Speeding up the diagnosis of antibiotic resistance

In the case of sepsis, each hour of delay in administering an effective antibiotic increases the probability of death by approximately 7%. But many antibiotics have been in use for so long that certain bacteria have evolved resistance to them. Worse yet: when a doctor encounters a patient with a potentially drug-resistant bacterial infection, they often don’t know it. They must make a decision based on very little knowledge, often using broad-spectrum antibiotics. This carries the risk of administering the wrong antibiotics to the wrong patient. For all these reasons, testing the pathogen's drug resistance as swiftly as possible is paramount. The current clinical procedure relies on bacterial culture, grown specifically for such testing. But this process takes over two days – far too long. Hence the urgent need for a much faster approach.

And that’s where a novel method called Genomic Neighbor Typing comes into play. Introduced by Karel Břinda  —and colleagues— during his postdoc at the Harvard T.H. Chan School of Public Health and Harvard Medical School, this groundbreaking technique was published^[1] in Nature Microbiology in 2020.

Genomic Neighbor Typing

"We leverage this technology in a very original way. Instead of trying to just identify the genes or mutations that ar responsible for antibiotic resistance, as most other methods do, and which still takes too long to inform patient treatment, we infer what are the closest known relatives of the pathogens found in the patient’s sample.” In other words: its nearest neighbor in a database of genomes sequenced previously.

“Pathogens are nowadays routinely studied and sequenced in laboratories, so we know which strains have been resistant or susceptible to which antibiotics. Now, in a patient’s sample, if you are faced with a bacterial strain for which you can identify the closest known relative, and if you already know the resistance profile of this closest relative from the past analyses, then you can make a pretty good guess on the properties of the strain in the sample.”

Diagnosis within 24h

By contrast to methods focusing on the identification of genes or mutations, Genomic Neighbor Typing comes with two key advantages. First, it does not require prior knowledge of the pathogen's complex biology. Second, it can make predictions even from very little data from the sequencer, and thus it’s ultrafast. “In the experiments in our paper, we showed that, once the sequencing had started, we were able to predict resistance and susceptibility within approximately 10 minutes.”

Prior to this sequencing run though, one must also factor in the time required for collecting the sample from the patient and its preparation, which takes about several hours overall, but this is gradually improving with better sequencing protocols. Sequencers are often compared to paper shredders that cut a whole library of books into a huge heap of small paper pieces. Piecing together the words, sentences, paragraphs, pages and books calls for a lengthy computational process. In that regard and besides portability, nanopore machines offer two additional advantages that Genomic Neighbor Typing benefits from. “With other sequencing technologies, you have to wait for many hours and then you get all the data at once when the process is completed. By contrast, with nanopore, you get a stream of data as soon as the device starts sequencing. And with our method, you can make a prediction nearly immediately from this early stream of data, so the diagnosis can come very rapidly. The other advantage is that nanopore sequencers provide very long reads. It’s like getting much bigger pieces of paper from the shredder. This is crucial, as this is the element that allows us to really zoom into the evolutionary tree and pinpoint the closest relatives. As soon as we get several sufficiently long reads of the pathogen, we are able to provide a well-informed guess on its resistance.”

A broader range of pathogens 

Now a permanent researcher within Genscale^[2], a bioinformatics team at Inria centre at Rennes University, in Rennes, Brittany, France, Břinda leads an Exploratory Action^[3] named BARDE. “Our goal is to develop more appropriate computational methods for the diagnosis of antibiotic resistance, combining both Genomic Neighbor Typing with other approaches, based, for instance, on the identification of resistance-associated genes and SNPs ^[4] as diagnostics based on nearest neighbors is not possible for all antibioticl and species.”

A broader range of pathogens must also be considered. “When we originally published our method, we gave examples for two bacterial species: Streptococcus pneumoniae and Neisseria gonorrhoeae. But for different species, we must adjust the method and develop representative databases of genomes and their resistance-related metadata. And to make everything work, we need experts on a given species with profound knowledge of its clinical aspects and its resistance mechanisms.”

Collaboration with Rennes University Hospital

In this context, BARDE has established a close partnership with Rennes University Hospital (CHU). “CHU Rennes is the home of one of France’s leading research groups in antibiotic resistance, led by Professor Vincent Cattoir.. Their specialty is enterococcal species, especially Enterococcus faecium, one of the major human pathogens. They also maintain the French national reference collection (CNR) for E. faecium, which gives us access to a high-quality national-level resource of genomes for diagnostic purposes.”

From there, the goal is to “connect Genscale’s computational expertise with Rennes CHU’s biological and clinical competencies in order to develop methods that can be applied directly in the context of hospitals, thus reducing the time that is currently necessary for resistance diagnosis in clinical settings.”

In the long run, the key question revolves around building large and truly representative databases of bacterial strains. “Right now, we work with databases of about thousands genomes. As sequencing becomes increasingly affordable and widespread, we will get more and more genomes, providing us with much larger databases. But we will need new computational methods, new software tools, to search through them if we are to compare, for instance, one particular strain from a patient's sample in real time against all the bacteria that have ever been sequenced and deposited in public archives. And in the end, everything will have to come together. So there is also a huge challenge lying ahead...”

^{[1] Read : Rapid inference of antibiotic resistance and susceptibility by genomic neighbor typing, par Břinda, K., Callendrello, A., Ma, K.C. et al. dans Nature Microbioly 5, 455–464 (2020).}

^{[2] Genscale is a project-team of Inria, Rennes University and CNRS, common to UMR Irisa. Its main goal is to develop scalable methods, tools, and software for processing genomic data. Within Genscale, a dedicated group of people work on the Barde project.}

^{[3] An Inria Exploratory Action is an internal mechanism to facilitate the emergence of new research themes by giving scientists the means to test original ideas.}

^{[4] A Single-Nucleotide Polymorphism (SNP) is a substitution of a single nucleotide at a specific position in the bacterial genome. SNPs can help explain some of the differences in susceptibility to antibiotics across a population.}