OUP user menu

Stress adaptation of Saccharomyces cerevisiae as monitored via metabolites using two-dimensional NMR spectroscopy

Woo Young Kang, Seol Hyun Kim, Young Kee Chae
DOI: http://dx.doi.org/10.1111/j.1567-1364.2012.00811.x 608-616 First published online: 1 August 2012

Abstract

Many studies on yeast metabolism are focused on its response to specific stress conditions because the results can be extended to the human medical issues. Most of those works have been accomplished through functional genomics studies. However, these changes may not show a linear correlation with protein or metabolite levels. For many organisms including yeast, the number of metabolites is far fewer than that of genes or gene products. Thus, metabolic profiling can provide a simpler yet efficient snapshot of the system's physiology. Metabolites of Saccharomyces cerevisiae under various stresses were analyzed and compared with those under the normal, unstressed growth conditions by two-dimensional NMR spectroscopy. At least 31 metabolites were identified for most of the samples. The levels of many identified metabolites showed significant increase or decrease depending on the nature of the stress. The statistical analysis produced a holistic view: different stresses were clustered and isolated from one another with the exception of high pH, heat, and oxidative stresses. This work could provide a link between the metabolite profiles and mRNA or protein profiles under representative and well-studied stress conditions.

Keywords
  • NMR
  • yeast
  • stress
  • metabolite profiling

Introduction

Saccharomyces cerevisiae has been one of the most extensively studied model eukaryotes for fundamental and applied studies because its basic cellular processes are similar to those of higher eukaryotic organisms including humans (Ostergaard et al., 2000; Devantier et al., 2005). From a commercial point of view, yeast is a valuable organism for the production of ethanol, which can be used as food or fuel (Peralta-Yahya & Keasling, 2010; Sicard & Legras, 2011). Among several important and interesting studies, those on yeast metabolism are focused on its response to specific stress conditions because the results can be extended to other higher eukaryotes. For example, oxidative stress is thought to be related to aging, apoptosis, cancer, or immunological responses (Valko et al., 2005).

Yeast cells express a common set of genes to protect themselves during stressful periods. This phenomenon is called the environmental stress response (ESR), which includes activation or suppression of around 900 genes (Gasch, 2003). In addition to the general physiological response, the expressions of those genes are precisely coordinated to protect the cells against the specific nature of the stress (Gasch, 2003). The response and adaptation of yeast to various stress conditions including ethanol (Alexandre et al., 2001), osmotic stress (Gasch et al., 2000), oxidative stress (Causton et al., 2001), extreme pHs (Causton et al., 2001), and heat shock (Boy-Marcotte et al., 1999; Gasch et al., 2000; Strassburg et al., 2010) have been well established. Recently, Berry and Gasch reported that any single stress could trigger the change of the gene expression not only to cope with the given stress but also to prepare for possible future stresses (Berry & Gasch, 2008).

Most of those works have been accomplished through functional genomics studies by observing the changes of gene expression patterns or mRNA levels. However, these changes may not show a linear correlation with protein or metabolite levels (Nicholson et al., 2004). In one report, there was little agreement between the gene expression and metabolite concentrations in the liver, urine, or plasma (Griffin et al., 2004). In another report, metabolic data were employed in the functional genomic strategy to reveal the phenotype of silent mutations (Raamsdonk et al., 2001). Recently, efforts have been made to integrate the ‘Omics’ technologies to describe organisms in a holistic and systematic way (Hannah et al., 2010; Saito & Matsuda, 2010).

Metabolomics is the ‘systematic study of the unique chemical fingerprints that specific cellular processes leave behind’, or more specifically, the study of their small-molecule metabolite profiles (Daviss, 2005). The metabolome represents the collection of all end products of the gene expression in a biological organism (Jordan et al., 2009). For many organisms including yeast, the number of metabolites is far fewer than that of genes or gene products (Raamsdonk et al., 2001). Thus, metabolic profiling can provide a simpler yet efficient snapshot of the system's physiology. Researchers around the world have been developing various metabolome databases to offer faster identification of interesting metabolites and their corresponding properties. These databases include HMDB (Wishart et al., 2007), MMCD (Cui et al., 2008), METLIN (Smith et al., 2005), Yeast Metabolome Database (YMDB) (Jewison et al., 2012), MDL (http://www.liu.se/hu/mdl/), and PRiMe (Akiyama et al., 2008) where YMDB is the specialized database for yeast.

One-dimensional1H nuclear magnetic resonance (NMR) spectroscopy has been used extensively as an analytical tool for identifying and quantifying small molecules (Lewis et al., 2007). When samples contain minimal peak overlap, 1D1H NMR can be employed because the peak intensity and its concentration maintain a linear relationship. Recently, high-throughput analysis of complex biological processes at the metabolic level by NMR has been receiving a great deal of attention. These studies, however, rely on 1D1H NMR and inevitably suffer from extensive peak overlap. As an alternative to the metabolomics by 1D NMR, a two-dimensional1H-13C HSQC (Heteronuclear Single Quantum Coherence) has been proposed, and this method showed a promising result on obtaining quantitative data of metabolite concentrations (Lewis et al., 2007). However, the semi-quantitative analysis like this report would suffice in many cases because the research was focused on differentiating responses from different stresses.

In this work, we applied several common stresses to yeast cultures to see whether we could classify those stresses in terms of the metabolite profiles (or specific markers) and whether the result could be integrated with currently available functional genomic data. This work could provide a link between the metabolite profiles and mRNA or protein profiles under representative and well-studied stress conditions.

Materials and methods

Yeast growth

DS10 (MATaGAL2 his3-11,15 leu2-3,112 lys1 lys2 Δtrp1 ura3-52), a wild-type strain of S. cerevisiae, was grown in YPD medium (1% yeast extract, 2% bactopeptone, and 2% glucose). As seed cultivation, yeast cells were grown overnight in 5 mL YPD at 25 °C. For main cultivation, the pre-cultivated cells were inoculated into 500 mL of fresh medium in 2.8 L baffled Fernbach flask at an initial OD600 of approximately 0.07 and cultivated for 24 h under each stress condition. Among the 500 mL cultures were seven representative stress conditions along with a control. Each growth condition had four replicates. The specific stress conditions were as follows: salt (1 M sodium chloride), osmotic (1.5 M sorbitol), ethanol (4% ethyl alcohol), oxidative (0.4 mM hydrogen peroxide), low pH (pH 4.5, adjusted with H3PO4), and high pH (pH 8.2, adjusted with Tris base) stresses. The cultures were harvested by centrifugation at 3380 g at 4 °C for 20 min. Cells were resuspended in 30 mL PBS buffer (10 mM sodium phosphate, 150 mM NaCl, pH 7.4) and centrifuged at 3380 g at 4 °C for 20 min. Resuspension and centrifugation were repeated twice. Cell pellets were stored at −70 °C. Frozen pellets were lyophilized later.

Metabolite extraction

To 1 g of the dried cell pellet in a 50 mL conical tube, 15 mL of boiling deionized water was added. The mixture was incubated at 121 °C for 20 min in an autoclave, then cooled on ice, and then centrifuged at 3380 g at 4 °C for 30 min. Fine debris was removed with 0.45 μm pore syringe filters. The resulting filtrate was then transferred to VIVASPIN 20 centrifugal filters (Sartorius Stedim, Bohemia, NY) and centrifuged at 3380 g at 4 °C until < 0.5 mL of solution remained in the upper compartment. The syringe filters and centrifugal filters were washed with 10 mL water three times before use. The final filtrate was frozen and lyophilized.

NMR sample preparation

The dried extract powder was weighed and dissolved in a HEPES buffer (5 mM HEPES, 0.2 mM DSS, 0.5 mM NaN3 in D2O) with the ratio of 35 μL per 1 mg of the extract except for low pH and high pH, with the ratio of 17.5, and for ethanol, with that of 140. The pH was adjusted to 7.40 ± 0.01.

NMR experiments and data processing

All experiments were performed at 298 K on Bruker Avance II 500 MHz. Sensitivity enhanced1H-13C HSQC spectra were collected with 112 scans, 256 increments with the echo-antiecho method. The spectral widths were 20 ppm for1H and 110 ppm for13C. The frequency offsets for1H and13C were set at 4.7 and 55 ppm, respectively. The DSS (4,4-dimethyl-4-silapentane-1-sulfonic acid) resonance was used to reference the chemical shift. The raw data were processed with Topspin 2.1 (Bruker, Germany). The resulting spectra were visualized and analyzed by rNMR (Lewis et al., 2009). The peaklist was sent to MMCD (http://mmcd.nmrfam.wisc.edu/) to find candidate metabolites. Peak intensity data were normalized by Microsoft Excel using one of the HEPES resonances as an internal standard. Multivariate analysis was performed with the R statistics software package (http://www.r-project.org).

Results and discussion

Sample preparation

The cultures grew until they reached the stationary phase (Supporting Information, Fig. S1). They were grown to stationary phase to observe the accumulated effect and to facilitate the equalization of the final growth state of each stress condition. However, as each given stress had its own severity, the final optical densities and dried masses of the cultures varied to some degree (Table 1). Some of this variation could be due to changes in morphology, as in the case of bacteria where osmotic stress changes cell shape and therefore alters optical density (Wood, 1999). The dried cell weight, the optical density, and the dried extract weight did not show clear correlation (Table 1). All samples were thought to have entered the stationary phase because there was no noticeable increase in the optical density around the time of harvest. The rich medium (YPD) was chosen as a growth medium because we tried to focus on the specific stresses such as NaCl or H2O2 and did not want cells to experience the nutrient-limiting stress in the minimal medium. We hoped that this background (the stationary phase and the rich medium) would give us a window through which we could observe the accumulated effect of only the specific stresses at the analysis stage because we had a control sample that was grown with the same condition except for the stress. We expected to be able to subtract the background effects that were common to every stress sample. The initial addition of 15 mL of boiling water was employed in an attempt to denature endogenous enzymes that might degrade or synthesize metabolites during the extraction procedure. The resulting mixture was heated in the autoclave machine, and only the small, soluble, and heat-resistant metabolites that came out of the cells were collected for NMR analysis. However, the heat-labile metabolites could have been disintegrated into simpler compounds, which was especially true for the phosphorylated metabolites. Nevertheless, this was a decision that we made to attain higher extraction yields and was also a compromise between sensitivity and resolution where we chose to achieve a higher total glucose level instead of each glucose-containing compound. Even with the hot extraction condition, we observed that only a fraction of the starting cells were lysed, and this was manifested in the ethanol stress where the number of cells was comparable to other cases, but the amount of extracted metabolites was much less than others. As shown in Table 1, most samples yielded more than 16 mg of extract. The buffer-to-extract ratio was chosen as (Lewis et al. (2007) suggested, but when the amount of the extract was not enough to make the NMR sample, we used the multiples of the suggested buffer volumes. This was most clearly evident with the ethanol stress sample, where the amount was as little as 1.8 mg. We observed similar results from four independent trials. We speculate that this low yield was not attributed to sample handling, but owing to the nature of the ethanol stress itself, which changed the cell membrane and prevented the metabolites from oozing out of the cells. This was reported to happen in E. coli where the fatty acid chain length was increased in response to ethanol stress (Dombek & Ingram, 1984). The intensities of resonances of such diluted samples were recalibrated during analysis according to the dilution factors. The average mass of the dried extract from 1 g of dried cells was 18.8 ± 3.3 mg. The yield of the final dried extract from the initial dried yeast cells was around 2%.

View this table:
1

Summary of cell growth, dried cell mass, and dried extract mass from eight different conditions

StressOptical density at 600 nm when harvestedTotal dried cell mass from 500 mL culture (g)Dried extract mass per g of dried cells (mg)
Control7.3 ± 0.30.74 ± 0.0422 ± 2.6
1 M NaCl4.3 ± 1.20.51 ± 0.0627 ± 1.6
1.5 M Sorbitol3.1 ± 0.10.25 ± 0.0118 ± 1.2
0.4 mM H2O2 8.1 ± 1.50.44 ± 0.0422 ± 2.6
37 °C5.0 ± 0.40.66 ± 0.0923 ± 2.9
4% Ethanol3.5 ± 0.80.41 ± 0.023.1 ± 1.2
pH 8.28.2 ± 0.20.49 ± 0.0127 ± 2.5
pH 4.54.1 ± 0.70.57 ± 0.0931 ± 2.7
  • The initial cell OD for inoculation was 7.3; the type of medium was YPD; the cultivation time was 24 h; and the number of duplication was four for each condition.

NMR data collection and processing

Each 2D HSQC experiment took about 14 h, which was attributed to the time share (only at night) of the NMR spectrometer. We believe that a 4-h experiment would have yielded a spectrum with acceptable sensitivity and resolution judging from the quality of the collected spectra. Compared with the more popular profiling method based on the 1D NMR data, these 2D NMR experiments require a much longer amount of time. For high-throughput analysis, the processing speed of 1D NMR would be the greatest advantage; however, if the accuracy and robustness are concerned, the 2D method can be considered as a potent alternative (Lewis et al., 2007; Motta et al., 2010; Martineau et al., 2011; Robinette et al., 2011).

NMR data analysis

Figure 1 shows the upfield portion of the NMR spectrum of a control sample along with the assignments of identified metabolites. Using rNMR and the MMCD, at least 31 metabolites were identified for all of the spectra by observing multiple peaks. As some of the samples were more diluted than others and the intensities of resonances had to be normalized carefully for proper semi-quantitative comparison, the internal standard HEPES acted as an excellent calibration reference to normalize intensity data; every NMR sample contained the same concentration (5 mM). The representative resonances of the identified metabolites were visualized, and their intensities were measured by the ROI feature of rNMR (Fig. 2). Figure 2 offers an ‘at-a-glance’ view of the metabolite changes. For example, methionine shows a stronger signal for the high pH stress than the control, while the maltose signal becomes weaker for the ethanol, sorbitol, and NaCl stresses. Those resonances in Fig. 2 were chosen because they were well isolated from other resonances (no overlap) and showed adequate intensities for a clear and accurate comparison. Figure 3 shows the changes in resonance intensities, and thus, concentrations of the metabolites under different stress conditions referenced to the normal condition. We can observe the ‘trends’ of the changes of metabolites upon the transition to the stress conditions. By putting cells under stresses in a general sense, we found that metabolites such as glycerol, methionine, trehalose, and valine increased while ADP, ethanolamine, glycine, lysine, putrescine, succinic acid, and UMP decreased. As a note, it was curious that we observed the phosphorylated compounds such as ADP and UMP even after the boiling extraction.

1

The upfield portion of the two-dimensional1H-13C HSQC spectrum of the control sample. The assigned resonances are labeled with the names of the corresponding metabolites.

2

ROI view of the representative resonances of identified metabolites. Each condition was repeated four times to be statistically meaningful. This view shows the raw data before the two normalization steps against HEPES and the concentration were applied.

3

Changes of resonance intensities referenced to the normal condition. Error bars were plotted according to the standard deviations of the peak intensities divided by the average of the normal condition. The bars belonging to one metabolite correspond to low pH, high pH, ethanol, heat, osmotic, oxidative, and salt stresses, from left to right.

From Figs 2 and 3, we found several interesting features comparable to previous reports about the metabolites that manifested themselves in the stressed environment. Glycerol and trehalose are widely known protectants in general stresses such as osmotic, oxidative, heat, ethanol, and low pH stress conditions (Hohmann, 2002; Devantier et al., 2005; Narendranath & Power, 2005; Shimizu et al., 2009; Hallsworth et al., 2010; Strassburg et al., 2010), while we found that trehalose was accumulated almost exclusively for the salt stress and that the glycerol level was more elevated for the osmotic stress than other cases. It was peculiar that trehalose was accumulated 100 times more in the salt stress condition than the simple osmotic stress, which should indicate trehalose was designed to protect cells against highly ionic solvents. The genes involved in the trehalose metabolism were reported to be induced by the saline stress (Posas et al., 2000), which is consistent with the increased intracellular trehalose concentration in our result. The trehalose signals from other conditions than the salt stress were weak but detectable. The reciprocal decrease in maltose under the salt stress was observed, which could be reasoned by the fact that maltose could be converted to trehalose by the trehalose synthase which was activated under osmotic stress in Pseudomonas syringae (Freeman et al., 2010). The signal from glycerol was strong in every case, which suggested its role as a housekeeping metabolite. We observed that betaine, another well-known osmoprotectant, increased under salt, osmotic, and high pH stresses. This finding is consistent with the report on a halotolerant cyanobacterium, which activated a betaine transporter at alkaline pH and high salt (Laloknam et al., 2006). This would be an indication of the universal role of betaine conserved throughout evolution. Proline is another solute protecting yeast from osmotic stress (Hohmann, 2002), but in our experiment, it increased for salt stress and decreased for osmotic stress, which was the reverse of the glycerol case. Proline and glycerol may counteract each other to fine-tune the response against the osmotic or salt stress. A significant increase in proline was also observed for the salt stress in the plant (Rentsch et al., 1996). Methionine was peculiar because its level was increased in all seven stress conditions unlike the other metabolites whose levels were either increased or decreased depending on the nature of the stress. The methionine biosynthetic gene MET22 was activated in the yeast cells under salt stress, resulting in increased methionine level; furthermore, the methionine supplementation improved salt tolerance (Glaser et al., 1993). Methionine is also involved in responding to starvation and oxidative stress (Petti et al., 2011). Our result is not only consistent with these reports but also implies the possibility of methionine as a general stress indicator. As for the lactic acid, yeast cells are known to prefer fermentations to oxidative phosphorylation even in the aerobic condition (Crabtree, 1928), and according to our result, this preference seemed to be more pronounced under the low pH or ethanol stress where the level of lactic acid was increased along with NAD, which could be indicative of the lactic acid fermentation. It was also interesting to observe the levels of glucose and acetic acid were elevated, which could be partially explained by the repression of the CIT2 gene under acid stress (Causton et al., 2001). We can speculate that the repression of CIT2. encoding a peroxisomal citrate synthase, would slow down the TCA cycle, and eventually glycolysis, leading to the accumulation of acetyl-CoA and then glucose. Acetic acid could have resulted from acetyl-CoA because of the hot water extraction method we employed.

In summary, our results indicate that (1) ethanol stress can be characterized by depletion of betaine, glucose, and maltose; (2) heat stress leads to depletion of lactic acid and UMP; (3) osmotic stress results in decrease in acetic acid, oxidized glutathione, lysine, maltose, proline and putrescine, and increase in betaine, glycerol, isoleucine, and threonine; (4) salt stress leads to decrease in maltose and increase in betaine, isoleucine, lactic acid, proline, threonine, and trehalose; (5) low pH causes depletion of betaine and increase in threonine; and (6) high pH results in increase in betaine, arginine, and methionine.

Multivariate analysis

As can be seen in Fig. 2, the amount of change varied within the same stress condition. In most cases, the relative deviation (standard deviation divided by average) was roughly in the range of 20–40%, but it could become as large as 88% for trehalose under the low pH stress. This large variation is believed to result from the sample preparation step, not from the NMR experiment. As noted by Dunn and Winder (Dunn & Winder, 2011), the intracellular metabolism should be quenched for accurate metabolite analysis, and they proposed an efficient quenching method based on the cold methanol/water mixture, which kept the cell pellet at a very low temperature, thus preventing any metabolite from being created or degraded. When we resuspended the yeast cells after the harvesting and the washing steps, the cells might have undergone some metabolic changes although we attempted to maintain a low temperature. And more importantly, each sample must have experienced slightly different variation in temperature during those steps, which could be the major reason regarding the variation between the samples under the same stress. If we should use the individual values of concentration changes of metabolites to characterize each stress condition, we would have to deal with large error bars and a small confidence window. Therefore, it would be very difficult to say whether a metabolite was an indicator of a specific stress. To better analyze and differentiate the stress conditions, we needed a better way to draw any meaningful pattern from these widely spread data. Principal component analysis (PCA) was employed in our study as a method of choice to deal with such data. The intensity data were read into the R software package, and the PCA was applied. The script was kindly written and provided by Ian Lewis (Princeton University, USA) and further modified in-house.

As shown in Fig. 4a, the same specific stress samples made their own clusters except for one outlier (low3) whose concentration was the lowest, possibly becoming the least accurate. The salt, osmotic, ethanol, and low pH stresses could be clearly distinguished from one another in the first 2 principal component spaces while the oxidative, heat, and high pH stresses were separated from the above-mentioned, but clustered together along with the control. It was interesting to observe that the salt, osmotic, and ethanol stresses were separated far from one another although those conditions could be characterized by the decrease in the surrounding water content, applying the osmotic pressure of transferring water molecules from inside the cells. When projected to the first and third component spaces, the overlap was reduced so that the controls were separated from the stresses. The heat and oxidative stresses made their own clusters while the high pH stress were overlapping and bridging between the two (Fig. 4b). These observations led to the conclusion that the heat, oxidative, and high pH stresses were similar in nature to the yeast cells so that they adapted themselves and coped with those stresses in a similar way. If more severe conditions had been used for those three stresses, there might have been better separation or tighter clustering. We are currently probing the stress conditions with differential severity.

4

PCA results. Scores plots along the first and second component axes (a) and along the first and third axes (b). Biplots along the first and second component axes (c) and along the first and third axes (d). For biplots, the symbols represent as follows: A, 4-aminobutyrate; B, acetic acid; C, adenosine; D, ADP; E, alanine; F, arginine; G, asparagine; H, aspartic acid; I, betaine; J, choline; K, ethanolamine; L, glutamine; M, glutamic acid; N, glycine; O, glycerol; P, histidine; Q, isoleucine; R, lactic acid; S, leucine; T, lysine; U, malic acid; V, maltose; W, methionine; X, NAD; Y, ornithine; Z, phenylalanine; a, proline; b, putrescine; c, serine; d, succinic acid; e, threonine; f, trehalose; g, UMP; h, valine.

From the biplots, we can see which metabolites are responsible for positioning the stress conditions in the principal component space. In Fig. 4c, trehalose, betaine, maltose, and lactic acid seems to be the dominant markers for the salt, osmotic, low pH, and ethanol stresses, respectively when viewed onto the first two principal components. Gene profiles for yeast under NaCl and sorbitol stresses have been described as similar (Hirasawa et al., 2006), but we can clearly see those two were separated by the actions of trehalose and/or betaine. Although many labels are overlapped, we can observe that most metabolites increased compared with the control when ethanol and low pH stresses were given. This indicates that there is more metabolic burden, and it is more difficult for the cells to deal with the two stresses, and the adaptation response is more complex in these situations. On the other hand, oxidative, heat, and high pH stresses showed much smaller metabolic changes. However, if we project onto the first and third component spaces (Fig. 4d), we can roughly say that heat and oxidative stresses were separated from others by glycerol and betaine, respectively. The high pH stress seems to be the mixture of heat and oxidative stresses according to its position.

Conclusions

We have shown that 2D HSQC spectra combined with a statistical analysis produced a robust and reliable result on the effects of different stresses. Among the seven stress conditions we studied, the osmotic, salt, ethanol, and low pH stresses separated themselves from others, indicating their metabolic consequences were distinct. The high pH, heat, and oxidative stresses clustered together, indicating that their metabolic consequences seemed to be similar. This kind of classification does not depend on finding a unique biomarker; instead, it uses ‘common’ metabolites and a combination of their concentrations as a whole. We predict that this method can lead to the classification of not only the yeast stress responses but also environmental responses of other organisms including humans.

Supporting Information

Fig. S1. Growth curves of yeast cultures under different stress conditions.

Acknowledgements

We thank Mr James Ellinger for his helpful advice on this manuscript and Profs. David Hindman and Julian Chung for proofreading. We also thank Mr Hyon Gyu Bok and Ms Suhyun Yoon for their technical assistance. This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science, and Technology (2010-0007161).

Footnotes

  • Editor: Hyun Ah Kang

References

View Abstract