Readthrough of stop codons under limiting ABCE1 concentration involves frameshifting and inhibits nonsense-mediated mRNA decay

Cell culture and transfection

HeLa cells were cultured in DMEM supplemented with 10% fetal calf serum (FCS) and antibiotics. Cells were grown in 5% CO₂ at 37°C. HeLa cells stably expressing TCRβ and the Ig-μ reporter genes were described previously (56,57). For readthrough experiments, cells were treated with 400 µg/ml G418 (Fisher Scientific) for 24 h before harvesting. SiRNA-mediated knockdowns were carried out by a double-transfection procedure. First, 3 × 10⁵ HeLa cells per well were seeded in six-well plates. One day later, the cells were transfected with 52 pmol of siRNA using Lullaby reagent (OZ Biosciences). After 2 days, the cells were re-transfected as before. Protein and total RNA were isolated after one additional day. SiRNAs with the sequence 5′-GAGGAGAGUUGCAGAGAUUU-3′, HSS10985 and HSS10986 (Invitrogen) for targeting ABCE1 and 5′-GAUGCAGUUCCGCUCCAUU-3′ for targeting UPF1 were used. Readthrough assays were performed using the p2luc plasmids as described previously (25). Briefly, 3 × 10⁵ HeLa cells per well were seeded in six-well plates. One day later, the cells were transfected with 52 pmol of siRNA using Lullaby reagent (OZ Biosciences). After 2 days, the cells were re-transfected with 52 pmol of siRNA and 800 ng of the indicated plasmids using Lipofectamine 2000 (Invitrogen). Protein and total RNA were isolated after two additional days. p2luc plasmids to assess readthrough combined with frameshift (p2luc + 1) were generated by site mutagenesis introducing an adenosine downstream the stop codon (UAA, UAG or UGA) between the two luciferases ORFs, and experiments with these new reporters were performed as before.

Knockdown efficiencies of the different targets were verified by western blotting using anti-ABCE1 (Abcam, ab185548), anti-Tyrosine Tubulin (Tub1A2) (Sigma, T9028), anti-RPS3 (Bethyl A303-840A), anti-RPL4 (proteintech, 11302-1-AP), anti-RACK1 (bethyl A302-545A), anti-RPS27A [(58) a kind gift of U. Kutay], anti-GFP (SicGen, AB0020-500) and anti-CPSF100 (custom made) antibodies. For IgM-μ protein detection, anti-IgM-μ (Jackson Imm. 115-005-020) was used in combination with Pierce Western Blot Signal Enhancer kit (Thermo Scientific).

Luciferase assay for readthrough quantification

Dual luciferase assays were performed according to the manufacturer's protocol (Promega). Luminescence was measured using TECAN Infinite M1000, two technical replicates for each sample. Readthrough efficiency was calculated by comparing the ratio of Firefly to Renilla luciferase activity relative to the plasmid without stop codon.

Puromycin incorporation assay

Translation level analysis was performed using an adapted protocol from a previous study (59). Briefly, HeLa cells were labeled with 10 μg/ml of puromycin (Santa Cruz) for 10 min in complete medium. After 30 min in puromycin free medium, cells were harvested and puromycin incorporation was detected by immunoblotting using a specific anti-puromycin-12D10 antibody (Millipore, MABE343). The western blot signal was quantified using the ImageJ program and displayed as mean ± standard deviation (SD). Student's t test was used to determine P-values.

Detection of out-of-frame readthrough product

ABCE1 knockdown using HSS10985 siRNA was performed as above. On day 4, cells were co-transfected with HSS10985 siRNA and pcDNA5/FRT/TO plasmid encoding for eGFP and C-terminal FLAG tag in 0 frame and SPOT tag in +1 frame using Lipofectamine2000. 24 h later, cells were harvested and resuspended in 500 μl lysis buffer (150 mM NaCl, 50 mM HEPES, pH 7.4, 0.5% Triton X-100, protease inhibitors). Cell lysate was sonicated 3 × 10 s, amp 45 (Vibra-Cell sonicator) and clarified by centrifugation (16 100 × rcf for 10 min at 4°C). The supernatant was incubated either with anti-SPOT beads (Spot-Trap magnetic agarose, Chromotek), or anti-FLAG beads (M2 antibodies coupled to Dynabeads M-270 Epoxy) to immunoprecipitate the out-of-frame or in-frame readthrough products, respectively. eGFP was immunoprecipitated using GFP-Trap magnetic agarose (Chromotek). In all cases, bound proteins were eluted in protein loading buffer and analyzed by western blotting using the indicated antibodies.

RNA manipulations and analysis

RNA was extracted using TriReagent according to the manufacturer's instructions. RT-qPCR analyses were performed using Brilliant III Ultra-Fast SYBR^® Green (Agilent) after DNase treatment using Turbo DNase (AMBION). All primers used in this study are listed in Supplementary Table S1. Data are displayed as mean ± SD. Student's t test was used and n values are indicated in the respective figure legends. GraphPad Prism v8 was used to create plots. For the analysis of total RNA by RNA-seq, rRNA-depleted mRNA was purified and used for library preparation using TruSeq Stranded mRNA Library Prep Kit (Illumina) and sequenced with Illumina HiSeq2500. Northern blot analysis was performed according to an adapted protocol from a previous study (60). Briefly, ∼10 μg RNA per sample was separated on a 1.2% (w/v) agarose gel containing 1× MOPS and 1% (v/v) formaldehyde and transferred to a positively charged nylon membrane by wet transferring. The ³²P-labeled ribo-probe was hybridized in ULTRAHyb buffer (AMBION) at 68°C overnight. The ribo-probes were transcribed from a NotI digested plasmids using SP6 polymerase (Fisher) in the presence of α³²P-UTP. The ribo-probe sequences are listed in Supplementary Table S1.

Polysome fractionations

The polysome profiling protocol was adapted from (61). Briefly, HeLa cells, with or without knockdown (two 150-mm culture dishes per condition) were treated with 100 μg/ml cycloheximide (Focus biomolecules) for 4 min at 37°C. Cells were then lysed in low-salt buffer (10 mM Tris–HCl [pH 7.5], 10 mM NaCl, 10 mM MgCl₂, 1 mM DTT, 1% Triton X-100, 1% Na-deoxycholate) supplemented with RNAse Inhibitor (NxGen), Protease Inhibitor cocktail (Biotools) and 100 μg/ml cycloheximide by quickly vortexing. After 2 min on ice, lysates were cleared by centrifugation for 5 min at 16 000 g. Lysates were loaded on 15–50% sucrose gradient tube and centrifuged at 40 000 rpm in a Beckman SW-41Ti rotor for 2 h at 4°C. Gradients were fractionated and monitored at absorbance 254 nm with a fraction collector (BioComp Instruments) at a 0.2-mm/s EM1 speed, with a distance of 3.71 mm per fraction. RNA was precipitated from gradient fractions with 3 volumes of 100% ethanol (overnight at −80°C). The next day, samples were centrifuged for 15 min at 16 000 g (4°C). Resulting pellets were reconstituted in 1 ml of TriReagent. RNA was precipitated from the aqueous fraction with isopropanol (–20°C, overnight). Proteins were precipitated from gradient fractions with 0.1 volume of 100% TCA and 1 volume of acetone (overnight at −80°C). The volumes of the monosome and disome fractions used for the protein analysis shown in Figure 6 were adjusted based on the OD 260 nm values (rRNA) to achieve equal overall protein amounts. The next day, samples were centrifuged for 15 min at 16 000 g (4°C). Resulting pellets were additionally washed 3 × 1 ml ice-cold acetone, followed by centrifugation for 5 min at 16 000 g (4°C). After the last wash, pellets were dried in a vacuum concentrator (Eppendorf, 15 min, no heating, no brake, V-HV vacuum high vapor program). Prior to mass spectrometric analysis, dried pellets were reconstituted in 8 M urea, 100 mM Tris pH 8, reduced, alkylated and digested with trypsin. For the western blot, pellets were reconstituted in 1.5× LDS, 50 mM DTT loading buffer.

AHA-labeling click-it experiment

Newly synthesized proteins were labeled with 40 µM l-azidohomoalanine (AHA, ThermoFisher C10102) under ABCE1 knockdown or control condition for 90 min in methionine-free medium at 37°C. Before starting the labeling, cells were pre-incubated in methionine-free medium for 30 min at 37°C. Labeled proteins were assessed using tetramethylrhodamine (TAMRA) protein analysis detection kit (ThermoFisher C33370). Labeled proteins were analyzed using SDS-PAGE and direct imaging using a gel doc system equipped with appropriate fluorescent filter. Additionally, to control for an equal loading, the gel was subsequently stained with SYPRO Ruby protein stain (ThermoFisher S12001). For the mass spectrometric analysis, newly synthesized labeled proteins were captured using Click-iT Protein Enrichment Kit (ThermoFisher C10416). Captured AHA-labeled proteins were digested with trypsin and characterized using mass spectrometry. All the procedures including AHA labeling and click-it chemistry were carried out accordingly to the manufacturer's instructions.

Ribosome profiling experiments

Cells for ribosome profiling and RNA sequencing were processed using an adapted protocol from previous publications (62,63). Cells were washed with ice-cold phosphate buffered saline (PBS) and flash-frozen in liquid nitrogen. Subsequently, cells were scraped and lysed in lysis buffer (20 mM Tris–HCl pH7.4, 150 mM NaCl, 5 mM MgCl₂, 1% Triton X-100, 1 mM DTT, 25 U/ul Turbo DNase, Turbo DNase buffer) on ice. Next, cells were triturated ten times through a 27-gauge needle of a syringe and clarified by centrifugation. For ribosome profiling, lysates (about 4 U₂₆₀) were subsequently treated with 200 U RNase I (Ambion) for 10 min at 23°C and shaking at 300 rpm. The digestion was stopped by addition of 100 U SUPERase In RNase inhibitor (Ambion). Monosomes were separated on Illustra Micro-Spin S-400 HR gel filtration columns (GE Healthcare Life Science) as previously described (62). TriReagent was added immediately to the eluates and samples were stored at −80°C until further processing. RNA was isolated according to the TriReagent protocol and separated on 15% Novex polyacrylamide gels (Invitrogen). Ribosome footprints were excised between 26 and 34 nucleotide RNA size markers. After RNA isolation and purification, rRNAs were removed using the RiboZero kit (Illumina) according to manufacturer's datasheet. Sequencing libraries from ribosome footprints were generated as previously described (62). RNA-seq libraries were prepared from cells lysed similar as the ribosome profiling samples. RNA was isolated from cleared lysates by addition of TriReagent as for the ribosome-protected fragments (see above). Total RNA was used for library generation with the TruSeq Stranded mRNA Library Prep Kit (Illumina) according to the manufacturer's instructions. Libraries were sequenced on an Illumina HiSeq2500 generating 100 nt single-end reads.

RNA-sequencing analysis

The human transcriptome annotation was downloaded from the Ensemble database, using version GRCh38.p10. Both transcripts annotated as coding and as non-coding RNAs were included in the reference. Reads were mapped to the reference transcriptome and transcript abundance was quantified using Salmon (64). Estimated transcript counts were imported into R using the tximport package (65). Transcripts with more than eight counts in at least three independent samples were kept for further analysis. For correlation and plotting, transcript counts were normalized by scaling to the mean of total counts across samples, then log₂-transformed after adding a pseudocount of 8. EdgeR was used for differential expression analysis (66), with raw, non-normalized counts as inputs. A likelihood-ratio test was used to determine differentially expressed genes.

Ribosome profiling analysis

The raw reads of the ribosome profiling experiments were trimmed based on quality and sequence adapters removed with Cutadapt v.1.8 (67). Reads with the expected read length (16–35 nt for the ribosome footprint (RPF) and 35–60 nt for total RNA) were kept for further analysis. Reads mapping Homo sapiens ribosomal RNA (rRNA) and transfer RNA (tRNA) databases (ENSEMBL v91) (68) using to bowtie2 (v.2.3.4.1) (69) were excluded. The remaining reads were aligned against H. sapiens transcripts database (ENSEMBL v91) using bowtie2 (v.2.3.4.1). Multi-mapping reads were filtered out and the remaining reads summarized at a gene level using an in-house script. Ribosome density (also called translational efficiency) was calculated in R (v.3.6.1). Raw genes ribosome footprints and total RNA counts were normalised using the DESeq2 package (70). Functional enrichment analysis (i.e. KEGG) was performed using the R packages biomaRt (71) and reactomePA (72). Read distribution analysis around termination codons was performed at transcript level, following a recently described strategy (73). To minimize the probability of falsely assigning RPFs originating from overlapping CDSs to stop codons and 3′ UTRs of related transcript isoforms of the same gene, one transcript was selected for each gene by applying the following selection criteria consecutively: (i) protein-coding genes must not overlap; (ii) select transcripts with the most downstream stop codon; (iii) select transcripts with the highest APPRIS score (74); (iv) for the few still non-unique cases (874 out of 12401 expressed genes), select one randomly. Heat-maps were performed using the ChIP-Extract tool from the ChIP-Seq toolkit (75). Subsequently, the 5′-end of each RPF read for the selected transcripts was shifted 15 bases downstream to represent the ribosome A site. Selected transcripts were then sorted according to the total number of ribosome A sites mapping in the stop codon, or 100 nucleotides downstream the termination codon.

Knockdown of the ribosome recycling factor ABCE1 renders many endogenous NMD targets immune to NMD

Based on previous work indicating a link between translation termination and NMD [reviewed in (3)], we hypothesized that interfering with translation termination would alter the abundance of many mRNAs, in particular NMD-sensitive transcripts. To test this hypothesis and identify the affected genes, we established an siRNA-based knockdown protocol that enabled us to reduce the protein level of the ribosome recycling factor ABCE1 to ∼10% in HeLa cells (Figure 1A). In three biological replicates, total RNA isolated from cells with an ABCE1 knockdown (ABCE1 KD) or cells with a control knockdown (CTRL KD) was depleted for rRNA, reverse transcribed and subjected to high throughput Illumina sequencing (RNA-seq) to assess the steady-state levels of mRNAs transcriptome-wide. Comparison of the three ABCE1 KD datasets among each other and with the three Ctrl KD datasets revealed an overall high correlation (Pearson correlation coefficients between 0.78 and 0.91), demonstrating not only good reproducibility among the biological replicates, but also indicating that ABCE1 KD does not cause a global change in RNA levels (Supplementary Figure S1). Of the totally detected 47 830 transcripts, 3722 (7.8%) changed by 2-fold or more and with a P-value ≤ 0.05 (Figure 1B). The detected changes occurred to a similar extent in both directions, with 1757 up- and 1965 down-regulated transcripts. Notably, prolonged depletion of ABCE1 reduces cell proliferation based on observing ∼30% less cells in cultures transfected with ABCE1-targeting siRNAs compared to control siRNAs.

To assess the effect of ABCE1 KD on NMD-sensitive transcripts, we used a previously identified high-confidence set of endogenous NMD targeted genes (5). This list comprises about 1000 NMD targets that were identified based on their RNA level increase in UPF1, SMG6 and SMG7 knockdowns, as well as their decrease in the respective rescue experiments (5). Of the 644 high-confidence NMD targets that we could detect in our analysis here, 109 (16.9%) changed significantly (P-value ≤ 0.05) by 2-fold or more upon ABCE1 KD, with the vast majority (96 transcripts) being upregulated (Figure 1B, red and green dots). Thus, compared to non-NMD targets, the NMD-sensitive transcripts are overrepresented by a factor of 4 among the transcripts that increased in abundance upon ABCE1 KD, suggesting that ABCE1 KD inhibits NMD on a significant fraction of the NMD-sensitive transcripts.

To validate RNA-seq results, we used reverse transcription followed by quantitative PCR (RT-qPCR) to measure the relative RNA levels of a panel of well-characterized NMD-sensitive transcripts in total RNA prepared from HeLa cells with a knockdown of ABCE1, or a knockdown of the central NMD factor UPF1 as a positive control. For HNRNPL, TRA2B, SRSF11 and TMEM208, the NMD-sensitive transcript (labeled _NMD; Figure 1C) arises from an alternative splicing event (intron retention, inclusion of a poised exon or usage of an alternative 5′ or 3′ splice site) as part of an autoregulatory feedback loop and the productively spliced form of the respective pre-mRNA encodes for the full length protein and is not targeted by NMD (labeled _PROT) (76). We included RNA measurements of the _PROT splice forms of these genes as a control and to exclude that the increases observed for the _NMD splice forms were due to changes at the level of transcription or splicing. The RT-qPCR results were in good agreement with the RNA-seq data, confirming the upregulation of a fraction of the NMD-sensitive transcripts (GADD45B, HNRNPL_NMD and TMEM208_NMD) in ABCE1 KD, while another fraction of NMD-sensitive transcripts (TRA2B_NMD and SRSF11_NMD) was not affected and SMG5 was weakly but not statistically significantly upregulated. Altogether, these data indicated that ABCE1 KD inhibits NMD on a sub-population of NMD-sensitive transcripts, but none of the known NMD-inducing features could readily explain why certain NMD targets were up-regulated by ABCE1 KD while others remained unaffected. To exclude that the observed changes arose from an off-target effect of our siRNA, we repeated the ABCE1 knockdown with two additional siRNA sequences (ABCE1 85 and 86, respectively) and obtained essentially the same result (Supplementary Figure S2).

ABCE1 knockdown also inhibits NMD of well-studied NMD reporter constructs

Since only a sub-population of endogenous NMD targets were upregulated by ABCE1 KD, we wondered whether widely used NMD reporter constructs were also affected by ABCE1 depletion. Therefore, we performed the ABCE1 knockdown in HeLa cells that stably express either of two well-characterized NMD reporters, namely the TCRβ (57) or immunoglobulin-μ (Ig-μ) minigenes (56). Of both reporter genes, there exists a PTC-free (so called ‘wild-type’) version that produces an mRNA which is not sensitive to NMD (TCRβ WT and miniμ WT), and a version harbouring a PTC in the middle of the coding sequence (TCRβ ter68, miniμ ter310) that strongly triggers NMD (Figure 2A, B, top). Again, our siRNA-based knockdown protocol resulted in a reduction of ABCE1 protein to about 10% of the normally present amount (Figure 2A, B, bottom). Measurements of the respective relative reporter RNA levels by RT-qPCR showed that both NMD-sensitive transcripts increased by 2.5-fold (TCRβ ter68) and 8-fold (miniμ ter310) upon ABCE1 KD, whereas the PTC-free versions remained unchanged (miniμ WT) or even slightly decreased (TCRβ WT) (Figure 2C, D; see also Supplementary Figure S2B, D for miniμ), indicating that NMD of these two reporter transcripts was inhibited by ABCE1 KD. Hence, we reasoned that these reporters would be useful to further investigate how ABCE1 inhibited NMD.

ABCE1 knockdown inhibits NMD before the induction of RNA degradation

The observed NMD inhibition of the two NMD reporters TCRβ ter68 and miniμ ter310 prompted us to further investigate at which point in the NMD pathway this inhibition occurs. In yeast, depletion of the ABCE1 homolog Rli1 has been shown to lead to the accumulation of a 3′ RNA decay fragment of an NMD-sensitive mRNA as a consequence of a defect in translation termination (77). The authors of this study speculated that in the absence of Rli1, stalled ribosomes at the premature stop codon would block the 5′-3′ exonuclease Xrn1, resulting in the accumulation of this 3′ RNA decay intermediate. To see whether ABCE1 KD would also cause the appearance of a similar decay intermediate in human cells, we examined by northern blot WT and PTC-containing TCRβ and miniμ transcripts expressed in HeLa cells (Figure 2E, F). For both transcripts, we used probes that hybridize to the RNA downstream the PTC, which allows the detection of both the full-length transcript and the putative 3′ fragment (Figure 2A, B, top). While the NMD-insensitive WT versions of TCRβ and miniμ mRNAs did not change significantly as expected, we detected a marked increase of the full-length transcripts of the NMD-sensitive versions (TCRβ ter68 and miniμ ter310) upon ABCE KD (Figure 2E, F). However, in contrast to yeast, no 3′ decay intermediates were detected, indicating that in human cells, ABCE1 depletion inhibits NMD at an earlier step, before the initiation of RNA degradation, which usually occurs by a SMG6-mediated endonucleolytic cleavage.

NMD-sensitive mRNAs are strongly associated with polysomes in ABCE1-depleted cells

Since NMD is dependent on active translation (78,79), we next examined whether ABCE1 depletion might inhibit NMD indirectly by causing a global translation inhibition. Notably, a previous study has reported that ABCE1 KD in human cells reduces the general translation level (80). To assess the overall translation activity in control and ABCE1 KD cells, we pulse-chased HeLa cells with puromycin and detected puromycylated nascent polypeptide chains using an antibody against puromycin (Figure 3A). The western blot results revealed a decrease of total protein synthesis by ∼40% in cells depleted of ABCE1. However, this overall reduction in protein synthesis is unlikely to account for the selective inhibition of NMD on a fraction of the NMD targets, and moreover the puromycylation assay does not address which step of translation is affected by ABCE1 KD. There is evidence that ABCE1 depletion causes ribosome stalling at termination codon and an increased ribosome occupancy in the 3′ UTR, suggesting mainly an impairment of translation termination (44,45,47). According to this model, mRNAs are expected to still be associated with polysomes upon ABCE1 KD, and strong association of mRNAs with polysomes would provide indirect evidence that these mRNAs are still actively translated. To test this hypothesis, we knocked down ABCE1 in HeLa cells (Figure 3B, top) and performed polysome profiling (Figure 3B, bottom) to assess the association of different transcripts with monosomes (M) and polysomes (P) (Figure 3C). The relative distribution between the monosome and the polysome fractions were quantified by RT-qPCR for two NMD-sensitive (HNRNPL_NMD, TMEM208_NMD) and two NMD-insensitive (HNRNPL_PROT, TMEM208_PROT) endogenous transcripts. Remarkably, compared to the control, the association of the NMD-sensitive transcripts HNRNPL_NMD and TMEM208_NMD with polysomes was not decreased under ABCE1 KD, similar to the protein coding isoforms that are actively translated under control and KDs conditions. We conclude that, even though ABCE1 KD reduces overall the translation rate in cells, this cannot account for the inhibition of NMD on these transcripts, since they are still actively translated as judged by their association with polysomes.

Depletion of ABCE1 induces ribosome stalling at termination codon and an enrichment in 3′ UTR ribosome occupancy

To gain more insight about translation changes and ribosome occupancy at TCs in the absence of ABCE1, we performed ribosome profiling in HeLa cells in which ABCE1 was knocked down as described before (Supplementary Figure S3A) using a modified version of the protocol published by Ingolia and coworkers (63). In contrast to commonly used ribosome profiling protocols, the use of cyclohexemide (CHX) was avoided, because CHX does not prevent disassembly of ribosomes located at TCs. Instead, cells were flash-frozen with liquid nitrogen, which has been shown to block elongating ribosomes as well as ribosomes located at TCs (21). Moreover, the use of CHX can bias the final output of ribosome footprints, as it seems to allow slow, concentration-dependent elongation prior to lysis according to recent studies (81,82). The experiment was conducted in triplicates with ABCE1 KD and Control KD cells and the mapped reads showed overall strong 3 nucleotide periodicity, confirming that most of them indeed represent ribosome-protected RNA fragments (Supplementary Figure S3B). Ribosome occupancy was examined by dividing the transcripts into four regions (5′ UTR, CDS from the start codon to the middle, CDS from the middle to the TC, and 3′ UTR) and aligning ribosome-derived reads to each specific region, relative to the total number of reads (Supplementary Figure S3C). While ribosome occupancy in the 5′ UTR was low and very similar between ABCE1 KD and Control KD, a slightly lower ribosome coverage in the CDS was detected in ABCE1 KD compared to Control KD (Supplementary Figure S3C), indicating overall reduced translation and confirming the results of the puromycylation experiment (Figure 3A). On the contrary, on average more ribosomes appear to be present in 3′ UTR in the ABCE1 KD compared to Control KD, indicating that depletion of ABCE1 increases the rate of TC readthrough and/or translation re-initiation. These results are in agreement with other studies that have observed ribosome stalling at TCs and an increased ribosome occupancy in the 3′ UTR upon ablation of ABCE1 (44,45,47).

To minimize the probability of falsely assigning ribosome-protected fragments (RPFs) that originate from overlapping CDSs of alternative transcript isoforms to the TC and the 3′ UTR, we followed a recently published strategy (73) and chose a single transcript for genes with multiple annotated coding transcripts. Initially, all transcripts with the 3′-most TC were selected and for genes with multiple transcripts sharing this TC, the primary APPRIS transcript was selected (74). This second filtering step resulted in one transcript for almost all genes (99.3%). For the remaining 874 genes whose transcripts did not have an APPRIS score, one transcript isoform was selected randomly. This approach resulted in a total of 12 401 transcripts which were used for further analysis. To obtain more detailed information about ribosome occupancy at TCs and in the 3′ UTR, we performed a metagene analysis of the RPFs in the range between 300 nucleotides upstream of the TC and 100 nucleotides downstream of the ABCE1 KD and the Control KD (Figure 4A, B, top). While average ribosome occupancy at the TC seemed to be the same for ABCE1 KDs and Control KDs, an increased ribosome occupancy in the ABCE1 depleted cells was observed 30 nucleotides upstream of the TC and in the first 100 nucleotides of the 3′ UTR (Figure 4A, B, top), suggestive of disome accumulation at TCs and increased readthrough, respectively. As revealed by the heatmap, ribosome occupancy, and by inference translation of the ∼12 000 transcripts, was affected differently by the depletion of ABCE1 (Figure 4A, B, bottom). When ordered according to ribosome coverage at the TC in ABCE1 KD (Figure 4A, bottom), ∼25% of the transcripts exhibited marked ribosome stalling at the TC compared to Control KD (top section labeled by green line), while more than half of the transcripts showed decreased ribosome occupancy throughout the CDS and at the TC, gradually decreasing to a level at which no ribosomes were detected anymore at the TC (bottom section labeled by red line). The combination of these two effects explains why in the metagene analysis, no difference in ribosome occupancy at the TC was visible between ABCE1 KD and Control KD. By contrast, the increased ribosome occupancy in the 3′ UTR in the metagene analysis was also clearly visible when individual transcripts were analyzed and ordered according to ribosome coverage in the 3′ UTR in the ABCE1 KD dataset (Figure 4B, bottom). Interestingly, of the 2000 transcripts with the most ribosomes in the 3′ UTR (Figure 4B, top section labeled by yellow line), one quarter also showed a concomitant higher ribosome coverage in the CDS. Therefore, to exclude that the increase of ribosome coverage in the 3′ UTR was simply due to a general high engagement of ribosomes with these specific transcripts, the ribosome-protected reads mapping in the 3′ UTR were normalized to the average ribosome occupancy in the CDS (Figure 4C). Confirming our previous notion, ABCE1 KD exhibited elevated ribosome occupancy in the 3′ UTR, even when scaled to ribosome occupancy in CDS, supporting the hypothesis that ABCE1 depletion might create an environment favourable for readthrough or re-intiation events. To exclude that the observed differences in ribosome engagement simply correlated with mRNA abundance, total RNA-seq data from the cells used for the ribosome profiling experiments were analyzed (Supplementary Figure S3D). When all transcripts together were assessed, no overall change in mRNA levels was detected between ABCE1 KD and CTRL KD conditions (Supplementary Figure S3D, All) and the same was true when different sub-populations of transcripts were analyzed separately: the 2000 transcripts presenting the highest ribosome occupancy at the TC in ABCE1 KD cells (High Stop; group labeled with green line in Figure 4A), the 2000 mRNAs showing the lowest ribosome occupancy at the TC (Low Stop; group labeled with red line in Figure 4A), and the 2000 transcripts presenting the highest ribosome occupancy in the 3′ UTR in ABCE1 KD cells (High 3′-UTR; group labeled with yellow line in Figure 4B). This analysis shows an overall similar pattern of mRNA expression changes between the different groups (Supplementary Figure S3D, left). Moreover, the mRNA levels of transcripts with a low ribosome occupancy at TC and in the CDS (Low Stop) are similar to all transcripts considered in the analysis in the ABCE1 KD condition (All), demonstrating that their low engagement with ribosomes is not simply due to a low abundance of these transcripts (Supplementary Figure S3D, right).

TC readthrough has been reported previously as a mechanism that can suppress NMD (36). Therefore, we checked where the previously identified endogenous NMD-sensitive transcripts (5) were located in the heatmaps shown in Figure 4A and B. When ribosome occupancy at the TC was considered, NMD targets were distributed homogeneously throughout the entire heatmap, being present in both the top (green) as well as the bottom (red) group (Figure 4A). It therefore seems that ribosome stalling in the absence of ABCE1 occurs neither preferentially nor specifically on NMD targets, consistent with the observation that the mRNA levels of only a fraction of NMD targets was upregulated in ABCE1 KD (Figure 1).

We next tried to search for those NMD targets upregulated in ABCE1 KD (identified in the RNA-seq experiment, Figure 1B) in the green and red groups of transcripts, but because several of them share most of their sequence with the corresponding protein-coding isoform, unambiguous transcript identification was not possible with the exception of the NMD-sensitive SMG5-encoding mRNA, which consists of a single isoform: It belongs to the top 2000 transcripts with a high ribosome occupancy in the 3′ UTR (Figure 4B, section labeled by the yellow line). To still look at NMD targets separately and compare it to other transcripts that are insensitive to NMD, the ribosome occupancy in 3′ UTRs, relative to CDS, was analyzed as before (Figure 4C), this time comparing the NMD targets against the rest of the transcripts (Figure 4D). As already indicated by the heatmaps, higher ribosome occupancy in the 3′ UTR was detected in both transcript categories. On NMD targets, translating ribosomes in 3′ UTR can prevent NMD activation. However, since ribosome profiling does not inform if the ribosome-protected RNA fragments detected in the 3′ UTR originate from continued translation after a readthrough event, from ribosomes that have re-initiated translation, or if they represent translationally inactive ribosomes without any peptide chain associated, we addressed this question using reporter genes.

ABCE1 KD promotes in-frame and out-of-frame readthrough

We noted that among the endogenous NMD targets (5) that we detected in our RNA-seq, UGA was the most frequently occurring TC. Among these transcripts, the occurrence of the UGA TC was even further enriched among the sub-population of endogenous NMD targets that was upregulated upon ABCE1 KD: ∼82% of them have UGA as the stop codon compared to 55% of the endogenous NMD targets that were not upregulated under ABCE1 depletion. According to previous studies, UGA is the least efficiently recognized stop codon of the three canonical stop codons, resulting in the highest readthrough rates (83). Because of the overrepresentation of the UGA stop codon among our ABCE1-dependent NMD targets, and since insertion of readthrough-promoting retroviral RNA elements into NMD reporter transcripts has been shown to antagonize UPF1 binding to their 3′ UTR and subsequent RNA decay (36), we hypothesized that increased translational readthrough could suppress NMD on the NMD targets that were upregulated upon ABCE1 KD. Notably, increased readthrough upon ABCE1 KD would also be consistent with previous findings in yeast (84). To measure readthrough, we used a well-established readthrough reporter gene (25) in which the ORFs for Renilla luciferase (Rluc) and Firefly luciferase (Fluc) are positioned either in frame or separated by one of the three stop codons (Figure 5A). The Fluc/Rluc ratio serves as a measure to quantify readthrough of stop codons under different conditions. First, we tested readthrough efficiency in cells transiently transfected with the dual-luciferase reporter and treated with geneticin (G418), an aminoglycoside antibiotic that promotes readthrough (85). Consistent with previous reports (86,87), basal readthrough frequency was lowest with UAA and somewhat higher with UAG and UGA. After treatment of the cells with 400 μg/ml G418 for 24 h, readthrough increased at each of the three stop codons, with UGA being the most permissive, reaching about 2.5% readthrough (Supplementary Figure S4A). When we measured readthrough occurrence under Control KD and ABCE1 KD conditions (Figure 5B), we observed increased readthrough at all three stop codons upon ABCE1 depletion (Figure 5C). ABCE1 KD caused the strongest increase in readthrough at the UAA stop codon, with UAA and UGA reaching 0.4% and 0.5% readthrough, respectively, while UAG was read through with about 0.7% efficiency. This result demonstrates that ABCE1 KD increases readthrough frequency in HeLa cells, consistent with previous findings (83). That the readthrough frequency in ABCE1 KD increased to a similar level with all three TCs, while G418 increased readthrough at the different TCs to different levels (compare Figure 5C with Supplementary Figure S4A), is consistent with our TC usage analysis performed on the ribosome profiling data (Supplementary Figure S4B): The ratio of the three TCs was similar among the previously described sub-populations of transcripts (High Stop, Low Stop and High 3′ UTR; see Figure 4A, B and Supplementary Figure S3D).

Because the reads from the Ribo-seq that mapped to the 3′ UTR in the ABCE1 KD samples did not have a strong 3 nucleotide periodicity, we hypothesized that the readthrough observed under ABCE1 depletion might frequently be associated with a frameshift, causing the ribosome to ensue translation out of the original reading frame. To test if such frameshift-associated readthrough of the TC indeed occurs, we introduced into the dual luciferase reporter gene an additional nucleotide downstream of the TC that separates the Rluc CDS from the Fluc CDS (Figure 5D). In order to translate Fluc in this reporter, the ribosome either needs to read through the Rluc TC and concomitantly shift the reading frame, or ribosomes would have to re-initiate translation at the Fluc start codon. Upon knockdown of ABCE1 (Figure 5E), we observed an increase of readthrough in the +1 frame under ABCE1 depletion (Figure 5F). Notably, the Fluc/Rluc ratio with the frameshift constructs was higher than with the original dual luciferase reporters (Figure 5, compare C and F), which most probably is due to the altered TC context. While the TC is followed by GAT in the original reporters (Figure 5A), it is followed by AGA in the +1 constructs (Figure 5D), and it is known that the three sequence of the three nucleotides downstream of the TC influence how well the TC is recognized during translation (38). The occurrence of readthrough-coupled frameshifting could in fact explain the NMD inhibition for several of the endogenous NMD targets that are upregulated in ABCE1 KD (Figure 1B). The majority of them has multiple in-frame TCs closely downstream of the NMD-triggering TC, which would quickly stop ‘escaping’ ribosomes and prevent them from translocating far enough to clear the 3′ UTR of NMD factors. However, in most of the cases, one of the other two frames would allow ribosomes to proceed past the last exon-exon junction, where termination is expected to no longer trigger NMD. We therefore postulate that the readthrough events on the endogenous NMD targets frequently are accompanied by a frameshift of the ribosome.

To further assess if readthrough can explain the observed NMD inhibition under ABCE1 KD, we looked for readthrough in an NMD-sensitive transcript. The miniμ NMD reporter was ideal for this, because of a polyclonal antibody that with high affinity and specificity recognizes the constant region of the encoded mouse Ig-μ heavy chain. Western blotting was performed with lysates from cells stably expressing miniμ WT or ter310 mRNA and treated with control or ABCE1 siRNAs, or with G418 (Figure 5G). While full-length Ig-μ was readily detected in comparable amounts in miniμ WT-expressing cells under all conditions, as expected, truncated Ig-μ chains were detected in cells expressing miniμ ter310. As a consequence of increased miniμ ter310 mRNA due to NMD inhibition upon ABCE1 KD or G418 treatment, the amount of truncated Ig-μ chains increased (compare lane 5 with lanes 6 and 7). Most importantly, in miniμ ter310-expressing cells depleted for ABCE1 or treated with G418, full-length Ig-μ chains were also detected in addition to the truncated ones, indicating that under these conditions, a sizable fraction of the ribosomes read through the PTC and translated all the way down to the normal termination codon, thereby presumably displacing UPF1, EJCs and other NMD-inducing factors from the region downstream of the PTC. Consistent with the occurrence of frameshift-associated readthrough events, additional bands were detected specifically under ABCE1 depletion in miniμ ter310 expressing cells (Figure 5G, lane 6) which migrate slightly higher the truncated one. Based on their size, they most likely represent such out-of-frame readthrough events in which the ribosome runs into TCs downstream the PTC. Noteworthy, the detection of full-length Ig-μ chains in miniμ ter310-expressing ABCE1-depleted cells suggests that ABCE1 not only functions in splitting and recycling of the small and large ribosomal subunits, but that it also may play a role in polypeptide release during translation termination.

While the results obtained with the frame-shifted dual luciferase reporter (Figure 5D) suggested increased readthrough-coupled frameshifting (Figure 5F) in the absence of ABCE1, we could not exclude the possibility that the increased Fluc activity might result from translation re-initiation rather than readthrough. To obtain unambiguous evidence for frameshift-associated readthrough, we therefore tested an eGFP reporter construct encoding in the +1 frame a C-terminal SPOT tag (Figure 5H and Supplementary Figure S4C). After immunoprecipitation (IP) with an anti-SPOT nanobody from lysates of ABCE1-depleted cells, SPOT-tagged eGFP was easily detected by western blotting using an anti-eGFP antibody (Figure 5H; band labeled eGFP+S). A very weak signal for this frameshift-associated readthrough product was also seen in the IP of the control KD cells, and under both conditions, similar amounts of the much more abundant normally terminated eGFP protein co-precipitated unspecifically (band labelled eGFP*). Noteworthy, while we could detect out-of-frame readthrough with this reporter (Figure 5H), we did not detect any in-frame readthrough product after IP with anti-FLAG antibody (Supplementary Figure S4C, D), suggesting that the sequence context around the TC might affect in which frame translation continues past the TC.

Depletion of ABCE1 leads to ribosome collision on highly translated transcripts

Stop codon readthrough associated with frameshifting has not been reported so far, but recently it was shown in yeast that ribosome collisions in the CDS result in +1 frameshifting in the absence of No-Go Decay (55). According to their model, transcripts with high ribosome density are more sensitive to frameshifting events, because ribosome collisions occur more frequently on these transcripts. That ribosome collision is a marker for recognition of aberrant translation and activation of ribosome quality control (RQC) has also been strongly indicated by recent work characterizing the ubiquitin ligase ZNF598 as a sensor for collided ribosomes (51). We hypothesized that ribosome collisions might also be the mechanism causing the observed TC readthrough (with or without frameshifting) in the absence of ABCE1. Therefore, we scrutinized our ribosome profiling data for evidence of ribosome collisions at TCs or of the presence of disome formation at TCs. Although our ribosome profiling protocol involved the isolation of monosomes and hence cannot detect disomes, we observed in the trace of the metagene analysis and specific for the ABCE1 KD a ‘bump’ approximately 30 nucleotides upstream of the TC peak (Figure 4A, B, top), which represents the accumulation of a ribosome immediately adjacent to ribosome located at the TC, and the distance from the TC matches with that recently reported for collided ribosomes (49).

Since only a subgroup of transcripts showed a pronounced increase of ribosomes in the 3′ UTR under ABCE1 KD (Figure 4B, High 3′ UTR), we re-analyzed the ribosome occupancy around the TC of the 3000 transcripts with the highest ribosome occupancy in the 3′ UTR. Metagene analysis of the ribosome-protected footprints ranging between 300 nucleotides upstream and 100 nucleotides downstream of the TC of these transcripts revealed overall a higher ribosome occupancy in the CDS, at TC and in the 3′ UTR in the ABCE1 depleted cells compared to the control (Figure 6A). This increased ribosome occupancy under ABCE1 KD is consistent with the suggested ribosome stalling at the TC, which in turn might lead to ribosome queuing in front of the TC and occasional pushing of a ribosome past the TC, where it will continue translation into the 3′ UTR in any of the three frames. As expected for frequent out-of-frame readthrough events, the ribosome-protected footprints mapped to the 3′ UTR did not show a clear periodicity.

To test our hypothesis that under ABCE1 depletion highly translated transcripts were more prone to ribosome collisions at TC and ribosomes being pushed into the 3′ UTR, we determined the ribosome density for the transcripts used in Figure 6A by normalizing the ribosome-protected footprints for each gene relative to the mRNA abundance and CDS length. As suspected, transcripts with higher ribosome occupancy in the 3′ UTR also showed slightly higher ribosome density (Supplementary Figure S5B), suggesting that high ribosome density in the CDS could contribute to increase ribosome collision and more readthrough events. A caveat with Ribo-seq data is, that it cannot distinguish if the detected ribosomes are really actively translating or if they are in a translationally inactive state. To measure actual protein synthesis, we therefore also performed a metabolic labelling with a methionine analogue (l-azidohomoalanine, AHA) in control or ABCE1 KD cells. After the labelling, newly synthesized proteins were enriched and analyzed by mass spectrometry. Consistent with the puromycin incorporation assay (Figure 3A), ABCE1 depletion caused a reduction in newly synthesized proteins (Supplementary Figure S5C, TAMRA), but did not affect the overall protein levels of the cell (Supplementary Figure S5C, SYPRO). To assess if the transcripts with high ribosome occupancy in the 3′ UTR are also highly translated, we compared the Ribo-seq data with the results of the AHA-labeling experiment. For this comparison, we considered the top 1000 transcripts from the ‘High 3′ UTR’ group (Supplementary Figure S5A) and proteins that exhibited a high normalized spectral abundance factor (NSAF) in ABCE1 KD sample in the AHA-experiment. Since detection of peptides derived from specific protein isoforms by mass spectrometry is difficult, the analysis was performed at gene level, considering unique gene ID from the Ribo-seq transcripts with high ribosome occupancy into the 3′ UTR (n = 1000) and unique protein ID from the AHA-experiment (n = 1372). About the 33% of the genes (n = 334) with high ribosome occupancy in the 3′ UTR were also highly translated in ABCE1 depleted cells (Supplementary Figure S5D). Altogether, these results support the idea that it might be colliding ribosomes at TC that cause readthrough by pushing the stalled ribosome into the 3′ UTR, where it can ensue translation stochastically in presumably any of the three frames.

Activation of ribosome quality control (RQC) in ABCE1 depleted cells

While the detection of ribosome collision in vivo is technically very challenging, recent reports have pointed out a connection between ribosome collision and the activation of RQC. In particular, collided disomes were shown to be the structural entity recognized by the E3 ubiquitin ligase Hel2/ZNF598 (49,51). Ubiquitination of several ribosomal proteins was reported as a consequence of ZNF598 recruitment to aberrant translation events, activating the RQC cascade (52,54). One well-documented example is the ribsomal protein RPS3, which is ubiquitinated by ZNF598 when ribosomes stall on a poly(A) stretch (49,52,54). Moreover, the ribosome associated protein RACK1 also seems to play a crucial role in ZNF598-dependent RQC activation (54). To obtain more evidence that ribosome collisions occur frequently in cells lacking ABCE1, we knocked down ABCE1 in HeLa cells and performed polysome profiling (Figure 6B) to subsequently assess ubiquitination of proteins isolated from the monosome or disome fractions. Compared to the control KD, we found in the monosome and disome fractions of the ABCE1 KD a significant increase in RPS3 ubiquitination and a strong decrease of ubiquitinated RPS27A, while the overall abundance of both proteins remained unchanged (Figure 6C, D). The RPS27A gene encodes a fusion protein consisting of ubiquitin at the N-terminus and ribosomal protein S27a at the C-terminus and the ubiquitin moiety serves as substrate for ubiquitination of other proteins. Therefore, reduction of ubiquitinated RPS27A indicates an elevated need for ubiquitin, for example to modify other ribosomal proteins in RQC or to target misfolded newly synthesized proteins to the proteasome. Moreover, the release factor eRF1 was elevated in monosomes and disomes of the ABCE1 depleted cells, suggesting that eRF1-bound ribosomes cannot recycle properly when ABCE1 is not available (Figure 6C). Likewise, also RACK1 was enriched in monosomes and disomes of ABCE1 depleted cells, further supporting the idea of RQC activation in the absence of proper ribosome recycling (Figure 6C, D). Collectively, these results suggest that in the absence of ABCE1, RQC is activated as a consequence of ribosome collisions at the TC, which is sensed by ZNF598 and leads to the ubiquitination of RPS3 (and probably additional ribosomal proteins), similar to what has been reported to occur when ribosomes collide inside the CDS or within poly(A) stretches.