Changing Mutation Rates

We have seen large changes in the spontaneous mutation rate in some of the LTEE populations.  These changes reflect an interesting tradeoff between short-term fitness and long-term evolvability.

Proximate causes.  Six of the 12 LTEE populations evolved to be so-called “hypermutators” by 50,000 generations.  The proximate (i.e., biochemical) causes of these changes are mutations in genes whose products are involved in DNA repair or the degradation of molecules that cause damage to DNA.

These mutations typically cause the rate of point mutations throughout the genome to increase by roughly 100-fold (Sniegowski et al., 1997, Wielgoss et al., 2013), so their effects are not at all subtle.  They also change the spectrum of mutations:  mutations in the mutS gene, which encodes a protein involved in mismatch repair, cause increased A·T–>G·C and G·C–>A·T transitions (Lenski et al., 2003); while mutations in mutT, which encodes an enzyme that degrades an oxidized nucleotide, cause A·T–>C·G transversions (Barrick et al., 2009).

Evolutionary effects.  The evolutionary effects of these hypermutators are subtle and interesting.  In essence, one can think of mutations that produce hypermutators as affecting the tradeoff between short-term fitness and long-term evolvability.

Short-term cost.  Of all the possible mutations that might occur, many more are deleterious than are beneficial. Therefore, hypermutators produce more maladapted progeny than otherwise identical cells with a lower mutation rate.  Hence, hypermutators suffer a fitness cost caused by the increased production of progeny with deleterious mutations.

However, the E. coli strain that was the ancestor to the LTEE has a low point mutation rate, which we’ve estimated as ~10^-10 per base-pair per generation (Wielgoss et al., 2011).  Given the genome contains ~5 x 10⁶ base-pairs, this rate translates to only ~0.0005 point mutations per genome per generation.  Therefore, even a 100-fold increase means that most hypermutator progeny are mutation-free.  Considering that only a fraction of genomic sites are subject to mutations that would be deleterious in the LTEE environment, we infer that the short-term cost to a 100-fold hypermutator is ~1% (Wielgoss et al., 2013).

Evolvability benefit.  Even a 1% cost is not trivial, so how can a hypermutator survive and spread through a population?  In fact, most hypermutators do not survive; the vast majority of mutations that cause hypermutators will die out as a consequence of that short-term cost.  However, hypermutators result from loss-of-function mutations, and a dozen or so large genes are targets for these mutations.  Hence, new hypermutators will continually be regenerated in large populations.  Absent other forces, an equilibrium frequency of hypermutators would be reached that reflects the balance between the rate of appearance of hypermutators by new mutations in the relevant genes and the rate at which they are removed by selection against the deleterious mutations they cause—in other words, the familiar mutation-selection balance of population-genetics theory.

But another force is at play: the populations in the LTEE are not sitting on a fitness peak, so there are on-going opportunities for beneficial mutations to appear.  And a hypermutator cell has a much higher probability of generating a beneficial mutation than does a “normal” cell.  In essence, there’s a race to produce the next winner.  If a hypermutable cell generates the next beneficial mutation that sweeps through the population, then the hypermutator will “hitchhike” along with it because, without sex, the two mutations are linked.

Combining forces.  So how do the short-term cost and the evolvability benefit play out together?  Mutations that knock out any one of the genes involved in DNA repair probably occur at a rate between 10^-5 and 10^-6 per generation, and the resulting hypermutable cells have a fitness disadvantage of ~1% owing to the production of deleterious mutations.  At mutation-selection balance, the frequency of hypermutators is between 0.01% (10^-4) and 0.1% (10^-3).  Let’s use 0.05% to illustrate.

Although the hypermutators are a small minority, on a per capita basis each of them has a 100-times higher probability than a normal cell of generating the next winner.  So 5% of the time, a hypermutator will be swept to fixation, but most of the time the winner will be produced by a normal cell.  Now consider the fact that each of the LTEE populations has had many beneficial mutations go to fixation over its history.  After 14 selective sweeps, the odds are better than 50:50 that at least one of those beneficial mutations was generated by a hypermutator.

King of the mountain.  After a hypermutator becomes common, it becomes very hard to dislodge it from the population.  This difficulty follows from the same logic as above.  Once the hypermutator reaches 1% of the population, it has a 50% chance of generating the next winner; by the time it gets to a 50% frequency, the odds are 100:1 in its favor.  Thus, a hypermutator only needs to get lucky once, and then it becomes extremely difficult to displace it … at least so long as the population is far from the fitness peak.

Nothing lasts forever.  Even before a population reaches a fitness peak, its rate of fitness improvement typically decelerates, at least in a constant environment like that of the LTEE (Wiser et al., 2013).  At some point, the magnitude of the benefit that would result from reducing the mutation rate and its associated fitness cost may become commensurate with the fitness advantages that are available from other mutations.  When that happens, selection to reduce the mutation rate becomes effective, and the hypermutable “king of the mountain” can be displaced by a genotype with a lower mutation rate.

Indeed, we have observed this displacement occurring in one of the LTEE populations (Wielgoss et al., 2013).  In that population, not one but two lineages independently arose (see Figure below) that reduced the mutation rate by about half, while reducing the fitness cost from ~1% to ~0.5%.  The population thus remains hypermutable, but less so than before.

What the future may hold.  In that paper, we hinted that it is probably easier to reduce the mutation rate in stages rather than to revert to the ancestral rate in a single step.  That’s because the population is continuing to adapt, albeit at a slower rate.  A genotype with a 50% reduction in the mutation rate will save half of the fitness cost of the full-blown hypermutator, yet it will continue to produce 50 times as many other beneficial mutations as would a genotype that reverted to the ancestral mutation rate.  In essence, the fitness costs and the evolvability benefits are on very different scales.

• Sniegowski, P. D., P. J. Gerrish, and R. E. Lenski. 1997. Evolution of high mutation rates in experimental populations of Escherichia coli. Nature 387: 703-705.
• Wielgoss, S., J. E. Barrick, O. Tenaillon, M. J. Wiser, W. J. Dittmar, S. Cruveiller, B. Chane-Woon-Ming, C. Médigue, R. E. Lenski, and D. Schneider. 2013. Mutation rate dynamics in a bacterial population reflect tension between adaptation and genetic load. Proc. Natl. Acad. Sci. USA 110: 222-227.
• Lenski, R. E., C. L. Winkworth, and M. A. Riley. 2003. Rates of DNA sequence evolution in experimental populations of Escherichia coli during 20,000 generations. J. Mol. Evol. 56: 498-508.
• Barrick, J. E., D. S. Yu, S. H. Yoon, H. Jeong, T. K. Oh, D. Schneider, R. E. Lenski, and J. F. Kim. 2009. Genome evolution and adaptation in a long-term experiment with Escherichia coli.  Nature 461: 1243-1247.
• Wielgoss, S., J. E. Barrick, O. Tenaillon, S. Cruvellier, B. Chane-Woon-Ming, C. Médigue, R. E. Lenski, and D. Schneider. 2011. Mutation rate inferred from synonymous substitutions in a long-term evolution experiment with Escherichia coli.  G3: Genes, Genomes, Genetics 1: 183-186.
• Wiser, M. J., N. Ribeck, and R. E. Lenski. 2013. Long-term dynamics of adaptation in asexual populations.  Science 342: 1364-1367.

The figure below shows the decelerating fitness trajectory (dark green curve, left axis) and the number of mutations (right axis) as the lineage with the ancestral mutation rate (blue) is replaced by a hypermutator lineage (red), which in turn is displaced by two independent lineages with somewhat lower mutation rates (light green and purple).  The figure comes from Wielgoss et al., 2013, Proc. Natl. Acad. Sci. USA; it is shown here under the doctrine of fair use.

Evolution of Novelty

These examples are particularly interesting—and surprising—because I chose the environment of the LTEE to be as simple as possible, thereby limiting the opportunity for novel functions and complex ecologies to emerge.  However, the evolving bacteria have proven me wrong by discovering new ways of making a living in the simple flask worlds where they live.

Novel function

• Blount, Z. D., C. Z. Borland, and R. E. Lenski. 2008. Historical contingency and the evolution of a key innovation in an experimental population of Escherichia coli. Proc. Natl. Acad. Sci. USA 105: 899-7906.
• Blount, Z. D., J. E. Barrick, C. J. Davidson, and R. E. Lenski. 2012. Genomic analysis of a key innovation in an experimental Escherichia coli population. Nature 489: 513-518.

Complex ecology

• Rozen, D. E., and R. E. Lenski. 2000. Long-term experimental evolution in Escherichia coli. VIII. Dynamics of a balanced polymorphism. American Naturalist 155: 24-35.
• Le Gac, M., J. Plucain, T. Hindré, R. E. Lenski, and D. Schneider. 2012. Ecological and evolutionary dynamics of coexisting lineages during a long-term experiment with Escherichia coli. Proc. Natl. Acad. Sci. USA 109: 9487-9492.

This photo shows the increased turbidity (cell density) of the population that evolved the ability to use citrate in the middle, along with two others from the LTEE.  Brian Baer and Neerja Hajela took this picture in my lab in 2008.

Repeatability of Evolution

These examples all bear on the reproducibility of evolution, which is one of the core questions that the LTEE seeks to address.  The answer is not a simple one with a dichotomous “yes/no” outcome, because evolution is an intriguing mix of random (mutation and drift) and directional (natural selection) processes.  But the LTEE offers the opportunity to examine this question more thoroughly than almost any other biological system studied to date.

Examples of Parallel Evolution in the LTEE

Fitness. The trajectories for fitness, as measured in the environment of the LTEE, have been very similar across the replicate populations, although they are certainly not identical (Lenski and Travisano, 1994; Travisano et al., 1995).  But perhaps that’s not too surprising because fitness integrates, rather than atomizes, the underlying changes.

Cell size. One of the most strikingly parallel trends has been in the size of the individual cells.  All 12 populations produce cells that are much larger than the ancestor (Lenski and Travisano, 1994).  If you had asked me, I would have thought the cells should become smaller based on surface-to-volume ratio considerations in a resource-limited environment.  But the bacteria obviously had a different “opinion” about this, so to say.

Genetics. And it’s not just phenotypic traits that show parallel evolution.  We’ve found three genes that have fixed mutations in all 12 populations (V. Cooper et al., 2001; Woods et al., 2005), although the exact mutations at the sequence level differ in almost every case.  By contrast, most of the 4,000+ genes retain the ancestral sequence in most or all of the lines because, while the LTEE is a long experiment, it’s still just a “drop in the bucket” of evolutionary time.

Gene expression profiles. Perhaps my favorite example of parallel evolution is at the level of changes in gene expression across the entire “transcriptome” (T. Cooper et al., 2003).  We examined only two of the LTEE lines (because of costs) and we used the old approach of microarrays (as opposed to new RNAseq methods).  The changes in the global expression profiles were strikingly parallel, so that after 20,000 generations (when this analysis was done) these two independently evolved lines were more alike than either was to its ancestor (see figure below).  The identity of the genes whose expression changed in parallel suggested a shared underlying cause—a change in a “global” regulon, a high-level pathway that coordinately regulates the expression of many genes. From there, we tracked down a mutation in a gene called spoT, a key gene in that regulon.

The figure below shows the comparisons in global gene-expression profiles between: (top left) the ancestor and itself, as a control; (top right) one evolved line and the ancestor; (bottom left) another evolved line and the ancestor; and (top right) the two independently evolved lines relative to one another.  The figure is modified from T. Cooper et al., 2003, Proc. Natl. Acad. Sci. USA; it is shown here under the doctrine of fair use.

And back to genetics. When the evolved version (allele) of spoT was moved to the ancestral genome, it conferred a significant competitive advantage, demonstrating that it was indeed a beneficial mutation.  Moreover, the ancestor with the evolved spoT allele recapitulated many of the changes in gene expression that we saw in the evolved lines and that led to its discovery, which provided satisfying closure to our inferences.  And when we sequenced spoT in all 12 of the LTEE lines, we found that 8 of them had substitutions in that gene.  Nonetheless, a mystery remained: one of the two populations with the expression profile that evolved in parallel, and which led to the discovery of the many parallel mutations in spoT, did not itself have a mutation in spoT.  A mutation in some other gene (not one of the other candidate genes we had sequenced) must “mimic” the effects of the evolved spoT mutation in the other line whose gene-expression profile we had studied.  The LTEE is not only a valuable resource for studying evolution, it also generates many mutations worthy of study from molecular, genetic, biochemical, physiological, and other perspectives.

Examples of Divergent Outcomes in the LTEE

Citrate utilization. The most striking case of divergence we’ve seen is that one of the populations evolved the ability to consume the citrate that has been present throughout the LTEE (Blount et al., 2008; Blount et al., 2012).  It took more than 30,000 generations for this innovation to arise in that population, and none of the other populations have figured it out even after almost 60,000 generations.

Growth on maltose and resistance to phage Lambda. There are many other, more subtle examples of phenotypic divergence. One that I find very interesting concerns the differences in adaptation to glucose and maltose (Travisano et al., 1995).  Maltose is simply a dimer of glucose. Glucose is the limiting resource in the LTEE (leaving aside the one line that evolved the ability to use citrate).  One might expect, therefore, that the bulk of fitness gains measured in the LTEE environment would carry over if maltose were substituted for glucose in the medium.  In fact, however, that is not the case.  After 2,000 generations, the variation among the replicate lines in their performance on maltose was at least an order of magnitude greater than their variation in glucose.  Now some of the lines cannot grow on maltose at all.  And the same mutations responsible for that complete loss of growth on maltose caused those lines to become resistant to infection by a virus, phage Lambda, even though the LTEE lines were never exposed to the virus (Meyer et al., 2010).

Inferences on Parallel and Divergent Evolution in Nature and in the Laboratory

Challenges in interpreting nature. Parallel and divergent outcomes are, of course, also seen in nature, but it is often difficult to interpret these cases.  If two or more lineages underwent parallel phenotypic changes, was it because they shared genetic variation that was present before the lineages split or by later gene flow?  If so, the parallel changes may not be truly independent evolutionary outcomes.  And even if shared variation can be excluded (e.g., the parallel phenotypic changes have different genetic bases), what’s the relevant denominator?  That is, how often did parallel evolution occur relative to how often it could have occurred? Also, if two or more lineages diverged phenotypically, does that reflect the random effects of mutation and drift?  Or might it reflect instead subtle differences in the environment (ones that may be imperceptible to us, but important to the organisms) or the ancestral genotypes (i.e., divergence that occurred prior to the lineages encountering similar environments)?

Easier inferences in the LTEE. By contrast, the 12 populations in the LTEE all started from the same ancestral strain of E. coli.  Although they share the same ancestor, the populations do not share genetic variation; in fact, there was no variation at the outset because each population was started from a single haploid cell.  In other words, all of the variation that underlies changes we observe in the LTEE arose by new mutations that occurred during the experiment itself.  And of course, the 12 populations have evolved under essentially identical conditions (or about as close as humanly possible), with a simple, defined, reproducible environment.

Further Reading

Parallel evolution

• Lenski, R. E., and M. Travisano. 1994. Dynamics of adaptation and diversification: a 10,000-generation experiment with bacterial populations. Proc. Natl. Acad. Sci. USA 91: 6808-6814.
• Cooper, V. S., D. Schneider, M. Blot, and R. E. Lenski. 2001. Mechanisms causing rapid and parallel losses of ribose catabolism in evolving populations of Escherichia coli B. J. Bacteriol. 183: 2834-2841.
• Cooper, T. F., D. E. Rozen, and R. E. Lenski. 2003. Parallel changes in gene expression after 20,000 generations of evolution in Escherichia coli. Proc. Natl. Acad. Sci. USA 100: 1072-1077.
• Woods, R., D. Schneider, C. L. Winkworth, M. A. Riley, and R. E. Lenski. 2006. Tests of parallel molecular evolution in a long-term experiment with Escherichia coli. Proc. Natl. Acad. Sci. USA 103: 9107-9112.

Divergent evolution

• Travisano, M., F. Vasi, and R. E. Lenski. 1995. Long-term experimental evolution in Escherichia coli. III. Variation among replicate populations in correlated responses to novel environments. Evolution 49: 189-200.
• Blount, Z. D., C. Z. Borland, and R. E. Lenski. 2008. Historical contingency and the evolution of a key innovation in an experimental population of Escherichia coli. Proc. Natl. Acad. Sci. USA 105: 899-7906.
• Meyer, J. R., A. A. Agrawal, R. T. Quick, D. T. Dobias, D. Schneider, and R. E. Lenski. 2010. Parallel changes in host resistance to viral infection during 45,000 generations of relaxed selection. Evolution 64: 3024-3034.
• Blount, Z. D., J. E. Barrick, C. J. Davidson, and R. E. Lenski. 2012. Genomic analysis of a key innovation in an experimental Escherichia coli population. Nature 489: 513-518.

An exciting new twist on the dynamics of adaptation by natural selection is the discovery that fitness can increase “forever” – or at least for a very long time – even in a constant environment.

A power-law model, which has no upper bound, gives a significantly better fit to the mean-fitness trajectories measured in the LTEE populations than does a model with an asymptote.

Moreover, the power law predicts the trajectory of fitness evolution with much greater accuracy.  That is, if we reduce the data so that it includes only the first 20,000 generations, the power law trajectory that fits this truncated dataset accurately predicts fitness out to 50,000 generations (blue trajectory in the figure below).  By contrast, the same procedure with the asymptotic model consistently underestimates the future fitness gains (red trajectory in the figure below).

Also, a dynamical model that incorporates clonal interference (competition between different beneficial mutations) and diminishing-returns epistasis (where the marginal effect of a beneficial mutation declines with increasing fitness) produces trajectories that have the same power-law form.  That, in turn, facilitates estimation of important population-genetic parameters including the rate of beneficial mutations and the average strength of the diminishing-returns epistasis.

• Wiser, M. J., N. Ribeck, and R. E. Lenski. 2013. Long-term dynamics of adaptation in asexual populations. Science  342: 1364-1367.

The figure below shows the grand-mean fitness data (symbols with error bars) over 50,000 generations of the LTEE.  It also shows the trajectories predicted by the power law (blue curve) and by a model with an asymptote (red curve) using only the first 20,000 generations of data.  The figure comes from Wiser et al., 2013, Science; it is shown here under the doctrine of fair use.

I’m sometimes asked to explain, in a general way, the main findings and discoveries from the long-term evolution experiment (LTEE) with E. coli.  So in a series of posts, I will briefly summarize a few of the most important findings and discoveries, as I see them, in very broad, conceptual terms.

I won’t provide the detailed results, but I will list some of the key papers that present them.  Interested readers can read those papers to see the methods, evidence, and data that support the findings and discoveries.

For anyone seeking a general introduction to the LTEE and the advantages that this experimental system offers for studying evolution, I recommend reading the 1994 paper by Lenski and Travisano (cited below) before all others.

Number 1.  Although no one (excepting cranks, zealots, and the uninformed) doubts the fact that adaptation by natural selection occurs, the LTEE provides a simple and compelling demonstration of its power and efficacy.  I’ve heard this point stated repeatedly after talks for the general public and even after seminars for other scientists.

• Lenski, R. E., M. R. Rose, S. C. Simpson, and S. C. Tadler. 1991. Long-term experimental evolution in Escherichia coli. I. Adaptation and divergence during 2,000 generations. American Naturalist 138: 1315-1341.
• Lenski, R. E., and M. Travisano. 1994. Dynamics of adaptation and diversification: a 10,000-generation experiment with bacterial populations. Proc. Natl. Acad. Sci. USA 91: 6808-6814.
• Wiser, M. J., N. Ribeck, and R. E. Lenski. 2013. Long-term dynamics of adaptation in asexual populations. Science 342: 1364-1367.

The figure below shows the step-like fitness trajectory measured for one of the LTEE populations over the course of the first 2,000 generations of the experiment.  The figure comes from Lenski and Travisano, 1994, Proc. Natl. Acad. Sci. USA; it is shown here under the doctrine of fair use.