Bioinformatics explores intestinal microbiota

Did you know that your intestines are home to over 1,000 billion micro-organisms, the vast majority of which are bacteria? Together, they make up the intestinal microbiota, an ecosystem whose major role in health has been recognized for many years: it contributes to digestion and the overall functioning of the digestive system, and also helps our immune system.

Researchers now know that the health of the intestinal microbiota and that of the individual who hosts it are closely linked. Understanding these bacterial communities and how they function has therefore become a central subject of study in biology: how do microbiota differ from one individual to another? What do they have in common? What does a healthy microbiota look like? These are just some of the questions scientists are trying to answer.

But what does this have to do with Inria? It can be summed up in one word: Pleiade. This is the name of the project-team at the Inria Centre at the University of Bordeaux (shared with Inrae) in which researcher Clémence Frioux works: "I started my research on gut microbiota in 2019, during my postdoc at the Quadram Institute in Great Britain," she explains. And when I took up a position at Inria six months later, I kept going! I loved the idea of applying my research to human health."

For three years, Pleiade, the Quadram Institute and the Earlham Institute worked together on a specific question: how can statistical learning contribute to obtaining a simple, interpretable model of microbiota composition? Since 2011, there has been the concept of enterotypes: a classification into three categories according to the type of bacteria predominating in the microbiota," explains Clémence Frioux. We wanted to propose a more detailed model that would be easy to reapply."

To develop this model, the researchers chose not only to focus on the dominant bacterial genera, but also to describe the combinations of bacterial communities and their abundance. Thanks to a previous study that gathered the metagenomes (i.e. all genes, all species combined) of the microbiota of over 5,000 individuals of all ages, across thirteen countries with Western lifestyles, they were able to build a learning base: data that includes the list of species in each individual's microbiota and their abundance.

Then machine learning came into play, using an algorithm designed for dimension reduction: "Instead of describing the microbiota on the basis of the 500 bacterial genera present, the algorithm will combine the genera into factors, and optimize their composition to best describe the initial data", explains Clémence Frioux. The result is remarkable: five "guilds" or bacterial communities are described. And the composition of the original microbiota is simplified to a combination of these "guilds".

In reference to enterotypes, Clémence Frioux and her colleagues call them "enterosignatures". "What's particularly interesting," stresses the researcher, "is that our five-enterosignature model proved to be generalizable." To prove this, the scientists applied it to another Western cohort of 888 bacterial metagenomes and demonstrated the ability of the five enterosignatures to describe the corresponding microbiota. The same success was achieved by testing it on a cohort of 1152 bacterial metagenomes from 12 non-Western countries. Challenge met!

But the researchers didn't stop there... " We had a lot of associated metadata. We had a lot of metadata associated with our samples, on the use of antibiotics or other drugs, the age of the individual, the mode of birth and diet for newborns, etc.," explains Clémence Frioux. This enabled us to associate our enterosignatures with certain characteristics.

The enterosignature dominated by bacteria from the Bacteroides group thus seems to play a predominant role after antibiotic treatment, for example, while that dominated by bacteria from the Firmicutes group is associated with characteristics of good health. The model also makes it possible to describe changes in the microbiota over time, highlighting variations in the proportions of microbiota enterosignatures as a function of individual age.

Continuing their research, the scientists were also able to observe, through the prism of their five-enterosignature model, the complementarities and redundancies of the metabolic functions (the chemical reactions taking place within cells) carried by micro-organisms. In a nutshell? Their model reduces the microbiota to just a few factors, enabling it to be interpreted and applied in the field of health... with precision. Which is exactly what they set out to do.

We also observed that a minority of samples were poorly described by the model, as they were associated with disturbances in microbiota composition," notes Clémence Frioux. The five enterosignatures thus enable us to identify atypical microbiota, which may have medical applications."

The scientists now want to take the model even further: to continue refining their results, they have set up a website dedicated to enterosignatures. "The aim is for colleagues to be able to apply the existing signatures to their own samples in order to confirm our model, but also to support it with new signatures, corresponding, for example, to certain pathologies."

As for Clémence Frioux, she intends to pursue her research, why not by applying the same methodology to other microbiota? The world of bacteria is vast, and understanding it is indispensable in many fields, from health to the environment.