|Home | About | Journals | Submit | Contact Us | Français|
RNA structural transitions are important in the function and regulation of RNAs. Here, we reveal a layer of transcriptome organization in the form of RNA folding energies. By probing yeast RNA structures at different temperatures, we obtained relative melting temperatures (Tm) for RNA structures in over 4000 transcripts. Specific signatures of RNA Tm demarcated the polarity of mRNA open reading frames, and highlighted numerous candidate regulatory RNA motifs in 3′ untranslated regions. RNA Tm distinguished non-coding versus coding RNAs, identified mRNAs with distinct cellular functions. We identified thousands of putative RNA thermometers, and their presence is predictive of the pattern of RNA decay in vivo during heat shock. The exosome complex recognizes unpaired bases during heat shock to degrade these RNAs, coupling intrinsic structural stabilities to gene regulation. Thus, genome-wide structural dynamics of RNA can parse functional elements of the transcriptome and reveal diverse biological insights.
Differential stability of RNA structures in the transcriptome corresponds to the diverse roles that RNA structures play in the cell. RNA structures can influence each step in the life cycle of a gene—from transcription, to pre-mRNA splicing, RNA transport, translation, and RNA decay (Wan et al., 2011). However, it is difficult to identify functional structural elements in the transcriptome because practically every RNA has the propensity to fold into extensive RNA structures. In addition to whether a base is paired, the stability of base pairing impacts the biological function of RNAs in important ways (Ringner and Krogh, 2005). Some RNAs, such as ribozymes and structural RNA scaffolds(Guo et al., 2004; Wang and Chang, 2011), form stable secondary and tertiary structures; other RNAs, such as RNA thermometers and riboswitches, undergo structural rearrangements at specific temperatures or in the presence of ligands, respectively, to mediate gene regulation (Breaker, 2010; Chowdhury et al., 2006). As such, differential RNA stability is one way to distinguish diverse RNA structures and to identify functionally important elements in the transcriptome. While RNA folding energies are difficult to predict computationally because of contributions from complex tertiary RNA structures and ligand interactions(Wilkinson et al., 2005), RNA folding energies have been experimentally probed by measuring RNA Tm via several methods(Luoma et al., 1980; Rinnenthal et al., 2010; Wilkinson et al., 2005). Tm is defined as the temperature at which half of the molecules of a double-stranded species become single-stranded. RNA structures of low Tm are more dynamic and exhibit lower energetic cost to unwind and access; conversely, RNA structures of high Tm are relatively more stable and demand higher energetic cost to unfold.
We recently reported genome-wide RNA structure data for the yeast transcriptome by coupling RNA footprinting, using RNase V1 and S1 nuclease, to high throughput sequencing (termed Parallel Analysis of RNA Structures, or PARS) (Kertesz et al., 2010). However, the relative stabilities of these structures and their influence on cellular biology remain unanswered. Inspired by the precedent of Tm measurement via RNA footprinting (e.g. SHAPE(Wilkinson et al., 2005), here we directly measure the melting temperature at single nucleotide resolution across the yeast transcriptome. We coupled RNA footprinting using RNase V1 to high throughput sequencing to probe for double stranded regions across 5 temperatures, from 23 to 75 Degrees Celsius (°C) (Fig. 1A). This approach, termed Parallel Analysis of RNA structures with Temperature Elevation (PARTE), revealed the energetic landscape of the transcriptome and its multiple roles in post-transcriptional regulation.
To carry out PARTE, we first defined conditions that allowed comparable results at different temperatures. RNA footprinting with RNase V1 of the well known, structured domains of the Tetrahymena ribozyme (Guo et al., 2004) revealed that RNase V1 retains its double-stranded specificity up to 75°C, and comparable footprinting results are obtained at different temperatures by correspondingly shorter incubation times with the enzyme to maintain single-hit kinetics (Fig. S1). Optimized conditions defined by these experiments were then used to probe RNA folding at different temperatures genome-wide. Next, we extracted total RNA from log-phase growth culture of yeast and performed polyA selection to enrich for mRNAs. Control RNAs, including domains of the Tetrahymena ribozyme and human long non-coding RNAs HOTAIR and HOTTIP were added into the reactions (Rinn et al., 2007; Wang et al., 2011). The RNA pool was folded in vitro at 23 and 30°C, and the 30°C pool i s split and shifted to 30, 37, 55, or 75°C for five m inutes. The RNA samples were then subjected to RNase V1 treatment with single hit kinetics, and the resulting fragments were cloned for deep sequencing on the SOLiD platform as previously described (Kertesz et al., 2010) (Fig. 1A). We performed two biological replicates for each temperature, yielding ten PARTE experiments in total. We generated over 3 million deep sequencing reads for each sample, and mapped the reads to the yeast transcriptome to identify the cleavage sites (Supp. Table 1). Footprinting of the doped-in P4P6 domain of the Tetrahymena ribozyme showed that PARTE signals closely reproduced the traditional gel-based RNA footprint patterns across temperatures, indicating that the sequencing information accurately captures dynamic changes in RNA structure (Fig. S1). We applied the PARTE data to examine the thermo-stability of the RNA subunit of the Signal Recognition Particle, SCR1 (Fig. 1B). Temperature transitions within local regions obtained by PARTE corresponded to those measured by UV spectroscopy (which detects base unstacking during melting of structured RNA) (Fig. 1C), and identified differential stabilities of specific bases and helices in SCR1 RNA structure (Fig. 1D, S2). This provides a nucleotide resolution view of RNA structural stability, with each interrogated base serving as a probe of its local structural context.
Pair-wise comparison of all ten PARTE samples showed that each biological replicate is most similar to each other with a single exception, while the samples showed progressively more differences as the temperature gradation increased (Fig. 2A). Analysis of data from all ten PARTE samples showed that stepwise increases in temperature led to stepwise losses of double-stranded regions, as expected (Fig. 2B). We developed a computational procedure to infer Tm from PARTE data (Methods). In brief, the data were normalized by the library sizes estimated by PoissonSeq (Li et al., 2011). Normalized data were fitted to an adaptive regression model to search for sharp transitions in read numbers at each base as a function of temperature (p<0.05, FDR=0.2, compared to running the same analysis on 100 permuted data sets)(Sahoo et al., 2007). Only bases that show a consistent single transition at a particular temperature across both biological replicates were included in downstream analysis. Indeed, single, sharp transitions allowed an approximate Tm to be determined for more than 350,000 bases in the yeast transcriptome, mapping to 4016 distinct yeast mRNAs and 65 non-coding RNAs (Fig. 2C). Approximately 80% of these bases had a transition from high RNase V1 reads to low V1 reads with temperature elevation (i.e. double stranded RNA (dsRNA) became single stranded RNA (ssRNA)), with the majority of the bases melting between 37 and 55°C (Fig. 2C). 20% of the Tm showed a transition for increased V1 reads at higher temperatures (55 and 75°C); these bases likely represent sites in thermodynamically stable dsRNA structures, which were either increasingly sampled at higher temperatures as other paired bases melted away or alternatively became accessible to RNase V1 upon dissolution of tertiary structures. We obtained an estimated melting temperature (eTm) per gene by averaging all the melting temperatures per base in that gene. Bases that do not melt by 75°C are assigned a Tm of 80°C for this calculation. To test the accuracy of our eTm, we determined the UV melting temperatures of twelve randomly chosen transcripts of approximately 100 bases, as UV spectrometry works best for short RNAs. Direct comparison of estimated Tm by PARTE versus UV spectroscopy for these RNAs showed good concordance (Spearman rank correlation, R=0.59, p<0.05); the concordance is even better for five of the transcripts with the most read coverage in PARTE data (>20% bases measured, Spearman rank correlation, R=0.9, Fig. S3A). These lines of evidence suggest that PARTE data are of good quality and are highly reproducible. PARTE estimates of Tm are demonstrably accurate at base-resolution, and PARTE estimates of Tm per gene requires sufficiently deep sequence reads to generate transcript-wide coverage.
As another independent validation, we compared PARTE Tm to Tm predicted by computational algorithm. We used RNAFold (Gruber et al., 2008) to simulate the folding of the same set of yeast transcripts into secondary structures at 23, 30, 37, 55, or 75°C, and then extracted the Tm per base from the predicted secondary structures. Because secondary structure is one of the important building elements of folding energy, we expect that the predicted Tm will correlate to some degree with measured Tm, but will also deviate from the measured Tm because the algorithm does not model tertiary interactions. Indeed, we found that PARTE data is significantly correlated with computationally predicted Tm, but there are also substantial differences (R=0.24, p<10−39, Fig. S4). For the twelve transcripts that we determined Tm by UV spectroscopy, computational prediction correlated poorly with measured Tm (Spearman rank test, R=-0.2), performing substantially worse than PARTE. Thus, we suggest that PARTE data can be used in addition to computational algorithms to better estimate RNA folding energy.
We first examined the thermodynamic properties of non-coding versus coding RNAs. Non-coding RNAs (ncRNAs) such as ribosomal RNAs, transfer RNAs, and RNase P are known to exhibit extensive secondary and tertiary structures of high stability (Clote et al., 2005). Indeed, the average melting profile of structured ncRNAs (rRNA, tRNA, snoRNA, snRNA) is distinct from that of mRNAs (Fig. 3A). While almost no base pairs in ncRNAs melted at < 37°C, ~25% of bases in mRNAs melted by 37°C (odds ratio =44:1, p<0.0001(Fisher Exact Test), for ncRNA over mRNA if <1% of bases in a transcript unpair at 30°C). ncRNAs also contain significantly more bases that remain paired at 75°C (odds ratio =82:1, p<0.0001(Fisher Exact Test), for ncRNA over mRNA if > 60% of bases remain paired at 75°C), which we term “stable bases”. Noncoding RNAs also showed higher melting temperature per gene as compared to mRNAs (p=0.000018, Wilcoxon rank sum test, Fig. 3B), suggesting that ncRNAs contain more thermodynamically stable structures than mRNAs. We faithfully recapitulated the high structural stability of the Tetrahymena ribozyme in our data (Schultes et al., 2005; Szewczak et al., 1998) (Fig. 3A). Two other yeast ncRNAs (TLC1, encoding telomerase RNA; SCR1, the RNA subunit of the Signal Recognition Particle) and the two doped-in human lincRNAs also show a profile that resembles the structured ncRNAs, indicating that structural stability may be a feature of some of these ncRNAs (Fig. 3A). Interestingly, the non-coding RNA SRG1, a product of antisense transcription that silences SER3 expression by transcription interference, has no known RNA function and showed a low stability profile that deviates from ncRNAs (Fig. S3B). This example provides further support that PARTE can identify functional RNAs whose structures are important for their biological roles. The difference between mRNAs and functional ncRNAs persisted even if the untranslated regions (UTRs) of mRNAs were removed from consideration, and UTRs alone showed an intermediate profile between coding sequences and ncRNAs. The differing trend of RNA Tm for coding versus structured ncRNAs provides global experimental support for prior computational predictions(Clote et al., 2005).
Because PARTE identifies RNA Tm with nucleotide level resolution, we next addressed the energetic landscape across specific portions of mRNAs (Fig. 4A). Previously, a static snapshot of global mRNA secondary structure showed that coding sequences (CDS) tend to be more structured than 5′ and 3′UTRs, and that two regions with least stable base pairing—at the start and stop codons— demarcated the CDS (Kertesz et al., 2010). However, the symmetric pattern of RNA structural landscapes at start and stop codons raised the question of whether the structural landscapes of mRNA also encode the 5′ to 3′ polarity of translation. Alignment of PARTE profiles for mRNAs at the start and stop codons revealed a striking pattern that extended prior descriptions (Fig. 4A-C). First, the PARTE profiles confirmed the global pattern of RNA structure across functional portions of mRNAs, and confirmed predictions that UTRs are generally more accessible (Kertesz et al., 2010; Ringner and Krogh, 2005). Second, the PARTE data showed that the two local regions with weakest pairings at start and stop codons are in fact flanked by diametrically opposite structural contexts. The most meltable part of the entire mRNA, on average, resides in nucleotides -3 to +3 surrounding the start codon. In contrast, the most stably paired region of the CDS maps to ~20 nucleotides immediately upstream and 10 nucleotides immediately downstream of the stop codon (p<0.01 for each, permutation test, Fig. 4C). This suggests the presence of a polarity in the energetic landscape of mRNA folding. Correlation of Tm with GC content showed that there is some correlation with GC content at the gene level (R=0.36), however these Tm differences at the ends of mRNAs are not simply explained by GC content as meta-gene analysis of the GC content of these genes do not display the same polarity in base pairing as seen in RNA melting. This polarity of structure stability is hence likely to be consequences of secondary or higher order structures of RNAs that are directly detected by PARTE (Fig. 4D, S4). Several reports have shown that both codon usage and RNA structure affects translation efficiency (Cannarozzi et al., 2010; Tuller et al., 2010). We and others have hypothesized that the 5′ UTR immediately proximal to the start codon is the energetically most favorable portion of mRNAs, which may facilitate ribosome access. Consistent with this idea, we find a significant and positive correlation between the propensity of bases upstream of the start codon to melt, and the translation efficiency of the gene (Fig. 5B). In contrast, near the end of the open reading frame, it is possible that the ribosome can encounter a plateau of energetic barrier to unwind the RNA before it reaches the stop codon.
To search for stable RNA structures within the yeast transcriptome, we identified continuous regions of RNA bases with Tm greater than 75°C. Consistent with the increased propensity to melt at the 5′ UTR, we notice that on average, the 5′UTR contains fewer stable bases (1.3%), as compared to the coding region (19.7%) and 3′ UTR (19.3%) (Fig. 5A). Out of 820 5′ UTRs for which we have Tm information, 35 contain stable bases, and they are significantly enriched in mRNAs that encode proteins localized to the cellular membrane (p=0.0017) (Fig. S5A). This finding raises the possibility that stable structures in the 5′ UTR may serve as targeting signals to the membrane fraction of the cells, although they may also influence the translation efficiency of these genes as a class. Stable bases in the 3′ UTR map to 337 genes, out of a total of 931 3′UTRs with Tm data, which is far more concentrated than expected by chance alone (Fisher’s exact test, p=2.12 ×10−67), suggesting that the stable bases in 3′ UTR, if present, tend to be clustered and may encode regulatory RNA motifs. Indeed, ranking 3′UTRs by the length of continuous regions of stable base pairing in 3′ UTRs identified known functional structured RNA regulatory elements, and nominated candidates in the yeast transcriptome, indicating the utility of our dataset (Supp. Table 2).
The most stable RNA structure in 3′UTRs is in HAC1, encoding a key transcription factor for the unfolded protein response (UPR)(Aragon et al., 2009); HAC1 mRNA contains a stretch of 26 stable bases in its 3′ UTR (Fig. 5C). Importantly, the region identified by our data coincides exactly with the structural RNA element required for HAC1 mRNA localization to the ER membrane during heat shock, leading to HAC1 mRNA splicing and protein production (Aragon et al., 2009) (Fig. 5D). The second hit on our list is RPS28B, encoding a ribosomal protein which regulates its own mRNA level via decapping (Badis et al., 2004). RPS28B has a RNA structure in its 3′UTR that is involved in recruiting decapping enzymes for mRNA degradation, and the region that we identified is immediately adjacent to this structure (Fig. S5B). Moreover, stable RNA structures in 3′ UTRs may direct the cytotopic localization of mRNAs, an important mechanism for the co-translational localization of proteins associated with membranes and organelles. Many genes with stable RNA structures in 3′ UTRs encode proteins that function in the endoplasmic reticulum membrane, Golgi apparatus, or the mitochondria (Supp. Table 2). These genes are significantly enriched for high membrane-to-free (MFI) index after cellular sub-fractionation, indicating that they are more likely to be membrane associated than cytoplasmic (Sylvestre et al., 2003) [p<0.05, GSEA test (Subramanian et al., 2005)]. Supporting the idea that stable bases are functional, these bases are also significantly more conserved across evolution than low Tm bases (p<10−20, student’s t-test, Fig. 5E). Table S3 contains the entire list of stable base pairing in 3′UTRs, as well as the precise location of candidate RNA motifs in each of these transcripts, which may serve as a useful resource as a starting point for finding localization and other structural elements in yeast mRNAs.
The energetic landscape of the transcriptome may facilitate the generation of diverse biological hypotheses. To identify functional classes enriched in genes with high or low eTm, we intersected sets of genes in the top 10 percentile (high) or bottom 10 percentile (low) eTm with Gene Ontology terms and genome-wide datasets of heat shock, and RNA binding proteins (Gasch et al., 2000; Hogan et al., 2008). Interestingly, low eTm genes are enriched for distinct functional classes, including ribosomal proteins (hypergeometric test, p=2.34 ×10−74, FDR<0.05), and targets of specific RNA binding proteins (RBPs) (hypergeometric test, p<2×10−4). The enrichment of ribosomal protein genes is not due to gene duplication (p<0.05, FDR<0.05 for all enrichments even if only one of two orthologs are considered), and gene location across the linear genome is not significantly clustered based on the eTm of the encoded transcripts.
Importantly, mRNA folding energies in vitro is predictive of the pattern of RNA decay during heat shock in vivo. In bacteria, a key element of the heat shock response is “RNA thermometers”—RNA structures occluding ribosome binding sites in mRNAs of key regulators that melt at the heat shock temperature to enable protein synthesis (Chowdhury et al., 2006). The location and number of “RNA thermometers” in a eukaryote are not known. In yeast, we identified 25721 bases (7.4 percent of all measured) that specifically unpair upon a shift from 30 to 37°C, which are clustered at more than 1800 sites in the transcriptome (Fig. 6A, Supp. Table 4). These bases are enriched in the 5′UTRs (p<2e-16, chi-square test), and specifically around the ribosomal binding site, suggesting that some of these bases may also unpair during yeast heat shock to regulate translation (Fig. 6B). Global gene expression profiling had identified gene sets that were induced or repressed in heat shock and other stresses (Gasch et al., 2000). Interestingly, our low eTm genes are highly enriched for these rapidly decreasing transcripts during the heat shock process (hypergeometric test, FDR<0.05, p-value of enrichment at 20, 30, 40, 60 and 80 min are 1×10−38, 1×10−36, 1×10−42, 1×10−27 and 1×10−4 respectively) (Fig. 6C). Interestingly, stratification of mRNAs by eTm over their lengths readily predicted the patterns of mRNA abundance during heat shock in vivo: mRNAs with the top quantile of eTm (most structurally stable) showed little decline in transcript level during heat shock, and mRNAs in quantiles of progressively lower eTm showed correspondingly larger and prolonged decrease in transcript levels (Fig. 6D).
The ability of eTm to predict the pattern of heat-shock induced RNA decay in vivo suggests that RNA structure modulates the activity of one or more components of the RNA decay machinery, a potentially new function of RNA thermometers. The exosome, the major 3′ to 5′ exonuclease in the cell, is a cage-like RNA processing machine; tunneling RNA into the exosome requires unpairing 31-33 nucleotides and is inhibited by structured RNAs with short 3′ overhangs (Bonneau et al., 2009). As such, we reasoned that the exosome is a good candidate for the factor that connects RNA thermometers to RNA decay. Indeed, exosome inactivation preferentially stabilized mRNAs with low eTm during heat shock, as shown by gene expression profiling (Fig. 7A, 101 of 224 RNA thermometers stabilized vs 44.8 expected by chance alone, p=1.3×10−22, hypergeometric test)(Gasch et al., 2000; Grigull et al., 2004; Houalla et al., 2006). mRNAs with progressively lower eTm are stabilized with greater magnitude in core exosome rrp41-1 mutant (Fig. 7B). The connection between the cytoplasmic core exosome and low eTm transcripts is highly specific. Deletion of nuclear exosome components RRP6 failed to stabilize low eTm mRNAs, and deletion of RRP47, another nuclear exosome subunit, only increased the level these mRNAs at a late time point but did not prevent the initial decay (Fig. 7A). Similarly, mutation of CCR4 deadenylase stabilized many more transcripts, and show weaker preference of whether the stabilized genes have low eTm and hence are easily degradable or not, as compared to the exosome complex (Fig. S6). These results suggest that the exosome complex acts as a “reader” to discriminate between structured and unstructured substrates for decay in vivo.
To test if the structural stabilities of the RNAs is sufficient to determine RNA decay, we tested the ability of the 10 subunit S. cerevisiae exosome reconstituted from recombinant purified proteins to discriminate between endogenous RNAs with high and low eTms. RPL1A mRNA, encoding a ribosomal protein and has low eTm, is rapidly degraded by the exosome in 10 minutes, whereas HAC1, the key UPR activator and possessing a very stable 3′ UTR RNA structure (Fig. 5D), is not processed by the exosome after one hour (Fig. 7C). To further test the structural sensitivity of the exosome, we designed mutations to increase the structural stability of a model RNA hairpin (Dziembowski et al., 2007) in a stepwise fashion, which we confirmed by UV spectroscopy. RNA structures that are increasingly stable progressively blocked exosome processing (Fig. 7D). Importantly, the exosome is active at 37°C to process a RNA hairpin engineered to melt between 30 and 37°C, but not a RNA hairpin with higher Tm (Fig. 7E). A structure-less polyadenylate sequence is constitutively processed by the exosome at low and high temperatures (data not shown). Thus, the substrate preference of the exosome allows this complex to select different RNA structures in a temperature-dependent fashion.
By performing PARTE at several carefully calibrated temperatures, we have measured the relative stabilities of RNA structures in the yeast transcriptome. In essence, PARTE is the genome-scale version of the classical RNA melting experiments that characterizes energetics of RNA folding. PARTE provides complementary information to global RNA structural probing methods (Kertesz et al., 2010; Lucks et al., 2011; Underwood et al., 2010), and can potentially overcome some limitations of static RNase footprinting strategies. The impacts of sequence bias and steric hindrance, two concerns with several footprinting reagents, are minimized by focusing on condition-specific changes of signals that relate to RNA structure. Our results also demonstrate the feasibility of probing dynamic RNA structural transitions genome-wide, which can be extended to study the impact of specific proteins, small molecules, or conditions to yield functional information of RNAs in the cell. This work sets the ground for future improvements in coverage and accuracy. Because we performed polyA selection to enrich for mature abundant mRNAs in wild type cells, introns and lowly expressed antisense or cryptic unstable transcripts were not well interrogated. Another important caveat of our procedure is that RNA structures requiring co-transcriptional folding or native protein-RNA interactions may not fold correctly, but our procedure preserves the in vivo pattern of RNA modifications (e.g. pseudouridines and methylated bases) that is difficult to reproduce with in vitro transcribed RNAs. We were able to accurately capture the known secondary structures of several yeast RNAs and Tetrahymena ribozyme added in as positive control (Kertesz et al., 2010). Classical studies of RNA folding energy are conducted in vitro, and we validated PARTE data with UV spectroscopy. Moreover, the ability of our Tm measurements to predict the pattern of heat-induced RNA decay in vivo and its regulation by the exosome demonstrate the physiological relevance of these findings. Finally, PARTE helps to infer but does not directly identify the base pairing partners; future efforts to combine PARTE with RNA mutagenesis may achieve this goal.
The genome-wide RNA Tm measurements provide a resource to annotate the transcriptome. By analyzing the patterns of RNA Tm, we found distinguishing features of RNA structural stability for several classes of transcripts and elements, including (i) coding versus non-coding RNAs, (ii) the beginning and ends of open reading frames in mRNAs, (iii) candidate regulatory elements in 3′ UTRs, and (iv) RNA thermometers associated with RNA decay during heat shock. Some of these features were previously predicted by computational analyses(Cannarozzi et al., 2010; Clote et al., 2005; Tuller et al., 2010), and our data provide genome-scale experimental support for them. Several of these findings were only evident based on genome-scale experimental data. Our data suggest that the transcriptome has evolved to possess extensive organization of its energetic landscape that can impact RNA cytotopic localization, translation, mRNA decay, and cellular response to stress.
The heat shock response is an evolutionarily conserved program for cells to adapt to temperature stress. Cells shut down general transcription and translation to decrease energy expenditure and accumulation of unfolded proteins, but need to selectively activate the UPR and produce heat shock proteins at the same time (Panniers, 1994). Our data indicate a role for RNA thermometers to direct RNA decay during heat shock. Messenger RNAs with low eTm, such as those encoding ribosomal proteins, are rapidly degraded during the heat shock while transcripts with higher eTm, such as key UPR activators HAC1 and PTC2, are long lived. We found that the exosome preferentially degrade unpaired RNA and hence is the “reader” of the RNA thermometer. Moreover, exosome is selectively required for heat-induced RNA decay in vivo. Because exosome pausing would occur after mRNA deadenylation, our data raises the intriguing possibility of polyA independent translation, which can proceed efficiently in yeast (Brown and Johnson, 2001). Alternatively, deadenylated but not degraded mRNAs may also be re-adenylated by the protein CID13 for rapid gene expression (Saitoh et al., 2002). The low Tm genes, on the other hand, are rapidly turned over in heat shock to shut down translation. These results connect RNA structure, the exosome, and the physiology of the heat shock response. Notably, this regulatory information is apparently not determined by primary sequence motifs or specific secondary structures, but rather may be encoded in differences in RNA folding stabilities that is coordinately organized across the transcriptome.
Total RNA was extracted from log phase S288C yeast using a slightly modified protocol that uses hot, acid phenol(Sigma). Poly(A) RNA was obtained by purifying twice using the Poly(A) purist MAG Kit according to manufacturer’s instructions (Ambion). RNA transcripts of the P4P6 and P9-9.2 domains of the Tetrahymena ribozyme, fragments of HOTAIR and HOTTIP are obtained by PCR followed by in vitro transcription using RiboMAX Large Scale RNA production Systems Kit according to the manufacturer’s instructions (Promega) and purified by PAGE purification. P4P6, P9-9.2, fragments of HOTAIR and HOTTIP were doped into 2μg of yeast poly(A)+ mRNA as controls. Briefly, the RNAs were heated to 90°C for 2min, cooled on ice for 2 min and incubated at 23°C for 20min, after adding 10X RNA structure buffer. The RNA pool is then probed using 0.005U of RNase V1 in 100ul reaction volume for 15min at 23°C.
The RNA pool to be shifted to different temperatures was heated to 90°C for 2min, cooled on ice for 2min, and slowly brought to 30°C in RNA structure buffer for 15min. The RNA pool was then either structure probed at 30°C using RNase V1(Ambion), or shifted to 37°C, 55°C or 75°C for 5min before undergoing structure probing at the respective temperatures using RNase V1. 0.04U, 0.028U, 0.014U, 0.005U of RNase V1 was added at 30°C, 37°C, 55°C, 75°C respectively for 1min. The nuclease reactions were inactivated and precipitated using an inactivation and precipitation buffer (Ambion). RNA ligation to SOLiD™ adaptors, amplification, and sequencing were as described (Kertesz et al., 2010) and detailed in Supplementary methods. The mapping results are provided in Supplementary Table 1.
PARTE data normalization, Tm estimation, and genomic analyses are detailed in Supplementary methods.
In-vitro transcribed RNA was 5′end labeled with P32 ATP using T4 PNK kinase (NEB) as previously described. The radio-labelled RNA was structure probed with RNase V1 in 1ug of poly(A)+ RNA at different temperatures as described below and in supplementary methods.
Yeast genes (SCR1, snR52, YCR024C-A, RDN58-1, YER138W-A, snR60, snR68, snR75, snR24, snR39B, snR18, snR13), as well as P9-9.2 domain of Tetrahymena ribozyme, were PCR amplified and in-vitro transcribed. Each RNA was heat denatured at 90°C for 5min in 900ul of water, snap cooled on ice for 10min before adding 100ul of 10X buffer (100mM sodium cacodylate pH7, 5mM MgCl2, 1M KCl ) and incubating the RNA at room temperature for 30min. UV absorbance was obtained by heating the RNA from 20-95°C using Cary 100 Bio UV-vis spectrometer at 1°C/min. Readings were taken at 260nM every 0.5min. Tm is predicted from curve fitting in Meltwin 3.0 program using the “non-self complementary” parameter. First derivative is obtained using KaleidaGraph and smoothed in excel by taking moving window of 20 data points.
Two 150mer fragments near the 3′ end of RPL1A and HAC1 gene respectively were PCR amplified and in-vitro transcribed. The 44mer RNA substrate was chemically synthesized (Dziembowski et al., 2007). Two point mutations were made to stabilize the stem loop structure in the RNA (44mut2); two more point mutations were made to lengthen the stem loop (44mut4). Melting temperatures of 44mer RNA constructs were obtained using UV absorbance by heating the RNA from 20-95°C, at 1°C/min, in buffer (final concentration: 50mM Hepes pH7.5, 50mM NaCl, 200uM Mg Acetate). Sequences of RNA substrates can be found in supplementary methods. 5′end labeled RNA was heated to 90°C for 2min, snap cooled on ice for 2min and incubated with exosome reaction buffer (final concentration: 50mM Hepes pH7.5, 50mM NaCl, 200uM MgAcetate, 10% Glycerol, 0.1%NP40, 1mM DTT) (Bonneau et al., 2009) at room temperature for 20min before adding exosome (Exo-10) to final 10ul reaction. 2ul aliquots were taken at indicated time points and the reaction was quenched by adding 8ul of RNA loading dye II (Ambion). The reaction products were resolved on a 15% TBE-urea PAGE gel and visualized by phosphorimaging.
We thank G. Sherlock, K. Schwartz, ET Kool, AR Hernández, W. Greenleaf, G. Zheng, B. Batista, MC. Tsai, M.H. Tan, the Life Technologies SOLiD team for assistance and critiques. We thank E. Conti for insights and support for the exosome experiment. This work was supported by National Institutes of Health grant (R01-HG004361). Y.W. is funded by the Agency of Science, Technology and Research of Singapore. E.S. is the incumbent of the Soretta and Henry Shapiro career development chair. H.Y.C. is an Early Career Scientist of the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCESSION NUMBERS The GEO accession number is GSE39680.
Author Contributions: Y.W, M.K., H.Y.C. and E.S. conceived the project; Y.W. and H.Y.C. developed the protocol and designed the experiments; Y.W. performed all experiments; Y.W, K.Q., Z.O., M.K., J.L.,R.T., E.S. and H.Y.C. planned and conducted the data analysis; D.L.M. and E.C. provided the reconstituted exosome complex, R.C.N. helped with sequencing; Y.W. and H.Y.C. wrote the paper with contributions from all authors.