Synopsis: In a recent paper from my lab [1], we report on watching gut bacteria get pummeled by low doses of antibiotics. The antibiotics induce changes in the spatial organization of the microbes, with major consequences for their ability to persist in the gut.
Motivations
At high enough doses, antibiotic drugs will kill bacteria, or at least stall their growth. That’s typically why we take them, of course. Antibiotic treatments induce large changes to the composition of the human gut microbiome. What about exposure to weak doses of antibiotics? Especially from studies in mice, it seems like low concentrations of antibiotics can also strongly perturb intestinal bacterial communities. The reasons for this remain mysterious, and it’s an especially important question given that low levels of antibiotics are commonly encountered as environmental contaminants. (A recent global study of rivers found that two-thirds are contaminated with antibiotics, often at alarming levels; link, National Geographic blurb.)
We set out, therefore, to illuminate the interactions between low doses of antibiotics and gut bacteria. We used larval zebrafish and three-dimensional microscopy to perform controlled experiments and directly observe bacterial responses — an approach that has been central to my lab’s work for the past several years (see e.g. here and here), but that remains surprisingly unique.
I’ll briefly describe our findings, which tie physical properties to antibiotic responses, and also comment on some aspects of the study that are only briefly noted in the paper. The project was joint between my lab and that of Karen Guillemin, our frequent and excellent collaborator, was led by the fantastic team of Brandon Schlomann, a Physics graduate student in my lab and Travis Wiles, a postdoc in Karen’s lab, and also involved Elena Wall, a technician in Karen’s lab.
Our approach
We focused our attention on two gut bacterial species that exhibit different extremes of physical behaviors. One is highly motile and individualistic:
(The movie above is of one optical slice, in real time. The bacteria are so dense that individuals are hard to see, especially on the left. For a different movie of the same species, with sparse labeling to make individuals more evident, see this movie.)
Bacteria of the other species form dense aggregates:
(The movie above is a 3D scan.)
We also focused on one antibiotic, ciprofloxacin, which is widely used and well studied. Cipro is also the antibiotic that was most commonly found to exceed safe levels in the river study noted above. Our experiments involved zebrafish larvae that started out germ-free, devoid of any gut microbes, after which they were colonized with one of these bacterial species, exposed to antibiotics (or not), and imaged with our home-built light sheet fluorescence microscope.
Of course, this system is very minimal: one antibiotic, one bacterial species, and larval fish that lead rather simple lives. However, the complexity of the “real” microbiome makes it extremely hard to gain insights into what actually shapes it. By building simple systems from the bottom-up, we aim to clearly identify mechanisms that govern gut microbes and their dynamics. With mechanisms and principles in hand, one can build up to ever greater complexity. This is the general approach that physics has had for centuries, and it’s been very successful.
Bacteria sticking together
What did we find?
Let’s first consider the very motile species of bacteria. Low levels of ciprofloxacin reduced the bacterial abundance in the water outside the fish by a factor of 10 compared to not having the antibiotic present, but in the fish gut, the bacterial population dropped by a factor of 100. Imaging inside the gut, we found that the normally peppy bacteria were now filamented and sluggish, and were pushed by the gut walls into aggregates that were transported out of the animal by intestinal flows.
This wasn’t too shocking; from in vitro experiments (i.e. experiments in a dish, not in an animal), it’s well known that antibiotic-induced stresses cause many bacteria to lose motility and elongate, growing without dividing. In the gut these filamentous cells entangle; like aggregates in many other experiments we’ve seen, they’re pushed along until expelled from the gut.
The more interesting observation came from the other species, the one that is naturally highly aggregated. We guessed that the antibiotic would have little effect; after all, the cells are already clumped together. Instead, we found a much stronger effect, a thousand-fold drop in the intestinal population following weak antibiotic treatment. Again looking at the bacteria themselves in the gut, we found that the nature of the aggregation had changed: small clusters were far less abundant, and the clusters that were present were large. I’ll say more about this in the next section, but for now I’ll just note that once again, the antibiotics have a large effect in the gut, tied to physical changes in the bacterial organization.
Polymers, gels, and models
(This section is somewhat technical, and can be skipped without consequence.)
This observation of changes in bacterial clusters led us to a general question about gut microbes: how can a species of aggregating gut bacteria persist in the gut, if aggregation is coupled to transport and expulsion? The answer must be that expulsion is balanced by the nucleation and growth of new clusters. Can we turn this picture into a rigorous model? Brandon realized that the answer is yes, and constructed and analyzed an excellent framework (on his own!). (Brandon is very good at this sort of thing.)
Watching gut bacteria, we realized that four processes are central to the dynamics of clusters: aggregation (in which two clusters merge into one), fragmentation (in which one cluster splits into two), growth (in which cells in a cluster divide) and expulsion (in which the cluster leaves the gut).
We made assumptions for the form of each of these processes: that aggregation is independent of cluster size, that fragmentation always takes the form of single cells breaking off of clusters, that growth follows a logistic form, and that the expulsion rate is size-independent. (The fragmentation assumption is the least obvious, bu it’s what we seem to consistently observe. The expulsion assumption is wrong, but relaxing it has little consequence, as explained in the paper.) With these simplifications, one can write kinetic equations describing the system — very similar, in fact, to models of polymer growth and colloidal dynamics considered especially in the 1980s. Stochasticity is important here, unlike typical polymer systems, where analytic, large-N solutions are useful. Here, stochastic simulations are necessary for modeling the system.
Unfortunately, the set of equations has five parameters — disturbingly large. As von Neumann famously stated,
With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.
Fortunately, however, we can estimate all five parameters from the data! (See the paper for details — the relevant data are growth rates, means and variances of populations, and other features that don’t involve spatial structure.)
It’s worth pausing for a moment and asking whether, at this point, we’ve learned anything from the model. The answer is certainly “no.” One can always fit any model to any dataset. The process is worthwhile only if there is some insight we gain from the parameter values, or if we can use this model to predict something non-trivial. This may seem obvious, but I’ve realized over the years that it is not. In biology especially it is fashionable to tack “mathematical models” onto papers, like the nutrient-free garnish on a fancy dinner. (These models are often bloated with ridiculous numbers of parameters, so perhaps a better picture is a garnish that occupies the whole plate.)
From our model, it turns out, there is something we can predict: the distribution of bacterial cluster sizes. No information about cluster sizes went into determining any of the model parameters, so it’s fair to look at the stochastic simulations’ outputs of this measure. From our 3D microscopy and computational image analysis, we can measure the actual cluster size distributions, and compare. Here are the results:
We were shocked at how well theory (curve) and data (points) agree — a strong indication that the model actually does capture the relevant dynamics governing aggregative bacteria in the gut.
The model predicts that there should be a boundary between persisting and extinct populations, as a function of the growth, fragmentation, and other parameters — somewhat similar to gelation transitions in polymer systems. With weak ciprofloxacin, our aggregating species is pushed across the phase boundary to extinction!
Implications
We uncovered a mechanism by which weak, sub-lethal antibiotics can dramatically perturb gut bacterial populations: altering bacterial aggregation dynamics. This had never been observed before, and was therefore pleasantly surprising. More broadly, why might we care? Where might this lead? Here are a handful of thoughts:
Spreading antibiotic resistance? The perturbed bacteria are alive. Aggregates are expelled from the gut but they’re not dead, and when the antibiotic is removed the bacteria will eventually return to their typically vibrant state. (We directly verified this.) Moreover, it’s those bacteria that “noticed” the antibiotic that are expelled. We wonder, therefore, if this provides a mechanism for enhancing the transmission of antibiotic resistance through an environment. I’m keen on testing this in the future.
Universality. It’s not in the paper, but we’ve looked a little bit at other antibiotics, and it seems like the phenomena we found is applies quite generally. This needs to be more carefully investigated, though — both similarities and differences across a variety of antimicrobial drugs will be important to map.
Mechanical models. One of the most important lessons from our experiments is that our findings could not have been predicted from test tube experiments. There are a lot of in vitro examinations of antibiotic responses, and they’ve led to great insights into molecular mechanisms, the evolution of resistance, and other topics. However, if one wants to predict responses in the intestinal environment, the physical properties of the bacteria, and of that space, are crucial.
Interestingly, our first version of the paper included more of a discussion of in vitro experiments we had done; the reviewers raised issues with these, and moreover pointed out that these studies were irrelevant — usually it’s us who make that argument! The reviewers of our paper, by the way, were excellent; critical but thoughtful and enthusiastic.
Can artificial, microfabricated environments with pumps and flows that mimic intestinal mechanics enable more realistic in vitro experiments? I think so, but my lab certainly doesn’t have the bandwidth to pursue this. Maybe you do, dear reader!
Connections to soft matter physics. Physics has been remarkably successful at describing materials of all sorts, from semiconductors to snowflakes. The gut microbiome, for all its complexity, is a material, physical object, and should be amenable to physical analysis. The success of our kinetic model of bacterial cluster suggests that merging soft-matter models with microbiota data may be a fruitful path.
Predictions for humans?! A legitimate criticism of our efforts is that our findings may just illustrate aspects of bacteria-zebrafish interactions, and not have any relevance more broadly, in particular to the human gut microbiome. I don’t think this is the case, mainly because the factors underlying what we’ve found are quite general — bacteria become filamented and less motile in response to antibiotics, and vertebrate intestines transport material. In addition to this, there may be a more specific mapping possible between our experiment and future human studies. We’ve shown that the distribution of bacterial aggregate sizes is set by physical processes in the gut. Could one invert this: measure the distribution of sizes of gut bacterial aggregates in animals not amenable to controlled experiments or imaging, like humans, and from that infer the underlying dynamics? (One would measure the aggregates from histology, or fecal samples; no, I’m not volunteering.) This would be challenging, but I wonder if we’ve stumbled upon a metric that goes beyond simple abundance measures to gain deeper insights into gut microbiome processes.
Final thoughts
Our first tentative experiments with antibiotics and gut microbes were in 2016, over three years ago! The experiments and their analysis have involved a lot of hard work, mainly from Brandon and Travis, who steered the project through many twists and turns. There are a lot of exciting experiments still to do. We’ll see what the future brings!
Addendum (Jan. 7, 2020): There’s a nice writeup on this paper and our earlier imaging-based work in Nature Reviews Microbiology (“Light sheets unveil host-microorganism interactions,” by Sean C. Booth & William P. J. Smith) — link here.
Today’s illustration
Not a great illustration, but one of the fastest I’ve made!
— Raghuveer Parthasarathy, October 18, 2019
References
[1] B. H. Schlomann^, T. J. Wiles^, E. S. Wall, K. Guillemin, R. Parthasarathy, “Sublethal antibiotics collapse gut bacterial populations by enhancing aggregation and expulsion.” Proc. Natl. Acad. Sci. (2019). [Link.] The bioRxiv preprint is very similar to the final published version.
[^ co-first authors]
