Adaptation to High Ethanol Reveals Complex Evolutionary Pathways

Abstract

Tolerance to high levels of ethanol is an ecologically and industrially relevant phenotype of microbes, but the molecular mechanisms underlying this complex trait remain largely unknown. Here, we use long-term experimental evolution of isogenic yeast populations of different initial ploidy to study adaptation to increasing levels of ethanol. Whole-genome sequencing of more than 30 evolved populations and over 100 adapted clones isolated throughout this two-year evolution experiment revealed how a complex interplay of de novo single nucleotide mutations, copy number variation, ploidy changes, mutator phenotypes, and clonal interference led to a significant increase in ethanol tolerance. Although the specific mutations differ between different evolved lineages, application of a novel computational pipeline, PheNetic, revealed that many mutations target functional modules involved in stress response, cell cycle regulation, DNA repair and respiration. Measuring the fitness effects of selected mutations introduced in non-evolved ethanol-sensitive cells revealed several adaptive mutations that had previously not been implicated in ethanol tolerance, including mutations in PRT1, VPS70 and MEX67. Interestingly, variation in VPS70 was recently identified as a QTL for ethanol tolerance in an industrial bio-ethanol strain. Taken together, our results show how, in contrast to adaptation to some other stresses, adaptation to a continuous complex and severe stress involves interplay of different evolutionary mechanisms. In addition, our study reveals functional modules involved in ethanol resistance and identifies several mutations that could help to improve the ethanol tolerance of industrial yeasts.

Fig 1. Experimental setup.

(A) Experimental evolution of prototrophic, isogenic populations of different ploidy (haploid (VK111), diploid (VK145) and tetraploid (VK202)) for increased ethanol tolerance was performed in a turbidostat. Every 25 generations, the ethanol concentration in the media was increased in a stepwise manner (starting at 6% (v/v) and reaching 12% at 200 generations). Increasing the ethanol concentration from 10% to 11% dramatically reduced growth rate of evolving cells. Therefore, instead of increasing ethanol levels, we first reduced the ethanol level to 10.7% after 100 generations. (B) Red circles represent sampling points (indicated as number of generations) for which whole-genome sequencing was performed. For each circle, heterogeneous populations as well as three evolved, ethanol tolerant clones were sequenced. Sequencing of the population sample of reactor 4 at 200 generations failed, so this data is omitted from the manuscript. For generation 80 of reactor 1, only population data is available.

Fig 2. Evolved populations are more ethanol tolerant.

Evolved populations from different reactors show increased fitness in EtOH. Fitness was determined for population samples of each reactor after 40 generations (blue) and 200 generations (red). Fitness is expressed relative to the ancestral strain of each reactor (haploid for reactor 1–2; diploid for reactor 3–4 and tetraploid for reactor 5–6). Data represent the average of three independent measurements, error bars represent standard deviation. Populations with mutator phenotypes showed higher standard deviations. This might be due to the large supply of mutations generated by these mutators, which could affect mutation-selection balance.

Fig 3. Haploid lineages diploidized during adaptation to EtOH.

Flow cytometry analysis of DNA content (stained by propidium iodide) of evolved populations, compared to ancestral haploid (red) and diploid (blue). For the clonal ancestral samples, two peaks are observed, corresponding to the G1 and G2 phase of the cell cycle. Evolved populations sometimes display three peaks, indicative of both haploid and diploid subpopulations.

Fig 4. Mutations present in evolved clones isolated from all reactors at 200 generations.

Circle diagrams depict the number and types of mutations identified in evolved clones at 200 generations of reactor 1, reactor 2, reactor 3, reactor 4, reactor 5 and reactor 6. Data for individual clones isolated at different time points can be found in S1 and S2 Files.

Fig 5. Copy number variation in evolved clones of reactor 2 and 3 isolated at 200 generations.

Genome view of yeast chromosomes and CNV patterns from a sliding window analysis. The y-axis represents log2 ratios of the coverage observed across 500bp genomic windows to the coverage expected in a diploid genome without CNVs. Area of the plot located between the red lines (from -0.23 to -1) marks putative CNV loss events, whereas region between the blue lines (from 0.23 to 0.58) marks putative CNV gain events. Data for other reactors is shown in S3 File. It should be noted that the amplified region of chromosome XII observed in some of our clones does not correspond to the ribosomal DNA genes.

Fig 6. Dynamics and linkage of mutations in evolved populations of reactor 2.

Mutations (reaching a frequency of least 20% in the evolved population samples) and corresponding frequencies were identified from population sequencing data. Muller diagram represents the hierarchical clustering of these mutations, with each color block representing a specific group of linked mutations. Indels are designated with ^I, whereas heterozygous mutations are in italics. Mutations present as heterozygous mutations in all clones of a specific time point and present at a frequency of 50% in the population, are depicted as a frequency of 100% in the population, since it is expected that all cells in the population contain this mutation. After 80 generations, a mutator phenotype appeared in this reactor (indicated by arrow under graph), which coincides with the rise in frequency of an indel in the mismatch repair gene MSH2. Frequencies of haplotypes can be found in S2 Table. Dynamics and linkage for reactor 1 is shown in S7 Fig.

Fig 7. Adaptive pathways are involved in cell cycle, DNA repair and protoporphyrinogen metabolism.

Shown is the sub-network that prioritizes putative adaptive mutations by applying PheNetic on all selected mutations, excluding those originating from the populations with a mutator phenotype i.e. reactor 2 and 6. The nodes in the network correspond to genes and/or their associated gene products. Node borders are colored according to the reactors containing the populations in which these genes were mutated. Nodes are colored according to gene function, for each gene the most enriched term is visualized (grey indicates no enrichment). Cell cycle related processes have been subdivided into DNA replication and interphase. The edge colors indicate different interaction types. Orange lines represent metabolic interactions, green lines represent protein-protein interactions, red lines represent protein-DNA interactions. Sub-networks extracted by separately analyzing the mutated genes observed in each of the different populations (reactors) are shown in S8–S13 Figs.

Fig 8. Single mutations present in evolved populations can increase ethanol tolerance of a non-adapted strain.

Plots show the average selection coefficient (s _mut) as a function of ethanol concentration for (A) pca1 ^C1583T, (B) prt1 ^A1384G, (C) ybl059w ^G479T, (D) intergenic ChrIV A1489310T, (E) hem13 ^G700C, (F) intergenic ChrXII C747403T, (G) hst4 ^G262C, (H) vps70 ^C595A, and (I) mex67 ^G456A. Superscripts denote the exact nucleotide change in each of the mutants tested. YECitrine-tagged mutants were competed with the mCherry-tagged parental strain (orange dots); dye-swap experiments were carried out by competing the mCherry-tagged mutants with the YECitrine parental strain (blue dots), except for (I). Error bars show the S.E.M. from three experimental replicates. Asterisk show P-values from the one-way ANOVA tests of the mean differences in 4–8% ethanol compared to fitness in 0% ethanol: * p < 0.05; ** p < 0.01; ***p < 0.005. P values can be found in S7 Table.