Ribosome stalling induced by giro
First, we set out to characterize giro’s inhibitory activity on protein synthesis in mammalian cells. Through metabolic labeling of newly synthesized proteins with O-propargyl-puromycin (OP-puro) (Fig. 1B), an aminoacyl-tRNA mimic carrying a click-chemistry compatible alkyne which allows conjugation to a fluorescent dye, we did observe a decrease in protein synthesis by giro in a dose-dependent manner. However, we also noticed that giro did not inhibit translation to the same extent as a conventional translation elongation inhibitor such as cycloheximide (CHX). When looking at the distribution of ribosomal populations via sucrose density gradient centrifugation (Fig. 1C), we observed a decrease in the 80S population but an increase in the polysome fraction. We conducted the same experiment in an in vitro translation system using [32P]-labeled rabbit β-globin mRNA. The data indicated an increase in the early polysome population (Fig. S1A). This increase in polysomes contradicts general translation initiation inhibition, which often results in polysome depletion. Rather, it is consistent with translation elongation inhibition, which stalls ribosomes in the middle of open reading frames (ORFs).
To get a global view of ribosome traversal along ORFs, we conducted monosome profiling, a deep sequencing based-method for ribosome footprints generated by RNase digestion12, on giro-treated HEK293T cells. In view of potential ribosome stalling by the compound, we also employed disome profiling which sequences the mRNA fragments protected by two closely adjacent ribosomes, caused by slow elongation of the leading ribosome13–16. Both monosome and disome footprints were increased at the 5′ end of the ORF (Fig. 1D), while the population around the stop codon was reduced (Fig. 1E). Especially the disome population decreased significantly towards the 3’ end. This suggested increased stalling towards the 5′ end with fewer ribosomes reaching the stop codon. We noted that there was no indication of stop-codon read-through (Fig. 1E), which contradicts earlier reports of giro interfering with translation termination4.
More quantitatively, we calculated the polarity score of footprint reads, a value indicating whether the average read distribution shifted towards the 5′ end (negative) or the 3′ end (positive)32. As expected, we saw a polarity score shift in direction toward the 5′ end in both monosome and disome profiling (Fig. 1F and 1G). The change in ribosome distribution towards the 5’ end could also be observed when representing the data as a cumulative fraction plot (Fig. 1H), showing that giro treated samples generally had lower polarity scores than control. Together, these data indicate that giro does not completely block translational activity, though the observed decrease in protein output may stem from increased ribosomal stalling.
Sequence-selective stalling
In addition to the global effect of giro on elongation, we investigated whether ribosomal stalling occurred on a particular sequence element. Therefore, we focused on the disome profiling data and analyzed overrepresentation of individual amino acids in the ribosomal A, P, and E sites (Fig. 2A). In the ribosomal P site, Pro (P) and Phe (F) appeared to increase while Gln (Q), Glu (E), and especially Lys (K) were overrepresented in the E site. This overrepresentation was also observed in our monosome data (Fig. S2A). Motif analysis for disome enriched codons further narrowed down the sequence context for giro-mediated ribosome pausing. Both 1 and 10 µM giro led to similar sequence enrichment: Lys/Pro-Pro-Pro and Lys/Pro-Phe-Pro across E,P and A sites (Fig. 2B). We then considered dipeptide motifs occupying E and P-site or P and A-site. High disome occupancy in E and P-site containing di-amino acid pairs Lys-Pro and Lys-Phe further underlined the importance of these sequences for giro’s activity (Fig. 2C). In a similar manner, we observed a significant increase of di-amino acid motifs containing P-site proline when considering P and A-site occupancy under giro treatment, including Pro-Pro. Disome formation on an individual Lys-Pro site inside the COX6C gene exemplifies the context specificity of giro (Fig. 2D). The mapped monosome data along the COX6C ORF also shows a decrease of ribosome density towards the 3’ end.
Given the enrichment of Lys in the E site, we wondered whether giro would show preference for one of the two lysine codons (AAA or AAG) in inducing ribosomal stalling. Intriguingly, we found a more pronounced increase in E-site AAA than AAG for disome formation (Fig. 2E). Sequence enrichment analysis for E-site AAA or AAG codons (Fig. S2B) led to the same conclusion. This was further highlighted by the disome occupancy of E and P-site with di-codon pairs, such as AAA-CCU (Lys-Pro), AAA-UUU (Lys-Phe), and AAA-CCG (Lys-Pro) (Fig. 2F). Moreover, unbiased motif analysis also showed enriched AAA, but not AAG, in the E site (Fig. S2C). These data demonstrated a unique context specificity for giro in inducing ribosomal stalling.
Displacement of eIF5A
We were furthermore intrigued by the increase in P and A-site proline. Increased stalling on poly-proline motifs is usually associated with insufficient levels of the translation factor eIF5A16,31−33 (Fig. S2D, E). Giro’s effect on translation looked like a combination of reduced eIF5A with additional stalling on E-site lysine. An involvement of eIF5A appeared possible, since the girolline binding site on the large ribosomal subunit of Haloarcula marismortui is known to lie in the vicinity of where eIF5A and the ribosome interact in eukaryotes (Fig. S3A)8,34. Furthermore, the binding site sits in an area of high sequence homology between archaea and eukaryotes (Fig. S3B) making binding in the same position on ribosomes of both phyla likely.
To determine whether giro binding affected the interaction between eIF5A and the ribosome, we used FLAG-tagged eIF5A for pulldown experiments. Tagged wild-type eIF5A did pull down ribosomes as shown by immunoblotting for small and large ribosomal subunit proteins, while the S51A mutant of eIF5A did not seem to interact with the ribosome (Fig. 3A, Fig. S3D). As most overexpressed eIF5A was reported to remain non-hypusinated35, we co-expressed DHS and DOHH, which are responsible for the generation of dehydroxyhypusine and then hypusine on lysine 5021,22, together with tagged eIF5A. Giro was able to interfere with the binding of hypusinated, and to a lesser extent also non-hypusinated eIF5A to the ribosome (Fig. 3A). It further blocked binding in a dose-dependent manner (Fig. 3B, S3E), when DOHH and DHS were co-expressed.
Thus far an involvement of eIF5A in the translation of the AAA codon has not been reported. We queried whether giro treatment induced ribosomal pausing in the same vicinity as pausing caused by eIF5A depletion (Fig. 3C). Focusing on significant pause sites detected under eIF5A depletion, we mapped pause sites induced by 10 µM giro treatment 30 nucleotides upstream and downstream of the giro-induced stall site. Both giro treatment and eIF5A depletion caused stalling on almost the same position, though with slight differences. eIF5A depletion showed a number smaller peaks 3’ of the main pause site.
This corresponds to our observation on an individual gene level. When we compared disome footprint distribution between giro treated cells and cells with RNAi-depleted eIF5A, we found a number of transcripts, such as ATP5B, COX7A2, and NDUFC1 where giro application produced ribosomal stalling in the exact same location as eIF5A depletion (Fig. 3D). In a few other instances, giro caused stalling at different codons than the eIF5A-knockdown-induced stall sites (Fig. 3E). Such was the case for PPT1, HIST2H2AA3, and RPL38. As can be seen in the sequence of the individual stall sites, Giro did not always induce stalling on AAA-codons. Rather it seemed to prefer areas where some stalling seems to occur naturally, as seen by a small number of disome footprints present even in the absence of external perturbation. It appears likely that translation slows in these areas, relying on eIF5A to maintain speed and fidelity.
AAA-specific stalling on reporter transcripts
To test whether we could replicate stalling on AAA-encoded lysine and test eIF5A involvement, we utilized fluorescent FACS reporter constructs, encoding enhanced green fluorescent protein (EGFP) and red fluorescent protein (RFP) connected by a short linker sequence 36. Flanked by P2A sites37, which allow the ribosome to proceed in-frame without amino-acid incorporation, the reporter produces three separate polypeptides from one transcript: EGFP as an internal control, the linker, and RFP as experimental readout (Fig. 4A)36. If translation stalls on the linker, less RFP is produced. Cells were transfected with vector constructs 48 h before cell sorting and treated with 1 µM giro for the last 16 h. We used empty linker K(O) for control and as expected did not see any effect of giro on either EGFP or RFP expression (Fig. 4B). Then we proceeded with a linker sequence containing twenty consecutive lysines encoded by (AAA)20 or (AAG)20 respectively (Fig. 4C, D). Twenty AAA codons in a row induced translational stalling with a modest reduction in RFP output compared to K(O) control, as reported previously36. However, giro treatment significantly reduced RFP production further (Fig. 4C). In many cases, RFP levels were reduced but not completely inhibited, visible in the FACS profile as the filling in of the “valley” between RFP negative cells and cells with high RFP expression (arrow). Giro affected only the expression of RFP when preceded by the (AAA)20-linker, while it had no discernible effect on a linker containing 20 consecutive AAG codons (Fig. 4D). Furthermore, on a shorter version of the linker containing only 12 AAA codons, 1 µM giro had very little effect while stronger reduction of RFP production required a 5 µM dose (Fig. S4A). This corroborated our suspicion that giro requires a slowdown in translation to act and 20 AAA-encoded lysines seem to cause more slowing than 12. These data suggest that giro presents a sequence context-dependent translation modulator, which primarily causes ribosomal stalling on AAA-encoded lysine.
We then assessed eIF5A involvement, testing our FACS vectors in cells with RNAi-reduced levels of eIF5A. On cells transfected with non-targeting siRNA, giro showed the same reduction in RFP expression observed in non-transfected cells (Fig. 5A top). Even in the absence of giro, eIF5A knocked-down cells showed reduced RFP production on reporters containing the (AAA)20-linker (Fig. 5A bottom). The FACS profiles of eIF5A-depleted cells are virtually indistinguishable from cells with normal levels of eIF5A after treatment with 1 µM giro (c.f. Figure 4C). Treating eIF5A knockdown cells with giro further reduced RFP expression, likely by preventing residual eIF5A from interacting with the ribosome (Fig. 5A bottom). eIF5A depletion only affected translation of the (AAA)20-linker; using the (AAG)20 construct, eIF5A depletion did not lead to detectable changes in RFP output (Fig. 5B).
Premature translation termination by RQC activation
The FACS reporters used in this study were originally designed for research on ribosome-associated quality control (RQC) and a recent study provided some evidence for eIF5A involvement in the RQC pathway38. RQC triggers when the cell detects collided ribosomes via the E3 ubiquitin ligase ZNF598 or the ATPase ASCC339. It involves separating the ribosomal subunits and destroying the nascent peptide by ubiquitination through the Ltn1 E3 ligase and the addition of non-templated alanine and threonine amino acids to the peptide’s C-terminus by NEMF, known as CAT-tailing40. Translation of poly-A tails in mammalian cells constitutes an established way of inducing RQC.
This led us to investigate the effect of giro on RQC induction. We modified our FACS reporter (Fig. 4A) to monitor premature translation termination and ribosomal stalling (Fig. 6A). We deleted the N-terminal P2A site and added a stop codon to the end of the linker. This construct now produced EGFP with a C-terminal tail sequence, consisting of either 20 AAA codons or 20 GAA codons. If translation was to proceed unimpeded, we expected to see a full-length band of ~ 48 kDa. Stalling and premature termination would produce a shortened protein, detectable by immunoblotting against EGFP. As expected, control cells showed some level of stalling made worse by the addition of giro (Fig. 6B), indicated by an increase in the detectable proteins smaller than 48 kDa.
eIF5A knock-down itself (Fig. S5A) led to an increase of these shortened proteins similar to treatment with 1 µM giro. Therefore, eIF5A appeared to aid the translation of poly-A stretches. This would support our hypothesis that giro acts in areas of translational slowing and would suggest that eIF5A plays a role in maintaining speed and fidelity of translation elongation beyond poly-proline stretches.
When RQC triggers ASCC3 or ZNF598 are knocked down (Fig. S5A), giro loses its effect, and translation proceeds to full length (Fig. 6B). The translational complexes stalling on a poly-A sequence should still be translation-competent and probably proceed to the end of the reading frame by themselves if RQC does not engage. When RQC factors Ltn1 and NEMF were depleted, more stalling was observed possibly because depletion of these downstream protein degradation factors slows down clearing of stalled ribosomal complexes further (Fig. 6B). Importantly, prematurely terminated peptides were only observed on the (AAA)20 construct, while the translation of (GAA)20 did complete unimpeded, irrespective of giro treatment or RQC factor knockdown (Fig. 6C). Corroborating our FACS data, even 20 consecutive AAG codons did not cause visible stalling irrespective of giro treatment or eIF5A-knockdown (Fig S5B). It is known that the AAA codon presents more challenges to the translation machinery than AAG41,42. This appears due to adenosine nucleotides potentially forming single-stranded helices, interfering with decoding43. In addition, the positively charged lysine, especially in case of poly-lysine sequences, appears to force the peptidyl tRNA into an unfavorable conformation.
This underlines that eIF5A is only required when translation slows down but does not act as a general elongation factor.
Effects on gene expression
Since giro-induced stalling occurs in a cellular context, we were curious about which translational response giro would induce. Simultaneous RNA sequencing from the same material used in ribosome profiling allowed us to measure changes in translation efficiency (TE). We observed significantly altered TE at higher giro concentration. Few genes appeared affected at 1 µM (Fig. S6A) while 10 µM giro produced a much stronger result. We observed increased expression of genes relating to protein synthesis, likely a result of the cell trying to compensate for decreased protein output (Fig. S6B and Supplementary Table 1)44,45. We also saw upregulation of adenosylmethionine decarboxylase (AMD1), a gene involved in polyamine synthesis46. Since the polyamine spermidine is a source for the hypusine modification on eIF5A21, this regulation may be another means to counteract giro’s function to inhibit eIF5A. At the same time, splicing factors and heat shock proteins appeared to decrease in expression (Fig. S6B and S6D). Stalling does not necessarily mean termination of translation and in a number of genes affected by giro we did not detect a significant reduction in protein levels (data not shown).
Mitochondrial phenotype
eIF5A also plays a role in mitochondrial maintenance29,30 and reduced hypusination levels on eIF5A during the aging process have been associated with mitochondrial deterioration. We compared giro’s impact on mitochondria to DHS inhibitor GC7, which significantly reduces eIF5A hypusination levels47. We tested whether giro would affect mitochondrial function by both fluorescent microscopy and metabolome analysis in comparison to GC7 treatment. Mitochondria were stained with MitoTracker Red and immunostained against the outer mitochondrial membrane protein TOMM20. Giro treatment displayed a modest effect on mitochondrial morphology with some thinning of the mitochondrial network (Fig. S7A). In contrast, treatment with 30 µM GC7 obliterated mitochondrial structure and led to a breakup of the mitochondrial network into small circular structures.
To further test whether giro inhibited mitochondrial translation, we used mitochondrial fluorescent non-canonical amino acid tagging (mito-FUNCAT) for metabolic labeling48. We detected a slight decrease in mitochondrial protein synthesis when treating cells with 1 and 5 µM giro overnight, while GC7 significantly inhibited mitochondrial translation (Fig. S7B). In terms of TCA intermediates, we did observe a reduction in α-ketoglutarate, malate, and fumarate, but did not see a reduction in citrate or isocitrate by giro treatment (Fig. S7C). In contrast, 30 µM GC7 treatment led to marked depletion of either metabolite (Fig. S7D).
Although our findings confirmed that giro treatment also impacts mitochondrial physiology, GC7 appeared to have a stronger effect than giro. 1 µM giro or 30 µM GC7 correspond to the IC50 for inhibition of cell proliferation. While GC7 treatment does decrease overall protein synthesis27, we could neither detect a change in ribosomal stalling behavior nor in the polarity of ribosome distribution (data not shown). We suspect that part of GC7’s impact on mitochondria may result from activity unrelated to eIF5A hypusination (Matsumoto et al. manuscript in preparation).