It looks like they screened 45 strains of bacteria to find 9 that passed their safety tests, and then only one of those had a 100-percent response. The sample size is also small: 5/5 sounds a lot less impressive than 100%. I'd expect the true response rate to be substantially lower ("winner's curse").

The bar for an acceptable side effect profile in an FDA-approved drug would also be a lot higher than "five genetically near-identical mice did not show evidence of pathology in a single study."

I'm not saying this work is bad (skimming, it seems fine for what it is, though haven't read in detail), but it's quite preliminary if we're talking about developing a medical treatment that could eventually be deployed in humans. There's a reason it ended up in a mid-tier microbiome journal.