To invoke rapid and full induction of fermentative capacity, respiratory, aerobic glucose-limited chemostat cultures (D = 0.1·h^-1) were shifted to fully fermentative conditions by sudden depletion of oxygen and addition of glucose. The glucose was added two min after sparging the continuous culture with pure nitrogen, when the dissolved oxygen concentration had decreased from 75–80% to 10–15% of air saturation (Fig. 1). This raised the glucose concentration to 200 mM and ensured that the residual glucose concentration after 2 h of cultivation would still be above 100 mM, thus maintaining strong glucose catabolite repression throughout the experiment. (Fig. 2A) 16. Indeed, the sudden shift to fermentative conditions resulted in fully fermentative metabolism within the first 5 min with CO₂, ethanol and glycerol as the major metabolic products (Fig. 2). This metabolic shift coincided with an increasing specific glucose consumption rate, up to 12-fold, over 2 h following the perturbation (Fig. 2B). The specific ethanol production rate, which under these anaerobic glucose-excess conditions reflects the culture's fermentative capacity, steadily increased to 19.6 mmol ethanol·g^-1·h^-1 (Fig. 2B). While the overall metabolic response was rapid and strong, the biomass concentration, the cell count and the cellular protein content did not change significantly throughout the experiment (Fig. 2C). During the two hours of the experiment, the growth rate did not exceed the starting growth rate of 0.1 h^-1. The biomass therefore only contributed to 5% of the total carbon flux, while the main metabolic products (i.e. carbon dioxide, ethanol and glycerol) accounted for ca. 90% of the total carbon produced.

To identify genome-wide transcriptional changes connected to the induced metabolic adaptation, micro-array analysis was performed on samples from two independent replicate steady-state chemostat cultures and on samples taken 5, 10, 30, 60 and 120 min after glucose addition. The coefficient of variation between replicates was below 20%, which is comparable with previous chemostat-based transcriptome analyses 1011.

A first main concern was with normalization of these microarray data from non-steady-state culture samples. In previous transcriptome studies on steady-state chemostat cultures using Affymetrix microarrays, setting the average signal intensity of all probe-sets to a fixed value (also called global scaling) provided a good normalization method 1011. As this normalization method might not be appropriate for dynamic cultivation conditions, we considered transcript levels of a few so-called 'house-keeping' genes commonly used as loading standards for Northern analysis and quantitative RT-PCR. After global scaling, the expression of ACT1, HHT2 and SHR3 (encoding respectively, for actin, histone and endoplasmic reticulum packaging chaperone protein) remained constant throughout the experiment with a variation coefficient around or below 20%. The stable transcript levels of house-keeping gene expression obtained with a global scaling approach indicated that no major changes in the total mRNA pools occurred during the experiment, which would require another type of normalization.

After global scaling, the significance of the changes in transcript levels during the dynamic experiment was estimated using the EDGE software (p-value threshold 0.005, 17, see Methods section for details). A set of 1923 genes was thus identified as being transcriptionally regulated in response to combined oxygen depletion and glucose addition (Additional file 1). This large group of genes was divided in several subgroups according to their expression profiles. 607 genes whose transcript levels increased after the perturbation were separated into four clusters according to their initial and later response (Clusters A-D; Fig. 3). 1316 genes with reduced transcripts responded rapidly to the perturbations (within 10 min) and were clustered according to their secondary response (Clusters 1–6; Fig. 3). All clusters were subsequently searched for overrepresentation of specific functional categories (as defined by MIPS 18), and of promoter elements corresponding to specific transcriptional regulation networks (see Methods section). Upon a first inspection, some of clusters, despite subtle differences in their time-dependent transcript profiles, showed an overrepresentation of genes from the same functional categories. These were pooled to further improve the enrichment analysis. Thus cluster A and B, as well as 2 and 3, and also 4, 5 and 6 were pooled (Table 1 and 2), resulting in a final set of six different clusters.

Sudden relief from glucose limitation enables yeast cells to accelerate to a higher specific growth rate. Although faster growth was not observed in the 2 h after the relief of glucose limitation, over one third of the initially up-regulated genes were related to protein synthesis (Fig. 3; cluster A, B and C). This included a massive and fast up-regulation of genes within clusters A and B that encode components of the translational machinery, including 126 genes involved in rRNA synthesis, processing and modification and 49 genes involved in ribosomal biogenesis (Table 1). Taking into account that the total RNA pool mainly consists of rRNA 19, an up-regulation of rRNA synthesis was confirmed by an increase of the RNA content of the biomass after the relief from glucose limitation (Fig. 4).

Genes in cluster C displayed a sustained, slower increase of their transcript levels than those in clusters A and B. 37% of the genes in cluster C encoded ribosomal proteins. The delay between the expression of ribosomal biogenesis/rRNA genes and ribosomal protein genes is in line with previous observations indicating the existence of different regulatory mechanisms for these two groups of genes 202122. Accordingly, PAC and RRPE regulatory elements were enriched in the promoter regions of genes in clusters A & B, whereas Rap1p/Sfp1p and Fhl1p motifs were overrepresented in the promoter regions of cluster C genes (Table 2). In addition to the translational machinery, 57 genes involved in amino acid metabolism and 46 genes involved in nucleotide metabolism were up-regulated. This was consistent with the overrepresentation of Met32p 23, Gcn4p 24 and Bas1p 25 binding sites in the promoter regions of these genes, and indicated the need for synthesis of building blocks for transcription and translation.

Among the 1316 genes with reduced expression, one cluster comprising 122 genes showed rapid and strong repression (Cluster 1, Fig. 3). Although this cluster appeared relatively heterogeneous, one functional category was clearly enriched. It consists of seven transcription factor genes (ACE2, PRP45, OAF1, GTS1, SWI5, MSN1 and STB1) involved in various cellular functions, like fatty acid oxidation, stress response and cell cycle progression 262728293031. A large number of known targets of these transcription factors were also present in the down-regulated clusters (Additional file 2). Most of the remaining 1194 down-regulated genes were associated to metabolism and energy generation. In addition, a large number of genes involved in protein degradation (97 genes in total) were down-regulated, indicating a decreased requirement for proteolytic activity. Interestingly, 73 genes involved in stress response were down-regulated, including 18 related to oxidative stress response. This observation suggests that anaerobicity per se does not evoke an immediate stress for yeast.

As expected, the initial response to fully fermentative conditions showed quite some overlap with published datasets for glucose pulses to aerobic cultures 1432, including induction of the translational machinery and repression of the respiratory chain 333435. With this study, we aimed to go beyond the primary response to see how yeast adjusted to its altered growth environment.

We did not identify genes whose transcript levels continuously increased or decreased in the 2 h following the perturbation. At 30 min after the shift, a pivotal point appeared to be reached at which the transcript profiles either indicated a reverse regulation mode (clusters A, B, 4, 5 and 6) or a stable mRNA level (clusters C, 2 and 3). In this respect, a striking difference was observed between the transcript profiles of genes encoding ribosomal proteins and those encoding ribosomal biogenesis components (Fig. 5). A steady transcript level after 30 min of ribosomal proteins was indicative for a constitutive requirement for translational building blocks to support faster growth. In contrast, transcriptional up-regulated genes involved in the synthesis, processing and modification of the translational machinery appeared only to be temporarily required for a rapid adaptation to the new environmental conditions. In addition to the ribosomal protein genes, genes involved in de novo purine biosynthesis, methionine metabolism, and tetrahydrofolate-dependent C₁ metabolism were continuously transcribed at an elevated level after 30 min. All three functional categories have previously been correlated with each other, and with a response to the decrease in the adenine nucleotide pool 14.

Also the initially down-regulated genes with a turning point after 30 min could be divided in two groups: steady pattern after 30 min (clusters 2 & 3) or again up-regulated after 30 min (clusters 4, 5 & 6). During the course of the experiment, the glucose concentration remained high and hence, functional categories known to be repressed by glucose were enriched among the clusters in which the transcript level remained low after 30 min. Regulatory factors involved in regulation of the respiratory chain (HAP2, HAP4 and HAP5) were down-regulated together with their targets 36. Stress-response genes also maintained low transcript level during the experiment. In contrast, transcripts of genes involved in lipid biosynthesis, reserve carbohydrate metabolism and protein degradation tended to increase again after 30 min. The large and coordinated transcriptional up-regulation of the translational machinery, specifically ribosomal proteins, complemented an opposite transcriptional regulation pattern of genes related to proteolytic activity. The down-regulation of target genes of the Mbp1/Swi6 complex, involved in G1 to S transition 37, correlated with a delay in cell cycle progression and correspondingly, a constant cell number over the two h monitored.

Many genes involved in the metabolism of storage carbohydrates (trehalose and glycogen) showed a decreased transcript level after the perturbation. To further investigate the observed changes in trehalose and glycogen metabolism, intracellular levels of trehalose and glycogen were measured. Both reserve carbohydrates were completely degraded within 30 min (Fig. 2D), consistent with a post-transcriptional activation of trehalose and glycogen phosphorylases 3839. Physiological interpretation of the trehalose and glycogen degradation however, is less straightforward, since trehalose and glycogen are known to be involved in flux regulation, stress response and cell cycle 39.

Eighty-seven of the 1923 genes that showed a significantly altered transcript level after the combined glucose pulse and oxygen depletion only showed an increased transcript level after 30 min (cluster D). One of the few functional categories enriched within this group involved modification by glycosylation (ALG7, GNT1, MNT4, OST5, PMT2, PMT4, PMT5, SEC53 and SWP1). PMT2, PMT4 and PMT5 are specifically involved in O-linked mannosyl glycosylation, which is indispensable for cell wall integrity 40. In addition, this 'delayed response' cluster contained five of the nine genes encoding anaerobically induced mannoproteins (DAN1, DAN4, TIR1, TIR2 and TIR4) 41. Two other anaerobically induced mannoproteins (DAN2 and DAN3) were initially down-regulated, whereas transcript levels of the gene encoding the major cell wall mannoprotein (TIP1) did not significantly change at all.

A strongly anaerobiosis-related character of the genes in cluster D was not only suggested by the presence of the abovementioned genes involved in cell wall maintenance, but additionally by the presence of several genes involved in lipid transport (AUS1, FAA4 and DNF2), heme biosynthesis (HEM13; Rox1p repressed), sterol metabolism and regulation (ARE1, HES1 and NCP1), and cell wall biosynthesis (EXG2). Accordingly, a high number of genes contained AR1 elements in their promoter (Table 2), indicating a role of Upc2p 942. The delayed up-regulation of these 'anaerobic genes' indicated that the response to anaerobiosis is slow compared to the fast response to the relief from glucose limitation (clusters A, B and C).

The response to the anaerobic shift described in this study was compared with a dataset from a previously published study 2021, in which the transcriptional response of batch cultures was monitored for several generations after a shift from aerobic to anaerobic conditions. Surprisingly, only 51 genes were overlapping with the significant up-regulated genes of our study. Half of these resided in our delayed response cluster D, which contains many anaerobiosis-related genes. The absence of a glucose pulse in the study of Lai et al. 2021 explains the absence of genes encoding components of the translational machinery among the up-regulated genes in their dataset. Similarly, the large group of genes related to protein degradation found in the present study was not observed among the down-regulated genes identified by Lai et al. 2021. A strong overlap (464 genes) was found between the down-regulated genes identified in the two studies. Most of this overlap resided in the constitutively low expressed clusters 2 & 3 of our study (45% of the genes overlapped), which include many genes related to oxidative stress response. The majority of genes within the functional category Stress Response responded slower in the anaerobic shift study of Lai et al. 2021 than in our study which included a step-up of the glucose concentration. Hence, we conclude that the observed regulation of stress response correlated with the relief from growth limitation rather than with a mere depletion of oxygen.

In an attempt to identify robust 'signature transcripts' that show a consistent response to anaerobiosis, the set of significantly responding genes in this dynamic study was compared with several datasets from glucose pulses and aerobic-to-anaerobic shift experiments (Fig. 6)142032. 457 genes of the 607 genes up-regulated in this study were previously identified in two other glucose-induced studies. Sixty-seven of the 150 non-overlapping genes resided within the delayed response of cluster D, indicating that more than 70% of the genes within cluster D were not responding to glucose. Twenty of these 150 genes are also up-regulated in the aerobic-to-anaerobic shift study of Lai et al. and can therefore been seen as specifically anaerobiosis-responsive (Fig. 6). These 20 anaerobic genes are involved in cell wall maintenance (DAN/TIR genes and related glycolysations), in membrane composition (WSC4, DAL5 and FET4) and metabolism (HEM13, MET13, ARE1, AUS1 and NCP1). Interestingly, 17 of those 20 genes resided in cluster D, and 12 genes contained the Upc2-binding promoter element (TCGTTTA), which earlier was associated with about 1/3 of anaerobic genes 9. Transcription factor Upc2 has been reported to be strictly regulated by heme and sterol levels 43. The delayed response of the anaerobic genes was likely due to the almost complete absence of growth during the experiment, thus sterol levels may not have been depleted rapidly through dilution over newly formed cells. A similar comparison with the down-regulated genes in our study resulted in 46 commonly responding genes (Fig. 6). The majority of the anaerobic down-regulated genes had functions related to mitochondrial function or oxidative stress response, of which 13 genes contained a Hap1 or Hap2/3/4/5 binding element 3644. Heme levels are likely to respond rapidly to the depletion of oxygen from the culture, consistent with the fast response of the anaerobically down-regulated genes.

The use of transcripts as a diagnostic tool for biotechnological applications has been proposed previously 81145. Based on steady-state cultivation experiments, a consistent response to anaerobiosis had been determined by analyzing aerobic and anaerobic chemostat cultures grown under different nutrient limitations (carbon-, nitrogen-, phosphorus-, and sulfur limitation), resulting in 65 anaerobically up-regulated and 90 down-regulated genes 11. Surprisingly, only 14 up-regulated genes and 11 down-regulated genes were also found in both our dynamic study and the dynamic study of Lai et al.. Therefore, these can be seen as true "signature" transcripts for anaerobicity both within dynamic and steady state conditions (Table 3).

The S. cerevisiae strain used in this study was a prototrophic haploid reference strain CEN.PK113-7D (MATa) 46. Stock cultures were grown at 30°C in shake flasks containing 100 ml of synthetic medium with 20 g of glucose per liter.

The synthetic medium contained per liter of demineralized water 5 g of (NH₄)₂SO₄, 3 g of KH₂PO₄, 0.5 g of MgSO₄·7H₂O, 0.15 ml of silicon antifoam (BDH), and trace element concentrations according to Verduyn et al. 47. After heat sterilization of the medium for 20 min at 120°C, a filter-sterilized vitamin solution 47 was added. The concentration of glucose in the reservoir medium was 7.5 g·l^-1. This glucose was added to the synthetic medium after separate heat sterilization at 110°C.

CEN.PK113-7D (MATa) was grown at 30°C in 2-l bioreactors (Applikon) with a working volume of 1.5 l via an electrical level sensor. Removal of effluent from the center of the culture ensured that biomass concentrations in the effluent line differed by less than 1% from those in the culture 48. The dilution rate was set at 0.10 h^-1. The pH was measured on-line and kept constant at 5.0 by the automatic addition of 2 M KOH using an Applikon ADI 1030 Biocontroller. A stirrer speed of 800 rpm and air flow of 0.75 liter·min^-1 were applied to keep the dissolved-oxygen concentration, as measured with an oxygen electrode, above 60% of air saturation in all chemostat cultivations performed. Steady-state samples were taken after ~10 volume changes to avoid strain adaptation due to long-term cultivation 4950. Biomass dry weight, metabolite, dissolved oxygen, and gas profiles were constant over at least three volume changes.

Anaerobic glucose-pulse experiments were started by sparging the medium reservoir of the fermentor of a steady-state glucose-limited aerobic chemostat culture (airflow of 0.5 liter·min^-1) with pure nitrogen gas (Hoek-Loos, Schiedam, <5 ppm O₂). Norprene™ tubing and butyl septa were used to minimize oxygen diffusion into the anaerobic cultures 51. Two min after nitrogen sparging and just before adding the glucose, the medium-supply and effluent-removal pump was switched off. The 200 mM (60 g of glucose monohydrate in 60 ml water) glucose pulse was injected aseptically through a rubber septum. Samples were taken 5, 10, 30, 60 and 120 min following glucose addition.

The exhaust gas was cooled by a condenser connected to a cryostat set at 2°C and dried with a Permapure™ dryer (Inacom Instruments) before analysis of the O₂ and CO₂ concentrations with a Rosemount NGA 2000 analyzer. The gas flow rate was determined with an Ion Science Saga digital flow meter.

Acetate, ethanol, glycerol, and glucose concentrations in supernatants were determined by HPLC analysis with a Bio-Rad Aminex HPX-87H column at 60°C. The column was eluted with 5 mM sulfuric acid at a flow rate of 0.6 ml min^-1. Acetate was detected by a Waters 2487 dual-wavelength absorbance detector at 214 nm. Glucose, ethanol and glycerol were detected by a Waters 2410 refractive index detector.

Culture dry weights were determined as described in 52 while whole cell protein determination was carried out as described in 53. Cell numbers were counted by a Coulter counter (Multisizer II; Beckman Coulter) by using a 50 μm aperture.

Trehalose and glycogen concentration measurements were performed as described previously 54 in duplicate measurements on two independent replicate cultures. Glucose was determined using the UV-method based on Roche kit no. 0716251.

Samples were collected during the pulse, washed three times with cold 5% trichloroacetic acid and the pellet is stored at -20°C. The samples were resuspended in 3% perchloric acid and heated at 90°C for 30 min. After centrifugation, the supernatant was mixed with 37% hydrochloric acid, containing 10 g l^-1 orcinol monohydrate (crystalline, Sigma-Aldrich, Germany) and 5 g l^-1 iron(III) chloride hexahydrate. The mixture was heated at 90°C for 20 min before measuring absorbance at 660 nm 55. Absorbance values were related to a concentration (expressed as μg·ml^-1) using a calibration curve of a standard yeast RNA solution (Sigma-Aldrich, Germany).

Sampling of cells from chemostats, probe preparation, and hybridization to Affymetrix Genechip^® microarrays were performed as described previously 10. The results for each time point after the perturbation (5, 10, 30, 60 and 120 min) were derived from two independently cultured replicates, while steady state data were derived from three independent chemostats. The complete dataset therefore comprised 13 arrays.

Acquisition and quantification of array images and data filtering were performed using Affymetrix GeneChip^® Operating Software version 1.2. Before comparison, all arrays were globally scaled to a target value of 150 using the average signal from all gene features. To eliminate insignificant variations, genes with expression values below 12 were set to 12 and genes for which maximum expression was 20 over the 13 arrays were discarded. From the 9335 transcript features on the YG-S98 arrays, a filter was applied to extract 6383 yeast open reading frames, as previously described 8. To represent the variation in the measurements, the coefficient of variation was calculated as the mean deviation divided by the mean 8. The array data used in this study can be retrieved at Genome Expression Omnibus 56 with series number GSE8187.

For additional statistical analyses, Microsoft Excel running the EDGE (version 1.1.208) add-in was used 17 for a time course differential expression analysis. To determine the genes called significantly changed according to EDGE a p-value of 0.005 was used. K-means clustering of the genes with significantly changed expression levels was subsequently performed using Genedata Expressionist^® Pro (version 3.1). The k-means algorithm used positive correlation as distance metric. The maximum number of iterations was set to 1000. Initially, the algorithm was run with k equal to 2, dividing the genes into an up- and a down-regulated cluster. Each cluster was then clustered again using k-means with k ranging from 2 to 10. The optimal k-value, i.e. 4 for the initially up-regulated and 6 for initially down-regulated genes, were based on the explained variance between clusters and the overrepresentation of functional categories (for detailed explanation please refer to Additional file 3).

Each cluster was consulted for enrichment in functional annotation and significant transcription factor (TF) binding (experimentally identified by Harbison et al. 57) as described previously 58. In addition, specific TF binding sites not present in the Harbison dataset were analyzed by using web-based Regulatory Sequence Analysis Tools 1159.