In science there are experiments that try to construct simple, tractable models of complex phenomena, with the goal of developing insights that will be relevant to the more complicated systems. Alternatively, there are experiments that try to infer insights amid a flurry of coexisting factors. Each of these approaches has its merits and flaws. I was reminded of this distinction by a paper that, in my opinion, falls squarely in the middle of this spectrum. The article was the subject of our last microbial ecology journal club meeting of the term:
A. G. Cheng, et al., Design, construction, and in vivo augmentation of a complex gut microbiome. Cell.185, 3617-3636.e19 (2022) doi:10.1016/j.cell.2022.08.003
It’s an excellent paper. This in itself is a remarkable statement; most of the papers we read are, to put it mildly, not excellent. (The journal club discussions are consistently great, however.)
The authors built a model bacterial community for studying the human gut microbiome and they examined the properties of this community in mice. There are several hundred species in the typical human gut, with lots of variation that makes it challenging to figure out what aspects of the community may be relevant to various readouts of health and disease. Several research groups have constructed model gut communities of a few bacterial species, where the small number enables experimental control and, hopefully, clear conclusions. A “small number” is typically 5-20, for example this with 17 species in mice, or this from my lab with 5 species in zebrafish. Cheng et al. suggest that this is too simple to be useful, and that the native gut microbiome is too complicated and uncontrolled to be useful.
The authors therefore develop a new sort of model system: a community of 119 bacterial species. All are culturable, and so one can grow up batches of each species, mix them up, and use the resulting cocktail for controlled experiments, inoculating germ-free mice (as in this paper), or perhaps humans. The method for selecting 119 species is clever: the authors first select 104 species that are from taxa commonly found in the human gut microbiome. Mixing these together, most (about 95) persist in mice and reach stable populations. Reasoning that there may be environmental niches left unfilled by this set of 104, the authors then “challenge” the mouse microbiome with human fecal samples (yes, microbiome research can be rather gross), determining from the resulting gut composition the challengers that persist and adding these to the model system recipe. The result is 119 bacterial species. The new set is more reistant to further challenge than the original set.
The authors look more at the behavior of this microbiome model community in terms of things like amino-acid-mediated interactions, resistance to a pathogen, and immune responses.
The paper is full of well done experiments, with substantial specimen numbers and assessments of replicability (which is rare in the literature). Even the visual display of information is great, with figures like this one (below) to show data on the relative abundance of species in the initial mix, in mice from weeks 1-4 post-inoculation, and in mice following the “challenge.” (The thin lines are useless, but the color coding of dots makes it clear that relative abundance is quite stable.)
Though the study and its presentation are great, there are a few broader aspects of it that give one pause:
First: 119 species? Is this really what it takes to have a tractable but useful experimental model of the gut microbiome? If so, is there any hope of actually understanding what’s going on, or discerning general, human-comprehensible principles?
Second: Is 119 species actually tractable? That’s a lot of culturing, troubleshooting, and assessment. The paper has 24 authors, and the resources that went into this work are substantial. Here’s the funding acknowledgement:
This may simply be what’s needed for meaningful work in this area. I’ve often thought, in fact, that a lot of work in biology / medicine is done on too small a scale, leading to problems of reproducibility and a proliferation of noisy studies. Still, it’s a bit discouraging, especially from the perspective of someone who runs a small lab that will never have the resources depicted above. In general, the gap in capabilities between the richest labs and institutions and the rest has been increasing; it’s not obvious whether this is good or bad for science.
These issues, however, are far outside the scope of the paper itself, and I’ll reiterate that the paper is excellent and a pleasure to read.
Obligatory book note! I have a whole chapter on the gut microbiome in my pop-science biophysics book. Less obviously, I write near the end about the distinction between conceptually simple studies and inference from “big data,” and how, historically at least, the former approach has been more successful as well as more aesthetically satisfying. The usual links: My description, Publisher, Amazon.)
Today’s illustration
Circles. If I kept count correctly, there are 119.
— Raghuveer Parthasarathy; December 12, 2022
The Eighteenth Elephant 