I’ve been thinking a lot about the long-term evolution experiment (LTEE) with E. coli lately – even more than usual. One impetus has been the paper by Mike Wiser, Noah Ribeck, and me that appeared today (14-Nov-2013) in Science (online publication in advance of print). Another reason is that I’m working on the competitive renewal for the NSF grant that funds this experiment.
The Experiment that Keeps on Giving
Both of these have got me thinking about the long-term fate of this long-term experiment. Should the experiment continue? For how long should it continue? Who will take it over when (or before) I retire? And after that person retires, then what? How will they sustain it? Will they rely on the usual competitive grants? Would an endowment be more suitable? How does one raise an endowment?
I like to say that the LTEE is the experiment that just keeps on giving. Between the element of time, the inventiveness of the bacteria (even in their simple, confined, little flask worlds), and the many talented students, postdocs, and collaborators who have worked on the LTEE, there seems to be no end to the insights this experiment can provide into fundamental questions about evolution. Why shouldn’t this experiment keep on giving, even after I’m gone?
When I started the LTEE in February of 1988, I had no idea that it would continue for more than 50,000 generations. I had previously done some other experiments that went for a few hundred generations, and I intended this one to run for at least 2,000 generations. That would deserve the “long-term” moniker. Although we made some fitness measurements along the way, most of the hard work comes after a milestone is reached. That’s when one begins the intensive assays to quantify the changes that occurred. And while we performed those assays, we continued the daily transfers. So by the time the first paper was prepared, submitted, reviewed, revised, and published in December of 1991, the LTEE was past 5,000 generations. And so it has gone: new milestones, new questions, new assays, new data, new analyses, and new papers.
And the new questions keep coming based on new hypotheses of students, postdocs, and collaborators (occasionally even me), new technologies such as genome sequencing, and new observations of what the evolving bacteria have done.
This latest paper is an interesting one because it uses our most old-fashioned assays – the kind that was the heart of the LTEE when it started, and which also formed the core of that first paper back in 1991. That is, the results are based on measurements of relative fitness, coupled with new models – both descriptive and dynamical. (Although this blog post emphasizes the descriptive model, the Science paper also presents new theory showing that the descriptive model can be derived from a dynamical model of evolution that incorporates two phenomena – clonal interference and diminishing-returns epistasis – that are known to occur in the LTEE and other studies of evolving asexual populations.)
Fitness is the central phenotype in evolutionary theory; it integrates and encapsulates the effects of all mutations and their resulting phenotypic changes on reproductive success. Fitness depends, of course, on the environment, and here we measure fitness in the same medium and other conditions as used in the LTEE. We estimate the mean fitness of a sample from a particular population at a particular generation by competing the sample against the ancestral strain, and we distinguish them based on a neutral genetic marker. Prior to the competition, both competitors have been stored in a deep freezer, then revived, and acclimated separately for several generations before they are mixed to start the assay proper. Fitness is calculated as the ratio of their realized growth rates as the ancestor and its descendants compete head-to-head under the conditions that prevailed for 500 … or 5000 … or 50,000 generations.
The exciting new result is that the fitness of these evolving bacteria shows no evidence of an upper bound or asymptote. A two-parameter power law fits the data much better than does a two-parameter hyperbolic model. According to both models, the rate of fitness increase decelerates over time, as it clearly does. However, the power-law model has no asymptote, whereas the hyperbolic model has an upper bound.
Even more striking and important, to my mind, is that the models differ in their predictive power. We fit these two models to truncated datasets that included only the first 20,000 generations of data and asked how well they predicted the fitness values observed over the next 30,000 generations of data. The unbounded power law beautifully predicts the fitness trajectory that it had not seen, whereas the asymptotic hyperbolic model underestimated later measurements. The underestimation of the asymptotic model becomes progressively worse as the temporal data are more and more truncated; that is, the evolving bacteria consistently pass right through the “limit” predicted from previous data. By contrast, even with only 5000 generations of data, the power-law model very nicely predicts the fitness trajectory all the way out to 50,000 generations.
How long can this continue? In our paper, we present the following thought-experiment. I’ve overseen 50,000 generations of the LTEE in my scientific life; now imagine another 49,999 generations of scientists, each one overseeing 50,000 more bacterial generations. That’s 50,000^2 generations, or 2.5 billion generations in total. (It will take about a million years to get there.) You’re probably thinking that the unbounded power-law model must predict some crazy high fitness that would imply a ridiculously fast growth rate.
In fact, the power law predicts that fitness relative to the ancestor will increase from ~1.7 after 50,000 generations to ~4.7 after 2,500,000,000 generations. If the bacteria eliminate the lag time associated with the transition from starvation to growth each day (which they have already largely done), then a fitness value of 4.7 implies that the bacteria will have to reduce their doubling time from the ancestor’s ~55 minutes to ~23 minutes. That’s very fast given that the LTEE uses a minimal medium where cells must synthesize everything from glucose, ammonium, and a few other molecules. But it’s not so fast that it suggests the bacteria would violate some biophysical constraint. Indeed, some bacteria can grow twice that fast, albeit in a nutrient-rich medium.
What Does the Future Hold?
I’d really like science to test this prediction! How often does evolutionary biology make quantitative predictions that extend a million years into the future? Maybe the LTEE won’t last that long, but I see no reason that, with some proper support, it can’t reach 250,000 generations. That would be less than a century from now. If the experiment gets that far, I’d like to propose that it be renamed the VLTEE – the very long-term evolution experiment.
And this prediction about the future fitness trajectory is not the only – or even the main – reason to keep the LTEE going. Some important things in evolution simply require a lot of time. In my presidential address to the Society for the Study of Evolution this past summer, I highlighted three findings where it proved to be important that the LTEE had continued for many years (and, if I’d had more time, I could have added more to that list). First, it takes a very long time series to distinguish between asymptotic and non-asymptotic fitness trajectories. Second, it took over 30,000 generations before the most dramatic phenotypic change occurred in one of the populations, which evolved the ability to use citrate – which has been present in the medium of the LTEE throughout its duration – as a second source of energy. Third, postdoc Zachary Blount is currently studying whether the refinement of that new function is leading to changes that would qualify the citrate users as a new, or incipient, species.
What other new traits might the bacteria evolve? Could they evolve some means of genetic exchange? Might the within-population competitive interactions ever take a turn toward predation? Who knows? Only time will tell – and only if we allow time, the bacteria, and future generations of scientists to do the work of evolution and science.
Some Additional Links
- Science Story by Elizabeth Pennisi on Richard Lenski and the LTEE
- Science Podcast Interview: Sarah Crespi with Richard Lenski
- MSU Press Release and Video Interview with Michael Wiser and Noah Ribeck
- NPR’s Nell Greenfieldboyce’s Report with Audio on the Morning Edition
- Carl Zimmer on The Loom Discusses this Story and his Previous Stories about the LTEE
- NewScientist‘s Bob Holmes Reports with Comments from Joachim Krug and John Thompson
- German Public Radio (Deutschlandfunk) and Lucian Haas Report in German
- George Dvorsky Blogs about our Findings at io9
- Wiser, M. J., N. Ribeck, and R. E. Lenski. 2013. Long-term dynamics of adaptation in asexual populations. Science 342: 1364-1367.
27 responses to “Fifty-Thousand Squared”
You kind of touch on this in the paper (but for the sake of discussion) there has to be an inherent upper bound on adaptation due to the probability of mutations to escape genetic drift. That bound is far off, but exists…so the model already needs some tweaking to match reality?
Hi Dave: Drift loss of new mutations is already in the model. To “count” a beneficial mutation has to occur, escape extinction when rare by drift, and escape extinction by clonal interference. There all in there!
Hi Rich! True, but at some point millions of years from now there stops being beneficial mutations unless population size expands…no? The model assumes that beneficial mutations are always possible (no asymptote)
although I guess it just becomes a situation where it takes insanely long for mutations with really small s to escape drift, but it is always theoretically possible
Yes, it assumes beneficial mutations are always possible, but their effect sizes get smaller and smaller … and as you added, they take longer and longer (even insanely long) to escape drift and fix. But there’s no theoretical upper bound.
Will the article be available as a “usual” Science paper soon? My university does not have access to ScienceXpress, and I am blocked behind the paywall… which is frustrating because the paper sounds really interesting!
Yes, but I’m not sure when it will appear in the usual form. However, if you email me (address easily found) I can send a preprint.
A few colleagues already sent me a pdf.
Pingback: Friday links: the history of “Big Data” in ecology, inside an NSF panel, funny Fake Science, and more | Dynamic Ecology
Pingback: Fifty-Thousand Squared | Media Cultures: Microb...
Pingback: A Long Way Left Up Darwin’s Mountain – Phenomena: The Loom
Pingback: Long term evolution of E.coli: Fifty-Thousand S...
Your article in the brazilian press – http://oglobo.globo.com/ciencia/nao-existe-perfeicao-nem-limite-para-melhorias-diz-pesquisa-10792370
Best, Claudio Motta
Does anyone out there think a small research team composed of systems/synthetic/molecular biologists armed with the latest tools of the day (genome-scale models, automated robots, and MAGE to give a few examples) could design a genome that would beat the fitness of Lenski’s most evolved lineages?
Rules: Anything can be engineered into or out of the ancestral strain (though I’m not sure if we should allow whole genomes Dr. Venter). You get 1 year to do it.
How might such a team go about this? If you think it won’t be possible, then why not?
Hello, Josh. Sounds like a fun, interesting challenge! I would suggest one additional rule. That is, one cannot add any functions that lead to “interference competition” via bacteriocins or secreted toxins that might inhibit or kill other cells. The long-term lines have improved in terms of “scramble competition” for the limiting resources, and toxins would be too easy to add and would change the “game” entirely.
It also seems to me you could split this into different competitions based on whether a team: (i) begins work from the ancestral strain only, and without using exogenous DNA; (ii) begins work from the ancestral strain only, and employs some exogenous DNA from other strains or species; and (iii) uses some of our evolved bacteria and/or information about them (from our papers) as a starting point for further improvements, with or without the addition of exogenous DNA.
I should also add, as per our Science paper, that my university requires the competition of a Material Transfer Agreement before we can share our strains.
Pingback: I’ve got your missing links right here (16 November 2013) – Phenomena: Not Exactly Rocket Science
Pingback: Morsels For The Mind – 15/11/2013 › Six Incredible Things Before Breakfast
Pingback: Fifty-Thousand Squared | AnnBot | Scoop.it
Pingback: Fifty-Thousand Squared | Complexity - Complex S...
Pingback: Links 11/20/13 | Mike the Mad Biologist
Incredible and very frightening to think of what the limits would be to increase in fitness. Just a question. Have you noticed any significant shift in codon usage between the ancestral and the recent cultures ?
I agree that potential increases to fitness are extremely interesting. But they should not be “frightening” because fitness, as measured in our study, is in the laboratory environment of the long-term experiment. The experiment began with a non-pathogenic strain of E. coli and, moreover, the populations are now highly “domesticated” bacteria adapted to this laboratory environment.
In regard to codon usage, let me point you to two papers. First, Wielgoss et al (2011, G3) show that there have been few synonymous changes in the lines that retained the low ancestral mutation rate. That’s because 50,000 generations is a long time for an experiment, but it’s still very short in terms of genomic evolution — except at those genomic sites that have been subjected to strong positive selection to change. Second, Wielgoss et al (2013, PNAS) show much more pronounced changes in codon usage in one of the populations that evolved hypermutability — and then also evolved a partial compensatory reduction in mutation rate — associated with changes in DNA metabolism and repair processes. There one sees changes in codon usage that are consistent with the known effects of the specific changes in DNA metabolism and repair processes on mutational spectra.
So as James A. Shapiro might suggest, are not some of these mutations self induced, or, as the use of “fitness” would indicate, are all of the stochastic variety, where the potential benefits are initially provided by some form of cosmic accident?
That’s a short question, but it will take many more words to answer it well. I’ll probably do so as a separate post in the future.
Pingback: Tre rekommendationer | There is grandeur in this view of life
Pingback: When We’re Sixty Four (Thousand) | Telliamed Revisited
Pingback: Another Birthday Haiku | Telliamed Revisited