Global transcriptional response after exposure of fission yeast cells to ultraviolet light

The two experiments

Two distinct experiments were performed (see also Additional file 1): First, G1-phase cells were irradiated and kept at the restrictive temperature (hereafter called the "restrictive-temperature experiment"). Thirty minutes after the time of irradiation the UVC-irradiated (UV30) and unirradiated control (C30) cells were collected. These cells were still in early G1 phase due to the continued inactivation of Cdc10. Second, G1-cells were released into the cell cycle after synchronisation by reducing the temperature to 25°C, thus reactivating Cdc10, allowing the cells to continue in the cell cycle. Cell samples were collected at 0, 30 and 90 min after irradiation (hereafter called the "time-course experiment"). Samples of irradiated and unirradiated cells were analyzed on microarrays and compared to a common reference pool (see Methods). Flow cytometry (Additional file 2) demonstrated that the control cells had entered S phase by 60 minutes after release, whereas the UVC-irradiated cells delayed in G1 phase and moved from G1 to S phase around 90 minutes after release, in agreement with previous data [13].

We have previously shown the existence of the G1/S checkpoint using several synchronisation methods [14], but none of the other methods provided good enough synchrony to perform similar analyses on global transcription. We have looked at our data for all known Cdc10 targets in the 12 arrays from the time-course experiment and observed no trend showing that UVC delays the occurrence of these transcripts. This is consistent with our RNA blots of the two selected transcripts cdc18 and cig2, which are not delayed by UVC in block-and-release experiments [Fig. 4D in [14]].

The restrictive-temperature experiment

In this experiment we searched for genes that altered their expression more than twofold as a consequence of UVC irradiation in G1 phase. Of almost 5000 genes represented on the microarrays as few as 74 genes were induced twofold or more and 43 of these were non-CESR genes (Fig. 1A). Most of the 74 genes were induced two- to threefold, and only three genes were induced more than fivefold. Most of the induced non-CESR genes are likely to be UVC-specific and not cell-cycle related, since the cells did not move into S phase during the time of the experiment. No non-CESR genes were found to be induced by both UVC and H₂O₂ [3], but two genes, SPCC132.04c and SPBC16A3.17c were induced after both UVC (this work) and IR treatment [4] (see Table 1). We categorised the 43 non-CESR genes into eight different groups according to the functions of their products (Table 1). These genes are involved in a variety of functions such as signalling and stress response, ribosome biogenesis and translation, DNA/RNA binding, and as many as 15 of the genes are involved in transport mechanisms. There were no obvious candidates for genes involved in the G1/S checkpoint amongst the 74 upregulated genes. Surprisingly, only one of the induced genes, rhp4b, encoding a nucleotide excision repair factor, is involved in DNA repair of UVC-induced lesions, strongly suggesting that the capacity to perform DNA repair is not regulated at the transcriptional level. Another possibly interesting induced gene is pyp1, a protein-tyrosine phosphatase that acts on Sty1, the MAPK (mitogen-activated protein kinase) that regulates various stress responses in S. pombe [15]. We have shown that the G1/S checkpoint does not require Sty1 [13], but it is possible that Pyp1 has additional targets in the cell besides Sty1.

We also identified 44 genes that were repressed more than twofold and almost all of them (40) were non-CESR genes (Fig. 1B). These 40 genes were categorised according to the function of their products (Additional file 3), and, like for the proteins encoded by the induced genes, were known to be involved in a wide variety of activities including transport, metabolism, mating and mitochondrial activities. Amongst the repressed genes no obvious candidate genes for checkpoint regulators were found.

The time-course experiment

Furthermore, we searched for genes differentially expressed between 0/30, 30/90 and 0/90 minutes in both the C and UV samples. To this end we used moderated t-statistics with a P-value cut-off of 0.05 (for details see Methods). In these experiments each gene could potentially be assigned 12 expression values (three time points, C and UV, two repeats). About 35% of the values were missing in the entire dataset. The reasons for the missing values will be discussed below. We decided to remove the data for a gene if four or more of the 12 possible data values were missing. This action reduced the dataset from 5266 to 2836 genes, and data for the resulting 54% of the genes was considered more reliable and was used for further analysis.

Cell-cycle-regulated genes

The RNA samples from irradiated cells reflect gene expression changes that occur for two separate reasons: first, the cells are progressing through the cell cycle and will necessarily change gene expression [16] and, second, the cells have been exposed to UVC light and will display stress-related and UVC-specific changes. In contrast, differences between the samples in the restrictive-temperature experiment (above) should reflect only the stress-related and UVC-specific changes. In the present analysis, we attempt to identify the genes specifically affected by UVC exposure. For this reason we identified the genes in unirradiated control cells whose expression varied after release into the cell cycle and these genes were classified as not specifically responding to UVC. In the unirradiated cells altogether 143 genes were found to be differentially expressed between G1 phase (C0 or C30) and S/G2 phase (C90) (Fig. 2A and Table 2). Only 4 genes (3+1) were differentially expressed at 30 minutes after release into the cell cycle (C0 compared to C30). In comparing C30 (late G1 phase) with C90 (S/G2 phase) 47 genes (32+14+1) were found to be differentially expressed. Finally, 129 genes (93+32+1) had a changed transcriptional level when comparing C0 (early G1 phase) to C90 (S/G2 phase). This means that, not surprisingly, most of the induced genes were found in cells entering S phase (the comparison of C0 with C90 and C30 with C90). Well-known G1-specific genes, like cdc18, cig2, cdc22 and cdt2, were induced normally during G1 phase (C30) in our study (data not shown), serving as convenient controls.

UVC-regulated genes

Using the P-value cut-off of 0.05, we identified as many as 241 genes in total that were differentially expressed in the UVC-irradiated cells after release into the cell cycle (Fig. 2B and Table 2). Of all the 241 genes only 11 were different between UV0 and UV30, so little was happening at the transcriptional level during the first 30 minutes after irradiation. No genes were determined to be differentially expressed when comparing UV30 (late G1 phase) and UV90 (S phase). However, as many as 230 genes were defined to be different between UV0 (early G1 phase) and UV90 (S phase). The apparent discrepancy between the numbers of regulated genes between UV0/UV30 and UV30/UV90 on the one hand and between UV0/UV90 on the other can be explained as follows: Expression of a number of genes is different between UV0 and UV30, but only 11 genes were significantly different (P < 0.05). Similarly, a number of genes changed their expression between UV30 and UV90, but no difference passed the threshold we had set. However, when comparing UV0 and UV90, a number of genes (230) had altered their expression sufficiently during the total interval. It follows that the level of change for all of these genes was low.

Profile analyses

To further analyse the kinetics of gene expression in the time-course experiment we applied the software package maSigPro [17] on the filtered data, which allowed a multiple comparison of all three time points. There were 133 genes that significantly changed their expression levels during the time-course (P ≤ 0.05), and the expression of almost all (105 genes) was changed in both control and UVC-irradiated cells. Fifteen genes were identified as specific for the irradiated cells. There was a good correlation between the genes identified in this profile analysis and the differentially expressed genes determined in the statistical analysis above (Fig. 2C).

The expression levels of the 133 genes with expression that varied during the time-course were subjected to a clustering analysis, forming a two-dimensional map (Fig. 3). Genes being up- or downregulated at the particular time point are represented by different colours, green indicating induction and blue repression. The map shows that analysis of the two biological replicates revealed that they were indeed similar and assembled into the closest branches of the cluster. This confirms a good reproducibility of the experiments. Furthermore, the profile analysis is also consistent with the conclusions of the pairwise statistical analyses (above), showing that the expression patterns in UV30 and UV90 were similar. The map shows little difference between control and irradiated samples at all time points, and the least closely related branches of the cluster refer to changes occurring during progression in the cell cycle. Thus, this analysis corroborates our conclusions from the above pairwise comparisons, that the cell cycle progression affects transcription profiles more than the UVC treatment and there is only a weak transcriptional response to UVC irradiation.

Pathway analyses

Gene ontology (GO) enrichment analysis was performed on the 43 UVC-induced genes in the restrictive-temperature experiment (Table 1) using the DAVID software after ID-conversion. Ten unique genes (20%) were members of enriched GO groups, suggesting upregulation of genes involved in ion transport or ion homeostasis in the restrictive-temperature experiment (Additional file 5). Analysis of over-represented GO annotations amongst the 172 upregulated genes in the time-course experiment (Table 2) showed that 38 unique genes (22%) are members of the enriched GO groups. The nature of these groups indicates regulation of genes that affect protein biosynthesis or structural components of ribosomes (Table 4). Further network analysis shows a concerted response involving direct protein-protein interactions between 20 of the 172 induced gene products. Thus, the analyses suggest a coordinated induction of rRNA biogenesis, ribosome assembly and components of the 60S and 40S ribosomal subunits in the time-course experiment (Fig. 4). (Additional file 7)

Category	GO number	Term	Count*	%	P-value
GOTERM_Cellular Component	GO:0005887	integral to plasma membrane	3	8.11	0,024
	GO:0044459	plasma membrane part	4	10.81	0,035
	GO:0031226	intrinsic to plasma membrane	3	8.11	0,038
GOTERM_Biological Process	GO:0015674	di-, tri-valent inorganic cation transport	4	10.81	0,001
	GO:0006812	cation transport	5	13.51	0,007
	GO:0030001	metal ion transport	4	10.81	0,008
	GO:0030003	cellular cation homeostasis	4	10.81	0,018
	GO:0055082	cellular chemical homeostasis	4	10.81	0,020
	GO:0050801	ion homeostasis	4	10.81	0,022
GOTERM_Molecular Function	GO:0008324	cation transmembrane transporter activity	5	13.51	0,011
	GO:0046873	metal ion transmembrane transporter activity	3	8.11	0,042

* Count denotes the number of genes in our dataset in each cluster. Percent coverage of these genes relative to the numbers on the gene ontology clusters were calculated.

Confirmation of the microarray data

RNA blotting and hybridisation was used to confirm our microarray results for four selected transcripts. The induction of SPAC2E1P3.05c, fip1 and gst2 found in the time-course experiment was verified (Fig. 5A). Fip1, an iron permease, and Gst2, a glutathione S-transferase, have also been shown to be induced after IR [4]. On the other hand, rhp4b had no values in any of the 12 arrays in our time-course experiments and was also not detected after RNA blotting and hybridisation. However, in the restrictive-temperature experiment, rhp4b transcription was found to be induced by UVC irradiation and this finding was also confirmed by RNA blotting (Fig. 5B). Therefore, RNA-blotting experiments with all four selected genes verified the results from the microarray experiments.

UVC does not induce the unfolded protein response

The unfolded protein response (UPR) is activated by the accumulation of unfolded proteins in the endoplasmic reticulum (ER). The UPR triggers a transcriptional response which serves to induce the production of ER components and to increase the degradation capacity to dispose of the unfolded proteins [18]. It is possible that UVC irradiation might stress the ER and thus activate the UPR. Furthermore, GCN2 has been shown to be required for the induction of a majority of UPR target genes during ER stress in S. cerevisiae [19]. We therefore used the present data from both the restrictive-temperature and time-course experiments to investigate whether UPR genes are induced by UVC in fission yeast. We identified UPR-genes in fission yeast as the homologues of the UPR-induced budding yeast genes [20]. There was little, if any, induction of the UPR-genes identified by this method (Additional file 6). The lack of transcriptional response of these genes after UVC strongly argues that the G1/S checkpoint is not a manifestation of the UPR.

Comparison of the data from the two experiments

162 genes were identified as specifically UVC-regulated in the time-course experiment and 43 in the restrictive-temperature experiment. Surprisingly, as few as 4 genes (see Additional file 4) were found to be common for the two datasets, and these 4 genes were not differently expressed in UV0 and UV30 in the time-course experiment (Table 3), which is the time period when the irradiated cells were arrested in the G1/S checkpoint. The UVC irradiation elicits a quite weak transcriptional response on the cells both when considering the number of genes affected and the level of the response for the affected genes. Thus, our assay must be considered to be rather sensitive and even a small change in the experimental setup might bring the marginal levels of gene expression over or under our threshold, which could be one reason for the poor overlap between the two experiments. This further underlines our conclusion that UVC has only a marginal effect on gene transcription when given in G1 phase. It is possible that the biological differences between the cells in our two types of experiments is dominating and that the dissimilar sets of regulated genes reflect a biological difference rather than an artefact of our data analysis. In the restrictive-temperature experiment cells were arrested in G1 phase by the inactivation of Cdc10, which means that cell-cycle dependent responses were inhibited. Thus, this experiment reveals transcriptional regulation exclusively due to UVC irradiation, at a specific stage in the cell cycle. In contrast, in the time-course experiment cells were allowed to progress into the cell cycle. Even though we have subtracted the cell-cycle-regulated genes when identifying the UVC-regulated genes, the cell cycle stage in which the cellular response is analysed was still different from that of cells held in the cdc10 block. Consistent with this line of reasoning, we have shown that little happened at the transcriptional level during the first 30 minutes after irradiation in the time-course experiment (only 11 regulated genes when comparing UV0 to UV30), and most of the 172 genes changed their transcription enough to satisfy the criteria of our analysis only by a later time point, as the cells entered S phase. Therefore these genes have not been identified as regulated in the restrictive-temperature experiment.

Missing values

In the data for the time-course experiment values were missing for many data points. This can be attributed to at least two reasons: lack of expression of the relevant open reading frames and technical problems with the microarrays. At any one time there are genes that are poorly expressed, so it was expected that there would be some missing values in our datasets. Low expression is the most likely reason for the absent microarray signal for the rhp4b transcript, which could also not be detected after RNA blotting. However, lack of expression cannot explain all the missing values. The 12 microarrays used in the time-course experiment comprised from 18% to 55% missing values (on average 35%), arguing that there were technical problems with at least some of the microarrays. Such a high level and difference in the number of missing values was not found for the two repeats of the restrictive-temperature experiment, which both had about 6% missing values, further suggesting technical problems with the arrays used in the time-course experiment. It should be noted that the microarrays used for this experiment came from a different batch/production than those used for the restrictive-temperature experiment. This problem did reduce the quality of the data for some of the genes in the time-course experiment, and we decided to remove all the data pertaining to genes where too many data points were missing (detailed in Results). However, the stringency of our analysis allows us to draw conclusions in spite of the missing data. Furthermore, any technical problem would have affected a random set of genes, and a strong transcriptional response would have been obvious even from the time-course experiment.

Comparisons with other organisms

This is the first report about the global transcriptional response after UVC irradiation in fission S. pombe and there are only a few reports about similar experiments in other organisms. The available data indicate that the weak transcriptional response to UVC stress we observed in fission yeast might be a conserved feature. For example, in human cells exposed to UVC only 155 of more than 7500 genes investigated changed their expression more than 2-fold [21]. An early microarray-report using Escherichia coli cells identified several differentially expressed genes after UVC-irradiation, but the response was generally not more than two-fold [22].

Many checkpoints, both in fission yeast and in other organisms, involve transcriptional regulation. For instance, in multicellular organisms one of the best characterised checkpoint targets is p53, a transcription factor which is mutated in over half of human cancers. p53 stimulates transcription of cell-cycle inhibitors such as p21 [23] and is essential for a persistent G1 arrest. Another tumour suppressor, pRb, targets the E2F-driven transcriptional programme in the G1/S checkpoint [23, 24]. In budding yeast activation of the G1/S checkpoint also impinges on transcriptional regulation in that the transcription factor SWI6 is phosphorylated and thereby inactivated by the checkpoint protein RAD53, leading to delayed transcription of CLN1 and CLN2 and a delayed entry into S phase [25]. Recent data show that in fission yeast the transcriptional response is important in the intra-S checkpoint after exposure to hydroxyurea [10, 26].

Comparisons with other stress agents

We have observed no obvious change of the transcriptional programme that could be responsible for the G1/S checkpoint and also no strong induction of genes involved in DNA repair or checkpoint function. It is interesting to note that other DNA-damaging agents, such as H₂O₂ and IR, also do not lead totranscriptional induction of many DNA repair- or checkpoint-related genes [3, 4]. Therefore, it is likely that there is no need to specifically induce transcription of DNA repair-related genes, suggesting that the DNA repair capacity is high already before the UVC exposure, as the case is for budding yeast [27].

We have compared our data from synchronised cells to data based on H₂O₂- and IR-treated asynchronous cells [3, 4], and there is little overlap in the spectrum of non-CESR genes differentially expressed after exposure to the three agents. As discussed above, it seems that the cell-cycle position is important for the transcriptional profile obtained when exposing the cells to a stress treatment. However, when comparing the differentially expressed genes in asynchronously growing cells exposed to UVC (our unpublished observations) to cells exposed to H₂O₂ [3] and IR [4] there is little overlap.

Pathway analyses

DNA-damaging agents, heat and other forms of stress give overlapping responses, described as the CESR, that involves 14% of the genome in S. cerevisiae [5]. Few UVC-specific expression changes have been reported [21] and the fold-changes observed are low [28]. It is therefore possible that transcriptional responses to heat stress combined with stringent statistical analyses and exclusion of the CESR-genes mask UVC-relevant gene expression changes in the restrictive-temperature experiment. Similarly, the absence of significant gene-expression changes after 30 minutes recovery in the time-course experiment indicates that UVC-relevant transcriptional responses are masked by the small fold-changes and extensive overlap between the UVC-specific response and the CESR. This is supported by appearance of differences on the transcriptional level first after 90 minutes. Our network analyses of gene-expression changes indicate that pathways involved in recovery after DNA damage are induced after 90 minutes (Fig. 4). The processes involved probably reflect the importance of the translation machinery in recovery after the insult. These recovery processes are likely to be important for survival after UVC treatment, as was shown in MMS-treated S. cerevisiae [29], although they may not be components of the G1/S checkpoint per se.

Regulation of the G1/S checkpoint

In the G1/S checkpoint the Gcn2 kinase is activated to phosphorylate the eukaryotic initiation factor 2α, eIF2α, thereby inhibiting translation [13]. There is a good correlation between eIF2α phosphorylation and checkpoint activation [30], but it is still unclear whether and, if so, how the checkpoint is dependent upon this phosphorylation and on the ensuing downregulation of translation. Gcn2 is best known for its role in the starvation response, where eIF2α phosphorylation leads to induction of the transcription factor GCN4 both in budding yeast and higher eukaryotes. Fission yeast does not have a GCN4 homologue, and it remains to be seen whether the GCN2-dependent G1/S checkpoint in budding yeast [31] involves activation of GCN4. One might expect a transcriptional response after Gcn2 activation also in S. pombe, but the finding that no such transcriptional response could be identified argues that in fission yeast the G1/S checkpoint does not operate like it does in budding yeast or higher eukaryotes and is not dependent upon the strong transcriptional induction of one or a set of genes. We have shown that the G1/S checkpoint in S. pombe is associated with a strong downregulation of translation and it might be relevant that among the few genes that are affected at the transcription level, 15% affect the translation machinery. Importantly, amongst the transcriptionally regulated genes none were detected that are likely to be directly involved in the G1/S checkpoint according to their annotations.

Induction or inhibition of transcription is a fairly slow response and regulation of the cell cycle in S. pombe should preferably occur rapidly in order to be efficient and meaningful. Therefore, it intuitively makes sense that the present data suggest that the G1/S checkpoint is regulated at the level of protein modification and/or translation, which is rapid, rather than a slower regulation of gene expression.

Yeast strains and cell growth

The cdc10-M17 strain is a derivative of the L972 strain [32]. The basic growth media were as described [33]. The temperature-sensitive cdc10 cells were grown exponentially in EMM to an optical density (595 nm) of 0.15 (about 3 × 10⁶ cells/ml), before they were synchronised by a four-hour temperature shift to 36°C and irradiated with 1100 J/m² UVC (254 nm), giving a cell survival of ~15% [14]. Samples of 25 ml of control or UVC-irradiated cells were harvested at different time points by centrifugation and snap-frozen in liquid nitrogen.

RNA isolation

Total RNA was extracted using a hot phenol method [34]. RNA concentrations and qualities were measured in a NanoDrop (NanoDrop Technologies) and in a 2100 Bioanalyzer (Agilent Technologies).

Microarray hybridization and data acquisition

Total RNA was reverse transcribed (GibcoBRL) in the presence of Cy3- or Cy5-labelled dCTP. The cDNA was hybridised onto glass DNA microarrays containing duplicate probes for 99.3% of all known and predicted open reading frames in the fission yeast genome (for details on protocols and microarrays, see Lyne et al. 2003 and http://www.sanger.ac.uk/PostGenomics/S_pombe). A GenePix 4000 B laser scanner was used for scanning the microarrays before analysis with GenePix Pro software (Axon Instruments). A Perl script was used for removing unreliable signals and normalization of the data [34]. All raw data are available under accession number A-MEXP-1666 and A-MEXP-1667 from ArrayExpress.

Experimental design

Two different types of experiment were performed (see Results) and for both types the RNA from two biological repeats were analysed with a dye swap. Samples from all three time points of the "time-course experiment" were hybridised individually against a reference pool containing equal amounts of RNA from the unirradiated cells, at all the three time points (0, 30 and 90 min). After normalisation (for details see Lyne et al. 2003), the ratio of the values for the actual sample and for the reference pool for each gene was divided by the corresponding ratio for untreated cells at time 0 (0 min control/reference pool). Samples from UVC-irradiated cells kept at 36°C, the "restrictive temperature experiment", were hybridized to the arrays against RNA from unirradiated cells.

Pathway analyses

Genes found differentially expressed in the time-course experiment (Table 4) were analyzed for gene ontology-enriched clusters using DAVID (Database for Annotation, Visualization and Integrated Discovery) http://niaid.abcc.ncifcrf.gov[35, 36]. As Schizosaccharomyces pombe gene names are not recognized, the gene names were converted to UniProt accession numbers using the YOGY (eukarYotic OrtholoGY) software http://www.bahlerlab.info/YOGY/[37]. The YOGY software was also used to find Saccharomyces cerevisiae orthologues for use in the pathway analyses. When several orthologues were found, the S. pombe sequence was used as template for a BLAST search and the best hit in S. cerevisiae used. As there are very few resources available to investigate functional interactions in S. pombe, the S. cerevisiae orthologues were used to map protein-protein interactions. These interactions were processed using FunCoup (networks of functional coupling) http://funcoup.sbc.su.se/[38].

Data evaluation

In both types of experiment genes were classified as differentially expressed when expression values were changed more than twofold (linear values) in both of the two biological repeats. Before statistically analysing the time-course experiment, a filtering of the data was performed. This filtering excluded all genes in the dataset that did not have a value in two-thirds of the arrays (i.e. missing more than 3 out of 12 values). The normalised expression values (see paragraph above) for the remaining genes were transformed from linear values to log2 values. Moderated t-statistics with a P-value cut-off of 0.05 was used to identify genes differently expressed during the time-course [39]. Benjamini and Hochberg's method was used to calculate adjusted P-values and to statistically correct for the occurrence of false positives [40]. The statistical analysis was performed using the programme R and Bioconductor [41]. The Bioconductor package maSigPro was used to perform profile analysis of the time-course experiment to show how gene expression changed with time. Benjamini and Hochberg's method was also used in the profile analysis and the P-value cut-off was set to 0.05.

RNA blots

Total RNA was run on agarose gels in formaldehyde, blotted onto Hybond-XL membranes (Amersham Bioscience) and cross-linked by UVC. Probes were prepared by PCR of genomic DNA and labelled with ³²P-d-CTP (Rediprime II Random prime labelling system, Amersham Bioscience). Phosphoimager screens were exposed to the washed blots and analyzed by a Pharos FX scanner (BioRad).