Prior to carrying out the Ribo-seq experiment, we first optimized the conditions of Api treatment to achieve a significant antibiotic effect while minimizing undesirable secondary events. Incubation of E. coli (BL21) cultures for 1 min with increasing concentrations of Api resulted in severe inhibition of protein synthesis (Figure 1—figure supplement 1A). Exposure to 1.25 µM of Api, which corresponds to the minimal inhibitory concentration (MIC) that prevents cell growth, reduced translation by as much as 75%. Treatment for 2 min with 4× MIC diminished protein synthesis to nearly 10% of the control and extending the treatment to 10 min achieved ~94% inhibition (Figure 1—figure supplement 1B). These results confirmed the previous assertion that Api acts as a general inhibitor of bacterial translation (Castle et al., 1999; Krizsan et al., 2014) and guided the selection of the treatment conditions for the subsequent experiments.

We then proceeded to the Ribo-seq experiments to obtain an unbiased view of how treatment with Api alters the global distribution of translating ribosomes in cellular mRNAs (Ingolia et al., 2009; Oh et al., 2011). Exponentially-growing E. coli cells were incubated for 5 min with 4× MIC of Api, conditions that led to nearly complete inhibition of translation (Figure 1—figure supplement 1B), and ribosome-protected mRNA fragments (ribosome footprints) were isolated, sequenced, and mapped to the E. coli genome (Becker et al., 2013; McGlincy and Ingolia, 2017). Exposure to Api caused a dramatic redistribution of ribosomes along mRNAs. The general pattern observed within individual genes revealed a buildup of ribosome density toward the ends of the ORFs, often spilling into the 3’-intergenic regions (Figure 1A, Source data 1). In operons with narrowly spaced genes, the ribosome density was primarily observed at the junctions spanning 3’-terminal codons of the upstream ORF and 5’-proximal codons of the downstream ORF (Figure 1B). Overall, these effects are drastically different from those observed previously with the antibiotic retapamulin, a specific inhibitor of translation initiation, which generates precise and sharp peaks representing ribosomes occupying start codons (Meydan et al., 2019). To mitigate the uncertainty of assigning reads at the junctions of adjacent or overlapping genes in the operons, we limited our subsequent analysis to actively translated genes that are separated by at least 50 nt from the nearest gene (see Materials and methods). The metagene analysis of ribosome coverage near the stop codons of the well-separated genes (Figure 2A) revealed several major effects of Api action that will be discussed in the following sections.

A prominent outcome of Api treatment is a significant (~tenfold) increase of the average normalized ribosome density at stop codons (Figure 2A). Approximately 11% of all the ribosome footprints in Api-exposed cells map to stop codons. This result reveals Api as a potent global inhibitor of translation termination in bacterial cells and expands the previous findings of the in vitro experiments which showed that Api could arrest ribosomes at the stop codon of several model genes during cell-free translation (Florin et al., 2017).

Given the disparity in ribosome occupancy at and around termination sites in individual genes (Figure 1A), we computed stop codon (SC) scores to systematically assess the extent of Api-induced ribosome stalling in these regions. The SC score reports the ribosome density within the last three codons of the genes (i.e., the last two sense codons and the stop codon of the ORF) relative to the average density across the entire gene (Figure 3A, Figure 3—figure supplement 1). In the Api-treated cells, the SC scores of 97% of genes shifted to higher values compared to the untreated control. Although SC scores varied significantly between the genes, more than half (56%) of the genes in the Api sample exhibited values above 8 (log₂ = 3), whereas only 1% of the genes in the control cells showed such high scores, reflecting the dramatic increase in stop codon occupancy caused by Api (Figure 3A).

We considered the possibility that the SC score variation between the genes could be due to a differential action of Api toward RF1 and RF2. Indeed, the Api sample SC scores at the RF2-specific UGA codons were somewhat lower than those at the RF1-specific UAG codons, but the difference was fairly modest and generally followed the trend observed in the untreated control (Figure 3—figure supplement 2). This result shows that Api efficiently acts upon both class 1 RFs, thereby contradicting a recent suggestion that Api preferentially traps RF1 over RF2 (Kuru et al., 2020). In search for other features that could be associated with a more pronounced Api-induced ribosome stalling at the ends of specific genes, we analyzed the C-terminal sequences of the proteins encoded in ORFs with high SC-scores (Figure 3B). The pLogo analysis showed statistically significant prevalence of glycine residues at the C-termini of such proteins, a tendency not observed in high SC-scoring ORFs in the untreated cells. This result indicates that Api does not simply exacerbate inherent difference in translation termination efficiency between the genes, but rather acts in a context-specific manner.

No remarkable sequence context features were detected at the regions downstream of the stop codons of the high-SC score genes (Figure 3—figure supplement 3A). Similarly, none of the other examined properties, including mRNA abundance, its propensity to form stable secondary structures, or the number of ribosomes per mRNA molecule show any strong correlation with the SC scores of the genes in the Api-treated cells (Figure 3—figure supplement 3B).

Taken together, our data revealed that the primary mode of Api action in the cell is inhibition of translation termination and that this effect is influenced by sequence context.

At least three distinct waves of ribosome density are observed in the metagene profile upstream from the stop codons in the Api sample (Figure 2A). The crests of the waves are spaced by ~27 nt, a distance corresponding to the peak of the footprint length distribution (Figure 2—figure supplement 1), suggesting that this upstream density likely represents elongating ribosomes queued behind those arrested at stop codons. Even though ribosome collisions have been observed in bacteria and eukaryotes exposed to translation elongation inhibitors (Kearse et al., 2019; Mohammad et al., 2019), the extent of ribosome queuing in the Api-treated cells is especially pronounced. Because the queued ribosomes occupying inner codons of the ORFs carry nascent polypeptides they should be refractory to Api binding (Florin et al., 2017). Yet, despite being Api-free, they are unable to complete the synthesis of the encoded protein and participate in new rounds of translation.

The metagene profile of the Api sample shows a dramatic increase in the number of ribosome footprints mapping to the regions downstream from the stop codons of the well-separated genes (Figure 2A). Similar to the SC score, we created the readthrough (RT) score to quantify the ribosome density within the first 40 nt of the 3’UTR relative to the ORF-associated density. In the control sample, the RT score of most genes is very low because few ribosomes are found downstream of stop codons. In contrast, in the Api sample, 90% of the genes have an RT score greater than 1, indicating higher relative density in the 3’UTR than in the ORF itself due to pervasive and efficient stop codon bypass (Figure 4A, Figure 4—figure supplement 1). We found no strong correlation between the stop codon identity and the RT score, indicating that all the stop codons are bypassed with comparable efficiency in the Api-treated cells, in contrast to the control cells, where UAA genes have significantly lower RT scores (Figure 4A, Figure 4—figure supplement 2).

The ribosome footprints in the 3’UTRs could represent post-termination ribosomes that have released the completed protein but somehow remained associated with mRNA and migrated past the stop codon. Alternatively, the 3’UTR footprints may correspond to ribosomes that failed to release the completed protein at the stop codon, bypassed it, and continued translation into the downstream segment of mRNA. Such a scenario could be caused by the depletion of available RF1/RF2 trapped by Api on terminating ribosomes. To test the feasibility of the latter model, we examined to which extent Api treatment affects the abundance of free RFs, more specifically of RF2, which by far outnumbers RF1 in the E. coli cytoplasm (Schmidt et al., 2016). Strikingly, unlike those of the control sample, the ribosome-depleted lysates of Api-treated cells essentially lacked free RF2 (Figure 2B). In addition, we reasoned that if 3’UTR-associated ribosomes are engaged in translation they should preferentially stall at the downstream stop codons. Indeed, metagene analysis shows a peak of ribosome footprints at the first in-frame downstream stop codon (Figure 4—figure supplement 3A), indicating that ribosomes that bypassed the termination signal of the main ORF are actively translating downstream mRNA sequences.

We used shotgun proteomics to test whether exposure of cells to Api leads to productive translation of the non-coding mRNA segments and generation of proteins with C-terminal extensions. To allow for the accumulation of proteins synthesized during Api treatment, we exposed cells for 1 hr to a sublethal concentration of Api (0.5× MIC) permitting protein synthesis to continue, even if at a reduced level. In addition, we also incubated cells for 1 hr with a higher Api concentration (4× MIC). We then analyzed the proteins synthesized in treated and control cells, using custom algorithms to search for tryptic peptides encoded in the 150 nt-long segments downstream from the stop codons of the annotated ORFs (see Materials and methods). Even when applying strict filtering criteria (Gessulat et al., 2019), we were able to identify a significant number of peptides encoded in the genomic regions downstream from the ORF stop codons in both of the Api samples (Figure 4B and Source data 3). The same analysis found no such peptides in the control sample. The number of identified peptides encoded in the 3’UTRs increased further upon applying a less strict filter (Figure 4—figure supplement 3B). These results argued that genome regions downstream of the stop codons of the annotated ORFs are actively translated in the cells exposed to Api. Approximately half (49%) of the identified ‘extension’ peptides were encoded in frame with the upstream ORF (Figure 4C) and thus, were likely produced by the ribosomes that misread the main ORF stop codon as a sense codon and continued translating the downstream mRNA sequence. Consistently, in the three identified tryptic peptides encoded in mRNA sequences spanning the ORF termination signal, the stop codon was mistranslated by a near-cognate aminoacyl-tRNA (Figure 4—figure supplement 4). The other detected downstream peptides were encoded in the (−1) or (+1) frames relative to the main ORF (Figure 4C and Source data 3). These peptides also likely belong to C-terminal extensions of the main-ORF proteins given that stop codons can be bypassed via frameshifting of the ribosomes stalled at termination sites in the pre-release state, as was observed in the experiments of Gross and coworkers (Baggett et al., 2017). The sample size of the ‘extension’ peptides was insufficient to systematically identify the mRNA or nascent protein features conducing to stop codon bypass via frameshifting. However, the presence of a slippery sequence preceding the stop codon (e.g. in arcB; Figure 4—figure supplement 5A) or the possibility of repairing of peptidyl-tRNA in a different frame (e.g. in rplE; Figure 4—figure supplement 5B) could be among the factors contributing to the synthesis of identified proteins’ C-terminal extensions.

Taken together, our results show that stop codon bypass, active translation of 3’ UTRs, and production of proteins with C-terminal extensions constitute major consequences of Api action.

The results of our analysis suggest that the complex metagene profile observed in the vicinity of stop codons in the Api-treated cells is accounted for by distinct populations of ribosomes: the post-release ribosomes in complex with RF1/RF2 arrested by Api at the stop codons of the main ORF and the ribosomes carrying peptidyl-tRNA (including the queued ribosomes, some of the ribosomes in 3’ UTRs, and a fraction of ribosomes stalled at the stop codons in pre-release state). We hypothesized that this complex profile could be at least partially simplified by clearing from mRNA the ribosomes carrying peptidyl-tRNA. For that, we used puromycin (Pmn), an antibiotic that releases the nascent protein from ribosomes whose PTC A site is accessible (Traut and Monro, 1964). Indeed, metagene analysis showed that Pmn treatment of lysate from Api-treated cells significantly reduced the relative amounts of queued ribosomes and 3’UTR ribosomes and, consequently, increased the relative occupancy of stop codons (Figure 2—figure supplement 2). Importantly, this does not imply that Pmn causes more ribosomes to associate with stop codons but that a higher fraction of the remaining mRNA-bound ribosomes are positioned at termination sites. However, even upon Pmn treatment, some ribosomes remained within the ORFs, suggesting that either some of the translation complexes are refractory to Pmn action or that Pmn treatment requires further optimization. The remaining footprints still found in 3’UTRs after Pmn treatment may represent the ribosomes arrested at the downstream stop codons: Api-trapped RFs would block Pmn binding.

Exposure of cells to Api leads to significant deregulation of the proteome. One of the most prominent effects is an increase in the relative content of ribosomal proteins and other factors related to the function and assembly of ribosomes (Figure 5-figure supplement 1A and Source data 4). Among the top 50 proteins whose abundance is increased in the cells exposed to 0.5× MIC of Api, 19 are related to translation (Figure 5—figure supplement 1B). This shift may reflect an attempt to compensate for the reduced protein synthesis capacity caused by the Api inhibitory action (Source data 4). Curiously, 13 other proteins among the top 50 are either associated with the inner or outer membrane or reside in the periplasm (Figure 5—figure supplement 1B), suggesting a still unknown link between Api action and protein secretion.

Analysis of the Ribo-seq data showed that the highest peak of ribosome occupancy in the Api-treated cells corresponds to the stop codon of the degron-encoding sequence in tmRNA (Figure 5A,B). tmRNA is a component of a conserved bacterial system that rescues stalled mRNA-associated ribosomes with an empty A site (Buskirk and Green, 2017; Moore and Sauer, 2005). Following tmRNA binding to the A-site, the ribosome switches templates, synthesizes the tmRNA-encoded degron sequence, and terminates translation, using class 1 RFs, at the UAA stop codon (Buskirk and Green, 2017; Moore and Sauer, 2005). tmRNA is known to operate on ribosomes that reach the 3’ end of non-stop mRNAs or, potentially, those stalled due to scarcity of A-site ligand (Ivanova et al., 2004; Janssen et al., 2013; Li et al., 2007; Subramaniam et al., 2014). Both types of tmRNA substrates are generated in the cells treated with Api. First, the Api-stimulated stop codon bypass enables the translating ribosomes to approach the 3’-termini of RNA transcripts (Figure 5C,D), where they could be recognized as requiring rescue. In addition, ribosomes stalled at stop codons in a pre-release state, and possibly the queued ribosomes as well, whose A site is vacant, could serve as tmRNA substrates.

The sharp increase of ribosome footprints at the tmRNA stop codons in the Api samples (Figure 5B) argues in favor of the massive recruitment of tmRNAs to the stalled ribosomes. However, due to the Api-mediated depletion of the available RFs (Figure 2B), the degron-tagged protein cannot be released, thereby preventing proper tmRNA function and recycling. Consistently, the cell attempts to activate the two other E. coli backup rescue factors, ArfA and ArfB (reviewed in Buskirk and Green, 2017). For the arfA gene in the Api-treated cells, we observed a nearly tenfold increase in the abundance of its transcript and a more than sevenfold higher number of ribosomes associated with its mRNA (Figure 5D and Source datas 5 and 6). The relative amount of arfB transcripts increases by ~40% and there are ~4 times more ribosomes per arfB transcript (Source datas 5 and 6). However, because ribosome rescue by tmRNA or arfA requires the availability of RFs and because ArfB, which acts as a termination factor itself, becomes trapped in the ribosome by Api (Chan et al., 2020), it is unlikely that either of the rescue systems can mitigate the consequences of Api action. Consistently, deletion of tmRNA, arfA, or arfB genes does not increase cell sensitivity to Api (Source data 7).

While the predominant effect of Api on cellular translation is the build-up of ribosome density in the vicinity of stop codons (Figure 2A), a metagene analysis of the 5’ ends of genes showed a moderate increase (~twofold) of the number of ribosome footprints at the start codons of the ORFs in the Api sample compared to the control (Figure 2—figure supplement 3A). Among the genes with the strongest start codon effects, we selected zntA, srmB, cysS, tyrS, and yhbL to test in vitro whether Api directly delays the ribosome departure from the initiator codon. While translation arrest at stop codons is readily detected by toeprinting analysis at 50 µM of Api (Florin et al., 2017), no start codon effects were observed at this concentration of the inhibitor in any of the tested ORFs. However, increasing Api concentration to 2 mM led to the appearance of a start codon toeprint bands in the zntA and tyrS templates (Figure 2—figure supplement 3B–C); the start codon effects were much weaker or completely absent with the other tested genes. The moderate effect of Api on translation initiation is reminiscent of the main mode of action of other PrAMPs (Gagnon et al., 2016; Roy et al., 2015; Seefeldt et al., 2016; Seefeldt et al., 2015) and could be explained by its ability to bind with a reduced affinity to the ribosome even in the absence of RF1/RF2 (Florin et al., 2017; Kolano et al., 2020; Krizsan et al., 2014). However, it remains to be elucidated whether the mechanism of the weak inhibition of translation initiation by Api is comparable to that of other PrAMPs.

Api137 (Berthold et al., 2013), which we refer to simply as Api, was synthesized by NovoPro Biosciences Inc DNA oligonucleotides were synthesized by Integrated DNA Technologies.

The inhibition of cellular protein synthesis by Api was analyzed by metabolic labeling as described previously (Meydan et al., 2019) with the following modifications. E. coli BL21 ∆tolC cells were grown overnight in MOPS medium (Teknova) lacking methionine (MOPSΔMet). Cells were diluted 1:200 into fresh MOPS-Met and grown at 37°C until the culture reached a density of A₆₆₀ of 0.2. Aliquots of the exponentially growing cells were added to tubes containing appropriately diluted Api in MOPSΔMet medium to obtain 0.75, 1.5, 3.1, 6.25, 12.5, 25, and 50 µM as final Api concentrations in the total volume of 100 µL. After 2 min incubation, 28 µL were transferred to another tube containing 2 µL MOPSΔMet medium supplemented with 0.3 µCi of L-[³⁵S]-methionine (specific activity 1,175 Ci/mmol; MP Biomedicals). Following 90 s incubation, the content was transferred onto Whatman 3 MM paper discs pre-wetted with 5% TCA and the procedure was continued as described previously (Meydan et al., 2019).

The time course of protein synthesis inhibition was performed as described above, except that Api was directly added to a final concentration of 6.25 µM to the exponentially growing culture in MOPSΔMet medium and aliquots were taken at 0, 2, 5, and 10 min.

The Ribo-seq experiments were performed as described previously (Meydan et al., 2020; Meydan et al., 2019) using cells grown in MOPS medium. Briefly, the overnight cultures of E. coli BL21 ∆tolC cells were diluted 1:200 in 150 mL of medium and grown at 37°C in flasks until reaching density of A₆₀₀ ~0.4. For Api-samples, 4.12 mg of dry Api powder was added directly to the cultures to a final concentration of 4× MIC and incubation with shaking continued for 5 min. Untreated (control) and Api-treated cells were harvested by rapid filtration, flash frozen in liquid nitrogen and cryo-lysed in 650 µL lysis buffer (20 mM Tris pH 8.0, 10 mM MgCl₂, 100 mM NH₄CL, 5 mM CaCl₂, 0.4% Triton X-100, 0.1% NP-40) supplemented with 65 U RNase-free DNase I (Roche) and 208 U Superase•In RNase inhibitor (Invitrogen). The pulverized cells were thawed at 30°C for 2 min and incubated in an ice water bath for 20 min and lysates were clarified by centrifugation at 20,000 × g for 10 min at 4°C. One aliquot of the clarified lysates was preserved frozen for RNA-seq analysis, another aliquot was immediately treated with MNase (see below); the third aliquot was treated by puromycin (Pmn) as follows. The cell lysates were supplemented with 1 mM Pmn (Millipore-Sigma), 10 mM creatine phosphate (Millipore-Sigma) and 40 µg/mL creatine kinase (Roche), and incubated for 10 min at 25°C with shaking at 1400 rpm. Cell lysates, treated or not with Pmn, were subjected to MNase treatment by adding 450 U MNase (Roche) per 25 A₂₆₀ units of lysate, 3 mM GMPPNP, and incubating for 60 min at 25°C. The MNase reaction was quenched by the addition of EGTA (5 mM final concentration). Following sucrose gradient centrifugation and collection of the 70S ribosome peak (Becker et al., 2013), subsequent isolation of ribosomal footprints and library preparation was performed as described (McGlincy and Ingolia, 2017).

A custom script was used to demultiplex the samples, remove the linker barcode and any nts 3' of the linker barcode, and then remove five nts from the 3' end and 2 nt from the 5' end, which were added as part of the library design (McGlincy and Ingolia, 2017; Meydan et al., 2020; Meydan et al., 2019). Bowtie2 (v2.2.9) within the GALAXY pipeline first aligned the trimmed reads to the non-coding RNA sequences. The remaining unmapped reads were aligned to the reference BL21 genome (GenBank ID CP001509.3; Meydan et al., 2020; Meydan et al., 2019). The 26 to 39 nt-long reads were used in the subsequent analyses. After analyzing the footprints at the start codons, the first position of the P-site codon was assigned 16 nt from the 3’ end of the read, in-line with the 15 nt offset previously proposed (Mohammad et al., 2019).

The metagene analyses at the annotated start and stop regions followed the previously described protocol (Aleksashin et al., 2019). For the inclusion in the analysis, the well-separated ORFs had to satisfy three criteria in all four datasets (two control and two Api-treated biological replicates): (i) The ORF length is at least 300 nt, (ii) at least 5% of the positions within the coding sequence had assigned reads values above zero, and (iii) the average number of rpm per nt in an ORF is ≥0.005. Ribosome footprint density at each nucleotide was normalized to the average coverage of the entire ORF plus 50 nt up- and downstream. The mean of the normalized values was calculated and plotted for the regions around the start and stop codons of the ORF.

The metagene analysis of the first in-frame stop codons of the downstream regions was carried out using 3769 genes with such codons present within the 225 nt downstream of the annotated ORF. Normalized read coverage, calculated for each nucleotide within a 200 nt window centered around the last nucleotide of the downstream stop codons, was determined by dividing ribosome footprint coverage at each position by the maximum coverage value within the window. Normalized coverage was then summed for each position in the windows and plotted.

To analyze the metagene coverage at the 3’ termini of mRNAs, we first converted the 3’ termini coordinates reported in the E. coli BW25113 strain (Dar and Sorek, 2018) to coordinates in the BL21 strain, used in this work. We then followed a similar normalization and summing procedure to that used in the first in-frame stop codon analysis. However, the window of nts for normalization included 100 upstream and 20 downstream of the mRNA 3’ termini.

The well-separated genes included in the analysis were required to meet two criteria in all four datasets (two control and two Api-treated biological replicates): (i) they were separated from the upstream and downstream adjacent genes by at least 50 nt, and (ii) they have a coverage density of at least 0.1 reads per nt in the protein-coding sequence, excluding the first and last 9 nt of the ORF. The stop codon (SC) region was assigned as nine 3’-terminal nt of the ORF (the last two sense codons and the stop codon); the readthrough (RT) region encompassed 40 nt downstream from the ORF stop codon. The SC and RT scores were calculated by dividing the ribosome density within the SC region or the RT region, respectively, by the total ORF density. The mean SC and RT scores of each gene were used in further analyses.

Aliquots of the clarified cell lysates from the samples prepared for the Ribo-seq analysis (see above) were used to extract and analyze total RNA as previously described (Aleksashin et al., 2019). Briefly, total RNA was purified by acidic phenol extraction, short RNAs and rRNA were subtracted using the MEGAclear (Ambion) and MicrobEXPRESS (Ambion) kits, respectively, mRNA was fragmented via alkaline fragmentation and, after size-fractionation and isolating fragment sin the 15–45 nt range, converted to the library for next-generation sequencing.

The RNA-seq analysis was performed as previously described (Aleksashin et al., 2019). Briefly, Cutadapt was used to process and demultiplex the reads. Kallisto (v.0.44; Bray et al., 2016) and Sleuth (v.0.30; Pimentel et al., 2017) were used to align the reads and perform a differential expression analysis, respectively. To align reads to the ssrA transcript, the GALAXY pipeline was used for alignment and quantifying reads per million after filtering out reads aligning to the non-coding RNA.

E. coli BL21ΔtolC cells were grown in 100 mL of MOPS media to a density of A₆₀₀ ~0.5, split into separated flasks and either treated with 4× MIC of Api for 5 min or left untreated (control). The flasks were cooled in an ice water bath and cell cultures were centrifuged in a rotor JA-25.50 (Beckman) at 4°C and 4000× g for 10 min. The cell pellet was resuspended in 400 µL cold lysis buffer (20 mM Tris pH 8.0, 10 mM MgCl₂, 100 mM NH₄Cl, 5 mM CaCl₂, 0.4% Triton X-100, 0.1% NP-40), 10 U/µL RNAse free DNase I (Roche) and 20 U/µL Superase•In RNase inhibitor (Invitrogen) frozen in liquid nitrogen and cryo-milled. The pulverized cells were thawed at 30°C for 2 min, incubated in an ice water bath for 20 min and lysates were clarified by centrifugation for 10 min at 20,000 × g and 4°C. About 150 µL of the lysate was layered onto a 600 µL sucrose cushion (25% sucrose, 20 mM tris pH 8.0, 10 mM MgCl₂, 100 mM NH₄Cl) in a TLA100.2 rotor (Beckman) and centrifuged for 1 hr at 100,000 rpm and 4°C. The supernatant was collected, and the ribosome pellet was resuspended in buffer (20 mM Tris pH 8.0, 10 mM MgCl₂, 100 mM NH₄Cl, 1% SDS). Samples were separated by 4–20% SDS-PAGE gels, then electroblotted onto nitrocellulose membranes followed by Ponceau staining and probing with rabbit anti-RF2 antibody. Antibody binding was visualized using a secondary antibody (goat anti-rabbit IgG labeled with horseradish peroxidase) and imaged chemiluminescence.

Individual fractions of the tryptic peptides were separated by a capillary C18 reverse phase column of the mass-spectrometer-linked Ultimate 3000 HPLC system (Thermo Fisher) using a 90 min gradient (5–85% acetonitrile in water) at 300 nL/min flow. Mass spectrometry data were acquired by an Orbitrap QExactive HF system (Thermo Fisher) with nanospray ESI source using data-dependent acquisition. Raw data were collected and processed by the ProLuCID search engine (Xu et al., 2015) and an in-house pipeline.

Raw files were converted into MSn files using in-house converter or mzml files using MSconvert (Chambers et al., 2012) and then searched by both ProluCID (Xu et al., 2015) and MSFragger (Kong et al., 2017) using standard and extended reference proteomes (see below). The identified peptide sequences were first mapped to the in silico-generated extended proteome, then cross-referenced to the standard reference proteome (Uniprot E. coli BL21 reference proteome downloaded 2/20/2020). The 'extension' peptides were identified by subtracting all the peptides found in standard reference proteome from the peptides found in the extended proteome. All peptides mapped to the 150 nt-long genomic regions downstream of the stop codons of each annotated ORF and those covering both the tail of annotated ORFs and head of the downstream of the stop codons were reported. All scripts processing the data were written in Python 3.7 and are available at http://git.pepchem.org/gaolab/api_ribo_ext.

The standard reference proteome was constructed using the annotated reference genome of E. coli, strain BL21 DE3 (NCBI GenBank: CP001509.3). Extended reference proteome was generated by amending standard reference proteome with the proteome that could be encoded in three frames in the 150 nt-long genomic regions downstream of the stop codons of the 2100 highly expressed annotated genes (>112 sequence counts or 17.2 RPMs, the median of all genes, in the Ribo-seq dataset) in the Api-treated sample. The bypass of the first downstream stop codon was presumed to take place in 0/ (+1)/(−1) frame using the rules described above for the primary stop codon. The subsequent downstream stop codons were presumed to be translated in 0 frame.

To investigate the readthrough pattern of the primary stop codon, another custom proteome database was constructed with the primary stop codon translated into each of the 20 natural amino acids. The mass spectrometry data were then re-processed the same way as described above except using this new extended database.

To elucidate the changes in the proteome among different conditions, all the proteins identified from the above-mentioned search, regardless of the extension status, were quantified using a semi-quantitative label-free quantitation method using normalized spectral abundance factor (NSAF) to represent the relative expression level of each protein (Zybailov et al., 2005). All the low confident proteins, that is, proteins with <5 spectral counts, were removed from the analysis. The rest of the proteins were compared using the NSAF ratio between two conditions and sorted using a gamma distribution cumulative distribution function. The top 50 genes chosen from each comparison were used as input for pathway analysis using String-DB (Szklarczyk et al., 2019). A heat map of all the protein expression levels across all conditions was generated using NSAFs (Figure 5—figure supplement 1A). Genes and conditions were both clustered using a hierarchical k-mean clustering algorithm (Gao et al., 2019). The color represents the relative expression level (NSAFs) of each protein quantified as mentioned above.

DNA templates for toeprinting containing T7 promoter and an annealing site for the toeprinting primer were generated via PCR by amplifying the corresponding gene sequences from the E. coli BL21 genomic DNA. The following DNA primers were used for preparing the respective gene templates: zntA (TAATACGACTCACTATAGGGAAACTTAACCGGAGGATGCCATGTCG and GGTTATAATGAATTTTGCTTATTAACTTTGAACGCAGCAAATTGAGGGGC), tyrS (TAATACGACTCACTATAGGGTTATATACATGGAGATTTTGATGGCAAGC and GGTTATAATGAATTTTGCTTATTAACGACTGGGCTACCAGCCCCCG), srmB (TAATACGACTCACTATAGGGGCCCCACACAGAGGTAGAACATGACTGTAACG and GGTTATAATGAATTTTGCTTATTAACGGCTTCCAGCAGGCTTTCGTCG), ybhL (TAATACGACTCACTATAGGGTATATCTTCAGGAGATCGTCATGGACAGATTCC and GGTTATAATGAATTTTGCTTATTAACGATTGCAAGCCAGCCCGGGGTTGTACG), cysS (TAATACGACTCACTATAGGGATGTCTAAACGGAATCTTCGATGCTAAAAATCTTC and GGTTATAATGAATTTTGCTTATTAACTGAATAGGCTTAAATTCCTCTTTTTGGCG). The toeprinting assay was performed as previously described (Orelle et al., 2013) using the PURExpress system (New England Biolabs). When needed, Api was added to the final concentration of 50 µM or 2 mM. The final concentration of edeine was 50 µM. Following 10 min of translation, reverse transcription was carried out for 10 min using toeprinting primer NV1 GGTTATAATGAATTTTGCTTATTAAC (Vazquez-Laslop et al., 2008). Reverse transcription products were separated on a 6% sequencing gel. The gel was dried, exposed to a phosphorimager screen, and visualized in a Typhoon phosphorimager (GE Healthcare).

E. coli BL21ΔtolC cells were grown in MOPS media to a density of A₆₀₀ ~0.4 and then treated for 30 min with 2× MIC of either Api, spectinomycin (60 µg/mL), or streptomycin (1 µg/mL). Control antibiotic treatment conditions were those affording maximum protein aggregation (Ling et al., 2012). Cells were lysed and protein aggregates were isolated as previously described (Tomoyasu et al., 2001). The aggregates were separated with SDS-PAGE and visualized by silver staining.