My interest in the effect of population size on the dynamics and repeatability of evolution has a long history. For my PhD (I graduated in 1996), I tested a then popular model for the evolutionary significance of sex and recombination, which required negative interactions among deleterious mutations – the type of mutations that would readily accumulate in small populations. When I next went for a postdoc to Rich Lenski’s lab in Michigan, I had to switch concepts and learned about evolution in larger populations, dominated by rare beneficial mutations. Rich and his PhD student Phil Gerrish had just coined the term ‘clonal interference’ for the competition between different adaptive mutants in sufficiently large asexual populations [1], which fundamentally changed the conventional population genetics’ view of adaptation via isolated selective sweeps of beneficial mutations. Together with postdoc Cliff Zeyl, we tested how clonal interference would affect the rate of adaptation by comparing adaptation in populations with different supplies of beneficial mutations by varying their size, mutation rate and initial level of adaptedness. We found that increases in the mutation supply hardly accelerated adaptation, demonstrating that clonal interference would more or less put a “speed limit” on the adaptation rate of asexual populations [2].

Now I was convinced of the impact of clonal interference on adaptation, I became interested in its consequences for the predictability of evolution, as clonal interference should increase the contribution of the same small number of large-benefit mutations. And here my interest in population size and mutation interactions came together. Postdoc Danny Rozen and PhD student Michelle Habets had done an evolution experiment comparing adaptation in small and 50-fold larger bacterial populations in two nutrient environments [3]. As expected, their large populations showed less variation in fitness improvements, but surprisingly, small populations with slow initial improvements surpassed the final improvements of the larger populations in the complex nutrient environment. These patterns suggested that small populations would sometimes avoid “local fitness traps” on a rugged fitness landscape. Later, postdoc Merijn Salverda confirmed this result using evolution experiments with a single antibiotic-degrading enzyme, which allowed us to analyze the underlying mutations [4].

Independently, long-term collaborator Joachim Krug from the University of Cologne had also become interested in the evolutionary role of population size on rugged fitness landscapes. Together with his postdoc Kavita Jain, he had published theoretical work confirming the potential of small-population adaptive benefits on rugged fitness landscapes [5]. Once we started to collaborate, this work was extended by Joachim’s postdoc Ivan Szendro, who simulated adaptation on the rugged fitness landscape of a fungus, which I had analyzed as a PhD student. Ivan’s simulations showed that the repeatability of mutation trajectories varied in a complex way with population size [6]. By analyzing the distinct roles of single and double mutants, Ivan found that repeatability initially increased with increases of the population size due to stronger clonal interference among mutants carrying alternative single beneficial mutations, but slightly decreased again in very large populations where double mutants would drive adaptation, since they were less constrained by the adaptive landscape.

Around the same time (2013), postdoc Martijn Schenk, who worked with Joachim and me on the evolvability of Merijn’s antibiotic-degrading enzyme, decided to start a “small side project”. Motivated by our previous findings, Martijn was interested in the joint influence of fitness landscape and population size on the dynamics and repeatability of adaptation, now by analyzing the genome-wide contribution of mutations in E. coli. He challenged 72 small and 24 100-fold larger populations with increasing concentrations of the antibiotic cefotaxime, measured resistance improvements and sequenced the genomes of the evolved bacteria. Different from our previous results for a single enzyme, Martijn found no small-population adaptive benefits, suggesting that E. coli’s much larger genome presented a smoother adaptive landscape. And then he left for a permanent position, as it goes in the uncertain lives of postdocs.

But luckily a new postdoc in the Cologne-based consortium, Mark Zwart, came to rescue this project. Mark’s analyses of the genomic changes revealed more parallel mutations in the large populations, as we expected from stronger clonal interference. However, when Mark looked in more detail at the different mutation classes, he was up for a surprise: whereas large populations showed more parallel point mutations, parallel structural variants (i.e. large deletions and duplications) occurred more often in small populations! How could this be? The simplest explanation we could think of, was that the joint influence of variation in the fitness effects of mutations (‘selection bias’) and variation in the rates of mutations (‘mutation bias’) had different impact in small and large populations. The stronger impact of clonal interference in large populations required that point mutations and structural variants would trade-off their rates and fitness effects: if point mutations came with lower rates, but larger benefits than structural variants, point mutations may dominate in large and structural variants in small populations.

Would our hypothesis stand the test? One piece of support was found by Mark’s analyses of samples from multiple time points of a subset of populations, which showed that structural variants appeared earlier, but fixed later than point mutations, especially in large populations. Other support came from Joachim’s postdoc Sungmin Hwang, who used evolutionary simulations to infer parameter values for the different mutation classes that best explained their observed frequencies in the 96 evolved strains. Sungmin’s best-fit estimates indicated roughly three-fold larger fitness effects combined with 300-fold lower rates for point mutations relative to structural variants, again supporting our hypothesis. Independent support came from zooming into the large class of point mutations, where we distinguished loss-of-function from putative gain-of-function mutations. The latter, arguably less frequent, mutations were found more often and more in parallel in large than in small populations, whereas the more numerous loss-of-function mutations showed no difference, consistent with larger benefits driving the preferential use of gain-of-function mutations in large populations.

What does this all mean? Our findings demonstrate the separation by clonal interference of mutation classes with different rates and fitness effects. As a consequence, populations of different size fix different mutation types. Interestingly, we found that these different mutation ‘choices’ may impact the tempo of evolution by changing its mode, because small populations often deleted the – initially inactive – antibiotic-degrading enzyme, while large populations more often activated this enzyme via gain-of-function point mutations, leading to 10-fold higher resistance levels (Fig. 1). While this has been a slowly maturing project, taking roughly 60 times longer than the 500-generation experiment itself to understand it, all time and effort have been more than worth it. Not only did the project yield several new methods and insights, it already spawned two spin-off studies, and several future plans have been inspired by it.

Link to Schenk, Zwart et al., 2022, Population size mediates the contribution of high-rate and large-benefit mutations to parallel evolution, Nature Ecology and Evolution, DOI 10.1038/s41559-022-01669-3

1. Gerrish, P.J. and R.E. Lenski, The fate of competing beneficial mutations in an asexual population. Genetica, 1998. 102/103: p. 127-144.
2. de Visser, J.A.G.M., et al., Diminishing returns from mutation supply rate in asexual populations. Science, 1999. 283: p. 404-406.
3. Rozen, D.E., et al., Heterogeneous adaptive trajectories of small populations on complex fitness landscapes. PLoS One, 2008. 3: p. e1715.
4. Salverda, M.L.M., et al., Adaptive benefits from small mutation supplies in an antibiotic resistance enzyme. Proceedings of the National Academy of Sciences, 2017. 114(48): p. 12773-12778.
5. Jain, K., J. Krug, and S.-C. Park, Evolutionary advantage of small populations on complex fitness landscapes. Evolution, 2011. 65: p. 1945–1955.
6. Szendro, I.G., et al., Predictability of evolution depends non-monotonically on population size. Proceedings of the National Academy of Sciences USA, 2013. 110: p. 571-576.