RNA Pol II transcription termination can occur by at least two alternative pathways. Cleavage and polyadenylation by the CPF/CF complex precedes mRNA transcription termination, while the Nrd1 complex is involved in transcription termination of non‐coding RNAs such as sno/snRNAs or cryptic unstable transcripts. Here we show that transcription of RPL9B, one of the two genes coding for the ribosomal protein Rpl9p, terminates by either of these two pathways. The balance between these two pathways is modulated in response to the RPL9 gene copy number, resulting in the autoregulation of RPL9B gene expression. This autoregulation mechanism requires a conserved potential stem‐loop structure very close to the polyadenylation sites. We propose a model in which Rpl9p, when in excess, binds this conserved 3′‐UTR structure, negatively interfering with cleavage and polyadenylation to the benefit of the Nrd1‐dependent termination pathway, which, being coupled to degradation by the nuclear exosome, results in downregulation of RPL9B gene expression.
In eukaryotes, regulation of gene expression can occur at virtually any level, from transcription to processing, nuclear export, RNA stability, translation and protein stability. In metazoans, a large part of the protein diversity is generated at the level of pre‐mRNA maturation. It is well recognized that alternative splicing plays a major role in the regulation of gene expression (Keren et al, 2010; Witten and Ule, 2011). Likewise, the well‐known involvement of alternative cleavage and polyadenylation sites (poly(A) sites) usage in gene regulation (see Edwalds‐Gilbert (1997) for review) has recently been underscored by the finding that it is a very widespread mechanism important for normal cell development as well as cancer progression (Mayr and Bartel, 2009).
mRNA 3′‐end formation involves the RNA polymerase II (Pol II)‐associated cleavage and polyadenylation complex (CPF/CF complex), a large complex composed of more than 20 proteins in yeast (see Kuehner et al (2011) for review). The poly(A) site sequences recognized by CPF/CF consist of loose composite consensus sequences (at least four separate degenerated motifs define a poly(A) site in yeast; Graber, 2002). The 3′‐end maturation process of mRNAs is initiated by the recognition of cis‐acting sequences (the poly(A) sites) on the nascent transcript by components of the CPF/CF complex that triggers endonucleolytic cleavage of the RNA. The 5′‐fragment of the RNA is polyadenylated and exported to the cytoplasm, whereas the 3′‐fragment is rapidly degraded by the 5′‐ to 3′‐exonuclease Rat1 (Xrn2 in human). In the torpedo model, termination is proposed to occur when Rat1, which attacks the 5′‐end generated by the cleavage of the nascent transcript, ultimately catch up with Pol II, resulting in its dissociation from the DNA template (Kim et al, 2004). The efficiency of termination correlates well with the strength of the poly(A) site. 3′‐end maturation of the mRNAs by cleavage and polyadenylation is thus intimately coupled to transcription termination. Yet, there is a second mode of transcription termination, which is carried out by another distinct Pol II‐associated complex comprised of Nrd1p, Nab3p and Sen1p (Nrd1 complex). Nrd1p and Nab3p are two RNA‐binding proteins that recognize GUAA/G and UCUU sequences, respectively, whereas Sen1p is an RNA helicase. The Nrd1 complex is required for the termination and synthesis of stable non‐coding RNAs, such as snRNAs and snoRNAs (Steinmetz et al, 2001), as well as unstable pervasive RNAs called CUTs for cryptic unstable transcripts (Arigo et al, 2006; Thiebaut et al, 2006). As for the CPF/CF complex, the Nrd1 complex interacts with the carboxy‐terminal domain (CTD) of the large subunit of Pol II (Vasiljeva et al, 2008). Although the actual mechanism of termination by the Nrd1 complex has not been shown, one idea is that the simultaneous binding of the Nrd1 complex to the nascent transcript and to Pol II could trigger the unwinding by Sen1p of the RNA/DNA duplex within the transcriptional R‐loop, resulting in the dissociation of Pol II. Moreover, the Nrd1 complex also interacts with the nuclear exosome and couples this mode of transcription termination to 3′‐ to 5′‐exonucleolytic trimming/degradation (Vasiljeva and Buratowski, 2006). In the case of snRNAs and snoRNAs, co‐transcriptionally assembled snRNP and snoRNP proteins restrict exonucleolytic trimming, which thus constitutes one of the maturation steps of these RNPs. In the case of CUTs, however, the exonucleolytic degradation of the RNA is not impeded by proteins, resulting in the complete degradation of the transcripts. These RNAs thus exhibit instability, being degraded as soon as their transcription terminates, hence their ‘cryptic’ nature since they never accumulate in wild‐type cells, at least under standard laboratory growth conditions (Wyers et al, 2005). Because of the pervasive nature of Nrd1/Nab3 recognition sequences, this degradation‐coupled termination mode is thought to constitute a ‘default’ transcription termination mode able to target undesired or spurious transcripts to degradation.
While the CPF/CF preferentially binds the CTD of the Rpb1p subunit of Pol II in its serine 2 phosporylated form, Nrd1p binds the same domain when it is phosphorylated on serine 5. This modification, present at the beginning of transcription elongation, decreases, concomitantly with an increase of serine 2 phosphorylation, while Pol II synthesizes the first few hundred nucleotides of RNA (Gudipati et al, 2008; Vasiljeva et al, 2008). This is consistent with the Nrd1 complex being dedicated to transcription termination of short non‐coding RNA. Indeed, very few examples have been reported that suggest the implication of the Nrd1 pathway in the termination of normal mRNAs in wild‐type cells. The Nrd1 termination mode has been reported to be able to serve as a ‘fail‐safe’ mRNA transcription termination mechanism in a rat1 mutant background in which the poly(A) site‐associated transcription termination mechanism is deficient or when the main poly(A) site of an mRNA has been deleted (Rondón et al, 2009). The Nrd1 mode of termination is then able to efficiently compete with the use of cryptic, weak poly(A) sites. The transcripts resulting from such aberrant events are unstable. However, it has recently been reported that, in a few cases, the Nrd1 termination pathway could be involved in termination of some mRNAs. Indeed, the recent genome‐wide mapping of Nrd1‐Nab3 crosslink sites has revealed the presence of these factors 3′ to several genes (such as AHP1, URA1 and SNA3), suggesting a role for the Nrd1 pathway in termination of their transcription (Jamonnak et al, 2011). Furthermore, CTH2 mRNA synthesis has been shown to involve a peculiar pathway. Transcription terminates in a Nrd1‐dependent manner, generating a 3′‐extended pre‐mRNA, subsequently trimmed by the nuclear exosome that pauses at a [GUn]5 site before the RNA gets polyadenylated by Pap1, leading to the production of a mature CTH2 mRNA (Ciais et al, 2008).
Here, we show that the ribosomal protein gene RPL9B is autoregulated by a novel mechanism that takes advantage of the ability of Pol II to terminate transcription by these two alternative modes, the CPF/CF mode that gives rise to stable mRNAs and the Nrd1 mode that generates unstable transcripts. Indeed, we show that RPL9B transcription naturally terminates by these two alternative pathways and is regulated by the modulation in the choice between these two alternate modes. This modulation requires the presence of a conserved sequence, potentially forming a stem‐loop structure, which is in very close proximity to the poly(A) sites. These observations suggest a model in which Rpl9p, when present in excess, binds, directly or indirectly, to this conserved hairpin, hampering the recognition of the poly(A) sites to the benefit of the use of downstream Nrd1/Nab3 termination sites, which trigger degradation‐coupled transcription termination, hence ensuring gene repression.
RPL9B generates, in addition to canonical poly(A) mRNAs, transcripts with the characteristics of CUTs
Previous genome‐wide analyses determined, at the nucleotide level, the 3′‐ends of both wild‐type poly(A) transcripts (Neil et al, 2009; Ozsolak et al, 2010) and transcripts from an RNA fraction highly enriched for CUTs (Neil et al, 2009). CUTs are normally highly unstable as a result of their Nrd1/Nab3 mode of transcription termination, which is coupled to fast nuclear degradation by the TRAMP–exosome complex (Jacquier (2009) for review). The CUT‐enriched fraction was obtained by selecting for nuclear capped RNAs (RNAs co‐precipitated with the affinity‐purified nuclear cap‐binding factor Cbp20p) from cells in which the CUTs are stabilized by hampering TRAMP and nuclear exosome activities (depleted‐Trf4, Δrrp6). For a number of mRNAs, 3′‐end tags could be found in the CUT fraction as well as in the poly(A) fraction. The tags recovered in both fractions essentially mapped to the same sites in both samples, suggesting that they simply represented the nuclear and the cytoplasmic fractions of normal mRNAs, respectively. RPL9B, however, behaved differently, its 3′‐end tags clustering at well distinct positions in the poly(A) and CUT fractions (Figure 1A). The poly(A) tags cluster into two main regions, separated by about 80 nucleotides, thus defining two transcripts, here called #1 and #2 (Figure 1B), corresponding to two main poly(A) sites. The precise mapping of these poly(A) sites, combining the 3′‐tags from Neil et al, 2009, the direct RNA sequencing tags (DRS tags) from Ozsolak et al (2010), as well as 3′‐RACE and RNase H mappings performed in this study, are shown in Supplementary Figure S1. In the CUT fraction, however, a different 3′‐tag cluster was found yet about 100 nucleotides downstream of the poly(A) sites. When RNAseq (performed according to Nagalakshmi et al (2008); Figure 1A) and RNase H mapping (Figure 1B) were used to analyse the 3′‐ends of RPL9B RNAs in a Δrrp6 strain, it confirmed that the 3′‐end cluster found in the CUT fraction corresponded to an extended form of the RPL9B gene transcript (Figure 1B, transcript #3; compare lanes1 and 2). That this extended form is absent in a wild‐type strain, but revealed by compromising the TRAMP‐nuclear exosome activities (Neil et al, 2009), suggests that it is derived from a Nrd1/Nab3‐dependent transcription termination (which is coupled to nuclear exosome degradation). Also consistent with this hypothesis, these extended RRP6‐sensitive transcripts terminate just after a region of the sequence harbouring a high density of Nrd1 and Nab3 recognition sequences (Figure 1A; green tags) that coincides with a high‐density cluster of Nrd1 and Nab3 crosslink sites (Figure 1A), according to Jamonnak et al, (2011).
RPL9p regulates the levels of endogenous RPL9B mRNAs
Like the majority of the ribosomal proteins in Saccharomyces cerevisiae, Rpl9p is encoded by two gene copies, namely RPL9A and RPL9B. Rpl9p is part of the large (60S) ribosomal subunit, and Rpl9Ap and Rpl9Bp are nearly identical (98.5%). In the absence of the RPL9A gene, the RPL9B transcripts #1 and #2 accumulate relative to a wild‐type strain (Figure 1B; compare lanes 1 and 3). This indicates that RPL9B expression is autoregulated, the amount of its mRNAs increasing in a Δrpl9A strain to compensate for the decrease in the gene copy number. In contrast, transcript #3 behaves just in the opposite way, being greatly diminished in the Δrpl9A, Δrrp6 strain relative to a Δrrp6 strain (compare lanes 2 and 4). Thus, accumulation of transcripts #1 and #2 appears to occur at the expense of transcript #3. Note also that RPL9A expression is not affected by the deletion of RPL9B (data not shown) and, thus, only one of the two duplicated genes is autoregulated.
The autoregulation of RPL9B is mediated by its 3′‐UTR
The results described above, which show that transcripts #1 and #2 are regulated in a way opposite to transcript #3, suggest that the autoregulation mechanism might operate by modulating the balance between the CPF/CF‐dependent transcription termination, giving rise to stable transcripts, and the Nrd1/Nab3 mode of termination, which generates highly unstable transcripts. According to this model, the cis‐regulatory elements would be expected to be located in the 3′‐end region of the gene in order to modulate transcription termination.
To directly test this hypothesis, we used a CUP1 reporter gene. Cup1p is a metallothionein that confers copper resistance to yeast (Winge et al, 1985). When the endogenous CUP1‐1 and CUP1‐2 genes are deleted, the cells are unable to grow on copper‐containing media (data not shown). This copper sensitivity can be complemented by transformation with a plasmid harbouring the CUP1 ORF, expressed from an exogenous TEF promoter and with a CYC1 transcription terminator (data not shown), as expected for a strain properly expressing Cup1p. However, when the 830 nucleotides downstream region of the RPL9B ORF (RPL9B 3′‐UTR) was inserted between the CUP1 ORF and the CYC1 terminator (Figure 2A), the cells were no longer able to grow in the presence of 0.2 mM CuSO4, showing that an element in this 830 nucleotide sequence was repressing the proper expression of the reporter gene (Figure 2B, CUP‐3′UTR plasmid in a WT strain). Deletion analyses showed that the first 308 nucleotides of these RPL9B downstream sequences were necessary and sufficient to elicit this repressive effect (data not shown). We then analysed this CUP1 construct in different genetic backgrounds. As can be seen on Figure 2B, the repressive effect exerted by the RPL9B 3′‐sequences on CUP1 expression is mostly obliterated when the RPL9 gene copy number is reduced from two to one by either deleting RPL9A or RPL9B. Note that, in the Δrpl9A strain, the endogenous RPL9B should still be autoregulated. Yet, the autoregulation of RPL9B expression is clearly not able to fully compensate for the RPL9A deletion in term of total Rpl9p level since there is still sufficient derepression of the CUP1‐3′UTR reporter to allow growth on the copper‐containing medium, although slightly less efficiently than in the Δrpl9B strain. Finally, and as expected, overexpression of Rpl9p from a plasmid fully restores repression of the CUP1‐3′UTR reporter, as judged by the total inability of the cells to grow on copper‐containing medium. Thus, fusing the 3′‐UTR of RPL9B downstream of a CUP1 gene is sufficient to recapitulate the autoregulation of RPL9B on this reporter gene. In addition, that the RPL9B 3′‐UTR is sufficient to allow regulation of CUP1 in response to the RPL9 gene copy number, even though it is expressed from an exogenous TEF promoter, is consistent with the hypothesis that RPL9B autoregulation is a post‐transcriptional mechanism.
Finally, deletion of RRP6 partially restores cell growth in the presence of copper (Figure 2B; Δrrp6 strain), indicating that at least part of the CUP‐3′UTR transcripts are normally degraded in the nucleus, consistent with the results obtained for the endogenous RPL9B gene (Figure 1B, lane 2). Indeed, analysis of the 3′‐ends of the CUP‐3′UTR transcripts by RNase H digestion followed by northern blot (Figure 2C) gave results very similar to those obtained for the endogenous RPL9B gene (Figure 1B).
Transcript #3 terminates by the Nrd1–Nab3 pathway
The initial observation that transcript #3 is sensitive to Rrp6p and found only in the CUT‐enriched fraction (Neil et al, 2009), together with the fact that it terminates shortly after a cluster of Nab3‐binding sites and Nrd1/Nab3 crosslink sites (Jamonnak et al, 2011), strongly suggested that its transcription is terminated by the Nrd1 pathway rather than the CPF/CF pathway. In order to directly test this hypothesis, we placed NAB3 under the control of a galactose‐inducible promoter (Pgal‐NAB3) and NRD1 under the control of a tetracycline‐repressible promoter (PTetO7‐NRD1), either in a wild‐type or a Δrrp6 background. Depletion of Nab3p alone (shift from galactose‐ to glucose‐containing medium for 9 h—Figure 3A) or both Nab3p and Nrd1p (shift from galactose to glucose plus doxycycline‐containing medium for 9 h; not shown) both lead to a decrease in the steady state level of transcript #3 (compare lanes 2 and 4 in Figure 3A). This diminution coincided with the appearance of a new band (labelled #4 on Figure 3A) that was revealed by RNase H digestion of the samples with an additional oligonucleotide (RG055) located downstream of the Nrd1–Nab3‐binding sites in the RPL9B 3′‐UTR. This new band must result from 3′‐readthrough transcripts that give rise to a single species upon RNase H cleavage of their 3′‐heterogenous distal sequences. Most importantly, band #4 is insensitive to the absence of Rrp6p, fully consistent with it corresponding to long transcripts that terminate in a Nab3p‐independent manner.
In order to get an independent confirmation of the involvement of the Nrd1–Nab3 pathway in the transcription termination of transcript #3, we mutated all the putative Nrd1‐ and Nab3‐binding sites (guaa/g, and ucuu sequences according to Carroll et al, 2004)within the 3′‐UTR of the CUP‐3′UTR reporter fusion (mutated version named CUP‐ΔBS; see Supplementary Figure S2 A for the precise nature of the 11 substitutions introduced). Surprisingly, northern blot analysis did not reveal obvious effect of the substitutions on the amount of transcript #3 (compare lanes 2 to 6 or 4 to 8 in Supplementary Figure S2B; the only noticeable effect was the diminution of transcript #1, which can be explained by the fact that one of the substitution in a potential Nab3 site resides precisely at the major poly(A) site of transcript #1 – see Supplementary Figures S1 and S2). Nevertheless, RNase H digestion in the presence of oligonucleotide RG055 revealed a readthrough band (#4) specific for the CUP‐ΔBS construct (lanes 14 and 16 in Supplementary Figure S2B). This shows that a fraction of the transcripts escape Nrd1–Nab3 termination at site #3 in the mutant. These readthrough transcripts, that give rise to band #4, are reduced upon deletion of RPL9A, along with transcript #3, consistent with the idea that they constitute readthrough of the latter. Of note, band #4 is stabilized by the deletion of RRP6 (in contrast to band #4 in Figure 3A), which means that the corresponding readthrough transcripts likely terminate at downstream Nrd1–Nab3 sites rather than at a cryptic poly(A) site. The observation that the effect of the substitutions is so weak likely illustrates the fact that the actual Nrd1‐ and Nab3‐binding site consensus sequences remain very poorly defined and is consistent with similar observations made on CUTs (FF and AJ, unpublished results).
Finally, we analysed how mutations in components of the CPF/CF complex affected the different RPL9B transcripts species. Mutations in RNA14 or RNA15 have been shown to lead to a transcriptional readthrough of termination sites (Mandart and Parker, 1995), the resulting extended RNAs being usually rapidly degraded in the nucleus (Libri et al, 2002). We thus analysed the effects of the temperature sensitive (ts) mutants rna14‐3 and rna15‐2 (Minvielle‐Sebastia et al, 1991)on the RPL9B transcript profile, either in an RRP6 wild‐type or an Δrrp6 background. Both rna14‐3 and rna15‐2 exhibited identical results (Figure 3B and data not shown). At the permissive temperature (25°C), the rna14‐3 mutation already exerted a marked effect on cleavage and polyadenylation, with significant readthrough of poly(A) site #1, resulting in a significant decrease in the amount of the corresponding transcript and a concomitant increase of transcript #2 (compare lanes 1 and 2). This suggests that this mutation, at 25°C, impairs termination at site #1 without significantly affecting termination at site #2. After a 30 min shift to the semi‐restrictive temperature (35°C), both transcripts #1 and #2 are decreased (compare lanes 5 and 6) and part of the termination probably occurs further downstream in a Nrd1–Nab3‐dependent fashion, as evidenced by the increase of transcript #3 in the Δrrp6 background (compare lanes 7 and 8) and by the fact that no band corresponding to longer transcripts (such as band #4 in Figure 3A) was revealed after RNase H digestion in the presence of oligonucleotide RG055 (not shown).
Altogether, these data confirm the hypothesis that transcript #3 is terminated by the Nrd1–Nab3 pathway, while transcripts #1 and #2 are terminated by the normal pathway for mRNAs, that is, the CPF/CF pathway.
A potential conserved stem‐loop structure within the 3′‐UTR is essential for the autoregulation
In order to identify functional elements within the RPL9B 3′‐UTR, we searched for sequences that would be conserved among related yeast species. This revealed a region of conservation, located between about 155 to 200 nucleotides from the end of the ORF (Figure 4A), which has the potential to form a stable hairpin (Figure 4B). Interestingly, many nucleotide differences between yeast species (in red in Figure 4B) exhibit compensatory base changes, a signature for the functional nature of this potential structure. Moreover, and importantly, deletion of this conserved sequence (see Figure 4C for the precise nature of the mutation) in the CUP‐3′UTR reporter construct (CUP‐Δhairpin) abrogates the repressive effect of the RPL9B 3′‐UTR, allowing proper Cup1p expression and growth on the copper‐containing medium (data not shown)and totally abolishes the ability of theCUP1 reporter transcript to respond to RPL9 gene copy number variations (Figure 4D and data not shown). RNase H mapping of 3′‐ends of the corresponding mRNAs shows that deletion of this potential hairpin generates a major stable transcript whose 3′‐end coincides with that of the endogenous RPL9B transcripts #1, as determined by 3′‐RACE (Supplementary Figure S1). This shows that deletion of the potential hairpin not only makes the gene refractory to Rpl9p‐dependent repression, but also transforms the weak poly(A) site of transcript #1 into a strong poly(A) site.
To further analyse the sequence elements involved in the feedback repression by the 3′‐UTR, we carried out random PCR mutagenesis of the RPL9B 3′‐sequences. The mutagenized RPL9B3′‐UTRs were cloned between the CUP1 ORF and the CYC1 terminator in the reporter construct (see Figure 2A) and selected for their ability to restore growth of cup1‐1 cup1‐2 cells on copper‐containing media with different molar concentrations of CuSO4. The selected clones were sequenced to determine the specific mutations that abrogated the repression and only the clones bearing a single mutation were taken into account. With the exception of a U to G mutation, four nucleotides upstream of the hairpin, 11 out of the 12 mutations that most strongly abrogate the repression (in red in Figure 4C) are located within the conserved stem‐loop structure. Mutations exhibiting weaker phenotypes were also found within the hairpin and up to 29 nucleotides upstream (in orange for the mutants moderately affected and in light blue for the mutants weakly affected, as determined by growth in the presence of various CuSO4 concentrations; data not shown). For some representative mutants, we analysed the steady state levels of the reporter transcripts by RNase H‐mediated 3′‐end analysis to determine whether the loss of repression correlates with changes in the abundance of reporter transcripts (Figure 4E). This gave rise to a more complex picture than expected. In some of the mutants, a simple increase in the amount of transcripts #1 and #2 was observed, as previously observed in the Δrpl9A strain (Figure 1B). This is in particular true for nucleotides that are not involved in Watson–Crick base pairing, such as mutants number 2 or 12. This is the behaviour expected for mutants that simply affect binding of the hairpin by a trans‐acting factor (possibly Rpl9p; see below). In contrast, the other mutants reveal an additional termination site, located a few nucleotides upstream of the 3′‐end of transcript #1 (labelled with a star on Figure 4E). While this novel 3′‐end was not mapped at the nucleotide level, the RNase H experiment locates it within a few nucleotides upstream of the 3′‐end of transcript #1, indicating that this new transcript terminates in the hairpin. Strikingly, all these mutants, with the exception of mutant number 11, disrupt a Watson–Crick base pair within the structure. Note that mutant 9, which, while involving a paired nucleotide, is conservative with respect to base pairing, substituting a G:C base pair by a G:U base pair, does not induce this new 3′‐end. This suggests that the destabilization of the hairpin might unmask a cryptic poly(A) site, not normally accessible. This effect on the derepression of the CUP1 reporter might thus be somewhat indirect. This is true for example for mutants 6, 7 and 10, which, apart from this new 3′‐end, do not show a marked increase of transcripts #1 and #2. Yet, many mutants, such as mutants 3 and 4, show both the new 3′‐end and also a strong increase of transcripts #1 and #2, suggesting that the corresponding nucleotides have a dual role, in maintaining the integrity of the structure and also allowing the recognition by a trans‐acting factor (although we are aware that the two effect are probably not independent from each other). In these examples, the mutation might not totally destabilize the hairpin, which thus would be in equilibrium between a folded and an unfolded state. In the folded state, the mutation might impair the binding of the trans‐acting factors, while in the unfolded state, it would unmask the cryptic poly(A) site. Only mutant number 11 does not fit well in this picture. It does very significantly increase the amount of transcripts #1 and #2, suggesting that it is involved in the recognition by a trans‐acting factor, yet it is not well conserved. Moreover, it induces the formation of the new 3′‐end, suggesting that it affects the stability of the hairpin, yet it is located outside the stem‐loop structure. Several hypotheses could explain these observations. For example, this mutation could favour alternative structures, competing with the formation of the hairpin. Folding programmes could not reveal obvious competing structures, but these predictions are not fully reliable, not taking into account tertiary interactions for example. Finally, weak mutations, located outside of the hairpin (number 5 and 8), also increase transcripts #1 and #2 (although more modestly than, for example, mutant 2, as expected from their weaker phenotype). This could suggest that the cis‐regulatory elements recognized by trans‐acting factor(s)might extend significantly upstream of the hairpin, although this region is outside the conserved sequences. Here again, other explanation could explain the phenotype. For example, these mutations could affect the overall stability of transcripts #1 and #2 in the cytoplasm, in a manner independent of the amount of Rpl9p. Clearly, more detailed work will be required to fully decipher all the parameters involved in the regulation. Yet, these experiments show that very few sequences, if any, are involved in the repression outside the hairpin, the deletion of which totally abrogates the regulation and where the vast majority of the single mutations were recovered.
In this work, we show that RPL9B transcription termination occurs by two alternative pathways, the CPF/CF pathway, which gives rise to normal polyadenylated mRNAs, and the Nrd1 pathway, which is coupled to rapid nuclear degradation involving the nuclear exosome factor Rrp6p. We further show that RPL9B expression is autoregulated and that this feedback loop involves a shift in the balance between the uses of the two alternative modes of termination. We show that fusing the 3′‐UTR region of RPL9B to a CUP1 reporter transcript is sufficient to recapitulate this regulation on CUP1 expressed from an exogenous promoter, which is fully consistent with this autoregulation being a post‐transcriptional process. We finally show that the RPL9B 3′‐UTR harbours an evolutionary conserved sequence potentially forming a stem‐loop structure, which is essential for the regulation and which overlaps or is very close to the two poly(A) sites.
From these observations, we would like to propose the model depicted in Figure 5. The key feature of the model is that the autoregulation of RPL9B expression results from the modulation of the balance between a productive mode of transcription termination, which involves the normal CPF/CF mRNA termination pathway, and the degradation‐coupled Nrd1/Nab3‐dependent mode of termination, normally associated with the transcription of non‐coding RNAs such as CUTs or sn/snoRNAs. Two elements are key in the model, a cis‐acting element, the conserved hairpin, which overlaps or is very close to the poly(A) site recognition sequences, and a trans‐acting factor that, by interacting with the hairpin, inhibits the recognition of the poly(A) site signal sequences by the CPF/CF complex. At this stage, the most parsimonious hypothesis is that this trans‐acting factor is Rpl9p itself, which has the required property of being an RNA‐binding protein. This model is reminiscent of the mode of regulation previously proposed for RPS28B (Badis et al, 2004). Rps28p binds a conserved hairpin within the 3′‐UTR of its mRNA, which promotes the cytoplasmic mRNA decapping and degradation of the RNA in an Edc3‐dependent manner. This model is also reminiscent of the mode of regulation of NPL3 (Lund et al, 2008). The NPL3 locus produces at least three different RNA species with varying 3′‐UTR lengths. The presence of an excess of phosphorylated Npl3p promotes the use of the most downstream poly (A) site, which results in the degradation of the transcripts by both a cytoplasmic and a nuclear degradation pathway, although not involving the Nrd1/Nab3/Sen1 complex.
Despite repeated efforts by various strategies, we were unable to biochemically demonstrate the interaction between the conserved hairpin and Rpl9p. Three‐hybrid did not give convincing results either. This suggests that the interaction between Rpl9p and the RPL9B 3′‐UTR is transient or that factors other than Rpl9p are required for this interaction to occur, similar to what has been proposed for Npl3p (Lund et al, 2008). Note also that, as for Rps28p, we could not find, within the region of the 25S rRNA secondary structures bound by Rpl9p (as determined from the crystal structure; Ben‐Shem et al, 2010), any convincing resemblance with the conserved RNA hairpin within the RPL9B 3′UTR. The same observation was made for Rps14B, which has been shown to directly bind a conserved structure of its own mRNA, structure that does not bear any obvious similarities with its known ribosomal RNA‐binding RNA structure (Fewell and Woolford, 1999).
Whatever the trans‐acting binding factor is, another parameter that might be essential for such a mechanism to be functional is its ability to efficiently compete, kinetically speaking, with the recognition of the poly(A) site elements by the CPF/CF machinery. The conserved hairpin might simply be located sufficiently upstream of the poly(A) site, such that, being transcribed significantly before the poly(A) site recognition sequences, it would be bound by the trans‐acting factor before cleavage and polyadenylation happen. But then, the bound protein might be too far to interfere, by simple steric hindrance for example, with poly(A) site recognition. Thus, the model must accommodate two seemingly contradictory parameters, first the close proximity between the conserved structure and the poly(A) site, second the competition between this factor and CPF/CF for binding to their respective target sequences. Indeed, the hairpin overlaps poly(A) site #1 recognition sequences (a potential Positioning Element, or Element 2, of poly(A) site #1, AAUAAA, resides within the loop of the hairpin), while the most probable Element 1 of poly(A) site #2 (UGCAUA according to the probabilistic model; Graber, 2002) is separated from the hairpin by only 35 nucleotides, which could be sufficient to allow interference between Rpl9p and CPF/CF binding. One could thus assume that the competing poly(A) site should be intrinsically weak. Since we have shown that transcription of RPL9B terminates at two poly(A) sites, the first one, which is the closest to the potential hairpin, must be weak to allow significant read through down to the second site. According to a probabilistic poly(A) site prediction model (Graber, 2002), site #1 would indeed be a very weak site, while site #2 would be stronger. This is also confirmed by the effect of the rna14‐3 mutation, which is stronger on site #1 than on site #2 (see Figure 3B). Independent experimental data from the literature suggest that both RPL9B poly(A) sites are intrinsically weak. Indeed, in very large‐scale Synthetic Genetic Array screens (Costanzo et al, 2010), Δrpl9Awas found among the mutations that exhibit the strongest synthetic genetic interaction with two thermosensitive mutants of Pcf11p, a key factor of the CPF/CF complex (Supplementary Figure S3). This observation can be readily explained by the fact that the RPL9B poly(A) sites are particularly sensitive to the pcf11 mutations, even in the permissive conditions used during the screens. In the absence of RPL9A, the cells can grow well with the RPL9B gene alone, given that it is not repressed. In the presence of the pcf11 mutations, however, RPL9B is not properly expressed, most likely as a result of a 3′‐end termination defect, and can no longer compensate for the absence of RPL9A. In contrast, RPL9A cleavage and polyadenylation does not seem to be sensitive to the pcf11 mutations in the same condition since the ability of RPL9A to compensate for the absence of RPL9B is not affected by these mutants, which even exhibit weak epistasis with the Δrpl9B deletion (Supplementary Figure S3).
The proposed mode of regulation of RPL9B is also reminiscent of the autoregulation of yet another RNA‐binding protein, Nab2p. In this system, the overexpression of Nab2p results in the destabilization of its own mRNA in a way dependent on an unusual encoded terminal A26 sequence and Rrp6p. Although the precise mechanism by which this regulation occurs remains to be elucidated, it has also been proposed that it might act at the level of mRNA 3′‐end formation at an intrinsically weak poly(A) site (Roth et al, 2009).
RPL9B represents the fifth example of an autoregulated ribosomal protein gene in S. cerevisiae. It is noteworthy that the four other known examples also involve post‐transcriptional mechanisms operating at the RNA level: RPS14B (Li et al, 1995) and RPL30 (formerly L32) are autoregulated at the level of pre‐mRNA splicing (Dabeva and Warner, 1993). The autoregulation of RPL30 also occurs at the translation level. RPS3 (Hendrick et al, 2001), RPL4A (formerly ribosomal protein L2; Presutti et al, 1995) and RPS28B (Badis et al, 2004) are autoregulated at the level of mRNA decay. In this context, the autoregulation mechanism of RPL9B, which involves the modulation in the balance between two modes of transcription termination, appears quite original.
Finally, it has previously been shown that the Nrd1/Nab3 mode of termination is most efficient at the beginning of transcription, when the C‐terminal heptameric repeats (CTD) of the RNA Pol II largest subunit are still phosphorylated at the Ser5 positions. It has experimentally been demonstrated that the progressive substitution of Ser5 phosphorylation by Ser2 phosphorylation of the CTD results in the Nrd1‐dependent pathway becoming very inefficient past the first 900 nucleotides of transcription (Gudipati et al, 2008; Vasiljeva et al, 2008). This suggests that the specific mode of regulation described here, relying on the modulation of the balance between the two modes of transcription termination, can only apply to relatively short transcripts, in a range where both the CPF/CF and the Nrd1 pathway can efficiently operate. Indeed, with a size of slightly more than 800 nucleotides, the RPL9B transcription unit might have the ideal size to be controlled by such a mechanism.
This raises the question of whether other genes are regulated by the same mechanism. Searching the data from Neil et al (2009), we could not identify obvious candidates with clearly separated CPF/CF‐ and Nrd1‐dependent termination sites. But RPL9B might have been in the ideal situation where it was expressed at an intermediate level of regulation, thus generating tags in both RNA populations.
Materials and methods
Construction and manipulation of yeast strains and plasmids
All the strains used in this study are derived from W303 or BY4741 and are listed in Supplementary Table S1. The strains with names beginning with DLY are gifts from D Libri. Stains with names starting with LMA are from this work or previous work from our laboratory. To obtain the plasmid borne CUP‐RPL9B‐3′UTR constructs, we first cloned the CUP1‐Tcyc1 gene under the control of the promoter TEF (translation elongation factor) generating plasmid PTEF‐CUP1‐TCYC1. RPL9B 3′‐UTR (832 nt) was subsequently introduced by homologous recombination using NotI linearized plasmid PTEF‐CUP1‐TCYC1 and PCR fragment generated by oligos RG005/RG006. All constructs were verified by sequencing.
Random mutations in the RPL9B 3′‐UTR were generated by random mutagenesis PCR, as described (Wilson and Keefe, 2001), using oligos RG05/RG006. The amplified fragments were cloned by homologous recombination as described above. The mutations were selected by picking the clones that were able to grow in the presence of copper and subsequently sequenced. Only the clones that have a single mutation are represented.
Yeast culture parameters
The thermo‐sensitive mutants were heat shocked, when required, by adding one volume of culture at 25°C into one volume of fresh medium at 35°C and then incubated at 35°C for 30 min. The strains containing plasmids were always grown in minimal medium.
To perform the metabolic depletion of Nab3p, strains carrying the PGAL‐NAB3 allele were grown in galactose‐containing rich medium and then shifted for 9 h into glucose‐containing rich medium.
RNA preparations and analyses
RNAs were analysed on 5 or 6% denaturing polyacrylamide. Hybridizations were performed using a commercial buffer (Ultrahyb, Ambion or Rapid‐hyb Buffer, GE healthcare). Strand‐specific probes were prepared by standard methods, either by phosphorylation of oligonucleotides using 32P γATP or by in vitro transcription of RNA by including 32P αUTP using T7 RNA polymerase. The templates for T7 RNA polymerase were prepared by standard PCR, by integrating T7 promoter into one of the PCR primers. RNAse H digestions were performed as described (Libri et al, 2002), with oligonucleotides as indicated in the figure legends. See Supplementary Table S2 for the sequence of the oligonucleotides used in this study.
Conflict of Interest
The authors declare that they have no conflict of interest.
Source data for Figures 2C and 4D
We thank the members of the GIM laboratory for helpful discussions, D Libri for the kind gift of strains (DLY strains in Supplementary Table S1) and Laurence Ma from the Institut Pasteur Genomics Platform (PF1) for running the cDNA deep‐sequencing samples. This work was supported by grant ANR ‘CutReg’ (ANR‐08‐BLAN‐0038) from the Agence Nationale de la Recherche, by the CNRS and the Institut Pasteur. RKG received a ‘Roux’ fellowship from Institut Pasteur. HN was supported by grant ARN ‘CUT’ (ANR‐ ANR‐05‐BLAN‐0062).
Author contribution: RKG, HN and FF constructed the plasmids and yeast strains. RKG and FF performed the RNA analyses and genetic screens, FF prepared the samples for cDNA deep sequencing and CM performed the bioinformatics analyses. RKG, HN, FF and AJ designed the experiments. RKG and AJ wrote the manuscript.
- Copyright © 2012 European Molecular Biology Organization