Thread by @cibiocm

Can we assess engraftment of donor's strains in FMT? What drives it?

In our latest study @NatureMedicine we examine the extent and determinants of strain engraftment upon FMT, and build ML models to predict the resulting microbiome composition (thread).

nature.com/articles/s4159…

FMT (Fecal Microbiota Transplantation) is very effective against recurrent Clostridioides difficile infection and is being explored for many other diseases, with divergent outcomes.

Very little is known on microbial engraftment dynamics, preventing higher FMT success.

We analysed publicly-available metagenomic data together with newly-collected cohorts (@bambinogesu and Fondazione Policlinico Gemelli) for a total of 1,371 samples and 226 triads of donor, pre-FMT recipient, and post-FMT recipient samples.

Using our latest software developments for strain-level metagenomics StrainPhlAn 4 (biorxiv.org/content/10.110…), we could track strain sharing among individuals:

So, what are the main determinants of strain engraftment upon FMT?

A mixed administration (upper+lower GI) resulted in higher engraftment rates. Antibiotic pre-treatment and infectious diseases (both hypothesized to reduce microbiome colonization resistance).

And does the extent of strain engraftment matter for the FMT recipient?

Yes, as we found it associated with FMT success when meta-analysing all cohorts

However, not all bacteria were engrafted equal: Bacteroidetes and Actinobacteria species displayed higher average strain engraftment rates than Firmicutes and Proteobacteria.
Those that engrafted best included so-far uncultured bacteria.

This was also linked to phenotypic characteristics:
Disease-associated, widespread and gram negative bacteria engrafted better, which calls for thorough donor screening to promote transfer of health-promoting bacteria.

We finally tried to predict post-FMT microbiome composition with machine learning models (random forest with LODO and CV).
While results varied by cohort, substantial prediction ability was maintained across datasets.

Feature importance analysis revealed that species composition matters: the presence of a species after FMT is dictated primarily by its amount in the donor and in the recipient, as well as taxonomy and general prevalence

Finally, we found that our ML framework has the potential to pinpoint particularly suitable donor individuals for improving microbiome features (e.g richness) in recipients based on their microbiomes.

With more and larger datasets becoming available, we hope our work will allow to further stratify FMT strain dynamics of specific conditions, and eventually guide choices in clinical practice to exploit the full potential of FMT.
Read the full paper at nature.com/articles/s4159…

Thanks to all co-authors, especially @gianluca1aniro @PuncocharM @NicolaiKarcher @MireiaVallesc @GiovanniCammar9 @nsegata.
We also thank all study participants for their commitment, and @ERC_Research #H2020 oncobiome and @MASTER_IA_H2020 @EMBO @CIBIO_UniTrento for their support.