State of the MinION: a pore update

[Script for the talk]

It's been a while since I talked in general about nanopore sequencing at the Malaghan Institute, and I figured that maybe it's time to pull out some of my old pictures and revisit some claims I've made about nanopore sequencing over the past few years.

Representations of DNA

To get myself warmed up, I'll talk a bit about sequencing. Despite what TV shows will tell you, DNA is not a bunch of letters on a computer screen.

Those letters represent a physical thing: fluffy, stringy whiteish stuff that frequently has a gel-like consistency and an amazing ability to clump together with itself.

DNA sequencing is the process of turning the physical thing of DNA into a model of that thing on a computer. All methods of DNA sequencing have their own underlying model about how DNA should behave.

The most common physical three-dimensional structure of DNA is a right-handed double helix, with hydrogen bonds providing a weak linking between complementary bases. If the precise details of that structure are not important, we can use a model of a single strand of DNA, a polymer chain that is composed of some basic building blocks.

This is about where nanopore sequencing sits. In my opinion it's as close as we can get, for now, to a model-free sequencing process: an observational, single-molecule method of sequencing.

Sequencing By Shape with ImageJ

I'm going to drill down a bit into how that signal is converted into a letter on a computer.

Imagine there is a device that can scan across an image of DNA, like a photocopier, and work out how much black there is in each line of the image. What you end up with is a squashed down representation of the DNA, showing how much stuff there is at a particular place.

As the shape of the DNA changes, the amount of signal at a particular location also changes.

If our line detector were sensitive enough, then we would be able to distinguish the four different bases just on this profile alone. A and G look similar, as do C and T. Similar, but a well-trained eye would be able to tell the difference.

Depending on whether or not you believe chemical structures to be a reasonable representation of physical reality, it may not be too surprising that the most common base call error the nanopore sequencers make is mistaking an A for a G (and vice versa).

Sequencing Using a Nanopore

But how do we get from the physical thing to the sequence?

The Oxford Nanopore MinION contains a consumable flow cell. This flow cell has thousands of small sequencing pools, each which has a waterproof cover over the top. This cover is pierced by protein tunnels, which we call nanopores.

An electric current is hooked up to the flow cell, which encourages DNA to move through the nanopores. Samples are prepared for sequencing by attaching a helicase to the end of the sequence. This helicase engages the pore, unwinds the double-stranded DNA and begins the sequencing process.

The shape of the DNA changes along its length, and this changes the electrical resistance. The change in resistance is recorded by the sequencer and sent to a computer over a USB cable.

Software is then used on a computer to convert the electrical traces back into a symbolic representation of DNA, most typically in the form of a sequence of letters stored in a computer file.

This sequencer is very portable: all it needs is some prepared DNA, and a little bit of electricity.

MinION Sequencing in 2014-2015

Here's a picture of a minion in our lab. I think this is a photo of the first time we attempted sequencing on the MinION.

My records on the Malaghan Wiki tell me that our first run was actually a dud flow cell, replaced at no charge by Oxford Nanopore. I ran it for six hours, hoping for a spark of life, but was only rewarded with a blank screen.

For our second run, we got about 3 Mb of 4T1 mitochondrial DNA sequence, which was the start of data that we used to confirm or reject genotyping that was done at Griffith University in Australia. That data was part of the mitochondrial DNA transfer paper that was published in January 2015.

It wasn't the first publication in the world that discussed nanopore sequencing, but it demonstrated that even at very small sequencing outputs, our MinION was useful and productive pretty much from the start.

Our first year or so of nanopore sequencing was spent mostly in pilot and feasibility studies. I was getting a bunch of free flow cells from Oxford Nanopore for my participation in their online community, and using them to test the limits of sequencing. At that time I could see that it wasn't ready for genomic or transcriptomic sequencing: both the sequence quality and yield needed to improve before I'd be able to do even a low-coverage genome-wide analysis.

Reality of Nanopore Sequencing (2015)

And that's the environment in which I gave my first nanopore sequencing talk here at the Malaghan Institute. I had to somehow drum up support for a device that had an amazing potential, but not all that much to show for it.

Even taking into account the things that were promised, the report card for Oxford Nanopore wasn't looking too good from the outside. The main things that it had going for it were that it was fast and portable, and I had trouble convincing people here that those qualities were useful.

MinION Sequencing in 2016-2017

Over the next couple of years I worked on a bit of outreach work with the MinION. I gave a TEDx talk on a tomato source in March 2016, set up a rapid sequencing table at a conference in October 2016, and helped out at a PoreCamp training session in Australia in February 2017.

During that time I was still chipping away at pilot studies for MinION sequencing at the Malaghan Institute, and we ended up putting out a few papers on our progress. I've also been trying to encourage other people at the institute to get involved in the sample preparation process, now that all the people who helped me out in the first couple of years have left.

MinION Sequencing in 2018 - mtDNA

So that brings me to this year. I talked in March about the analysis work I'd done on 4T1 breast cancer cell line samples that Carole and Olivier had prepared, discovering that there was no expression of mitochondrial genes when mitochondrial DNA is missing from a cell.

This was such a startling result that Rebecca repeated the experiment, and as of this week I can confirm we've now got seven ρ0 samples that had no detectable mitochondrial DNA transcripts.

MinION Sequencing in 2018 - data analysis

But I didn't just look at mitochondrial transcripts.

I'll step aside here to point out that there can be a lot of decisions to make that affect the outcome of results, even at the analysis stage. As one example, reads should have a polyA primer sequence at one end, and a strand-switch primer sequence at the other
end. However, some of the reads from the most recent sequencing run had a polyA primer tag at both ends of the sequence.

It turns out that a gene which is somewhat common in transcriptome sequencing, beta actin, has a polyT sequence about 2/3 of the way in. This was presumably binding to the reverse complement of the polyT primer and getting amplified in the PCR reaction, without the need of a strand-switching primer to replicate the other strand of cDNA.

If I throw out all similar sequences where I'm a little bit unsure of the direction of transcription for various reasons, I lose about 40% of them. Olivier reminded me that one of the things I still need to do is see if the transcript distribution for the "ideal" reads is
similar to that of the "possibly valid" reads; if it's not, I'll need to stick to the smaller, more reliable, subset of transcript sequences.

MinION Sequencing in 2018 - genome

In addition to the mitochondrial genome, I also looked at genomic differences for Carole's run. You may remember that the real surprise from that run was seeing gene expression differences between wild type and ρ0 cells throughout the entire genome.

After looking at Rebecca's run, the new ρ0 results were fairly consistent with the old ones: 14 of 20 genes with high expression in ρ0 had roughly similar expression in Rebecca's samples, and 19 of 20 genes with low expression in ρ0 were similarly low in Rebecca's samples.

Nanopore Sequencing (Mar 2018)

I don't want to stall too long on this slide because this is the same slide I showed in March this year. This device is now very versatile, and can usually be encouraged to work for anything that can be thrown at it.

That information was current as of March this year. But in the nanopore sequencing world, things can change over the course of a few months.

London Calling 2018

Towards the end of May, I attended the large yearly Oxford Nanopore conference in London. There are \emph{always} exciting things that are just on the horizon, and this year produced the usual smorgasbord of enticing things.

On the left here is a picture I took of a MinION Mk1C. This is a single box, about the size of a small pencil case, that can do sequencing, base calling, and data analysis off a small 12V power supply. Loaded into this MinION is a flow cell dongle (also called a "flongle"), and loaded into the flongle is a small, cheap sequencing flow cell that will be available for about $150.

If that doesn't seem cheap to you, bear in mind that currently up to 12 samples can be run on a MinION, so with existing sample preparation kits that works out to be about \$25 per sample. And that's for something that produces per sample about thirty times the amount of data we were getting out of our first MinION run in 2014.

Nanopore Sequencing (July 2018)

Going back to that other slide, a few things have changed since March that work towards bringing down the sequencing cost further. We were told at London Calling that community members are testing out the flongle flow cells now, and those cells should be commercially available in a few months.

In our swag bag at the conference, we were given a voucher for what Oxford Nanopore calls a "Series D" flow cell. This is the first major electronics upgrade that they've made on their flow cells, ever, and fixes a number of issues with the previous cells. John Tyson has
already taken his Series D for a spin, and reported the run results on Twitter, claiming over 30 gigabases from one flow cell in a 3-day run. Without any bulk discounts, that's equivalent to ten times coverage of the human genome for about 1.5 thousand New Zealand dollars.

In other news, Alex Payne and Nadine Holmes looked carefully at MinION data, and found there was a software bug that was incorrectly chopping long reads up into smaller chunks. The longest read they got was 2.3 megabases; that's about 3/4 of a millimetre, which is really long. It's even more impressive given that the theoretical maximum read length from that run was 3.2 megabases between run restarts every 2 hours.

Reality of Nanopore Sequencing (2018)

So we've now got a device in our hands that can give us cheap, long, fast sequencing for almost all applications. Sample preparation is still a sticking point, so I'm still grateful for all the help I can get in that regard.

With the right experimental design, MinION sequencing works really well. I've been able to take a sample of plasmid DNA, prepare it in about 15 minutes, load it onto a used flow cell, run it for about half an hour, then spend another couple of hours on sequence analysis. That gave me something I could compare to an existing reference to determine whether or not our plasmid was likely to have the same sequence, all done on a whim in the space of a Friday afternoon.

Acknowledgements

So I want to thank the people who have supported the technology development of the MinION at the Malaghan Institute. Mike, Melanie, Graham, Franca and Giulia have all given me the opportunity to work with their samples and experiment with nanopore sequencing.

And, a big thank you to the people who have braved the sample preparation protocols of Oxford Nanopore, and allowed me to return to my friendly computer for data analysis.

And of course you, for being such a wonderful attentive audience. READ LESS