This Article Abstract Full Text (PDF) Supplemental material Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Reprints and Permissions Copyright Information Books from ASM Press MicrobeWorld Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Lai, L.-C. Articles by Kwast, K. E. Search for Related Content PubMed PubMed Citation Articles by Lai, L.-C. Articles by Kwast, K. E.

 Previous Article  |  Next Article 

Molecular and Cellular Biology, May 2005, p. 4075-4091, Vol. 25, No. 10
0270-7306/05/$08.00+0     doi:10.1128/MCB.25.10.4075-4091.2005
Copyright © 2005, American Society for Microbiology. All Rights Reserved.

Dynamical Remodeling of the Transcriptome during Short-Term Anaerobiosis in Saccharomyces cerevisiae: Differential Response and Role of Msn2 and/or Msn4 and Other Factors in Galactose and Glucose Media

Department of Molecular & Integrative Physiology,¹ Department of Computer Science, University of Illinois, Urbana, Illinois²

Received 16 September 2004/ Returned for modification 19 October 2004/ Accepted 23 February 2005

	ABSTRACT

Top Abstract Introduction Materials and Methods Results Discussion References

 
In contrast to previous steady-state analyses of the O₂-responsive transcriptome, here we examined the dynamics of the response to short-term anaerobiosis (2 generations) in both catabolite-repressed (glucose) and derepressed (galactose) cells, assessed the specific role that Msn2 and Msn4 play in mediating the response, and identified gene networks using a novel clustering approach. Upon shifting cells to anaerobic conditions in galactose medium, there was an acute (10 min) yet transient (<45 min) induction of Msn2- and/or Msn4-regulated genes associated with the remodeling of reserve energy and catabolic pathways during the switch from mixed respiro-fermentative to strictly fermentative growth. Concomitantly, MCB- and SCB-regulated networks associated with the G₁/S transition of the cell cycle were transiently down-regulated along with rRNA processing genes containing PAC and RRPE motifs. Remarkably, none of these gene networks were differentially expressed when cells were shifted in glucose, suggesting that a metabolically derived signal arising from the abrupt cessation of respiration, rather than O₂ deprivation per se, elicits this "stress response." By 0.2 generation of anaerobiosis in both media, more chronic, heme-dependent effects were observed, including the down-regulation of Hap1-regulated networks, derepression of Rox1-regulated networks, and activation of Upc2-regulated ones. Changes in these networks result in the functional remodeling of the cell wall, sterol and sphingolipid metabolism, and dissimilatory pathways required for long-term anaerobiosis. Overall, this study reveals that the acute withdrawal of oxygen can invoke a metabolic state-dependent "stress response" but that acclimatization to oxygen deprivation is a relatively slow process involving complex changes primarily in heme-regulated gene networks.

	INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

 
In response to environmentally stressful conditions, all organisms mount defensive strategies essential for survival. The first line of defense often consists of low-molecular-weight compounds (e.g., trehalose) and proteins (e.g., chaperones), which afford immediate protection against a variety of environmental insults. Activation of stress signaling pathways, in turn, leads to the transcription of genes that serve additional protective roles. These can include both stress-specific genes (e.g., oxidative stress genes) and a common set involved in a more general response to stress. Genomic studies in Saccharomyces cerevisiae have revealed a large, overlapping set of genes that are differentially expressed in response to variety of environmental insults, including temperature shock, osmotic stress, oxidative stress, low pH, glucose or nitrogen starvation, and DNA-damaging agents (12, 28, 63, 70). These studies identified Msn2 and Msn4 as key regulators of both the environmental stress response (28) and common environmental response (12). Here, we examined the remodeling of the transcriptome during acclimatization to short-term anaerobiosis under different metabolic states (catabolite repressed or derepressed) and assessed the specific role these factors play in mediating the response.

Msn2 and Msn4 are Cys₂His₂ zinc finger proteins that activate the expression of a number of stress-inducible genes (60, 73). Although often considered to be functionally redundant in part because they activate gene expression through a common site, the stress response element (STRE), they are differentially regulated and may play distinct roles under different environmental conditions (26, 29). Their activity is regulated by their subcellular localization, residing in the cytosol under standard growth conditions and translocating to the nucleus under stressful conditions (30). This translocation is thought to be controlled by their phosphorylation state, which may dictate their interactions with cytosolic anchoring proteins (Bmh2 and Bmh1) (9).

Several condition-specific signaling pathways are thought to influence the activity of Msn2 and/or Msn4 (hereafter referred to as Msn2/4 for simplicity) and other factors responsible for the environmental stress response (reviewed in reference 29). These include the target of rapamycin (TOR) pathway (9), the protein kinase C-mitogen-activated protein (MAP) kinase pathway (34, 44, 55, 65), the high-osmolarity glycerol-MAP kinase cascade (70), the Snf1 protein kinase pathway (61), and the protein kinase A-MAP kinase pathway (30, 79, 80). Additional signaling pathways that respond to changes in other environmental parameters may also feed in to the regulation of Msn2/4 and the general stress response. Although seemingly complex, such a multiplex of signaling cascades is likely required for dictating specificity in the cellular response to an environment in which changes in a multitude of different parameters (e.g., temperature, osmolarity, pH, O₂ availability, etc.) can occur simultaneously.

Previous studies of the oxygen-responsive transcriptome in yeast have focused on genes that are differentially expressed between steady-state aerobic and anaerobic conditions (8, 51, 68, 77) and the gene networks controlled by key regulators, such as Rox1 (8, 27, 51, 78), Upc2 (87), Hap1 (8, 78), and others (8). Although these studies have helped to identify O₂-responsive genes and the role these factors play in controlling expression, the dynamics of remodeling activity elicited by acute changes in oxygen availability remain largely unexplored. In this study, we examined dynamical changes in the transcriptome associated with the acute withdrawal of oxygen, compared the response under different metabolic states (respiro-fermentative and fermentative), and identified the gene networks involved using a novel clustering approach.

In regard to Msn2/4, although our previous analyses indicated little overlap between genes that are differentially expressed in response to O₂ deprivation and those activated by other environmental insults, a substantive fraction of anaerobically induced genes are annotated under "cell stress" and little is known about their regulation. This poor overlap could simply reflect differences in the time courses examined (e.g., acute versus chronic) or, alternatively, O₂ deprivation may activate entirely different gene networks than those activated by other environmentally stressful conditions. To distinguish between these possibilities and explore the dynamics of the response, we compared genomewide expression profiles obtained with 70-mer oligonucleotide microarrays between an msn2/4 strain and its isogenic parent during short-term acclimatization to anaerobiosis. To aid in the identification of gene networks, we present a novel method for determining a clustering approach (algorithm and distance metric) and number of clustering divisions that results in the most nonrandom configuration of transcription factor motifs (TFMs) among gene clusters.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

 
Yeast strains, media, and growth conditions. The following yeast strains were used: JM43 (MAT leu2-3,112 his4-580 trp1-289 ura3-52 [rho⁺]) (wild type) and the isogenic strain KKY8, which contains msn2::LEU2 and msn4::KanMX disruptions. KKY8 was constructed by transforming JM43 with a BglII fragment from pmsn2::LEU2 and then transforming the resulting strain with a BamHI-EcoRI fragment from pmsn4::KanMX (plasmids were a gift from R. S. Zitomer). Gene disruptions were confirmed by Southern blot analysis.

Cultures were grown in a semisynthetic galactose or glucose medium containing Tween 80, ergosterol, and silicon antifoam (SSG-TEA and SSD-TEA, respectively) (10). Amino acids and nucleotides were added, as appropriate, at a concentration of 40 mg/liter. Liquid precultures were grown at 28°C on a shaker (200 rpm) and kept in early to mid-exponential growth phase (<100 Klett units, optical density at 600 nm <1.0) for 3 to 4 days prior to inoculating a New Brunswick BioFlo III fermentor (3.5-liter working volume) (51). The fermentor was inoculated with an appropriate volume of preculture so as to yield an initial cell density of 0.2 Klett unit. The cultures were sparged with air (1.2 vol/vol of medium per min) for six generations. A sample (aerobic control) was then harvested before switching the gas to 2.5% CO₂ in O₂-free N₂ for 2 generations of anaerobic growth. This procedure ensured the cell density upon final harvesting was 60 Klett units. The dissolved O₂ concentration was calculated from the output current of a 12-mm Ingold polarographic O₂ sensor, which was calibrated for 0% O₂ saturation with Ingold sensor-checking gel and for 20.94% O₂ with air-saturated medium. Corrections were made for temperature and barometric pressure (for SSG-TEA medium at 101 kPa and 28°C, 20.94% O₂ = 238 µM O₂).

To compare the response of the wild-type and msn2/4 strains, samples were harvested at the same time points (0, 10, 20, 45, and 480 min) after the shift to anaerobiosis as well as after the same relative number of cell mass doublings (0, 0.04, 0.08, 0.19, and 2 generations) as assessed by turbidity measurements (Klett meter). As discussed in Results, only the latter comparison provides good alignment of the temporal signatures between the strains. Thus, this same "generation-specific" sampling regimen (0, 0.04, 0.08, 0.19, and 2 generations) was used for comparing the effects of carbon source (galactose or glucose) in the wild-type strain. For sampling, cells were harvested using a vacuum filtration apparatus (11) onto AcetatePlus membranes (GE Osmonics, Minnetonka, MN). The filtered cells were washed with either sterile oxygenated or deoxygenated water, as appropriate, flash-frozen in liquid N₂ within one minute of initiating the sampling, and stored at –80°C for later RNA isolation (11). At least three independent fermentor experiments were completed for each strain and growth condition examined.

RNA extraction, cDNA synthesis, and microarray hybridization. Total RNA was extracted from the filtered cells using hot phenol as described previously (11). Thirty micrograms of total RNA was used for first-strand cDNA synthesis and microarray target preparation following previously described methods (32) with the following modifications. For cDNA synthesis, Superscript III reverse transcriptase (Invitrogen, Carlsbad, CA) was used with a 4:1 ratio of amino-allyl-dUTP to dTTP so as to yield about 1 dye molecule per 15 to 20 nucleotides. QIAquick mini-PCR purification columns (QIAGEN, Valencia, CA) were used for cDNA purification and unincorporated dye removal. For the latter, an additional first wash step with 750 µl of 35% guanidine hydrochloride was included for more efficient free-dye removal. The amount of cDNA obtained and dye (Cy3 or Cy5) incorporated was quantified using wavelength-scanning (750 to 200 nm) spectrophotometry with a 50-µl microcuvette.

Fluorescent cDNA targets were dried under vacuum (SpeedVac) and resuspended in an appropriate volume of hybridization buffer (50% formamide, 5x SSC [1x SSC is 0.15 M NaCl plus 0.015 M sodium citrate], 0.1% sodium dodecyl sulfate, and 0.5 µg/µl of tRNA) so as to yield equal amounts of Cy3 (query cDNA) and Cy5 (reference cDNA) dyes. A reference design was used for all microarray hybridizations. The reference consisted of a pool of equal masses of RNA from each time point sampled. The 80-µl samples (2.5 µl per cm² of coverslip) were denatured at 95°C for 2 min, centrifuged (16,000 x g for 1 min), and immediately applied to the microarrays. The arrays were sealed in a humidified aluminum chamber and placed in a 42°C water bath for 16 h.

After hybridization, the arrays were washed individually with mild agitation in 50 ml each of 2x SSC and 0.1% SDS at 42°C for 5 min, 0.2x SSC at room temperature for 1 min, and 0.1x SSC at room temperature for 1 min. The slides were dried by centrifugation (1,000 rpm for 5 min) and scanned within 2 h of processing using a GenePix 4000B confocal laser scanner (Axon, Union City, CA). The custom microarrays consisted of QIAGEN/Operon's yeast genomic 70-mer oligonucleotide set (QIAGEN, Valencia, CA; version 1.1) spotted in duplicate at a concentration of 20 µM in 150 mM sodium phosphate (pH 8.5), Arabidopsis oligonucleotide spike controls (Stratagene SpotReport, La Jolla, CA) spotted in quadruplicate, and 10 human and 10 yeast oligonucleotide negative controls spotted in duplicate. The oligonucleotides were printed on Codelink slides (Amersham, Piscataway, NJ) by Microarrays, Inc. (Nashville, TN). Postprint processing was conducted according to the manufacturer's recommendations.

Microarray and statistical analyses. GenePix Pro software (v4.1) was used for spot identification and fluorescence intensity quantification. Spots with aberrant measurements due to array artifacts or of poor quality were manually flagged and removed from analyses. Local background fluorescence was subtracted from the median Cy3 and median Cy5 fluorescence intensity values. Any resulting negative intensity values, which accounted for <0.5% of the total observations here, were set to zero, and a constant of one fluorescent unit was then added to all intensity values. Outliers were identified using SAS (SAS Institute, Cary, NC) as those observations whose fluorescence intensity deviated significantly (P < 0.01) from the average of all six observations (duplicate spots on each of three independent replicate slides per treatment).

After removing outliers, the log₂ Cy3 intensity (query cDNA) for all observations on a slide was normalized against that of the log₂ Cy5 intensity (reference cDNA) using locally weighted linear regression (Loess) in SAS. The linearity of the resulting Cy3 and Cy5 intensities across each slide was checked against the Arabidopsis spike controls (fluorescence intensity versus spike mRNA amount [ranging from 0.02 to 2 ng]) and a slope adjustment was made if the latter deviated significantly (P < 0.01) from a value of 1 (zero occurrences in this study). The log₂ (Cy3/Cy5) ratio for each spot was calculated, and the mean log₂ (Cy3/Cy5) ratio across all observations on a slide was normalized to a value of zero. The mean of the normalized log₂ (Cy3/Cy5) ratio for each gene was then calculated by averaging the duplicate observations on each slide and pooling replicate slides by strain, medium, and sampling time.

Statistical analyses were performed as a three-factor analysis of variance (ANOVA) using the SAS MIXED procedure with repeated measures. The factors were medium (galactose or glucose), strain (wild-type or msn2/4), and time (0, 0.04, 0.08, 0.19 or 2 generations after the switch to anaerobiosis). A step-down Bonferroni post hoc P value adjustment was used to minimize false positives. Unless otherwise noted, all P values reported were adjusted using this procedure. Postmodel analyses included promoter searches (–1 to –800 bp excluding upstream ORFs) using regulatory sequence analysis (RSA) tools (83) or other freely available, web-based bioinformatic tools. Note that the full data set has been deposited at GEO with accession number GSE1879 (http://www.ncbi.nlm.nih.gov/projects/geo/query/acc.cgi?acc=GSE1879), and the results of all statistical comparisons (see items S1 to S5 in the supplemental material) are available.

Data clustering and gene network discovery. To aid in the identification of the gene networks, we developed a method that uses two metrics to assess the quality of clustering results obtained with different algorithms and distance metrics and determine an optimal number of clusters (K) based on the distribution of transcription factor motifs (TFMs): consensus share (CS), the percentage of genes that are consistently grouped together over multiple runs of a clustering algorithm, and the motif configuration statistic (MCS), a novel metric that determines which clustering approach and cluster number (K) results in the most nonrandom configuration of transcription factor motifs (TFMs) among gene clusters.

To begin, the temporal profiles in gene expression were clustered ten times with different algorithms (K-means, K-medoids, or self-organizing map [SOM]) and distance metrics (Euclidean, Manhattan, Sup, or correlation) using a range of K values, in this case 2 to 50. These metrics were then calculated for each value of K and clustering approach (algorithm and distance metric) examined. Rather than using mean expression values from the microarrays, or models for the inclusion of the variance estimate (38, 89), individual replicates (n 3) were used as features in clustering. Genes that were not consistently clustered together over replicate runs of a given clustering approach were placed into a separate group and excluded from MCS calculations. The configuration of 1,813 transcription-factor consensus binding sequences (see Table S1 in the supplemental material), taken from both experimental and comparative phylogenic studies (15, 46, 54, 69, 76, 91), was assessed among the gene clusters by calculating the MCS for each.

The MCS metric is presented using the following definitions. Let n = (n₁... n_i) be a vector of cluster sizes and m = (m₁... m_k) be a vector of motif counts per cluster (motif configuration). The vector of relative cluster sizes is defined as q = (q₁... q_k), where

For the null hypothesis we assume that m came from the multinomial distribution M(q), i.e., larger clusters have higher motif counts. In this case the value of the probability mass function for each motif configuration m can be calculated as:

(1)

and the MCS is then given by the probability P(m) = P[p(r) p(m)], where r is a sample drawn from M(q). Because it is not feasible to compute the statistic exactly, we use the following algorithm to approximate it. Starting with counter c = 0, we repeat the following steps 10⁶ times: distribute the total motif count uniformly among genes; calculate the vector of motif counts per cluster r = (r₁... r_k) by grouping gene counts per cluster; compute p(r) according to equation 1; if p(r) p(m), then increase counter c. The approximation of MCS P value is given by c/N. An average MCS value is calculated for all motifs in the list. By comparing the values of MCS and CS for different clustering approaches (both algorithms and distance metrics), we can determine which approach consistently uncovers the most structure from the expression profiles (highest CS) and which value of K yields the most nonrandom configuration of TFMs (lowest MCS) among gene clusters.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

 
Dynamics of the response to short-term anaerobiosis in galactose medium. In contrast to previous steady-state analyses of the O₂-responsive transcriptome, here we examined dynamical changes elicited by the acute withdrawal of O₂ in sparged batch cultures of yeast. To minimize the impact of changing medium components in batch, cultures were maintained at low cell densities (<60 Klett units) and harvested in early log-phase growth. Inoculation volumes were adjusted to achieve identical cell densities among treatments (strains or media) for all samples compared. Finally, cells were allowed to acclimate to fermentor conditions for six generations before harvesting the aerobic control and switching the sparge gas from air to O₂-free N₂.

By design these experiments examined the dynamics of the genomic response to a rapid change in O₂ availability during the transition from pseudo-steady-state aerobiosis to anaerobiosis, and, out of necessity, cultures were grown in batch because of the rapid sampling regimen and quantity of cells required at each time point using our 3.5-liter fermentation apparatus. As shown in Fig. 1, even with relatively high gas sparge rates (1.2 vol gas/vol of medium per minute) several minutes are required to purge O₂ from the medium after switching the gas from air to 2.5% CO₂ in O₂-free N₂. Previous studies of the O₂ dependency for transcription of aerobic and hypoxic genes suggest 1 µM O₂ is a critical threshold (50), with maximal expression of aerobic genes above this value and that of hypoxic genes below it. Thus, the first sampling in this study (10 min) roughly corresponds to the time at which this oxygen concentration is achieved.

View larger version (12K): [in this window] [in a new window]   FIG. 1. Change in O2 concentration over the first 10 min of the shift to anaerobiosis. The change in dissolved O2 concentration (µM) in the fermentor is plotted as a function of time over the first ten minutes after switching the sparge gas from air to 2.5% CO2 in O2-free N2. The O2 concentration was calculated from the dissolved oxygen level measured with a 12-mm Ingold polarographic O2 sensor and is based upon the solubility of O2 in the media at 28°C and ambient barometric pressure.

View larger version (21K): [in this window] [in a new window]   FIG. 2. Dynamics of oxygen-responsive gene induction and repression during short-term anaerobiosis in galactose medium. The number of genes that responded significantly (P < 0.01) to the shift in O2 concentration in the wild-type strain (JM43) grown in galactose medium (SSG-TEA) is plotted as a function of both time and relative number of doublings in cell mass (generations) after the shift to anaerobiosis. Genes are divided into those that were significantly up-regulated and those that were significantly down-regulated. Black bars indicate the number of genes that were identified for the first time at that time point to exhibit a significant change in expression from that of the aerobic control (time zero). Gray bars indicate the number of genes that were differentially expressed relative to the aerobic control yet had already been identified at an earlier time point to have responded significantly to the shift in O2 concentration. The combined height of the black and gray bars indicates the total number of genes at each time point that showed a significant change in expression relative to the aerobic control.

Figure 3 compares CS (upper panel) and MCS (lower panel) as a function of cluster number for three different algorithms: SOM with one-dimensional (1D) ring topology (solid line), K-means (dashed line), and K-medoids (dotted line). The results were obtained from unbiased clustering of the temporal profiles of the 938 differentially expressed genes (P < 0.01) in the wild-type strain grown in galactose media (SSG-TEA) using Pearson correlation as the distance metric and replicates as features. From this figure it is clear that the SOM algorithm finds far more structure in the gene expression profiles than either K-means or K-medoids as evidenced by higher CS values for all K > 2. Moreover, the SOM algorithm partitions the gene expression profiles in a manner that results in a more nonrandom configuration of TFMs among clusters (lower MCS) for nearly all values of K examined (2 to 50). We interpret the inferior performance of both K-means and K-medoids to be due, in part, to the fact these algorithms do not use topological information.

View larger version (28K): [in this window] [in a new window]   FIG. 3. Comparison of the performance of different clustering algorithms in terms of consensus share and the motif configuration statistic as a function of cluster number (K). Three clustering algorithms were evaluated in terms of the resulting gene-to-cluster consensus (consensus share) and the motif configuration statistic (MCS) as a function of cluster numbers (K = 2 to 50) using Pearson correlation as the distance metric: SOM (Kohonen map) algorithm with 1D ring topology (solid line), K-means (dashed line), and K-medoids (dotted line). Consensus share (upper panel) is the percentage of genes that were consistently grouped together over ten runs of the algorithm. An average MCS P value (lower panel) for 1,813 transcription factor consensus binding motifs (TFMs) was calculated using a compiled list of both experimentally verified and putative motifs (see Table S1 in the supplemental material) as described in Materials and Methods. Lower MCS values indicate more nonrandom configurations of TFMs among gene clusters. The clustered data are individual replicate expression levels of all genes that significantly (P < 0.01) responded to the shift to anaerobiosis in SSG-TEA (galactose) medium.

In interpreting the results presented in Fig. 3, it is important to realize that genes were clustered based solely on temporal changes in their expression (unbiased clustering) and these quality metrics (CS and MCS) were calculated from the results obtained. As presented, the MCS is a global metric that examines the configuration of all TFMs provided among gene clusters to determine which clustering approach and K value results in the most nonrandom configuration of TFMs among gene clusters (lowest MCS). Although MCS can be approximated by other, computationally more tractable, methods (e.g., Chi-square fitness test), we developed the generally more applicable Monte-Carlo sampling approach given that our motif analyses frequently did not meet the central limit theorem (CLT) criterion for normal approximation, i.e., that each cluster contain at least 5 expected occurrences of each motif provided in the TFM list. The resolution of this approach depends on the performance of the algorithm in terms of its ability to partition genes into coregulated groups based upon expression profiles alone and, second, on the use of a TFM list that includes all motifs that are responsible for the observed expression differences.

In regard to the latter, we use a comprehensive list of 1,813 consensus sequences (see Table S1 in the supplemental material) taken from both experimental and comparative phylogenic studies (15, 46, 54, 69, 76, 91). Although the list includes experimentally unverified and thus perhaps dubious motifs, their inclusion should have little effect on the choice of an appropriate clustering approach and K value; such motifs or ones that are "inactive" under the experimental conditions examined are expected to have a random distribution among network defined clusters, and, thus, their inclusion should have little influence on the average value of MCS. The lower the average MCS value for all TFMs the more biased the distribution of those that are most likely to be responsible for the observed differences in expression profiles.

From the clustering quality assessment shown in Fig. 3, we chose to further analyze the gene networks recovered with the SOM algorithm for K = 17 as this yields the lowest MCS P value (0.37) while retaining high consensus share (94%). Panel D in Fig. 4 shows the clustered expression profiles as heat maps for the 938 genes that responded significantly (P < 0.01) to the shift to anaerobiosis in the wild-type strain grown in galactose medium. Because of the 1D ring topology, expression profiles of genes in adjacent clusters are most similar to each other, with cluster 17 to cluster 1 serving to close the ring. The 54 genes shown in cluster 0 are those that were not consistently grouped together over ten runs of the SOM algorithm and were omitted from the MCS calculations. From this figure it is clear that the algorithm nicely divides the temporal signatures into those that are primarily down-regulated (clusters 1 to 9) from those that are up-regulated (clusters 10 to 17), with further partitioning based upon differences in the timing of the response. Comparison of these temporal signatures to those obtained in glucose medium (Fig. 4A to C) and to an msn2/4 strain (F to H) will be discussed in separate sections below.

View larger version (47K): [in this window] [in a new window]   FIG. 4. Heat maps and statistical comparisons of oxygen-responsive genes in a wild-type strain grown in both galactose and glucose media and in an msn2/4 strain grown in galactose. The temporal profiles of genes whose transcript levels responded significantly (P < 0.01) to the shift to anaerobiosis (wild-type strain, SSG-TEA) were clustered using an SOM algorithm with 1D ring topology (K = 17). Panel D is a heat map of the temporal changes in gene expression relative to the aerobic control (time zero) plotted as a function of the number of generations of anaerobiosis (0, 0.04 0.08, 0.19, and 2). Green indicates down-regulated expression and red indicates up-regulated expression. Genes are sorted into 17 primary clusters as indicated in panel E. Cluster 0 contains genes that exhibited unstable cluster membership. Panel B shows the response of the same genes shown in panel D but in glucose (SSD-TEA) as opposed to galactose medium (wild-type strain). The black bars in panel A indicate genes that responded significantly (P < 0.01) to the shift in O2 concentration in glucose medium. Black bars in panel C indicate genes that showed a differential response (P < 0.01) in the two media (galactose versus glucose) as assessed by ANOVA. The purple bars in panel F identify putative Msn2/4-regulated genes, that is, genes that were anaerobically induced (P < 0.01) in the wild type but expressed at significantly (P < 0.01) lower levels in the msn2/4 strain and contained one or more STRE sites. Panel G is an expansion of the heat map shown in panel D for Msn2/4-regulated genes in the wild-type strain. Panel H is the expression profiles of the Msn2/4-regulated genes in the mutant strain, with yellow boxes indicating samples for which the transcript level was significantly (P < 0.01) lower in the msn2/4 strain compared to its wild-type parent.

An example where this is apparent is the partitioning of genes between clusters 1 and 2, where there is only a slight difference in the timing of the response (Fig. 4D) yet no overlap in the TFMs enriched in each. In general, this clustering approach can be used to screen for the predominate TFMs that are most likely to be responsible for the observed expression patterns. However, caution must be exercised in interpreting the results, as one cannot deduce from the mere presence of an enriched motif that changes in the activity of its associated transcription factor are directly responsible. Motifs that are "inactive" under the experimental conditions examined may also be enriched, especially in clusters that contain functional regulons that are controlled by a multiplex of transcriptional networks. Thus, the clustering results are best interpreted in the context of any additional knowledge of regulation and/or function that may be available.

In terms of the specific genes networks identified here, several TFMs and functional categories were predictably enriched based upon previous experimental studies (reviewed in references 50 and 92), including steady-state analyses of the O₂-responsive transcriptome (51, 68, 77). These include HAP1 and HAP2/3/4/5 in clusters 7 and 8, respectively, which were significantly enriched for genes involved in respiration and energy metabolism; UPC2 in clusters 15 to 17, which were enriched for lipid, fatty acid, and isoprenoid metabolism, and cell wall genes; and ROX1 in cluster 16, which was enriched for cell wall and cell rescue, defense and virulence genes. Moreover, there was a temporal delay (45 min) in the response of these gene clusters, a predicted result based upon studies of their regulation by heme (reviewed in references 50 and 92). In addition to these motifs, others not previously associated with the anaerobic response were also enriched, notably in clusters of transiently responding genes (clusters 1 to 5 and 12 to 13). These included MCB, SCB, MBP1, SWI4, and SWI6 in cluster 1, which was significantly enriched for DNA synthesis/replication and the cell cycle genes; PAC and RRPE in clusters 2 to 5, which were enriched for rRNA processing genes; and MSN2/4 in clusters 12 to 14, which were enriched for carbohydrate and reserve energy metabolism genes.

Interestingly, many of the same motifs have been found in sets of genes that transiently respond to other environmental challenges (12, 28, 63, 66, 67, 70, 90). Indeed, a comparison of these transiently responding gene networks with those identified in the environmental stress response (12, 28) reveals substantive overlap: 51% (173 of 340) of those that were transiently down-regulated are also down-regulated in the environmental stress response (585 total) and 23% (59 of 259) of those that were transiently up-regulated are also up-regulated in the environmental stress response (283 total). In contrast, there is very little overlap between the environmental stress response and the set of genes that were more chronically down-regulated (1%) or up-regulated (2%) here. From these analyses it is clear that there is a bifunctional response to the shift in oxygen: one that consists of acutely yet transiently responding gene networks that appear to function in a general stress response and a more delayed and chronic one comprised of heme-responsive gene networks that are associated more directly with acclimatization to oxygen deprivation. In the following sections, we explore functional attributes of these transiently responding gene networks.

DNA synthesis and repair and the cell cycle. Some of the first genes to respond to the O₂ shift were transiently down-regulated (cluster 1 in Fig. 4D) and are associated with DNA synthesis/repair and the cell cycle (Table 1). Overall, there is remarkable enrichment for TFMs whose factors, i.e., Mbp1p-Swi6p (MBF) and Swi4p-Swi6p (SBF), are known to coregulate genes required for progression through the G₁/S transition of the cell cycle (36, 39, 42). Genes associated with such functions include several for chromosomal replication initiation (CDC45, CDC54, TAH11, and YLR003C), DNA replication (POL1, -12, and -30, DBP2, and RNH201), concomitant repair (LRP1, RFA2, and RDH54), checkpoint function (RFA2 and MEC3), and chromosomal structure (IRR1, MCD1, ASF1, and PDS5). DNA synthesis and chromosome maintenance are closely linked and monitored at the replication fork and, thus many of the genes affected are associated with telomeres and their regulation (e.g., ESC8, DOT1, RIF1, and TBF1) or more than one process (e.g., CDC9, CTF18, HO, MSH2, RAD27, SMC5, TOF1, TOP1, TOS4, and TRF5). Genes that supply nucleotides for DNA synthesis were also transiently down-regulated (ADE1, -4, -12 and -17, GUA1, PRS3, RNR1, TRZ1, and URA2 and -7), as were salvage pathway genes (DCD1, HPT1 and URK1) and the uridine transporter (FUI1). Finally, genes associated with cytokinesis (e.g., CBF2, NKP1, SLK19, GIN4, and KCC4) (6) and bud site selection (STE20, RHO4, BEM3, PWP2, and SKG6) were also transiently down-regulated. Given the function of these genes, it would appear that the abrupt cessation in respiration during the shift to anaerobiosis in galactose medium results in a transient arrest in the cell cycle, predictably at the G₁/S transition, i.e., before the cells commit to another round of DNA replication.

rRNA processing genes. In conjunction with cell cycle-related genes, a large number involved in rRNA processing were also transiently down-regulated (clusters 2 to 5, Table 1). Remarkably, 96% of these have one or more PAC sites in their promoter and about half have RRPE sites (37, 69). Studies of environmental stress responses (12, 28) have observed a similar down-regulation of such genes, followed by a Rap1-mediated down-regulation of ribosomal protein genes. However, unlike the environmental stress response, few genes encoding structural components of the ribosomes (only RPL4A, RPP2B, and RPS2 at P < 0.01) were significantly affected here. Notably, these rRNA processing genes include the majority that encode the U3 snoRNP complex (small subunit processome; BUD21, DIP2, ECM16, IMP3 and -4, MMP10, NOC4, NOP14, PWP2, RRP9, SAS10, and UTP4, -5, -7, -8, -14, -18, -20, and -21), which functions in the earliest steps of ribosome biogenesis and is essential for pre-18S rRNA processing (20). In addition, several involved in 20S pre-rRNA processing (EMG1, ENP1, NOP7 and -14, RIO2, UTP22, and YGR272C), 35S primary transcript processing (DBP3, -6, -8, -9 and -10, DRS1, FAL1, MRD1, MTR3 and 4, PXR1, RNT1, and RRP3), large ribosomal subunit biogenesis, processing, and assembly (ARX1, IPI3, MAK16, NMD3, NOC2, NOG2, NOP16, NSA2, RLP7 and -24, RPF1 and -2, RRB1, SQT1, SSF1 and -2, TIF6, and YTM1), and a large group involved in general rRNA processing and modification (CGR1, EBP2, ENP2, ERB1, IFH1, IMP3 and -4, IPI1, KRR1, LCP5, MAK5, MPP10, MRT4, NHP2, NOC3, NOP6 and -12, NSR1, NUG1, PNO1, POP1, RCL1, REX4, RRP1, -8 and -9, RRS1, SPB1, TSR1, and YJL010C) were also similarly affected. Genes for tRNA synthesis and/or processing (ALA1, DUS1, FRS2, GCD14, KRS1, MES1, PUS7, TRM1, TRM10, TRM2, TRM82, and YKR079C) as well as subunits of RNA polymerases I (RPA12, -14, -43, and -49), II (R-3), and III (RPC19, -40, and -53 and RPO31) were also down-regulated, similar to that observed during the environmental stress response.

From a functional viewpoint, rather than chronically down-regulating the capacity for translation under anoxia, it would appear there is a brief interruption in the processing and de novo synthesis of the cytoplasmic translational apparatus. Such a transcriptional program makes functional sense given the acute interruption in the steady-state rate of energy production during the switch from mixed respirofermentative to strictly fermentative growth in galactose. Although no factors have been identified that directly bind to PAC or RRPE, Sfp1, and/or Sch9 are likely involved given their role in regulating nucleolar genes for ribosomal biogenesis and translation at START (22, 43, 59).

In contrast to the cytoplasmic ribosomal complexes, the majority of genes encoding or associated with the mitochondrial 15S complex (MRP10, -13, -20 and -51, MRPS17, -18 and -35, RSM7, -18, -24, -26 and -28, and NAM9) and 21S complex (MRPL1, -3, -7, -8, -15, -19, -20, -22, -23, -33, -35, -36 and -51, MRP20, IMG1 and -2, RML2, and YML6) were chronically down-regulated after a substantial delay. Their chronic down-regulation is predictable as there would be little apparent benefit in supporting the mitochondrial translational apparatus to full aerobic capacity when mitochondrial function is restricted due to O₂ lack. Although these genes appear to be coregulated given their tight clustering pattern (primarily cluster 9), no TFMs were significantly enriched and directed promoter searches failed to reveal any overrepresented sequences. Hap2/3/4/5 may regulate some of them (31), which could account for the delay in their transcriptional down-regulation.

Carbohydrate utilization and reserve energy metabolism. Concomitant with the transient down-regulation of rRNA processing and cell cycle genes was the transient up-regulation of genes associated with carbohydrate and reserve energy metabolism (clusters 11 to 14, Table 1). Their up-regulation here suggests a program designed for preserving cellular energy status during the metabolic transition from mixed respiro-fermentative to fermentation growth. This is evidenced in part by the transient induction of genes for the import and catalysis of hexose sugars (GAL2 and HXT3, -4, -6, -7, -9, -11, -13, -15, -16, and -17), as well as the exploitation of alternative carbon sources such as maltose (MAL11, -12, -31, -32, and -33) and xylose (XKS1). Genes encoding key regulators in hexose dissimilation were also transiently up-regulated, including those for glucokinase (GLK1), hexokinase (HXK1), glucose-6-phosphate isomerase (PGI1), phosphofructokinase (PFK2), and 6-phosphofructo-2-kinase (PFK26 and -27).

Evidence for a drop in the cellular energy state includes the transient induction of negative regulators of gluconeogenesis (GID7, FYV10, VID24, -28 and -30, and UBC8) and genes that accumulate under glucose-limiting conditions (DCS1, DCS2, GPX1, SIP2, and the regulator SNF3). Genes encoding catalytic and regulatory subunits of trehalose-6-phosphate synthase (TPS1, TPS2, TPS3, and TSL1) were also transiently induced, similar to that observed during the environmental stress response. Trehalose is not only an important reserve energy store but is also involved in controlling glycolytic flux (23, 81), facilitating the transition from respiratory to fermentative growth (64), preserving protein structure during anoxia (14), and maintaining membrane integrity in the face of increasing ethanol concentrations (2). Unlike the environmental stress response, genes involved in its degradation (ATH1 and NTH1 and -2) were not differentially expressed here. Glycogen is also an important reserve energy source and stress protectant, especially during restricted growth (23, 67), and several genes encoding key regulators (MRK1, PCL6 and -10, PSK1, and GLC8) or enzymes involved in its elongation (GSY2) or branching (GLC3) were significantly induced. The cyclins PCL6 and PCL10 for Pho85 kinase, which may be responsible for the accumulation of glycogen and trehalose in G₁ for later use in the cell cycle (23), were also transiently up-regulated.

Overall, the transient induction of genes for trehalose and glycogen processing in conjunction with those for the dissimilation of sugars suggests a transcriptional program designed for ensuring adequate energy supplies during the transition from mixed respiro-fermentative to strictly fermentative growth. Indeed, when all transiently responding gene networks are viewed together, it is clear they are involved in balancing energetic supply and demand during this transition. Although this response is phenotypically similar to what has been referred to as the "environmental stress response" (28) or "common environmental response" (12), here it is clear that the "stress" encountered is one associated with the acute cessation of respiration and associated changes in energy metabolism during this metabolic transition. Given the commonalities in these responses to environmental change, we sought to define the specific role that the stress-inducible factors Msn2 and Msn4 play in mediating acclimatization to anaerobiosis.

Role of Msn2/4 in mediating the anaerobic response. Msn2/4 appear to play a ubiquitous role in regulating the response to environmentally stressful conditions. Thus, it was not surprising to find significant enrichment for Msn2/4 binding sites in groups of genes that were transiently induced in response to anaerobiosis (clusters 12 to 14, Table 1). To further define the role these factors play, we compared the temporal profiles of an msn2/4 double mutant strain to that of its isogenic wild-type parent. For these studies, cells were grown on the nonrepressing substrate galactose (SSG-TEA medium) to both facilitate comparisons to our previous steady-state analyses of the O₂-responsive transcriptome (51) and circumvent confounding effects of carbon source regulation on specific subsets of O₂-responsive genes (e.g., Hap2/3/4/5-regulated genes). However, as has been reported for other msn2/4 strains (21), the mutant grew significantly (P < 0.001) more slowly than its parent, especially under anaerobic conditions (Table 2).

Compared to the wild-type strain, preliminary genomic analysis of the msn2/4 strain revealed a substantial delay (3x) in its response to the shift in O₂ availability, corresponding well to its three times slower anaerobic growth rate (Table 2). Given this difference, we harvested anaerobic samples from each strain after the same number of doublings in cell mass (generations), and at the same time points for comparison, and then determined how best to align the temporal profiles for statistical comparisons. A time-warping algorithm (1) provided poor alignment (data not shown), presumably due to the limited number of time points sampled. However, comparison of the temporal profiles on a generation-specific versus time-specific scale (ANOVA) revealed the former provides good alignment between the strains, with far fewer genes (5% versus 14%) exhibiting a significant (P < 0.01) strain-time interaction. These results are perhaps not surprising given that we are assaying the end products (mRNAs) of a highly energy-dependent process (transcription) and the primary defect in the mutant is one associated with limiting rates of galactose fermentation. This same generation-specific time scale was used for comparing the effects of carbon source (galactose versus glucose) on the anaerobic response of the wild-type strain, as described in a separate section below.

Statistical comparison (mixed ANOVA with repeated measures) of the msn2/4 and wild-type strains revealed a total of 93 putative Msn2/4-regulated genes, i.e., genes that were anaerobically induced (P < 0.01) in the wild type but expressed at significantly (P < 0.01) lower levels in the mutant (see items S2 and S3 in the supplemental material for full results). Promoter searches (–1 to –800 bp) revealed that 65% (60 genes) contain one or more STRE sites (AGGGG), corresponding to a binomial enrichment P value of 4.8 x 10^–10. Predictably, most of these genes are found in clusters 11 and 12 (Fig. 4D), i.e., within clusters of genes that were transiently induced and significantly enriched for Msn2/4-binding sites (Table 1).

The expression profiles of these 60 genes are shown in panels G and H of Fig. 4 for the wild type and mutant, respectively. The yellow boxes in panel H indicate samples for which the transcript level in the mutant was significantly (P < 0.01) lower than that of its parent. From this figure it is clear that Msn2/4-regulated genes begin to respond by 0.04 generations (10 min in the wild type) after the shift to anaerobiosis, are maximally induced at 0.08 generation, and their transcript levels diminish thereafter. This time course is similar to that observed for the transient induction of Msn2/4-regulated genes in response to other environmental challenges (12, 28) and fits well with the kinetics of Msn2 nuclear import and export determined in previous studies (40). From this comparison it is also clear that factors in addition to Msn2/4 likely regulate these genes given that nearly half (25 of 60) are anaerobically induced in the mutant but to a significantly (P < 0.01) lower extent than in the wild type. Comparison of the genes identified here with those from studies of heat shock (28), H₂O₂ addition (28), or acidic conditions (12) reveals only moderate overlap (22 genes of 60), indicating modularity in the network of genes these factors control depending on the specific nature of the stress encountered.

In regard to the function of these genes, most are annotated under categories of energy metabolism (carbohydrate and reserve energy metabolism) or general cell stress. They include genes for import and utilization of hexose sugars (HXT13, -15, -16, and -17) or maltose (MAL31 and -32), reduction of pentose sugars (YJR096W and GRE3) or aldose (YDL124W), and for processing of mannitol (YEL070W) and fructose-2,6-bisphosphate (YLR345W, DSC2, and GPX1). Several are also involved in the processing of trehalose (TPS1, -2 and -3, TSL1) and glycogen (GDB1, GLC3, GSY2, and MRK1). Finally, others are associated with the mitochondrial function, including ubiquinone synthesis (COQ1 and -6), chaperone activity (TCM62 and MBA1), division (FIS1), aldehyde processing (ALD5), and other functions (SFT2 and CBP6). The function of the remaining Msn2/4-regulated genes is not known and, thus, this study should help with their annotation. Overall, given the function of the genes affected here, it would appear that Msn2/4 are involved in the retooling of catabolic pathways and energy reserves, presumably to ensure adequate energy supplies during the metabolic transition from mixed respiro-fermentative metabolism to the slower growth rates supported by fermentative metabolism alone. In the following section, we further explore this hypothesis.

Response to anaerobiosis in glucose medium. As noted above, the acute response to anaerobiosis in galactose medium is phenotypically similar to that invoked by a number of other environmental insults. This includes the transient down-regulation of genes for rRNA processing genes (PAC and RRPE) and DNA transcription and repair (MCB and SCB), as well as the transient up-regulation of Msn2/4-regulated genes involved in carbohydrate and reserve energy metabolism (Table 1). Although a number of explanations have been proposed for such "stress-induced" changes (12, 28), from a functional standpoint it would appear to be simply part of an energy balancing measure. In this study, this would be required as diminishing O₂ availability limits respiration, resulting in a shift to slower growth rates supported by galactose fermentation alone (Table 2). A similar response might also be elicited by any stress-induced decrease in energy production, whether chronic in nature, as here, or transient. Thus, rather than directly mitigating the environmental stress per se, Msn2/4 appear to orchestrate acute changes in catabolism (18, 49). To further explore this hypothesis, we asked if a similar genomic response would be elicited during the shift to anaerobiosis when growth rate, and the corresponding cellular energy status, is unaltered during the shift (see Table 2 for O₂- and medium-dependent growth rate comparisons). To examine this, we shifted cells to anaerobic conditions in glucose medium and repeated the microarray analyses.

ANOVA revealed that far fewer genes responded to the shift in O₂ availability in glucose (352 genes) than in galactose medium (938 genes) (see items S4 and S5 in the supplemental material for full results). For direct comparison, panel B of Fig. 4 shows the glucose-dependent response of those genes that responded significantly to the shift in galactose (panel D). Note that only those genes indicated by a bar in panel A were differentially (P < 0.01) expressed with respect to O₂ in glucose. Bars in panel C indicate genes that showed a differential response (P < 0.01) in the two media. From this figure it is clear that the majority of genes that transiently respond to the O₂ shift in galactose medium (clusters 1 to 5 and 10 to 15) fail to do so in glucose. Indeed, the biphasic nature of the response observed in galactose (Fig. 1) is compressed to a single, delayed response in glucose, as shown in Fig. 5. From Fig. 4C it is also clear that the majority of genes that respond similarly in the two media exhibit a delayed response to the O₂ shift. These comparisons reveal substantive carbon-source-dependent differences that are specific to the acute phase of the anaerobic response. This suggests that glucose either represses the expression of the transiently responding networks observed in galactose or that a signal other than the change in O₂ availability is responsible for eliciting such a response.

View larger version (20K): [in this window] [in a new window]   FIG. 5. Dynamics of oxygen-responsive gene induction and repression during short-term anaerobiosis in glucose medium. The number of genes that responded significantly (P < 0.01) to the change in O2 concentration in the wild-type strain (JM43) grown in glucose medium (SSD-TEA) is plotted as a function of the relative number of cell doublings (generations) after the shift to anaerobiosis. Genes are divided into those that were significantly up-regulated and those that were significantly down-regulated. Black bars indicate the number of genes that were identified for the first time at that time point to exhibit a significant change in expression from that of the aerobic control (time zero). Gray bars indicate the number of genes that were differentially expressed relative to the aerobic control yet had already been identified at an earlier time point to have responded significantly to the shift in O2 concentration. The combined height of the black and gray bars indicates the total number of genes at each time point that showed a significant change in expression relative to the aerobic control.

To further explore these carbon source-specific differences in the networks that respond to the shift in O₂ availability, we clustered the temporal profiles of the 352 genes that were differentially (P < 0.01) expressed in glucose. From the clustering quality assessment shown in Fig. 6 (panel A), we chose to analyze the results obtained with K = 11 as it results in the most nonrandom configuration of TFMs among clusters (lowest MCS P value) while retaining high consensus share (CS). As shown in the heat map of Fig. 6 (panel B), the SOM algorithm nicely partitions the transcript profiles into temporally shifted clusters of up- and down-regulated genes. Bars to the left of the heat map indicate genes that failed to respond (P > 0.01) to the O₂ shift in galactose whereas bars to the right indicate genes that were differentially (P < 0.01) expressed in the two media. Overall, this figure further illustrates similarities in the genes that exhibit a delayed response to the O₂ shift in the two media (clusters 4 to 6 and 8 to 10) yet substantive differences for those that more acutely respond (clusters 1, 2, and 11).

View larger version (18K): [in this window] [in a new window]   FIG. 6. Clustering quality assessment (A) and heat maps (B) for genes that significantly responded to the shift to anaerobiosis in glucose medium. The temporal profiles of genes whose expression responded significantly (P < 0.01) to the shift to anaerobiosis in glucose medium (SSD-TEA) were clustered using an SOM algorithm with 1D ring topology. Panel A shows the motif configuration statistic (solid line, left ordinate) and consensus share (dotted line, right ordinate) as a function of cluster number. Panel B is the heat map of the temporal changes in gene expression relative to the aerobic control (time zero) partitioned with 11 clusters. Green indicates down-regulated expression and red indicates up-regulated expression. Bars to the left of the heat map indicate genes that failed to respond (P > 0.01) to the O2 shift in galactose medium, and bars to the right indicate genes that exhibited a differential response (P < 0.01) in glucose versus galactose medium.

Delayed yet chronically expressed gene networks. In contrast to the transiently responding gene networks, there is very little overlap between the environmental stress response and the gene networks that were more chronically down-regulated (1%) or up-regulated (2%) in galactose, or indeed any of the genes that were either down-regulated (1%) or up-regulated (4%) in glucose. Changes in these networks include the chronic down-regulation of Hap1- and Hap2/3/4/5-regulated ones involved in mitochondrial function and the up-regulation of Rox1- and Upc2-regulated ones required for more long-term acclimatization to anaerobiosis. Given that we have only a limited view of the dynamical changes in these gene networks here, we limit our functional analyses to a few groups below.

Transition metal transport and utilization. Transition metals, such as iron, copper, zinc, cobalt, and manganese, are essential to all organisms for participation in a variety of redox reactions, and their transport and processing are closely monitored because the metals themselves can be toxic. The bulk of proteins that utilize these metals are associated with the processing of oxygen or its by-products and, thus, both the transport of these ions and synthesis of proteins that require them is affected by oxygen availability.

Here we observed the down-regulation of number of genes for the import of metals, including those encoding high-affinity iron transporters (FRE1, UTR1, FTR1, and FET3), yet up-regulation of the relatively nonspecific (Fe, Cu, Mn, or Zn) low-affinity transporter encoded by FET4, which is regulated by Rox1, Aft1, and Zap1 (41, 86). Fet4 may thus assume a pivotal role in regulating general metal-ion homeostasis during anaerobic conditions. However, increased Fet4 activity can lead to transition metal sensitivity (56), which may account for the simultaneous up-regulation of the metallothionine Cup1 and the copper chaperone for Sod1 (LYS7) here. A number of other genes for transporting or processing metals were chronically down-regulated (e.g., ARN1, CCC2, CCH1, COX17, HMX1, ISU2, OCT1, and SMF1). The up-regulation of IZH4, which is a plasma membrane protein induced by high Zn (57) or hypoxia (45), may serve a critical role in coordinating both sterol and zinc metabolism under anoxia (57).

Lipids and sterols. Although it has long been known that an exogenous source of unsaturated fatty acid and sterol is essential for long-term anaerobic growth in S. cerevisiae (3, 4), a more complete picture of the anoxia-induced remodeling in these pathways is revealed here. Notably, this remodeling is fairly specific for sterol and sphingolipid pathways, with very few genes for phospholipid or fatty acid synthesis (save for AAC1 and OLE1) showing changes in expression. For sterol, genes in the early portion of the pathway (for isoprenoid synthesis) exhibited complex expression patterns, with some showing transient (ERG10) or chronic up-regulation (IDI1 and HMG2) and others showing transient (ERG8 and MVD1) or chronic down-regulation (ERG13 and -20 and HMG1). However, nearly all of the genes (ARE1, ERG1, -2, -3, -6, -11, -24, -25, -26 and -28, and NCP1) in the later portion of the pathway were chronically up-regulated, as were genes involved in transport (PDR11 and AUS1) and their primary regulator (UPC2).

Apparently exogenous ergosterol is imported and cycles between membranes and lipid droplets but does not affect control points in the endoplasmic reticulum for its synthesis (5, 75), given that ergosterol synthesis is not possible without oxygen. In terms of their regulation, Upc2 and Rox1 control many of them (51, 84, 87), and binding sites for these factors were significantly enriched in clusters containing these genes (Tables 1 and 3). Finally, for sphingolipid synthesis, a number genes encoding the middle portion of the pathway, linking dihydrosphingosine to ceramide, and the putative signaling molecules (dihydrosphingosine-1-phosphate and phytosphingosine-1-phosphate) were affected here (e.g., SUR2, YSR3, LCB4, and YDC1). The net effect of these changes may be to increase phytosphingosine during the transition to slower anaerobic growth rates.

Cell wall, vesicle transport, and secretory networks. Similar to previous steady-state analyses of the anaerobic transcriptome (51, 77), here we see substantive evidence for remodeling of the plasma membrane and cell wall during the shift to anaerobiosis. This is reflected in the delayed but chronic up-regulation of genes for cell wall structure, modifying enzymes, protein secretion, vesicle trafficking, as well as lipid and sterol metabolism. These networks are largely controlled by Upc2 and/or Rox1, whose binding sites were significantly enriched in Gal-clusters 15 to 17 and Glu-clusters 8 to 10. Notably, this includes the up-regulation of nearly all of the seripauperin and TIP1 gene family (DAN1, -2, -3 and -4; TIR1, -2, -3 and -4; PAU1, -2, -3, -4, -5, -6 and -7; and 10 other open reading frames) as well as other genes for biogenesis or modification of the cell wall (e.g., CSR1, GSC2, and YOL155C).

Although the expression patterns of secretory genes were more complex, and probably reflect the slowing of growth, a number of such genes were up-regulated here (e.g., EUG1, CPR4, SRO77, AKR2, YOL075C, and PLB2). Genes for modifying transported proteins were also up-regulated, including several for glycosylation (PMT3, PMT5, KRE9, UGP1, and KTR4), prenylation (BET2), and GPI anchors (GPI12). Finally, genes involved in endocytosis (SRO77 and AKR2) were also up-regulated, including several (e.g., IZH4, CLC1, AKR1, RVS161, AFR1, and YBR108W) that were induced much earlier in the time course. Overall, such remodeling activity can be expected to alter cell wall porosity to accommodate the transport and processing of essential nutrients that are required during prolonged periods of anaerobiosis.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

 
In this study, we have found that the dynamics of the genomic response to acute oxygen deprivation consists of two distinct phases when cells are shifted from aerobiosis to anaerobiosis in galactose medium: (i) an acute, transitory one in which genes associated with the retooling of metabolism (respiro-fermentative to strictly fermentative and reserve energy) and slowing of growth rate are differentially expressed; and (ii) a delayed, chronic phase in which genes associated with long-term acclimatization to anaerobiosis are differentially expressed. The acute, transitory phase is absent when cells are shifted to anaerobiosis in glucose medium, conditions in which little catabolic remodeling is required and no change in the growth rate is observed.

Using a novel metric that we call the motif configuration statistic, which determines what clustering approach and number of clustering divisions results in the most nonrandom distribution of transcription factor motifs (TFMs) among gene clusters, we recover a transcriptional network from SOM clustering that shows that these phases are distinct from one another, both in terms of the gene networks that are involved and the signaling pathways that elicit these responses. This clustering approach identifies several gene networks, notably MBF, SBF, PAC/RRPE, and MSN2/4 networks, that were not previously known to be involved in the anaerobic response while simultaneously redefining those controlled by heme-responsive transcription factors, such as Hap1, Hap2/3/4/5, Rox1, and Upc2. Examination of an msn2/4 strain shows these stress factors (i.e., Msn2 and Msn4) are involved in the metabolic remodeling that occurs in galactose medium but do not directly respond to O₂ deprivation per se, as evidenced by their conspicuous absence in playing a role in the O₂-dependent remodeling of the transcriptome under glucose-repressed conditions.

Physiological acclimatization to anaerobiosis. In comparison to other environmentally stressful conditions examined in S. cerevisiae (12, 28, 49, 66, 67, 70, 90), acclimatization to anaerobic conditions occurs more slowly, over multiple generations. Perhaps this is not surprising, given that O₂ deprivation per se does not pose an immediate threat to cell survival or directly damage cellular components, unlike the case for oxidative, osmotic, or acid stresses, temperature shock, glucose and nitrogen starvation, or DNA-damaging agents. The major challenges facing a yeast cell growing anaerobically on a fermentable carbon source are apparently those associated with (i) the time-dependent depletion of essential cellular components that require molecular O₂ for their biosynthesis, (ii) maintenance of cellular redox balance, and (iii) mitigation of damage associated with the accumulation of anaerobic end products.

In regard to the first, molecular oxygen is required for the activity of various hydroxylases, desaturases, and oxygenases involved in the de novo biosynthesis of membrane components such as sterols and unsaturated fatty acids (reviewed in references 50 and 92). Although cells growing on rich medium can typically undergo several (two to three) anaerobic divisions without supplementation, an exogenous source of unsaturated fatty acids and sterol is essential for long-term anaerobic growth (3, 4). One of the most striking patterns in gene expression observed during anaerobic growth is the remodeling activity associated with the cell wall and plasma membrane. This remodeling is required, in part, for the efficient import and processing of these supplements in order to combat the compromised ability to regulate membrane fluidity (51). However, these changes are slow to occur and, as will be described elsewhere (L.-C. Lai, P. V. Burke, and K. E. Kwast, unpublished data), take several generations (>4) for completion.

Molecular oxygen is also required for the biosynthesis of nicotinic acid (33), ubiquinone (62), and some enzymes (e.g., hemoproteins), as well as for the activity of oxidases involved in Fe³⁺ uptake and the reduction of free radicals (e.g., ascorbate) (reviewed in r