An extremely long post, mainly written to have something to point people to as a commentary on some recent work.
A new paper from my lab came out recently in PLOS Biology, on watching and learning about the competition between gut microbes. I like the paper a lot, and, with one possible exception, it took more hard labor than any other paper I’ve worked on. It’s reassuring, therefore, that it’s gotten a good amount of attention — for example, a writeup in phys.org and over 3000 article views within a week of its debut. (Scroll down for a graph.) I thought I’d use this space to comment on some things that aren’t obvious from the paper itself. First, a summary:
We now know that our intestines are home to a vast multispecies microbial ecosystem, the composition of which influences health and disease in a variety of ways. As in any ecosystem, competition between species should be a major determinant of composition. What drives competition between gut microbes? There are lots of potential mechanisms, but what’s often assumed is that biochemical interactions, like vying for the same nutrients or secreting some toxin, are the relevant drivers. We decided to look, literally, at two species of gut bacteria using larval zebrafish as a tractable model vertebrate animal and our beloved light sheet microscopy methods to see the microbes, in three dimensions, over many hours. See here for a blurb on our earlier work using this approach to look at a single bacterial species.
The two species we examined were natives of the zebrafish gut that, on their own, colonize to high abundances. When fish are exposed to one of the species (“Vibrio“) after being well-colonized by the other (“Aeromonas“), we find that the Aeromonas population becomes small and highly variable. This we can learn from straightforward (though not trivial) dissection and plating methods. But what is actually happening in the gut that leads to this outcome?
Watching the two species, we find that their spatial structures are vastly different. Vibrio exists as beautiful clouds of fast-swimming individuals:
Live imaging of a single optical plane in the intestinal midgut of a 6 dpf larval zebrafish inoculated at 4 dpf with GFP-labeled Vibrio. Scale bar: 50 μm.
Aeromonas, in contrast, is mostly found in large, immobile aggregates. These aggregates are easily pushed around by the peristaltic motion of the intestine — the motion that in all of us mixes up gut contents and sends them downstream.
When challenged by Vibrio, Aeromonas happily persists, except when suddenly dislodged and expelled from the gut. These expulsions are rare, and seem random in time. With Vibrio present, the expulsions occur with twice the frequency than when it’s absent, and moreover Aeromonas typically can’t recover to its original population size.
Time-series of maximum intensity projections of images from a larval zebrafish, initially colonized at 4 dpf with Aeromonas (magenta), challenged 24 hr later by inoculation with Vibrio (cyan), and then imaged every 20 min for 14 hr. Times indicate hours post-challenge. The yellow line roughly indicates the lumenal boundary of the intestine; the two bacterial fluorescence channels are overlaid inside this region. Scale bar: 200 μm.
This leads to a picture in which the two species’ different responses to their host’s peristalsis is the key determinant of their competition. If so, we should be able to do two things: (1) construct a quantitative model that relates the observed frequencies and magnitudes of Aeromonas population collapses to the statistics of their final population levels, and (2) turn off the peristalsis of the gut and see the bacterial competition go away. We were able to do both! “Turning off” peristalsis actually involved using fish with a particular gene mutation that leads to a lack of neurons lining the gut, and weaker peristalsis; interestingly, the same mutation is found in many people with Hirschprung’s Disease, a gastrointestinal condition. (See the paper for details.)
Why construct a mathematical model of some observed phenomena? There are many reasons to do so (and many reasons not to), but here, as is often the case, the goal was to see whether some relatively simple but well-defined process could recapitulate key features of the data, and moreover to see if the parameters that emerge could be independently tested, giving us further confidence that the model corresponds to reality.
I wrote the growth-and-collapse model and simulation myself. It’s pretty simple, and would make a good homework assignment in a biophysics class. In it, a single species (Aeromonas) obeys logistic growth, meaning that it has some intrinsic growth rate, and some carrying capacity at which the actual growth rate goes to zero. Punctuating these are collapses in which the population decreases by some fraction. These collapses occur randomly, but with some constant probability per unit time (i.e. it’s a Poisson process). The model has four parameters, which isn’t bad, but what’s even better is that we know the intrinsic growth rate from measurements, and that the collapse fraction and temporal probability can both be merged into one effective parameter. That means, therefore, that there are only two free parameters in the model. (Again, see the paper for details, especially the Supplementary Materials.) Two is a wonderful number, almost as good as zero or one — we have some hope that two parameters will actually be well constrained by the data. (As Enrico Fermi famously said, “With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.” Some time ago, I read a microbiome modeling paper with 287 parameters, which I’m sure made both Fermi and his elephant turn in their graves.)
I fit this model to data on Aeromonas abundance from dissection experiments — i.e. from the guts of several fish dissected at a single point in time. This gave some value for the collapse parameter. Separately, we could measure the collapse parameter from our movies — i.e. watching individual fish over an extended period of time. The two values for the collapse parameter are in remarkable agreement. I was really stunned by this — that a single snapshot of lots of guts matches time-series of individual guts. It’s like ergodicity in statistical physics!
What’s the point? (and a model of readership)
The overall message of our paper is that we’ve shown, by direct observation, that the response of bacteria to the physical environment of the intestine can be a major determinant of apparent inter-bacterial competition. From a macroscopic ecological point of view, this is obvious: no one is surprised that the winners and losers at a tide pool are determined in part by who can resist the crashing of the waves. This perspective is, however, generally lacking in studies of the gut microbiome. It shouldn’t be! More constructively, we suggest a route by which enteric nervous system disorders (e.g. in Hirschprung’s disease) can influence the composition of the microbiome by changing this physical environment. Furthermore, we suggest that physical traits of gut bacteria, such as aggregation and motility, are important to try to understand, as these can determine the composition of the microbiota.
As mentioned, the paper seems to have gotten a good amount of interest, including far more article views than I’m used to. Here’s the number, plotted versus days since the article appeared:
The solid curve is a fit to a exponential-plus-linear model noted here. I am, as before, surprised that this model fits so well. (The best-fit parameters, by the way, give an initial decay time for the decline of interest of 4 days, and a later linear rate of readership of 70 views/day.)
Is it Biology? (On rejection.)
Before publishing in PLOS Biology — which, I’ll point out, was excellent in its handling of the review process and in its promotion of the paper after publication — we were rejected by two other journals. One took several weeks to reject us with no explanation at all. The other criticized our model for being “simple” — apparently it’s better to have 289 parameters — and criticized our lack of a “molecular mechanism” explaining Vibrio’s ability to enhance the rate of Aeromonas collapses. On the one hand, it would be interesting to figure this out — more on this later. However, in general I strongly disagree with the emphasis on “molecular mechanisms” in much of modern biology.
A rather technical analogy: A stellar accomplishment of biophysics is the runs-and-tumbles model of E. coli motion developed by Berg, Purcell, and others, which not only connects to predictions for bacterial dispersal, led to insights on fundamental limits on sensing and detection, and more, but which was immensely useful many years before a “molecular” picture of bacterial motility was pieced together. Moreover, its utility transcends the underlying molecular details — no one would argue that to understand how fluctuations of nutrient concentrations constrain bacterial sensing, we have to understand chemotaxis gene networks. The same philosophy is at work throughout biophysics and, we would argue, “classical” molecular biology. Similarly, in our work, we’re aware that we lack “molecular” pictures, but we aim for general insights into the importance of stochasticity (i.e. the population collapses) and host-derived physical perturbations that should serve as a much needed framework for the field to progress. (The above is part of a reply to the journal editors, which got us nowhere.) A shorter analogy: to understand an incandescent lightbulb, it’s silly to stress the “molecular” properties of a tungsten filament rather than the more general physical phenomenon of thermal radiation.
As yet another caution about “molecular mechanisms:” I’m sure that we will, in fact, find a gene that influences Vibrio’s ability to out-compete Aeromonas. (In fact, I think we already have one.) However, I’m sure it’s not unique — as everyone knows, but routinely ignores, gene activities are connected by intricate networks, so it’s in general unenlightening to pretend that showing an effect from knocking out some gene implies that that gene is special.
While initially frustrating, the end result of our paper submission process is very satisfying. In general, though, it highlights interesting philosophical struggles in many areas of science about what “explanation” or “understanding” really mean.
Is it Physics?
I also sometimes deal with the question of whether this sort of work is Physics — after all, it doesn’t seem to match at all my colleagues’ experiments on elementary particles or nonlinear optics. My two answers to this are an emphatic “yes,” or a similarly emphatic “I don’t care.”
About “yes:” One of our goals as physicists should be an understanding of how structure forms and develops — how snowflakes grow, or crystals melt, for example. Here, we’re concerned with the structure and dynamics of bacterial communities. It’s true that these happen to be alive, and so can fit under the heading of “biology,” but what matters is not the type of system we look at, but the type of questions we ask. This is, I would argue, what has made physics so successful as a field.
I’ll note, by the way, that from a physical perspective, we got lucky. It turned out that the physical structure of the bacteria we were examining was crucial to the behavior of the system. We didn’t know this would be the case when we started, however, and in fact what we expected was some simple smooth decline of one species together with the smooth rise of the other, indicative of some mechanism of nutrient competition, for example. This would be worthwhile to characterize, but certainly not as “physics-y” as what we actually found. The moral: you won’t find unexpected physics without leaping into experiments on new systems. This seems like an obvious statement. Often, however, when talking to new physics graduate students, I’m struck by their desire to work on conventional things that are guaranteed to map onto things they’ve learned in classes and for which the outcomes are more or less known. I find this shocking, and a bit sad.
About “I don’t care:” More importantly, the boundaries between fields are human-made constructions, and we should happily ignore them when there are mysteries to be solved. The relevant question is not “Is it Physics?,” or “Is it Biology?,” but rather “Is it Science?” Again, pretty obvious, but perhaps worth pointing out.
There’s a lot to do that builds on this experiment. Obviously, we’re working with a very simplified model system, with just two bacterial species competing in the gut. We have hopes and plans to expand this number, which can take us towards neat questions of overall ecosystem stability, predictions based on pair-wise interactions, and more. We’re also working with bacteria whose physical properties can be altered, seeing the impact of this on overall population dynamics. Also, as mentioned, we have strong indications of particular cellular mechanisms that can strongly influence inter-species competition. We’re also making better tools for looking at gut microbes and extracting information from these images, developing analytic solutions to the population collapse model described above (!), and more. There’s a lot to do!
This work was led by two amazing people: Travis Wiles, a postdoc in the lab of our close collaborator, Karen Guillemin, and Matt Jemielita, a Physics graduate student in my lab who has since moved on to a postdoctoral position at Princeton, split between biologists and physicists, looking at bacterial biofilms. Many more people were involved, some of whom are hard at work right now illuminating more mysteries of the gut microbiota!
… a Thorny Seahorse, a watercolor based on a photo I found in
“Seahorses: A Life-Size Guide to Every Species,” by Sara Lourie, which was on the new book shelf at the UO Science Library.
— Raghuveer Parthasarathy, September 8, 2016