it could be -- one thing that would help would be to mount the camera on a tripod so it doesn't move! It's hard -- at least with my eyes -- to see exactly what is going on. A side-on view movie would also be helpful. Registration marks on the Scotch tape would be great too so it is clear when the clear tape is moving and when just stages are moving. And while I'm kvetching on videos, AVI format seems to be poorly supported on major platforms -- two of the videos are in it rather than mp4 and are frustrating my attempts to watch them.
The preprint shows some basic statistics on the data. Near the edges things go badly, but most of the surface area delivers very respectable Q-scores using BGI's chemistry. The flowcells are large, with 225 square centimeters of operating surface -- 15 cm by 15 cm -- with 5x10^10 DNBs on the surface -- about 5-fold more than a NovaSeq S4 flowcell. Their estimate is that the surface coating technique reduces reagent waste from 99.9% to 99% -- with a hope to get down to 90%. Driving chemical reactions to completion by mass action seems to be a sausage factory item -- you're far less happy about it once you know the details!
MGI is touting fast cycling -- 1.5 minutes total of chemistry per cycle. They believe they can complete 2x100 chemistry in 12 hours, perhaps able to crunch that down to 8.8 hours. In that time they claim a single chip is good for 20 terabases of DNA. This is when the numbers get crazy. Since the system is broken into different stages, multiple chips could be run simultaneously. They propose that the strip would have 2 meters by 0.4 meters of surface area for reagents to service 266 of these chips, which would yield 8 petabases per day. But, there is a catch -- they would need 425 imaging subsystems. From this, and some claims about lower reagent losses to dead volumes in tubing, they are claiming a reagent cost of $15 for a 30X human genome is possible.
The idea reflects the Complete Genomics approach to sequencing -- a company BGI acquired many years ago. Build highly integrated sequencing workcells that generate huge numbers. The economics of this will always be highly centralized -- high capital cost instruments, specialized technicians to keep it running and huge throughput on which the cost estimates rely -- you're not going to get that $15/human WGS running one genome a day! I'm not totally averse to such an approach -- we love to show off our integrated workcells at the Strain Factory. But it is certainly a very different model than individual sequencers widely distributed to labs that want to use them.
But I do always like any sort of playing with boundaries of how to build a sequencer. The reagent delivery on films seems like something that might see other embodiments. Have ink jet heads been explored for reagent delivery in an optical sequencer? Or some sort of squeegee sliding across the surface?
Late addition note: for those of us in the U.S., it's very much in doubt we could get one of these anytime soon -- Illumina has tied BGI up in knots with patent litigation.