Main

The discovery of pervasive transcription and the revelation of the large number of non-coding RNAs (ncRNAs) raised the question of their functionality4,5. ncRNAs are divided into small (shorter than 200 nucleotides) and long ncRNAs. The lncRNAs are represented globally in life and are in their entirety perhaps the least-well-understood class of transcripts. Importantly, the deregulation of lncRNAs has been associated with human diseases such as cancer and neurodegenerative diseases6. They are more like mRNAs than they are different: they are transcribed by RNA polymerase II with similar chromatin states and undergo 5′-capping, splicing and 3′-polyadenylation7, but most lack an open reading frame and therefore also the coding potential of mRNAs8. Several reports indicate nuclear functions for lncRNAs in S. cerevisiae in regulating coding genes by suppressing transcriptional leakage9,10,11 or by inducing transcription in response to environmental changes12,13. However, many antisense lncRNAs travel into the cytoplasm for reasons unknown, as becomes evident after mutation of the cytoplasmic exonuclease Xrn1 (ref. 3). Xrn1-mediated lncRNA degradation follows RNA translation and subsequent recognition through the non-sense-mediated decay (NMD) system, which detects the lack of an open reading frame14. Although some asRNAs have been shown to affect translation in human cells15,16, no cytoplasmic function for asRNAs is currently known in yeast. Thus, the export and cytoplasmic degradation of bulk asRNAs would seem to be a waste of energy for cells. We have discovered that asRNAs act as regulatory RNAs for their sense counterparts through dsRNA formation, which leads to their preferential nuclear export and a subsequent boost in gene expression.

To obtain a global picture of nucleo-cytoplasmic RNA distribution in the eukaryotic model organism S. cerevisiae, we fractionated cells and determined the amount of RNA in the cytoplasmic fraction relative to the total RNA by RNA sequencing (RNA-seq) (Fig. 1a, Extended Data Fig. 1a–d and Supplementary Fig. 1a,b). As expected, ribosomal RNAs (rRNAs) increased in the cytoplasmic fraction, because they are part of the translating ribosomes, whereas small nucleolar RNAs (snoRNAs) were underrepresented, because they function in the nucleus. Interestingly, we found that 23.69% of cellular mRNAs are greatly enriched in the cytoplasm, with many more being nuclear (42.14%). By contrast, 66.33% of asRNAs are greatly enriched in the cytoplasm (Fig. 1b and Extended Data Fig. 1e,f). This finding reveals striking differences in the overall cellular distribution of mRNAs and asRNAs, and led to the surprising observation that bulk asRNAs shuttle out of the nucleus. Furthermore, mRNAs with an equally highly expressed asRNA show on average an equivalent enrichment in the cytoplasmic fraction, whereas mRNAs with no considerable asRNA transcript are more likely to be nuclear (Fig. 1c). This indicates a possible determining role for asRNA transcripts in the cellular localization of their mRNA, perhaps by dsRNA formation.

Fig. 1: dsRNAs are mainly localized in the cytoplasm.
figure 1

a, Spatial RNA detection after fractionation RNA-seq experiment. The log2-transformed fold change of the cytoplasmic fraction compared with total lysate indicates the nucleo-cytoplasmic distribution. From n = 3 biologically independent samples. b, asRNAs are enriched in the cytoplasm. The percentage of unevenly distributed mRNAs and asRNAs from the fractionation RNA-seq is shown. NS, not significant. c, mRNAs localize to the cytoplasm when the asRNA is present in equal or higher amounts. Average nucleo-cytoplasmic distribution of mRNAs, based on their relative asRNA expression. d, J2 RIP experiments enriched asRNAs, as determined by the log2-transformed fold change of the J2 eluate relative to the unbound fraction. From n = 3 biologically independent samples. e, Percentage of significantly changed transcripts from the J2 RIP-seq. f, J2-enriched dsRNAs are cytoplasmic, as determined by their distribution in fractionation RNA-seq. WT, wild type. g, RNAs grouped by their enrichment in RNAi-seq46 analysed for their mean change in fractionation RNA-seq. h, PHO85 mRNA forms dsRNA with PHO85 asRNA (asPHO85) expressed from a plasmid. GFP pull-down with GFP-tagged MS2 loop-binding protein precipitates MS2-tagged PHO85 mRNA and co-precipitates PHO85 asRNA, determined by qPCR. From n = 4 biologically independent samples. i, Overexpression of PHO85 asRNA shifts PHO85 mRNA into the cytoplasm relative to no PHO85 asRNA (dotted line). The qPCR result after the fractionation experiment and RNA isolation of the total and unspliced PHO85 transcripts is shown. From n = 4 biologically independent samples. j, Hybridization with 12 Cy3-labelled probes detects a single GFP-tagged PHO85 mRNA and 15 Alexa647-labelled probes detects the MYC(15×)sequence of PHO85 asRNA. k, smFISH analysis shows dsRNA formation and a cytoplasmic shift of the PHO85 mRNA signal with PHO85 asRNA expression. The pho85∆ cells have a galactose-inducible PHO85–GFP and either an empty plasmid or the galactose-inducible asPHO85–MYC(15×) plasmid after 8 min galactose induction. The white arrowheads indicate colocalizing signal. From n = 4 biologically independent experiments with similar results. l, smFISH quantification of nuclear and cytoplasmic PHO85 signal at different time points after galactose induction with or without simultaneous asRNA induction; n > 52 cells over 3 biologically independent experiments.

Source Data

To identify dsRNAs in yeast we used the dsRNA-specific antibody J2 (ref. 17) in RNA co-immunoprecipitation (RIP) and subsequent RNA sequencing (RIP-seq) experiments (Fig. 1d and Extended Data Fig. 2a–c). This antibody recognizes dsRNAs 40 base pairs (bp) long and has already been used to determine the dsRNA transcriptome of Escherichia coli18. Remarkably, in yeast, J2 RIPs enriched more than 60% of all asRNAs in the eluate, but only around 13% of the mRNAs (Fig. 1d,e and Extended Data Fig. 2d,e), indicating that most of the asRNAs are part of a double strand, whereas this is only the case for a minority of the sense transcripts. Noticeably, snoRNAs and rRNAs were depleted in the J2 eluate (Fig. 1f), confirming the specificity of the antibody to dsRNAs formed by mRNA and asRNA pairs in yeast. The predominant enrichment of asRNAs is in line with the observation that asRNAs are on average expressed at 10-fold lower levels than the coding transcripts19 and simply have a higher probability of being present in a dsRNA. Moreover, single-cell analyses have shown that in only 10% of the cells is a sense RNA present together with its corresponding asRNA, but around half of the asRNAs are present with their sense transcript (Extended Data Fig. 3a). Most importantly, more than 80% of the RNAs that were enriched in the J2 eluate were cytoplasmic (Fig. 1f), indicating the increased cytoplasmic presence of dsRNAs. By contrast, most of the ssRNAs that did not bind to the J2 antibody were nuclear.

In yeast, dsRNA formation has previously been detected by a screen based on RNA interference (RNAi) in which the dsRNA-degrading RNAi system was artificially established, because S. cerevisiae had lost this system during evolution20,21. It allowed dsRNAs to be identified after Dicer expression by detection of the accumulating degradation products. The two datasets, the RNAi-seq and our J2 RIP-seq data, show a high correlation (Spearman’s rank correlation, r = 0.72; Extended Data Fig. 3b). We then grouped RNAs, on the basis of their enrichment in RNAi-seq, into ten clusters (Extended Data Fig. 3c) and applied fractionation RNA-seq. It became apparent that the more an RNA is prone to form a double strand, the more it is likely to localize to the cytoplasm in wild-type cells (r = 0.52; Fig. 1g and Extended Data Fig. 3d). Gene-coverage analysis of the RNAi-seq data furthermore revealed that most of these dsRNAs are formed with Xrn1-sensitive unstable transcripts (XUTs), which show greatest cytoplasmic enrichment and longest overlap with their mRNA in lncRNAs, fewer with stable unannotated transcripts (SUTs) and even fewer with cryptic unstable transcripts (CUTs)22 (Extended Data Fig. 3e–g). Thus, asRNAs, along with mRNAs that have a high probability of being present in a dsRNA with their counterpart, show on average cytoplasmic enrichment.

To investigate whether an asRNA could indeed increase the cytoplasmic localization of its sense transcript, we randomly chose to study the cellular localization of the cyclin-dependent kinase PHO85 mRNA. Based on its fractionation RNA-seq data, the PHO85 transcript is expressed with average amounts and is mostly nuclear, whereas its asRNA (SUT412) is only barely detectable. First, we validated the double-strand formation of both transcripts in RIP experiments when the asRNA is expressed ectopically from a plasmid. For this, we used a PHO85 mRNA tagged with 12 MS2 loops and placed the PHO85 asRNA downstream of a galactose-inducible promoter (PGAL1) to control its transcription and raise the expression level. The GFP-tagged MS2-binding protein MCP was co-expressed, subsequently precipitated and the co-purified RNA was analysed. PHO85 mRNA and its asRNA were present in the eluate, confirming that we had precipitated the RNAs as a double strand (Fig. 1h). Next, we carried out cytoplasmic fractionation experiments while overexpressing PHO85 asRNA, and these revealed that the PHO85 mRNA transcript notably shifted its distribution into the cytoplasm through elevated asRNA expression and dsRNA formation (Fig. 1i, Extended Data Fig. 4a,b and Supplementary Fig. 1c,d) For further validation, we placed both transcripts downstream of the inducible GAL1 promoter, which enabled us to trace their export in single-molecule fluorescence in situ hybridization (smFISH) experiments. Because it is difficult for probes to bind in the dsRNA region, both transcripts were tagged and 12 (for the GFP tag) or 15 (for the MYC tag) probes with different fluorophores were designed to be complementary to the respective single-stranded tag (Fig. 1j,k). After induction, cells were collected and fixed at different time points for smFISH. The simultaneous induction with and without asRNA resulted in similar signal intensities for the PHO85 mRNA over the course of the experiment (Fig. 1l). However, in combination with asRNA expression, the cytoplasmic pool of the sense transcript increased substantially compared with no asRNA expression (Extended Data Fig. 4c). The observed colocalization of both transcripts indicates that the effect is due to dsRNA formation. Indeed, a similar observation had already been made in human skin cutaneous melanoma cells, in which an increased transcription of the antisense RNA TTN-AS1 resulted in increased cytoplasmic localization of its mRNA TTN. Remarkably, knockdown of the antisense form reduced tumour growth and metastasis23. Notably, allowing dsRNA formation to occur inhibited neither splicing nor nuclear quality control of the PHO85 mRNA transcript, because no increased unspliced mRNA was detected in the total lysate or in the cytoplasmic fraction (Fig. 1i). Thus, asRNA can shift the nucleo-cytoplasmic distribution of its mRNA towards the cytoplasm, which is a previously unknown phenomenon that reveals a new layer of gene expression. To investigate whether dsRNA formation might alter the stability of the transcripts, we analysed previously published experimental data24 from a study that explored RNA stability through a precise pseudo-uridine labelling method and combined that study’s data with those of the RNAi-seq-based dsRNA probability and J2 RIP-seq. We observed no correlation between RNA stability and its occurrence in a double strand (r = −0.04, r = −0.13; Extended Data Fig. 5a,c). Interestingly, mRNAs with a long half-life (r = −0.42) and a high expression level (r = −0.43) tend to be more nuclear (Extended Data Fig. 5b,c). Importantly, because the relative cytoplasmic distribution is not caused by higher stability of the dsRNA, a faster export rate of dsRNAs might rather be a possible explanation for their increased cytoplasmic presence.

It was already known that mRNAs use multiple molecules of the heterodimer Mex67–Mtr2 (human TAP–p15) for export, and these are recruited during quality control by the guard proteins. The karyopherin Crm1/Xpo1 supports export through interaction with the cap-binding complex attached to the 5′ end of the transcript1,2,25,26,27,28. It is evident from high-throughput analyses that Mex67 is also implicated in ncRNA export29. Furthermore, ncRNAs such as the lncRNA TLC1 and snRNAs have been shown to be exported via the Mex67, Xpo1 pathway28,30. To determine whether dsRNAs also depend on this pathway, we repeated the fractionation RNA-seq in the mex67-5 xpo1-1 double mutant that was shifted to its restrictive temperature at 37 °C for 1 h (Extended Data Fig. 6a,b). Indeed, all types of RNA were trapped in the nucleus, regardless of their probability of forming dsRNA, demonstrating their dependence on both export factors (Fig. 2a and Extended Data Fig. 7a). Furthermore, immunofluorescence and dot-blot experiments using the J2 antibody confirmed the cytoplasmic localization of dsRNAs in wild-type cells and the nuclear retention in mex67-5 xpo1-1 (Fig. 2b,c and Extended Data Fig. 7b, c). Importantly, inhibiting translation in the ribosomal subunit export mutant nmd3-2, the ribosome subunit joining mutant rpl10G161D and the inhibitor cycloheximide increased the cytoplasmic dsRNA content, confirming the reliance on translation for resolution (Fig. 2b–d, Extended Data Fig. 7d and Supplementary Fig. 2a). These findings indicate the dependence of dsRNA export on the Mex67 mRNA export pathway and the need for translation to resolve dsRNAs.

Fig. 2: Mex67 preferentially binds dsRNAs for faster nuclear export.
figure 2

a, dsRNAs accumulate in the nucleus of mex67-5 xpo1-1, as determined by fractionation RNA-seq experiment. The average reduction of the cytoplasmic RNA relative to the wild type is based on groups defined by RNAi-seq. From n = 3 biologically independent samples. b, Exported dsRNAs reach the ribosome. J2 antibody IF with a Cy3-labelled secondary antibody (dsRNA) and FISH with a Cy3-labelled oligonucleotide d(T) probe were done for 1 h at 37 °C. From n = 3 biologically independent experiments. Scale bars, 3 μm. c, Fluorescence intensity quantification in J2 IF for 1  h at 37  °C for each condition shown in b; n > 20 cells over 3 biologically independent experiments. From left to right: P = 1.27 × 10−8, P = 3.03 × 10−25, P = 1.98 × 10−23 and P = 2.39 × 10−18. d, dsRNA levels increase when translation is inhibited. Dot-blots with RNA from the indicated strains and treatments after 1  h at 37  °C were detected with the J2 antibody. From n = 3 biologically independent experiments with similar results. e, dsRNAs contact ribosomes before ssRNAs. We shifted mex67-5 to 37 °C for 1 h to block RNA export. It was subsequently released by lowering the temperature to 25 °C. RIP experiments with Rps2–GFP from different time points were done. Three ssRNA and three dsRNA targets were analysed by qPCR. From n = 3 biologically independent experiments. f, Induction of PHO85 asRNA expression changes both protein and mRNA levels of PHO85, as shown by western blot quantification and qPCR. Hem15 served as loading control and for normalization. From n = 3 biologically independent experiments. g,h, More Mex67 molecules can bind to dsRNA than to ssRNA. EMSA with FAM-labelled ssRNAs (g) or dsRNAs (h) was carried out by adding increasing amounts of recombinant TAP-tagged Mex67–Mtr2. From n = 3 independent experiments with similar results. i,j, Competition assay detects preferential Mex67–Mtr2 binding to dsRNA. Cy5-labelled ssRNAs (red) were pre-incubated with Mex67–Mtr2. Subsequently, increasing amounts of a FAM-labelled dsRNA (green) were added as a competitor (i). Increasing amounts of Cy5-labelled ssRNA were added to pre-bound FAM-labelled dsRNA (j). From n = 3 independent experiments with similar results.

Source Data

To determine whether dsRNAs are exported in preference to ssRNAs, we did an experiment. First, we blocked RNA export in the mex67-5 mutant by shifting it to 37 °C, causing mislocalization of Mex67 from the nuclear rim to the cytoplasm1 and in turn nuclear accumulation of ssRNAs and dsRNAs (Fig. 2e). When we lowered the temperature to 25 °C, functional Mex67 returned to the nuclear rim and transport was restored. To see which transcripts reach the ribosome, and thus the cytoplasm, first, we precipitated Rps2 and purified the co-precipitated RNA at specific time points after the export block release (Extended Data Fig. 7e and Supplementary Fig. 2b,c). We found that dsRNAs arrived at ribosomes notably earlier than ssRNAs, providing experimental evidence for the preferential export, and thus translation, of dsRNAs (Fig. 2e). Next, we analysed how faster export induced by asRNA expression would influence the protein and RNA level of the sense gene in quantitative PCR (qPCR) and western blots (Fig. 2f and Supplementary Fig. 2d,e). After PHO85 asRNA induction, the protein level of Pho85–GFP increased (3.1-fold at 60 min), whereas the level of the PHO85–GFP mRNA decreased slightly (0.32-fold at 60 min), because it might no longer be held in the nucleus for potential export. This is an indication of the gene-expression boost of the sense RNA by its asRNA.

The faster export may be due to a different occupancy of the RNAs by export receptors. As suggested previously, the more export proteins cover an RNA, the faster transport is initiated26,31. To analyse Mex67 binding to ssRNA and dsRNA, we carried out electrophoretic mobility shift assays (EMSAs). First, we purified recombinant Mex67–Mtr2, which was shown to bind directly to mRNAs32. Mex67–Mtr2 was added in varying molar excess to fluorescein amidite (FAM)-labelled 36-nucleotide-long ssRNA or dsRNA (Extended Data Fig. 7f and Supplementary Fig. 4a–c) and incubated for 15 min before the samples were analysed on gel electrophoresis. A 2-molar excess of Mex67–Mtr2 fully upshifted ssRNA, which was saturated at 5-molar excess (Fig. 2g and Supplementary Fig. 3a). By contrast, dsRNA reached binding saturation at a 10-molar excess, with a full upshift at a 3-molar excess (Fig. 2h and Supplementary Fig. 3b). This indicates that dsRNAs can bind about two heterodimers every 7 bp, whereas ssRNAs accommodate only one in the same length. Subsequently, we carried out competition assays in which we added either FAM-labelled dsRNA or Cy5-labelled ssRNA to a Mex67–Mtr2 pre-formed complex on similarly labelled dsRNA or ssRNA, respectively. Mex67–Mtr2 dissociated from ssRNA and associated with dsRNA in the presence of increasing dsRNA amounts, but it showed almost no displacement properties when ssRNA was titrated against the dsRNA complex (Fig. 2i,j and Supplementary Fig. 3c,d). Switching the labels made no difference and resulted in the same effect (Extended Data Fig. 7g,h and Supplementary Fig. 4d,e). These findings show that Mex67–Mtr2 binds preferentially and more extensively to dsRNA, explaining its preference in export.

Our findings indicate that asRNAs can boost the expression of individual transcripts. The broad existence of annotated asRNAs in the yeast genome indicates a general usage of asRNA boosts to navigate gene expression, which would explain their prevalence. It is likely that not all boosting asRNAs have been discovered yet, because they may be transcribed only under specific conditions to steer the cells in a particular direction, such as during development or stress. For instance, a previous study found a new set of asRNAs that are suppressed by the chromatin modifier Set2 (ref. 33). These down-regulated transcripts, termed Set2-repressed antisense transcripts (SRATs), are antisense to stress response and ageing-related transcripts. To determine whether these pairs could form dsRNAs, we carried out J2 immunofluorescence (IF), J2-IP and dot-blot experiments in set2∆ and found a significant increase in the total amount of dsRNA in the cytoplasm (Extended Data Fig. 8a–c,e and Supplementary Fig. 5a), indicating that the induced SRATs form double strands with their sense counterparts and are exported as dsRNA. Cell fractionation experiments and qPCRs with a Set2-responsive transcript, SEG2 asRNA, confirmed this observation and showed an up-regulation of the asRNA when SET2 was deleted (Extended Data Fig. 8d), resulting in dsRNA formation (Extended Data Fig. 8e) and a subsequent increased presence of SEG2 mRNAs in the cytoplasm (Extended Data Fig. 8f,g and Supplementary Fig. 5b).

Changes to the cellular expression programs are sometimes necessary, for instance for development, during ageing and in response to stress. Each situation might generate new asRNA transcripts that boost the expression of individual mRNAs. Analysis of previously published RNA-seq data that were generated after osmotic shock34 revealed not only that stress-responsive mRNAs increased during periods of stress, but so did asRNAs13 (Fig. 3a). Indeed, 95% of significantly increased mRNAs with significantly changed asRNA show an up-regulation of the asRNA (Fig. 3b). To visualize the transcriptome-wide changes to the expression program, we determined the dsRNA formation in cells shocked with salt (0.7 M NaCl) or ethanol (10% EtOH) through J2-IF experiments (Fig. 3c,d). Although bulk mRNA accumulates in the nucleus and dissociates from Mex67 under stress conditions32, the newly produced dsRNA reached the cytoplasm in the first 5 min. These dsRNAs represent the preferentially exported stress-induced mRNAs that hybridized with their asRNA.

Fig. 3: dsRNA formation is essential for cells changing their expression program.
figure 3

a, Stress increases asRNA levels. A genome-wide RNA analysis of cells incubated with 0.6 M NaCl (ref. 34) for 30 min was used to find changes in sense mRNA and asRNA expression compared with unstressed conditions, and is shown here by the log2-transformed fold change. From n = 2 biological independent samples. b, Stress-responsive mRNAs are accompanied by increased asRNA expression. Significantly changed asRNAs of significantly increased mRNAs from the RNA-seq34 are shown after 30 min exposure to 0.6 M NaCl. c, The amount of dsRNA increases under stress conditions. J2-IF is shown for the wild type exposed to the indicated stress conditions. From n = 3 biologically independent experiments. d, Quantification of the J2-IF displayed in e. From n = 30 cells over 3 independent experiments. Left to right: P = 6.91 × 10−9, P = 1.59 × 10−15, P = 7.57 × 10−4, P = 5.47 × 10−12. e, dsRNA degradation by the bacterial ribonuclease RNaseIII in the nucleus of yeast cells is lethal. We spotted 10-fold serial dilutions of the wild type containing the indicated plasmids onto glucose (no induction) or galactose (with induction) plates and incubated for 3 days. From n = 3 biologically independent experiments with similar results. f, J2-IF and localization of the GFP- and transport signal-tagged RNaseIII fusion proteins in yeast cells. Plasmid-containing wild-type cells were grown to the logarithmic phase before the RNaseIII expression was induced by adding galactose. From n = 3 biologically independent experiments with similar results. g, Cytoplasmic RNaseIII is not tolerated in cellular stress situations. We spotted 10-fold serial dilutions of wild-type cells containing either a constitutively expressed RNaseIII–NES from the ADH1 promoter or the RNAi system onto plates and incubated for 3 days at 25 °C. From n = 3 biologically independent experiments with similar results. h, Stress-induced dsRNA is degraded by cytoplasmic RNaseIII. J2-IF and oligonucleotide d(T) FISH are shown either without stress or after 30 min incubation with 0.7 M NaCl. From n = 3 biologically independent experiments. All scale bars, 3  μm.

If this mechanism of regulated, boosted gene expression resulting from dsRNA formation were of a general nature, the degradation of dsRNA should be hazardous to cells. Previous work has shown that double strands of mRNA and asRNA are degraded by RNaseIII from E. coli18. In S. cerevisiae, such studies found unexplained toxicity of cells expressing RNaseIII in growth analyses that could not be attributed to the degradation of rRNA or general mRNA35, but it might be caused by unrecognized dsRNA degradation. To confirm the specificity of RNaseIII for dsRNA, we exposed isolated total RNA to recombinant RNaseIII and subsequently detected the remaining dsRNA by J2 dot-blot. After RNaseIII digestion, the dsRNA was noticeably reduced (Extended Data Fig. 8h and Supplementary Fig. 5c). To analyse its effects on living cells, we expressed RNaseIII in vivo from the inducible GAL1 promoter and directed the protein to different compartments. Tagging the enzyme with a nuclear localization signal resulted in cell death (Fig. 3e). Expressing RNaseIII with both a nuclear localization signal and a nuclear export signal (NES) was equally toxic. This fusion protein can shuttle between the nucleus and the cytoplasm and was visible mostly at the nuclear rim (Fig. 3f). Interestingly, a construct that restricted the dsRNA-degrading enzyme to the cytoplasm (RNaseIII–NES) was tolerated in rich medium. This may be because the cytoplasmic dsRNA is somehow protected or because the subsequent translation is quite fast. The possibility that the cytoplasmic construct is not functional is unlikely because reduced dsRNA levels were detectable in the translation-defective strain rpl10G161D (Extended Data Fig. 8n and Supplementary Fig. 5d), in which the cytoplasmic dsRNA accumulates (Fig. 2b). Most importantly, although the cytoplasmic RNaseIII–NES was tolerated in rich medium, its expression during osmotic stress resulted in severe growth defects and dsRNA reduction (Fig. 3g,h). This was also the case for the cytoplasmic-operating RNAi system. Together, these data indicate that the asRNA boost of mRNA expression through dsRNA formation is particularly important in changing and challenging situations.

An enzyme that mediates dsRNA formation is most likely to be an RNA helicase, because these enzymes are known not only for their RNA unwinding but also for their dsRNA-binding, dsRNA-annealing and dsRNA-clamp activities36,37. Interestingly, deletions of the two nuclear helicases Dbp2 or Mtr4 have been reported to increase XUT-asRNA levels38. Mtr4 is an enzyme known to be involved in RNA degradation39, whereas for Dbp2, strand annealing under the mediation of Yra1 has already been demonstrated in vitro and has been suggested to be important for messenger ribonucleoprotein (mRNP) assembly40,41. Furthermore, Yra1 was shown to recruit Mex67 for mRNA export42. To investigate whether Dbp2 and/or Mtr4 might be involved in dsRNA biogenesis, we used a cold-sensitive DBP2-knockout strain41 (Extended Data Fig. 9a) and the temperature-sensitive mtr4G677D mutant43 in J2 IF studies and J2 dot-blot experiments. Interestingly, the mtr4 mutant accumulated dsRNA in the nucleus, indicating that this helicase is not required for dsRNA formation but instead for its decay after quality control (Fig. 4a,c,d, Extended Data Fig. 9b and Supplementary Fig. 6a). Most importantly, however, the absence of Dbp2 resulted in a clear decrease in dsRNA formation and a simultaneous nuclear accumulation of poly(A)+ RNA (Fig. 4b–d and Extended Data Fig. 9c), identifying Dbp2 as the dsRNA-forming helicase in vivo and showing its mRNA export-supporting function. Indeed, it has been shown that the sense and corresponding antisense transcripts changed by the deletion of DBP2 strongly correlate44, which supports our findings. Thus, we reconsidered previously published high-throughput analyses of dbp2, in which RNA secondary structures were determined through the labelling of free nucleotides with dimethyl sulfate in the wild type and dbp2∆ (ref. 45). We applied this dataset to the RNAi-seq data21 and found that the more an RNA is prone to be in a double strand, the more nucleotides became accessible and were labelled in the dbp2∆ compared with the wild type (Extended Data Fig. 9d). Interestingly, individual-nucleotide resolution cross-linking and immunoprecipitation sequencing (iCLIP-seq) data45 showed similar binding of Dbp2 to all transcripts when analysed in the context of RNAi-seq21 (Extended Data Fig. 9e), indicating that the helicase contacts all transcripts initially, possibly for unwinding as previously suggested45. The potential for subsequent dsRNA formation then depends on the availability of the respective asRNA and on the contact with Yra1.

Fig. 4: Dbp2 induces dsRNA formation.
figure 4

a, We found that dsRNAs accumulate in the nucleus of mtr4G677D at 37 °C. J2-IF and oligonucleotide d(T)-FISH are shown. b, Disturbed dsRNA formation in dbp2∆. Strains were changed to the non-permissive temperature for dbp2∆ of 25 °C. c, Signal quantification from a and b. From n > 40 cells over 3 biologically independent experiments. Left to right, P = 7.83 × 10−35, P = 8.56 × 10−34. d, J2 dot-blot of isolated RNA. From n = 3 biologically independent experiments with similar results. e, Dbp2 binds to dsRNA. Western blot of J2 Co-IP from cells expressing MYC-tagged DBP2 with or without the addition of recombinant RNaseIII. Grx4 is negative control. From n = 3 biologically independent replicates with similar results. f, dsRNA formation and cytoplasmic shift of PHO85 mRNA after PHO85 asRNA expression was lost in dbp2∆. Wild type and dbp2∆ carrying the galactose-inducible PHO85–GFP plasmid and either an empty vector or the galactose-inducible asPHO85–MYC(15×) plasmid were used for smFISH after 8 min galactose induction. From n = 3 biologically independent experiments. g, Quantification of smFISH for the cytoplasmic/total signal ratio of PHO85 mRNA either with or without (dotted line) simultaneous asRNA expression. From n > 23 cells examined over 3 biologically independent experiments. h, The increased presence of PHO85 mRNA in the cytoplasm after PHO85 asRNA overexpression was abolished in dbp2∆. After the induction of PHO85 asRNA, cells were shifted to 25 °C for 1 h before cytoplasmic fractionation, RNA isolation and qPCR. From n = 3 biologically independent samples. i, Model for the preferential export of dsRNAs. After transcription, ssRNAs are eventually bound by Mex67, leading to low-level export and translation in the cytoplasm. Gene expression is boosted by the transcription of asRNA and subsequent dsRNA formation of sense–antisense pairs by the helicase Dbp2 and its co-factor Yra1. The dsRNA preferentially binds to Mex67 for nuclear export and mRNA translation. Ribosomes recognize the non-coding property of the asRNAs and subsequent NMD-mediated degradation. This mechanism ensures preferential gene expression. All scale bars, 3 μm.

Source Data

To find further experimental evidence, we first confirmed that Dbp2 binds to dsRNA in vivo through J2-IP (Fig. 4e and Supplementary Fig. 6b). This binding was lost in the presence of recombinant RNaseIII. To validate the in vivo relevance for the dsRNA-formation function of Dbp2, we repeated the experiment from Fig. 1i–k in the absence of Dbp2. We found that although the asRNA was still highly enriched in dbp2∆ after galactose induction, it was not able to manipulate the localization of its sense PHO85 mRNA (Fig. 4f–h, Extended Data Fig. 9f,g and Supplementary Fig. 6c,d). This finding indicates that Dbp2 is the key factor for dsRNA formation that enables the preferential export of dsRNAs.

In conclusion, our findings reveal a new layer of regulated gene expression. Boosting asRNAs anneal with their sense counterpart through Dbp2-mediated dsRNA formation. These dsRNAs are preferentially exported and subsequently the respective sense transcripts are preferentially expressed (Fig. 4i). This mechanism is particularly important for effective cellular adaptation and adds preferential export as a new layer of regulated gene expression. Furthermore, it could also explain how pervasive transcription controls gene expression, and why so many asRNAs are generated and travel into the cytoplasm.

Methods

Yeast strains, plasmids and oligonucleotides

All the yeast strains used in this study are listed in Supplementary Table 1 and the plasmids are listed in Supplementary Table 2. Strains were cultivated and grown in standard medium at 25 °C. The diploid strain HKY2065 was sporulated and subjected to tetrad dissection followed by analysis of haploid spores for their genetic markers.

FISH

The experiments were essentially carried out as previously described47. To detect poly(A)+ RNA, a Cy3-labelled oligonucleotide d(T)18 probe (Sigma) was used. Cells were grown to mid-logarithmic phase (around 1 × 107 cells per ml) before they were treated as indicated. Cells were fixed by adding formaldehyde to a final concentration of 3.7% for 40 min at room temperature. After washing, permeabilization and pre-hybridization, the Cy3-labelled d(T)18 probes were added and hybridized overnight at 37 °C. DNA was stained with DAPI (Sigma) for 2 min. Microscopy studies were carried out using a Leica AF6000 microscope and an HCX PL APO CS ×63 objective lens. Pictures were obtained using a LEICA DFC360FX camera with a resolution of 1,392 × 1,040 px and LAS AF 2.7.3.9 software (Leica). Images were quantified using Fiji software.

smFISH

The experiment was conducted largely as described above. Cells were grown in 2% raffinose to the logarithmic-growth phase. The expression of PHO85 mRNA and its asRNA were induced by adding 2% galactose. Cells were collected after the indicated times and fixed for 20 min in 3.7% formaldehyde. The probes used are listed in Supplementary Table 5. They were incubated for 3 h at 37 °C. Thereafter, the washing steps with SSC were carried out for 15 min each, as described in the FISH protocol. Quantification of the signal was carried out using Fiji software. To determine the nuclear signal, the DAPI signal was used as a reference. The boundary of the total cell was determined using Nomarski optic. The cytoplasmic signal was calculated by subtracting the nuclear signal from the total signal. The background signal was measured three times per image and subtracted from the measured signal of the cell as follows: integrated density − (selected area) × (mean fluorescence of background readings), which resulted in the final signal strength that was used for all images. For every time point, cells from three independent biological repetitions were quantified.

Immunofluorescence

Cells were grown, collected and treated as described for the FISH experiment. After permeabilization, cells were blocked in ABB (0.1 M Tris, pH 9.0, 0.2 M NaCl, 5% FCS, 0.3% Tween, 500 µg ml−1 transfer RNA) and incubated for 1 h at 37 °C, followed by ABB with the addition of 1/200 µl of the J2 antibody (1 µg µl−1) from Scicons17 and 0.2% Triton for 2 h at 37 °C. The addition of Triton prevented binding to membrane-bound glycan RNA, which was already known to be an antigen of the J2 antibody. Subsequently, cells were washed with 0.5% Triton in 1× PBS for 15 min, twice with 1× PBS for 15 min and finally with ABB for 30 min. The secondary Cy3-conjugated anti-mouse antibody in ABB (1:200) was thereafter incubated for 1 h at room temperature. Subsequently, cells were washed with 0.5% Tween in 1× PBS for 10 min and twice in 1× PBS for 10 min. Nuclei were stained with DAPI (Sigma) and mounting, microscopy and quantification were carried out as described in the FISH experiment.

GFP microscopy

The visualization of GFP-tagged proteins in vivo was essentially done as previously described47. Cells were grown in glucose (2%)-containing medium until the early logarithmic phase (0.5 × 107 cells per ml), washed once with 1 ml sterile H2O, transferred into galactose (2%)-containing medium and grown for 6 h to induce the expression of RNaseIII constructs. Next, cells were fixed with 3.7% formaldehyde for 1 min at room temperature and washed twice with 1 ml P-Solution (0.1 M potassium phosphate buffer, pH 6.5, 1.2 M sorbitol) before adding 20 µl on a polylysine-coated slide for 15 min at room temperature. Permeabilization, DNA staining, microscopy and quantification were carried out as described in the FISH experiment.

Cytoplasmic fractionation

To detect RNAs in the cytoplasm, cells were grown to mid-logarithmic phase (2 × 107 cells per ml), washed once with 1 ml YPD/1 M sorbitol/2 mM DTT and resuspended in YPD/1 M sorbitol/1 mM DTT with the addition of zymolyase (100 mg ml−1) to spheroplast cells. Before cytoplasmic fractionation, 200 µl of the cell suspension was taken for total lysate control. For the analysis shown in Fig. 1a and similar work, after spheroblasting, cells were diluted in 50 ml YPD/1 M sorbitol for 30 min at 25 °C before they were shifted to 37 °C for 1 h. After shifting, 10 ml was taken for total cell lysis. Next, cells were cooled on ice and centrifuged for 5 min at 2,000 rpm. For cytoplasmic fractionation, the cell pellets were resuspended in 500 µl Ficoll buffer (18% Ficoll 400, 10 mM HEPES, pH 6.0) and cells were lysed by adding 1 ml buffer A (50 mM NaCl, 1 mM MgCl2, 10 mM HEPES, pH 6.0) and 1 µl Ribolock RNase Inhibitor (Thermo Fisher). The suspension was vortexed and centrifuged for 10 min at 2,000 rpm. The resulting supernatant reflects the cytoplasmic fraction. To verify correct fractionation, samples were analysed in western blots for the presence of the cytoplasmic Zwf1 (anti-Zwf1 in TBS-T (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.1% Tween 20), 1:4,000) and the nuclear proteins Yra1 (anti-Yra1 in TBS-T, 1:1,000) and Nop1 (anti-Nop1 in TBS-T, 1:4,000). RNA was isolated using a Nucleo-Spin RNA Kit (Macherey and Nagel).

J2 RNA co-immunoprecipitation experiment

Yeast strains were grown to mid-logarithmic phase (2 × 107 cells per m) followed by ultraviolet cross-linking with a wavelength of 254 nm for 7 min. Cells were collected and lysed in RIP buffer (25 mM Tris-HCl, pH 7.5, 150 mM NaCl, 2 mM MgCl2, 0.5% (v/v) Triton X-100, 0.2 mM PMSF, 0.5 mM DTT, 10 U RiboLock RNase Inhibitor (Thermo Fisher) and protease inhibitor (Roche)) by using a FastPrep-24 machine (MP Biomedicals) with shaking three times for 30 s at 5.5 m s−1. After centrifugation, 30 µl of the supernatant was taken for input control and the remaining lysate was incubated with or without 3 µl of the J2 antibody (1 µg µl−1)17 from Scicons and the addition of recombinant ShortCut RNaseIII (NEB) for 30 min at 4 °C. After the first incubation, the lysates were transferred to prewashed G-sepharose beads and incubated for another 90 min at 4 °C. The beads were then washed five times with RIP buffer (0.25% Triton). The supernatant was removed and SDS loading dye (125 mM Tris-HCl, pH 6.8, 2% SDS, 10% glycerol, 5% 2-mercaptoethanol, bromphenolblue) was added. Subsequently, samples were incubated at 95 °C for 5 min and loaded onto an SDS gel followed by western blotting and staining with MYC (anti-MYC in TBS-T, 1:,) and Grx4 (anti-Grx4 in TBS-T, 1:4,000).

J2 RIP for RNA-seq

All yeast strains were grown to mid-logarithmic phase (2 × 107 cells per ml). Total RNA was isolated with TRIzol reagent. After the first ethanol precipitation, a DNaseI treatment was conducted followed by a second precipitation overnight. The obtained RNA was eluted in RNase-free water. Then 90 µg RNA and 3 µg J2 antibody were incubated in 500 µl PBST for 120 min at 4 °C (1× PBS, 0.5% Tween-20). After the first incubation, the RNA–antibody mix was transferred to prewashed G-sepharose beads and incubated for another 120 min at 4 °C. The beads were centrifuged for 1 min at 4,000 rpm at 4 °C. The supernatant was transferred a second time, together with 3 µg J2 antibody, to freshly washed beads and incubated for 120 min. Subsequently, these beads were centrifuged and the supernatant was used as the unbound fraction. The beads from the first incubation were washed five times with 1 ml PBST. Between each step, the beads were centrifuged for 1 min at 4,000 rpm at 4 °C. Finally, the RNA was purified from the unbound fraction and from the eluates by TRIzol-chloroform (Ambion RNA by Life Technologies) extraction and forwarded to RNA sequencing. We repeated the experiment three times, and it showed a high reproducibility.

J2 dot-blot

Cells were grown to the logarithmic-phase and shifted, if necessary, as indicated. RNA isolation was carried out with TRIzol reagent. Then 1 µg of the isolated RNA was applied onto a nylon membrane, which was blocked in PBST (1× PBS, 1% Tween-20), 0.05 mg ml−1 ssDNA and 5% (w/v) non-fat dried milk. Subsequently, the J2 antibody (anti-dsRNA in PBST, 1:5,000) was added and incubated for 2 h at room temperature. Finally, there were two washing steps with PBST, each for 15 min at room temperature, before the HRP-coupled goat anti-mouse secondary antibody was added in PBST for 1 h. Finally, the membrane was washed again three times with PBST for 10 min at room temperature, before the ECL detection was carried out with a Fusion FX7 Edge 18.06c (Vilber). Quantification was finalized with the analysis software Bio-1D from Vilber Lourmat.

Protein isolation and purification

Transformed Rosetta 2 E. coli cells were grown in 200 ml LB medium with ampicillin (100 µg ml−1) and chloramphenicol (25 µg ml−1) overnight, diluted to OD600 = 0.1 in 1,200 ml Terrific Broth medium (28.8 g yeast extract, 24 g Trypton, 9 ml 50% glycerin, 17 mM KH2PO4, 72 mM K2HPO4) and 100 µg ml−1 ampicillin. The diluted cells were incubated at 32 °C and 130 rpm for 3 h, followed by 37 °C and 130 rpm for 1 h. For protein induction, 1.2 ml of 1 M IPTG was added and the culture was further incubated at 16 °C and 130 rpm overnight. After induction, cells were washed in 200 ml IMAC loading buffer (50 mM NaH2PO4, 500 mM NaCl, 10 mM Imidazol, pH 7.8) and finally resuspended in 75 ml IMAC loading buffer with Roche complete protease inhibitor (one tablet per 50 ml). Cells were lysed using a microfluidizer with the setting 3 times at 700 bar. Thereafter, the lysate was centrifuged at 15,000g for 90 min. Cleared lysate was loaded onto a 5 ml HisFF column and subsequently washed with IMAC exchange buffer, then 1 M LiCl, again with IMAC exchange buffer, and finally with IMAC loading buffer. The proteins were eluted with IMAC elution buffer (50 mM NaH2PO4, 500 mM NaCl, 400 mM Imidazol, pH 7.8) and dialysed against heparin base buffer (40 mM HEPES KOH, 100 mM KCl, pH 7.5) overnight. After dialysis, the eluate was loaded onto a heparin column and again eluted with heparin elution buffer (40 mM HEPES-KOH, 100 mM KCl, 2 M NaCl, pH 7.5). Finally, the eluate was dialysed in dialysis buffer (30 mM HEPES-KOH, 160 mM KCl, pH 7.6) for 2 days. Protein concentration was determined by measuring the optical density at 280 nm.

EMSA

Either ordered FAM or Cy5-labelled RNAs (Sigma Aldrich) were used. Every RNA contained 36 nucleotides and had the same amount of C, G, T and A (Supplementary Table 4). dsRNAs were formed by incubating 20 µM of the labelled and 20 µM of the reverse complementary non-labelled RNA in dialysis buffer (30 mM HEPES-KOH, 160 mM KCl, pH 7.6) at 65 °C for 5 min and immediate subsequent cooling on ice. Next, 4 µM dsRNAs or ssRNAs were incubated with purified Mex67–Mtr2 and 2 µl Ribolock RNase Inhibitor (Thermo Fisher) in dialysis buffer, resulting in a final volume of 20 µl, at 30 °C for 15 min. For Fig. 2c, Mex67–Mtr2 was added in increasing amounts from 4 µM to 44 µM. For the competition assay depicted in Fig. 2d, 12 µM Mex67–Mtr2 was added to 3 µM substrate RNA, resulting in a molar ratio of 1:4 between substrate RNA and Mex67. The competitor RNA was added after the first incubation and further incubated at 30 °C for 15 min. Finally, a 6× loading dye (10 mM Tris-HCl, pH 7.6, 60% glycerol, 60 mM EDTA, 0.03% bromophenol blue) was added and the samples were loaded onto a 0.5% agarose gel with 1× TAE (40 mM Tris, 1 mM EDTA, 20 mM acetic acid, pH 9.5) running in 1× TAE, pH 9.5. Complexes were separated by running native gels at 300 V and 4 °C for 40 min. In-gel detection was carried out with a Fusion FX7 Edge 18.06c (Vilber) using the filter F-595 YR and Epi-Light module C530, or filter F-710 and Epi-Light module C640, with Evolution-Capt. Edge software.

Export release assay

The mex67-5 xpo1-1 RPS3-GFP strain was grown to the mid-logarithmic phase (2 × 107 cells per ml) and shifted to 37 °C for 2 h. Cells were collected either directly after shifting (0 min) or after shifting them back to 25 °C for 5 min, 10 min, 15 min, 30 min and 60 min. The cell pellets were frozen in liquid nitrogen and subsequent RIP experiments were carried out as described for the J2 RNA co-immunoprecipitation experiments, with the exception that GFP Trap beads were used and no antibody was added. After the final washing step, the beads were split in half for RNA isolation with TRIzol reagent and subsequent qPCRs and for SDS-PAGE and western blot analysis of GFP (anti-GFP in TBS-T, 1:4,000) and Aco1 (anti-Aco1 in TBS-T, 1:2,000). For qPCR measurements, the ssRNAs RPS17A, RPS6A and TDH1 and the dsRNAs FRE5, HPF1 and PRY3 were analysed. dsRNA targets were chosen using three criteria: the asRNA had a higher RPKM (reads per kilobase million) than the sense RNA; they were identified as dsRNA in an RNAi-seq experiment21; and they are enriched in J2 RNA-seq. The ssRNA targets were chosen because of the opposed criteria: the level of the asRNA is less than 1:100 compared with the mRNA and they are not enriched in either RNAi-seq nor J2-seq.

Cell lysis for protein and RNA quantification

Cells (30 ml) were grown overnight in synthetic medium containing 2% raffinose until the logarithmic phase. For asRNA induction, 2% galactose was added. At each indicated time point, a sample of 5 ml was taken and centrifuged at 4,000 rpm and 4 °C for 4 min. Cells were lysed in 400 µl RIP buffer (25 mM Tris-HCl, pH 7.5, 150 mM NaCl, 2 mM MgCl2 0.5% (v/v) Triton X-100, 0.2 mM PMSF, 0.5 mM DTT, 10 U RiboLock RNase inhibitor (Thermo Fisher) and protease inhibitor (Roche)) and divided into two samples. SDS loading dye (125 mM Tris-HCl, pH 6.8, 2% SDS, 10% glycerol, 5% 2-mercaptoethanol, bromphenolblue) was added to one of the samples. Subsequently, samples were incubated at 95 °C for 5 min and loaded onto an SDS gel followed by western blotting and staining of GFP (anti-GFP in TBS-T, 1:4,000) and Hem15 (anti-Hem15 in TBS-T, 1:5,000). The RNA was isolated from the second sample using the NucleoSpin RNA kit (Macherey Nagel) and quantified by qPCR.

Strand-specific cDNA synthesis and qPCR

To exclusively measure either mRNA or asRNA in qPCR, RNA-specific reverse primers were used (Supplementary Table 3) in cDNA synthesis (Nippon Genetics) and two separate samples were created. Each contained either the asRNA primer or the mRNA primer, resulting in separate asRNA and mRNA cDNAs. Furthermore, actinomycin D was added together with the reverse transcriptase because it prevents non-specific transcription from DNA and thereby secures strand-specific transcription, as reported previously48,49. In the qPCRs, the corresponding cDNA of mRNA and asRNA from one gene were measured with the same primer pair.

Drop-dilution analysis

Cells were grown to the logarithmic phase (2 × 107 cells per ml) and diluted to 1 × 106 cells per ml. Then, 10-fold serial dilutions to 1 × 103 cells per ml were prepared and 8 μl of each dilution was spotted onto selective plates. The plates were incubated for 3 days at the indicated temperatures and conditions. Pictures were taken after 2 or 3 days with an Intelli Scan 1600 (Quanto Technology) and the SilverFast Ai program.

RNA-seq

The sequencing of RNA samples was conducted at the NGS-Integrative Genomics Core Unit of the University Medical Center Göttingen. Samples were prepared with the TruSeq RNA Sample Prep Kit v.2, according to the manufacturer’s protocol (Illumina). Single-read (50 bp) sequencing was conducted using a HiSeq 4000 (Illumina). Fluorescence images were transformed to BCL files with the Illumina BaseCaller software (v.3.6.3) and samples were demultiplexed to FASTQ files with bcl2fastq (v.2.17).

Differential gene-expression analysis

Sequences were aligned to the genome reference sequence of Saccharomyces cerevisiae (sacCer3, obtained from UCSC; https://hgdownload.cse.ucsc.edu/goldenPath/sacCer3/bigZips/) using STAR software50 v.2.5, allowing for two mismatches. Subsequently, abundance measurement of reads overlapping with exons or introns was conducted with featureCounts51, subread v.1.5.0-p1, Ensembl (EF4.68) supplemented with the coordinates of UTRs, CUTs, SUTs22,52,53 and XUTs3,29. Data were processed in the R/Bioconductor environment (www.bioconductor.org, R v.3.6.1) using the DESeq2 package54; v.1.24.0). The sequencing data and abundance measurement files have been submitted to the NCBI Gene Expression Omnibus database. For null-hypothesis testing, the Wald test was used with multiple comparison adjustments using the Benjamini and Hochberg method. In downstream analysis, only transcripts with an average count above 40 were considered.

Sense–antisense-pair identification

Overlapping sense–antisense pairs were identified using BEDTools intersect (v.2.3.1)55, requiring overlaps to occur on the opposite strand with a minimum overlap of 0.5. lncRNAs were considered in analysis as SUTs, XUTs or CUTs only if they do not overlap with other transcripts of the other types on the same strand.

RNAi coverage analysis and classification

For gene coverage of RNAi degradation products, reads were trimmed using Cutadapt (v.2.1)56 and aligned to the reference genome with TopHat2 (v.2.1.1)57. For gene coverage, the geneBody_coverage module of the RSeQC package was used (v.2.6.4)58. The input BED file was filtered by lncRNA classes (SUT, CUT or XUT) or by RNA enrichment in RNAi-seq. Overlapping features on the same strand were excluded. To calculate the enrichment in RNAi-seq, the read densities of a transcript in the RNAi strain was divided by its read densities in the wild type. Subsequently, the logarithm to base 2 of this ratio was calculated. For subsequent analyses, transcripts were grouped into ten groups from −5 to 5 without 0, on the basis of their log2-transformed fold change (log2 [RNAi/wild type]). Group 1 contained transcripts with changes between 0 and 1; group 2 contained changes between 1 and 2, and so on. Finally, group 5 contained transcripts that have a log2-transformed fold change above 4. In the negative range, the classification was made in the same way.

Dimethyl sulfate reactivity analysis

The dimethyl sulfate reactivity assay was carried out as previously described45. The dimethyl sulfate reactivity for each transcript was summed in wild type and in dbp2∆. The average reactivity in the wild type was subtracted from the average reactivity in dbp2∆ to obtain the structural change between the strains.

Statistics and reproducibility

Experiments from which a significance was calculated were conducted independently at least three times. In Figs. 1h,i, 2f, 4c,h and Extended Data Figs. 4b, 7b,c,e, 8d,e,g and 9g, data are presented as mean values ±s.d., two-sided t-test P < 0.05*, P < 0.01**, P < 0.001***. In Figs. 1l, 4g and Extended Data Fig. 4a, the box plots are defined by the median as the centre line, the 25th and 75th percentiles as the box boundaries and the 10th and 90th percentiles as the whiskers. Two-sided Welch’s t-test, P < 0.05*, P < 0.01**, P < 0.001***, P < 0.0001****. In Figs. 1a–d, 2c,e, 3a,d, 4c and Extended Data Figs. 4c, 5a,b and 8b, the box plots are defined by the median as the centre line, the 25th and 75th percentiles as the box boundaries and the minimum and maximum values as the whiskers. Two-sided t-test, P < 0.05*, P < 0.01**, P < 0.001***, P < 0.0001****. In Fig. 2e, the centred asterisks show a significant enrichment compared with time point 0. One-sided t-test, P < 0.05*, P < 0.01**, P < 0.001***, P < 0.0001****. Spearman’s rank correlation analysis in Extended Data Fig. 5c and associated and between repetitions in Extended Data Figs. 1d, 2c and 3b were calculated using GraphPad PRISM.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.