The importance of context in analyzing antibiotic resistance

When analyzing genomic data, I usually try to push further and get the most of each approaches. Here is the story behind some of the methods we used in our recent work investigating the resistance genes found in gut microbiomes, particularly in the lower abundance bacteria.
The importance of context in analyzing antibiotic resistance

Our paper recently published in Microbiome is the third of what I call my trilogy on the impact of antibiotics upon the microbiome. In the first article (The ISME Journal, 2016), we determined how the intake of a commonly used antibiotic modulated the fecal microbiome. One of the strengths of this study was that we used deep shotgun metagenomics (up to 15 Gb per sample), which allowed us to not only do taxonomical profiling of the microbiomes, but also to mine these metagenomes for antibiotic resistance genes. We detected more than 43 000 putative antibiotic resistance genes!

Yes, 43 000 is a really big number. That’s a lot of potentially antibiotic resistant bacteria in only 24 healthy individuals (at three time points). At first, I was worried. Today, I see this number from a wholly different perspective.

A few months after the publication of this first article (and of the second also), I had the pleasure to attend the Advanced course on antibiotics organized by the Fondation Mérieux and the Institut Pasteur at Les Pensières, in Annecy, France (see picture above). In a nutshell, this was a two-week course about antibiotics and antibiotic resistance, given by more than a dozen experts in different areas related to the topic. It was more than a course, but also a human experience. At some point in the training, I was invited to briefly present about antibiotic resistance and the microbiome. I showed unpublished figures related to our 2016 ISME paper where specific antibiotic resistance genes were listed as prevalent in our microbiomes. At this point, an eminent microbiologist raised his hand and commented: “Many of these are present in all strains of E. coli. They are not clinically relevant. They are everywhere!”

The culprit figure from my presentation
The culprit figure from my presentation

This got me thinking and, gradually, I became aware that these putative antibiotic resistance genes needed much more than cataloguing. 43 000 was just a scary number, it didn’t mean anything. These genes needed context.

Genomic context.

Epidemiological context.

Clinical context.

In the following year, my colleagues and I investigated approaches to conduct comparative genomics from a big data perspective (for example by using our Ray Surveyor software). This led me to wonder how we could use this expertise to better interpret the gene content of microbiomes based on genomes available in public databases. This is how we proceeded:

  • Performed taxonomical identification of contigs containing antibiotic resistance genes.
  • Annotated antibiotic resistance genes in all the available genomes from a bacterial species (for example E. coli).
  • Compared the antibiotic resistance genes found in contigs potentially originating from E. coli to the genes that were actually found in E. coli genomes.

I won’t repeat here what we observed from these analyses (if you are interested, the paper is here), but it did help us in determining which antibiotic resistance genes were potentially relevant, and which had the most potential for horizontal gene transfer. How did we do that? By considering the genomic environment of these genes along with their distribution in different isolates.

The take-home message here is a single word: context.

There are so many sequences available to help contextualize our data, that we would be crazy not to use them. Truly, the only limitation here is our imagination. And maybe our computing power…

For more details, see our manuscript here: