Promoter-dependent nuclear RNA degradation ensures cell cycle-specific gene expression

Cell cycle progression depends on phase-specific gene expression. Here we show that the nuclear RNA degradation machinery plays a lead role in promoting cell cycle-dependent gene expression by triggering promoter-dependent co-transcriptional RNA degradation. Single molecule quantification of RNA abundance in different phases of the cell cycle indicates that relative curtailment of gene expression in certain phases is attained even when transcription is not completely inhibited. When nuclear ribonucleases are deleted, transcription of the Saccharomyces cerevisiae G1-specific axial budding gene AXL2 is detected throughout the cell cycle and its phase-specific expression is lost. Promoter replacement abolished cell cycle-dependent RNA degradation and rendered the RNA insensitive to the deletion of nuclear ribonucleases. Together the data reveal a model of gene regulation whereby RNA abundance is controlled by promoter-dependent induction of RNA degradation.

C ells proliferate through a well-coordinated cycle ensuring the replication and redistribution of all cellular components [1][2][3] . This cell cycle requires the synthesis of a panel of proteins required for replication, many of which are expressed only when they are needed. Consequently, progression through the cell cycle is accompanied by dramatic reorganization of gene expression that is often referred to as cell cycle-regulated transcription 3 . 20% of yeast genes are expressed in a cell cycledependent manner, guided by the activation of transcription factors and cycling genes 4 . These waves of expression are often divided into four events that take place in G1, S, G2/M, and in the transition between the M and G1 phases. Genes associated with these waves are classified into clusters based on expression patterns and promoter motifs that promote cell cycle-regulated transcription 3 . Furthermore, shared common promoter elements are co-regulated by specific subsets of transcription factors, like the G1/S SCB-binding factor and MCB-binding factor 5,6 . These promoters can be used experimentally to induce phase-specific expression of reporter proteins 7,8 .
More recently, cell cycle changes have been associated with changes in RNA stability 9 . Metabolic RNA labeling and dynamic transcriptome analysis identified hundreds of genes in budding yeast with periodic changes in, transcription and mRNA degradation, rates [9][10][11][12][13] . In general RNA degradation follows peaks of mRNA synthesis at defined times during the cell cycle. Strikingly, it was shown that the promoters of two mitotic genes (SWI5 and CLB2) trigger cell cycledependent RNA degradation in the cytoplasm 14 . In these two cases, it was proposed that RNA is marked for decay by factors loaded on the RNA during transcription, leading to cytoplasmic degradation at the end of the cell cycle. However, it is not clear how the factors are recruited to the promoter, nor how the RNA is degraded in the cytoplasm. It is also unclear if cotranscriptional RNA degradation in the nucleus contributes to cell cycle repression. A few examples of targeted nuclear RNA degradation have been linked to nutritional changes and other stresses [15][16][17] . However, it remains unclear generally how nuclear RNA degradation contributes to transcriptional regulation or how it regulates specific cell cycle genes.
Most biochemical assays provide estimates of average RNA amounts in a population of cells but cannot distinguish between nuclear and cytoplasmic RNA degradation. In such assays it is difficult to determine the location and timing of gene expression and the resolution of the expression cycle is often blurred by stochastic differences between cells. In this study, we monitored the synthesis and degradation of the phase-specific axial budding gene AXL2 during the cell cycle using single molecule analysis to determine the location, timing, and mechanism of phase-specific gene expression. Like many cell cycle genes AXL2 is thought to be expressed in the G1 phase of the cell cycle in a promoter-dependent manner 18,19 . However, in vitro studies suggested that this gene might also be regulated post-transcriptionally 20 . Here, we demonstrate that despite the sharp decline of AXL2 transcripts at the end of G1, it is transcribed albeit at a lower level throughout the cell cycle; live cell analysis and chromatin immunoprecipitation of RNA polymerase II confirmed that transcription of AXL2 continues in the S and G2/M phases of the cell cycle. Strikingly, deletion of nuclear ribonucleases completely blocked the characteristic periodicity of mature AXL2 mRNA levels. Ribonucleases were associated with the chromatin linking the RNA degradation machinery to transcription. Consistently, substitution of the AXL2 promoter impaired the cycling of gene expression. Together our data reveal a model of gene expression, in which cell cycle expression is achieved through promoter-dependent RNA degradation.

Results
RNA decay drives the cell cycle expression of AXL2. To determine the regulatory events controlling the expression of cell cycle-associated genes, we monitored the life cycle of the axial budding gene AXL2 using single-molecule fluorescence in situ hybridization (smFISH) 21 . AXL2 expression is required in the G1 phase of the cell cycle, and defects in its expression are easily scored by changes in budding patterns 18 . The Axl2 mRNA was detected using two independent sets of single labeled 20 oligodeoxynucleotides probes (Fig. 1a). The first probe set hybridizes to the 5′ end of the mRNA, while the other hybridizes to the 3′ end. Signals generated by the 5′ end probe at the transcription site in the nucleus identify nascent RNA, while complete or mature RNA, which is mostly cytoplasmic is identified by signals generated from both 5′ and 3′ end probes. To determine the number of probes sufficient for accurate RNA detection we compared the signal intensity and the number of RNAs detected by 12-and 24probe sets. The results indicated that while a set of 24 probes resulted in higher signal intensity indicating a higher number of RNAs detected, the use of 12 probes provided a narrower peak and less inter-experiment variation without changing the overall conclusion ( Supplementary Fig. 1). Accordingly, we used the 12probe set for all subsequent experiments comparing either nascent (nuclear) or complete (mostly cytoplasmic) RNA in different conditions and phases of the cell cycle. We also confirmed the accuracy of nuclear targeting of the transcripts by comparing compressed 2D-with uncompressed 3D images, as 2D stacking of the images, which simplify quantification, do not result in more than 10% inaccuracies ( Supplementary Fig. 2).
Examining the expression pattern of Axl2 during the cell cycle using either the 5′ or the 3′ end probe reproduced the established expression pattern of Axl2 mRNA, which peaks in G1 and is repressed in later phases (Fig. 1b) 18 . We next analyzed the Axl2 expression dynamics quantitatively as our technique allows quantification of the number of nascent RNAs (hybridizing to the 5′ end probe only in the nucleus) and completed RNAs (hybridizing to both 5′ and 3′ end probes in the cytoplasm) per cell. Both nascent and full-length RNA decreased as the cells exited from the G1 phase of the cell cycle ( Fig. 1c and Supplementary Fig. 3). On average 4 RNA molecules per cell were detected in G1, while only 1 molecule per cell was detected in S phase. Together these data illustrate the overall accuracy of the Axl2 mRNA targeted probes. We also found the amount of Axl2 mRNA varies depending on the cell size ( Supplementary  Fig. 3c). In general, more Axl2 mRNA was detected in larger G1 cells. In all cases RNA peaked in G1 and gradually dipped over S phase to a minimum in G2/M. However, cytoplasmic and nascent RNA did not decrease at the same rate; full length RNA decreased by 6-fold, whereas nascent RNA decreased only 3-fold (Fig. 1c). Notably, RNA was detected in the nucleus in all phases of the cell cycle suggesting that transcription continues throughout the cell cycle (Fig. 1d). To verify the continuing transcription of the AXL2 gene beyond the G1 phase of the cell cycle, we synchronized cultures and compared the association of RNA polymerase (RNAPII) at the AXL2 locus to accumulation of its RNA.
As indicated in Fig. 1e, the association of RNAPII with the promoter of AXL2, did not decrease but instead, increased in the S and G2/M phases. On the other hand, the association of RNAPII with the AXL2 open reading-frame was modestly reduced but continued to be detected in the S and G2/M phases. This data clearly indicate that Axl2 mRNA transcription continues beyond G1. Consistently, monitoring the accumulation of Axl2 mRNA in the same synchronized culture by RT-qPCR once again indicated that Axl2 mRNA abundance decreases but remains detectable in S and G2/M (Fig. 1f) Together these results clearly indicate that the cycling of AXL2 expression is not produced by strict repression and induction of transcription.
To monitor the dynamics of AXL2 expression, we examined the accumulation of Axl2 mRNA in real-time ( Fig. 2 and Supplementary Figs. 4 and 5). We did this by inserting repeats of the MS2 coat protein binding site into the 5′UTR of the gene (Fig. 2a), which then acted as a reporter of the mRNA by attracting a coat protein-EGFP fusion in live cells 22,23 . We have used the tag version and location previously certified for the detection of transcription and nascent RNA 22 . Tagging Axl2 mRNA slightly increased its abundance but did affect its expression cycle (Fig. 2b, c). Indeed, the overall number of spots detected by the tagged version was similar to those detected by the untagged version using FISH (compare, Figs. 1b and 2c). We monitored Axl2 mRNA every 10 min across the cell cycle and every minute across the G1/S transition. Interestingly, de novo transcription was detected in the nucleus at all phases of the cell cycle and was not completely repressed in S and G2/M (Fig. 2c). On the other hand, cytoplasmic RNA was clearly reduced in S and G2/M. Interestingly, while there were only 1 and 0.6 Axl2 RNAs per cell in S and G2/M phases respectively on average, transient pulses of RNA expression were detected in live cells in these phases (Fig. 2d). This indicates that RNA is both made and degraded during S and G2/M. It should be noted that since the pictures were taken every 10 min and the transcription pulses take~2 min followed by the export or degradation of the RNA  Fig. 1 Single mRNA molecules analysis reveals synthesis of Axl2 mRNA across the cell cycle. a Schematic representation of the probes used to detect AXL2 mRNA in the fluorescence in situ hybridization (FISH) assay shown in (b). Probes complementary to the 5′ end of the target are shown in green and those to the 3′ end are red. b Monitoring AXL2 expression through the cell cycle using single cell FISH assay. The RNA was detected using the probes illustrated in (a) and the nuclei stained with DAPI (blue). The cell cycle phases were determined using the bud size and nucleus position. The white bar equals 2 µm. c AXL2 expression is downregulated during S and G2/M. Nascent mRNAs (dark gray) are defined as RNAs detected using the 5′ probe in the nucleus, while complete cytoplasmic mRNAs (light gray) are detected by both 5′ and 3′ probes in the cytoplasm. The number of RNAs per cell is shown in the violin plot for 150 cells analyzed per phase. d Examples of cells transcribing Axl2 mRNA in the S and G2/M. RNA was detected as described in (b). e Cells were synchronized in G1, S, or G2/M and the enrichment of RNAPII at the AXL2 locus was analyzed by ChIP and normalized to the G1 enrichment. f The amount of Axl2 mRNA from the synchronized cultures used in (e) was determined using qRT-PCR and presented relative to the amount detected in G1. The bar graphs represent the average of 3 independent biological replicates and the value of each replicate is shown as dot. (*p < 0.05 and **p < 0.01 by two-tailed paired t test) many RNA transcription events may escape detection 23 . Indeed, acquiring images every minute reveals similar number of transcription pulses in the G1 and S phases (Fig. 2e). This clearly indicates once again that the rapid decrease in Axl2 mRNA amounts is not only due to transcription repression. We conclude that the phase-specific expression of AXL2 is not only created by peak transcription in G1 but through accelerated RNA decay in S phase.
Nuclear ribonucleases promote AXL2 cell cycle-dependent repression. The data above indicate that at the G1/S transition there is a discrepancy between the decrease in number of Axl2 mRNAs in the nucleus and its decrease in transcription, suggesting increased nuclear RNA decay. To evaluate the contribution of nuclear RNA degradation to cell cycle-dependent expression, we examined the effect of abolishing the activity of different nuclear ribonucleases on AXL2 expression cycle. We deleted or inactivated 4 ribonucleases: three exoribonucleases (Rrp6p, Rat1p, and Xrn1p) and one endoribonuclease (Rnt1p). Rrp6p is a component of the nuclear exosome implicated in the 3′-5′ end degradation of non-polyadenylated RNA with free 3′ OH 24 . On the other hand, Rat1p is a 5′-3′ exoribonuclease that degrades uncapped RNA with 5′ monophosphate involved in transcription termination 25 . Unlike Rrp6p and Rat1p that are exclusively nuclear, Xrn1p is a mostly cytoplasmic paralog of Rat1p, but it has also been linked to nuclear RNA degradation 26 . Finally, Rnt1p is a nuclear endoribonuclease that exits from the nucleolus to the nucleoplasm at the end of the G1 phase of the cell cycle and cleaves RNA with conserved NGNN stem loop structures including one in Axl2 mRNA 20 . Mutating each of the 4 ribonucleases individually increased the abundance of both nascent and complete RNA in S phases when compared to that detected in G1 or when compared to the wild type RNA in the same phase (Fig. 3a). Similarly, certain ribonuclease mutations (e.g., rat1-1 and rnt1Δ) increased the ratio of the RNA in G2/M phase compared to G1 (Fig. 3b). However, only RNT1 deletion increased cytoplasmic Axl2 mRNA in the G2/M phase compared to wild type while the detected RNA remained stable at around 0.2-0.5 copies per cell in the other ribonuclease deletions ( Supplementary Fig. 6a). Surprisingly, XRN1 deletion not only altered the amount of complete Axl2 mRNA in the cytoplasm but it also increased nascent RNA in the nucleus in G1 and S (Fig. 3b, c). This suggests that Xrn1p directly or indirectly contributes to Axl2 degradation in the nucleus. Double and triple deletions or mutations of exoribonucleases (e.g., RRP6 and XRN1) increased the effect of the endoribonuclease RNT1 in G2/M but not in S phase (Supplementary Fig. 6b). Therefore, RNA that is not cleaved by Rnt1p in S phase can be degraded by the other exoribonucleases later in the cell cycle. The other nuclear ribonucleases may also complete the degradation of RNA fragments generated by Rnt1p cleavage. Together these observations suggest that nuclear ribonucleases work together to ensure the rapid degradation of Axl2 mRNA in S phase. Strikingly, double deletion of RNT1 and XRN1 completely abolished the periodic expression of AXL2 resulting in the production of increasing amounts of RNA as the cells grow ( Supplementary Fig. 6a, b). We conclude that nuclear ribonucleases are required for the cell cycle-dependent repression of AXL2.  Nuclear ribonucleases are essential for the cell cycle-dependent repression of AXL2. a smFISH analysis of cells with the various ribonuclease mutations shown. The expression of Axl2 mRNA was monitored in wild type cells (WT) or cells with single (rrp6Δ, rat1-1, rnt1Δ, and xrn1Δ) or double (rrp6Δ rnt1Δ, rat1-1 rnt1Δ, xrn1Δ, and xrn1Δ rnt1Δ) mutations in different nuclear or nuclear/cytoplasmic ribonucleases. FISH analysis was performed as described in Fig. 1 and examples of cells in the S and G2/M phases, when AXL2 is normally repressed, are shown. The white bar equals 2 µm. b Quantification of ribonuclease-dependent changes in nascent RNA levels (5′ end probe) in S (top panel) or G2/M (bottom panel). Nascent RNA detected in the nucleus was quantified in different phases and the Log2 change between S or G2/M relative to G1 is plotted for each mutant strain. c Effect of ribonuclease mutations on the abundance of cytoplasmic Axl2 mRNA. The total amount of full-length RNA detected in different phases was quantified and the Log2 change between the S (top panel) or G2/M (bottom panel) relative to G1 is plotted for each mutant strain. All error bars represent the uncertainties calculated from the standard errors of the mean (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by two-tailed unpaired t test with Welch's correction) Rnt1p is required for AXL2 cell cycle-dependent repression. We next focused on Rnt1p as it is the only ribonuclease of the four tested with well-defined substrate selectivity 27 , a deletion phenotype of net increase in the Axl2 mRNA ( Supplementary  Fig. 6a), and previously shown Axl2 mRNA cleavage activity in vitro 20 . We thus studied if and how Rnt1p endoribonuclease represses AXL2 during the cell cycle. To test this, we first reevaluated the impact of RNT1 deletion on Axl2 mRNA in vivo and the specificity of Rnt1p's cleavage of Axl2 mRNA in vitro. Total RNA was extracted from wild type cells, cells lacking RNT1 (rnt1Δ) or cells expressing a mutation in the Axl2 mRNA stem-loop recognized by Rnt1p (rnt1Δ LoopM, Fig. 4a). We then incubated the RNAs with, or without, recombinant Rnt1p. As expected, deletion of RNT1 increased the overall abundance of Axl2 mRNA, confirming the results obtained by smFISH (Fig. 4b, c). Recombinant enzyme generated the previously seen cleavage product (P) 14-16 nucleotides from the loop 20 . The cleavage was dependent on both the presence of the site that Rnt1p cleaves (rnt1Δ LoopM+) and the catalytic activity of Rnt1p (rnt1Δ c) (Fig. 4c). This clearly illustrates the accuracy and specificity of Rnt1p's cleavage of Axl2 mRNA.
The stem-loop mutation (LoopM) that blocks Axl2 mRNA cleavage in vitro (Fig. 4c) also increased the abundance of Axl2 mRNA, but only in the absence of the exoribonucleases Xrn1p and Rrp6p (Supplementary Fig. 7a). Deleting RNT1 in the strains carrying the stem-loop and exoribonuclease mutations (rnt1Δ rrp6Δ xrn1Δ LoopM) did not increase the abundance of nascent or full-length RNA in S phase ( Supplementary Fig. 7b). In contrast, deletion of RNT1 increases the abundance of Axl2 mRNA in the G2/M phase even when the loop is mutated. Therefore the site that Rnt1p cleaves is required for Axl2 repression in S phase and Rnt1p is required for maintaining the repression during G2/M. The reason why Rnt1p loop mutation by itself did not increase RNA abundance is likely because of b a The position of the rRNA size marker is indicated on the right. c In vitro cleavage assay of Axl2 mRNA. Total RNA extracted from rnt1Δ cells (rnt1Δ) or rnt1Δ cells expressing Axl2 mRNA with mutated cleavage site (rnt1Δ LoopM) was incubated alone (−), with recombinant Rnt1p (+), or with catalytically inactive (rnt1-D245R) mutant of Rnt1p (c). The RNA was detected using probes against Axl2, Rnt1p substrate Mig2, or Act1 mRNA as negative control. The position of the uncleaved RNA and the 3′ cleavage product (P) is indicated on the left. d Short-term inactivation of Rnt1p increases the abundance of Axl2 mRNA. Axl2 mRNA was quantified by qRT-PCR in WT, rnt1Δ or mutated thermosensitive RNT1 (rnt1-1) cells grown at the permissive (26°C) or restrictive temperature (37°C). e Axl2 mRNA mimics the expression kinetics of Rnt1p substrates. The abundance of Axl2 and Rnt1p substrate Mig2 and unrelated Toa1 mRNA was examined after shifting to the restrictive temperature (37°C). Axl2 and Mig2 mRNAs were analyzed in 3 N while Toa1 mRNA was analyzed in 2 N. f Rnt1p exit from the nucleolus to the nucleoplasm is required for the repression of AXL2. The abundance of Axl2 mRNA was determined in cells expressing wild type Rnt1p (RNT1) or a mutated version that accumulates in the nucleolus (rnt1-ΔNT2). The bar graphs represent the average of 2-3 independent biological replicates and the value of each replicate is shown as dot. (**p < 0.01, ***p < 0.001 by two-tailed unpaired t test) degradation by other exoribonucleases that are otherwise inactivated by RNT1 deletion, as previously shown for other substrates of Rnt1p 15 . Notably, the catalytically inactive Rnt1p mutant that nevertheless preserves RNA binding, increased AXL2 mRNA abundance and partially inhibited its cycling, clearly indicating Axl2 cleavage is required for its repression (Supplementary Fig. 7c).
To differentiate between direct and secondary effects of RNT1 deletion, we examined the impact of short-term inactivation of Rnt1p activity on the abundance of Axl2 mRNA and compared it to the effect of RNT1 deletion by qRT-PCR. As expected, the deletion of RNT1 (rnt1Δ) had the most impact on Axl2, increasing its mRNA 4-fold (Fig. 4d). Strikingly, the increase in Axl2 RNA was also observed shortly after the inactivation of a temperature-sensitive allele of RNT1 (rnt1-1) 28 , consistent with direct cleavage by Rnt1p (Fig. 4d). Indeed, the increase in Axl2 mRNA was observed after only 1 h of shift to the restrictive temperature, at the same time as change in the abundance of the known substrate Mig2 mRNA 16 is observed and before any increase in an unrelated mRNA, Toa1 (Fig. 4e). Together these data confirm the direct link between Rnt1p catalytic activity and the increase in Axl2 mRNA abundance. Since Rnt1p accumulates exclusively in the nucleus its deletion would be expected to increase the number of RNAs leaving the nucleus without affecting the overall half-life of the majority of the RNA that accumulates in the cytoplasm. Accordingly, there was no significant difference between the half-life of Axl2 RNA extracted from wild-type and rnt1Δ cells ( Supplementary Fig. 8). This further confirms the specificity of Rnt1p, on the nascent Axl2 transcripts, observed above by smFISH. To further prove the specificity of Rnt1p effect on Axl2 mRNA we examined the impact of Rnt1p localization on the abundance of Axl2. Axl2 mRNA is produced in the nucleoplasm and Rnt1p exits from the nucleolus to the nucleoplasm in S phase. Therefore, blocking this exit of Rnt1p from the nucleolus should prevent the cleavage of Axl2 and increase it incrementally in S and G/2M. As indicated in Fig. 4f, deletion of the N-terminal domain of Rnt1p, which blocks the cell cycle-dependent exit of Rnt1p from the nucleolus 29 , increased Axl2 mRNA. We conclude that repression of AXL2 requires Rnt1p to leave the nucleolus in S phase.
Live-cell analysis also indicated that RNT1 deletion perturbs the cycling of Axl2 mRNA ( Fig. 5a and Supplementary Fig. 5). The presence of the tag did not inhibit the effect of RNT1 deletion as an increase in MS2 tagged Axl2 mRNA was still observed albeit to a lower extent upon deletion of RNT1 in the tagged strain (Fig. 5a). Both nuclear and cytoplasmic RNA continued to accumulate before and after budding (Fig. 5b). However, it was not clear whether Rnt1p induced degradation of Axl2 mRNA led to changes in the abundance of Axl2 protein or affected Axl2 function. Therefore, we examined the impact of RNT1 deletion on Axl2-dependent axial budding pattern. Cells were stained using the bud scar stain calcofluor and number and position of bud scars were compared between wild type and rnt1Δ cells. As indicated in Fig. 5c, deletion of RNT1 increased the number of random budding pattern normally associated with AXL2 overexpression 18 . As expected, the defect in budding pattern was not exaggerated by multiple nuclease deletion confirming that the different ribonucleases affect budding through the same pathway ( Supplementary Fig. 9). We conclude that nuclear mRNA degradation inhibits the production of Axl2 protein to maintain budding patterns.
Rnt1p cotranscriptionally degrades Axl2 mRNA. It was previously shown that Rnt1p cleaves its substrates cotranscriptionally 16,30,31 explaining the rapid decrease in nascent AXL2 transcripts upon the exit from G1. To evaluate this possibility, we examined the association of Rnt1p with the AXL2 locus using chromatin immunoprecipitation (ChIP). As indicated in Fig. 6a, b, Rnt1p was modestly enriched in the promoter region and Rnt1p binding site but the enrichment was noticeably increased when the turn over of the enzyme is blocked (rnt1-D245R). Indeed, impairing the catalytic activity of Rnt1p increased by 3 folds the enzyme association with AXL2 chromatin (Fig. 6b probes 1 and 2). Active enzyme rapidly turnovers decreasing the stability of its association with chromatin while the catalytically impaired version does not [32][33][34] . Replacement of the stem loop structure (LoopR) inhibited Rnt1p association with AXL2 chromatin near the cleavage site, while increasing the interaction with promoter region (Fig. 6b probes 1, 3, and 4). Surprisingly, when Rnt1p binding site was replaced, Rnt1p association with AXL2 promoter and the open reading-frame 5′ end increased dramatically. This suggests that Rnt1p associates with the chromatin of AXL2 using two mechanisms: the first depends on the RNA binding site, which is linked to cleavage, and the other depends on the promoter sequence, perhaps through association with transcription factors.
To evaluate the contribution of Rnt1p to the transcription of AXL2, we monitored the impact of RNT1 deletion on RNAPII occupancy by ChIP. As indicated in Fig. 6c, deletion of RNT1 does not affect the overall level of RNAPII occupancy near the promoter and the 5′ end of the AXL2 open reading-frame (Fig. 6c  probes 1 and 2). Instead, the deletion increased RNAPII occupancy with the region surrounding the cleavage site in AXL2 RNA, suggesting that cleavage may result in RNAPII release (Fig. 6c probes 3 and 4). This is consistent with the established role of Rnt1p in inducing transcription termination 30 . Therefore, it is the cleavage by Rnt1p, rather than its association with the chromatin, that decreases RNAPII occupancy. Indeed, inactivation of Rnt1p catalytic activity increases RNAPII association with AXL2 RNA suggesting that, in the absence of RNA cleavage, association of Rnt1p with chromatin may lead to transcriptional pausing (Fig. 6b, c). The data is therefore consistent with the model whereby cleavage of RNA by Rnt1p initiates transcription termination by generating an entry point for the ribonuclease Rat1p, as previously shown for the NPL3 gene 30,31,35 .
Since Rnt1p associates with chromatin near the promoter of AXL2, we examined the role of the promoter sequence on its association with Rnt1p and on AXL2 expression. For this we used the promoter of the TOA1 gene as it is expressed at similar levels to AXL2 independently of the cell cycle [36][37][38] . Substitution of AXL2 promoter with the unrelated TOA1 promoter impaired the cycling of AXL2 expression both in the presence and absence of Rnt1p. Remarkably, expression of AXL2 from the TOA1 promoter did not decrease in any phase of the cell cycle and the deletion of RNT1 had no additional effect on the expression of the TOA1p-AXL2 gene (Fig. 7a). A ChIP assay indicated that the substitution of the AXL2 promoter inhibited the association of Rnt1p with the promoter region, but not its overall association with chromatin (Fig. 7b-d and Supplementary Fig. 10). This indicates that the promoter sequence helps recruit Rnt1p to the transcription site. Strikingly, promoter substitution also decreased the occupancy of Rnt1p at its cleavage site on AXL2 RNA indicating once again that the promoter sequence is required for cleavage by Rnt1p (Fig. 7d). We conclude that AXL2 expression is regulated by cell cycle-induced promoter-dependent RNA decay.

Discussion
It is believed that modulation of gene expression is always linked to changes in transcription rates. For example, the cell cycle is often viewed as a periodic sequence of waves of transcription of phase-specific genes 3 . Here, we challenge this notion and demonstrate that the cell cycle may modulate gene expression of constitutively transcribed genes. We have shown that the transcription of the cell cycle-regulated gene AXL2 continues despite the sharp decrease in the number of mature RNAs accumulating in S phase ( Fig. 1 and Supplementary Fig. 3). The decrease in RNA depends on a network of nuclear ribonucleases, ensuring rapid and maintained repression of AXL2 expression in S and G2/ M (Fig. 3). Surprisingly, the cell cycle-dependent degradation of Axl2 mRNA depends on the specific promoter sequence (Fig. 7). A heterologous promoter impaired the recruitment of nuclear endoribonuclease Rnt1p to the transcription start site and rendered the AXL2 insensitive to nuclear ribonucleases. These observations reveal a model of cell cycle-dependent regulation of gene expression whereby the promoter regulates RNA stability when transcription is not repressed.
Previous studies indicated that promoter sequence may control RNA stability in the cytoplasm 14,39 . In all these examples, accelerated RNA decay is coupled to transcription repression and the degradation events occur in the cytoplasm after transcription is completed. Therefore, in these examples accelerated RNA decay functions to accentuate transcriptional repression. In the case of AXL2, the role of RNA decay takes center stage as the promoter is directly responsible for the recruitment and/or activation of the RNA decay machinery to the transcription site (Fig. 7). Most importantly, there was a decrease but no arrest in AXL2 transcription as Axl2 mRNA decreases ( Fig. 1 and Supplementary  Fig. 3). Therefore, unlike the Dbf2p and Rap1p promoter-induced decay, the repression of AXL2 expression is achieved in the nucleus while the gene continues to be transcribed.
It is unlikely that this direct feedback between synthesis and degradation is restricted to AXL2. Rnt1p RNase has hundreds of potential cleavage targets in the yeast transcriptome, including in cell cycle genes 20 . Equally, there are at least two other mRNAs coding for protein required for cell polarity and bud site selection (BEM2 and BNI1) that were identified using high-throughput assay as targets for nuclear RNA degradation 20 . However, it remains to be seen if these degradation events act as the primary mechanism of gene regulation or function as fail-safe mechanisms to ensure clean transcription repression. Distinguishing between transcriptional repression and cotranscriptional RNA degradation of transcribed genes is challenging since both mechanisms reduce the amount of RNA and, as we show also, may both depend on the promoter sequence. In addition, bulk measurements of total RNA decay cannot easily differentiate between these two nuclear mechanisms since most mRNA is cytoplasmic. Live cell analysis coupled with single molecule counting and RNAPII ChIP assay, as we performed here, will be required to clearly establish the regulatory mechanism of these candidate genes. Comparing both untagged RNA FISH assay and live cells assay of tagged RNA increases confidence in the data obtained and reduces the chance of technique specific artefacts. In addition, comparing between conditions or ribonuclease deletion using the same probe set or tag reduces the chance of errors of these techniques. In any case, the findings in this study caution against the automatic interpretation of the dependency of gene expression on promoter sequence as evidence for transcription regulation. RNA stability is regulated by a large network of ribonucleases and blocking the mutation by one may activate the degradation by another. This explains why mutation of the RNA site that Rnt1p cleaves in the presence of exoribonucleases does not increase RNA expression (Supplementary Fig. 7b). RNA accumulating in the nucleus upon the mutation of Rnt1p loop are targets of the DRN or the nuclear RNA degradation mechanism that include a component of the nuclear cap binding complex (Cbc1p) and nuclear ribonucleases like Rrp6p and Rat1p 40,41 . Indeed, deletion of exoribonucleases including Rrp6p increases RNA abundance in a Rnt1p cleavage site-dependent manner (Supplementary Fig. 7b).
The data presented in this study suggest a model of cell cycledependent gene expression (Fig. 8). In this model, in the absence of Rnt1p, which is mostly nucleolar in the G1 phase of the cell cycle 29 , Axl2 mRNA accumulates and is translated in the cytoplasm. Upon the exit of Rnt1p from the nucleolus to the nucleoplasm in the S phase of the cell cycle, it is recruited to the promoter region of AXL2 to survey RNA synthesis and cleaves it upon the transcription of the cleavage signal. This cleavage leads to a sharp decrease in the number of complete RNA exported to the cytoplasm and releases RNAPII. It is not clear how Rnt1p is recruited to the promoter. We have shown that the interaction between Rnt1p and the promoter does not depend on the RNA sequence of the open reading-frame. Indeed, disrupting the RNA site that Rnt1p cleaves increases the interaction with the promoter region (Fig. 6). This reinforces a model whereby Rnt1p binds to the promoter region before translocating to the AXL2 stem loop when it is synthesized. Rnt1p may be recruited by cell cyclespecific transcription factors or directly by interaction with RNAPII itself. Protein interaction assay identified binding of Rnt1p with several transcription factors and interaction with RNAPII CTD was also proposed 30,42 . Unlike standard, transcription-based models of cell cycle-dependent regulation, this RNA decay-dependent mechanism provides faster response time permitting rapid cycling of gene expression.

Methods
Yeast strains. Yeast strains were generated in the BY background 43 starting from strain LLY36 44 using standard procedures 45 . 24xMS2 tags 23 were introduced in the 5′ UTR of AXL2 by transforming a PCR fragment amplified from plasmid pMS2-AXL2 synthesized by BioBasic. Likewise, 24xPP7 tags 23 were introduced in the 3′ UTR of AXL2 by transforming a PCR fragment amplified from plasmid pAXL2-PP7, which was synthesized by BioBasic. Strains expressing RNA tags were crossed to BY4720 43 to remove TRP1 to use plasmid pNoel-NLSred or pNoel. The LoopR mutant was produced by replacing the cleavage site of Rnt1p in AXL2 with a GFP cassette. The LoopM mutant was produced by introducing five silent point mutations in the sequence of the cleavage site of Rnt1p in AXL2 to disrupt its predicted secondary structure without changing the coding sequence. The AXL2 promoter replacement by the promoter of TOA1 was performed using a synthesized plasmid as template to amplify a replacement cassette. The chromatin was precipitated using antibodies against Rnt1p 46 and different fragments of AXL2 were amplified using qPCR analysis using AXL2 specific probes. The position of the different amplicons is illustrated in (a). c The deletion of RNT1 increases RNAPII association with the AXL2 locus. Chromatin was immunoprecipitated from the different cells described in (b) and rnt1Δ with antibodies against RNAPII and the enrichment of the different AXL2 fragments was evaluated using qPCR as described in (b). The enrichments were measured relative to a non-coding region of chrV. The bar graphs represent an average of 3 biological replicates for the wild type strains and 2 biological replicates for all other strains. The data points are shown in the form of dots Yeast cultures. All yeast strains were grown in yeast complete (YC) media lacking the appropriate nucleotide or amino acid when needed for plasmid maintenance (1.7 g/L yeast nitrogen base without amino acids and ammonium sulfate, 1 g/L sodium glutamate, 100 mg/L cysteine, 100 mg/L threonine, 85 mg/L tryptophan, 80 mg/L leucine, 60 mg/L lysine, 50 mg/L aspartic acid, 50 mg/L isoleucine, 50 mg/L methionine, 50 mg/L phenylalanine, 50 mg/L proline, 50 mg/L serine, 50 mg/L tyrosine, 50 mg/L uracil, 50 mg/L valine, 20 mg/L adenine, 20 mg/L arginine, 20 mg/L histidine, pH 6) with 2 g/L dextrose. Cultures were grown at 26°C to accommodate thermosensitive strains such as rnt1Δ and rat1-1 unless restrictive conditions were required in which case precultures grown at the permissive temperature were incubated at the restrictive temperature of 37°C for 4 h prior to cell collection. Cell cycle arrest was performed with cultures grown in YEPD (10 g/L yeast extract, 20 g/L bacto-peptone, 100 mg/L adenine, 2 g/L dextrose) at 30°C. G1 arrest was performed by adding 5 µg/ml α-Factor to the culture and monitoring the arrest by looking at the budding index of sonicated culture samples. After 3 h of treatment, >95% were un-budded and cells were collected. S phase arrest was performed by the addition of 200 mM hydroxyurea powder to the cultures and monitoring the arrest as for G1. After 3 h of treatment, >90% were budded and cells were collected. The G2/M arrest was performed by the addition of 15 µg/ml nocodazole from 1.5 mg/ml stock in DMSO to the cultures and monitoring the arrest as above. The cell cycle-dependent degradation of Axl2 mRNA is induced by its promoter. a The promoter of AXL2 is required for Rnt1p-dependent repression of gene expression. The abundance of Axl2 mRNA was monitored in wild type or rnt1Δ cells expressing AXL2 from its own promoter (WT and rnt1Δ) or a heterologous promoter (TOA1p-AXL2 and TOA1p-AXL2 rnt1Δ). The mRNA was quantified as described in Fig. 1c and is shown on the left while examples of cells captured in S and G2/M phases are shown on the right. White bar equals 2 µm. b Schematic representation of the AXL2 locus depicting the position of the different fragments amplified by qPCR after chromatin immunoprecipitation. The position of the RNA target Rnt1p cleavage site (RCS) is shown at the top and the TOA1 promoter is indicated by a gray box. The probe specific to the substituted sequence is indicated by 1′. c The promoter substitution increases the association of RNAPII with the AXL2 locus. The RNAPII association pattern was examined in wild type cells (WT) or cells expressing AXL2 from a heterologous promoter (TOA1p-AXL2) as described in Fig. 6c. d Substitution of the AXL2 promoter inhibits the recruitment of Rnt1p to the AXL2 locus. Rnt1p association with the AXL2 locus, expressed from a wild-type or heterologous promoter, was examined as described in Fig. 6b with primers for amplicon 1 specific to the AXL2 or TOA1 promoter as required. Shown is the enrichment of Rnt1p relative to RNAPII. The bar graphs represent the average values obtained from 3 independent biological replicates shown as dots. Asterisks indicate significant difference between strains (*p < 0.05, **p < 0.01, ***p < 0.001 by two-tailed unpaired t test) bathocuprioinedisulphonic acid followed 10 min later by the addition of 10 µg/ml thiolutin with 150 nM CuSO 4 . Samples were collected 0, 5, 10, 20, 40 min after the addition of thiolutin.
Single-molecule fluorescence in situ hybridization. The yeast smFISH protocol was performed essential as previously described 21  Live cell imaging. Yeast cultures expressing MS2 repeats in the 5′ UTR of Axl2 mRNA and containing a plasmid encoding MCP-GFP and NLS-td-tomato were grown to an OD 600 of 0.4-0.6, concentrated by centrifugation to an OD 600 of 2 and spotted on 1.2% agarose pads made with growth media. Cells expressing MS2-AXL2 or AXL2-PP7 and with a plasmid containing MCP-GFP and PCP-td-tomato were grown and prepared likewise. The coverslips were sealed with petroleum. Slides were visualized in a 26°C incubation chamber on a Zeiss spinning disk microscope with an oil immersion alpha Plan-Apochromat 100×/1.46 objective. An Evolve 512 camera with EM gain set at 500 was used and z-stacks of 5-7 images spaced by 500 nm were collected using the Zen 2012 software. GFP was illuminated for 100 ms with a 100 mW 488 nm laser set at 10-20% through a 500-550 nm emission filter. Td-tomato was illuminated for 100 ms with a 40 mW 561 nm laser set at 20% through a 598-660 nm emission filter. The stacks of images were converted to 2D images by maximum projection and exported with the Zen 2012 software. RNA molecules were counted manually using the profile analysis tool in Zen 2012 on maximum projected images and scored when spots were twice as bright as the cellular background. Brightness and contrast were adjusted using Photoshop CS6 to set a common background level to the time series, for display purposes. The arbitrary cut-off was determined by contrast adjustment of live cell images to identify spots comparable in size to those observed by FISH. To compensate for the higher background of the live cell images in the GFP channel  Chromatin immunoprecipitation. 100 ml cultures were grown to OD 600 0.4-0.6 and fixed for 20 min at room temperature by the addition of 1% formaldehyde to the culture. The crosslinking reaction was stopped by the addition of 3.6 mM unbuffered Tris and 360 mM glycine. The cell pellets were washed in ice-cold TBS (20 mM Tris, 150 mM NaCl, pH 7.5), split into two aliquots and kept frozen at −80°C. Frozen pellets were resuspended in 500 µl lysis buffer (50 mM HEPES-KOH pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% Na-deoxycholate) with proteases inhibitors (1 mM PMSF, 1 mM benzamidine, 1 µg/ml Aprotinin, 1 µg/ml Leupeptin, 1 µg/ml Pepstatin A, 1 µg/ml Antipain) and 500 µl acid-washed glass beads, and lysed using a Bertin Precellys 24 with 6 rounds of 10 s of shaking at 6500 rpm, spaced by 60 s pauses on ice. The lysates were adjusted to 750 µl with lysis buffer and sonicated 12 times for 10 s at 20% power amplitude with a Branson digital sonifier spaced by 2 min pauses on ice. The sonicated lysates were cleared by centrifugation at 13,000g for 10 min at 4°C and 500 µl of cleared lysates were mixed with magnetic beads coated with the appropriate primary antibody and incubated overnight at 4°C with rotation. For RNAPII ChIP, 50 µl of Pan mouse IgG Dynabeads (ThermoFisher Scientific, #11041) were coupled to 3 µg of purified mouse anti-RNA Polymerase II RPB1 antibody (BioLegend, #664906, clone 8WG16). For Rnt1p ChIP, 30 µl of M-280 sheep anti-rabbit IgG Dynabeads (ThermoFisher Scientific, #11203D) were coupled to 3 µl of pre-cleared anti-Rnt1p antibody. The next day, the beads were washed twice with lysis buffer, twice with high salt buffer (50 mM HEPES-KOH pH 7.5, 500 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% Na-deoxycholate), twice with wash buffer (10 mM Tris pH 8, 250 mM LiCl, 1 mM EDTA, 0.5% Igepal CA-630, 0.5% Na-deoxycholate) and once with TE (10 mM Tris pH 8, 1 mM EDTA). The de-crosslinking and elution of DNA was performed by re-suspending the beads in 100 µl TE containing 1% SDS and incubating overnight at 65°C. Input DNA was de-crosslinked at the same time by incubation of samples of the cleared lysate diluted in TE containing 1% SDS overnight at 65°C. The DNA was purified by RNase A and proteinase K treatments followed by 2 phenol:chloroform:isoamyl alcohol (25:24:1) extractions, chloroform:isoamyl alcohol (24:1) extraction, and ethanol precipitation with glycogen. The purified DNA pellets were resuspended in 10 mM Tris pH 8 and further diluted in water at 1:200 for the input DNA, 1:20 for the RNAPII ChIP, 1:10 for the Rnt1p ChIP.
Quantification and statistical analysis. smFISH analysis: smFISH experiments were performed on 3 independent cultures and a target of 50 cells per phase were scored coming from each culture. In the figures, all data points are presented in the form of violin plots unless stated otherwise in the legends and Supplementary Data 1. In Supplementary Fig. 3, the frequency plots show the mean from the 3 cultures (each 50 cells). FISH analysis was performed on the maximum projection of the raw images adjusting for background as described below. 2D images (maximum projections) were imported on a Perkin Elmer Columbus server and analyzed as follows. The nuclei were detected using the DAPI staining and cytoplasms were identified from the cellular background using the orange filter. Cells touching the borders of the image as well as cells smaller than 2 µm 2 or larger than 20 µm 2 were removed, while cells with a nucleolar pre-rRNA control signal were kept. Spots were detected and quantified for the 5′ and 3′ probes, then their intensity (corrected to local background) and localization data was exported in a table format. After analysis of the 2D images, cells were manually assigned to G1, S, or G2/M phases of the cell cycle according to the bud size and nucleus position. The value of fluorescence intensity corresponding to 1 RNA molecule was established for each strain by taking the median value of the cytoplasmic RNAs of for all its dataset. The RNA molecule number per spot was calculated by dividing the spot fluorescence intensity by the value corresponding to 1 RNA molecule and rounding to the closest integer. As the primary object detection was based on nuclei identification, many G2/M cells had to be reconstructed from two cells identified from dividing nuclei. Nascent RNA per cell numbers were calculated from the brightest 5′ probes spot in the nucleus when more than 1 nuclear spot was detected. Cytoplasmic RNAs were scored when a spot detected with the 5′ probe also scored positive for the 3′ probe signal.
RT-qPCR and chromatin immunoprecipitation analysis: Relative mRNA quantification was achieved through normalization to the Act1 mRNA using the formula: RQ = 2(CT WT −CT mutant ) Axl2 /2(CT WT −CT mutant ) Act1 , where CTs are the cycle threshold values determined by the Mastercycler software. PCR reactions were performed in technical duplicates or triplicates for cDNA produced from 3 independent cultures for both the wild-type (WT) and mutant strains. Relative enrichment of RNAPII or Rnt1p at the AXL2 locus was achieved through normalization to a region of the chromosome V using formula: RE = 2 (CT input −CT ChIP ) Probe /2(CT input −CT ChIP ) ChrV . PCR reactions were performed in technical duplicates or triplicates for ChIP performed at least 2 times in independent cultures.
Statistics and reproducibility: Statistical analyses of mean values of measurements were conducted using GraphPad Prism version 7.03 for Windows using Student's t test if so stated in the figure legends with Welch's correction for FISH data comparison, where variance between strains was unequal. The number of cells analyzed per FISH experiment, means and fold changes to wild-type with pvalues are reported in Supplementary Data 1. Experiments were repeated 3 times unless stated otherwise.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.