BashTheBug

This is the third year that the NIHR Oxford Biomedical Research Centre (BRC) held its Showcase at the popular Westgate Shopping Centre in the centre of Oxford. Like last year it was held jointly with the NIHR Oxford Health BRC.

Our research software engineers had redesigned and rewritten our dance mat game so now you took on the role of a doctor firing different coloured antibiotics at incoming bacteria. So far so easy: over time, however, the bacteria get resistant to one or more of the antibiotics (denoted by a coloured shape appearing on them) meaning a lot of foot stamping is required to keep changing to the right colour antibiotic!

The game starts nice and slowly but the difficulty increases rapidly. On the day a few people scored over a hundred but anything more than that was too challenging…

It was half-term which was great as it meant lots of parents, grandparents and children stopped by and talked to us about our research. By the end of day we’d spoken to 130 people; half of them had heard about the Showcase beforehand, three members of our PPIEP committee and some ex-members of MMM dropped in to say hi, the Deputy Head of Mabel Prichard School (whose school we visited earlier in the year) and finally we spoke to one person who was halfway through a 10km run!

What is the problem?

Studying the genetic code of bacteria (DNA) can give us vital information like the genes they carry and how they are related. This can help us detect outbreaks and track how antimicrobial resistance genes spread, so we can contain them better. Putting the whole genome together gives us a clearer story than just looking at the short DNA fragments that come out of a sequencer, just like binding a book tells a clearer story than looking at the scattered pages. We can be more certain of where certain genes are, which can indicate whether it was borrowed from another bacteria, for example. 

Traditionally, the most accurate way to put together, or assemble, a bacterial genome into the correct sequence of the A, T, C and G nucleotide ‘rungs’ of the DNA ladder, has been to use a combination of two different technologies in a ‘hybrid’ method: highly accurate but hard-to-assemble short-read Illumina sequences, and error-prone but easy to put together long-reads. Long read sequences make it easier to get the overall structure right- a bit like doing a 1000-piece instead of a 500,000-piece jigsaw puzzle of a modern art painting. The bigger ‘jigsaw’ pieces are more likely to include a unique part that helps orientate where you fit in the overall picture. The highly-accurate illumina short-reads then swoop in to spell-check everything. The problem with using both methods in this way is that it is expensive to run two different experiments for a single sample, and this acts as a barrier to using bacterial whole genome sequencing on a larger scale in public health surveillance. 

Luckily, continued improvements in long-read sequencing utilising powerful computers and the latest machine learning techniques have driven down the error-rates of this technology.

What did we do?

To test this out, we assembled the genomes of 96 bacteria, taken from human bloodstream infections across England, using both hybrid and long-read only methods. We found that whilst both methods allowed us to create very high-quality genomes, the long-read only method was actually better than some of the hybrid methods at putting the genomes together.

So what?

This is great news for public health professionals! It means that creating complete and accurate bacterial genomes is now much more cost-effective than it was before, and takes us one step closer to integrating this into our routine surveillance of disease-causing bacteria. It could help us detect outbreaks earlier and pinpoint where they came from more accurately, especially when different bacteria start swapping smaller bits of DNA with each other and may not be linked to an outbreak with our current methods. Ultimately, looking at whole bacterial genomes better enables us to track their spread at a higher resolution compared to the more fragmented genomes from short-read sequencing, and will allow us to more effectively prevent people from getting infected.

This has been a long time coming but last month, the Ellison Institute of Technology launched EIT Pathogena. This is a website where anyone anywhere in the world can work out what species of Mycobacteria are in a sample and, if it is Mycobacterium tuberculosis, which, as the name suggests, is the causative agent of tuberculosis, also work out which antibiotics are likely to be effective. Lastly it also tells you if the genome of that sample is sufficiently similar to any other samples you’ve uploaded that they could be part of the same outbreak.

So how does it do this? Well you have to have put your Mycobacterial sample through a genetic sequencing machine — this gives you two output files (called FASTQ files) which contain lots of short stretches of DNA found in the sample which will have come from the patient, other bacteria, the odd virus and probably some Mycobacteria. Historically sifting through these files and working out what is what and then seeing if you can build a genome from some of the short stretches (a bit like a really big jigsaw, just one where the pieces overlap and some have mistakes) is the job of a Bioinformatician and is difficult.

EIT Pathogena makes that simple; all you have to do is drag and drop the FASTQ files onto the web portal and it will upload them, then automatically remove and forget any bits of human DNA (as these could be used to identify the patient in theory) before working out what species are present etc.

We have written all the computer code that handles all the short stretches of DNA. Much of the software used to predict which antibiotic is likely to work was originally written as part of our earlier CRyPTIC project but has been rewritten by our Research Software Engineers (RSE) to bring it inline with modern software engineering practices.

If you like looking at code, head over to GitHub and check out gnomonicus which in turn uses gumpy and piezo. All of these are written in Python3 — Jeremy Westhead who is one of our RSEs noticed that we could speed up this part of the pipeline significantly by rewriting gumpy in Rust. He called this new version grumpy of course! All of this software has a license allowing anyone to use it for research but prohibits using it for commercial purposes.

When we test a sample taken from a patient to see which antibiotics will work (and which will not) we test many thousands of bacteria all at once; if the antibiotic kills most of them we say it is susceptible. But in some cases that isn’t good enough: the few that are left (because they are resistant) can grow and multiply so all you’ve done is buy a little time.

What if instead you could look at the effects of an antibiotic on a single bacterium?

That, in essence, is what the interdisciplinary team drawn from both the Department of Physics and the John Radcliffe hospital in Oxford did with this project. Using fluorescent staining and super-resolution microscopy they can image individual bacteria and ones which are resistant to an antibiotic “look different”, providing you’ve stained the right parts of the bug.

Humans, of course, are really good at looking at photographs and so they also set up a Citizen Science project on the Zooniverse called, you guessed it, Infection Inspection. They asked volunteers to classify of E. coli which had been fluorescently stained and then treated with an antibiotic as either resistant or susceptible.

If you want to read more about this please go and read their paper.