A unique No-Go Decay cleavage in mRNA exit-tunnel of ribosome produces 5’-OH ends phosphorylated by Rlg1

The No-Go Decay (NGD) mRNA surveillance pathway degrades mRNAs containing stalled ribosomes. An endoribonuclease has been proposed to initiate cleavages upstream of the stall sequence. However, primary site of cleavage remains unknown. Indeed, direct evidence that two RNA fragments resulting from a precise and unique cleavage has never been obtained. We used mRNAs expressing a 3’-ribozyme to produce truncated transcripts in vivo that mimic naturally occurring truncated mRNAs, known to trigger NGD. We analysed ribosome associated NGD cleavage products at single-nucleotide resolution and show that a precise endonucleolytic cleavage event occurs within the mRNA exit tunnel of the ribosome, 8 nucleotides upstream of the first P-site residue. The first two stalled ribosomes are apparently not competent for mRNA cleavage. We demonstrate that NGD cleavage within the third or upstream ribosomes produces 5’-hydroxylated RNA fragments that are phosphorylated by the Rlg1/Trl1 kinase. The resulting 5’-phosphorylated RNA fragments are digested by the 5’-3 exoribonuclease Xrn1, but surprisingly, can also be trimmed by the 5’-3’ exoribonuclease activity of Dxo1 in Xrn1 deficient cells.


INTRODUCTION
The No-Go Decay (NGD) mRNA surveillance pathway degrades mRNAs containing stalled ribosomes 1,2 . NGD occurs when translation elongation is blocked by the presence of stable intra-or intermolecular RNA structures, enzymatic cleavage, chemically damaged sequences, rare codons or mRNA depurination 1,[3][4][5][6][7][8] . This mRNA degradation process is dependent on translation and involves an unidentified endoribonuclease that cleaves just upstream of the stall sequence 1,5,6,9 . Other mRNA surveillance pathways can also ultimately lead to NGD. For instance, transcripts synthesized without a stop codon due to premature polyadenylation have stalled ribosomes that are initially detected by the Non-stop decay (NSD) decay pathway 9,10 . NSD targeted mRNAs are cleaved by an uncharacterized mechanism and become targets of NGD when ribosomes reach the new 3'-end and stall 9,11 ,12 .
NGD thus plays a key role in resolving translational issues potentially detrimental to cellular homeostasis. When mRNAs are truncated, the stalled ribosomes are rescued in a process mediated by the Dom34/Hbs1 complex that dissociates the ribosomal subunits 5 . Their association with the 60S subunit is recognized by the Ribosome Quality Control (RQC) pathway leading to the rapid degradation of the nascent peptide [13][14][15] . However, despite extensive study, the precise location of NGD cleavage and the mechanism of degradation of the resulting RNA fragment remain unknown.
In this paper, we focused on the fate of NGD-cleaved mRNAs, with an initial goal of mapping the sites of mRNA cleavage with accuracy. Two major obstacles to achieving this objective are that NGD fragments are rapidly attacked by 5'-3' and 3'-5' exoribonucleases after ribosome dissociation 5 and that simultaneously blocking the 5'-3' and 3'-5' exoribonuclease decay pathways is synthetic lethal 16 . It has been shown, however, that the stability of such mRNAs is largely dependent on the Dom34/Hbs1 complex 5,17 . In dom34 mutant cells, ribosomes stalled at the 3'-end of truncated mRNAs inhibit the degradation by the exosome and facilitate the detection of sequential endonucleolytic cleavages upstream of the ribosomal stall site 5 . Interestingly, dom34 and xrn1 mutations (inactivating the main 5'-3' exonucleolytic degradation pathway) are not synthetic lethal 1 . Moreover, NGD endonucleolytic cleavages still occur in the absence of Dom34 2,3 . The limited 3'-5' degradation of specific mRNA targets (in the absence of Dom34) combined with 5'-3' exoribonuclease mutants thus allows an accumulation of RNA fragments resulting from endonucleolytic cleavages whose extremities can be mapped accurately. We created a series of truncated mRNAs in vivo by insertion of a hammerhead ribozyme sequence (Rz), known to generate NGD targeted mRNAs 5 . These constructions mimic chemically or enzymatically cleaved mRNAs, or those resulting from abortively spliced mRNAs that are processed by the NGD pathway 5,9,18 . As anticipated, these designed truncated 3'-ends block ribosomes at determined positions and, because ribosomes guide NGD mRNA cleavages 5,19 , we were able to detect 3'-NGD RNA fragments of specific sizes. By analysing these RNAs in detail, we demonstrated the existence of a unique cleavage site 8 nucleotides (nts) upstream of the first P-site nt and therefore propose that this event occurs within the mRNA exit tunnel of the ribosome. We show that the 3'-NGD cleavage products have a hydroxylated 5'-extremity and are then phosphorylated by the RNA kinase activity of the tRNA ligase Rlg1/Trl1 to allow degradation by the 5'-3' exoribonuclease Xrn1. In the absence of Xrn1, the alternative 5'-3' exoribonuclease Dxo1 takes over.

Mapping the 5'-ends of 3'-NGD RNA fragments at single-nucleotide resolution
To generate 3'-truncated mRNA substrates for NGD in vivo, we inserted a hammerhead ribozyme sequence 20 in the 3'-sequence of URA3 gene ORF (mRNA1Rz). This results in the production of an mRNA that lacks a stop codon and a polyadenylated tail, called mRNA1 in Fig. 1a and Supplementary Fig. 1a, and known to be an NGD target 5 . We first verified that we could detect NGD cleavages in the 3'-proximal region of mRNA1, by northern blotting with a probe corresponding to the 3'-end (probe prA, Fig. 1a and Supplementary Fig. 1a). The upstream and downstream cleavage products are referred to as 5'-NGD and 3'-NGD RNAs, respectively (Fig. 1a). We indeed detected a ladder of 3'-NGD RNA fragments in dom34 mutant cells (Fig. 1b), in the presence or absence of active 5'-3' or 3'-5' exonucleolytic decay pathways, i.e. xrn1 or ski2 mutations, respectively 21 . In agreement with the current NGD model in which endonucleolytically cleaved 3'-NGD fragments are primarily degraded by the 5'-3' exoribonuclease Xrn1 5 , inactivation of the 5'-3' RNA decay pathway (xrn1 mutant cells) produced a different ladder of 3'-NGD RNAs compared to WT or the ski2 mutant. This was confirmed by a higher resolution PAGE analysis followed by northern blotting (Fig. 1c). The PAGE analysis was completed by mapping the 5'-ends of the 3'-NGD RNA fragments in the dom34 and dom34/xrn1 mutants by primer extension experiments with prA ( Fig. 1d). We showed that the truncated mRNAs produce several discrete 3'-NGD RNA bands (B1-5) that can be mapped to single-nucleotide resolution. B5 (77 nts) and the major RNA species B1 (47 nts) were only detected in the presence of active 5'-3' exoribonuclease Xrn1 (Fig. 1b, 1c and 1d); B3 (68 nts) and B2 (65 nts) RNAs were exclusively observed in the xrn1 mutant cells, and B4 (71 nts) was detected in all three strains. The sizes of the major B1 and B5 RNAs differ by 30 nts (Fig. 1d), consistent with the length of mRNA covered by an individual ribosome 22 . We therefore surmised that the difference in size is most likely due to the presence of an extra ribosome protecting the B5 RNA species from 5'-3' degradation by Xrn1 23 , compared to B1. This prompted us to analyse how these 3'-NGD RNAs are protected by ribosomes in vitro.

Ribosome protection of 3'-NGD RNA fragments from RNase activity
To assess the protection of 3'-NGD RNAs by ribosomes, we first performed an Xrn1 protection assay (Fig. 1e). We focused on the fate of the major RNA species B1 (47 nts) and B5 (77 nts) detected in dom34 cell extracts that likely correspond to RNAs protected from Xrn1 digestion in vivo by trisomes and disomes, respectively. Indeed, an association of a 47nt RNA species with disomes was deduced from ribosome profiling experiments 9,19,[22][23][24] and is explained by the approximate size of the trailing ribosome protecting a full ribosome footprint (28-30 nt) and the leading ribosome protecting a half footprint to the site of mRNA truncation (16-17 nts with no RNA or an incomplete codon in the A-site). An additional ribosome would thus be expected to protect a ~77-nt RNA. We thus verified that Xrn1 treatment of RNA present in dom34 cell extracts had no impact on B5 and B1 RNAs in vitro (Fig. 1e), since we anticipated that these RNAs were already protected from Xrn1 digestion by ribosomes in vivo, and thus should be still protected in cell extracts in vitro. Interestingly, the persistence of B4 RNA after Xrn1 treatment suggests that this RNA is also potentially protected by ribosomes in the dom34 background (Fig. 1e). We also added purified Xrn1 to cell extracts of the dom34/xrn1 strain in vitro and showed that it can efficiently recapitulate the production of the B1 species observed in Xrn1-containing cells in vivo. The appearance of the B1 RNA was inversely correlated to the amount of B4, B3 and B2 RNAs remaining, suggesting that these three species have unprotected 5'-protruding RNA extremities in vivo, due to the absence of the 5'-3' exoribonuclease Xrn1 (Fig. 1e). The B5 RNA was also generated at the expense of some larger species by Xrn1 treatment in vitro, consistent with the presence of trisomes on this species in the dom34/xrn1 cell extracts (Fig. 1e). Based on these experiments, we propose that B1 (47 nts) and B5 RNAs (77 nts) correspond to Xrn1-trimmed RNAs protected by two and three ribosomes, respectively 23 and that the 71 nt Bd4 RNA is also protected from Xrn1 by three ribosomes.
To validate the presence and number of ribosomes on 3'-NGD RNAs by a second method, and particularly the presence of trisomes on the 71-nt B4 RNA, we also performed RNase I protection assays on cell extracts of dom34 and dom34 xrn1∆ strains. We hypothetised that the B5 and B4 RNAs protected from RNase I by three ribosomes should be detectable with both probes prA and prD (Fig. 1f). The presence of two or three ribosomes on the major RNA species B1, B4 and B5 in dom34 cells (deduced from primer extension experiments and Xrn1 treatment in vitro) would preferentially conduct RNase I to cleave at three major sites, Cut1, 2 and 3 in Fig. 1f. After RNase I treatment (Fig. 1g), the accumulation of RNase I protected RNAs of similar size to B5 suggested that this RNA is covered by trisomes in dom34 (Fig. 1g). It is known that RNase I and Xrn1 cleave about 15 nts 9 and 17 nts 23 upstream of the first nt of the ribosomal A-site, respectively. We thus expected 5'-end of the B5 RNA to be 2 nt shorter after RNase I treatment compared to Xrn1, if we had accurately surmised the positions of ribosomes on this RNA. Indeed, primer extension experiments confirmed that ribosomes protect 75-nt RNAs against RNase I in the dom34 background (Supplementary Fig. 1b and Supplementary Fig. 1c). The equivalent of the B1 RNA was 45-46 nt in size after RNase I treatment ( Supplementary Fig. 1d) and, surprisingly, the B4 RNA was not detected using probe prA (Fig. 1g). We thus wondered whether RNase I cleaved preferentially at Cut3 (Fig. 1f), thus preventing the detection of B4 using probe prA. We probed the membrane in Fig. 1g with prD ( Fig. 1h), and two distinct RNA species were detected, corresponding to B5 and B4 processed by RNase I, and inefficiently cleaved at the Cut2 site (Fig. 1f). We thus proposed that the 5'-extremities of B4 RNA are protected by ribosomes. We conducted the same experiment on the B4 RNA from xrn1/dom34 cell extracts. These RNAs were sensitive to Xrn1 treatment in vitro (Fig. 1e), and using probe prD to detect the B4 RNA specifically, we observed that these RNAs were also protected from RNase I to a similar extent as B4 in dom34 cell extracts (Fig. 1h). Together these results allow us to infer the precise positions of ribosomes on RNA species in the different mutant backgrounds (Fig. 1i). In the dom34 context, the B1 and B5 RNAs correspond to RNAs covered by disomes and trisomes, respectively, after processing by Xrn1 in vivo (Fig. 1i). We propose that three ribosomes cover the 71-nt B4 RNA in dom34 mutant cell extracts as this species is resistant to Xrn1 (Fig. 1e) and its 5'-region is protected from RNase I digestion in vitro (Fig. 1h). Whether two or three ribosomes dwell on the 71-nt Bd4 RNA in xrn1/dom34 mutant cell extracts is unclear as this species is sensitive to Xrn1 in vitro (i.e. have 5'ribosome-free extensions that can be pared down to B1 by Xrn1 digestions) (Fig. 1e), which would be consistent with protection by two ribosomes, but resistant to RNase I (Fig. 1h), more consistent with three.

Dxo1
We thus strongly suspected that B4 RNAs were original NGD products, and because they were exclusively detected in Xrn1 deficient cells, we speculated that the B3 and B2 RNAs might be derived from B4 RNAs by an alternative 5'-3' exoribonuclease. We therefore asked whether the 5'-3' exoribonucleolytic activity of Dxo1, on the margins of its important role in 5'-end capping quality control 25  and B2 RNA production to a significant extent, but a catalytic mutant failed to do so (Fig. 2b).
We took advantage of the almost exclusive presence of the B4 3'-NGD RNAs in dom34/xrn1/dxo1 mutant cells to re-question how this RNA is protected by ribosomes, by adding Xrn1 to cell extracts as described above. Some of the B4 RNA was Xrn1-resistant (Supplementary Fig. 2b) in accordance with our hypothesis that a portion of this species is still protected by three ribosomes in xrn1/dom34 cell extracts ( Supplementary Fig. 1c).
Remarkably, most of the B4 RNA was also degraded to a 47-nt species ( Supplementary Fig.   2b), strongly suggesting that disomes persist on the majority of the 3'-end of truncated RNAs in dom34/xrn1/dxo1 cells in vivo. We thus conclude that two populations of B4 RNAs coexist in vivo in Xrn1 deficient cells, with one resembling B4 RNAs covered by three ribosomes like in dom34 mutants and the other having 5'-protuding RNA extremities due to the absence of 5'-3' exoribonucleases.
We performed a number of experiments to probe the role of Dxo1 under conditions where Xrn1 is still present, but when its activity is attenuated. Inhibition of the 5'-3' exoribonuclease activity of Xrn1 occurs upon accumulation of the metabolite 3'-phosphoadenosine-5'-phosphate (pAp), for example in Met22 deficient cells, or in cells exposed to high levels of toxic ions such as sodium or lithium 26,27 . Remarkably, in cells containing Xrn1, the met22 mutation or the addition of lithium led to the accumulation of B3 and B2 RNAs, while still maintaining the production of the B5 and B1 species ( Fig. 2c and Supplementary   Fig. 2c). Hence, Dxo1 can participate in 3'-NGD RNA trimming even under conditions where Xrn1 is still partially active.
We asked whether Dxo1 trimming of NGD-cleaved mRNAs is dependent on the nature of the nucleotide sequence. To do this, we modified the nucleotide sequence encompassing the site of the potential endonucleolytic cleavage by building derivatives of mRNA1 (called mRNA2, 3 and 4). We probed the resulting decay intermediates in dom34, dom34/xrn1 and dom34/xrn1/dxo1 mutants, and observed that although the B2 and B3 3'-NGD RNA intermediates produced by Dxo1 from these RNAs were slightly different in levels and in size compared to their equivalents from mRNA1, the production of the B4 RNA was remarkably preserved in all four constructs ( Fig. 2d and Supplementary Fig. 2d). Taken together, these results provide evidence that B2 and B3 RNAs are products of Dxo1 activity and that Dxo1 plays a general role in supporting Xrn1 in the degradation of the B4 3'-NGD RNA.

Identification of the primary NGD endonucleolytic cleavage site
The results described above suggested that the B4 RNA is the major 3'-product of NGD cleavage in our constructs and that it is trimmed to smaller sizes by Xrn1 and Dxo1.
While its resistance to Xrn1 in vitro (Fig. 1e) could be explained by a third ribosome dwelling after cleavage, we also considered the possibility that its 5'-phosphorylation state could contribute to its stability, since both Xrn1 and Dxo1 require 5'-phosphorylated extremities to degrade RNA 25,28 . We therefore asked whether the B4 RNA naturally has a monophosphate or a hydroxyl group at its 5'-end by treating RNA purified from dom34 cell extracts with T4 polynucleotide kinase to see whether this would stimulate attack of B4 by Xrn1 in vitro.
By definition endonucleolytically cleaved NGD mRNAs result in 5' and 3'-NGD RNA fragments. We thus searched for the corresponding 5'-NGD fragment for the B4 3'-NGD RNA. To map the 3'-end of 5'-NGD RNAs, total RNA preparations from ski2 and ski2/dom34 mutants were ligated to a pre-adenylated oligonucleotide linker using truncated RNA ligase (Fig. 3b) 30 . The ski2 mutant context was used to limit 3'-trimming of these RNAs in vivo. RNAs were pre-treated with T4 polynucleotide kinase to modify 2'-3' cyclic phosphates to 3'-OH to permit RNA ligation 30 . Linker-ligated RNAs were reverse transcribed, amplified by PCR and cloned for sequencing, using a method called 3'-RNA ligase mediated RACE (called 3'-RACE below) (Fig. 3b) 31 . The major RT-PCR product was of the expected size (66 bp; Fig. 3c and Supplementary Fig. 3a) and verified by sequencing resulting clones (Fig. 3d). The identification of a matching 5'-NGD fragment for the B4 3'-NGD RNA, confirms that an endonucleolytic event occurred at this precise position.
Remarkably, the same procedure performed on RNAs isolated from ski2 mutants where Dom34 is still active yielded the same major PCR product, also verified by sequence ( Supplementary Fig. 3b).

The fate of 5'-NGD RNAs
We anticipated that following NGD cleavage of mRNA1, ribosomes that had initiated translation on the 5'-NGD fragments would advance to the new 3'-end and the RNA subjected to Xrn1 trimming, similar to the process that generates B1 and B5 (Fig. 4a). Since the B4 3'-NGD RNAs are cut in the +1 reading frame (Fig. 1i), upstream ribosomes on these 5'-NGD RNAs would be expected to stall with one nucleotide in ribosome A-site ( Fig. 4a and Supplementary Fig. 4) and as result produce new RNA fragments 47+1, 77+1 nts, protected by two or three ribosomes, respectively (see Supplementary Fig. 4). Indeed, using probe prG, complementary to the new 3'-ends generated by NGD cleavage, we detected RNA fragments consistent with a 1-nt increase in size compared to prA by northern blotting the same membrane (Fig. 4b). We mapped the 5'-ends of these new ribosome protected fragments by primer extension assays using prG ( Fig. 4c and Supplementary Fig. 4). The 48-nt (and 78-nt) cDNAs only detected in cells containing active Xrn1 (Fig. 4c) strongly suggest that the 5'-NGD endonucleolytic products are covered by two and three ribosomes, respectively. The production of cDNAs of exactly the predicted sizes (48 and 78 nts) is an independent confirmation that the 3'-extremity of the 5'-NGD product corresponds precisely to the proposed NGD endonucleolytic cleavage site ( Supplementary Fig. 4). Remarkably, the 3'extremity of the 5'-NGD RNA was detected in the context of active 3'-5' exonucleases, meaning that ribosomes run on and cover the 3'-extremity before any 3'-5' attacks can occur.
In summary, we propose that the B4 RNA is produced when associated with three ribosomes and that at least two upstream ribosomes promptly protect the resulting 5'-NGD fragment from degradation (Fig. 4d).
In contrast, we observed that a major portion of B4 RNAs from dom34/xrn1 and dom34/xnr1/dxo1 cell extracts can be degraded by Xrn1 in vitro without prior 5'phosphorylation ( Supplementary Fig. 1e, 2b), suggesting that B4 accumulates as a 5'phosphorylated species in this mutant background. We believe that the B4 RNAs in dom34 mutant cells and dom34/xrn1 mutant cells (i.e in dom34/xrn1/dxo1 mutant cells) differ by their association with three and two ribosomes, respectively (Fig. 1e, 1i and Supplementary   Fig. 1c, 2b) Fig. 5a). We therefore did an experiment to show that the accumulation of B4 seen upon thermo-inactivation of the rlg1-4 allele is due to accumulation of 5'-OH RNAs and not a simple inactivation of Dxo1. We treated total RNA from rlg1-4/dom34, xrn1/dom34 and rlg1-4/xrn1/dom34 with Xrn1 in vitro and showed that the accumulation of B4 RNA lacking a 5'-phosphorylated extremity (i.e. Xrn1 resistant in vitro) at the non-permissive temperature correlated with the presence of the rlg1-4 ts mutation ( Fig.   5b and 5c). We also observed that rgl1-4 mutants grow for several hours at 37°C, and that 100% of the B4 RNA species lacks a 5'-phosphorylated extremity after a 16-hour shift at 37°C (Supplementary Fig. 5b, 5c). The RNA kinase activity of Rlg1 is thus required for the 5'-exoribonucleolytic digestion of 3'-NGD RNA products.

Identification of NGD cleavage products on mRNAs containing rare codons
We asked whether we could identify endonucleolytic cleavages on another NGDtargeted mRNA, using what we learned about this process on truncated mRNAs. We chose an mRNA containing four contiguous rare CGA codons, which we call (CGA) 4 -mRNA, as an NGD target (Fig. 6a and Supplementary Fig. 6a) 5 . Similarly to the truncated mRNAs, ribosomes were shown to stall when decoding rare codons and 5'-and 3'-NGD RNAs were produced (Fig. 6a). As previously demonstrated 5,32 , we verified that 3'-NGD RNAs fragments can be detected in dom34 or DOM34 genetic contexts by northern blotting experiments using probe prB ( Supplementary Fig. 6b). The precise identification of endonucleolytic cleavages by primer extension experiments is known to be challenging 5 probably because, in contrast to truncated mRNAs, the positioning of ribosomes on contiguous rare codons is less precise. We first asked whether we could detect the resulting 5'-NGD RNAs (Fig. 6a) using the same procedure as for NGD-targeted truncated RNAs (Fig. 4a and 4b). By probing the (CGA) 4 -mRNA in a large region upstream of the four CGA codons ( Supplementary Fig. 6a), we detected RNA bands using a probe annealing 71 nts upstream of the first rare codon (probe prH, Fig. 6b and Supplementary Fig. 6a). Similar to the 5'-NGD RNAs produced from NGDtargeted mRNA1 (Fig. 4b), RNA detection required dom34 genetic backgrounds (Fig. 6b).
The profile of the 5'-NGD RNAs resulting from endonucleolytic cleavages of (CGA) 4 -mRNA are remarkably similar to B1, B4 and B5 RNAs from the truncated mRNA1. We then treated these RNAs with Xrn1 and, as anticipated, we showed that the ~71-nt RNAs, like B4 RNAs, are Xrn1-resistant, and that ~47-nt and ~77-nt RNAs, like B1 and B5 RNAs, were Xrn1sensitive. These results strongly suggest that NGD-targeted (CGA) 4-mRNAs are a source of truncated RNAs which are, in turn, processed like mRNA1 by the NGD pathway.
The detection of short RNA species by prH probe suggested that endonucleolytic cleavages occurred just downstream, in a region located ~70 nts upstream of the cluster of rare codons (Supplementary Fig. 6a). We thus set out to map the NGD cleavage sites on the (CGA) 4-mRNA, using 3'-RACE for the detection of 5'-NGD RNA 3'-ends in ski2 and ski2/dom34 mutant cells (Supplementary Fig. 6c). We obtained major RT-PCR products of about 45 bp that were purified, cloned and sequenced ( Fig. 6d and Supplementary Fig. 6d).
The sequences of the 3'-ends (Fig. 6e) formed three clusters, C1, C2 and C3 (Fig. 6g), that map to ~71 nts upstream of the second, third and fourth rare codon, respectively, consistent with cleavage within the footprint of the third ribosome as seen for the truncated mRNA1. No 3'-ends were detected within the region covered by the two ribosomes, comforting the notion that disomes are not competent for NGDase activation. Xrn1 arrests mapping to 17-18 nts upstream of the A-site of the two first ribosomes positioned with either the second or third CGA codon in the A-site were also detected through primer extension experiment (Fig. 6f, 6g and 6h). The strongest Xrn1 arrests corresponded to those where the lead ribosome contains the third CGA codon in the A-site (Fig. 6h), suggesting that the major stall occurs on this codon. Typically, Xrn1 is preferentially blocked 17 nts upstream of the first ribosomal A-site residue 23 . We speculate that this 1-nt difference reveals distinct conformations of stalled ribosomes on rare codons versus truncated mRNAs. All these results taken together suggest that the (CGA) 4 -mRNA and truncated mRNA1 are NGD-targeted in a highly similar process that results in cleavage within the footprint of the third ribosome, 71 nts upstream of the stall site for the leading ribosome.

Discussion
In this study, we first characterized the 3'-NGD RNA fragments produced near the 3'end of truncated mRNAs that mimic natural cleaved mRNAs known to be NGD targets. One advantage of studying the 3'-NGD products of truncated mRNAs is that the positioning of stalled ribosomes results in 3'-NGD RNA fragments of specific sizes. Using a ribozyme to efficiently generate precise 3'-ends within an open reading frame, we were able to obtain detailed information about the ribosomal positioning on 3'-NGD RNAs, and provide the first precise mapping of the original site of endonucleolytic cleavage on an NGD substrate. Our model suggests that this occurs 8 nts upstream of the first P-site nt of the third ribosome stalled at the 3'-end of the truncated mRNA ( Fig. 1i and Fig. 7). This localizes the cleavage within the mRNA exit tunnel, 4 nts before the RNA emerges from the ribosome and becomes available for cleavage by RNase I, classically used in ribosome foot-printing studies. This site is consistent with the idea that the NGD endonuclease, NGDase 19 , might be the ribosome itself. However, we cannot fully exclude the possibility the stalled ribosome allows access to an external nuclease with a specific conformation to penetrate this far into the mRNA exit tunnel.
The NGD endonucleolytic cleavage detected within the third stalled ribosome, suggests that the first two stalled ribosomes are not competent for the activation of the NGDase. Here we use the term disome to designate the two first ribosomes blocked in elongation and inactive for RNA cleavage (Fig. 7). The 3'-RACE experiments did not amplify DNA products corresponding to RNAs having the predicted sizes of NGD-cleaved RNAs with the second (41 nts) or first stalled ribosome (15 nts) (predicted sizes 95 and 125 nts), suggesting that they do not occur to any significant level. The major ~65-bp RT-PCR products obtained corresponded perfectly to RNAs cleaved 71nt upstream of the 3'-extremity of mRNA1, suggesting this is the primary site of NGD cleavage.
Xrn1 treatment of various mutant cell extracts suggested that the predominant configuration on truncated mRNAs is disomes. Interestingly, the existence of disomes on truncated mRNAs has been previously reported in ribosome profiling analysis 19 and stacking of two or more ribosomes has recently been proposed as a prerequisite for the activation of the endonuclease 24 . The latter observation led to the proposition that ribosome collision triggers NGDase. We propose another model in which disomes are not competent for NGDase activation, but that three or more ribosomes are required. This suggests that the conformation of disomes is particular and incompatible with NGDase activation. On the contrary, the repetitive band pattern in our northern blot analyses of stalled complexes ( Fig.   1c and 4b) suggests that five or more ribosomes are present on the mRNA1 construct and that the third ribosome and those further upstream are all competent for NGD cleavage. The ability to induce NGD cleavage appears thus to be a normal property of stalled ribosomes, with disomes being an exception.
The dom34 mutation can exaggerate the ribosome stalling and can facilitate the cleavage beyond what would naturally be observed. As discussed in the introduction, analysis of NGD RNA fragments is facilitated by the dom34 mutation and is crucial for RNA stabilization when analysing truncated mRNAs by northern blotting experiments 5,19 . In the presence of Dom34, and more efficient ribosome dissociation, the exosome would certainly be more actively involved once the first endonucleolytic cleavage event has occurred 5,19 . Importantly, the 3'-RACE experiments confirmed the existence of 5'-NGD products having 3'-ends matching to the 5'-extremity of the 3'-NGD B4 RNA in Dom34 active cells (Fig. 3c and 3d). These observations were used to map endonucleolytic cleavages that occur on a second NGD-target mRNAs containing rare codons, in a DOM34 genetic context. Endonucleolytic cleavages were mapped 71nts upstream of the first residue in the first ribosome A-site, in the region potentially covered by the third ribosome. We propose that in this case also, a particular conformation, or associated factor of disomes are responsible for their inability to trigger NGDase activity.
Our experiments also show that NGDase produces downstream cleavage products bearing a 5'-hydroxyl group (Fig. 3a), typical of cleavage reactions not involving a metal ion.
We show that the Rlg1/Trl1 kinase, in addition to its role in tRNA splicing, phosphorylates the 3'-NGD fragment to allow degradation by Xrn1 and Dxo1 (Fig. 5 and Supplementary Fig.   5). The resistance of the 3'-NGD B4 RNA fragments to Xrn1 attacks in vitro and in vitro ( Fig. 1e) is likely to be a direct consequence of the presence of the third ribosome preventing access to Rlg1/Trl1. Accordingly, extremities of the B4 RNAs were shown to be predominantly 5'-hydroxylated in a dom34 context, while an important portion of B4 RNAs are 5'-monophosphorylated in dom34/xrn1 and dom34/xrn1/dxo1 mutant cell extracts, where these RNAs accumulate in the absence of Xrn1 and are mostly associated with disomes. Rlg1 is thus a far more general RNA kinase than previously suspected, and also acts in the NGD pathway.
The unexpected function of Dxo1 in the NGD pathway might appear secondary but its implication can explain the heterogeneity of 3'-NGD RNA fragments observed in xrn1 or xrn1/dom34 mutant cells. Xrn1 and Dxo1 activities, combined with Rlg1/Trl1 RNA kinase activity have likely masked the detection of the major NGD cleavage products in previous studies. Additionally, the fact that we observed that the 5'-3' exoribonucleolytic activity of Dxo1 is naturally thermosensitive raises the question of its known decapping activity on unmethylated-capped mRNAs and the biological consequences of their accumulation at high temperature 25 .
In conclusion, NGDase cleavage does not randomly occur upstream of the ribosomal stall and is not an artefact of dom34 genetic context. It can also be precisely mapped as a unique cleavage event within the ribosomal mRNA exit tunnel at 8 nt of the P-site residue.
Remarkably, mRNAs containing rare codons are processed similarly, but cleavage accuracy is slightly affected and might be explained by a particular conformation of the first stalled ribosomes that correlates with specific Xrn1 arrests (Fig. 6f). Ribosomal A-site and occupancy of the mRNA entrance tunnel by mRNAs containing rare codons could explain some of the minor differences observed with truncated mRNAs. We also learned that 5'-OH NGD RNA fragments are produced, phosphorylated by Rlg1/Trl1 and degraded by Xrn1 and alternatively by Dxo1. Indeed, we learned that the inactivation of Xrn1 can lead to a Dxo1 trimming that can mask many original cleavages and can scramble global analysis. Moreover, Rlg1-Xrn1-(Dxo1) may potentially play a general role in the healing and degradation of a multitude of other 5'-hydroxylated RNAs, such as those identified by Peach et al. 33 . It will be also interesting to confirm a role of all these factors for these substrates in yeast and for its homologous kinase hClp1 in human cells 34 . In conclusion, this study provides very important new mechanistic insights that will help to go further in the comprehension of all mRNA surveillance pathways in connection to NGD.

METHODS
Yeast Media, plasmids, strains, and oligonucleotides. The media, plasmids, strains of S.
cerevisiae, and oligonucleotides used in this study are described in supplemental information.
Northern blot analysis. RNA Extracts and northern blots were performed as described previously (Sinturel et al., 2012). Total RNA was resolved by 8% TBE-Urea polyacrylamide or 1.4%TBE-Agarose gels. Blots were exposed to PhosphorImager screens, scanned using a Typhoon FLA 9500 (Fuji), and quantified with ImageJ software.
In vitro RNA digestion. 5µg of total RNA or 2OD 260nm of cell extracts were digested by 1 unit of Xrn1 (Biolabs) in NEB buffer 3 at 25°C during 30 min unless otherwise indicated. NEB Buffer 3 was replaced by Kinase NEB buffer in all kinase assays in the presence or absence of Xrn1 (Fig. 5a). For RNase I treatment of cell extracts, 2OD 260nm of extracts (prepared without heparin) were incubated with 0.5 µl, 1 µl and 2 µl RNase I (Invitrogen, 100 units/ml) 30 min at 25°C. For total RNA treatment, 5µg of RNA were digested by 0.5 µl of RNase I, 30 min at 25°C. All RNase treatments were followed by RNA extraction and northern blot analysis as described above.
Primer Extension. Radiolabeled primers (primers prA and prE for RNA1, and primer prJ for for (CGA) 4 -mRNA) were used and reverse transcriptase (ThermoFisher) was used to synthesize a single-stranded DNA toward the 5'-end of the RNA. The size of the labeled single-stranded DNA was determined relative to a sequencing ladder (ThermoFischer Sequenase sequencing kit) on 5% TBE-Urea polyacrylamide gel. Oligonucleotides were radio-labeled with [γ-32P]ATP with the T4 polynucleotide kinase (Biolabs).

3'-end RNA mapping.
Mapping was performed according to the 3'-RNA ligase mediated RACE method described previously 31 with minor modifications: Total RNA preparations were first 3'-dephosphorylated using T4 PNK 1h at 37°C without ATP and pre-adenylated linker (Universal miRNA cloning linker, NEB) ligation was performed during 4h at 22°C in the presence of truncated ligase 2 (NEB) 30 . Reverse transcriptase reactions were performed using reverse primer prE complementary to the linker sequence. PCR primer prF specific to RNA1, or primer prK specific to (CGA) 4 -mRNA, were used with primer prE in PCR reactions ( Supplementary Fig. 3a and 6c). PCR products were purified, cloned into zero Blunt TOPO PCR Cloning vector (Invitrogen), transformed and plasmids sequenced.