MCTS2 and distinct eIF2D roles in uORF-dependent translation regulation revealed by in vitro re-initiation assays

We first used the Klhdc8a 5′ UTR (and the first 6 codons of its CDS, fused in-frame to firefly luciferase) to establish a set of three different reporter genes that would allow us to estimate re-initiation and leaky scanning rates (Fig. 1I). Thus, the WT uORF reporter contained the full, unmodified Klhdc8a 5′ UTR. Mutating the uORF start codon (no uORF reporter) allowed us to quantify the luciferase signal in the absence of the inhibitory uORF. A third reporter contained a mutated uORF stop codon, extending the uORF coding sequence all the way into the luciferase CDS, but in a different frame (overlapping uORF reporter). Thus, this reporter would give a luciferase signal only through leaky scanning of ribosomes across the uORF start codon. After reporter delivery into HeLa cells (via lentiviral constructs that also contained a Renilla luciferase reporter for internal normalisation), re-initiation signal can be extracted from the individual reporter readouts, as shown in Fig. 1J. The relative reduction of signal from no uORF to WT uORF reporters thus represents the overall inhibitory activity of the uORF, with the remaining signal corresponding to the sum of leaky scanning- and re-initiation-mediated reporter translation. By further subtracting the overlapping uORF reporter signal that derives from leaky scanning, we calculate an estimate for re-initiation. Using WT and Denr KO HeLa cells (Fig. 1H), this assay indicated strong inhibition by the Klhdc8a uORF, and the remaining signal that was largely DENR-dependent, as well as very low leaky scanning rates (Fig. 1J). The data further allowed us to propose a model of how ribosomal fluxes occur on the Klhdc8a reporter transcript in HeLa cells (Fig. 1K). Our analyses suggest that ~98% of ribosomes initiate on the uORF, and only ~2% undergo leaky scanning. Assuming further that ribosomes committed to uORF translation can undergo the two fates after termination—recycling or CDS re-initiation—and that there is no interference between uORF ribosomes, we would estimate that about 19% re-initiate onto the main CDS, a process that is largely DENR-dependent. Based on considerations that we further develop in the Discussion, we believe that this estimate of re-initiation probability may further represent a lower limit for re-initiation, with actual rates quite possibly higher.

We concluded that the translation of endogenous and reporter-encoded KLHDC8A was highly re-initiation-dependent and that our system based on three reporters, together with the use of WT vs. Denr KO cells, allowed the quantification of DENR-dependent re-initiation in vivo.

First, we carried out the in vitro translation assays on the set of Klhdc8a reporters (no uORF, WT uORF, overlapping uORF), using HeLa extracts prepared from both wild-type and Denr KO cells. These experiments revealed clear uORF- and DENR-dependence (Fig. 2E, left side; identical outcomes were seen for ‘long Klhdc8a 5′ UTR’ reporter, see Appendix Fig. S2C), closely resembling our observations from HeLa cells in vivo (Fig. 1J). The calculated ribosomal fluxes showed excellent correspondence as well (Fig. 2F). In particular, the uORF strongly impeded CDS expression with a very low rate of leaky scanning (1%). Re-initiation-mediated reporter expression (11%) was quantitatively lower than in vivo (19%, see Fig. 1K), yet remained highly DENR-dependent (i.e. in Denr KO extracts, the signal from the WT uORF reporter was similar to the leaky scanning signal from the overlapping uORF reporter; a summary of calculated re-initiation rates from all in vitro experiments of this study is also given in Appendix Fig. S3). Importantly, complementation with recombinant MCTS1-DENR (Appendix Fig. S2D,E) fully rescued re-initiation in Denr KO extracts (Fig. 2E, right), confirming the direct involvement of these proteins in re-initiation in vitro. Note also that in Denr KO, MCTS1 is strongly co-depleted (Fig. 1H) as a result of its instability in the absence of its binding partner (as previously observed in (Ahmed et al, 2018; Bohlen et al, 2020b)), making a rescue with both proteins necessary. We concluded that our in vitro assays showed very good correspondence with re-initiation parameters observed in HeLa cells in vivo.

Next, we assayed in vitro the start-stop uORF-containing Asb8 reporter set (Fig. 2G). These experiments revealed that this uORF was rather permissive to leaky scanning, likely due to poor Kozak context of its start codon; in particular, given that all stop codons (TGA, TAA, TAG) begin with a T nucleotide, start-stop uORFs by default have a (suboptimal) T in Kozak sequence +4 position. The lower efficiency of inhibition by the Asb8 uORF (as compared to the Klhdc8a uORF) was in line with the translation efficiencies measured in cells by Ribo-seq (Fig. 1F). According to our calculations (Fig. 2H), 58% of ribosomes underwent leaky scanning and 42% engaged in uORF translation. About two-thirds of uORF termination events would lead to re-initiation at the CDS and the remaining one-third would not (29 vs. 13%). Re-initiation was mostly MCTS1-DENR-dependent, as judged from the signals obtained in Denr KO extracts and from recombinant MCTS1-DENR rescue (Fig. 2G; Appendix Fig. S3).

MCTS1-DENR selectivity for certain uORFs has been linked to specific penultimate codons (Bohlen et al, 2020b). The Klhdc8a uORF terminates with the sequence ‘… GTG(Val) TAG(Stop)’, yet penultimate GTG was not observed as specifically associated with DENR-dependence previously (Bohlen et al, 2020b). In order to test penultimate codon influence in our assay, we tested reporters in which the GTG(Val) was mutated to GTC(Val) (among the reported significantly depleted penultimate codons in DENR-dependent uORFs (Bohlen et al, 2020b)) and to GCG(Ala) (most strongly and significantly enriched in DENR-dependent uORFs according to (Bohlen et al, 2020b)) (Fig. 2I). We found that both the inhibitory impact of the uORF and the re-initiation rates were only weakly affected by these codon changes in vitro. We noted that in the depleted penultimate codon case, GTC(Val), DENR-dependence of re-initiation appeared somewhat reduced, yet overall re-initiation capacity was still intact. This observation is compatible with the idea that the depleted penultimate codon is able to partially decouple the re-initiation process from its dependence on these particular re-initiation factors. Moreover, it may point to the involvement of MCTS1-DENR in the process of specific deacylated tRNA removal.

Next, we examined Asb8 reporter variants, converting the 1-aa uORF to a 2-aa uORF (and thus from start-stop to elongating uORF) and testing different penultimate codons. We inserted either a TGT(Cys), which is (according to (Bohlen et al, 2020b)) the only significantly enriched penultimate codon starting with ‘T’ (hence preserving the +4 Kozak context of the start codon), or GCG(Ala), which is a highly enriched codon and additionally improves the Kozak context for uORF initiation (Fig. 2J). In the in vitro translation assays, GCG insertion indeed led to strongly decreased leaky scanning rates, converting the weakly translated 1-aa uORF to a strongly translated 2-aa uORF (Fig. 2J, right), compatible with the improved Kozak consensus associated with the +4 position ‘G’. Lengthening the uORF with TGT(Cys) had little consequences on leaky scanning and overall re-initiation rate, yet DENR-dependence was significantly reduced (Fig. 2J, middle). We concluded that re-initiation rates were not per se different between start-stop and elongating uORFs, but the DENR-dependence of re-initiation was modulated, which is in agreement with the reported preferential MCTS1-DENR regulation of start-stop uORFs (Bohlen et al, 2020b; Schleich et al, 2017).

The phosphorylation of eIF2α plays a crucial role in the integrated stress response (ISR) by reducing overall translation initiation and selectively enhancing the translation of stress-related mRNAs. This process operates through a mechanism involving uORFs, re-initiation, and start codon selection (Dever et al, 2023). Following the termination of uORF translation and partial ribosome recycling—specifically the dissociation of the 60S subunit and removal of deacylated P-site tRNA—the 40S ribosomal subunit regains its scanning competence and can eventually re-initiate translation at a downstream start codon. According to the ‘delayed translation re-initiation model’, which is based on studies of transcripts such as Atf4/Gcn4, in conditions where non-phosphorylated eIF2α is abundant, a new eIF2-GTP-Met-tRNA^Met_i complex (also known as ternary complex, TC) is quickly recruited during the 40S scanning process. This rapid acquisition leads to translation initiation at a nearby second uORF (87 nt away) rather than at the more distant CDS, which is 184 nt downstream. Phosphorylated eIF2α inhibits the guanine nucleotide exchange factor eIF2B, which is necessary to convert inactive GDP-bound eIF2α to its active GTP-bound form, thereby limiting the availability of TC. In this scenario, the likelihood of timely TC recruitment for translation of the second uORF is diminished, increasing the chances of re-initiation at the downstream CDS. Recent studies have underscored the crucial roles of MCTS1-DENR and eIF2D in facilitating Atf4 CDS translation (Bohlen et al, 2020b; Vasudevan et al, 2020); however, the specific stage at which these proteins exert their effects— whether during ribosome recycling, TC recruitment, or re-initiation itself—remains unclear. Intriguingly, eIF2D was originally identified as having the ability to bind initiator tRNA in a GTP-independent fashion and deliver Met-tRNA^Met_i to the P-site in vitro (Dmitriev et al, 2010). This has led to the hypothesis that MCTS1-DENR and eIF2D might perform a similar function in vivo, potentially making TC activity for initiator tRNA delivery redundant during re-initiation—a model that has seen some critical review recently (e.g. (Grove et al, 2024)), prompting the question of whether TC/eIF2α activity is a limiting factor in MCTS1-DENR-mediated re-initiation, independent of its effects on global initiation.

To further investigate the above observations on potential eIF2D targets, we selected several candidate transcripts to test if their TE regulation was 5′ UTR-mediated. Of these, Cenpa, Ndc80, Med23 and Rps20 were picked as candidate mRNAs with low TE from Fig. 4C. We also included Hoxa3, which has been identified as a bona fide eIF2D-regulated mRNA in a previous study (Alghoul et al, 2021). According to the published findings, eIF2D is implicated in uORF-mediated inhibition, rather than in re-initiation, of Hoxa3 CDS translation. Indeed, our in vivo data in cells confirmed Hoxa3 CDS TE up-regulation upon Eif2d depletion (Fig. 4E; note that Hoxa3 is very lowly expressed in our cells and did not feature among the significantly changed TE transcripts in Fig. 4C, but the differences were statistically significant in the direct comparison of TEs shown in Fig. 4E). Luciferase reporter assays in cells using the 5′ UTRs of the selected endogenous candidate transcripts did not reveal significant regulation by eIF2D (Fig. 4F). We also analysed Klhdc8a (endogenous and 5′ UTR reporter), but we found no indication that eIF2D was involved in its re-initiation in cells (Fig. 4E,F). Endogenous Asb8 showed an increase in translation efficiency upon Eif2d depletion (Fig. 4E), contrasting the effect seen upon Denr depletion.

We next tested the outcome of eIF2D loss-of-function and rescue in our in vitro assay. In lysates from Eif2d KO HeLa cells (Fig. 1H), no difference in re-initiation activity or leaky scanning was detectable for the Asb8 reporter (Fig. 4G). By contrast, for the Klhdc8a reporter, re-initiation rates in Eif2d KO lysates were reduced by about half (Fig. 4H). However, neither recombinant eIF2D nor recombinant MCTS1-DENR were able to fully rescue the low signal obtained with the WT uORF reporter in Eif2d KO lysates (Fig. 4H). These observations were most compatible with the model that lower re-initiation signal was an indirect effect of eIF2D loss, e.g. due to secondary effects related to the widespread gene expression reprogramming seen in Fig. 4C. Thus, we would not expect indirect effects to be rescuable in vitro, whereas a direct role in the re-initiation process itself should be rescued by adding the needed re-initiation factor as a recombinant protein.

In Eif2d knockout extracts, MCTS1-DENR is abundant and may mask the re-initiation function of eIF2D. To address this possibility, we carried out two experiments. First, using re-initiation-deficient Denr KO extracts and recombinant eIF2D, we were indeed able to partially rescue re-initiation activity (Fig. 4I), although not as efficiently as with recombinant MCTS1-DENR (compare with Fig. 2E). Second, we used Denr + Eif2d HeLa double knockout cell lysates. In vitro translation reactions in these extracts showed reduced re-initiation activity (Fig. 4J, left) that was very similar to Denr single knockouts (Fig. 2E). Addition of recombinant MCTS1-DENR (0.5 μM) showed the expected rescue (Fig. 4J, right). Interestingly, also recombinant eIF2D (0.5 μM) showed rescue of the double knockout (Fig. 4J, middle). Although rescue with recombinant eIF2D appeared less potent than rescue with recombinant MCTS1-DENR at first sight—i.e., in the right panel to Fig. 4J, there is no significant difference on WT uORF reporter between MCTS1-DENR-supplemented WT vs. MCTS1-DENR-supplemented double KO lysates, whereas in the middle panel, double KO lysate supplemented with recombinant eIF2D still showed significantly lower signal than WT lysate supplemented with eIF2D—two-way ANOVA did not uncover a significant interaction and we concluded that eIF2D can potentially rescue as potently as MCTS1-DENR. Taken together, these results indicate that eIF2D is likely not an essential component of the re-initiation machinery in cells in vivo and in vitro under wild-type conditions where MCTS1-DENR is abundantly present. However, when MCTS1-DENR is absent, providing eIF2D protein can compensate for MCTS1-DENR loss. Thus, it is a probable scenario that eIF2D has, in principle, retained shared activity with MCTS1-DENR, but we deem it likely that its main cellular functions under physiological conditions lie outside of uORF re-initiation.

Finally, it is also possible that eIF2D is active in re-initiation in a conditional fashion; in particular, upon induction of the integrated stress response that leads to the unavailability of eIF2α, eIF2D may be able to deliver initiator-tRNA to the 40S ribosome. We addressed this possibility by carrying out in vitro translation in Eif2d knockout extracts (without/with recombinant eIF2D rescue) in the absence of GADD34Δ1-240 (Fig. 4K). Also, under these conditions, we observed the (presumed) indirect/secondary effects of eIF2D loss that could not be rescued by recombinant eIF2D, and the experiment revealed no further indications for a specific re-initiation role of eIF2D.

Human and mouse MCTS1 protein sequences are identical apart from one leucine to isoleucine substitution at amino acid position 30 (Fig. 5F). In MCTS2, divergence from MCTS1 can be found at 14 positions when aligning mouse, rat, human and macaque protein sequences, with four substitutions common to MCTS2 across species and other changes specific to the analysed rodents or primates (Fig. 5F). Using available structural data from in vitro reconstituted MCTS1-DENR in complex with the 40S ribosome, an initiator-tRNA molecule and a viral IRES-containing model RNA (Weisser et al, 2017) (please note that no such structure has so far been reported for a natural, non-viral transcript), we found that MCTS2-specific amino acid changes affected positions that all cluster in the same region of the fold but outside of the interaction surface between MCTS1-DENR and also outside of that between MCTS1-DENR and the 40S ribosomal subunit (Fig. 5G). Hence, the capacity of MCTS2 vs. MCTS1 to interact with DENR, and of MCTS2-DENR vs. MCTS1-DENR to interact with 40S can be expected to be comparable. However, the substitutions—of which in particular the change at amino acid position 75 from glutamate (negative charge) in MCTS1 to lysine (positive charge) in MCTS2 (Fig. 5F) would have the potential to significantly alter protein properties—could affect additional, so far unknown interactions taking place at this solvent-exposed surface within the complex (Fig. 5G). To gain insights into whether MCTS1-DENR and MCTS2-DENR could redundantly act in re-initiation, we tested both as recombinant proteins on Klhdc8a and Asb8 reporters. Addition of recombinant DENR alone showed only low rescue activity, likely due to residual MCTS1/2 in Denr KOs (see Fig. 1H), yet supplementing with either of the MCTS paralogs together with DENR led to a comparable and full rescue (Fig. 5H,I). Taken together, we concluded that Mcts2, although previously categorised as a pseudogene (Nandi et al, 2006), is expressed to relevant levels and produces a protein that assembles with DENR into protein complexes in cells in vivo. In vitro, the paralogous complexes show similar activity on our two re-initiation model substrates, indicating redundancy in activity, at least for certain substrates. Finally, it is noteworthy that our anti-MCTS1 antibody recognises recombinant MCTS1 and MCTS2 with similar efficiency (Appendix Fig. S6B), indicating that western blot quantifications of endogenous MCTS1 in our and other previous studies likely already reported on the combined protein levels of the two paralogs.

A pBluescript II SK(+) vector was ordered as a synthetic clone to contain the knock-in cassette 3×FLAG-NeonGreen-dTAG. Denr 5′ and 3′ homology arms were amplified from mouse genomic DNA by PCR using primers reported in Table 4 and ligated in the ‍pBluescript II SK(+) first digested with SpeI and BamHI for the 5′ homology arm, then with NdeI and KpnI for the 3′ homology arm. Positive clones were identified by colony PCR using PCR homology arm primers and by Sanger sequencing.

NIH/3T3, HEK293FT and HeLa cells were cultured under standard conditions (DMEM; 10% FCS, 1% penicillin/streptomycin, all from Invitrogen; 37 °C; 5% CO₂). The HeLa Denr knockout (KO) cells have been published (Bohlen et al, 2020b). For Eif2d KOs, wild-type HeLa cells were seeded in a six-well plate (200,000 cells per well) in DMEM with 10% FBS and 1% Pen/Strep. After 24 h, the cells were transfected with plasmid pX459 v2.0 into which a sgRNA sequence targeting Eif2d was cloned (sequence: CACCGCTGTGGACTGGAAACACCCG) using Lipofectamine 2000 following the manufacturer’s instructions. Twenty-four hours later, lipofectamine was removed, and cells were kept in DMEM for recovery for 2 days. On the third day, puromycin was added to a final concentration of 1.5 µg/ml. After another 3 days, puromycin was removed, and a fresh medium was added. After 2 days, cells were seeded into a 96-well plate at limiting dilution, so that only one cell per well was seeded. Wells were checked under the microscope for single-cell colonies. Upon reaching confluency, wells with only one colony were reseeded into six-well plates and kept for growing. Then immunoblots on these monoclonal lines were performed to identify the ones with loss of eIF2D. The knockout was verified at the genomic level via PCR using the following primers: 5′-GGGGAGGGCTCGGGTATGAC-3′ and 5′-AAGGCAGGGGCTGTCTCATATC-3′ to amplify the genomic region. This revealed premature stop codons in all alleles. For generating Eif2d / Denr double knockout cells, the same procedure was performed to introduce the Eif2d knockout into the previously published Denr-knockout HeLa line (Bohlen et al, 2020b).

Lentiviral shRNA plasmids targeting Eif2d were purchased from Sigma (Cat. No. TRCN0000119844 for shRNA1 and TRCN0000119845 for shRNA2). The Denr shRNA and the con1 shRNA have been previously described (see (Castelo-Szekely et al, 2019), where they are named “Denr shRNA2” and Scramble shRNA, respectively). The lentiviral shRNA plasmid for con2 shRNA was purchased from Sigma (Cat. No. TRCN0000119842) and was originally designed as an shRNA that can target Eif2d, yet as it shows no knockdown activity whatsoever (non-functional/nf shRNA) it is used as an additional control shRNA. Production of lentiviral particles in HEK293FT cells using envelope pMD2.G and packaging psPAX2 plasmids and viral transduction of NIH/3T3 cells were performed following published protocols (Castelo-Szekely et al, 2019; Salmon and Trono, 2007), with puromycin selection at 5 μg/ml for 4 days for shRNA-transduced cells.

Total protein extracts from NIH/3T3 and HeLa cells were prepared according to the NUN method (Lavery and Schibler, 1993). For analysis of phosphorylated proteins, PhosSTOP™ (Cat. No. 4906845001, Roche) was included in the NUN buffer. Antibodies used were rabbit anti-KLHDC8A 1:5000 (ab235419, Abcam), rabbit anti-VINCULIN 1:5000 (ab129002, Abcam) mouse anti-U2AF65↗ 1:5000 (U4758, Sigma), rabbit anti-eIF2D 1:750 (12840-1-AP, Proteintech), rabbit anti-DENR 1:5000 (ab108221, Abcam; used for all immunoblots apart from Appendix Fig. S6B), rabbit anti-DENR 1:1000 (PA5-54507, Invitrogen; used for Appendix Fig. S6B), rabbit anti-phospho-eIF2alpha (Ser51) 1:1000 (#9721S, Cell Signaling), mouse anti-FLAG 1:10000 (M2F3165, Sigma), rabbit anti-RPL23 (ab112587, Abcam), mouse anti-beta-Tubulin 1:5000 (T5201, Sigma) and secondaries goat anti-guinea pig-HRP 1:10000 (SA00001-12, Proteintech), anti-rabbit-HRP (W4011, Promega) and anti-mouse-HRP 1:10000 (S3721, Promega). The guinea pig anti-MCTS1 antibodies (used at 1:500) have been described (Ahmed et al, 2018). Blots were visualised with the Fusion Fx7 system.

The overall protocol for extract preparation was an adaptation of that published by Gurzeler et al, 2022 (Gurzeler et al, 2022). Briefly, confluent 15-cm dishes of HeLa cells were washed once with PBS at 37 °C, then trypsinised with 0.05% Trypsin-EDTA (Cat. No. 25300054, Gibco™). Harvested cells were collected by centrifugation at 500 × g for 5 min at 4 °C. The cell pellet was washed twice with ice-cold PBS and centrifugated at 200×g for 5 min at 4 °C. The cell pellet was resuspended in 1:1 lysis buffer (10 mM HEPES pH 7.3, 10 mM potassium acetate, 500 μM MgCl₂, 5 mM DTT and 1× cOmplete™ EDTA-free Protease Inhibitor Cocktail (Cat. No. 4693132001, Roche)). Cells were lysed by dual centrifugation at 500 RPM for 4 min at −5 °C using a ZentriMix 380R centrifuge (Hettich AG) with a 3206 rotor and 3209 adaptors. The resulting lysate was then centrifuged in a table-top centrifuge at 13,000 × g at 4 °C for 10 min, and the supernatant was aliquoted and snap-frozen in liquid nitrogen; storage at −80 °C for up to 6 months.

Recombinant eIF2D, MCTS1 and DENR protein purification has been described (Weisser et al, 2017). Recombinant MCTS2 was expressed and purified identically to MCTS1. Recombinant protein purity can be appreciated in Appendix Fig. S2D. The protein sequences of the recombinant proteins are given in Appendix Fig. S2E. GADD34Δ1-240 expression plasmid has been described (Gurzeler et al, 2022); briefly, for protein preparation, bacterial strain BL-21 (DE-3) (pLysS) was transformed with the plasmid, grown in 1000 ml Luria broth at 37 °C until the OD600 reached 0.4–0.6. After induction (1 mM IPTG), bacteria were cultured for another 4 h and cells were harvested. Expression of recombinant GADD34Δ1-240 was validated on Coomassie gel. The cell pellet was resuspended in 50 ml lysis/wash buffer (50 mM Tris-HCl pH 7.5, 300 mM NaCl, 5% glycerol, 25 mM imidazole), lysed by sonication with 240 × 1 s pulses at 40% amplitude on ice and centrifuged at 18,000×g for 20 min at 4 °C. The supernatant was mixed with 1 ml HisPur Ni-NTA washed resin slurry (Cat. No. 88222, Thermo Fisher Scientific) for at least 2 h at 4 °C under continuous rotation. Unbound proteins and background were removed by washing the resin 2× with 50 ml of lysis/wash buffer, then bound proteins were eluted 2× with 3 ml elution buffer (lysis/wash buffer supplemented with 400 mM imidazole). The proteins were dialysed in protein reconstitution buffer (30 mM NaCl, 5 mM HEPES pH 7.3) overnight using a Slide-A-Lyzer Dialysis Cassette (MWCO 10 K) (Cat. No. 88400TS, Thermo Fisher Scientific) at 4 °C under agitation. Protein concentration was quantified with a nanodrop device.

Reporter DNA templates were generated by PCR amplification from the cloned reporters (forward primer: GCGCGTTGGCCGATTCATTAATGC; reverse primer: T₁₀₀GGGAGCTCGCCCCCTCGGAG). In vitro transcription was carried out for 2 h using the MEGAscript® T7 Transcription Kit protocol (AM1334, Invitrogen). Capping was carried out co-transcriptionally using the CleanCap Reagent AG (3′ Ome) (N-7413, TriLink) and the Yeast Inorganic Pyrophosphatase (Cat. No. M2403S, New England BioLabs) following the CleanCap protocol. The capped RNAs were treated for 15 min at 37 °C with 0.14 U/μl Turbo DNase (AM2238, Invitrogen), then purified using the Monarch RNA Cleanup Kit (T2040L, New England Biolabs). The quality and size of the produced RNAs were assessed using an Agilent Fragment Analyzer.

About 6.5 µl of translation-competent extracts were freshly supplemented with 0.4 mM amino acids (L4461, Promega), 15 mM HEPES pH 7.3, 6 mM Creatine Phosphate, 102 ng/ml Creatine kinase, 28 mM potassium acetate, 1 mM MgCl₂, 24 mM KCl, 1 U/µl RNasin® Plus Ribonuclease Inhibitor (N2611, Promega) and 2 fmol/µl of Renilla luciferase reporter in a final volume of 12.5 µl prior to each in vitro translation assay. Unless specified, 16 ng/µl of recombinant GADD34Δ1-240 was included in the supplemented extracts. About 30 fmol/µl of firefly luciferase reporter were used per reaction. The in vitro translation assays were carried out at 37 °C for 50 min. Luminescence activity was quantified using the Dual-Glo® Luciferase Assay System (E2940, Promega) and a Tecan Safire2™ plate reader.

NIH/3T3 cells were first stably transduced with the lentiviral dual-luciferase reporters containing the 5′ UTRs to be assayed (upstream of firefly luciferase) and Renilla luciferase (internal normalisation), bidirectionally driven by the Pgk1 promoter. After establishing the reporter-expressing bulk cell population and several passages to ensure stable expression, cells were transduced with the shRNA lentiviruses and selected on puromycin for 4 days before lysis in 1× passive lysis buffer (E1941, Promega) and dual-luciferase readout using the Dual-Glo® Luciferase Assay System (E2940, Promega) and a Tecan Safire2™ plate reader.

One 15-centimetre dish of confluent NIH/3T3 cells was used per replicate. Ribosome profiling and parallel RNA-seq were performed in triplicate for cells transduced with Denr shRNA, Eif2d shRNA1, Eif2d shRNA2, con1 shRNA and con2 shRNA, closely following our previously reported protocol (Castelo-Szekely et al, 2019) with minor modifications as described below. Cells were washed in cold PBS (without cycloheximide), harvested by scraping down in a small volume of PBS (without cycloheximide), pelleted by brief centrifugation, and then flash frozen. Cell lysates were prepared as described (with cycloheximide) (Castelo-Szekely et al, 2019). For the generation of ribosome-protected fragments, cell lysates were treated with 5 units Ambion™ RNase I (Cat No. AM2295, Invitrogen) per OD260, and monosomes were purified on MicroSpin S-400 columns (GE Healthcare). For both Ribo- and RNA-seq, 2 µg of RNA was used for rRNA depletion following the RiboCop rRNA Depletion Kit V1.2 (Cat. No. 037.24, Lexogen). Finally, libraries were amplified using 10-14 PCR cycles, and sequenced (HiSeq2500 platform).

Sequencing reads were processed as described previously (Arpat et al, 2020). Briefly, sequencing adaptors were trimmed using Cutadapt (Martin, 2011), and the reads were size-filtered by a custom Python script (26–35 nt for RPF and 21–60 nt for RNA). The reads were then aligned subsequently to mouse rRNA, human rRNA, mouse tRNA and mouse cDNA from Ensembl mouse database release 100 using Bowtie2 (Langmead and Salzberg, 2012). In order to estimate expressed isoforms, the RNA-seq reads were mapped in parallel to the mouse genome using STAR mapper version 2.5.3a (Dobin et al, 2013) and processed with StringTie (Pertea et al, 2015) to measure FPKM for each transcript. A custom Python script classified the transcripts into single expressed isoforms or multiple transcript isoforms. The mapped reads were counted on 5′ UTR or CDS of the protein-coding genes using a custom Python script. The location of the putative A-site of RPFs were estimated as 5′ position +16 for reads shorter than or equal to 31 nt and +17 for reads longer than 31 nt. Transcripts for which there were less than 10 counts in total of all samples were filtered out of the further analysis. Read counts were normalised using the trimmed mean of M-values (TMM) from edgeR (Robinson and Oshlack, 2010), then transformed to transcripts per million (TPM) to normalise for transcript length. Translation efficiencies were calculated as the log₂-transformed ratio of normalised CDS footprint counts to normalised CDS RNA counts and averaged over replicates. All analyses were carried out on R version 4.2.1.

Ribosome profiling sequencing data were analysed using the deltaTE method (Chothani et al, 2019) in order to assess significant changes in translation efficiency and RNA abundance. The significance threshold was set to FDR <0.1.

For uORF annotation, transcripts were used for which reads could be unambiguously assigned to the 5′ UTR, i.e. transcripts that are the only protein-coding isoform expressed (single expressed isoform) (N = 6841). In order to include as many transcripts as possible for the annotations, genes with multiple protein-coding isoforms were also considered if all expressed isoforms had the same CDS start, in which case the transcript with the longest 5′ UTR was selected (N = 2425). For the selected transcripts, uORFs were annotated and considered as translated with the following criteria: (i) they started with AUG, CUG, GUG or UUG, (ii) they had an in-frame stop codon within the 5′ UTR, and (iii) they had a coverage of at least 33%. When several potential uORFs were overlapping, the one with the highest coverage (read count/uORF length) was considered.

The cell proliferation assay was performed using the Click-iT® EdU Flow Cytometry Assay Kits (C10424↗, Invitrogen) following the vendor’s protocol with some modifications. Briefly, NIH/3T3 cells transduced with con1 shRNA, con2 shRNA, Denr shRNA and Eif2d shRNA1 and shRNA2 were seeded into 6-cm plates under puromycin selection. After 4 days, the cells were treated for 1 h with 10 µM of EdU, washed with PBS, trypsinised and collected into FACS tubes. The cells were then centrifuged at 1200 × g for 5 min and resuspended in PBS. The cells were counted, and all conditions were adjusted to the same number of cells. After washing with 3 ml of 1% BSA in PBS, the cells were resuspended in 100 µl fixative buffer and incubated overnight at 4 °C. Another wash with 1% BSA in PBS followed by a 15 min incubation in 100 µl of 1×-saponin-based permeabilisation and wash reagent. About 200 µl of Click-iT reaction cocktail were added to each tube, and the cells were incubated for 30 min at room temperature. After a wash in 3 ml of 1% BSA in PBS, the cells were resuspended in 100 µl of 1×-saponin-based permeabilisation and wash reagent complemented with 5 µl of 20 mg/ml RNaseA and 2 µl of 1 mg/ml propidium iodide and incubated for 30 min at room temperature. Fluorescence was measured using the Accuri C6 FACS machine.

C-terminal tagging of endogenous DENR with 3×FLAG-NeonGreen-dTAG in NIH/3T3 cells was carried out using CRISPR-Cas9 homology-directed repair. 10⁵ NIH/3T3 cells were seeded in a six-well plate 24 h prior to transfection. About 0.9 µg of the pBLU donor plasmid (containing 400 bp 5′ and 3′ homology arms directly upstream and downstream of Denr stop codon) as well as 0.5 µg of the pCas9-gRNA and 0.1 µg of the control pRRE reporter were transfected into NIH/3T3 using Lipofectamine™ 3000 (L3000008, Invitrogen). The medium was changed after 24 h and cells were FACS-sorted based on mCherry and EGPF expression in a 10 cm dish 48 h post-transfection. Cells were counted and diluted to have 1 cell per well of a 96-well plate. When cell clones became visible, they were preselected based on NeonGreen expression, then split into a 96-well plate for genotyping and a 12-well plate for maintenance. For genomic DNA extraction, cells were washed with PBS and then lysed in 20 µl of DNA lysis buffer (10 mM Tris-HCl pH 8, 0.5 mM EDTA, 0.5% Triton X-100, 0.5 mg/ml proteinase K). After incubation for 1 h at 55 °C, cooling on ice and further incubation for 10 min at 95 °C to inactivate proteinase K, this prep was used to check the insertion of the cassette into the Denr locus by PCR using primers upstream of the 5′ and downstream of the 3′ homology arms.

Triplicate 15 cm plates of 80% confluent NIH/3T3 DENR-3×FLAG-NeonGreen-dTAG cells were treated with 500 nM dTAG-13 or control-treated for 5 h prior to harvesting. After washing with PBS, cells were scraped in 1 ml PBS (4 °C). Cells were collected by centrifugation at 500×g for 5 min at 4 °C and lysed in 400 µl IP buffer (50 mM Tris-HCl pH 7.5, 120 mM NaCl, 1% NP-40, 0.5 mM DTT, 10 mM MgCl₂, 1× PhosSTOP™ (4906837001, Roche), 1× cOmplete™ EDTA-free Protease Inhibitor Cocktail (Cat. No. 4693132001, Roche), 0.04 U/µl RNasin® Plus Ribonuclease Inhibitor (CN2611, Promega)) for 30 min on ice with mixing by pipetting every 10 min. Lysed cells were centrifuged for 7 min at 5000 × g at 4 °C, and the supernatant was collected. Proteins were quantified using Pierce™ BCA Protein Assay kit (Cat. No. 23225, Thermo Fisher Scientific) and 500 µl of 2 mg/ml proteins were incubated with washed anti-FLAG magnetic beads (Cat. No. M8823, Millipore) at 4 °C for 3 h on a rotating wheel. The flow-through was kept for Western Blot, and beads were washed twice in wash buffer (20 mM Tris-HCl pH 7.5, 100 mM NaCl, 0.5 mM DTT, 10 mM MgCl₂, 1× cOmplete™ EDTA-free Protease Inhibitor Cocktail (Cat. No. 4693132001, Roche), 0.04 U/µl RNasin® Plus Ribonuclease Inhibitor (N2611, Promega)) with change of tubes between the first and the second wash. Proteins were eluted in 80 µl elution buffer (NH₄OH pH 11–12) and incubated for 15 min at 4 °C. The elution step was repeated twice and eluates were pooled, frozen and dried in a SpeedVac system.

Samples were digested following a modified version of the iST method (named miST method). Briefly, dried material was redissolved in 50 µl miST lysis buffer (1% Sodium deoxycholate, 100 mM Tris pH 8.6, 10 mM DTT), heated 5 min at 95 °C and diluted 1:1 (v:v) with water. Reduced disulfides were alkylated by adding 0.25 vol. of 160 mM chloroacetamide (32 mM final) and incubating for 45 min at RT in the dark. Samples were adjusted to 3 mM EDTA and digested with 1.0 µg Trypsin/LysC mix (Promega #V5073) for 1 h at 37 °C, followed by a second 1 h digestion with an additional 0.5 µg of proteases. To remove sodium deoxycholate, two sample volumes of isopropanol containing 1% TFA were added to the digests, and the samples were desalted on a strong cation exchange (SCX) plate (Oasis MCX; Waters Corp., Milford, MA) by centrifugation. After washing with isopropanol/1% TFA, peptides were eluted in 200 µl of 80% MeCN, 19% water, 1% (v/v) ammonia, and dried by centrifugal evaporation. Tryptic peptide mixtures were injected on an Ultimate RSLC 3000 nanoHPLC system interfaced via a nanospray Flex source to a high-resolution Orbitrap Exploris 480 mass spectrometer (Thermo Fisher, Bremen, Germany). Peptides were loaded onto a trapping microcolumn Acclaim PepMap100 C18 (20 mm × 100 μm ID, 5 μm, Dionex) before separation on a C18 custom-packed column (75 μm ID × 45 cm, 1.8-μm particles, Reprosil-Pur, Dr. Maisch), using a gradient from 4 to 90% acetonitrile in 0.1% formic acid for peptide separation (total time: 140 min). Full MS survey scans were performed at 120,000 resolution. A data-dependent acquisition method controlled by Xcalibur software (Thermo Fisher Scientific) was used that optimised the number of precursors selected (“top speed”) of charge 2+ to 5+ while maintaining a fixed scan cycle of 2 s. Peptides were fragmented by higher energy collision dissociation (HCD) with a normalised energy of 30% at 15,000 resolution. The window for precursor isolation was of 1.6 m/z units around the precursor and selected fragments were excluded for 60 s from further analysis. Data files were analysed with MaxQuant 1.6.14.0 incorporating the Andromeda search engine. Cysteine carbamidomethylation was selected as a fixed modification, while methionine oxidation and protein N-terminal acetylation were specified as variable modifications. The sequence databases used for searching were the mouse (Mus musculus) reference proteome based on the UniProt database (www.uniprot.org↗, version of April 6, 2021, containing 55,341 sequences RefProt _Mus_musculus_20210604.fasta) supplemented with the sequences of common contaminants, and a database of mammalian immunoglobulin sequences (https://www.imgt.org↗ version of 20th Nov. 2018, 2243 entries). Mass tolerance was 4.5 ppm on precursors (after recalibration) and 20 ppm on MS/MS fragments. Both peptide and protein identifications were filtered at 1% FDR relative to hits against a decoy database built by reversing protein sequences. All subsequent analyses were done with the Perseus software package (version 1.6.15.0). Contaminant proteins were removed, and intensity iBAQ values were log₂-transformed. After assignment to groups, only proteins quantified in at least three samples of one group were kept. After missing values imputation (based on normal distribution using Perseus default parameters), t-tests were carried out among all conditions, with permutation-based FDR correction for multiple testing (Q-value threshold <0.05). Log₂FC of non-treated Denr-3×FLAG-NeonGreen-dTAG vs. dTAG treated Denr-3×FLAG-NeonGreen-dTAG was calculated and the p value was corrected for multiple testing by false discovery rate (significance threshold was set to FDR <0.05).