The combined deletion of BEM3 and NRP1 has been shown previously to cause defects in polarity establishment that exceed the summed effects of their individual deletion (Laan et al., 2015), meaning they exhibit epistasis (Phillips, 2008). The existence of epistatic interactions between these mutations suggests a functional relation between Bem3 and Nrp1. This is surprising, because while Bem3 is known as a GTP Activating Protein (GAP) for Cdc42, the master regulator of cell polarity (Etienne-Manneville, 2004), Nrp1 has never been implicated to be involved in polarity establishment before. Instead, based on the current knowledge about its function, Nrp1 is best described as a prion forming protein that localizes to stress granules formed under conditions of glucose stress (Buchan et al., 2008; Kroschwald et al., 2015). This led us to hypothesize that the deletion of BEM3 and NRP1 may have consequences for the ability of the polarity network to respond to environmental cues that signal different growth modes.

We tested this hypothesis in the context of the ability of S. cerevisiae to perform diauxic growth between glucose and ethanol. In the presence of extracellular glucose, S. cerevisiae maintains a rapid mode of growth by alcoholic fermentation of glucose. The ethanol produced during alcoholic fermentation can be used as an alternative energy source when extracellular glucose drops below a critical level, but only in the presence of extracellular oxygen. The transition from the fermentation of glucose to the respiration of ethanol, a growth phase known as the diauxic shift, is characterized by several physiological changes (Galdieri et al., 2010), which includes changes in the polarized distribution of the actin cytoskeleton (De Virgilio and Loewith, 2006; Galdieri et al., 2010).

We qualitatively determined the effects of deleting BEM3 and NRP1 on the coupling of (diauxic growth) glucose sensing to cell polarity by imaging bem3Δnrp1Δ cells during the diauxic shift (Figure 1A). The diauxic shift was induced by switching from growth media containing glucose as the sole carbon source to one where ethanol was the sole carbon source using a microfluidic device. A wild-type strain subjected to these conditions displayed the expected behavior, which consisted of rapid growth on glucose followed by a short growth pause at the time of the media switch, after which growth was resumed on ethanol media, but at a slower rate compared to growth on glucose media (Brauer et al., 2005). Overall, bem3Δnrp1Δ cells followed the same pattern that we observed for the wild-type cells, but critically failed to produce buds during growth on ethanol media. Instead, isotropic growth was sustained in these cells up to the point where it induced cell death by lysis. Based on the link between polarity defects and an increase in cell size, we deduced that our observations for the bem3Δnrp1Δ phenotype are the result of the inability of the polarity network to respond appropriately to the physiological changes that occur during the diauxic shift. Specifically, while bem3Δnrp1Δ mutants are generally less fit than the wild-type strain, the cellular defect that leads to a lower fitness differs between conditions of standard vegetative growth and conditions where the cells must respond to the diauxic shift. During vegetative growth (2% glucose in Figure 1A) bem3Δnrp1Δ cells proliferate, but do so at a slower rate than wild-type cells. After the media switch (transition from 2% glucose to 3% ethanol in Figure 1A) bem3Δnrp1Δ cells enlarge, but are unable to divide.

Next, we quantified the effect of deleting BEM3 and NRP1 on the diauxic shift using Optical Density (OD) measurements of population growth in order to obtain growth curves for each strain (Figures 1B, D). The diauxic shift was clearly visible in the growth curves as a transition period between two exponential growth phases with different growth rates. For technical reasons (see Supplementary Figure 1), we used media containing a high glucose concentration (2%) to quantify growth before the diauxic shift and a lower glucose concentration (0.1%) to quantify growth after the diauxic shift. We extracted the growth rate during the exponential phase before and after the occurrence of the diauxic shift by calculating the slope of the linear portion of the growth curve when plotted on a semi-log scale. These values were subsequently converted into their corresponding doubling times (T_PRE–shift and T_POST–Shift). This analysis revealed that bem3Δnrp1Δ populations have a significantly longer doubling time than the wild-type both before and after onset of the diauxic shift (Figures 1C, E). While it is expected that the overall fitness defect of bem3Δnrp1Δ mutants will lead to longer doubling times both before and after the diauxic shift, we argue based on our microfluidic experiment (Figure 1A) that the physiological cause that leads to a lower doubling time is different between the two conditions. Before the diauxic shift, bem3Δnrp1Δ cells divide at a slower rate than the wild-type strain due to a defect in polarity establishment which causes a longer T_PRE–shift. In contrast, T_POST–Shift is affected by both the slower division rate and the higher death rate of bem3Δnrp1Δ cells as the polarity defect becomes much more severe at the onset of the diauxic shift. In support of this idea, we found that bem3Δnrp1Δ populations stop growing at a significantly lower OD than wild-type populations after the diauxic shift (ratio wild-type: bem3Δnrp1Δ = 2.25, Figure 1F), while both strains enter the diauxic shift at approximately the same density (ratio wild-type: bem3Δnrp1Δ = 1.15, Figure 1F). We therefore interpret these results as indications that the defects in polarity establishment caused by the deletion of BEM3 and NRP1 makes the polarity network insensitive to environmental cues that signal the onset of the diauxic shift. The loss of responsiveness to these cues causes an inability to establish a polarity site when the physiological changes related to the diauxic shift have taken place, leading to prolonged isotropic growth and an increase in cell size.

We sought to determine whether the environment is the decisive factor that controls the adaptive value of restoring the cellular response to the diauxic shift during evolution. To do this, we took an experimental approach and evolved several parallel wild-type and bem3Δnrp1Δ populations in two frequently used set-ups for experimental evolution (Figure 2). In the first set-up, the batch culture, nutrient levels vary over time and cells experience periods of glucose depletion several times throughout the experiment (Brauer et al., 2005; Gresham and Dunham, 2014). Mutations that allow cells to correctly coordinate the physiological changes necessary to pass through the diauxic shift with those that regulate polarity establishment are therefore expected to be beneficial during evolution in a batch culture set-up, as this extends the overall number of progeny that a cell can produce before each passage. In the second set-up, the glucose limited continuous culture, nutrient concentrations remain constant after a steady state is reached and growth is maintained at a constant rate (Brauer et al., 2005; Gresham and Dunham, 2014). These constant environmental conditions have the consequence that cells do not induce the majority of the cellular responses that are associated with the diauxic shift (Brauer et al., 2005). Thus, the ability to perform diauxic growth appears as a dispensable trait during evolution in a continuous culture. Based on the theoretical assumptions that traits that do not experience selective pressure (1) tend to deteriorate and (2) are unlikely to fix mutations that improve their function during evolution, we expect restoration of diauxic growth by bem3Δnrp1 populations to emerge only during batch culture evolution.

We evolved a total of 14 bem3△nrp1△ populations and 2 wild-type populations in the glucose limited continuous culture for 70 generations. The parameter T_PRE–Shift was used as a proxy for adaptations that restore the polarity defect caused by the deletion of BEM3 and NRP1, but that do not necessarily improve the ability of the polarity network to respond to cues that signal the onset of the diauxic shift. Alternatively, T_POST–Shift was used as a proxy for adaptations that improve the response of cells to the diauxic shift. The values of T_PRE–Shift and T_POST–Shift of the evolved cell lines were determined by reviving the evolved population from a frozen stock and measuring the change in OD over time in media containing 2% and 0.1% glucose, respectively. This procedure is the same as what was done to determine T_PRE–Shift and T_POST–Shift for the ancestral wild-type and bem3△nrp1△ populations (see Figures 1B, D, and the section “Materials and methods”).

Comparison of T_PRE–Shift between the evolved populations and their ancestor (Figure 3A) revealed that all evolved populations had either a similar or lower value for T_PRE–Shift relative to their ancestor. In contrast, we find that for T_POST–Shift half of the evolved bem3△nrp1△ populations (7/14) had a lower doubling time, while the other half (7/14) had a longer doubling time relative to their ancestor, indicating that changes in diauxic growth do not affect fitness in a continuous culture. A similar trend for T_PRE–Shift and T_POST–Shift was visible for our two evolved wild-type populations. Taken together, these observations support our initial view that a continuous culture selects for faster vegetative growth, but not diauxic growth.

The finding that some of bem3△nrp1△ populations evolved in the continuous culture show improvements in T_POST–Shift could be explained by a possible interdependence of T_PRE–Shift and T_POST–Shift: improvements in T_PRE–Shift may be caused by mutations that increase the overall rate of cell division and these mutations will therefore also lead to improvements in T_POST–Shift. However, these mutations do not necessarily also resolve the high death rate of bem3△nrp1△ mutants at the start of the diauxic shift (Figure 1A), which may be a major factor that determines T_POST–Shift. To verify that a decrease in T_POST–Shift relates to adaptations that resolve the high death rate, we imaged cells from the evolved population with the lowest (fastest growing, CCE1) and highest (slowest growing, CCE2) value for T_POST–Shift during the diauxic shift (Figure 3C) using the same microfluidic set-up we used in Figure 1. In agreement with our expectations, the results showed that the phenotype of CCE1 after switching to ethanol media was qualitatively more similar to that of our ancestral wild-type strain and CCE1 cells were able to resume proliferation after the diauxic shift. Alternatively, the phenotype of CCE2 was more similar to the ancestral bem3△nrp1△ strain, as CCE2 cells enlarged and were frequently unable to divide after the onset of the diauxic shift.

We evolved 8 bem3△nrp1△ and 2 wild-type populations in a batch culture with a daily passaging procedure. We initially maintained the same number of generations for evolution as we had done for the continuous culture (70 generations), but after assessing our proxies for fitness we were unable to identify any significant changes in the values of T_PRE–Shift and T_POST–Shift between the evolved populations and their ancestors (see Supplementary Figure 2). We assumed that this is due to the frequent population bottlenecks that occur during the passaging of the populations, which can slow down the rate of adaptation by purging beneficial mutations from the population (Wein and Dagan, 2019). We provide an estimate of the effect of population bottlenecks on the fixation dynamics of beneficial mutations in Supplementary Section 1, which shows that bottlenecks vastly increase the expected number of generations that are required before a beneficial mutation that fixates in the population will emerge. To compensate for this effect of population bottlenecks, we allowed our batch culture experiment to run for an additional 230 generations such that the total number of generations was 300.

We found that all evolved populations grew faster than their ancestors, both before and after the diauxic shift (Figure 3B). The fact that we do not observe populations that evolve toward a state where the doubling time after the diauxic shift becomes longer suggests that these pathways are inaccessible during evolution in a batch culture. Taken together, these results imply that the environmental variability that exists in the batch culture imposes constraints on the diauxic growth pattern that can be attained during evolution, allowing only those where growth on both nutrients is improved, while the stable environment of the continuous culture releases some of these constrains. As a result, phenotypes that have evolved to perform well during the diauxic shift, presumably through evolutionary repair of the polarity defects caused by deleting BEM3 and NRP1, only reproducibly emerge in a batch culture. However, although the degree of reproducibility is lower, similar phenotypes do frequently evolve in a continuous culture. This indicates that evolutionary constraints imposed by the environment are not sufficient to explain the restoration of the responsiveness of the polarity network to cues of the diauxic shift during evolution.

To understand the molecular basis of the different adaptations of T_PRE–Shift and T_POST–Shift we observed in our continuous and batch cultures, we performed Whole Genome Sequencing (WGS) on the 22 evolved bem3△nrp1△ lines and the 4 evolved wild-type controls and compared them to the genome of their wild-type ancestor (see section “Materials and methods”). We looked for patterns of parallel evolution by restricting our analysis to genes that were mutated in at least 2 different populations evolved in the same environment. This resulted in a total of 88 genes that acquired non-synonymous mutations or indels in 2 or more evolved populations (including the wild-type lines).

The most notable environment-specific mutations were the early stop codons in WHI2 that frequently occurred in the populations evolved in the continuous culture: 12 out of 14 evolved bem3△nrp1△ and both evolved WT controls had mutated WHI2. Disruptive mutations in WHI2 have also been reported in other experimental evolution studies that used nutrient limited continuous cultures (Kvitek and Sherlock, 2013; Hong and Gresham, 2014) and these mutations therefore likely provide a general advantage during adaptation to nutrient-limiting conditions.

Because we saw the same phenotype emerge in the batch culture and continuous culture (populations that decreased T_POST–Shift), we questioned whether the molecular basis of these adaptations were similar. In total, 22 genes were mutated in at least 2 of the bem3△nrp1△ lines evolved in each environment. We grouped these genes according to their Biological Process Gene Ontology (GO) annotation on the Saccharomyces Genome Database. This revealed that populations evolved in a continuous culture had more mutations in genes involved in the stress response, while populations from the batch culture had slightly more mutations in genes related to transcription and translation. We then split the evolved populations into two groups: those that evolved to decrease T_POST–Shift (7/14 populations of the continuous culture and 8/8 populations of the batch culture) and those that evolved to increase T_POST–Shift (7/14 populations of the continuous culture and 0/8 populations of the batch culture). Interestingly, populations with a decreased T_POST–Shift had more mutations in lipid metabolic genes than those that did not decrease T_POST–Shift. Of the 14 bem3△nrp1△ populations that were evolved in the continuous culture, 6/7 populations with a decreased T_POST–Shift had mutations in the IPT1, while we only found mutations in this gene in 1/7 populations with an increased T_POST–Shift (Figure 4). In the batch culture populations, 2/8 had acquired mutations in IPT1, while 3/8 had mutations in SUR1. Notably, Ipt1 acts directly downstream of Sur1 in the pathway for the synthesis of complex sphingolipids (Thevissen et al., 2000; Dickson et al., 2006; Morimoto and Tani, 2015).

Based on this correlation we hypothesize that, after the deletion of Bem3 and Nrp1, the robustness of the polarity module during the diauxic shift can be (partially) restored by changes in the lipid composition of the plasma membrane. Interestingly, this strategy appears to be dominant for repairing the defect caused by the deletion of BEM3 and NRP1 regardless of whether diauxic growth is part of the selective environment.

All strains used in this study are derived from the W303 background and are MATa haploid cells. We used yLL132a as our WT strain and yLL143a as our bem3△nrp1△ strain (Laan et al., 2015), which has the same genetic background as yLL132a, but with BEM3 and NRP1 replaced with, respectively, the natMX4 (clonNAT-Nourseothricin resistance) and hphMX4 (Hygromycin B resistance) cassettes. For batch culture evolution experiments, standard rich media (10 g/L Yeast Extract, 20 g/L Peptone and 20 g/L Dextrose) was used and was prepared by dissolving 50 g/L from a premixed batch of ingredients (Sigma-Aldrich) in H₂O. For chemostat evolution experiments the same premix was used, but supplemented with 19 g/L extra Yeast Extract and 9.5 g/L extra Peptone to obtain a final dextrose concentration of 1 g/L. A total of 0.1 mg/mL Ampicillin was added to the chemostat media as a safeguard against bacterial contamination. Microscopy experiments were performed in Synthetic Complete (SC) media prepared from Complete Supplement Mixture without amino acids, riboflavin and folic acid (750 mg/L), Yeast Nitrogen Base (6.9 g\L) and either Dextrose (2%w/v) or Ethanol (3%v/v) as a carbon source. All media was filter sterilized to avoid degradation of components during autoclaving.

Batch culture evolution experiments were started with 10 bem3△nrp1△ and 2 WT cultures. The cultures were derived from a single bem3△nrp1△ and a single WT liquid culture initiated from a glycerol stock and grown to saturation for 2 days in YP 2 %D in a roller drum (set at 40 RPM) at 30°C. After the cultures reached saturation, 10 μL of each starter culture was diluted into 10 mL of fresh YP 2 %D and were placed back into the roller drum. The cultures were diluted by 10 μL into 10 mL of fresh YP 2 %D every 24 ± 2 h. After each dilution, the OD at 600 nm of the remaining culture was measured using the same procedure as described above for the chemostat evolution experiment. Because batch cultures involve frequent population bottlenecks that can reduce genetic variation and possibly purge beneficial mutations (Gresham and Dunham, 2014; Gresham and Hong, 2014), it might take longer for an adaptive mutation to settle in the population. To compensate for this effect, the number of generations that the populations were evolved in a batch culture setting was increased to 300 generations (an additional 230 generations compared to the populations evolved in a continuous culture).

Growth curves were obtained by measurements using a plate reader (Tecan Infinite 200 Pro). Cells were inoculated from a glycerol stock in YP 0.1 %D liquid media and grown to saturation for 2 days in a roller drum at 30°C. On the day of the measurement, the saturated cultures were diluted 1000X into either fresh YP 0.1 %D or fresh YP 2 %D, depending on whether we wanted to measure the doubling time before (T_PRE–Shift) or after (T_POST–Shift) the diauxic shift. A total of 100 μL of this culture was pipetted into each well of a sterile 96-well plate (Nunc™Edge 2.0, Thermo Scientific™) with the edge moats filled with 1.7 mL of sterile H₂O. Each plate contained multiple technical replicates of each sample. As a control for contamination and to allow for background subtraction for downstream processing, 8 wells were filled with blank medium. Measurements were taken during incubation at 30°C in the plate reader using the following protocol: First, the cells were shaken for 1000 s (linear shaking, 1 mm amplitude) without measurement. After this, the absorbance of each well was measured every 7 min with intermittent shaking (260 s, linear, 1 mm amplitude) for 48 h.

Doubling times for pre-diauxic (T_PRE–Shift) and an post-diauxic (T_POST–Shift) growth were extracted from the growth curve measurement in YP 2 %D and YP 0.1 %D, respectively. First, the measured OD values were blanked using the time average value of one of the wells containing blank media. Then, the data was converted to semi-log data by taking the natural logarithm of the blanked OD values. A home written MATLAB script was used to fit a line to the linear portion of the semi-log data to obtain the growth rate during pre-diauxic or post-diauxic growth. These growth rates were converted into doubling times using the following relation:

where T_d is the doubling time corresponding to the growth rate μ obtained from the slope of the linear fit.

Cells were grown to log phase in SC media containing 2% dextrose. Clumps of cells were dissociated prior to imaging by sonicating (Q500 Sonicator, QSonica) in a sealed Eppendorf tube using a cup horn at 70% amplitude for 2 min (cycle of 30 s pulse on, 15 s pulse off). After sonication, each sample was diluted to the same optical density in fresh Synthetic Complete media containing 2% dextrose. Cells were trapped in a microfluidic culture chamber (CellASIC ONIX Y04C-02, Merck–Millipore) after flushing the culture chambers with fresh media for 20 min using a pressure of 8 psi. Brightfield images were taken with a Nikon Eclipse Ti-E inverted Microscope using a 60x objective (Plan Apo λ 60X oil, NA: 1.40) with 1 min intervals. During imaging, cells were maintained in a constant flow of media using a pressure of 1 psi. Cells were subjected to a media switch by changing from an inlet with SC media containing 2% dextrose to an inlet with SC media with 3% ethanol after 8 h of imaging.

We extracted genomic DNA from liquid cultures grown for two overnights for each of the 16 chemostat samples, 10 serial dilution samples and a non-evolved yLL132a ancestor with the MasterPure Yeast DNA Purification Kit (Epicentre, Madison, WI, USA) following the manufacturer’s protocol. We included a RNase A (Qiagen, Hilden, Germany) treatment step in the protocol and collected DNA in a final volume up to 30 μL H₂O. We pooled up to three extractions per sample using the Genomic DNA clean and Concentrator kit (Zymo Research, Irvine, CA, USA), following the supplied protocol. We eluted DNA in a final volume of 30 μL. We assessed DNA quality by 0.8 % agarose gel electrophoresis and quantity by fluorometry using a Qubit 4.0 Fluorometer (Invitrogen, Carlsbad, CA, USA). Samples were individually barcoded and pooled into a single library with the NEB Next Ultra DNA Library Prep Kit (New England Biolabs, Ipswich, MA, USA) and sequenced on a HiSeq machine (Illumina, San Diego, CA, USA) by Novogene (Beijing, China).

We first checked raw paired-end reads (150 bp) for quality with the FASTQC toolkit (version 0.11.7).¹ We removed low quality ends (Quality scores <20; and first 9 bases of all reads), and removed duplicates with the FastX toolkit (version 0.0.14).² We downloaded the R64-1-1 S. cerervisiae genome from the Saccharomyces Genome Database (SGD)³ and used it as our reference. We indexed the reference genome with the Burrows-Wheeler Aligner [BWA; version 0.7.17; (Li and Durbin, 2010)], and SAMtools [version 1.8; (Li and Durbin, 2009; Li, 2011)], and generated a dictionary with Picard (version 2.18.5).⁴ We mapped sequences from all samples individually to the reference with BWA-MEM sorted and indexed mapped reads into a BAM file with SAMtools. We performed multisample SNP calling and additional indexing with SAMtools and BCFtools (version 1.8). We plotted and checked statistics, e.g., TS/TV and quality of sites and read depth, with BCFtools. These statistics were used to filter out SNPs and Indels with low quality sites (QUAL > 30), low read depth (DP > 20), and variants in close proximity to gaps (SnpGAP 10). We annotated the VCF file with snpEff [version 4.3T; (Cingolani et al., 2012)] with R64- 1-1.86. We then retrieved variants (SNPs and indels) of interest through comparison of variants between the reference strain, the ancestor strains, and the evolved strains. We excluded variants that were different between R64 and all our W303 samples, as these merely display differences between the two genetic backgrounds [see e.g., (Ralser et al., 2012)]. Synonymous variants, variants in non-coding regions, and stop retained variants were excluded. Mutations in telomeric regions and in Long Terminal Repeats (LTRs) were excluded from analysis due to the natural variation that occurs in the genomic sequence of these regions. To find causative mutations, we looked for genes that mutated in at least two evolved lines, excluding those that appeared only in the mutant line(s) from one environment and the wild-type line(s) of the other environment. From the resulting list of genes, genes corresponding to dubious or uncharacterized Open Reading Frames (ORFs) were removed according to their description on SGD. Two genes (RPS29B and ECM33) had acquired the same mutation across all 22 parallel evolved bem3△nrp1△ lines that sweeped the population, suggesting that these mutations were acquired in the ancestor before the different cell lines were split. Although these mutations might have some fitness benefit in the bem3△nrp1△ background, they do not explain the adaptation we observe during our evolution experiments and we therefore excluded them from further analysis. We used the OncoPrint function from the ComplexHeatmap package (Gu et al., 2016) available in R (version 4.2.3) to visualize the relevant mutations in our evolved lines as a heatmap.