Choose three general pathways in which eukaryotic mrna is typically degraded in eukaryotes.

Abstract

RNA degradation pathways enable RNA processing, the regulation of RNA levels, and the surveillance of aberrant or poorly functional RNAs in cells. Here we provide transcriptome-wide RNA-binding profiles of 30 general RNA degradation factors in the yeast Saccharomyces cerevisiae. The profiles reveal the distribution of degradation factors between different RNA classes. They are consistent with the canonical degradation pathway for closed-loop forming mRNAs after deadenylation. Modeling based on mRNA half-lives suggests that most degradation factors bind intact mRNAs, whereas decapping factors are recruited only for mRNA degradation, consistent with decapping being a rate-limiting step. Decapping factors preferentially bind mRNAs with non-optimal codons, consistent with rapid degradation of inefficiently translated mRNAs. Global analysis suggests that the nuclear surveillance machinery, including the complexes Nrd1/Nab3 and TRAMP4, targets aberrant nuclear RNAs and processes snoRNAs.

https://doi.org/10.7554/eLife.47040.001

eLife digest

Cells contain a large group of DNA-like molecules called RNAs. While DNA stores and preserves information, RNA influences how cells use and regulate that information. As such, regulating the quantities of different RNAs is a key part of how cells survive, grow, adapt and respond to changes. For example, messenger RNAs (or mRNAs for short) carry genetic information from DNA which the cell reads to produce proteins. RNAs that are not needed can be degraded and removed from the cell by RNA degradation proteins.

Most RNA degradation proteins need to be able to bind to RNA in order to work. A technique called “photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation”, often shortened to PAR-CLIP, can detect these proteins on their targets. The PAR-CLIP technique irreversibly links RNA-binding proteins to RNA and then collects those proteins and their bound RNAs for analysis. As with DNA, the RNAs can be identified using genetic sequencing. Degradation often starts at RNA ends, where specialized structures protect the RNA from accidental damage.

Using PAR-CLIP, Sohrabi-Jahromi, Hofmann et al performed a detailed study of 30 RNA degradation proteins in the yeast Saccharomyces cerevisiae. The results highlight the specialization of different proteins to different groups of RNAs. One group of proteins, for example, remove the protective ‘cap’ structure at the start of RNAs. Those mRNAs that are not efficiently producing proteins attracted a lot of these cap-removing proteins. The findings also identify proteins involved in RNA degradation in the cell nucleus – the compartment that houses most of the cell’s DNA.

Together these findings provide an extensive data resource for cell biologists. It offers many links between different RNAs and their degradation proteins. Understanding these key cellular processes helps to reveal more about the mechanisms underlying all of biology. It can also shed light on what happens when these processes fail and the diseases that may result.

https://doi.org/10.7554/eLife.47040.002

Introduction

The abundance of the different eukaryotic RNA species controls cell type and cell fate, and is determined by the balance between RNA synthesis and RNA degradation. Multiple mechanisms exist for RNA degradation (Parker, 2012). RNAs are generally exported to the cytoplasm, where they are degraded with an RNA-specific rate. Such canonical turnover is critical for RNA homeostasis. However, RNAs that are defective with respect to their processing, folding, assembly into RNA-protein particles, or their ability to be translated, are identified and rapidly degraded by surveillance pathways. Surveillance occurs both in the nucleus (Schmid and Jensen, 2018) and in the cytoplasm (Zinder and Lima, 2017). In the nucleus, aberrant RNAs resulting from upstream antisense transcription are quickly degraded. Moreover, some non-coding RNAs (ncRNAs) require processing by the degradation machinery (Houseley et al., 2006). Cytosolic RNAs also vary in their life-time, with mRNAs encoding cell cycle regulators or transcription factors having reported life-times in the range of minutes (Geisberg et al., 2014; Miller et al., 2011), whereas ribosomal RNAs live for days (Turowski and Tollervey, 2015). Therefore, RNA degradation kinetics need to be actively regulated to ensure the optimal life time for each transcript.

RNA degradation can occur from both ends of a transcript, and these processes are often coupled. During canonical mRNA turnover, degradation is thought to be initiated by shortening of the 3´ poly-adenylated (polyA) tail through two major deadenylation complexes, the Pan2/Pan3 complex and the multi-subunit Ccr4/Not complex (Ccr4, Not1, Pop2, Caf40) (Wolf and Passmore, 2014). Specific mRNAs can recruit selected deadenylating factors after loss of the polyadenylate-binding protein (Pab1) that protects the mRNA from degradation on the 3′ end (Finoux and Séraphin, 2006; Goldstrohm et al., 2006; Semotok et al., 2005), but the choice of deadenylation pathway remains unclear. Studies have pointed out a direct link between translation termination and mRNA degradation (reviewed in Huch and Nissan, 2014), in particular deadenylation, which is dependent on Pab1 and Ccr4 (Webster et al., 2018). A proposed stepwise model for deadenylation suggests that first the average yeast polyA tail length of 90 nucleotides (nt) is reduced to 50 nt by the Pan2/Pan3 complex, before further shortening via the Ccr4/Not complex (Beilharz and Preiss, 2007; Brown and Sachs, 1998; Tucker et al., 2001). When the polyA tail reaches a length of 10–12 nt, the mRNA is decapped (Chowdhury et al., 2007; Tharun and Parker, 2001), or subjected to exosome catalyzed degradation (Bonneau et al., 2009).

The second step in mRNA degradation is thought to be the removal of the 5´ cap by the decapping complex (Dcp1, Dcp2, Dcs1). The cap protects the mRNA from degradation by the 5´→3´ exonuclease Xrn1, which requires a 5´ monophosphate at the terminal residue (Stevens and Poole, 1995). Decapping is highly regulated by decapping enhancers such as the DEAD box helicase Dhh2, Edc2 and Edc3. Different potential mechanisms to trigger decapping include interference with translation initiation factors, facilitated assembly of the decapping machinery, and stimulation of Dcp2 catalytic function. Assembly of the decapping machinery occurs mainly after shortening of the polyA tail, which triggers decapping complex formation on the deadenylated 3´ end of mRNA and opening of the mRNA closed-loop structure (Caponigro and Parker, 1995; Morrissey et al., 1999). In this closed-loop model, the 5′ and 3′ ends of the mRNA are thought to be in close proximity by forming a complex between translation initiation factors binding to the 5´ cap and Pab1 associated with the 3´ end, thereby contributing to mRNA expression regulation (Vicens et al., 2018; Wells et al., 1998).

An alternative pathway of mRNA degradation after deadenylation is 3′→5′ degradation by the exosome and its auxiliary factors (Anderson and Parker, 1998). The exosome is a multi-subunit complex that consists of 10 core factors, comprising six members of the RNase PH protein family (Rrp43, Rrp45, Rrp42, Mtr3, Rrp41, Rrp46), three small RNA-binding proteins (Csl4, Rrp40, Rrp4) (Allmang et al., 1999), and the Rrp44/Dis3 protein, which harbors an exonuclease and an endonuclease domain (Lebreton and Séraphin, 2008; Schaeffer et al., 2009). In addition to its functions in the cytoplasm, the exosome fulfills multiple roles in nuclear RNA processing and degradation (Lykke-Andersen et al., 2009; Ogami et al., 2018) for which it is additionally bound by Rrp6, another 3′→5′ exonuclease, Rrp47, and Mpp6 (Milligan et al., 2008; Mitchell et al., 2003; Synowsky et al., 2009).

For RNA degradation by the exosome, RNA first passes through either the TRAMP or the Ski complex. TRAMP is a nuclear poly-adenylation complex (Houseley and Tollervey, 2009) that is involved in many of the RNA maturation and degradation processes and exists in two isoforms, TRAMP4 (Trf4, Air2 and Mtr4) and TRAMP5 (Trf5, Air1 and Mtr4). These complexes harbor a pA polymerase (Trf4 or Trf5), a zinc-knuckle putative RNA-binding protein (Air1 or Air2), and an RNA helicase (Mtr4). Defective nuclear RNAs are tagged with a short polyA tail by TRAMP, making them a more favorable substrate for the exosome core (Vanácová et al., 2005). The Ski complex is required for cytoplasmic exosomal degradation. The Ski7 protein is stably bound to the cytoplasmic exosome through the Ski4 subunit (van Hoof et al., 2002). The Ski2, Ski3, and Ski8 proteins form a subcomplex interacting with Ski7, which is required for 3′→5′ degradation of mRNAs (Araki et al., 2001; Brown et al., 2000; Wang et al., 2005). The Ski2 protein is an ATPase of the RNA helicase family that generates energy by ATP hydrolysis to unwind secondary structures and dissociate bound proteins to deliver the RNA to the exosome.

To degrade eukaryotic mRNAs that are defective in translation, cytoplasmic quality control mechanisms exist (Doma and Parker, 2007). Normal and aberrant mRNAs can be discriminated by the translation machinery, and translationally defective mRNAs are guided to a degradation pathway. mRNAs with aberrant translation termination due to a premature translation termination codon are subjected to nonsense-mediated decay (NMD) (Losson and Lacroute, 1979). Substrates for NMD are identified by the Upf1 protein interacting with the translation termination complex followed by binding of the Upf2 and Upf3 proteins, which enhances the helicase activity of Upf1 (Baker and Parker, 2004; Chakrabarti et al., 2011). During NMD, the mRNA can be subjected to enhanced deadenylation, deadenylation-independent decapping and rapid 3′→5′ degradation (Cao and Parker, 2003; Mitchell and Tollervey, 2003; Muhlrad and Parker, 1994).

The large variety of different RNA degradation factors poses the question how RNA degradation pathways are selected and whether the RNA sequence can influence this selection. Answering this question requires a systematic analysis of the RNA-binding profiles of the involved protein factors. Although several transcriptome profiles of the RNA degradation factors Xrn1, Rrp44, Csl4, Rrp41, Rrp6, Mtr4, Trf4, Air2 and Ski2 have been reported (Delan-Forino et al., 2017; Milligan et al., 2016; Schneider et al., 2012; Tuck and Tollervey, 2013), we lack transcriptome-wide binding profiles for components of the deadenylation, decapping, and NMD machineries, as well as other subunits of the exosome complex. Thus, the task of systematically analyzing the binding of subunits from all known factors involved in RNA degradation to a eukaryotic transcriptome (‘transcriptome mapping’) has not been accomplished thus far.

Here we used photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP) to systematically generate transcriptome-wide protein binding profiles for 30 general RNA degradation factors in the yeast Saccharomyces cerevisiae (S. cerevisiae). In-depth bioinformatic analysis and comparisons with previously reported PAR-CLIP data provide factor enrichment on different RNA classes and the binding behavior for mRNAs and their associated antisense transcripts. The results also give insights into how the various degradation complexes, and also different subunits in these complexes, may be involved in the degradation of different RNA species. Several conclusions can be drawn with respect to degradation pathway selection, new functions for known factors can be proposed, and several hypotheses emerge that can be tested in the future. Finally, our dataset provides a rich resource for future studies of eukaryotic RNA degradation pathways, mechanisms, and the integration of mRNA metabolism.

Results

Transcriptome maps for 30 RNA degradation factors

In order to get a better understanding of RNA processing and degradation in a eukaryotic cell, we measured transcriptome-wide binding locations of 30 RNA degradation factors involved in mRNA deadenylation, decapping, exosome-mediated degradation, and in RNA surveillance pathways including nuclear RNA surveillance and cytoplasmic nonsense-mediated decay (NMD) (Figure 1A,B). We performed PAR-CLIP in S. cerevisiae using our published protocol (Battaglia et al., 2017), with minor modifications (Materials and methods). The high reproducibility of these PAR-CLIP experiments is revealed by a comparison of two independent biological replicates that we collected for all 30 degradation factors (Figure 1—figure supplement 1), with Spearman correlations between 0.87 and 1.00 (mean: 0.94). We typically obtained tens of thousands of verified factor-RNA cross-link sites with p-values≤0.005 (Figure 1B). These transcriptome maps represent an extensive, high-confidence dataset of in vivo RNA-binding sites for factors involved in RNA degradation.

Overview of PAR-CLIP experiments performed in this study.

(A) Overview of degradation pathways studied. (B) Number of high-confidence PAR-CLIP cross-link sites obtained for each factor after merging data from replicates.

https://doi.org/10.7554/eLife.47040.003

Degradation factors exhibit transcript class specificity

We first compared degradation factor binding over different RNA classes. These included messenger RNA (mRNA), where we distinguished the 5´ untranslated region (5´ UTR), the coding sequence (CDS), introns, and the 3´ untranslated region (3´ UTR). We also included several classes of ncRNAs: ribosomal (r), transfer (t), small nucleolar (sno), and small nuclear (sn) RNAs, as well as stable unannotated transcripts (SUTs), cryptic unstable transcripts (CUTs), and Nrd1-unterminated transcripts (NUTs) (Neil et al., 2009; Pelechano et al., 2013; Schulz et al., 2013) (Figure 2).

Distribution of degradation factor cross-link sites over the yeast transcriptome.

(A) Fractions of high confidence PAR-CLIP sequencing reads of 30 yeast degradation factors fall into various transcript classes. Depicted classes are the following: messenger RNA (mRNA) in turquoise (n = 4,928), ribosomal RNA (rRNA) in antique pink (n = 24), transfer RNA (tRNA) in dark blue (n = 299), small nucleolar RNA (snoRNA) in yellow (n = 77), small nuclear RNA (snRNA) in green (n = 6), stable unannotated transcripts (SUTs) in red (n = 318), cryptic unstable transcripts (CUTs) in light brown (n = 637), Nrd1-unterminated transcripts (NUTs) in dark brown (n = 298)(Materials and methods). (B) Enrichment z-scores of high confidence PAR-CLIP cross-link sites of 30 yeast degradation factors (rows) in various segments of mRNA transcripts (left columns; UTR: untranslated region; intron; CDS: coding sequence), or other transcript classes as in A (other columns). The color-coded enrichment score shows the column and row normalized enrichment values of binding preferences of each factor for each transcript class (color encoded, depleted in blue and enriched in red). The coefficient of variation on top is the standard deviation divided by the mean for each transcript class.

https://doi.org/10.7554/eLife.47040.006

A first analysis revealed that most PAR-CLIP sequencing reads fall into the mRNA transcript class, although many of the factors also show a considerable number of sequencing reads in ncRNAs, in particular rRNAs (Figure 2A). To obtain a more quantitative comparison, we defined log enrichment scores that reflect the preferences of factors in binding to a specific transcript class in comparison to other factors and classes. To correct for the different sizes of classes and different numbers of measured factor binding sites, we normalized the log enrichment scores by subtracting class- and factor-specific offsets, such that the mean for each class and each factor vanishes (Figure 2B, Materials and methods). This analysis highlights differences between degradation factors with respect to binding to various transcript classes, as will be discussed in detail below.

Distinct factor distribution along mRNA

We next focused on degradation factor distribution on mRNAs. We prepared metagene profiles showing the average occupancy of each factor around the mRNA transcription start sites (TSS) and the poly-adenylation (pA) sites, respectively (Figure 3). The Pan2/Pan3 deadenylase complex and the Ccr4/Not subunits Pop2, Not1, and Caf40 all cross-link upstream of the 3´ end of mRNA with the highest enrichment at the pA site, as expected from their function in shortening the polyA tail. The catalytic subunit Ccr4 binds strongly in the 5´ region of mRNAs. All 5´ decapping factors bind upstream of the pA site, and all but the catalytically active subunit Dcp2 show increasing occupancy towards the 3´ end of mRNAs. These patterns can be explained if decapping factors are pre-bound to mRNAs that form a closed loop that holds the RNA ends in proximity. In contrast, Dcp2 binds almost exclusively at the pA site, suggesting that it might be recruited only upon active mRNA degradation. The cytoplasmic 5´ exonuclease Xrn1 has the highest occupancy towards the 3´ end, similar to the previously published crosslinking and cDNA analysis (CRAC) data (Tuck and Tollervey, 2013), thereby resembling the binding profiles of the decapping factors. Comparison of the binding profiles aligned at the pA site or alternatively with profiles aligned at the translation stop codon shows that the binding preference indeed lies at the end of the 3´ UTR independent of the stop codon position (Figure 3—figure supplement 1B,C).

Metagene analysis of degradation factor binding on mRNAs.

Averaged occupancy profiles of degradation factors over mRNAs aligned around their transcription start site (TSS) (n = 3,193, left) and around their poly-adenylation (pA) site (n = 3,193, right) in a window of [±700 nt]. Regions that have neighboring transcripts on the same strand were removed to avoid contaminating profiles (Materials and methods). Factors are grouped according to their functional role; from top to bottom: deadenylation, decapping, Xrn1, exosome, TRAMP complex, Ski complex, or NMD. The color code shows the average occupancy normalized between the minimum (blue) and maximum (red) values per profile.

https://doi.org/10.7554/eLife.47040.009

The exosome core subunits (Csl4, Rrp40, and Rrp4) and the catalytically active subunits (nuclear: Rrp6, cytoplasmic: Rrp44) cross-link to the 5´ end of the transcript (Figure 3), possibly because the exosome binds to the 5´ end while digesting the 3´ end, or more likely because the exosome slows down towards the remaining 5´ end of mRNAs after rapid degradation from the 3´ end. Both TRAMP complexes bind mainly in the 5´ region of mRNAs near the TSS, as previously observed for Mtr4 and Trf4 (Tuck and Tollervey, 2013).

The Ski complex components Ski7 and Ski8 occupy the entire mRNA with increasing occupancy towards the pA site, whereas Ski2 and Ski3 show more discrete binding towards the polyA tail (Figure 3). The NMD factors Upf1 and Upf3 show binding over the entire mRNA with highest occupancy at the pA site, consistent with their role in scanning for premature stop codons in mRNAs and remodeling of the 3´ end of protein-RNA complexes and completion of mRNA decay (Franks et al., 2010). In addition, Upf2 and Nmd4 show strongest binding near the 3´ ends of mRNAs. Taken together, the distribution of cross-links along mRNA transcripts differs between degradation complexes and in some cases also between their subunits.

Surveillance of aberrant nuclear ncRNA

Pervasive transcription of the genome leads to many short-lived aberrant RNAs that must be rapidly detected and degraded in the nucleus. We previously reported that the RNA surveillance factors Nrd1 and Nab3 strongly cross-link to aberrant upstream antisense RNA that stems from bidirectional transcription (Schulz et al., 2013). In order to find factors cross-linking to aberrant ncRNAs, we plotted the occupancy of all 30 investigated factors on the antisense strand of known mRNAs (Figure 4). For comparison, we plotted the published Nrd1 and Nab3 profiles in the first two lanes of Figure 4. The factors are involved in processing and degradation of Nrd1-unterminated transcripts, or NUTs (Schulz et al., 2013), and are expected to show similar binding to upstream antisense RNA as Nrd1 and Nab3. Indeed, we observed a similar binding pattern for all exosome subunits (Rrp6, Csl4, Rrp40, Rrp4, Rrp44) and subunits of the TRAMP4 complex (Mtr4, Air2, Trf4). Consistent with this, these factors also bind strongly to previously annotated NUTs and CUTs (Figure 2) and show strong enrichment of Nrd1 and Nab3 motifs (GTAG, CTTG) around their cross-link sites (Figure 4—figure supplement 1).

Surveillance of aberrant nuclear antisense RNAs by the exosome and the TRAMP4 complex.

Averaged occupancy profiles of degradation factors binding to transcripts antisense of mRNAs aligned around transcription start site (TSS) (n = 3,076, left) and around their poly-adenylation (pA) site (n = 2,705, right) in a window of [±700 nt]. Regions with annotated genes on the antisense strand are removed to avoid contaminating the profiles (Materials and methods). The color code shows the average occupancy normalized between the minimum (blue) and maximum (yellow) values per profile. On top, previously published PAR-CLIP profiles for Nrd1 and Nab3 are included for comparison (Schulz et al., 2013).

https://doi.org/10.7554/eLife.47040.013

It has been shown that Nrd1 is involved in terminating transcripts upstream of the TSS. We also observe a strong signal for binding upstream of TSS on the sense strand for Air2 and Mtr4 (Figure 3). This suggests that the TRAMP4 complex is involved in degradation of those Nrd1-regulated upstream sense transcripts. To investigate this hypothesis, we compared the binding profiles around the TSS of 459 protein-coding genes, previously annotated as having upstream Nrd1-unterminated transcripts, or NUTs (Schulz et al., 2013), with the profiles obtained for all mRNAs (Figure 3—figure supplement 2A,B). TRAMP4 and the exosome subunits show a strong preference for binding to the upstream promoter region of genes that are controlled by the Nrd1/Nab3 complex. To further confirm that this upstream signal originates from NUTs and CUTs, we excluded cross-link sites that fall within such previously annotated regions. We then compared the binding profiles generated from the remaining binding sites on mRNAs (Figure 3—figure supplement 3). Upon filtering, the signal upstream of the TSS for Air2 and Mtr4 decreases, showing that Nrd1-mediated regulation is the primary cause for this upstream signal. Comparison of the observed antisense profiles (Figure 4) with those obtained after excluding cross-link sites in previously annotated NUT and CUT regions (Figure 4—figure supplement 2) confirms that most of the signal originates from transcripts that are targeted by the Nrd1/Nab3 machinery.

These results are consistent with the idea that the nuclear RNA surveillance machinery involves, in addition to Nrd1 and Nab3, the TRAMP4 complex and the nuclear exosome. Indeed, it was reported that TRAMP4 can add a short polyA tail on aberrant RNAs (Wyers et al., 2005), which may trigger degradation by the nuclear exosome. It was also recently shown that Nrd1 and Trf4 interact, providing a basis for coupling surveillance-mediated termination to RNA degradation (Tudek et al., 2014).

Interactions between RNA processing machineries

To find out which groups of factors can work together in degrading transcripts, we analyzed their tendency to co-occupy the same transcripts by calculating the Pearson correlation of their occupancy across all transcripts (Figure 5A). We also analyzed their co-localization, that is the tendency of a factor to bind near to another factor’s binding sites using a range of ±40 nt from each cross-link site (Figure 5B). To relate these profiles to those of other factors, we included previously published PAR-CLIP profiles from our lab (Supplementary file 1). Profiles were available for factors that function in nuclear RNA surveillance (Nrd1, Nab3), cap binding (Cbc2), mRNA transcript elongation (Bur1, Bur2, Ctk1, Ctk2, Cdc73, Ctr9, Leo1, Paf1, Rtf1, Set1, Set2, Dot1, Spt5, Spt6, Rpb1), pre-mRNA splicing (Ist3, Nam8, Mud1, Snp1, Luc7, Mud2, Msl5), pre-mRNA 3´ processing (Pab1, Pub1, Rna15, Mpe1, Cft2; Yth2), transcription termination (Rat1, Rai1, Rtt103, Pcf11), and mRNA export (Hrp1, Tho2, Gbp2, Hrb1, Mex67, Sub2, Yra1, Nab2, Npl3) (Baejen et al., 2017; Baejen et al., 2014; Battaglia et al., 2017; Schulz et al., 2013).

Global co-occupancy and co-localization analysis reveals unexpected cooperation between factors from different complexes and pathways.

(A) Matrix of pairwise correlation coefficients of factor occupancies evaluated over all transcripts. (B) Matrix of co-localization based on the enrichment of factor x binding within 40 nt of the cross-link site of factor x´ (Materials and methods). (C) Two-dimensional embedding of the co-occupancies in (A) analyzed for 74 RNA processing factors with tSNE, including 30 factors from this study (highlighted in bold), and 44 factors from previous studies (Baejen et al., 2017; Baejen et al., 2014; Battaglia et al., 2017; Schulz et al., 2013) (Supplementary file 1). Factors that are plotted in close proximity show a preference for binding to the same transcripts. Clusters present factors involved in RNA synthesis (1), splicing (2), 3´ processing (3), deadenylation (4), decapping (5), nuclear ncRNA processing (6), and surveillance (7).

https://doi.org/10.7554/eLife.47040.016

Co-occupancy and co-localization plots for all factors can be found in Figure 5—figure supplements 1 and 2, respectively. A two-dimensional embedding of co-occupancy profiles between all these processing factors is shown in Figure 5C. It represents the degree of similarities between co-occupancy of transcripts (Figure 5A) in terms of the distance in two dimensions. The two-dimensional embedding of the co-localization matrix in Figure 5B shows a similar clustering (Figure 5—figure supplement 3). This extensive global analysis suggests which factors reside in functional complexes and which functional complexes may interact during RNA processing and degradation. The analysis recovers several established interactions between subunits of known complexes and between different complexes, providing a positive control. For example, all factors of the decapping complex show very high co-occupancy and co-localization, as do Air2 and Mtr4, which reside in the TRAMP4 complex.

The analysis contains a lot of new information, forcing us to focus here on a few interesting, novel findings (Figure 5C). First, the largest cluster is formed by the previously analyzed factors involved in transcription elongation by RNA polymerase II (cluster 1) and in co-transcriptional pre-mRNA processing, including cap-binding complex (Cbc2), 3´ processing, transcription termination, and RNA export. The degradation factors Ccr4 and Air1 also reside in this cluster, maybe reflecting the role of Ccr4 in transcription elongation (Kruk et al., 2011). A second cluster is formed by splicing factors (cluster 2). Factors involved in nuclear and cytoplasmic exosomal degradation (Rrp6, Csl4, Rrp4, Rrp40 and Rrp44) form a third cluster (cluster 3). Close to cluster 3, we find the TRAMP4 complex subunit Trf4, the elongation factors Dot1, Paf1, Leo1, and the termination factors Pcf11 and Rai1. Rai1 has been shown to detect and remove incomplete 5´ cap structures, to subject aberrant pre-mRNAs to nuclear degradation (Jiao et al., 2010).

A fourth cluster is formed by mRNA deadenylation factors together with polyA tail binding proteins (Pab1 and Pub1), Ski7, Ski8, Trf5, and the export factor Yra1 (cluster 4). This is consistent with coupled mRNA deadenylation and subsequent degradation from its 3´ end by the exosome with the Ski or TRAMP complex as adaptors. The fifth cluster is formed by mRNA decapping factors, which cluster together with Xrn1, suggesting a coupling of mRNA decapping with degradation from the 5´ end by Xrn1 (cluster 5). The NMD-involved factors Upf1, Upf2, Upf3 and Nmd4, and Ski2 and Ski3 are also found in cluster 5. The high correlation between Xrn1 and Ski2 has been reported in a CRAC experiment (Tuck and Tollervey, 2013). The elongation factor Ctr9, the 3´ processing factor Mpe1 and the export factors Tho2, Mex67 and Nab2 are also found in cluster 5. A last cluster (cluster 6) is formed by factors involved in nuclear RNA surveillance, including Air2, Mtr4 and the Nrd1/Nab3 complex. Taken together, these findings are consistent with known functional associations and physical interactions between factors and suggest intriguing new associations to be investigated in future work.

5´ degradation machinery senses translation efficiency

To study the link between cytosolic mRNA translation and degradation, we compared the occupancy of degradation factors on mRNAs to their average codon-optimality score (‘transcript optimality’) (Figure 6A, Figure 6—figure supplements 2–8A). We found that the 5´ decapping machinery and Xrn1 preferentially bind transcripts with low transcript optimality. In contrast, the 3´ deadenylation machinery and the exosome bind more strongly to optimal transcripts. We asked whether this correlation with codon optimality is introduced by only a few differentially bound codons or by global enrichment/depletion of optimal codons. For this purpose, we introduced a ‘codon enrichment score’, which measures a codon’s enrichment in the set of transcripts bound by the factor relative to the yeast mRNA pool. For Dcp2 this enrichment score is high on non-optimal codons, and low on optimal codons, whereas the opposite trend is observed for Ccr4 and most degradation factors (Figure 6B, Figure 6—figure supplement 1–7) This is consistent with a model that ribosome stalling on translationally inefficient codons can lead to recruitment of Dcp2 and Xrn1 and subsequent 5´ degradation of the transcript (Heck and Wilusz, 2018).

Binding preferences reveal a link between decapping-mediated degradation and translation.

(A) Total occupancy per mRNA (according to TIF-seq annotation) for six factors as a function of the average mRNA codon optimality (transcript optimality). The occupancy of factors from the 5´→3´ degradation machinery (decapping and Xrn1, left) decreases with increasing transcript optimality, whereas the occupancy of factors from the 3´→5´ degradation machinery (Ccr4 and Caf40 deadenylation complex subunits and exosome subunit Rrp44, right) increases with increasing average codon optimality. (Gray shading: 95% confidence intervals generated by bootstrapping mRNAs). (B) Codon enrichment in transcripts bound by Dcp2 and Ccr4 compared to the average frequency over all mRNAs. The bar colors represent codon optimality, with highly optimal codons shown in dark red. (Thin gray lines: 90% confidence intervals generated by bootstrapping coding sequences.) (C) Significance of correlations between the binding strength of degradation factors and transcript length, transcript optimality (Pechmann and Frydman, 2013), expression level (Baejen et al., 2017), and half-life derived by multivariate linear regression analysis (Materials and methods). Bars are separated according to the direction of correlation with positive correlation marked by a red background and negative correlation marked by a blue background.

https://doi.org/10.7554/eLife.47040.020

To investigate the significance of the correlation between transcript optimality and binding of the 5´ degradation machinery, we compared the contribution of several mRNA features in explaining the occupancy patterns retrieved from PAR-CLIP experiments. Since mRNA expression, half-life, and translation optimality are inter-correlated (Figure 6—figure supplement 8), a causative effect of one of these features on binding strength may lead to correlations with all three features. To better distinguish correlation from causation, we used linear regression analysis to explore whether correlations between factor binding and optimality are better explained with other mRNA features (Figure 6—figure supplement 9). We assessed the significance of features via the likelihood ratio test on the multi-variate linear regression model for occupancy. The likelihood ratio test calculates the significance of a feature from the change of the likelihood (quantifying the prediction quality) upon removal of that feature from the regression model. For decapping enhancers (Edc2, Edc3, and Dhh2) and Xrn1, low codon optimality is the most determining feature for binding (Figure 6C). The same is true for NMD factors Upf1 and Upf3, which are known to bind non-optimal transcripts (Celik et al., 2017). This result confirms the importance of the translation efficiency for the stability of cytosolic mRNAs and strengthens our finding that transcripts with low average codon optimality are preferentially targeted by the decapping machinery and degraded from the 5´ end.

Decapping factors are enriched upon RNA degradation

Although decapping occurs at the 5´ end of mRNAs, decapping factors show a strong occupancy near the 3´ end (Figure 3). To investigate this further, we compared metagene profiles of decapping factors between stable (top 25%) and unstable (bottom 25%) transcripts, using mRNA half-life estimates (Figure 7A, Materials and methods). On both stable and unstable mRNAs, Dcp1, Edc2, Edc3, and Dhh2 show increased binding near the 3´ end, but unstable RNAs show a higher occupancy in the transcript body. The catalytically active subunit Dcp2 binds almost exclusively at the 3´ end and has a higher occupancy on unstable transcripts. Moreover, A-rich 4-mers are abundant around the proximity (eight nt) of Dcp2-cross-link sites (Figure 7C), indicating a binding preference of Dcp2 for A-rich RNA sequences. Overall, these binding patterns suggest that decapping factors are bound in transcript bodies and near the 3´ end of transcripts, and that through closed-loop formation of the mRNA they are in close proximity to the 5´ end. Decapping factors might also travel with the 5´→3´ exonuclease Xrn1 upon RNA degradation.

Location and recruitment of the decapping complex Dcp1/Dcp2 and decapping enhancers Edc3, Dhh2, and Edc2.

(A) Smoothed, transcript-averaged PAR-CLIP occupancy profiles aligned at TSS and pA sites [±750 nt] of unstable and stable transcripts (first and fourth quantile of half-life distribution, respectively). (B) Dependence of total occupancy of factors on the transcripts half-life. The fitting function is plotted in red and the fitted value for b is marked with a dashed gray line. (Gray shade: 95% confidence intervals generated by bootstrapping transcripts). (C) Sequence binding preference for the catalytically active subunit of decapping complex (Dcp2), illustrated with the five most enriched and the 3 most depleted 4-mers. The color code shows the log2 enrichment factor of 4-mers around PAR-CLIP cross-link sites [±8 nt]. Dark red represents strong enrichment and dark blue shows strong depletion of a 4-mer. Infeasible combinations are shown with gray. The most highly enriched field is binding AAAAU with the cross-link at the U, which is enriched over random expectation approximately 22.3 = 5-fold.

https://doi.org/10.7554/eLife.47040.030

Decapping factors may bind to complete mRNAs or to transcripts that are in the process of being degraded. To quantify these two behaviors, we combined our PAR-CLIP occupancy data with RNA half-life estimates (Materials and methods). We modeled the occupancy of factors on mRNA as the sum of binding to all transcripts (b) and surplus binding to transcripts that are in the process of degradation (at1/2). Therefore, we can model occupancy as a function of half-life with a linear equation (occupancy=at 1/2+b). In cases where there is no surplus binding upon active degradation, that is the occupancy is the same as in intact RNAs, ‘a’ will be zero. For 5´ decapping factors, this model closely fits the occupancy patterns retrieved from our experiments (Figure 7B), other degradation factors also follow this pattern to varying degrees (Figure 6—figure supplements 1–7). In particular, Dcp2 shows a very high a/b ratio, revealing that it cross-links preferentially to transcripts that are being degraded. This analysis strongly suggests that the 5´ decapping machinery, although present to some extent on complete mRNAs, is enriched when mRNAs are degraded.

Discussion

Here we report transcriptome-wide binding maps for 30 RNA degradation factors in yeast. A detailed bioinformatics analysis of these maps revealed how degradation factors vary in their binding specificities for different classes of RNAs and with respect to their preferred locations on RNA transcripts. Global comparisons of the profiles alongside previously published profiles of other RNA-binding factors revealed clusters of factors that co-occupy RNAs or co-localize on RNAs. Our data are consistent with a large body of published results and extend these to a global scale. In addition, the results revealed several unexpected, novel findings, which we discuss below. Although our data reflect factor cross-linking signal and measure occupancy on transcripts, and do not directly reveal function, the correlations of occupancies between factors and with transcript properties indicate functional aspects and suggest functional associations between factors that may guide future studies.

With respect to canonical mRNA turnover in the cytoplasm, the Pan2/Pan3 deadenylation complex and subunits of the Ccr4/Not complex bound preferentially to the pA site, reflecting a function in polyA tail shortening at early stages of RNA degradation. In contrast, the Ccr4/Not complex subunit Ccr4 bound also strongly in the 5´ region of transcripts. This pattern may reflect functional differences between Ccr4 and Pop2 during deadenylation (Webster et al., 2018) or an additional function of Ccr4 in transcription elongation (Kruk et al., 2011). Ccr4/Not subunits also show variations in their RNA-binding specificities, suggesting several isoforms of the complex that vary in composition and function, or RNA-specific conformational rearrangements. Factors involved in decapping show higher cross-linking near the RNA 3´ end, consistent with previously proposed binding near the pA site (Chowdhury et al., 2007). The profiles of decapping factors resemble those of Xrn1, suggesting formation of a complex with this 5´→3´ exonuclease or a fast mRNA decay by Xrn1 from 5´→3´ and slowing down towards the 3´ end. Complex formation of Xrn1 and the decapping factors is consistent with a function of Xrn1 in the buffering of mRNA levels in cells (Sun et al., 2013), which may be explained if Xrn1 is a regulatory component of the decapping complex.

The preferred localization of 5´ decapping and 3´ deadenylation factors near the mRNA 3´ and 5´ end, respectively, seems counterintuitive, but may be explained by formation of an mRNA closed-loop structure by messenger ribonucleoproteins (mRNPs), where 5´ and 3´ ends are in proximity (Gallie, 1991). It is possible that mature mRNPs carry decapping factors near their 3´ end, and that upon polyA tail shorting the decapping complex is activated, leading to decapping and rapid RNA degradation from the 5´ end. In this model, decapping would open the RNA closed-loop structure, providing access for exonucleases and allowing for rapid RNA removal. Further, our result that decapping factors are enriched on translationally non-optimal codons agrees with previous findings that suggest a link between translation and RNA degradation from the 5´ end through the decapping enhancer Dhh2 (Radhakrishnan et al., 2016). We note however that our approach does not detect binding events within the polyA tail, limiting further insights.

Comparison of our data with previous profiles of the nuclear surveillance factors Nrd1 and Nab3 reveals many factors that show a similar binding to aberrant non-coding nuclear RNAs, in particular antisense RNA upstream of known promoters, suggesting that these factors are part of the nuclear surveillance machinery. These results are consistent with published data (Schmid and Jensen, 2018) and with the following model for nuclear surveillance. First, the Nrd1/Nab3 complex recognizes aberrant, antisense RNAs (Schulz et al., 2013). These RNAs then get adenylated by the TRAMP4 complex, consistent with an interaction of the Nrd1/Nab3 complex and the TRAMP subunit Trf4 (Tudek et al., 2014). The short polyA tail targets the RNA for degradation by the nuclear exosome. We note that the degradation of introns and ncRNAs upstream of mRNAs on the same strand, which were annotated as NUTs and CUTs, is likely to use the same degradation mechanism because these are bound by the same factors (Figure 2, Figure 3—figure supplement 2). Thus, our results are consistent with the idea that degradation of short-lived ncRNAs in the nucleus involves Nrd1/Nab3, TRAMP4, and the exosome.

The exosome is involved in 3´→5´ cytoplasmic mRNA degradation, and in the processing and degradation of long-lived transcripts such as tRNAs, rRNAs and sn(o)RNAs. The differences in exosome co-factor binding to different RNA classes (Figure 2) support the hypothesis that these factors confer specificity for processing and degradation of various RNA species (Delan-Forino et al., 2017). The exosome co-factor Ski2 shows cross-linking towards the 3´ end of mRNAs (Figure 3). This indicates that the subunit is important for initial RNA degradation by using its helicase activity to dissolve secondary structures and allowing the exosome to start degradation of the transcript (Schneider and Tollervey, 2013). Like the exosome complex, the exosome co-factor TRAMP5 binds mRNAs towards the 5´ end of transcripts, and thus some mRNAs may be targeted by TRAMP5 for exosomal degradation. The catalytic exosome subunits Rrp44 and Rrp6 and the exosome core cross-link near the mRNA 5´ end, probably because the exosome moves rapidly from 3´→5´ and then slows down, causing extensive cross-linking. In summary, our resource of transcriptome-binding profiles for 30 RNA degradation factors reveals several hypotheses for the function of these factors that can be tested case by case in the future.

Materials and methods

Data availability

Sequencing data have been deposited in GEO under accession codes GSE 128312.

The following data sets were generated

The following previously published data sets were used

References

Decision letter

[Editors’ note: a previous version of this study was rejected after peer review, but the authors submitted for reconsideration. The first decision letter after peer review is shown below.]

Thank you for submitting your work entitled "Transcriptome maps of general eukaryotic RNA degradation factors" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by a Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: David Tollervey (Reviewer #2).

Our decision has been reached after consultation between the reviewers. All reviewers acknowledge the extensive data sets obtained, the quality of the analysis and the potential value for the RNA turnover field. There are, however, concerns about the reproducibility of the data and the significance of the biological insights, given the absence of any independent validation of the results. We are therefore obliged to reject this paper for publication in eLife. However, new submission of a revised manuscript would be welcome if all the reviewers' concerns can be addressed.

We hope that the individual reviews appended below will be helpful in deciding how to move forward.

Reviewer #1:

In this manuscript the authors describe the genome-wide binding sites for 30 different proteins acting in multiple RNA degradation pathways. The data will be a valuable resource for researchers in this area and should be published. However, there are a numerous places were the manuscript needs to be improved by relatively minor rewrites such that the text more accurately reflects the data. Some of the issues are:

1) The description of the reliability and quality of the data in the "Transcriptome maps for 30 RNA degradation factors" section of the Results is not completely congruent with the results.

1a) "The high reproducibility of these PAR-CLIP experiments is revealed by a comparison of biological replicates that we collected for eleven out of the 30 degradation factors (Figure 1—figure supplement 1)." The supplementary figure actually shows high reproducibility for 4 of the proteins (all cytoplasmic decapping factors), medium reproducibility for 3 factors and modest reproducibility for 4 factors.

1b) 19 of the factors were not analyzed in duplicate and thus the reproducibility of these datasets is hard to evaluate.

1c) Furthermore, the Materials and methods indicate that "Reads from PAR-CLIP experiments with replicates were merged", thus it is not clear what part of the analyses were actually preformed in duplicate.

2) At least seven of the factors studied here have been previously studied by the Tollervey lab (in what they call CRAC, which is fundamentally the same as PAR-CLIP). Thus, a valuable extension of this study and the validity of the data would be to compare reproducibility between the author's PAR CLIP data and the published CRAC data.

3) "we typically obtained tens of thousands of verified factor-RNA cross-link sites (Table 1)." Table 1 actually shows that about half of the factors meet this tens of thousands threshold while the remaining factors vary between 561 and 16663 sites.

4) In the description of "RNA end-processing complexes differ in their targets" section the highlighted results that cytoplasmic factors like Caf40 and Ccr4 bind sparsely to nuclear RNAs such as introns, snRNAs and snoRNAs is not surprising. The authors attribute this to a "strong binding preference", but this may simply reflect that these factors are not exposed to these RNAs instead of binding constants.

More importantly, the z-scoring in Figure 1B appears to be inappropriate for these kinds of statements. If I understand the description of the z-scoring correctly the unexpected high z-score of Ccr4 for tRNAs in Figure 1B means that tRNAs have a strong preference for Ccr4 compared to the other 29 factors, not that Ccr4 has a strong preference for tRNAs compared to all other RNAs as implied in the Results section. Another example of problematic description of Figure 1B is "Rrp6.…. shows binding to rRNAs, tRNAs, snoRNAs and snRNAs." However Figure 1A does not appear to show appreciable binding to rRNAs.

5) "Dcp2, which harbors the hydrolase activity that removes the cap, cross-links frequently to mRNAs, in particular their 3 ́UTR, and to short noncoding transcripts, in especially to unstable SUTs. This suggests that Dcp2 is involved not only in cytosolic mRNA degradation but also in decapping during nuclear RNA surveillance." However, the defining characteristic that differentiates SUTs from CUTs is that SUTs are degraded by cytoplasmic Dcp2-dependent pathways as shown by several different labs (e.g. Pubmed IDs 17074811, 21610164, 21826286, 24931603). Thus rather than suggesting a new role for Dcp2, these findings confirm a different known role.

6) "The analysis recovers established interactions between subunits of known complexes" but this is not supported by any quantitative analysis. Among the factors the authors analyze, some proteins that are thought to form stable complexes indeed cluster (e.g. Dcp1/2, Pan2/3) while other known stable complexes do not (e.g. Ski2/3/8 or the Air1 subunit of TRAMP).

7) Some of the names of the clusters in Figure 4C are poorly justified. For example, out of the six proteins in the "nuclear ncRNA processing" cluster four are cytoplasmic proteins that act mostly or exclusively on mRNA (Ccr4, Upf1, Ski3 and Ski7). They also function in degradation, rather than processing, thus at least the first two but possibly all three words in this name should be changed.

8) "For decapping factors (Dcp1, Edc3, Edc2, and Dhh2) and Xrn1, low codon optimality is the most determining feature for binding (Figure 5C).” However Figure 5C appears to show that this is true for three of these factors, but the opposite is true for Edc2, where high codon optimality appears to be determining. The authors should clarify.

9) "Moreover, A-rich 4-mers are abundant around the proximity ( ± 8 nt) of Dcp2-cross-link sites (Figure 6C)." However Figure 6C and the legend say the window is ± 5 instead of ± 8. This might be a typo.

10) "As expected, most PAR-CLIP cross-link sites fall into the mRNA transcript class, although about one third of the factors also show a considerable number of cross-links to non-coding RNAs". I suggest deleting "as expected" or replacing it with "surprisingly". Several of the factors have known roles in ncRNA processing/degradation and mRNA is only a minor fraction of cellular RNA and therefore I would not have expected most sites to fall into the mRNA class.

A related issue may be that the analysis in Figure 1 appears to only include "Fractions of high-confidence PAR-CLIP cross-link sites" and not the actual number of or fraction of high-confidence reads. e.g. if there are 9 sites, in mRNA with 100 reads each and one in a ncRNA with 900 reads this would count as 90% of the sites being in mRNA, but only 50% of the reads are in mRNA. Since the complexity of mRNA is probably the highest (more than half the genome), the number of distinct sites in mRNA for a given factor is probably a systematic overrepresentation of that factor's importance in mRNA processing.

11) In Materials and methods "various non-overlapping transcript classes c" needs to be clarified. Several of the transcript classes in fact do overlap in the genome. Many snoRNAs are encoded within introns. Many CUTs, SUTs and NUTs overlap with mRNAs and each other. The methods need to be clarified.

Reviewer #2:

The authors present the PAR-CLIP analysis of 30 proteins implicated in RNA turnover or surveillance. In principal, these results should be of considerable value to the field, but this is much reduced by the lack of biological or even technical replicates for most of the experiments. Moreover, there are some concerns about the mapping of the data to the yeast genome. The downstream bioinformatic analysis of the data appears to have been very well performed, with imaginative approaches being applied and interesting conclusions. The 2D clustering analysis (Figure 4C and Figure 4—figure supplement 3) is particularly good.

I regret that the technical problems with the manuscript prevent me from recommending acceptance of the current version.

Specific points:

1) Subsection “Transcriptome maps for 30 RNA degradation factors”: It appears that biological replicates were performed for only eleven out of the 30 degradation factors tested. For 19 of the factors analyzed, the conclusions are seemingly based on single experiments. While it is the case that the experiments reported involve a lot of work, it is quite unusual for the bulk of data in a manuscript to be published without clearer evidence of reproducibility. It would have been better to analyze fewer proteins in more depth. The authors present data indicating the consistency of the results from the proteins for which replicates were performed. However, while the replicates appear to cluster well in the 2D diagrams, inspection of the metagene analysis reveals appreciable differences between duplicates (Figure 1—figure supplement 1), which would presumably be much more marked at the level of individual genes.

This is particularly an issue when the results are not as expected, e.g. the apparent binding of the deadenylase Ccr4 to a high proportion of tRNAs, which is based only a total of only 4,540 confident Ccr4 cross-linking sites.

2) The authors report that "most PAR-CLIP cross-link sites fall into the mRNA transcript class". This is unexpected, particularly for pre-rRNA processing factors, including the exosome and Mtr4, that should recover high level of rRNA hits. Inspection of the methods indicates that the reads were mapped to a unique database. This would have the effect of eliminating most rRNA and tRNA reads, as well as a subset of mRNAs. In the yeast genomic database, there are 2 copies of the rDNA repeat, so unique mapping will eliminate most rRNA hits. For tRNAs the situation is more complex; most tRNAs are present in multigene families and will be eliminated from unique hits, whereas the few single copy tRNA genes will be retained. This might be a particular issue for the reported Ccr4-tRNA interactions. For mRNAs, duplicated genes such as R-proteins will also be excluded.

This point should be clarified. If uniquely-mapped reads were indeed used, the full mapping data should be presented.

3) Subsection “RNA end-processing complexes differ in their targets”: SUTs can be cytosolic, so this might not be sufficient to show that Dcp2 has nuclear activity.

4) Subsection “Distinct factor distribution along mRNA”, Figure 1B (and end of subsection “PAR-CLIP experiments of S. cerevisiae proteins”): It would be helpful to discuss more why Csl4, Rrp4 and Rrp40 have distinct profiles. In particular, the Rrp4 profiles looks very different from all other degradation factors. This raises question of the PAR-CLIP experiment really worked for each cofactor. It is notable that some of these have low numbers of XL sites (561 for Rrp6, 923 for Rrp40, 7,373 for Csl4, while Rrp44 has much more 44,721).

5) Subsection “Interactions between RNA processing machineries”, Figure 4 and Figure 4—figure supplements 1-2: It would be helpful to discuss why Trf4/Air2 is bound to introns (Figure 1) but not Trf5 while Trf5/Air1 is showing higher co-occupancy / colocalization with splicing factors

Reviewer #3:

The paper by Sohrabi-Jahromi et al. presents genome-wide RNA binding datasets for S. cerevisiae RNA decay factors. Analyses of the obtained data are often in-line with current knowledge but also uncover several unexpected binding patterns.

The presented data are certainly of interest to a wide audience. However, there is no independent validation of any of the unexpected binding patterns. This is especially problematic, given that many datasets are based on single biological replicates. Moreover, some observations challenge our current thinking about the function of the investigated factor(s), which, to be suitable for publication in eLife, should be supported by independent experiments.

- Functional relevance:

The manuscript presents a series of unexpected and somewhat controversial observations. One example concerns a possible role for Pop2 and Not1 in decay of CUTs. Any functional relevance of such binding should be trivial to validate. This would be relevant for other of the most striking findings. Such validation may be possible by re-analysis of published datasets, for example by intersecting occupancy with RNAseq upon depletion/mutation of the same factors.

- Replicates and Technical artefacts:

PAR-CLIP is a complex technique, where technical artefacts are not uncommon. For example, binding to highly abundant tRNA, snRNA and snoRNA species is often observed, even for factors where such binding is unlikely to occur or be relevant in vivo. Hence, it is quite possible that binding to such species, especially when based on a single biological replicate, does not reflect biology. Such considerations should be mentioned in the manuscript. Ideally, all factors for which single replicates are presented, should be backed up by additional data. This may be simply an additional replicate, or some kind of independent validation.

More generally, more information on how the authors scrutinize the quality of their PAR-CLIP experiments should be given (i.e. more explicitly than simply referring to another publication).

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for resubmitting your work entitled "Transcriptome maps of general eukaryotic RNA degradation factors" for further consideration at eLife. Your revised article has been favorably evaluated by James Manley as the Senior Editor, and three reviewers, one of whom is a member of our Board of Reviewing Editors.

The manuscript has been considerably improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

1) Even though some of the more controversial suggestions of the original draft are now resolved, comparison with available knock-out or mutant / RNA-seq datasets would greatly help to support, or dismiss, some of the mentioned novel findings.

2) If available, western blots and/or autoradiographs of the different PAR-CLIP experiments would be nice to include to allow readers to judge the quality of different IP experiments.

3) Some of the conclusions presented in the Results sections still appear vague or unclear. This concerns for example interpretation of the RNA exosome binding profiles.

a) i.e. In the first paragraph of the subsection “The exosome and surveillance factors”: unclear what different binding patterns "reflect" the distinct functions of the RNA exosome.

b) In the second paragraph of the subsection “Distinct factor distribution along mRNA”: How/why would TRAMP4 binding upstream of the TSS affect early termination of the mRNA? Isn't it more plausible that it serves a role in termination of (unannotated) upstream ncRNA? Similarly, 5' bias of exosome and TRAMP subunits could arise from unannotated NUT-like transcripts.

4) Figure 3 and Figure 3—figure supplement 1B: What RNAs generate the high signals for Mtr4 and Air2 5' to the TSS? This region is expected to give rise to very few transcripts. The upstream ncRNAs are further upstream and on the opposite strand. There must be a concern that some sequences have been mis-mapped. Were the sequences mapped only to a small number of loci, which would strengthen this concern? The authors should re-check the data supporting these distributions.

5) The second paragraph of the subsection “The exosome and surveillance factors” refers to Figure 2—figure supplement 3 and not Figure 3—figure supplement 3.

6) Figure 5C: 3' degaradation => 3' degradation

https://doi.org/10.7554/eLife.47040.044

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

Reviewer #1:

1) The description of the reliability and quality of the data in the "Transcriptome maps for 30 RNA degradation factors" section of the Results is not completely congruent with the results.

1a) "The high reproducibility of these PAR-CLIP experiments is revealed by a comparison of biological replicates that we collected for eleven out of the 30 degradation factors (Figure 1—figure supplement 1)." The supplementary figure actually shows high reproducibility for 4 of the proteins (all cytoplasmic decapping factors), medium reproducibility for 3 factors and modest reproducibility for 4 factors.

1b) 19 of the factors were not analyzed in duplicate and thus the reproducibility of these datasets is hard to evaluate.

1c) Furthermore, the Materials and methods indicate that "Reads from PAR-CLIP experiments with replicates were merged", thus it is not clear what part of the analyses were actually preformed in duplicate.

Thank you for this important comment, which we took very seriously. To ensure the highest possible data quality and reproducibility, we generated 49 new datasets, obtaining between two and four highly reproducible biological replicates for each degradation factor (Figure 1—figure supplement 1). Upon comparing the replicates, we found poor reproducibility for Pop2, Not1, Rrp4 and Caf40 and therefore omitted old replicates from the analysis. Furthermore, we applied a more conservative threshold for including data and eliminated six of the previous replicates that had less than 10.000 cross-link sites (Csl4, Rrp40, Rrp6, Ski2, Ski3, and Ski7). As a result, only highly reproducible data with a high number of crosslink sites were included in the final data set. This has led to a clearly visible improvement of the analysis and revised manuscript.

We now write:

“The high reproducibility of these PAR-CLIP experiments is revealed by a comparison of two independent biological replicates that we collected for all 30 degradation factors (Figure 1—figure supplement 1), with Spearman correlations between 0.87 and 1.00 (mean: 0.94).”

The extent of improvement of the PAR-CLIP data can be appreciated by comparing the old Figure 2 with new Figure 3.

2) At least seven of the factors studied here have been previously studied by the Tollervey lab (in what they call CRAC, which is fundamentally the same as PAR-CLIP). Thus, a valuable extension of this study and the validity of the data would be to compare reproducibility between the author's PAR CLIP data and the published CRAC data.

We have included the comparison between metagene profiles obtained from CRAC data and PAR-CLIP in Figure 1—figure supplement 1. The agreement between these studies is indeed good.

3) "we typically obtained tens of thousands of verified factor-RNA cross-link sites (Table 1)." Table 1 actually shows that about half of the factors meet this tens of thousands threshold while the remaining factors vary between 561 and 16663 sites.

Compare comment above. We have generated new biological replicates for most of the PAR-CLIP experiments. Our new datasets contain between 21,311 and 590,328 cross-link sites per factor (Figure 1).

4) In the description of "RNA end-processing complexes differ in their targets" section the highlighted results that cytoplasmic factors like Caf40 and Ccr4 bind sparsely to nuclear RNAs such as introns, snRNAs and snoRNAs is not surprising. The authors attribute this to a "strong binding preference", but this may simply reflect that these factors are not exposed to these RNAs instead of binding constants.

More importantly, the z-scoring in Figure 1B appears to be inappropriate for these kinds of statements. If I understand the description of the z-scoring correctly the unexpected high z-score of Ccr4 for tRNAs in Figure 1B means that tRNAs have a strong preference for Ccr4 compared to the other 29 factors, not that Ccr4 has a strong preference for tRNAs compared to all other RNAs as implied in the Results section. Another example of problematic description of Figure 1B is "Rrp6.…. shows binding to rRNAs, tRNAs, snoRNAs and snRNAs." However Figure 1A does not appear to show appreciable binding to rRNAs.

Thank you very much for this remark, which prompted us to come up with a better way to analyse the enrichment of factor binding among transcript classes. The new score now normalizes both by factor and by transcript class and is simply interpretable as a log enrichment score:

“For each factor, we counted the number of high confidence reads falling in each transcript class. […] The normalization can be interpreted as subtracting from the log enrichment matrix X’ the first singular component of its singular-value decomposition.”

Moreover, we have used non-unique mapping for this analysis to include duplicated genomic regions like rRNAs and tRNA isodecoders. The new procedure and its consequences for the analysis are described in our response to point 2 of reviewer 2.

5) "Dcp2, which harbors the hydrolase activity that removes the cap, cross-links frequently to mRNAs, in particular their 3 ́UTR, and to short noncoding transcripts, in especially to unstable SUTs. This suggests that Dcp2 is involved not only in cytosolic mRNA degradation but also in decapping during nuclear RNA surveillance." However, the defining characteristic that differentiates SUTs from CUTs is that SUTs are degraded by cytoplasmic Dcp2-dependent pathways as shown by several different labs (e.g. Pubmed IDs 17074811, 21610164, 21826286, 24931603). Thus rather than suggesting a new role for Dcp2, these findings confirm a different known role.

Thanks for pointing this out. We have revised the text accordingly:

“Whereas they [all decapping-related factors] show slight enrichment in binding to mRNAs, in particular to CDS and 3´ UTR, they preferentially bind to SUTs. This is consistent with previous findings that SUTs get degraded via Dcp2-dependent pathways in the cytoplasm (Marquardt et al., 2011; Smith et al., 2014; Thompson and Parker, 2007).”

6) "The analysis recovers established interactions between subunits of known complexes" but this is not supported by any quantitative analysis. Among the factors the authors analyze, some proteins that are thought to form stable complexes indeed cluster (e.g. Dcp1/2, Pan2/3) while other known stable complexes do not (e.g. Ski2/3/8 or the Air1 subunit of TRAMP).

The new replicate data has led to a generally better agreement between known protein complexes and clusters in our co-occupancy map (Figure 5C), such as the decapping and deadenylation complexes. We have updated the Results part to reflect this:

“The analysis recovers several established interactions between subunits of known complexes and between different complexes, providing a positive control. For example, all factors of the decapping complex show very high co-occupancy and co-localization, as do Air2 and Mtr4, which reside in the TRAMP4 complex.”

However, known complexes do not always correspond to clusters, as highlighted in the Discussion section:

“Although our data reflect factor cross-linking signal and measure occupancy on transcripts, and do not directly reveal function, the correlations of occupancies between factors and with transcript properties indicate functional aspects and suggest functional associations between factors that may guide future studies.”

7) Some of the names of the clusters in Figure 4C are poorly justified. For example, out of the six proteins in the "nuclear ncRNA processing" cluster four are cytoplasmic proteins that act mostly or exclusively on mRNA (Ccr4, Upf1, Ski3 and Ski7). They also function in degradation, rather than processing, thus at least the first two but possibly all three words in this name should be changed.

With the new datasets, clusters have gotten more in line with prior expectations. We have also tried to rename the clusters to reflect the biological functions more clearly (Figure 5C). However, we still see some factors amongst clusters that do not reflect the general function of the cluster. These are important findings that can be followed up by the community in the future. We highlighted and related unexpected findings in the text, e.g.:

“Factors involved in nuclear and cytoplasmic exosomal degradation (Rrp6, Csl4, Rrp4, Rrp40 and Rrp44) form a third cluster (cluster 3). […] Rai1 has been shown to detect and remove incomplete 5´ cap structures, to subject aberrant pre-mRNAs to nuclear degradation (Jiao et al., 2010).”

8) "For decapping factors (Dcp1, Edc3, Edc2, and Dhh2) and Xrn1, low codon optimality is the most determining feature for binding (Figure 5C).” However Figure 5C appears to show that this is true for three of these factors, but the opposite is true for Edc2, where high codon optimality appears to be determining. The authors should clarify.

Thank you for pointing this out. We generated two new replicates for Edc2 and those now show the same trends as other decapping factors with low codon optimality as the most determining feature for RNA binding (Figure 6—figure supplement 3). We do not know the reasons why Edc2 was an outlier in the previous data.

9) "Moreover, A-rich 4-mers are abundant around the proximity ( ± 8 nt) of Dcp2-cross-link sites (Figure 6C)." However Figure 6C and the legend say the window is ± 5 instead of ± 8. This might be a typo.

The correct range is ± 8. This typo was fixed in the manuscript.

10) "As expected, most PAR-CLIP cross-link sites fall into the mRNA transcript class, although about one third of the factors also show a considerable number of cross-links to non-coding RNAs". I suggest deleting "as expected" or replacing it with "surprisingly". Several of the factors have known roles in ncRNA processing/degradation and mRNA is only a minor fraction of cellular RNA and therefore I would not have expected most sites to fall into the mRNA class.

A related issue may be that the analysis in Figure 1 appears to only include "Fractions of high-confidence PAR-CLIP cross-link sites" and not the actual number of or fraction of high-confidence reads. e.g. if there are 9 sites, in mRNA with 100 reads each and one in a ncRNA with 900 reads this would count as 90% of the sites being in mRNA, but only 50% of the reads are in mRNA. Since the complexity of mRNA is probably the highest (more than half the genome), the number of distinct sites in mRNA for a given factor is probably a systematic overrepresentation of that factor's importance in mRNA processing.

Thank you for this important comment. We included read distribution over annotated regions in Figure 2A and noted the change in the text:

“A first analysis revealed that most PAR-CLIP sequencing reads fall into the mRNA transcript class, although many of the factors also show a considerable number of sequencing reads in non-coding RNAs, in particular rRNA (Figure 2A).”

11) In Materials and methods "various non-overlapping transcript classes c" needs to be clarified. Several of the transcript classes in fact do overlap in the genome. Many snoRNAs are encoded within introns. Many CUTs, SUTs and NUTs overlap with mRNAs and each other. The methods need to be clarified.

This is indeed correct. We clarified how we retrieve our non-overlapping set in the Materials and methods section:

“We removed overlapping annotations with the following priority list: rRNA, tRNA, snRNA, snoRNA, intron, CDs, UTR, SUT, CUT, NUT.”

Reviewer #2:

Specific points:

1) Subsection “Transcriptome maps for 30 RNA degradation factors”: It appears that biological replicates were performed for only eleven out of the 30 degradation factors tested. For 19 of the factors analyzed, the conclusions are seemingly based on single experiments. While it is the case that the experiments reported involve a lot of work, it is quite unusual for the bulk of data in a manuscript to be published without clearer evidence of reproducibility. It would have been better to analyze fewer proteins in more depth. The authors present data indicating the consistency of the results from the proteins for which replicates were performed. However, while the replicates appear to cluster well in the 2D diagrams, inspection of the metagene analysis reveals appreciable differences between duplicates (Figure 1—figure supplement 1), which would presumably be much more marked at the level of individual genes.

This is particularly an issue when the results are not as expected, e.g. the apparent binding of the deadenylase Ccr4 to a high proportion of tRNAs, which is based only a total of only 4,540 confident Ccr4 cross-linking sites.

Thank you very much for this important comment. We took this very seriously, collected many new datasets, and applied a more conservative threshold to eliminate weak old datasets. We are certain that only high-quality, highly reproducible data are included in the manuscript. Here is the reply to the same issue, raised by reviewer 1, point 1:

To ensure the highest possible data quality and reproducibility, we generated 49 new datasets, obtaining between two and four highly reproducible biological replicates for each degradation factor (Figure 1—figure supplement 1). Upon comparing the replicates, we found poor reproducibility for Pop2, Not1, Rrp4 and Caf40 and therefore omitted old replicates from the analysis. Furthermore, we applied a more conservative threshold for including data and eliminated six of the previous replicates that had less than 10.000 cross-link sites (Csl4, Rrp40, Rrp6, Ski2, Ski3, and Ski7). As a result, only highly reproducible data with a high number of crosslink sites were included in the final data set. This has led to a clearly visible improvement of the analysis and revised manuscript.

We now write:

“The high reproducibility of these PAR-CLIP experiments is revealed by a comparison of two independent biological replicates that we collected for all 30 degradation factors (Figure 1—figure supplement 1), with Spearman correlations between 0.87 and 1.00 (mean: 0.94).”

The extent of improvement of the PAR-CLIP data can be appreciated by comparing the old Figure 2 with new Figure 3.

2) The authors report that "most PAR-CLIP cross-link sites fall into the mRNA transcript class". This is unexpected, particularly for pre-rRNA processing factors, including the exosome and Mtr4, that should recover high level of rRNA hits. Inspection of the methods indicates that the reads were mapped to a unique database. This would have the effect of eliminating most rRNA and tRNA reads, as well as a subset of mRNAs. In the yeast genomic database, there are 2 copies of the rDNA repeat, so unique mapping will eliminate most rRNA hits. For tRNAs the situation is more complex; most tRNAs are present in multigene families and will be eliminated from unique hits, whereas the few single copy tRNA genes will be retained. This might be a particular issue for the reported Ccr4-tRNA interactions. For mRNAs, duplicated genes such as R-proteins will also be excluded.

This point should be clarified. If uniquely-mapped reads were indeed used, the full mapping data should be presented.

Again, many thanks for this important observation. Unique mapping indeed contributed to low rRNA reads. However, since non-unique mapping negatively impacts the quality of cross-link sites in mRNA regions, we used two mapping strategies in the manuscript for Figure 2 and the rest of presented data (Materials and methods):

“We used two alternative approaches for mapping reads using Bowtie (Langmead et al., 2009): For all analyses except the ‘transcript class enrichment analysis’ in Figure 2, reads are uniquely mapped with up to one mismatch. […] For Figure 2, we therefore allowed Bowtie multi-mapping in two regions with –best, –starra options and discarded reads shorter than 30 nt."

This approach resulted in many more mapped reads in rRNA regions for Figure 2:

“A first analysis revealed that most PAR-CLIP sequencing reads fall into the mRNA transcript class, although many of the factors also show a considerable number of sequencing reads in non-coding RNAs, in particular rRNA (Figure 2A).”

3) Subsection “RNA end-processing complexes differ in their targets”: SUTs can be cytosolic, so this might not be sufficient to show that Dcp2 has nuclear activity.

We have changed the text to better reflect this point:

“For all decapping-related factors we observed similar binding preferences. […] This is consistent with previous findings that SUTs get degraded via Dcp2-dependent pathways in the cytoplasm (Marquardt et al., 2011; Smith et al., 2014; Thompson and Parker, 2007).

4) Subsection “Distinct factor distribution along mRNA”, Figure 1B (and end of subsection “PAR-CLIP experiments of S. cerevisiae proteins”): It would be helpful to discuss more why Csl4, Rrp4 and Rrp40 have distinct profiles. In particular, the Rrp4 profiles looks very different from all other degradation factors. This raises question of the PAR-CLIP experiment really worked for each cofactor. It is notable that some of these have low numbers of XL sites (561 for Rrp6, 923 for Rrp40, 7,373 for Csl4, while Rrp44 has much more 44,721).

We have repeated the experiments for members of the exosome to obtain 2-4 biological replicates with high reproducibility. In our new data-set we have at least 175,000 cross-link sites for each of the core exosome subunits and the metagene profiles show a consistent trend between them (Figure 3).

5) Subsection “Interactions between RNA processing machineries”, Figure 4 and Figure 4—figure supplements 1-2Figure 4: It would be helpful to discuss why Trf4/Air2 is bound to introns (Figure 1) but not Trf5 while Trf5/Air1 is showing higher co-occupancy / colocalization with splicing factors

With our new definition of enrichment (taking into account both row and column normalization), we see a stronger enrichment for Mtr4, Air1, and Air2 for intronic regions. This is reflected in the colocalization analysis (Figure 5—figure supplement 2), where both TRAMP4 and 5 show high correlation with splicing factors. In contract, co-occupancy reflects the tendency to bind to the same transcripts regardless of the binding position. Because TRAMP4 and 5 bind to most mRNAs, not just intron containing ones, its low co-occupancy with splicing factors is expected.

Reviewer #3:

[…] The presented data are certainly of interest to a wide audience. However, there is no independent validation of any of the unexpected binding patterns. This is especially problematic, given that many datasets are based on single biological replicates. Moreover, some observations challenge our current thinking about the function of the investigated factor(s), which, to be suitable for publication in eLife, should be supported by independent experiments.

Thank you very much for this important comment. Here is the reply to the same issue, raised by reviewer 1, point 1:

To ensure the highest possible data quality and reproducibility, we generated 49 new datasets, obtaining between two and four highly reproducible biological replicates for each degradation factor (Figure 1—figure supplement 1). Upon comparing the replicates, we found poor reproducibility for Pop2, Not1, Rrp4 and Caf40 and therefore omitted old replicates from the analysis. Furthermore, we applied a more conservative threshold for including data and eliminated six of the previous replicates that had less than 10.000 cross-link sites (Csl4, Rrp40, Rrp6, Ski2, Ski3, and Ski7). As a result, only highly reproducible data with a high number of crosslink sites were included in the final data set. This has led to a clearly visible improvement of the analysis and revised manuscript.

We now write:

“The high reproducibility of these PAR-CLIP experiments is revealed by a comparison of two independent biological replicates that we collected for all 30 degradation factors (Figure 1—figure supplement 1), with Spearman correlations between 0.87 and 1.00 (mean: 0.94).”

The extent of improvement of the PAR-CLIP data can be appreciated by comparing the old Figure 2 with new Figure 3.

- Functional relevance:

The manuscript presents a series of unexpected and somewhat controversial observations. One example concerns a possible role for Pop2 and Not1 in decay of CUTs. Any functional relevance of such binding should be trivial to validate. This would be relevant for other of the most striking findings. Such validation may be possible by re-analysis of published datasets, for example by intersecting occupancy with RNAseq upon depletion/mutation of the same factors.

Thank you for the comment. We have generated replicates and sometimes triplicates to show reproducibility of our data and significance of our observations. We observed problems with reliability of some factors in the older dataset, particularly Not1 and Pop2. We repeated the experiments for these factors at least twice. According to our new dataset, Pop2 and Not1 behave similarly to other members of the deadenylation machinery (Figure 2-4).

- Replicates and Technical artefacts:

PAR-CLIP is a complex technique, where technical artefacts are not uncommon. For example, binding to highly abundant tRNA, snRNA and snoRNA species is often observed, even for factors where such binding is unlikely to occur or be relevant in vivo. Hence, it is quite possible that binding to such species, especially when based on a single biological replicate, does not reflect biology. Such considerations should be mentioned in the manuscript. Ideally, all factors for which single replicates are presented, should be backed up by additional data. This may be simply an additional replicate, or some kind of independent validation.

More generally, more information on how the authors scrutinize the quality of their PAR-CLIP experiments should be given (i.e. more explicitly than simply referring to another publication).

Thank you for the comment. We have generated replicates and sometimes triplicates to show reproducibility of our data and significance of our observations. We observed problems with reliability of some factors in the older dataset, particularly Not1 and Pop2. We repeated the experiments for these factors at least twice. According to our new dataset, Pop2 and Not1 behave similarly to other members of the deadenylation machinery (Figure 2-4).

[Editors' note: the author responses to the re-review follow.]

The manuscript has been considerably improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

1) Even though some of the more controversial suggestions of the original draft are now resolved, comparison with available knock-out or mutant / RNA-seq datasets would greatly help to support, or dismiss, some of the mentioned novel findings.

While there are some RNA-seq experiments available to investigate functions for some of these factors, they were performed with different strains and under different conditions. However, we previously published a dataset of microarray expression measurements for the same strain background. In this work, synthesis and degradation rates upon knock-out of several degradation factors have been determined (Sun et al., 2013).

We correlated the PAR-CLIP-measured occupancy across mRNA transcripts with the change in degradation rate upon knock-out from this dataset. The heat map shows the Pearson correlations.

As expected, these values are anti-correlated, meaning that transcripts that are bound more by degradation factors experience a decrease in degradation rate upon knock-out. But unfortunately, we see little structure in the correlation matrix that would connect a PAR-CLIP profile to its respective knock-out experiment. Interestingly, we observed strongest anti-correlation of degradation rates with other degradation factor knock-outs for the deadenylation factor Pop2 with deadenylation and exosome components and the 5’→3’ exonuclease Xrn1 with decapping and NMD factors. This indicates a function for Pop2 in RNA degradation from the 3’ end, and an effect on RNA degradation from the 5’ end for Xrn1. We have previously shown a Xrn1 dependent buffering of transcript levels (Sun et al., 2013).

We believe this rather unstructured pattern is largely due to secondary effects based on the extensive buffering that balances changes in degradation rates by adapting synthesis rates when one of the factors is knocked-out (Sun et al., 2013). Moreover, the viability of these knock-out strains suggests that such bound transcripts could get processed by alternative degradation pathways. Finally, the lack of data on non-coding transcripts makes it difficult to validate further hypotheses. We have therefore not included this analysis in the manuscript.

2) If available, western blots and/or autoradiographs of the different PAR-CLIP experiments would be nice to include to allow readers to judge the quality of different IP experiments.

We included one representative Western Blot of the immunoprecipitation for each factor to show the quality of the different experiments as Figure 1—figure supplement 2.

3) Some of the conclusions presented in the Results sections still appear vague or unclear. This concerns for example interpretation of the RNA exosome binding profiles.

a) i.e. In the first paragraph of the subsection “The exosome and surveillance factors”: unclear what different binding patterns "reflect" the distinct functions of the RNA exosome.

Thank you for pointing this out. We have changed the text to make the conclusions clearer: “This complex distribution of cross-links for different exosome subunits to different RNA classes reflects the distinct functions of the exosome in nuclear RNA surveillance, processing of stable non-coding RNAs, and cytoplasmic mRNA degradation (Zinder and Lima, 2017).”

b) In the second paragraph of the subsection “Distinct factor distribution along mRNA”: How/why would TRAMP4 binding upstream of the TSS affect early termination of the mRNA? Isn't it more plausible that it serves a role in termination of (unannotated) upstream ncRNA? Similarly, 5' bias of exosome and TRAMP subunits could arise from unannotated NUT-like transcripts.

Many thanks for pointing this out. Prompted by point 4 – as explained below – we have shown better now that the binding of TRAMP4 to regions upstream of TSS originates from NUT-like transcripts (compare Schulz et al., 2013). Since this explains the signal more readily, we have substituted the analysis with the one concerning attenuated genes. We have adapted the text in the second paragraph of the subsection “Surveillance of aberrant nuclear ncRNA”.

We also observed that upon removal of the cross-link sites that fall into previously annotated NUTs, and CUTs, the 5’ bias of the exosome remains intact (see new Figure 3—figure supplement 3). This suggests that such transcripts are apparently not the source of the binding bias on the 5’ end for exosome factors.

4) Figure 3 and Figure 3—figure supplement 1B: What RNAs generate the high signals for Mtr4 and Air2 5' to the TSS? This region is expected to give rise to very few transcripts. The upstream ncRNAs are further upstream and on the opposite strand. There must be a concern that some sequences have been mis-mapped. Were the sequences mapped only to a small number of loci, which would strengthen this concern? The authors should re-check the data supporting these distributions.

Thank you for this word of caution, which prompted us to investigate in more detail the origins of this signal. The reads are mapped uniquely to the whole genome with a minimum length of 20 bases and just one mismatch allowed. This strict mapping criteria makes it very unlikely for reads to get mis-mapped. However, we showed before that upon Nrd1 knock-out aberrant transcription upstream of TSS becomes visible (Schulz et al., 2013). To show that the responsible Nrd1-unterminated transcripts (NUTs) are indeed the RNAs for which we see TRAMP4 binding signal, we performed two further analyses that are explained in the text:

“It has been shown that Nrd1 is involved in terminating transcripts upstream of the TSS. […] Comparison of the observed antisense profiles (Figure 4) with those obtained after excluding cross-link sites in previously annotated NUT and CUT regions (Figure 4—figure supplement 2) confirms that most of the signal originates from transcripts that are targeted by the Nrd1/Nab3 machinery.”

5) The second paragraph of the subsection “The exosome and surveillance factors” refers to Figure 2—figure supplement 3 and not Figure 3—figure supplement 3.

Thank you. We corrected the text accordingly.

6) Figure 5C: 3' degaradation => 3' degradation

Thank you. We corrected the figure accordingly.

https://doi.org/10.7554/eLife.47040.045

Article and author information

Author details

1. Salma Sohrabi-Jahromi

   Quantitative and Computational Biology, Max-Planck-Institute for Biophysical Chemistry, Göttingen, Germany

   Contribution

   Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing

   Contributed equally with

   Katharina B Hofmann

   Competing interests

   No competing interests declared
   "This ORCID iD identifies the author of this article:" 0000-0002-8417-8230

2. Katharina B Hofmann

   Department of Molecular Biology, Max-Planck-Institute for Biophysical Chemistry, Göttingen, Germany

   Contribution

   Data curation, Validation, Investigation, Methodology, Writing—original draft, Writing—review and editing

   Contributed equally with

   Salma Sohrabi-Jahromi

   Competing interests

   No competing interests declared
   "This ORCID iD identifies the author of this article:" 0000-0002-0683-6277

3. Andrea Boltendahl

   Department of Molecular Biology, Max-Planck-Institute for Biophysical Chemistry, Göttingen, Germany

   Contribution

   Data curation, Methodology

   Competing interests

   No competing interests declared

4. Christian Roth

   Quantitative and Computational Biology, Max-Planck-Institute for Biophysical Chemistry, Göttingen, Germany

   Contribution

   Formal analysis, Visualization, Methodology

   Competing interests

   No competing interests declared

6. Carlo Baejen

   Department of Molecular Biology, Max-Planck-Institute for Biophysical Chemistry, Göttingen, Germany

   Contribution

   Data curation, Methodology

   Competing interests

   No competing interests declared

7. Johannes Soeding

   Quantitative and Computational Biology, Max-Planck-Institute for Biophysical Chemistry, Göttingen, Germany

   Contribution

   Conceptualization, Supervision, Funding acquisition, Methodology, Writing—original draft, Project administration, Writing—review and editing

   For correspondence

   Competing interests

   No competing interests declared

Metrics

• 5,021 Page views
• 960 Downloads
• 11 Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

1. Salma Sohrabi-Jahromi
2. Katharina B Hofmann
3. Andrea Boltendahl
4. Christian Roth
5. Saskia Gressel
6. Carlo Baejen
7. Johannes Soeding
8. Patrick Cramer

(2019)

Transcriptome maps of general eukaryotic RNA degradation factors

eLife 8:e47040.

https://doi.org/10.7554/eLife.47040