We studied wild type invasive-growth competent yeast subjected to acute glucose starvation. Cells were starved for 0, 10, 20, 30, 60, or 120 minutes before processing. For each time point, polysome profiles were generated to monitor global translation. In agreement with previous reports (Ashe et al., 2000
; Kuhn et al., 2001
), a collapse of the polysome region and concomitant increase in the 80S monosome peak occurred after 10 minutes of glucose withdrawal, indicating a bulk reduction in translation initiation. Previous investigations of the translational response to acute glucose withdrawal using either bulk measurements (Ashe et al., 2000
; Holmes et al., 2004
) or examination of a few genes (Brengues et al., 2005
) suggested that little translation initiation occurs 10–20 minutes following glucose removal. In light of prior studies of glucose-stimulated transcriptional regulation, as well as genetic evidence that many of the genes induced upon glucose depletion are required for growth in the absence of glucose, we reasoned that such transcripts must at some time associate with polysomes in glucose-starved cells (Zaman et al., 2008
). We extended the time course of the experiment to learn when bulk translation recovers as cells adapt to growth in low glucose concentrations. Polysomes remained low between 10 and 30 minutes post glucose withdrawal, and showed signs of partial recovery after 60 minutes, with further increases by 120 minutes ().
Regulation of transcription and translation in glucose-starved cells
Changes in Polysomal mRNA Levels Closely Parallel Changes in Transcript Abundance
To examine the kinetics of gene-specific translation activity following glucose withdrawal, we isolated RNA from polysome fractions and total cell lysates, and prepared fluorescently labeled cDNA for competitive hybridization on custom microarrays. In order to determine relative changes in translation activity, mRNA abundance, and the fraction of total mRNA associated with polysomes (ribosome occupancy), we performed the following comparisons for each time point after glucose withdrawal: polysomal RNA starved with polysomal RNA mock-starved (Px/P0); total RNA starved with total RNA mock-starved (Tx/T0); and polysome starved x-min with total starved x-min (PxTx), respectively (). All experiments were performed twice (biological replicates), with each RNA sample processed in duplicate (technical replicates). ‘Noisy’ genes for which mRNA abundance or polysome association varied by > 1.5-fold between biological replicate experiments in unstarved cells were omitted from further analysis (495 genes). Reproducible results were obtained for 5,590 genes.
Glucose withdrawal led to changes in the relative mRNA abundance of hundreds of genes within 10 minutes, consistent with prior studies of carbon source-mediated regulation of yeast transcription (Zaman et al., 2008
). These changes persisted and were amplified over the course of 120 minutes of starvation. Our data do not distinguish between transcriptional induction/repression and mRNA stabilization/destabilization as the mechanism responsible for changing total mRNA abundance. Both mechanisms likely contribute. Many of the genes that showed relatively increased mRNA levels following glucose withdrawal are known targets of glucose regulated transcription factors (Fig. S1
). For most genes, changes in their relative mRNA abundance in polysomes (P/P) mirrored changes in overall mRNA levels (T/T) (). Notably, for the more than 1,000 genes whose relative mRNA abundance increased by 2-fold or more after glucose withdrawal, relative polysomal abundance similarly increased. Furthermore, most induced genes did not show a noticeable lag between the increase of mRNA in the total RNA pool and appearance in the polysomal fraction. Thus, despite the reduction in the rate of ‘bulk’ protein synthesis, translation initiation occurred on transcriptionally up-regulated mRNAs as early as 10 minutes following glucose withdrawal. Reduction in polysomal mRNA closely paralleled reduction in total mRNA levels for almost all down-regulated genes. These data contradict the model that glucose withdrawal leads to widespread sequestration of mRNAs in stable translationally repressed mRNPs (Brengues and Parker, 2007
; Brengues et al., 2005
; Hoyle et al., 2007
; Teixeira et al., 2005
Unsupervised hierarchical clustering revealed small groups of genes that appear to be regulated primarily at the post-transcriptional level: cluster I genes showed relatively reduced polysomal mRNA levels despite increased total mRNA abundance due to low ribosome occupancy; conversely, cluster II genes had relatively unaffected polysomal mRNA levels despite reduced total mRNA abundance due to high ribosome occupancy (). Nevertheless, there was strong overall agreement between the relative increase or decrease of an mRNA in the cell at any time point after glucose withdrawal (T/T) and the change in polysome association at that time point (P/P). In contrast, ribosome occupancy (P/T) at any time point was more highly correlated with ribosome occupancy at other times than with either P/P or T/T at the same time point after glucose withdrawal (), suggesting that it is largely an intrinsic property of each mRNA, although some starvation-induced changes in P/T were observed. At the global scale, fold-changes in polysomal mRNA levels were somewhat compressed compared to changes in total mRNA levels (, S2
). Thus, our data do not indicate widespread ‘potentiation’, whereby changes in total mRNA levels are amplified by homodirectional changes in translation efficiency, as was suggested in studies of yeast subject to rapamycin treatment or heat shock (Preiss et al., 2003
). Stress specificity of ‘potentiation’ was previously noticed in comparison of the translational responses to amino acid starvation and butanol stress (Smirnova et al., 2005
Relationships Between Changes in Transcript Levels and Ribosome Occupancy
The complex groupings of genes produced by combining the analysis of transcriptional and post-transcriptional regulatory behavior using unsupervised hierarchical clustering did not readily reveal the logic underlying the relationship between the two modes of regulation. To investigate this relationship more directly, changes in total mRNA levels (T/T) and changes in ribosome occupancy (P/T) were analyzed separately using k
-means clustering to identify groups of genes displaying similar behavior for each mode of regulation. Experimenting with various group numbers (k
= 2–20) revealed that k
= 7 for the T/T comparisons and k
= 3 for the P/T comparisons gave robust solutions reflecting a reasonable compromise between preserving the complexity of the data and simplifying the subsequent analysis (Fig. S3
). Clustering genes by their ribosome occupancy (P/T) produced simple divisions into groups with high, low, and neutral P/T ratios ().
Ribosome occupancy and mRNA abundance are divergently regulated
Although we anticipated that kinetic analysis of the total mRNA changes in response to glucose starvation would reveal temporal distinctions between genes, the groups of co-regulated genes identified by k
-means clustering at k
= 7 differed primarily by the magnitude of relative increase/decrease rather than by the timing of the maximal changes in gene expression (). Clustering with values of k
> 12 did reveal kinetically distinct patterns of mRNA accumulation that are consistent with current understanding of transcriptional responses to glucose withdrawal (Fig. S3
). For example, genes subject to repression by Mig1 in the presence of glucose (e.g. SUC2
) were de-repressed within 10 minutes following glucose withdrawal, whereas known targets of the Cat8 transcription factor (e.g. PCK1
, and FBP1
) accumulated mRNA strongly only after 120 minutes (Fig. S1
). Groups of genes clustered based on starvation-induced changes in total relative mRNA levels ranged from strongly induced (119 genes, median induction at t = 60 min of 19.7-fold) to strongly repressed (311 genes, median repression at t = 60 min of 11.9-fold). The use of multiple time points as well as both biological and technical replicates allowed the confident identification of genes displaying modest yet consistent changes in mRNA levels. The most weakly induced category includes more than 100 genes encoding mitochondrial proteins known to be important for growth in the absence of glucose, highlighting the potential biological significance of coordinated small changes in gene expression.
Changes in mRNA relative abundance and ribosome occupancy were not independent of one another (Χ2 = 179, p < 0.0001 for T/T vs. P/T). Starvation-induced genes were more likely to show low ribosome occupancy after glucose withdrawal, and repressed genes were more likely to show high ribosome occupancy (). Despite the statistical interdependence of changes in total mRNA levels and ribosome occupancy, for each group of genes having similarly induced/repressed total mRNA levels there were many genes displaying each of the possible post-transcriptional regulatory behaviors.
Functionally Distinct Groups of Genes Are Co-Regulated at the Post-Transcriptional Level
To investigate the possibility that differences in post-transcriptional regulatory behavior are biologically significant, the function of genes in each category was examined by gene ontology (GO) analysis. Notably, the GO terms that were significantly enriched (p < 0.01 with Bonferroni correction for multiple hypothesis testing) for the seven mRNA abundance-based groups (highly induced, moderately induced, weakly induced, strongly repressed, moderately repressed, weakly repressed, and unchanged) segregated within these groups along post-transcriptional regulatory divisions. A complete list of significant GO terms for each of the 21 regulatory groups (7 T/T × 3 P/T) is provided in Supplemental Table 1
. Rarely were GO categories split between multiple post-transcriptional (P/T) regulatory groups, and where such a split occurred, the GO category spanned two out of three most similar groups – ‘high’ and ‘neutral’ or ‘low’ and ‘neutral’, not ‘high’ and ‘low’. This suggests that the post-transcriptional behavior of a gene is related to its biological function in the starved cell.
The post-transcriptional partitioning of functionally related genes may derive from mechanistic similarities in their gene expression pathways. For example, the nuclear-encoded mitochondrial protein genes showed consistently low ribosome occupancy despite the fact that these genes are transcriptionally up-regulated in response to glucose withdrawal and encode proteins required for the cellular adaptation to low glucose conditions. mRNAs of some nuclear-encoded mitochondrial proteins are translated on cytosolic ribosomes that associate with mitochondria (Marc et al., 2002
; Sylvestre et al., 2003
). Subcellular localization of these mRNAs is driven by cis-acting elements in their 3′UTRs, and requires both the trans-acting RNA-binding protein Puf3 and the translocase of the mitochondrial outer membrane complex for full localization (Eliyahu et al., 2010
; Saint-Georges et al., 2008
). The apparent lag between the appearance of these transcriptionally induced mRNAs in the cell and their association with polysomes in our experiments could be explained by translational silencing of messages in transit from the nucleus to the periphery of mitochondria. In light of this explanation for the low P/T ratios of mRNAs encoding mitochondrial proteins, it is interesting to consider whether other transcriptionally induced yet poorly translated mRNAs identified in our analysis might also be subject to localization-dependent translational control.
The most striking example of gene function partitioning according to post-transcriptional regulatory behavior was the separation of cytoplasmic ribosomal protein genes (RPGs) from ribosome biogenesis factors (RBGs) (, , S4
). Genes from both functional groups were moderately to strongly reduced in both the total and polysomal mRNA pools after glucose withdrawal. This down-regulation is likely due to the greatly reduced demands for new ribosome synthesis as the cells transition from rapid growth and division, requiring the assembly of ~200,000 new ribosomes every 90 minutes (Warner, 1999
), to cellular differentiation and slower growth in the invasive filamentous form (Cullen and Sprague, 2000
). The two groups diverged in their post-transcriptional responses to glucose starvation. The P/T ratios of RBG mRNAs increased after glucose withdrawal and remained relatively high throughout the two-hour experiment. In contrast, the mRNAs encoding RPGs showed very low P/T ratios between 10 and 30 minutes after glucose withdrawal, and these ratios increased between 30 and 120 minutes (). Low P/T ratios indicate that a population of non-translating mRNA exists in the cell. To test this interpretation directly, we performed qRT-PCR on polysome gradient fractions after 10 minutes of glucose starvation. RPG mRNAs (low P/T genes) accumulated in ribosome-free fractions at the top of the gradient, whereas mRNAs from genes with high P/T ratios did not (Fig. S4
). In principle, the apparent increase in ribosome occupancy for RPGs after 60–120 minutes could result from improved translation (increased P) or from degradation of the non-polysomal pool of mRNA. Given that P/P and T/T ratios for RPGs were divergent at early times minus glucose and converged after 60 minutes of starvation (), the second interpretation is more plausible, suggesting that this sub-population of mRNA is only transiently stable as a non-translating pool. Notably, the messages that were most dramatically down-regulated upon glucose withdrawal and had low P/T values in acutely starved cells are the most abundant (). Transient preservation of these mRNAs in a non-translating pool (Fig. S4
) may reflect a bet-hedging strategy (see Discussion).
RPGs and RBGs differ in their post-transcriptional responses to glucose withdrawal
Highly expressed mRNAs are preferentially retained in the non-translating pool
Only a Subset of mRNAs Can Return to Polysomes Following Starvation and Re-Feeding
Previous reports showed that the global reduction in protein synthesis caused by ten minutes of glucose withdrawal can be reversed within five minutes of glucose re-addition (Ashe et al., 2000
; Brengues et al., 2005
). It was proposed that this rapid recovery of translation is due to mobilization of stored, translationally silenced mRNPs from P-bodies (cytoplasmic RNP granules) to polysomes. Only two reporter genes were tested for the capacity to return to polysomes in the absence of new transcription (Brengues et al., 2005
). Our data suggest that most mRNAs are depleted from the total mRNA pool coincident with their loss from polysomes (, S4D
), and that the capacity for translational resurrection of pre-existing messages is thus narrowly restricted to a select sub-population that includes the RPG mRNAs. To test this hypothesis directly, we examined global as well as gene-specific recovery of translation following glucose re-addition to starved cells. New transcription was inhibited with the drug thiolutin (Grigull et al., 2004
), to assess the capacity of pre-existing mRNAs to return to polysomes. Treatment with thiolutin slightly blunted translational recovery upon glucose re-addition, resulting in fewer heavy polysomes, more disomes, and persistence of a larger 80S monosome peak (). These results indicate that initiation of translation on newly synthesized mRNAs accounts for some of the ‘recovery’ of translation after glucose re-addition, even after only 10 minutes of glucose starvation.
Translational resurrection is restricted to a subset of genes for a limited time
Nevertheless, substantial polysome recovery occurred even when new transcription was inhibited (). To determine which genes participate in this recovery, the polysomal and non-polysomal mRNA pools were examined using microarrays following 10 minutes of glucose starvation and again after 5 minutes of glucose repletion, with thiolutin present throughout. A gene’s P/T ratio during glucose starvation predicted its ability to return to polysomes upon glucose re-addition. Genes that showed high or neutral P/T ratios following glucose removal () were depleted from the non-polysomal as well as the polysomal mRNA pools after 10 minutes of glucose starvation, and showed very little mobilization into polysomes 5 minutes after glucose re-addition (). This argues that genes like the RBGs, which have high P/T ratios under conditions of decreasing P, do so because of rapid depletion of the non-polysomal mRNA. Consistent with this interpretation, we found by qRT-PCR that 80–95% of the total mRNA from these high P/T genes is gone from the total mRNA pool by 10 minutes after glucose starvation. In contrast, the genes that showed low P/T after glucose withdrawal were preferentially depleted from the polysomal compared to the non-polysomal pool (Fig. S4D
). This group of genes also showed significantly greater mobilization into polysomes upon glucose re-addition by microarray (p < 0.0001), and by qRT-PCR analysis of select genes’ mRNA abundance in polysome gradient fractions (). Similar results were observed for more moderately down-regulated genes, with the low P/T sub-population showing significantly greater recovery than either the high or neutral P/T genes, although the extent of recovery was somewhat less. The RPGs as a class showed ~6-fold more recovery than all other starvation-repressed genes (T/T clusters 5, 6 and 7 from ) combined (p < 6.0 × 10−54
), whereas RBGs did not recover ().
Quantitative RT-PCR validation of select genes’ mRNA abundance in polysome fractions following glucose starvation and re-feeding
We verified that P-bodies formed under these conditions based on the localization of previously characterized protein and RNA reporters (Fig. S5
). In addition, we examined the localization of RPS26A-U1A, a low P/T gene capable of returning to polysomes upon re-feeding, and of RRP4-U1A and LSG1-U1A, high P/T genes that are largely depleted within 10 minutes of glucose withdrawal and are not capable of returning to polysomes in the absence of new transcription. We detected P-body localization for RPS26A-U1A and RRP4-U1A. We did not detect LSG1-U1A, which was very lowly expressed by Northern blot (data not shown). These reporter experiments don’t distinguish whether the P-body localized portion of RPS26A mRNA is the fraction of intact mRNA that returns to polysomes upon re-feeding (~50%, ), or alternatively, if the P-body localized portion is comprised of decay intermediates of the fraction (~50%, Fig. S4D
) that disappeared from the total mRNA pool, based on qRT-PCR using primers within the ORF. We found that stabilizing 3′UTR decay intermediates led to prominent P-body localization (Fig. S5
), which is consistent with previous reports (Sheth and Parker, 2003
). Thus, P-bodies likely contain endogenous mRNAs undergoing decay. Whether or not they also contain stored translationally repressed RPG mRNAs remains an open question. Wherever translationally repressed mRNAs reside in the cell, only select mRNAs are able to persist in a non-translating pool and resume active translation following a short period of glucose starvation.
The glucose starvation microarray time course data suggest that the non-translating sub-population of mRNAs that accumulates at early times is depleted after more prolonged glucose starvation (, ). This interpretation of the data predicts a turning point after which any recovery of polysomes upon glucose re-addition would require new transcription. To test this prediction, we subjected cells to varying periods of glucose starvation, with and without inhibition of new transcription by the drug thiolutin, and examined global translation activity after five minutes of glucose re-addition. After 45 or 60 minutes of starvation in the absence of new transcription, polysome recovery was greatly reduced compared to cells starved for only 10 minutes (). If new transcription was allowed to occur during extended starvation, translational recovery upon glucose re-addition increased (). These observations, together with the indication that specific mRNAs (RPGs) are lost between the 10- and 60-minute time points (), suggests that translational activation of these silenced mRNAs contributes substantially to polysome recovery upon re-feeding of briefly starved cells. These data argue that cells’ inability to rapidly restore translation without new transcription after longer periods of starvation is due to the disappearance of a population of non-translating mRNAs. Between 30 and 60 minutes appears to be a switch point for P/T ratios in our data, regardless of whether the ratios are increasing or decreasing. This suggests that the timing of the ‘turning point’ for cells to recover translation of stored RPGs mRNAs may relate to widespread changes in the activity of factors that influence P/T through effects on mRNA stability.
Molecular Insights Into Selective Preservation of Non-Translating RPG Transcripts
How are certain mRNAs able to persist in the cell, even transiently, following glucose starvation, when the majority of mRNAs are depleted from the total RNA pool_coincident, within the time resolution of our experiments, with their loss from polysomes? Comparison with genome-wide mRNA half-life measurements indicates that stability in glucose-replete conditions is probably not the determining factor for an mRNA’s capacity to persist in a stable non-translating pool after glucose withdrawal (Fig. S6
). In particular, the RPGs as a class have short half-lives in glucose-replete conditions compared to most genes (Grigull et al., 2004
; Wang et al., 2002
Alternatively, the ‘post-transcriptional operon’ hypothesis posits a role for RNA-binding proteins (RBPs) in coordinating the post-transcriptional fate of functionally related genes (Keene, 2007
; Keene and Tenenbaum, 2002
). To investigate the possibility that specific RBPs affect the post-transcriptional behavior of mRNAs following glucose withdrawal, we examined data from a recent genome-wide association study that identified mRNA targets of 40 yeast RBPs (Hogan et al., 2008
). Comparing the ‘transcriptional’ (changes in total mRNA abundance) and post-transcriptional (P/T) regulatory groups identified in our study with RBP-mRNA target groups revealed concordance between RBP association patterns and P/T but not T/T (). For simplicity, only the seven RBPs that showed significant enrichment or de-enrichment (p < 0.01, Bonferroni corrected) of target mRNAs within at least one group are displayed. Similar to GO terms, which rarely spanned opposing P/T groups, RBP-association profiles were similar for “high” and “neutral” P/T groups for many RBPs, and for “low” and “neutral” for Puf3, whereas the “high” and “low” groups had almost no enrichments in common and frequently appeared to be mirror images of each other. In contrast, organizing RBP-association patterns by T/T group resulted in a ‘checkerboard’ appearance, despite the fact that T/T groups contain functionally and cytotopically related genes. This suggests that the coherence of the association between P/T behavior and particular RBPs might be the result of direct effects of these RBPs on mRNA translatability and/or mRNA stability outside of polysomes.
Resurrection-competent mRNAs associate with Pab1 and have longer poly(A) tails
Pab1-associated mRNAs were notably enriched with groups having low P/T ratios, and correspondingly depleted in groups with high and neutral P/T ratios (). Comparison of P/T groups with genome-wide measurements of poly(A) tail lengths (Beilharz and Preiss, 2007
) revealed a lack of short-tailed and an enrichment of long-tailed genes among mRNAs with low P/T values. Conversely, genes with long poly(A) tails were significantly under-represented among groups with high P/T ratios (). The association of Pab1 and long poly(A) tails with groups of mRNAs with low P/T ratios was counterintuitive given Pab1’s role in enhancing translation initiation and the positive correlation between poly(A) tail length and ribosome density (Beilharz and Preiss, 2007
). However, genome-wide ribosome occupancy (P/T), as measured in this study or by others (Arava et al., 2003
), does not correlate with measures of translational efficiency (number of ribosomes/mRNA) determined by polysome fractionation (Arava et al., 2003
) or by ribosome footprint profiling (Ingolia et al., 2009
) (Fig. S7
). Thus, we interpret the low P/T values of RPGs and high P/T values of RBGs after 10–30 minutes without glucose as consequences of differences in the stabilities of non-translating mRNAs, rather than as differences in translational activity following glucose withdrawal.
Consistent with a role for Pab1/poly(A) in stabilizing non-translating mRNAs, poly(A) tail length was positively correlated with the capacity for translational resurrection following glucose re-addition (). The intersection between the two groups that show increased total mRNA levels, low P/T ratios, and enrichment with Pab1 (as well as Puf3) is dominated by nuclear-encoded mitochondrial protein genes, previously described as having long poly(A) tails (Beilharz and Preiss, 2007
) and thought to transit the cytoplasm in a translationally repressed state before being translated on peri-mitochondrial ribosomes (Eliyahu et al., 2010
; Marc et al., 2002
; Sylvestre et al., 2003
). A potentially unifying explanation for the enrichment of Pab1 with various low P/T groups is that Pab1 association may stabilize non-translating mRNAs.