Thread by @TSBSchm

Very pleased (& proud) that our meta-study of microbiome dynamics following #FMT is out today in @NatureMedicine

Congrats to @simone_s_li @BorkLab & collaborators @embl @amsterdamumc @WUR & CDD.

What happens after FMT? It's messy...

1/

nature.com/articles/s4159…

We started from the realisation that diff smaller studies on #FMT had come to v diff conclusions on microbiome dynamics. So we collected public and newly generated data from 316 FMTs across different indications and conducted a metagenomic meta-study.

2/

We built #pangenome|s from 47k newly generated MAGs and 25k ref genomes for a total of >1k species and profiled strain populations in each sample based on microbial Single Nucleotide Variants and gene presence/absence.

3/

We reasoned that there are different possible strain-level outcomes for each species in an FMT (see below): persistence of recipient strains, colonisation by donor, coexistence or 'influx' of novel (prev. unobserved) strains...

4/

...and that these outcomes can be quantified based on characteristic patterns in strain composition (based on determinant SNVs) in each triad of donor (blue) recipient pre-FMT (yellow) & post-FMT (purple) sample.

5/

OK, enough technical stuff. Results.

1st thing we noticed was that while all species seemed to experience all outcomes, some were prone to donor-recipient coexistence whereas others tended towards exclusionary patterns.

6/

Across FMTs, outcomes were a broad gradient. Donor colonisation was common in rCDI patients, but confounded with antibiotics usage.

Correcting f confounders & multiple testing, we did not observe an association of donor colonisation with clinical success.

7/

We next trained LASSO models to predict the fractions of persisting vs colonising strain populations. By re-training from variable subsets, we were able to pinpoint the most important factors.

8/

We took this a step further and trained LASSO models for strain-level outcomes *for every species*. Interestingly, rec. persistence and don. colonisation were only moderately predictable, but rec. 'turnover' could be predicted v accurately.

9/

The best predictors for donor strain colonisation in a species were donor:abd ratio of that sp, don:rec strain dissimilarity & rec strain diversity. Moreover, several 'gatekeeper' species in the rec seemed to have exclusion effects.

10/

In other words, if a rec carries a diverse strain pop (several strains of the same species), donor colonisation is less likely.
If the strain pop btw donor and rec are v dissimilar, it is more likely.

11/

Finally, we put all these variables into an ecological framework to link them to possible underlying processes. While preliminary, this can give an idea where to tune to enhance donor colonisation (if that is a desired outcome).

12/

If you're interested in the topic, here's a bit further reading.

@AonghusLavelle & @h_sokol discussed our work in a News & Views piece in the current @NatureMedicine :

13/

nature.com/articles/s4159…

They also discuss another FMT metastudy, led by @NicolaiKarcher @cibiocm , published back to back:

14/

nature.com/articles/s4159…

A few weeks ago, @DanielPodlesny @WFlorianFricke et al also published an #FMT metastudy:

15/

sciencedirect.com/science/articl…

All three studies started from a similar idea, but used different methods and focussed on different questions. Some findings are congruent, others complementary.

Or put bluntly: same same, but different.

Yay for reproducibility in science!

16/

Finally, please see this older thread on the @biorxivpreprint for more details (and different figures):

https://twitter.com/TSBSchm/status/1443848216131350558?s=20&t=Q1gLCYey7YuB_IA_qVTLpg

PS: @embl comms have also covered our study and probably do a better job at explaining it than me ;-)

https://twitter.com/embl/status/1570439819636908033?s=20&t=dbuL0T294npZIoA_fkL3pw