CAT tails drive on- and off-ribosome degradation of stalled polypeptides

Stalled translation produces incomplete, ribosome-associated polypeptides that Ribosome-associated Quality Control (RQC) targets for degradation via the ubiquitin ligase Ltn1. During this process, the Rqc2 protein and large ribosomal subunit elongate stalled polypeptides with carboxy-terminal alanine and threonine residues (CAT tails). Failure to degrade CAT-tailed proteins disrupts global protein homeostasis, as CAT-tailed proteins aggregate and sequester chaperones. Why cells employ such a potentially toxic process during RQC is unclear. Here, we developed quantitative techniques to assess how CAT tails affect stalled polypeptide degradation in Saccharomyces cerevisiae. We found that CAT tails improve Ltn1’s efficiency in targeting structured polypeptides, which are otherwise poor Ltn1 substrates. If Ltn1 fails, CAT tails undergo a backup route of ubiquitylation off the ribosome, mediated by the ubiquitin ligase Hul5. Thus, CAT tails functionalize the carboxy-termini of stalled polypeptides to drive their degradation on and off the ribosome.

we observed substantial CAT tail dependence independent of Ltn1 (in the ltn1∆ 2 background) (Extended Data Fig. 1c and subsequent text). To ensure that the CAT tail 3 dependence observed for RQCsub LONG was not an artifact of changes in ribosome 4 stalling, we designed a quantitative stalling reporter similar to one used by the Hegde 5 group 10,23 : RQC2 or LTN1 mutations did not affect stalling relative to control (Extended 6 Data Fig. 1d). These data suggest that CAT tails facilitate degradation of RQCsub LONG 7 but are dispensable for RQCsub SHORT . 8 9 A previous study proposed that CAT tails aid degradation by extending the RQC 10 substrate so that lysine residues buried in the ribosome exit tunnel emerge from the 11 ribosome and can be ubiquitylated by Ltn1 24 . RQCsub LONG has a lysine 24 amino acids 12 away from the stall sequence, placing it in the 35-40 amino-acid-long exit tunnel 25 at the 13 point of stalling. However, mutation of the buried lysine to arginine (which cannot be 14 ubiquitylated) preserved the bulk of CAT tail dependence (from 24% to 19%) ( Fig. 1f;  15 Extended Data Fig. 1e). Similarly, mutation of ten lysine residues in the C-terminal half 16 of RQCsub LONG maintained CAT tail dependence (from 24% to 22%) (Fig. 1f). Although 17 these mutations placed the proximal lysine 150 residues from the stall sequence, ltn1∆ 18 still stabilized this substrate (Extended Data Fig. 1e). These data suggest that Ltn1 19 activity is not restricted to lysine residues close (in primary sequence) to the exit tunnel. 20 Therefore, CAT tails can mediate degradation of RQC substrates without displacing 21 lysine from the exit tunnel. 22

Substrate structure affects degradation 1
We next sought to find the properties of RQCsub SHORT and RQCsub LONG that drive their 2 differences in CAT tail dependence. These substrates differ in their capacity to co-3 translationally fold (Fig. 2a). After stalling occurs, the linker in RQCsub LONG is long 4 enough to displace ten of the eleven GFP beta strands out of the exit tunnel, which 5 enables the nascent GFP to adopt a stable conformation 26  To evaluate the hypothesis that CAT tails promote degradation of structured RQC 5 substrates, we tested how modulating folding of the substrate changes CAT tail 6 dependence. To modulate folding of RQCsub LONG , we took advantage of the 7 temperature-sensitive folding of the GFP variant (GFP-S65T) used in the substrate 27 . As 8 incubation temperature increased (decreasing GFP-S65T folding capacity 27 ), CAT tail 9 dependence for RQCsub LONG decreased (Fig. 2b). When we replaced GFP-S65T with 10 the less temperature-sensitive superfolder-GFP 27,28 , RQCsub LONG-superfolder had high CAT 11 tail dependence even at high temperatures (Fig. 2b). These data support a model where 12 CAT tails enhance degradation of structured substrates but are dispensable for 13 unstructured substrates. 14 15 Next, we tested this hypothesis on spectrin, a protein whose co-translational folding is 16 well-understood and easily modulated 29,30 . We designed RQCsub SPECTRIN using the 17 same RFP-T2A module followed by a spectrin domain, a C-terminal GFP beta strand 18 (β11), and lastly a polyarginine stall sequence. We quantified RQCsub SPECTRIN levels 19 using a "split-GFP" strategy, co-expressing the N-terminal GFP fragment from another 20 transcript 31 . We observed that RQCsub SPECTRIN was 12% CAT tail dependent, while a 21 folding-disrupting mutation 29,30 eliminated CAT tail dependence ( Fig. 2c; Extended Data 22 If CAT tails mediate degradation of structured RQC substrates, we expect that addition 4 of unfolded domains to a structured substrate will relax its CAT tail dependence for 5 degradation. To test this hypothesis, we appended spectrin variants to the N-terminus of 6 RQCsub LONG . Folding-disrupted spectrin eliminated CAT tail dependence (from 24% to 7 1.7%) while folding-competent spectrin did not (from 24% to 28%) ( Fig. 2d; Extended 8 Data Fig. 2b). This finding prompted us to inquire whether the presence of unfolded 9 domains alone or, instead, flexible lysine residues within the domains abrogate CAT tail 10 dependence. To distinguish between these possibilities, we added unstructured FLAG-11 tag variants (with or without lysine residues) to the N-terminus of RQCsub LONG . The 12 lysine-containing FLAG-tag decreased CAT tail dependence (from 24% to 4.9%), 3.8-13 fold more than did lysine-free FLAG-tag (19%) (

Ltn1-independent ubiquitylation 18
While impairing CATylation affected the stability of some RQC substrates differently 19 when Ltn1 was intact, impairing CATylation in the ltn1∆ background dramatically 20 increased the stability of every substrate we measured (Extended Data. Fig. 1,2). 21 Furthermore, treatment with the proteasome inhibitor bortezomib also increased the 22 stability of RQCsub LONG in ltn1∆, but only when CATylation was intact (Fig. 3a). These 23 data suggest that CAT tails target proteins for proteasomal degradation by Ltn1-1 dependent and -independent mechanisms. 2 3 Either the process of CATylation or CAT tails themselves could serve as an Ltn1-4 independent degradation signal. To distinguish between these possibilities, we 5 employed a strategy to remove a substrate's CAT tail in vivo without disrupting the 6 process of CATylation. We co-expressed RQCsub LONG and TEV protease in vivo to 7 cleave RQCsub LONG 's C-terminus and remove its CAT tail. TEV co-expression 8 increased RQCsub LONG 's mobility on SDS-PAGE (Extended Data Fig. 3a), confirming 1 TEV activity in vivo. RQCsub LONG was stabilized by TEV co-expression in cells with 2 Rqc2 intact but not in cells incapable of CATylation (rqc2-d98a) (Fig. 3b). Therefore, 3 CAT tails themselves target RQC substrates for Ltn1-independent degradation. 4 5 We next searched a set of candidate genes from the ubiquitin-proteasome system to 6 identify an E3 ligase that ubiquitylates CATylated proteins in the absence of Ltn1. 7 Deletion of the proteasome-associated E3 ligase HUL5 32 stabilized RQCsub LONG in the 8 ltn1∆ background as much as removing CATylation (rqc2-d98a) (Extended Data Fig.  9 3b). Furthermore, during a cycloheximide chase, hul5∆ slowed decay of RQCsub LONG in 10 the ltn1∆ background as much as impaired CATylation (Extended Data Fig. 3c). This 11 indicates that the cell continuously degrades CATylated proteins when Ltn1 is limiting, 12 dependent on Hul5. The stabilization we observed after loss of Hul5 was not due to 13 perturbed CATylation, as RQCsub LONG had identical amino acid composition in ltn1∆ 14 and hul5∆ ltn1∆ (Extended Data Fig. 3d). By microscopy, disruption of HUL5 did not 15 affect aggregation of RQCsub LONG (data not shown), suggesting that the disruption of 16 degradation did not result from a change in solubility. When Ltn1 was intact, hul5∆ 17 specifically stabilized the CATylated species; hul5∆ significantly stabilized RQCsub LONG 18 with RQC2-WT but not rqc2-d98a (Fig. 3c). These modest effects observed in rqc2-19 d98a were likely non-specific, as hul5∆ also weakly stabilized a non-stalling degradation 20 sequence ("degron") 33 (Extended Data Fig. 3e). These results support a role for the E3 21 ligase Hul5 in Ltn1-independent degradation of CATylated proteins. 22 Hul5 has E4 ligase activity, which extends existing ubiquitin conjugates to create poly-1 ubiquitin chains that boost proteasome processivity 32,34,35 . It is thus possible that we 2 identified Hul5 because degradation of CATylated proteins requires extension of a 3 mono-ubiquitin mark left by another E3 ligase. A hallmark of E4 ligase activity is 4 stabilization of the mono-ubiquitylated substrate after loss of the ligase 36,37 , resulting in 5 an 11kDa shift (His-Myc-Ubiquitin) by SDS-PAGE. Purification of RQCsub LONG in the 6 ltn1∆ background revealed that hul5∆, like rqc2-d98a, diminished detection of 7 ubiquitylated conjugates without stabilizing an apparent mono-ubiquitylated species 8 (Fig. 3d). Thus, Hul5 is required for an E3 ligase activity that ubiquitylates CATylated 9 proteins. 10 11 While hul5∆ did not stabilize a mono-ubiquitylated species in the ltn1∆ background, 12 hul5∆ intensified a crisp band within the CATylated smear, ~1kDa above the lowest 13 band (Fig. 3d). This band disappeared after disruption of CATylation (rqc2-d98a) (Fig.  14   3d), suggesting that the corresponding protein contains short CAT tails of relatively 15 uniform size (~10-14 residues). To test whether short CAT tails support Hul5-dependent 16 degradation, we monitored RQCsub LONG stability in the presence of rqc2-d9a, an RQC2 17 mutant that produces short CAT tails (Extended Data Fig. 4b). In the ltn1∆ background, 18 rqc2-d9a preserved the majority of RQCsub LONG stabilization after hul5∆ that we 19 observed for RQC2-WT (Extended Data Fig. 4c). We posit that short CAT tails mark 20 proteins for Hul5-dependent ubiquitylation. 21 1

CAT tails are Hul5-dependent degrons 2
We wondered whether short tracts of alanine and threonine residues were sufficient to 3 confer the Hul5-dependent degradation we observed for CATylated proteins. To test 4 this, we replaced our model RQC substrates' stalling sequence with three non-stalling 5 arginine residues (preserving the stalling sequence charge) and appended defined 6 alanine and threonine sequences followed by a stop codon (Fig. 4a). These "hard-7 coded" CAT tails simulated natural CAT tails but had manipulable sequences and were 8 not RQC substrates. If a hard-coded CAT tail suffices for Hul5-dependent degradation, 1 its stability will be higher in hul5∆ cells than wild-type. We define this Hul5-dependence 2 as stability hul5∆ -stability wt (Fig. 4a). While the arginine C-terminus control and 3 alanine/threonine two-mers were not Hul5-dependent, "RRRATA" yielded weak Hul5-4 dependence (13%) (Fig. 4b). Doubling this motif to form "RRR(ATA) 2 " increased Hul5 5 dependence to 80%, but the "RRR(ATA) 4 " motif (54%) was weaker than "RRR(ATA) 2 " 6 ( Fig. 4b). Therefore, short hard-coded CAT tails suffice for destabilization by Hul5. 7 8 We then performed mutagenesis experiments to identify additional CAT tail properties 9 that confer Hul5-dependence. After making modifications to "RRRATATA," we found 10 that Hul5-dependence decreased after mutating alanine and threonine to glycine and 11 serine (especially alanine adjacent to arginine), replacing arginine residues with non-12 basic residues, or capping the C-terminus with two leucine residues (a relatively stable 13 C-terminal amino acid 38 ) (Fig 4b,4c). This mutagenesis series suggests that CAT tails 14 are effective degrons when: 1) adjacent to basic amino acids, especially when alanine is 15 directly adjacent, and 2) C-terminal. 16

17
Given that Hul5 destabilizes hard-coded CAT tails, we investigated whether these 18 proteins are ubiquitylated similarly to naturally CATylated proteins. Stability increased 19 upon bortezomib treatment for "RRRATATA" but not the arginine C-terminus control 20 (Fig. 4d). Purification of the most Hul5-dependent hard-coded CAT tail we tested, 21 "RRR(ATA) 2 " recovered ubiquitin conjugates whose detection was abolished upon 22 hul5∆ (Fig. 4e). As for the naturally CATylated RQCsub LONG , hul5∆ diminished all 23 ubiquitin conjugation and did not stabilize a mono-ubiquitylated species (Fig. 4e). Thus, 24 hard-coded CAT tails are sufficient to mark proteins that are not RQC substrates for 1 Hul5-dependent ubiquitylation. This sufficiency suggests that Hul5-dependent 2 ubiquitylation of RQC substrates can occur off the ribosome, unlike Ltn1-dependent 3 ubiquitylation 39 . 4 5 Figure 5 | Decomposition of the contribution of CAT tails to Ltn1 and Hul5 function. a, Model for how CAT tails enable degradation of RQC substrates by Ltn1 and Hul5. For unstructured substrates, ubiquitylation by Ltn1 occurs efficiently without CAT tails. CAT tails facilitate ubiquitylation of structured substrates, which are otherwise poor Ltn1 substrates. If Ltn1 fails, substrates released from the ribosome can be ubiquitylated, dependent on Hul5. b, Mean stability of RQCsubRz substrate whose mRNA self-cleaves and leaves thus stalls ribosomes without a polybasic tract (see schematic above). Error bars indicate s.e.m. from n = 3 biological replicates and p-and F-statistics from ANOVA (4 d.o.f.) are shown above bars. c, Scheme to decompose the contribution of CAT tails to Ltn1 and Hul5 function through combined perturbations to delete HUL5, remove long CAT tails with in vivo TEV cleavage, then mutate RQC2. d, Estimation of the the contribution of Ltn1 and Hul5 to CAT tail-mediated degradation of RQCsubLONG and RQCsubLONG with the C-terminal GFP lysine mutated (RQCsubLONG-Kl astR, as in Figure 1f). Data are presented as mean, and error bars indicate s.e.m. from n = 3 biological replicates. Raw data are presented in Extended Data Figure 5a. contain lysine in unstructured regions regardless of whether CATylation takes place 3 (Fig. 5a). However, CATylation enhances Ltn1's ability to ubiquitylate structured 4 substrates. If Ltn1 fails to ubiquitylate the substrate, short (~1kDa) CAT tails mark that 5 substrate for Hul5-dependent ubiquitylation, which can occur off the ribosome. To 6 support this model, we sought to test its key predictions and quantify how much CAT 7 tails contribute to Hul5 and Ltn1 function. 8 9 Our model predicts that CAT tails do not facilitate degradation of substrates that: 1) are 10 unstructured, and 2) terminate in a non-basic residue (preventing Hul5-dependent 11 degradation). We constructed such a substrate, RQCsub Rz , encoded by an mRNA 12 containing a hammerhead ribozyme that self-cleaves to produce a non-stop transcript 13 with a truncated 3' end that stalls ribosomes 40 . Before CATylation, the C-terminal 14 residue of RQCsub Rz is neutrally charged valine and there are too few residues 15 between GFP and the C-terminus to enable formation of the stable GFP conformation 26 . 16 As predicted, neither disruption of CATylation (rqc2-d98a) nor loss of Hul5 stabilized 17 RQCsub Rz (Fig. 5b). Thus, unstructured substrates terminating in non-basic amino acids 18 are not CAT tail dependent. 19 20 We next revisited RQCsub LONG to dissect how CAT tails mediate its degradation via 21 Hul5 and Ltn1. We first estimated CATylation's contribution to Hul5 and Ltn1 by 22 measuring the stabilization caused by hul5∆, assuming that Hul5 and Ltn1 activities are 23 independent (Fig. 5c). Using this analysis, we estimated that Hul5 mediates 40% of 1 CAT tail-dependent degradation and Ltn1 mediates the remaining 60% (Fig. 5d). We 2 were additionally interested in analyzing the size of CAT tails that facilitated Ltn1-3 mediated degradation. To estimate the contribution of long CAT tails, we co-expressed 4 TEV to cleave RQCsub LONG from the ribosome (and evade Ltn1) if its CAT tails were 5 long enough to expose the buried TEV-cleavage-site (greater than 21 residues) (Fig.  6 5c). TEV co-expression further decomposed Ltn1-mediated degradation into 19% 7 contributed by long CAT tails (TEV-sensitive) and 41% by short CAT tails (TEV-8 insensitive) (Fig. 5d). We repeated this analysis after mutating RQCsub LONG 's exit 9 tunnel-buried lysine. This mutation eliminated the contribution of short CAT tails to Ltn1 10 function, but increased the relative contributions of CAT tails to Hul5 and long CAT tails 11 to Ltn1 (Fig. 5d). We conclude that CAT tails enable RQC substrates to be targeted by 12

Conclusions 19
We propose that CAT tails facilitate ubiquitylation of RQC substrates in two stages. The 20 first stage occurs on the ribosome and is mediated by Ltn1; the second occurs off the 21 ribosome and depends on Hul5. 22 Our work reveals that the folding states of RQC substrates dictate how cells degrade 1 them. In particular, Ltn1 ubiquitylates structured RQC substrates inefficiently. 2 CATylation enhances Ltn1's ability to target these substrates, which may arise when 3 translation fails after synthesis of a folding domain (e.g. stalling within polyA when the 4 entire reading frame is synthesized). We propose two models that could explain how 5 CATylation facilitates on-ribosome Ltn1 activity on structured substrates: 1) CATylation 6 impairs substrate folding, or 2) an accessory factor binds CAT tails to promote substrate 7 unfolding or relax Ltn1's specificity for unstructured substrates. The latter is formally 8 possible because our data suggests that CAT tails long enough to emerge from the 9 ribosome exit tunnel, exposing potential accessory factor binding sites, contribute to 10 Ltn1 function. Future studies will elucidate the mechanisms by which CAT tails enable 11 Ltn1 to ubiquitylate structured substrates. 12

13
We discovered that short CAT tails made on the ribosome enable a backup route of 14 RQC substrate ubiquitylation when Ltn1 fails or is limiting. Rather than acting as an inert 15 extension of the RQC substrate, the alanine and threonine residues in CAT tails (along 16 with adjacent basic residues) form a Hul5-dependent degron. Critically, this same 17 alanine and threonine content mediates toxic aggregation of CATylated proteins when 18 the CAT tail is sufficiently long 17-19 . To maximize ubiquitylation by Ltn1 and Hul5 while 19 avoiding aggregation, we speculate that cells coordinate CAT tail synthesis with RQC 20 substrate degradation. How cells accomplish this regulation and ensure CAT tails are of 21 manageable length will be the focus of future research.   RQC2 mutants were constructed by first replacing 1.5kb 5' and 300bp 3' of the RQC2 1 start codon with a NATMX6 cassette. This gap was repaired in transformants by 2 transformation with a PCR product containing a LEU2 cassette and pRQC2-RQC2 3 variant N-terminus, amplified from plasmids containing these RQC2 variants. Log phase yeast growing in synthetic defined media were measured on a BD Accuri C6 19 flow cytometer (BD Biosciences) with 3-5 biological replicates (isolated clones) per 20 condition. Data were analyzed using custom Matlab scripts. Plasmid-expressing yeast 21 were selected by gating based on RFP fluorescence. Background signal bleeding from 22 the RFP channel into the GFP channel was calculated using an RFP-only control strain 23 1 Data Availability 2 The datasets generated during this study are available from the corresponding author 3 upon request. 4