We used a custom microfluidic device to monitor the growth of yeast cells exposed to periodic environmental fluctuations for up to 24 hr. Cells were imaged every 6 min in microfluidic chips containing five independent sets of channels connected to five growth chambers (Figure 1b; Figure 1—figure supplement 1), allowing five different conditions per experimental run, with five technical replicates for each condition. Computer-controlled fluidic valves were programmed to generate temporal fluctuations of the media dispensed to cells with rapid transitions (<2 min) from one medium to another (Figure 1—figure supplement 1). The rate of cell division was then quantified using automated image analyses (see Materials and methods). With this experimental system it is possible to determine not only how temporal fluctuations of individual parameters of the environment (e.g. a repeated stress or carbon source fluctuations) impact cell proliferation but also what are the impacts of the dynamic interactions of two environmental parameters. Here, we specifically study how periodic fluctuations of a metabolic resource (glucose concentration switching between 0% and either 2% or 0.1% wt/vol) and osmotic stress (sorbitol concentration switching between 0 and 1 M) interact to alter the proliferation of yeast cells (Figure 1).

To determine how the temporal dynamics of osmotic stress altered cell proliferation, we first measured the division rate of yeast cells exposed to fluctuations between 1 M sorbitol and no sorbitol at periods ranging from 12 to 480 min. In these experiments the time-averaged osmotic concentration was constant (i.e. cells were exposed to 1 M sorbitol half of the time in all conditions), which is important when studying the effects of the frequency, and not intensity, of osmotic stress on cell dynamics. The average division rate strongly decreased as the frequency of osmotic shock increased (Figure 2a), both in 2% glucose (2.2-fold reduction of division rate between periods of fluctuation T=192 min and T=12 min) and in 0.1% glucose (3.8-fold reduction of division rate between periods of fluctuation T=192 min and T=12 min). These results are consistent with findings from Mitchell et al., 2015, who attributed the drastic decrease in cellular growth observed at high frequency of osmotic shocks to overactivation of the HOG pathway. However, we also observed a clear negative relationship between the frequency of hyperosmotic stress and the division rate of HOG pathway mutants (Figure 2—figure supplement 3), indicating that the growth slowdown was not only explained by overactivation of the HOG pathway. The temporary reduction of division rate observed in response to a hyperosmotic shock in wild-type (Figure 2e–f; Figure 2—figure supplement 1; Figure 2—figure supplement 2a-c) and mutant (Figure 2—figure supplement 3) cells could also contribute to the negative relationship between division rate and frequency of osmotic stress. This negative relationship and the fact that cellular growth can rapidly recover after exposure to high osmolarity (Figure 2e–f) both indicate that yeast cells are more sensitive to repeated than persistent hyperosmotic stress.

Figure 2 with 3 supplements see all

Download asset Open asset

Figure 2—source data 1

Source data for Figure 2.

https://cdn.elifesciences.org/articles/88750/elife-88750-fig2-data1-v1.zip

Download elife-88750-fig2-data1-v1.zip

Next, we wondered whether the frequency of a different type of environmental fluctuation would also affect cell division rate. To answer this question, we quantified the division rate of cells exposed to periodic transitions between a medium without carbon source and the same medium complemented with either 0.1% or 2% glucose at periods ranging from 12 to 480 min. In contrast to the negative effect of osmotic stress frequency, we observed a positive relationship between the frequency of glucose availability and division rate (Figure 2b): cells divided faster when glucose availability fluctuated rapidly (0.0051 division/min at a fluctuation period T=12 min, corresponding to a doubling time of 136 min) than slowly (0.0027 division/min at a fluctuation period T=192 min, corresponding to a doubling time of 257 min). However, this behavior was only observed in 2% glucose: the frequency of glucose availability did not significantly impact the division rate in 0.1% glucose (Figure 2b). Under periodic fluctuations of 2% glucose, the division rate was lower during half-periods without glucose than during half-periods with glucose (Figure 2—figure supplement 2), as expected. However, this difference depended on the frequency of glucose fluctuations: the average division rate during half-periods without glucose was higher at high frequency (small period) than at low frequency (large period) of fluctuations (Figure 2—figure supplement 2). Therefore, the effect of the frequency of glucose availability on the division rate in 2% glucose is likely due to a delay between glucose removal and growth arrest: cell proliferation never stops when the frequency of glucose depletion is too fast.

Overall, we observed two opposing patterns of cell proliferation when we varied the temporal dynamics of the metabolic environment and external osmolarity. The division rates were highest for low-frequency sorbitol fluctuations (0.0064 division/min at a fluctuation period T=384 min) and high-frequency 2% glucose fluctuations (0.0051 division/min at a fluctuation period T=12 min); both of these values are close to the division rate in constant 2% glucose (0.0066 division/min). Therefore, with respect to their division rate, cells behave as a low-pass filter for osmotic stress but as a high-pass filter for glucose fluctuations. Moreover, the division rate is similar when the frequencies of glucose availability and sorbitol exposure are both equal to 0.039 min^–1 (intersection of the two curves on Figure 2c), corresponding to a period of 26 min and a division rate of 0.004 division/min. Since current models of the hyperosmotic stress response do not consider interactions with glucose metabolism, whether simultaneous fluctuations of glucose availability and osmotic stress affect cell growth additively or synergistically remains an open question. More generally, characterizing how cells respond to the dynamic phasing of two environmental components is fundamental for understanding how a living system can adapt to complex environmental changes. For these reasons, we next used our microfluidic system to quantify the division rate of cells exposed to dual periodic fluctuations of glucose availability and osmotic stress.

To determine whether glucose availability during hyperosmotic stress impacted cell growth in dynamic conditions, we compared cell division rates under two regimes of dual periodic fluctuations that only differed in the phasing of glucose and sorbitol fluctuations. In the ‘IPS’ regime, glucose depletion and 1 M sorbitol stresses were applied simultaneously for half a period followed by the addition of 2% (or 0.1%) glucose and the removal of sorbitol for the other half of each period of fluctuations (Figure 1e). In the ‘AS’ regime, glucose depletion and 1 M sorbitol were applied alternatively for half a period each (Figure 1f). We first subjected cells to dual fluctuations at a period of 24 min with 2% glucose, corresponding approximately to the period at which the division rate was the same when we only varied glucose availability or osmolarity (Figure 2c). Under both IPS and AS conditions, the division rate was more than twofold lower than under periodic fluctuations of only glucose or sorbitol (Figure 3a), showing that dual environmental fluctuations have a non-additive, synergistic impact on cell growth. Strikingly, cells divided about twice as fast under IPS condition (1.67×10^–3 division/min, corresponding to an average doubling time of 415 min) than under AS condition (9.4×10^–4 division/min, corresponding to an average doubling time of 737 min) when the fluctuation period was 24 min (t-test, p=1.35 × 10^–5; Figure 3a, Figure 3—figure supplement 1a and b) or 96 min (2.98×10^–3 division/min in IPS vs 1.83×10^–3 division/min in AS; p=4.10 × 10^–5; Figure 3b). A similar pattern of faster growth was observed under IPS and AS conditions when we used 0.1% glucose instead of 2% glucose, for both a fluctuation period of 24 min (0.84×10^–3 division/min under IPS vs 0.54×10^–3 division/min under AS; t-test, p=8.03 × 10^–5; Figure 3—figure supplement 1c) and 96 min (2.24×10^–3 division/min under IPS vs 1.17×10^–3 division/min under AS; t-test, p=6.80 × 10^–3; Figure 3—figure supplement 1d). Cells also displayed strikingly different temporal dynamics of division rates under IPS and AS conditions (Figure 3c and d). Under IPS condition, the division rate fluctuated largely over time: after the transition to 2% glucose, the division rate quickly increased to reach a plateau (4.86×10^–3 division/min on average during the half-period with 2% glucose); after the transition to 1 M sorbitol in the absence of glucose, cell division was greatly slowed down (1.34×10^–3 division/min during the half-period without glucose). In contrast, the division rate remained much more constant over time under AS condition: the average division rate was 1.69×10^–3 division/min during the half-period with 2% glucose and 1 M sorbitol and 1.89×10^–3 division/min during the half-period without glucose and sorbitol. Therefore, cells appear to use glucose more efficiently for growth under IPS than AS conditions. Collectively, these results further demonstrate that the timing of both glucose availability and osmotic stress matters: cells grow more slowly when facing periodic alternation of the two stresses (AS) than when facing periodic co-occurrence of these stresses (IPS).

Figure 3 with 1 supplement see all

Download asset Open asset

Figure 3—source data 1

Source data for Figure 3.

https://cdn.elifesciences.org/articles/88750/elife-88750-fig3-data1-v1.zip

Download elife-88750-fig3-data1-v1.zip

The slower cell division rate observed under AS when compared to IPS could be explained by the allocation of intracellular glucose to the osmotic stress response under AS when cells are exposed to glucose and sorbitol simultaneously, leaving less glucose available for growth. Indeed, in response to hyperosmotic stress glycerol is synthesized from a glycolysis intermediate (DHAP) derived from glucose (Norbeck et al., 1996). Under this hypothesis, glucose would only be fully allocated to growth in the absence of hyperosmotic stress, which occurred under IPS but not AS.

To test the hypothesis that the allocation of glucose toward glycerol synthesis explained the slower division rate observed under AS relative to IPS, we compared the division rate of mutants with impaired glycerol regulation under IPS and AS conditions. These mutant strains carried deletions of HOG1 (HOG pathway MAPK), PBS2 (MAPKK upstream of Hog1p), STE11 (MAPKKK upstream of Pbs2p), FPS1 (aquaglyceroporin regulated by Hog1p), GPD1 (glycerol-3-phosphate dehydrogenase regulated transcriptionally and post-transcriptionally by the HOG pathway) or GPD2 (paralog of GPD1). As expected, these mutants showed no growth defect in the absence of hyperosmotic stress and most mutants showed decreased division rates when exposed to constant hyperosmotic stress (Figure 2—figure supplement 3). At a fluctuation period of 24 min, the division rate was significantly lower under AS than IPS for almost all mutants (hog1Δ, pbs2Δ, gpd1Δ, gpd2Δ, gpd1Δ; gpd2Δ and fps1Δ) both with fluctuations of 2% glucose (Figure 4a) and 0.1% glucose (Figure 4—figure supplement 1a). However, the ste11Δ mutant exhibited similar division rates under AS and IPS (Figure 4a). At a fluctuation period of 96 min, the division rates of the two mutants we tested, ste11Δ and pbs2Δ, were significantly lower under AS than IPS (Figure 4—figure supplement 1b and c). In addition, the temporal dynamics of division rates were similar for the wild-type strain, pbs2Δ mutant (Figure 4b and c) and ste11Δ mutant (Figure 4—figure supplement 1d and e). In conclusion, mutations known to reduce intracellular accumulation of glycerol did not attenuate the growth differences that we observed in the wild-type strain between IPS and AS conditions. Therefore, allocation of glucose toward glycerol synthesis during hyperosmotic stress is not responsible for the lower division rate observed under AS than IPS.

Figure 4 with 2 supplements see all

Download asset Open asset

Figure 4—source data 1

Source data for Figure 4.

https://cdn.elifesciences.org/articles/88750/elife-88750-fig4-data1-v1.zip

Download elife-88750-fig4-data1-v1.zip

We next tested alternative hypotheses to understand why cells grew slower under AS than IPS condition. Glucose starvation was previously shown to induce fast inhibition of transcription (Jona et al., 2000) and translation initiation (Ashe et al., 2000; Joyner et al., 2016), leading to cell growth reduction. This phenomenon may explain the slower division rate observed in AS condition than in IPS condition, because rapid arrest of the cell cycle after glucose starvation could have smaller impact on global division rate when occurring concurrently (IPS) rather than alternatively (AS) with hyperosmotic stress that also leads to fast growth reduction. We tested this hypothesis by growing wild-type cells under IPS and AS conditions with galactose instead of glucose as a carbon source, because transcriptional and translational inhibition was not observed after galactose starvation in previous studies (Ashe et al., 2000; Jona et al., 2000). We observed a significant reduction of division rate under AS condition relative to IPS condition when using galactose as a carbon source, similar to what we observed in glucose (Figure 4—figure supplement 2). Therefore, fast inhibition of transcription and translation occurring after glucose starvation but not after galactose starvation does not contribute significantly to the slower growth in AS condition.

Yeast cells accumulate carbohydrate reserves such as glycogen to cope with nutrient starvation (Wilson et al., 2010; François and Parrou, 2001), and this reserve of glycogen can be mobilized during hyperosmotic stress (Bonny et al., 2021; Parrou et al., 1997). Glycogen may accumulate less under AS condition because glucose is only available during hyperosmotic stress, leading to slower growth. To test the hypothesis that glycogen storage may contribute to the difference of division rates observed between AS and IPS conditions, we quantified the division rates of glc3Δ mutant cells with impaired glycogen synthesis in these two conditions. Once again, we observed a significantly lower division rate of glc3Δ cells in AS condition relative to IPS condition similar to what was observed in wild-type cells, suggesting that glycogen storage was not significantly contributing to this difference.

Third, we tested the impact of point mutations in the cyclin inhibitor Sic1p on division rates in AS and IPS conditions. In response to hyperosmotic shock, residue 173 of Sic1p is phosphorylated by Hog1p, resulting in Sic1p stabilization and cell cycle arrest at the G1 phase (Escoté et al., 2004). Since hyperosmotic stress and glucose starvation both lead to cell growth arrest, cell division is expected to halt twice more frequently when hyperosmotic stress and glucose starvation are applied alternatively than when they are applied simultaneously, which could lead to the difference of division rates observed between AS and IPS conditions in a way that depends on Sic1p regulation. However, sic1(T173A) mutant cells (unphosphorylatable Sic1p) and sic1(T172E) mutant cells (constitutive Sic1p stabilization) showed a similar decrease of division rate in AS condition relative to IPS condition as observed in wild-type cells (Figure 4—figure supplement 2). The mechanism(s) responsible for the lower division rate in AS condition relative to IPS condition therefore remain(s) elusive.

We noticed a high proportion of wild-type cells dying under AS (Figure 5a): some cells suddenly burst with their nucleus staying in the growth chamber, others became opaque and stopped growing with their nucleus remaining completely still (the nucleus of living cells wobbled over time). These death events mostly occurred within minutes of the transition from medium containing 2% glucose and 1 M sorbitol to medium without glucose and sorbitol (Figure 5—figure supplement 1d), suggesting cell lysis occurred due to hypo-osmotic shock following removal of 1 M sorbitol. Cell death was less frequent under IPS than AS conditions for the wild-type strain (Figure 5a–c), even though the frequency of hypo-osmotic shock was the same in the two conditions (Figure 5—figure supplement 1c and d). We reasoned this could be due to lower intracellular accumulation of glycerol under IPS, when hyperosmotic stress is applied in the absence of glucose. Under AS, the presence of glucose during hyperosmotic stress could lead to faster intracellular accumulation of glycerol, resulting in stronger hypo-osmotic shock and cell lysis when the sorbitol concentration suddenly drops. Several pieces of evidence support this hypothesis. First, the rate of cell death should be reduced in mutants with lower glycerol synthesis. Indeed, we observed significantly lower rates of cell death for all mutants tested (ste11Δ, hog1Δ, pbs2Δ, gpd1Δ, gpd2Δ and gpd1Δ; gpd2Δ) relative to the wild-type strain under AS, but not under IPS for which glycerol synthesis is frustrated even in wild-type cells due to the absence of glucose during hyperosmotic stress (Figure 5c). In particular, the death rate decreased from 2.4×10^–3 min^–1 for wild-type to 5.2×10^–5, 3.4×10^–4, and 3.5×10^–4 min^–1, respectively, for the hog1Δ, pbs2Δ, and gpd1Δ; gpd2Δ mutants under AS when the fluctuation period was 24 min. A similar pattern was observed for the pbs2Δ mutant when the fluctuation period was 96 min (Figure 5—figure supplement 1a), although the reduction in the death rate was less pronounced than for the period of 24 min. Conversely, we observed a higher death rate for the fps1Δ mutant (3.5×10^–3 min^–1) under AS (Figure 5c), which is consistent with higher intracellular accumulation of glycerol in this mutant lacking the Fps1 aquaglyceroporin channel involved in glycerol export. Over a 10 hr AS experiment with a fluctuation period of 24 min, the death rate was lowest at the beginning of the experiment and was maximal during the last 5 hr of the experiment (Figure 5—figure supplement 1e). However, when the AS fluctuation period was 96 min, the maximum death rate occurred earlier (and stopped increasing after the second period) and the dynamics of cell death remained constant over multiple periods of osmotic fluctuation (Figure 5—figure supplement 1f). Again, these observations are consistent with cell death being due to glycerol accumulation, since it takes time for cells to accumulate an amount of glycerol sufficient to cause bursting after hypo-osmotic shock.

Figure 5 with 2 supplements see all

Download asset Open asset

Figure 5—source data 1

Source data for Figure 5.

https://cdn.elifesciences.org/articles/88750/elife-88750-fig5-data1-v1.zip

Download elife-88750-fig5-data1-v1.zip

Bonny et al., 2021, showed that sic1 mutants (sic1Δ and sic1(T173A)) could adapt faster than wild-type cells to a hyperosmotic shock at the expense of increased cell death under repeated osmotic stresses. Consistent with their finding, we observed higher death rate of sic1(T173A) and sic1(T173E) mutant cells during repeated exposure to 1 M sorbitol at a period of 24 min in constant 2% glucose (Figure 5—figure supplement 2a). Surprisingly, we did not observe an increased death rate of these mutants under AS and IPS conditions (Figure 5—figure supplement 2b), when both hyperosmotic stress and glucose availability fluctuated periodically over time. In fact, under AS condition, the death rate of sic1(T173A) cells was even lower than the death rate of wild-type cells. Under this condition, the particularly low division rate of sic1(T173A) cells may lead to strengthened cell wall, decreasing the probability of cell bursting after hypo-osmotic shocks.

The rates of cell division and cell death both contribute to fitness (i.e. the adaptive value) of a genotype in a particular environment. Since HOG pathway mutants exhibited different cell division and death rates compared to the wild-type genotype under AS, we calculated the population growth rate (division rate minus death rate) as a fitness estimate. Under IPS, the population growth rates of most mutant strains and of the wild-type strain were not significantly different; the only exception being the slightly lower growth rate of the ste11Δ mutant (Figure 5d). However, under AS with a fluctuation period of 24 min, several mutants had higher population growth rates than the wild-type strain (Figure 5d). In fact, the population growth rate was negative for the wild-type strain (–1.4×10^–3 min^–1) as cells died faster than they divided and positive for the hog1Δ mutant (1.0×10^–3 min^–1) and pbs2Δ mutant (5.5×10^–4 min^–1). These differences are clear in the microscopy images, as the population of wild-type cells visually shrank over time under AS (Figure 5a), while the population of pbs2Δ cells clearly expanded (Figure 5b). Thus, the hog1Δ and pbs2Δ genotypes are better adapted and would quickly outcompete wild-type cells under these dynamic conditions. However, this is only true when the frequency of environmental fluctuations is sufficiently high, since we did not observe significant differences in the population growth rate between the wild-type and pbs2Δ mutant under AS when the fluctuation period was 96 min (Figure 5—figure supplement 1b). We conclude that mutants that were first characterized by an inability to adapt to prolonged hyperosmotic stress can be well adapted when hyperosmotic stress rapidly fluctuates in antiphase with glucose availability. Therefore, the genetic mechanisms that contribute to adaptation under steady-state conditions could be detrimental under dynamic conditions, highlighting the importance of investigating how organisms adapt to dynamically changing environments.

The ability of cells to sense environmental fluctuations and to execute an adaptive response has been mainly studied using fluctuations of one stress cue at a time. How cells sense and respond to dual fluctuations of two interacting stresses remains a fundamental open question to understand how cells cope with complex environmental dynamics. Our findings suggest that the cell response to dual stress fluctuations can be very different depending on the phasing of the two stresses. Indeed, yeast cells appear to accumulate more glycerol under AS than under IPS. This could be either because of an impaired ability of cells to sense hyperosmotic shocks in absence of glucose or because of an impaired capacity to respond to hyperosmotic shocks in absence of glucose. Glycerol synthesis is regulated by the HOG pathway; thus, we investigated whether the activity of this pathway differed under IPS and AS conditions.

In the presence of glucose, activation of the HOG pathway in response to hyperosmotic shock triggers phosphorylation and nuclear translocation of Hog1p, which regulates transcription of multiple genes. A negative feedback loop dephosphorylates Hog1p, which leaves the nucleus in less than 15 min even if hyperosmotic stress is maintained (Hersen et al., 2008; Muzzey et al., 2009). To track HOG pathway activity, we quantified the nuclear enrichment of Hog1p over time by monitoring the subcellular location of Hog1 protein fused to a fluorescent marker (Hog1-GFP) in cells that also expressed the nuclear marker Htb2-mCherry. We did not detect any enrichment of Hog1-GFP in nuclei when only glucose fluctuated over time (Figure 6—figure supplement 1a), suggesting cells did not perceive glucose fluctuations as a significant osmotic cue. Although previous studies observed small transient (<2 min) peaks of Hog1-GFP nuclear localization after glucose was added back to the medium following glucose depletion (Sharifian et al., 2015; Piao et al., 2012), the temporal resolution in our experiments (one image every 6 min) may have been too low to detect these peaks.

In contrast, enrichment of Hog1-GFP fluorescence in nuclei was observed within minutes after exposure to 1 M sorbitol under both AS (Figure 6a) and IPS (Figure 6b). Therefore, cells can sense hyperosmotic shock, activate the HOG MAPK cascade, and phosphorylate Hog1 MAP kinase both in the presence (AS) and absence (IPS) of glucose. However, the adaptation dynamics of Hog1p (i.e. its exit from the nucleus) were remarkably different (Figure 6a–c): under AS, nuclear enrichment of Hog1-GFP peaked at 6 min following hyperosmotic shock and then quickly decayed and became undetectable after 30 min (Figure 6a and c)—essentially the same dynamics observed under periodic fluctuations of osmotic stress without glucose fluctuations (Figure 6—figure supplement 1a). Under IPS conditions, nuclear enrichment also peaked 6 min after hyperosmotic shock, but Hog1-GFP returned to the cytosol much more slowly; strong nuclear enrichment was still observed 48 min after exposure to hyperosmotic stress in the absence of glucose (Figure 6b and c). When the hyperosmotic stress was released, Hog1-GFP returned to the cytosol in less than 12 min.

Figure 6 with 1 supplement see all

Download asset Open asset

Figure 6—source data 1

Source data for Figure 6.

https://cdn.elifesciences.org/articles/88750/elife-88750-fig6-data1-v1.zip

Download elife-88750-fig6-data1-v1.zip

To determine whether Hog1-GFP eventually returns to the cytoplasm during hyperosmotic stress in the absence of glucose, we applied a single pulse of 1 M sorbitol without glucose for 4 hr. Nuclear enrichment of Hog1-GFP reached basal levels about 2 hr after the onset of hyperosmotic stress (Figure 6—figure supplement 1b). The delayed exit of Hog1-GFP out of the nucleus under IPS suggests that the activity of the feedback loop regulating Hog1p dephosphorylation is impaired in the absence of glucose. We hypothesized that delayed nuclear export of Hog1-GFP under IPS could be due to impaired osmoregulation. In support of this hypothesis, we observed no recovery of cell size during hyperosmotic stress under IPS conditions (Figure 6d). Cells only returned to their initial size when sorbitol was removed and glucose was added back to the medium, which also corresponded to the moment when Hog1-GFP returned to the cytosol. In contrast, under AS, both the recovery of cell size and nuclear export of Hog1-GFP occurred while cells were still exposed to hyperosmotic stress (Figure 6), showing that osmoregulation was not impaired under these conditions. These results suggest glucose is necessary for the rapid osmoregulation that usually occurs in the first 20 min following hyperosmotic stress. This fast osmoregulation has been proposed to rely on the induction of glycerol synthesis via Hog1p-dependent post-translational mechanisms (Schaber et al., 2012). Since glucose is a metabolic precursor of glycerol, the absence of glucose may prevent glycerol synthesis and thereby fast osmoregulation. Further work will be necessary to test this hypothesis and study how glucose stored in the cell is used (or not) for glycerol production.

Temporal fluctuations of two different stresses may have different or even opposite effects on gene expression, raising the question of how dual fluctuations of these two stresses would affect gene expression. Fast periodic fluctuation of osmotic stress was previously shown to cause hyper-activation of the STL1 promoter (P_STL1) regulated by Hog1p (Mitchell et al., 2015), while glucose depletion is known to inhibit transcription (Jona et al., 2000) and translation initiation (Ashe et al., 2000; Janapala et al., 2019). We therefore asked whether dual fluctuations of glucose depletion and osmotic stress had additive effects on P_STL1 expression or whether one of the dynamic cues had a dominant effect. To address this question, we quantified the expression dynamics of a P_STL1-mCitrine fluorescent reporter gene under IPS and AS conditions. As expected, we observed transient expression of P_STL1-mCitrine in response to both short (48 min) and long (10 hr) pulses of 1 M sorbitol (Figure 7): the mean fluorescence level peaked at a 2.8-fold change 115 min after exposure to a short sorbitol pulse and 2.4-fold change 122 min after a long sorbitol pulse; then, fluorescence gradually returned to basal levels, even when osmotic stress was maintained. This expression pattern is consistent with previous quantifications of P_STL1 transcriptional activity during hyperosmotic stress (Wosika and Pelet, 2020; Ben Meriem et al., 2019). When hyperosmotic shocks were periodically applied for 48 min every 96 min, P_STL1-mCitrine expression constantly increased to reach 10.2-fold change after 11 hr (Figure 7b), suggesting a lack of adaptation to repeated osmotic shocks. Interestingly, hyper-activation of P_STL1-mCitrine transcription was also observed under AS conditions where P_STL1-mCitrine expression reached a maximum of 4.1-fold change after 413 min and remained as high as 3.1-fold change after 11 hr under the AS regime with a fluctuation period of 96 min (Figure 7a and b).

Figure 7

Download asset Open asset

Figure 7—source data 1

Source data for Figure 7.

https://cdn.elifesciences.org/articles/88750/elife-88750-fig7-data1-v1.zip

Download elife-88750-fig7-data1-v1.zip

Conversely, we observed much weaker and slower induction of P_STL1-mCitrine expression under IPS: the maximal fold change was 0.7 after 11 hr of IPS with a fluctuation period of 96 min (Figure 7a and b). We observed a similar pattern (i.e. faster, stronger induction of P_STL1-mCitrine under AS than IPS) when the fluctuation period was 24 min (Figure 7c). Therefore, the STL1 promoter is not activated by hyperosmotic stress in the absence of glucose, despite nuclear translocation of the MAP kinase Hog1p. This result suggests that the global repression of expression in response to abrupt glucose starvation is dominant over the hyper-activation of P_STL1 transcriptional activity induced by periodic osmotic stress.

All S. cerevisiae strains used in this study are derived from BY4741 or BY4742 (Brachmann et al., 1998) and are listed in Tables 1 and 2. The reference strain yPH_132 expresses the mCherry fluorescent reporter fused to histone H2B to label cell nuclei. To obtain this strain, the mCherry_pAgTEF-KanMX4-tAgTEF linear DNA fragment was amplified from plasmid pYM35 (PCR_TOOLBOX collection from EUROSCARF). This DNA fragment was inserted at the HTB2 locus in strain BY4741 using a classic LiAc transformation protocol (Gietz and Woods, 2002) and a clone resistant to G418 was stored at –80°C as strain yPH_132.

All gene-deletion strains of yeast used in this study were obtained by directed mutagenesis of the yPH_132 strain (HTB2::HTB2-mCherry). Gene deletions were performed using CRISPR/Cas9 gene editing according to the method described in Laughery et al., 2015, that relies on the co-transformation of (i) a plasmid vector derived from pML104 (Addgene #67638) allowing expression of the Cas9 protein and a guide RNA specific to the target gene and (ii) a DNA repair fragment designed to delete the coding sequence of the target gene via homology-directed repair. For each gene deletion, a sequence containing the 20 nucleotides of the sgRNA recognition site in the coding sequence of the target gene was cloned between BclI and SwaI restriction sites in the pML104 plasmid by ligation of hybridized oligonucleotides. The DNA repair fragment was obtained by PCR amplification (Phusion Hot Start Flex 2X Master Mix, NEB M0536S) of two oligonucleotides with 20 bases of reverse complementarity to each other at the 3’ end and 70 bases homologous to the region either immediately upstream or downstream of the coding sequence of the target gene at the 5’ end. The repair fragment and the CRISPR/Cas9 plasmid specific to each gene were transformed together in exponentially growing yPH_132 cells using a standard lithium acetate method. Transformants were isolated on CSM-ura (Formedium, DCS0271) agar plates and gene deletions were confirmed by PCR screening and by Sanger sequencing. Positive clones were grown on YPG plates (10 g/L yeast extract, 20 g/L peptone, 5% [vol/vol] glycerol, 20 g/L bacto agar) to counter-select petite cells (p-phenotype) and then transferred on Complete Supplement Mixture (CSM) agar plates containing 0.8 g/L 5-fluoroorotic acid (Thermo Scientific R0812) to counter-select the CRISPR/Cas9 plasmid carrying the Ura3 gene. Two independent clones were stored at –80°C in 15% glycerol for each gene deletion (the clones used in this study are listed in Table 1).

Yeast cells were cultivated and imaged in custom-made microfluidic devices for all time-lapse microscopy experiments described in this study. One day before each experiment a new microfluidic chip was made by casting a mixture of 10 g of polydimethylsiloxane (Sylgard 184 kit, Neyco) and 1 g of curing agent on a master wafer made by soft lithography. The chip was then degased, cured at 65°C for 4 hr, peeled off, punched with a 1.2-mm-diameter needle at all positions of inlets and outlets and then bonded onto a 24×60 mm² coverslip after plasma activation of the surfaces. The chip pattern is shown in Figure 1—figure supplement 1a. In brief, it consists of five independent pairs of flow channels (800 µm wide×50 µm high) connected each to five growth chambers (400×400×3.8 µm³; L×W×H) where yeast cells are constrained to proliferate in monolayer.

Yeast strains were thawed from glycerol stocks kept at –80°C onto YPG agar plates (10 g/L yeast extract, 20 g/L peptone, 5% [vol/vol] glycerol, 20 g/L bacto agar) at least 3 days and no more than 2 weeks before each microscopy experiment. After 2 days of incubation at 30°C, YPG plates were kept at room temperature. The day before an experiment, a small amount of cells (~10⁵ cells) was inoculated in 4 mL of CSM medium (6.7 g/L Yeast Nitrogen Base without amino acids [BD Difco], 2% glucose [Euromedex], 0.8 g/L CSM of amino acids [MP Biomedicals]) and incubated overnight at 30°C with 250 rpm orbital shaking. 100 µL of cell culture was then transferred to 5 mL of CSM medium and incubated for another 4 hr at 30°C to reach the exponential phase of growth. Cells were loaded in the microfluidic chip by injecting 50 µL of cell culture through each inlet of the chip using a small syringe.

Next, each pair of inlets was connected to a three-way solenoid valve (the Lee Company, LFAA1201418H) that was itself connected to two bottles of medium via tubings with 0.5 mm inner diameter and 1.5 mm outer diameter (Cole-Parmer Microbore Tubing). A custom-made valve controller piloted by a Node-RED application was used to dynamically control which medium was dispensed to the cells based on a predetermined schedule of valve state switching. Media were sterilized by filtration (0.22 µm) to remove large particles and their composition varied depending on the experiment. All media were based on 6.7 g/L Yeast Nitrogen Base without amino acids (BD Difco) and 0.8 g/L CSM of amino acids (MP Biomedicals) in ddH₂O. This base was complemented either with 2% glucose, 0.1% glucose, or no glucose and with 1 M sorbitol or no sorbitol. The outlets of the microfluidic chip were connected via tubings to a peristaltic pump (Ismatec IPC 12) and to an empty beaker to collect the flow-through.

The chip was then mounted on a motorized inverted microscope (Olympus IX83) equipped with LEDs for fluorescence excitation (CoolLED pE-300ultra), a Zyla 4.2 sCMOS camera (Andor—Oxford Instruments) and an autofocus module (IX3-ZDC2, Olympus). The microscope, the microfluidic system, and culture media were all placed inside an incubation chamber maintained at 30°C. Immediately after mounting the chip, the pump was turned on with a flow rate set at 120 µL/min for each outlet tube, resulting in a flow rate of 240 µL/min in the growth chambers since they were each connected to two outlets. The medium dispensed to cells by default was CSM with either 2% glucose or 0.1% glucose. The microscope was controlled using iQ v3.6.3 software (Andor Technology) and all images were obtained using a 60× oil immersion objective (Olympus PlanApo N 60×/1.42) and a ×1.6 magnification changer. Using live bright-field imaging, we selected 25 positions (25 fields of view) of the motorized stage (Prior Scientific ProScan III) that captured 10–50 cells in each of the 25 growth chambers of the chip and were focused slightly below the median cell plane based on cell wall contrast. We used an iQ program that automatically scanned each position every 6 min for 12 or 24 hr and acquired bright-field and fluorescence images after autofocus adjustment. In parallel, we executed the program controlling the timing of valve state switches. This program always started with 1 hr of the default state corresponding to CSM+2% glucose or CSM+0.1% glucose so that cells could acclimate to the experimental settings and images recorded in the first hour were excluded from analyses. To detect YFP fluorescence, samples were exposed to blue LED illumination at an intensity of 10% using a 514/10 nm excitation filter and fluorescence was acquired with an exposure time of 250 ms using a 545/40 nm emission filter. To detect mCherry fluorescence, samples were exposed to green LED illumination at an intensity of 14% using a 560/40 nm excitation filter and fluorescence was acquired with an exposure time of 150 ms using a 630/75 nm emission filter. Microscopy images were saved in TIFF format and are available upon request.

Microscopy images were acquired following the same procedure for experiments aiming at quantifying division and death rates, for experiments aiming at quantifying the fluorescence of cells expressing P_STL1-mCitrine and for experiments aiming at quantifying the nuclear enrichment of Hog1-GFP fusion protein. However, image analysis was performed differently for each type of experiment as described below.

We used ilastik v1.3 for the segmentation and tracking of cell nuclei expressing HTB2-mCherry on fluorescence images. Three time-lapse movies of 120 fluorescence images of yPH_132 cells grown in CSM+2% glucose medium were used to train the machine learning algorithms to recognize the nuclei of single cells and the nuclei of dividing cells at various numbers of cells per image. The segmentation and tracking procedure was then applied to frames from all experiments after manually removing out-of-focus images that occasionally occurred due to autofocus failure. The output for each experiment was a CSV file for each of the 25 field of views where rows corresponded to all objects (i.e. cell nuclei) detected on all images and columns corresponded to parameters such as image identity, cell nucleus identity, parental nucleus identity, size of the nucleus, and XY coordinates of nucleus centroid. For all experiments involving combined stresses and AS (Figures 3—6), we manually added to CSV files a parameter indicating when cell death was observed based on visual inspection of bright-field images. This manual step was necessary because the nucleus of a dead cell could remain fluorescent and could be tracked for several hours after cell death. This time-consuming step was not performed in the analyses of other experiments where death events remained very rare (results shown in Figure 2). Next, we computed cell division rate and death rate using homemade R scripts. We designed and implemented an Eulerian measure of division rate that was robust to rare tracking errors and to cell saturation in the field of view. In this approach, we first define a tracking window of 1928×1928 pixels centered on each image of 2048×2048 pixels. We can then establish the relation:

where $N_{windo{w}_{\left(t\right)}}$ and $N_{windo{w}_{\left(t+1\right)}}$ are the number of cell nuclei in the tracking window at frames $t$ and $t+1$, respectively, $N_{ne{w}_{\left(t\right)}}$ is the number of divisions that occurred in the window between $t$ and $t+1$, $N_{i{n}_{\left(t\right)}}$ is the number of cell nuclei that entered the window, and $N_{ou{t}_{\left(t\right)}}$ is the number of cell nuclei that exited the window between $t$ and $t+1$. From this relation, we computed $\frac{N_{ne{w}_{\left(t\right)}}}{N_{w_{\left(t\right)}}}$ which is the number of division events relative to the total number of cell nuclei in the window between $t$ and $t+1$. The slope of the linear regression of the cumulative sum ${\sum }_{t=t_0}^{t_1}\frac{N_{ne{w}_{\left(t\right)}}}{N_{windo{w}_{\left(t\right)}}}$ over time corresponded to the average division rate between $t_0$ and $t_1$ . When death events were added to the input file, we only counted the nuclei of living cells when calculating the division rate. We excluded frames at the end of experiments that corresponded to an incomplete period of environmental fluctuation. The average death rate was calculated using a similar approach: it corresponded to the slope of the linear regression of the cumulative sum ${\sum }_{t=t_0}^{t_1}\frac{N_{deat{h}_{\left(t\right)}}}{N_{windo{w}_{\left(t\right)}}}$ over time where $N_{deat{h}_{\left(t\right)}}$ was the number of death events observed in the tracking window between $t$ and $t+1$.

We used a custom image analysis approach based on the segmentation and tracking of cells in bright-field images to quantify P_STL1-mCitrine expression in single cells (yPH_091 strain) over time. First, a preprocessing step was necessary to eliminate out-of-focus images that occasionally occurred. This was done using a U-NET model trained to estimate the area of cells on each frame and to detect sudden changes of cell area between consecutive frames indicative of autofocus failure. Next, a proper cell segmentation was achieved via a machine learning algorithm based on the StarDist method (GitHub, https://stardist.net/) (StarDist, 2021; Schmidt et al., 2018; Weigert et al., 2020) version 0.7.3 that uses star-convex shape prior. This method was well suited to the round shape of yeast cells and performed slightly better than the U-NET network allowing more reliable tracking of cells. Training of StarDist and U-NET algorithms was performed using a GPU NVIDIA RTX Quadro. For each cell in a frame $F_t$, the tracking was performed by detecting which cell in the previous frame $F_{t-1}$ was closest to the cell on frame $F_t$ based on Euclidean distances between cell centroids. This simple method worked best for cells that did not move too much between consecutive frames: it could fail when the distance between two different cells in two consecutive frames was smaller than the distance between the same cell in the two frames. For this reason, we manually selected by visual inspection cells with correct tracking over at least 6 hr and excluded cells with wrong tracking in further analyses. We computed the mean fluorescence of each cell as the total fluorescence of the cell divided by the area of the cell. The area corresponded to the number of pixels classified as belonging to the cell in the segmented bright-field image. The total fluorescence of a cell was the sum of intensities of all pixels classified as belonging to the cell in the fluorescence image. Pixel classification of fluorescence images was performed using a numpy.array function in Python that applied the segmentation masks obtained from the analysis of bright-field images to the corresponding fluorescence images. R scripts were used to plot the mean fold change of fluorescence over time for all cells analyzed in images that were taken at the same time point in different growth chambers sharing the same regime of environmental fluctuations. Fold change was calculated for each cell as the difference of mean fluorescence observed for that cell in a given frame and the mean fluorescence observed among all cells in the first frame divided by the mean fluorescence among all cells in the first frame.

The nuclear enrichment of Hog1-GFP was quantified in cells expressing both the Hog1-GFP reporter and the nuclear marker Htb2-mCherry. Cell segmentation and tracking was performed on bright-field images following the same procedure as described above for the quantification of P_STL1-mCitrine expression. In addition, another segmentation was done for cell nuclei that were detected with the red fluorescence channel using a simple thresholding step. Each nucleus contour was then associated by contour overlapping comparison to its corresponding cell contour obtained by segmentation of bright-field images. We then computed the mean fluorescence in the green channel for each cell and for each nucleus. The mean fluorescence of a cell was calculated as the total intensity of all pixels classified as belonging to the cell (including the nucleus) divided by the number of these pixels. The mean fluorescence of a nucleus was calculated as the total intensity of all pixels classified as belonging to the nucleus divided by the number of these pixels. Finally, the nuclear enrichment of fluorescence was calculated for each cell as the mean fluorescence of the nucleus divided by the mean fluorescence of the cell. R scripts were used to plot the nuclear enrichment of Hog1-GFP fluorescence over time for all cells analyzed in images that were taken at the same time point in different growth chambers sharing the same regime of environmental fluctuations.

We used a fluorescein assay to characterize the temporal dynamics of medium fluctuations inside microfluidic chips. In this assay, we connected a microfluidic chip to CSM medium complemented with 50 nM fluorescein and to CSM medium without fluorescein. We programmed the valve to dispense CSM with fluorescein to the chip for 20 min followed by CSM without fluorescein for another 20 min and repeated this treatment twice. We tried different flow rates on the peristaltic pump but only showed results for the optimal flow rate of 120 µL/min. We used the same microscopy setup as described above to image the growth chamber at the center of the chip, except that a 20× objective (Olympus Plan Achromat) was used to be able to visualize both the flow channel and the growth chamber in the field of view. One bright-field image and one fluorescence image were taken every 12 s. The fluorescence channel consisted of blue LED illumination at an intensity of 10% with a 514/10 nm excitation filter and acquisition with an exposure time of 250 ms using a 545/40 nm emission filter. We used ImageJ to quantify the mean fluorescence in two circular regions with a diameter of 280 pixels. One region was in the center of the growth chamber and the other region in the flow channel. A script in R was used to plot the relative level of fluorescence over time in each region. The relative fluorescence at time $t$ ($R{F}_t$) was calculated as $R{F}_t=\frac{F_t-F_{min}}{F_{max}-F_{min}}$, where $F_t$ is the mean fluorescence at time $t$, $F_{min}$ is the mean fluorescence observed between 30 and 40 min and between 70 and 80 min when fluorescein was at its minimal concentration in the chip and $F_{max}$ is the mean fluorescence observed between 10 and 20 min and between 50 and 60 min when fluorescein was at its maximal concentration in the chip.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

The authors would like to thank their team members for their critical reading of this manuscript. We also thank Williams Brett who helped us design the microfluidic control system. This work was supported by the European Research Council grant SmartCells (724813) and received support from grants ANR-11-LABX-0038, ANR-10-IDEX-0001-02, and ANR-16-CE12-0025-01.

1. Preprint posted: February 17, 2023
2. Sent for peer review: May 18, 2023
3. Reviewed Preprint version 1: July 26, 2023
4. Reviewed Preprint version 2: February 28, 2024
5. Version of Record published: April 3, 2024

You can cite all versions using the DOI https://doi.org/10.7554/eLife.88750. This DOI represents all versions, and will always resolve to the latest one.

© 2023, Duveau et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.