Yeast glucose pathways converge on the transcriptional regulation of trehalose biosynthesis

Background Cellular glucose availability is crucial for the functioning of most biological processes. Our understanding of the glucose regulatory system has been greatly advanced by studying the model organism Saccharomyces cerevisiae, but many aspects of this system remain elusive. To understand the organisation of the glucose regulatory system, we analysed 91 deletion mutants of the different glucose signalling and metabolic pathways in Saccharomyces cerevisiae using DNA microarrays. Results In general, the mutations do not induce pathway-specific transcriptional responses. Instead, one main transcriptional response is discerned, which varies in direction to mimic either a high or a low glucose response. Detailed analysis uncovers established and new relationships within and between individual pathways and their members. In contrast to signalling components, metabolic components of the glucose regulatory system are transcriptionally more frequently affected. A new network approach is applied that exposes the hierarchical organisation of the glucose regulatory system. Conclusions The tight interconnection between the different pathways of the glucose regulatory system is reflected by the main transcriptional response observed. Tps2 and Tsl1, two enzymes involved in the biosynthesis of the storage carbohydrate trehalose, are predicted to be the most downstream transcriptional components. Epistasis analysis of tps2Δ double mutants supports this prediction. Although based on transcriptional changes only, these results suggest that all changes in perceived glucose levels ultimately lead to a shift in trehalose biosynthesis.


Supplementary Experimental Procedures
All protocols related to DNA microarray expression profiling are also available from ArrayExpress (http://www.ebi.ac.uk/microarray) [1] with the accession numbers as indicated in the relevant sections.

Data availability
Raw data and normalised gene expression profiles are available from ArrayExpress extracting the files and opening the .cdt file. When viewing the data, individual strains and transcripts can be found quickly using the "Find" function.

Yeast Strains
All strains are isogenic to S288c, BY4742 [3]. The selection of glucose signaling components was manually curated from literature, resulting in 91 viable deletion strains. Haploid MATα gene deletion strains were initially obtained as two separate copies from the deletion collections Euroscarf (Frankfurt, Germany) or Open Biosystems (Huntsville, USA). Different problems, detailed below, were encountered for 16 strains in the collection. These were remade. New deletion mutants were constructed using the kanamycin cassette from pFA6a-kanMX6 [4]. All strains used are described in Supplementary Table 1.

Quality control on strains from the deletion collection
In 16 strains from the collection, the gene expression profiles revealed different defects, annotated in the strain list in Supplementary Table 1. Note that such defects may be common to all copies of the collection but could also have arisen due to our handling of these strains. All these strains were re-made. Three different types of defects were encountered and are described below. All 91 strains selected for expression profiling passed our quality control criteria.

Incorrect deletion
In some deletion strains the supposed deleted gene was not obviously downregulated. This is usually due to already low expression in WT. In all such cases the deletion strain was checked by two PCRs, one using two primers outside the presumed deleted gene and another using one primer outside the gene and the other primer in the middle of the marker.
PCR reactions were positive for four strains and the gene expression profiles were kept. Six deletion strains did not have the desired gene deleted and were subsequently re-made and re-profiled.

Aneuploidy
Aneuploidy is revealed in the gene expression profiles by analysis of expression changes in the context of chromosome location, one of the standard quality controls (QCs) performed on all gene expression profiles. Expression profiling revealed aneuploidy in three deletion mutants in the collection, sometimes as part of a chromosome, sometimes with one or more complete chromosomes involved. Aneuploid mutants were re-made and re-profiled.

Spurious mutations
Four deletion strains passed the QC criteria of correct deletion and no aneuploidy but had surprising gene expression profiles in light of what was previously known about the knockout gene. These were re-made and re-profiled.

Yeast growth for expression profiling
During this study the protocol for yeast growth was optimized to grow more mutants at the same day in a less labour intensive way. Two slightly different protocols for yeast growth were thus used in this study. These protocols only differed in culture volume and equipment.
Detailed comparative analysis of the same mutants grown with both protocols revealed no difference in expression. Both protocols are described below. For strains grown in the Tecan platereader, an automated method for RNA purification was used. WT inoculates, grown in parallel. Mutant and WT cells were harvested by centrifugation (4000 rpm, 3 min) at mid-log phase at an OD600 of 0.6, and pellets were immediately frozen in liquid nitrogen after removal of supernatant. No more than four cultures were harvested simultaneously to decrease processing time. Note that OD600 measurements of cells are spectrophotometer dependent. OD600 0.6 (±0.1) corresponds to early mid-log phase for these cultures. It is essential to harvest at an OD600 that corresponds to early mid-log phase for WT cultures. High resolution time-course analysis of the entire growth curve showed that early to mid-log phase represents the window during which WT gene expression profiles are identical along the growth curve. WT cultures start showing significant changes in gene expression (after mid-log phase) long before growth slows down (see also [5]). Adherence to a strict early mid-log OD600 of 0.6 window for harvesting is particularly important for (a minority of) mutants that have significantly slower growth compared to WT, since for some of these mutants, OD600 of 0.6 is further along their relative growth curve. Problems associated with this are overcome if all mutants are harvested at an OD600 that represents early mid-log for WT cultures. for external control normalisation and quality control [6].

RNA purification for growth in Erlenmeyers (ArrayExpress accession P-UMCU-37)
RNA purification was performed using Qiagen's RNAeasy kit using the following protocol.  1 µl of cleaned and normalised RNA is used to check integrity by running on a QiaXcel system.
 All plates are snap-frozen and stored at -80C until further use.

External controls
External control poly-A+ RNAs were added in equimolar amounts to the total RNA to enable monitoring of global changes in mutants [6]. Constructs containing Bacillus subtilis genes (ycxA, yceG, ybdO, ybbR, ybaS, ybaF, ybaC, yacK and yabQ) cloned between the XbaI and
All subsequent steps were performed by the following robot script:  Labelled cRNA product is purified with RNA Clean (Agencourt, GC biotech) according to manufacturers' protocol.
 RNA concentration and labelling incorporation are measured (SpectraMax 190).  Drying: blow with nitrogen for 3 minutes at 30C.

Scanning and image analysis
Slides were scanned using a G2565AA scanner (Agilent, California, USA) at 100% laser power and 30% PMT (ArrayExpress accession P-UMCU-40). After scanning, the intensities for the Cy5 (Red) and Cy3 (Green) channels were automatically extracted using the batch-

Microarray quality control
Each hybridisation performed within this project was subjected to a number of quality controls. Some of these are based on the data from one single hybridisation, while others are based on comparing data from one single hybridisation against the WT grown in parallel.

Quality controls for a single hybridisation
For each hybridisation a quality report was generated that contained a number of quality controls. For all of these quality controls either raw non-background corrected mean intensity values were used, or data normalised on all gene probes using the print-tip LOESS [7] algorithm (marray R package version 1.20.0) with a window span of 0.4 and excluding genes with nearly saturated signals (i.e. mean intensity > 215) for the loess curve estimation was used. The following plots are generated:

Quality controls for the entire project
To assess the performance of individual hybridisations within this project, hybridisations were ranked according to the percentage of good gene probes. Individual hybridisations for deletion mutants with fewer than 93% good gene probes where considered outliers and were removed. Each deletion mutant gene expression profile was compared against the gene expression profile of the WT grown in parallel to ensure that the effect that we observe in the deletion mutant is specific for the deletion mutant and is not present in the WT grown in parallel. If a significant overlap was found between the significantly regulated genes in the deletion mutant and the WT grown in parallel as determined by a hypergeometric test (p < 0.01), the hybridisations for the deletion mutant were removed. Using the quality controls used for all slides. The slide bias F was estimated for each hybridisation separately, using two groups of probes, having the strongest red or green biases, defined as those with an iGSDB in the top and bottom 5th percentiles of iGSDBs, respectively. The slide bias is the mean of the red and green probe group's median (M / iGSDB)-ratio.

Statistical analysis of gene expression profiles
For each deletion mutant the replicate hybridisations from two independent cultures were compared to the WT cultures grown on the same day in parallel and a pool of 200 WT replicates [10] grown throughout the project through a common reference. Due to a change in the shaker used for yeast growth during this project, two different WT pools were constructed. P values were obtained from the limma R package version 2.12.0 [11] after Benjamini-Hochberg FDR correction. Genes were considered significantly changed when the fold change (FC) was > 1.7 and the p value < 0.01. Transposable elements and mitochondrial genes were excluded from further downstream analysis, as well as YDL196W, since it is frequently upregulated indicating that expression of this gene was downregulated in the common reference WT culture, likely due to a spurious mutation. For determining which deletion mutants have a significant effect on mRNA expression levels, all deletion mutants and the 56 WT cultures grown in parallel were ranked by the number of significantly changing genes. No WT had twelve or more genes changing and based on this the deletion mutants were classified into two groups: profiling (≥ 12 genes changing) and non-profiling (< 12 genes changing). R (template) scripts for running the limma statistical analysis are available upon request.
For the glucose WT time-course, replicate hybridisations from two independent WT cultures at different time points were compared to the same cultures at time point 0 using the limma R package as described above. As for the deletion mutant dataset, transposable elements, mitochondrial genes, and YDL196W were excluded.