Ribosome collision is critical for quality control during no-go decay

NGD results in multiple endonucleolytic cleavages that map well upstream of the stall site

A collection of reporters has been utilized by a number of groups to study NGD in Saccharomyces cerevisiae by presenting a genetically-encoded translational block. We elected to use a PGK1 reporter, initially engineered by Parker and colleagues, which expresses a stable stem loop (SL) known to inhibit ribosome movement along the mRNA (Doma and Parker, 2006). We also constructed two additional reporters, which contain a stretch of the inhibitory codons CGA (PGK1-CGA₁₂) encoding poly-Arg or AAA sequence (PGK1-AAA₁₂) encoding poly-Lys; both of these sequences have been well documented to inhibit translation and elicit NGD (Doma and Parker, 2006; Kuroha et al., 2010; Letzring et al., 2010; Lu and Deutsch, 2008; Tsuboi et al., 2012). As a negative control, we constructed a reporter with a polyU stretch (PGK1-UUU₁₂) encoding poly-Phe instead. As has been seen previously, all of the reporters that present a block to the ribosome were found to produce shorter mRNA products in a ski2Δ background; these strains are defective for 3′–5′ exonuclease activity by the cytoplasmic exosome (Anderson and Parker, 1998). Also consistent with earlier reports, these products ran as smears on the gel suggesting that they are heterogeneous in nature (Figure 1A). In the absence of XRN1 (Stevens, 1980), 3′-PGK1 fragments accumulated in the presence of the stalling reporters (PGK1-SL, PGK1-CGA₁₂ and PGK1-AAA₁₂) (Figure 1B). Similar to the 5′ intermediates, the 3′ intermediates ran as heterogeneous products on the gel.

Although noted earlier by others (Chen et al., 2010; Doma and Parker, 2006; Tsuboi et al., 2012), we were surprised by the extent of the smear length (> 500 nt) (Figure 1A). These observations suggested to us three different scenarios: 1) The cleavage reaction is inefficient relative to the elongation rate of the ribosome so that by the time cleavage initiates multiple ribosomes had protected the mRNA; 2) The endonuclease requires multiple stacked ribosomes for efficient cleavage; 3) Rescue of ribosomes behind the primary ribosome is relatively inefficient resulting in secondary stalls that trigger more NGD cleavages.

To explore these different scenarios, we sought to map cleavage sites for the reporters using RACE assays. Previous studies focused on mapping the 3′-fragments in the xrn1Δ background. In Figure 1C we used the data reported by the Parker and Song groups (Chen et al., 2010) to plot the number of reads from the (CGA)₄ reporter as a function of their relative position from the stall site. Consistent with the northern analysis, the 3′-fragments mapped to a broad region of the reporter (~0–250 nt upstream of the stall site). In these experiments cDNA was prepared by ligating an RNA oligo at the 5′ end of the mRNA and reverse transcribing using a primer that anneals downstream of the stall sequence (Chen et al., 2010). As a result, secondary cleavage reactions resulting from inefficient rescue are lost during this step since they do not possess the complementary sequence for the RT primer, thus excluding the third scenario.

The corresponding 5′ fragments have not been mapped to single-nucleotide resolution and are likely to provide further hints about the endonucleolytic reaction. We mapped these fragments by first ligating an adenylated DNA oligo to the 3′ end. Following first-strand synthesis with a primer complementary to the ligated oligo, the cDNA was amplified from the start of the PGK1 gene (eg. Figure S1A). Alignment of the resulting reads revealed some interesting clues about the cleavage reaction. In contrast to the 3′ fragments, for which most clustered within 200 nt of the stall site, the 5′ fragments mapped to a broader region upstream of the stall site (Figure 1D) suggesting that the 5′ fragments are subject to multiple cleavage reactions, presumably due to inefficient recycling by Dom34 and Hbs1. In agreement with this proposal, inhibition of ribosome rescue by deleting DOM34 leads to even more dramatic heterogeneity of the 5′-fragments (Figure S1B) (Ikeuchi and Inada, 2016).

Mapping of the 5′-fragments for the SL and (AAA)₁₂ reporters was largely consistent with that for the (CGA)₁₂ reporter. However, unlike the (CGA)₁₂ reporter, for which the reads were uniformly distributed across a broad region upstream of the stall site, the reads for the SL and (AAA)₁₂ reporters are clustered either 100–175 nt upstream of the stall (Figure 1E) or close to the stall site (Figure 1F). These findings suggest that the endonucleolytic reaction is to a certain extent dependent on the nature of stall. Some of these distinctions are likely to be rationalized by differences in the precise location of where the ribosome stalls as well as the makeup of the polyA sequence, especially given its slippery nature (Koutmou et al., 2015). These distinctions between AAA- and CGA-mediated stalling are likely to be the result of subtle differences between bona-fide-no-go decay and non-stop decay (NSD) (Frischmeyer et al., 2002; van Hoof et al., 2002).

Cleavage appears to takes place between stacked ribosomes

To gain further insight into the nature of the cleavage reaction, we decided to carry out high-throughput sequencing to map the 5′-fragments from the NGD reporters in the ski2Δ background. Overall, three main characteristics of the cleavage reaction are readily discernable from our analysis. First, very few cleavage products map downstream of the stall site. We mapped more than 7 × 10⁶ reads to the (CGA)₁₂ reporter with greater than 99.7% mapping upstream of the stall site (Figure 2A). The first major peak of reads mapped at about −43-nt (likely behind the secondary stalled ribosome). Second, we observe a predominant peak at 75 – 150 nt upstream, depending on the identity of the stall sequence (Figure 2B, S2). Third, and arguably most interesting, is the observation that cleavage takes place significantly in frame (Figure 2C, S2). This latter observation is similar to what has been observed recently using ribosomal profiling (Guydosh and Green, 2017) and suggests that the endonuclease is using the ribosome as a ruler. Consistent with this proposal, smoothing of the (CGA)₁₂ mapping data revealed a regular oscillation of read peaks with a median period of 29-nt (Figure 2B, 2D) and a tight distribution (standard deviation of 2.5-nt). The 29-nt periodicity has important ramifications for the nature of the endonuclease as it suggests that the cleavage reaction takes place among stacked ribosomes. This observation argues against the potential proposal that the heterogeneity of the cleavage products is the result of secondary cleavage reactions as this would result in a 15-nt periodicity instead; during secondary cleavage reactions ribosomes are expected to run to the end of the message and hence protect only one half the length of an 80S footprint.

For the SL and (AAA)₁₂ reporters, we were able to map ~5 × 10⁶ and ~ 4 × 10⁶ reads to their corresponding sequence, respectively (Figure S2). Interpretation of the mapping results, however, was not as straightforward as for the CGA reporter, presumably due to imprecise stalling on these sequences (Figure S2). This can be rationalized in light of earlier reports showing that tandem CGA sequences present harder stalls relative to other motifs in yeast (Chen et al., 2010). Notwithstanding, we also document a similar correlation between the cleavage reaction and frame to that observed for the CGA reporter. Collectively our analysis suggests that cleavage on NGD substrates is intimately coupled to the translational machinery and that the endonuclease may sense stacked ribosomes to initiate its reaction.

Multiple stacked ribosomes are required for robust NGD

Perhaps one of the most important findings to come out of our sequencing efforts is the observation that the primary cleavage is likely to take place well upstream of the stall site. This is especially true for the SL reporter, for which the cleavage sites appear to cluster to regions >100 nt from the stall (Figure S2A). These results are in agreement with a model whereby multiple stacked ribosomes are required to activate the endonuclease. To empirically provide support for this proposal we constructed new PGK1 reporters that place the SL sequence at distances close to the start codon, which allowed us to control a maximum number of ribosomes that can load onto the reporter (Figure 3A). In particular we positioned the SL at 10, 22, 48, 60, 105, 135, 175, 200 and 1040 nts from the AUG

In agreement with our sequencing results, we only observe robust cleavage on reporters having the SL at sites that are >135 nt from the start codon as evidenced by the accumulation of the 5′-fragments in the ski2Δ background (Figure 3A). Although we can detect minimal cleavage products in the SL₁₀ through SL₁₀₅ reporters, cleavage was more evident in the SL₁₃₅ reporter and was much stronger in the SL₁₇₅ reporter. To rule out the possibility that the absence of cleavage products in the reporters with SL close to the start codon is due to probing issues or due to possible trimming by an unknown exonuclease, we looked at the accumulation of the 3′-fragments in the xrn1Δ background. Similar to what we observe for the 5′-fragments, robust accumulation of the 3′-fragments was only detected for reporters expressing the SL at sites that are >105 nt from the AUG codon (Figure 3B). The effect of its position within the ORF on cleavage efficiency was not dependent on the identity of the stall. Indeed similar to the SL stall, cleavage efficiency was significantly reduced for the CGA and AAA stalls when they were placed <105 nucleotides from the initiation codon (Figure S3).

As NGD is fully dependent on translation (Doma and Parker, 2006), we next sought to verify that reporters that are not subject to NGD are capable of associating with the ribosome. We carried out polysome analysis on the SL₆₀, SL₁₃₅, SL₁₇₅ and SL₁₀₄₀ mRNA reporters to assess their distribution across the different ribosomal fractions. As expected the SL60 mRNA was found to associate with monosomes and disomes, whereas the SL135 was found to associate with trisomes and tetrasomes. The SL₁₇₅ and SL₁₀₄₀ mRNAs, including their stabilized 5′-fragments due to the ski2Δ background, could be detected in the heavier fractions of the polysomes (Figure 3C). We note that the endogenous PGK1 mRNA, which migrates faster and can be detected with our probes, was found to associate with the heavy fractions in all of our samples offering an internal control. These observations suggest that although the SL₆₀ reporter harbors a roadblock for translation and associates with ribosomes, it is not a good target of NGD. Collectively, our data argues that robust endonucleolytic initiation of NGD requires multiple ribosomes to load onto the mRNA target.

Reduction of ribosome density inhibits NGD

Although our data so far suggest that ribosome collision is key for initiating NGD, the mechanism by which this signal is communicated to the endonuclease is unclear. The L1 stalk of the ribosome, which protrudes away from the ribosome facing the mRNA exit channel, is an obvious candidate for sensing collision events. Structures of polysomes from human cells revealed a likely interaction between the ribosome and the small subunit of the upstream neighboring ribosome near the mRNA entry tunnel (Brandt et al., 2010). Furthermore, recent studies from the Grayhack group showed that deletion of RPL1B, one of two genes that encodes ribosomal protein L1 and is more highly expressed than its paralog RPL1A, allows for read through of inhibitory CGA codons suggesting that it may play a role in NGD (Letzring et al., 2013). However, an alternative possibility is that deletion of RPL1B is affecting overall ribosome homeostasis, by reducing the levels of available 60S subunits. Indeed studies from multiple groups have documented a profound relationship between ribosomal protein availability and ribosome assembly. In particular, deletion of ribosomal protein paralogues has been shown to result in defects in ribosome biogenesis (Ferreira-Cerca et al., 2005; Moritz et al., 1990; Mueller et al., 1998; Rudra et al., 2007).

Before exploring the effect of the RPL1B deletion on ribosome homeostasis we sought to assess its effect on the endonucleolytic reaction. Consistent with its effect on readthrough of inhibitory codons (Letzring et al., 2013), we observe a significant reduction in 5′-fragment accumulation from the PGK-SL reporter in the rpl1bΔski2Δ strain (Figure 4A). The disappearance of NGD intermediates can be rescued by reintroducing RPL1B on a plasmid suggesting that the effect is due to the loss of RPL1B. To evaluate whether the effect of RPL1B deletion is due to limiting levels of available ribosomes or altered ribosome composition, we isolated ribosomes from wild-type as well as rpl1bΔ cells and performed mass spectrometry to quantify the relative levels of the ribosomal proteins. We measured almost identical stoichiometries of the near complement of ribosomal proteins between the wild-type and the mutant strains (Figure 4B). In particular, in the rpl1bΔ strain, the level of RPL1 was found to be reduced by a modest 15% relative to its parent (Figure S4). This confirms that the effects of RPL1B deletion on NGD cannot be explained simply by altered ribosome composition and instead are likely due to changes in the cellular concentration of ribosomes. The polysome profile of rpl1bΔ provided support for this model; the profile exhibited a significant reduction of polysomes relative to the 80S monosome peak when compared to profiles from the wild type strain (Figure 4C). As expected, we observe a gradual increase in polysome levels in the wild type strain whereas in the rpl1bΔ strain an opposite steep and immediate drop in polysomes levels was noted, indicating that ribosome density on mRNAs is significantly reduced in the rpl1bΔ strain. This in turn could explain the effect of the deletion on NGD in a manner that is consistent with our collision-based model for triggering cleavage; ribosome density and hence “traffic” has a profound impact on the likelihood of ribosomes running into each other.

To provide further support for the effect of RPL1B deletion on ribosome homeostasis, we looked at the relative level of another ribosomal protein as a proxy for gauging the cellular levels of ribosomes. As predicted, the level of the small subunit ribosomal protein RPS9 is significantly reduced in the rpl1bΔ strain relative to wild-type cells (~30% reduction) as judged by western blotting (Figure 4D). This apparent reduction in ribosome concentration was corroborated by assessing the global rate of protein synthesis using ³⁵S-methionine incorporation. The rate by which the radiolabel was incorporated into nascent proteins was observed to be ~fivefold slower in the rpl1bΔ strain relative to wild-type cells (Figure 4E). We note that, superficially, these deficiencies in translation can rationalize the observed inhibition of cleavage; after all NGD is a ribosome-based surveillance mechanism. However, this rationale falls short of explaining that in the absence of RPL1B we observe increased accumulation of full-length protein products from a number of NGD reporters. We utilized a reporter similar to the one used by Grayhack and colleagues (Letzring et al., 2013), for which a dual luciferase construct is interrupted with CGA codons between the renilla and firefly luciferase coding sequences. In the absence of RPL1B, the ratio of firefly- to renilla-dependent luminescence increased more than twofold (Figure 4F). Similar effects of the deletion were seen with the PGK1 constructs; both the SL and (AAA)₁₂ reporters produced > twofold more protein in the rpl1bΔ strain (Figure 4G). These findings suggest that although decreasing ribosome density per mRNA has an overall negative effect on translation, it increases protein output on NGD reporters because it reduces the probability of ribosome collision and hence cleavage efficiency.

The likelihood of elongating ribosomes running into each other, in principal, depends on two parameters (assuming the elongation rate is similar): 1) the strength of the stall, which affects the dwell time of the ribosome on the pause sequence, and 2) ribosome density, which affects the arrival time of the ribosome behind the primary ribosome. We manipulated the first parameter by presenting stall sites of varying strengths to the ribosome. Previous work suggested that stretches of CGA and AAA codons present harder blocks to translation as compared to SL structures (Chen et al., 2010). Indeed in our hands, we observe little to no full-length protein products from these reporters, whereas we observe significant full-length protein product from the SL reporter (Figure 4G). In agreement with our model, the effects of RPL1B deletion on NGD efficiency was completely dependent on the strength of the stall sequence. Cleavage was significantly inhibited for the SL reporter but moderately inhibited for the (CGA)₁₂ reporter and the (AAA)₁₂ reporter (Figure 4H). These findings suggest that by increasing the stalling time, NGD becomes less affected by ribosome density.

To provide further support for our model we deleted other ribosomal protein genes encoding RPS26 and RPS28. Both are encoded by duplicated genes, and deletion of RPS28B appears to impart defects on the biogenesis of the 40S as evidenced by accumulation of 18S rRNA precursors in a ski2Δ rps28bΔ strain (Figure S5) (Ferreira-Cerca et al., 2005). Similar to what we observe for the rpl1bΔ strains, the stoichiometry of ribosomal proteins in the 80S ribosomes was largely unaffected in the rps26bΔ and rps28bΔ strains (Figure S5) as evidenced by mass spectrometry analysis suggesting that ribosome composition is not compromised in these strains.

Consistent with our proposal, deletion of RPS28B inhibited NGD on the SL and, to a lesser extent, on the (CGA)₁₂ and (AAA)₁₂ reporters (Figure 5A). In contrast deletion of RPS26B had little effect on the accumulation of the 5′-fragments. This difference between the two deletions can be easily rationalized by the severity of the RPS28B deletion relative to the RPS26B deletion. This was evident in the polysome profiles, for which the rps28bΔ strain exhibited a much more pronounced peak for the 60S subunit relative to the rps26bΔ strain (Figure 5B). Furthermore, in contrast to the rps28bΔ strain, the rps26bΔ strain did not accumulate 18S rRNA precursor. Corroborating these observations is the finding that the growth rate of the rps26bΔ strain was only slightly reduced, whereas that of the rps28bΔ strain was significantly reduced (Figure S5). Collectively our data on the ribosome mutants suggests that NGD efficiency is to a large extent dictated by ribosome density as a result of its direct consequence on the frequency of ribosome collision.

So far our analysis of the effect of ribosome density on cleavage focused on altering ribosome homeostasis, which is likely to have pleiotropic effects on cellular fitness. In our next set of experiments we focused our efforts on changing ribosome density per mRNA by modifying the reporter itself. We reasoned that by significantly increasing the UTR length, resulting in prolonged scanning by the small subunit, initiation rates should be reduced to an extent that can considerably limit the number of elongating ribosomes on the reporter (analogous to highway on-ramp signals to manage traffic). In particular, we replaced the UTR of the PGK1 gene in our control and (CGA)₁₂ reporter with that of the RPO21 and SCH9 genes (516- and 454-nts, respectively). As expected, increasing the length of the UTR led to a profound decrease in protein synthesis; in particular whereas induction of the PGK1-UTR control reporter resulted in an eightfold increase in PGK1 protein levels, induction of the RPO21- and SCH9-UTR reporters resulted in a dismal twofold increase of protein levels (Figure 5C). As important, these experiments also established that the addition of the long UTRs does not abrogate recognition by the translation machinery and the new reporters are translated, albeit inefficiently. We note that we did not observe an effect on protein accumulation of the (CGA)₁₂ NGD reporter, most likely because these reporters do not produce significant amounts of protein regardless of the UTR sequence (Figure 5C). Nevertheless, in contrast to the PGK1-UTR NGD reporter, for which short abortive products are observed, the reporters with the long UTRs failed to produce observable short protein products suggesting that they are not subject to NGD. In agreement with these observations, we detected little to no 5′-RNA fragments in the presence of the reporters that harbor the long UTRs (Figure 5D). These observations provide independent support for our model that cleavage is sensitive to ribosome “traffic” on NGD targets.

Induction of global collision events is accompanied by ubiquitination of the ribosomal protein RPS3

Two recent reports from the Hegde and Bennett groups (Juszkiewicz and Hegde, 2017; Sundaramoorthy et al., 2017) provided evidence to support that monoubiquitination of a number of ribosomal proteins is critical for resolving stalled ribosomes. Deletion of the E3 ligase ZNF598, or introducing mutations to its ribosomal proteins targets allows for readthrough on polyA sequences and inhibits downstream RQC events. In agreement with these proposals, deletion of HEL2, the yeast orthologue of ZNF598, was also previously shown to result in a similar phenotype (Letzring et al., 2013; Saito et al., 2015). Motivated by these observations, we sought to explore whether ribosome monoubiquitination is also triggered by ribosome collision. To accomplish this, we took advantage of the fact that substituting proline for glutamine at residue 56 (P56Q) of ribosomal protein RPL42 (eL42) renders ribosomes resistant to cycloheximide (Roguev et al., 2007; Shirai et al., 2010). Fortuitously, in yeast RPL42 is encoded by two paralogues, RPL42A and RPL42B, and as a consequence mutating one should result in a mixed population of ribosomes with respect to cycloheximide sensitivity. Hence, addition of cycloheximide to these strains should result in global ribosomal collision.

We initially set out to assess the relative expression of RPL42A and RPL42B in our strain. However, since the protein sequence is identical we opted to use qRT-PCR with gene-specific primers to assess their transcript levels. Both genes express to substantial levels (as compared to taf10), with RPL42B expressing at approximately twice the levels of RPL42A (Figure 6A). With this in mind, we chose to introduce the P56Q mutation to RPL42B as we hypothesized this will lead to increased ribosome stacking upon cycloheximide addition. We also introduced the same mutation to both genes to make a strain that is fully resistant to cycloheximide. As expected, a strain carrying wild-type copies of both genes was very sensitive to cycloheximide, for which no growth was observed when the antibiotic was added to concentrations higher than 160 ng/mL (Figure 6B). A strain harboring mixed populations of ribosomes was slightly more resistant to cycloheximide but overall sensitive; robust growth was observed at an antibiotic concentration of 160 ng/μL. Finally a strain carrying P56Q-mutant copies of both genes was fully resistant to cycloheximide even when the antibiotic was added to 10 μg/mL.

Having established that the RPL42 strains display the expected sensitivity towards cycloheximide, we next explored the effect of the antibiotic on ribosomal protein ubiquitination. We focused our efforts on RPS3, which has been shown to be ubiquitinated in response to stalling (Juszkiewicz and Hegde, 2017; Sundaramoorthy et al., 2017). Tagging the endogenous RPS3 with a FLAG tag allowed us to follow its modification as a function of cycloheximide addition using western blotting. We note that tagging RPS3 with the longer 3 × FLAG tag altered the modification pattern, suggesting that ubiquitination is likely dependent on the context of the C-terminus of RPS3. As expected, in the absence of cycloheximide we observe little to no modification of RPS3 regardless of the strain background (Figure 6C). In agreement with our collision-based model, we only observe robust modification of RPS3 upon addition of cycloheximide to the mixed-ribosome strain. The appearance of a band that corresponds to a mass shift of ~8 kDa (Figure 6C) is consistent with monoubiquitin being added to the protein as has been shown by others (Juszkiewicz and Hegde, 2017; Sundaramoorthy et al., 2017). In support of these ideas, we also detect an accompanying accumulation of ubiquitinated proteins as judged by anti-ubiquitin blotting. The addition of ubiquitin to RPS3 was further confirmed by immunoprecipitation of the protein (using anti-FLAG resin) and subsequent blotting for ubiquitin (Figure 6D). Consistent with our earlier western blotting analysis (Figure 6C), substantial signal for ubiquitin was only observed for the mixed strain in the presence of cycloheximide. Furthermore as expected, the modification was no longer detected when HEL2 was deleted from the mixed-ribosome strain, and the accumulation of ubiquitinated proteins was also absent (Figure 6E).

The observation that cycloheximide on its own does not result in modification of RPS3 suggests stalling on its own cannot activate the RQC pathway. Having said that, we reasoned that modification of RPS3 should be observed in the wild-type strain by titrating cycloheximide to a concentration where it no longer saturates ribosomes. In other words, concentrations at which binding of the antibiotic is dynamic should lead to collision of ribosomes since at any given point some ribosomes will be bound by the antibiotic whereas others will not. We titrated cycloheximide from 100 μg/mL down to 5 ng/mL and looked at the modification of RPS3. As expected, no ubiquitination was observed at the high and low concentrations of cycloheximide (Figure 6F). Instead, and in agreement with our model, we observed appreciable modification of RPS3 when the antibiotic was added at intermediate concentrations (0.13–3.7 μg/mL or 0.45–13 μM). Interestingly, these concentrations are close to the measured K_D values for cycloheximide binding to the eukaryotic ribosome (Garreau de Loubresse et al., 2014; Schneider-Poetsch et al., 2010), and as such association between the antibiotic and the ribosome is most dynamic. In summary, these findings on the RQC pathway strongly suggest that ribosome collision is somehow detected by the E3 ligase HEL2 to ubiquitinate certain ribosomal proteins.

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources or reagents should be directed to the lead contact, Hani Zaher ([email protected]).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

METHOD DETAILS

QUANTIFICATION AND STATISTICAL ANALYSIS

DATA AND SOFTWARE AVAILABILITY

High throughput sequencing data were deposited into the GEO database, accession # GSE101667.

Raw gel images are available from the Mendeley Database at http://dx.doi.org/10.17632/w8tgnf8rr5.1

KEY RESOURCES TABLE