Thread by @jackwadden

Super excited to finally share this pre-print of our work demonstrating <30 minute sequencing-based somatic variant calls using @nanopore sequencers!

We had a need for speed. A (long) thread... 1/

https://twitter.com/biorxivpreprint/status/1394348745437835264

Why focus on speed? Intraoperative sequencing. Being able to quickly (<1hr) identify actionable mutations *during surgery* has a host of amazing potential use-cases. Imagine generating a personalized molecular diagnostic during biopsy… 2/

...combining biopsy/resection surgeries in real-time, confirming indeterminate frozen section results, intra-operative clinical trial enrollment to test targeted therapies, molecularly targeted chemo delivery directly into resection cavities. The list goes on… 3/

Why sequencing? qPCR assays are highly specific (and fast), but only target a single mutation and don’t really scale/multiplex well. Especially considering oncogenes where any mutation in any location is of interest (e.g. TP53), targeted amplicon sequencing is the way to go. 4/

Sequenced reads are also interrogable, and can directly quantify variant allele fractions (tumor content). Other stuff like Raman Spec. (super cool tech) exists but is also currently a bit narrow wrt discovery power 5/

Our research question eventually emerged as “Can we call a variant in <1hr?” and “how low can we go?” @nanopore MinION stood out because it offers 1) immediate, streaming access to sequencing data and 2) rapid (~10min) library preparations. 6/

An on-paper “workload characterization” of existing targeted sequencing protocols pointed to target amplification (i.e. PCR) and ligation library prep being the major bottlenecks to sub-1hr variant calls. So first question: can the rapid kit work on short (<300bp) amplicons? 7/

Yes! We varied amplicon size and hotspot location over a target of interest (HIST1H3B K27M) and showed that the rapid kit works (with acceptable efficiency trade-offs) down to 187bp product. Do you lose info via fragmentation? Yes. Is that fine? Also yes! 8/

So how can we reduce target amplification time? (other than optimizing PCR cycling parameters etc… which we did)? Needle-in-a-haystack analogy time... 9/

PCR exponentially amplifies the “needles” in the target genomic “haystack”. If you can amplify needles, when should you stop amplifying needles (PCR) and start searching the haystack (sequencing) to guarantee an optimal time-to-result? 🤔 10/

It turns out this is a pretty easy optimization problem: 1) model ONT sequencer 2) characterize assay performance over time, 3) identify amplification time that leads to the lowest end-to-end diagnostic time. We call this optimization technique “Threshold Sequencing” 11/

Threshold sequencing models indicated our assay produced time optimal results after 28 cycles of PCR. The resulting protocol (from tissue to 250x mut+wt variant call) took ~52 minutes. Sub-1hr 😎 #nice 12/

BUT….PCR is still like, a whole *50%* of total diagnostic time! That's still a big bottleneck! Can we do better? In comes LAMP 13/

Loop-mediated Isothermal Amplification (LAMP) is...er….difficult to explain over twitter, but tl;dr it’s a much faster target amplification approach than PCR and generates concatemeric amplicons (allow me to coin a term: lamplicons) 14/

One major downside to LAMP is that its complex machinery can malfunction and generate self-amplifying, spurious product. For some assays, we saw lots of amplification, but most of it was unmappable garbage. So what is going on? In comes LAMPrey (github.com/jackwadden/LAM…) 15/

LAMPrey is a tool specifically designed to diagnose and analyze LAMP concatemer product. It’s loosely based on the seed/chain/align paradigm with some cute visualizations that help identify when the LAMP machinery malfunctions. 16/

It helped us quickly diagnose product and figure out that one of our assays was indeed misbehaving, resulting in a large fraction of useless reads even though it was amplifying like a champ. 17/

We soon after realized that LAMPrey was also recovering much more (>40% !!!) diagnostically relevant information than standard minimap2 alignment, which was exciting because more recovered reads == faster diagnostic. 18/

So we applied the Threshold Sequencing methodology to our new LAMP assay (also using the LAMPrey tool) and it predicted ~14 min of amplification time was optimal (~2x faster than our PCR assay). Now we’re cookin' with gas 🔥🔥🔥 19/

Coupled with a few more DNA extraction and library preparation shortcuts and a steady-ish hand we were able to verify a known H3F3A K27M mutation (250x mut+wt support) from tumor tissue within 30 minutes. 20/

Other LAMPrey benefits: lamplicon polishing (combining concatemer sections to improve aligned read error rates from ~96% to >99%). We have some other fun projects in the works utilizing LAMP and the LAMPrey tool, so stay tuned! 21/

This was an amazing, fairytale interdisciplinary collaboration between the Precision Health research group @UMichCSE 🖥️ and @KoschmannLab 🧠🏥@umichmedicine @Umichpeds 22/

@KoschmannLab works tirelessly to fight pediatric brain tumors and develop and improve targeted diagnostics and therapies. They’re the real MVPs and I’m super excited to continue collaborating on development of novel rapid diagnostics to improve patient outcomes 23/

Also, huge, huge thanks to my @UMichCSE PIs for letting me buy a DNA sequencer (lol) and set up a wet lab (lol) and sequence DNA (lol). How amazing is that!!! @UMichCSE is at the cutting edge and @nanopore is so easy, a computer engineer can do it. 24/

Finally, if you're interested in hearing more, come check out my poster at @NanoporeConf #nanoporeconf #LC21 END/