An RNA structural switch controlling bacterial toxin translation

TL;DR

The study reveals a new RNA-based switch that controls toxin production in bacteria without needing enzymes.

Contribution

A novel structural RNA switch mechanism is identified for controlling bacterial toxin translation.

Findings

Translation initiation at timP requires a pseudoknot in the 5′ untranslated region.

A long-range interaction destabilizes the ribosome-binding-site-sequestering stem-loop.

An alternative RNA interaction locks the mRNA in an inactive state.

Abstract

Type I toxin–antitoxin systems (T1TAs) rely on tight posttranscriptional control to prevent inadvertent toxin synthesis, yet the molecular mechanisms underlying this control are highly diverse. Here, we uncover an RNA-based mechanism that controls translation initiation in the enterobacterial timPR system. Unlike most T1TAs, which typically rely on ribonucleolytic messenger RNA (mRNA) processing to relieve ribosome binding site sequestration, the primary timP toxin mRNA is activated through a purely structural RNA switch. Using a FASTBAC-Seq loss-of-function screen with biochemical and phenotypic assays, we here identify key RNA interactions that govern this switch. Translation initiation at timP requires formation of (i) a pseudoknot in the 5′ untranslated region, and (ii) a long-range interaction that destabilizes the ribosome-binding-site-sequestering stem-loop, rendering the Shine–Dalgarno sequence accessible for pre-initiation complex formation. Conversely, an alternative interaction locks the mRNA in an inactive state. Our findings reveal a structural RNA switch that controls toxin expression without the need for enzymatic processing and demonstrate an alternative mechanism for translation initiation in bacteria.

Graphical Abstract

Figures

FASTBAC-Seq identifies single point loss-of-function (LOF) mutations in timP mRNA. (A) Schematic representation of the timP 5′UTR secondary structure in its active and inactive conformation. Yellow: the pseudoknot-forming sequences; gray: the TimR-binding site; blue: SD and start codon. (B) Schematic representation of the FASTBAC-Seq method. The timP gene from Salmonella enterica, flanked by either a constitutive promoter or a promoter-less sequence and homologous recombination sites (brown boxes), was amplified under error-prone conditions and recombined into the Escherichia coli chromosome. DNA was extracted from the resulting colonies, followed by polymerase chain reaction (PCR) amplification, high-throughput sequencing and statistical analysis to identify LOF mutations (see the “Materials and methods” section for details). (C) LOF point mutations identified by FASTBAC-Seq throughout the timP mRNA sequence. (D) LOF mutations identified at the start and stop codons of timP. (E) LOF mutations identified in the timP pseudoknot. N/A: not applicable. Green color indicates LOF mutations.

LOF mutations in the timP 5′UTR abolish translation. The two top panels show Western blot (WB) analysis of TimP3xFLAG expression in the indicated LOF mutants. The timP variants were expressed from an arabinose-inducible promoter on a plasmid. GroEL served as a loading control. The two bottom panels show Northern blot (NB) analysis of timP3xflag mRNA levels for the indicated LOF mutants. Probing of 5S rRNA served as loading control.

An interaction between a previously predicted single-stranded region and SL4 is required for timP translation. (A) Secondary structure representation of the timP 5′UTR. Mutations designed to disrupt and restore the respective interactions are highlighted in red. (B) Western (WB) and Northern blots (NB) monitoring expression from the indicated timP3xflag constructs. The timP variants were expressed as in Fig. 2. An unspecific band served as loading control for Western blot. Probing of 5S rRNA served as loading control in the Northern blot. (C) Toxicity resulting from arabinose-induced overexpression of wild-type or mutant timP. (D) Secondary structure representation of timP SL4. Mutations predicted to destabilize SL4 are highlighted in red. (E) Western blot monitoring TimP3xFLAG expression. The tested timP variants were expressed from an arabinose-inducible promoter on the chromosome. GroEL served as loading control. (F) Quantification of heteroduplex formation between an antisense oligo and the SD sequence as a function of RNase H cleavage. Cleavage of the indicated mutants was normalized relative to wild type. Statistical analysis was performed using two-tailed t-test, *P < .0005, P < .005, *P < .05, ns: nonsignificant.

Site 1–2 interaction determines the inactive conformation of timP 5′UTR. (A) Secondary structure representation of the timP 5′UTR. Mutations designed to disrupt and restore the respective interactions are highlighted in red. (B) Alignment of complementary regions of sites 1, 2, 3, and 4 in the indicated enterobacterial species, S. enterica serovar Typhimurium, Serratia proteamaculans, R. aquatilis. Nucleotides in red highlight changes compared to the Salmonella sequence. (C) RNase T1 cleavage of in vitro transcribed timP and indicated mutants in the presence or absence of TimR. (D) Quantification of RNase T1 cleavage at position G76 based on two independent experiments. For each lane, the band intensity at position G76 was first normalized to that of position G131. Then, for each timP variant, values were divided by the timP sample lacking TimR (control). Average values and standard deviation are shown. (E) Western blot monitoring TimP3xFLAG expression. The timP variants were expressed from an arabinose-inducible promoter on a plasmid. GroEL served as loading control.

Schematic model for posttranscriptional regulation of the timPR T1TA system. The translationally inactive conformation of the timP 5′UTR is characterized by the interaction between sites 1 and 2, and SL4, which prevents 30S access to the SD. A structural rearrangement results in the formation of a pseudoknot structure between sites 1 and 3, which in turn allows site 2 to bind to site 4 to release the SD sequence from SL4. This permits 30S binding to the RBS and synthesis of the TimP toxin. Alternatively, the sRNA TimR binds on SL2, which abolishes pseudoknot formation and results in inhibited translation.

Full text

Introduction

Translation initiation is the rate-limiting step of protein synthesis in bacteria [1, 2]. Central to this process is the binding of the 30S small ribosomal subunit at the messenger RNA (mRNA) ribosome binding site (RBS), which includes the Shine–Dalgarno (SD) sequence and the start codon. The complementarity between the SD sequence in mRNA and the anti-SD sequence of 16S rRNA specifies the binding of 30S to the RBS. The subsequent joining of initiator fMet-tRNA^fmet^ at the start codon and the initiation factors establishes the pre-initiation complex [1, 3, 4]. The central function of the RBS in initiation makes it a hot spot for regulation. Trans-acting factors such as small RNAs (sRNAs) and RNA-binding proteins affect the availability of the RBS for 30S binding to tune translation rates [5–7]. Likewise, intramolecular stem-loop structures forming at the RBS are a common feature in posttranscriptional control [8–10]. Since formation of the pre-initiation complex requires an unfolded RBS, the initiation rate is inversely proportional to the stability of RBS-sequestering structures [11]. Regulatory mechanisms that rely on folding at the RBS include translational riboswitches [12, 13], where ligand binding-induced switching between mutually exclusive RNA structures either promote or abolish 30S access to the RBS, and RNA thermometers [14], in which RBS-blocking stem-loops unfold at specific temperatures to allow translation initiation. RBS-sequestering structures can also be affected by translation itself: translation of, or ribosome stalling within, upstream open reading frames (ORFs) can destabilize a downstream RBS-sequestering stem-loop, thereby promoting initiation [15, 16]. Alternatively, 30S can be recruited to so-called ribosome standby sites [10, 17]. This was first described in the MS2 coat protein-encoding mRNA, where an unusually stable stem-loop structure prevents direct 30S access to the RBS [18]. Translation initiation instead relies on 30S binding to “standby” sites—single-stranded regions in the vicinity of the stem-loop—and subsequent accommodation of 30S at the RBS when the stem-loop transiently unfolds [18].

Posttranscriptional control involving RBS-sequestering structures is particularly common among genes encoding toxins of type I toxin–antitoxin systems (T1TAs). Here, the primary mRNA is typically translationally inert due to intramolecular RBS sequestration [9, 10]. Translation occurs only after the mRNA has undergone processing, during which a 5′ or 3′ sequence is removed by ribonucleolytic cleavage. The processed mRNA becomes translationally active either by a structural change that renders the RBS available, or by the formation of a ribosome standby site. Concomitantly, the mRNA becomes amenable for binding by the antitoxin sRNA, which inhibits translation by steric hindrance of the RBS or standby site, and/or by promoting RNase III-mediated cleavage of the mRNA–sRNA duplex to produce a translationally inactive mRNA [9, 10].

We have recently characterized the enterobacterial T1TA system timPR. The TimP toxin inserts into the bacterial inner membrane to induce membrane leakage and cell growth inhibition [19]. TimR, the antitoxin sRNA, inhibits translation initiation by direct binding to timP mRNA. Initial characterization of the timP 5′ untranslated region (UTR) indicated that synthesis of TimP relies on a mechanism of translation initiation that does not follow the canonical model of ribonucleolytic activation. The timP 5′UTR folds into four stable stem-loops (SL), one of which, SL4, sequesters the SD sequence [20] (Fig. 1A). Biochemical data further suggested that the primary timP mRNA can adopt more than one structural conformation, and that one of these features a pseudoknot formed between SL1 and SL3 (Fig. 1A). Formation of the pseudoknot is essential for translation, and strongly stimulates TimR binding, which destabilizes the SL1–SL3 interaction and, hence, inhibits translation [20]. Although these observations offered a plausible route for ribosome engagement in the absence of an exposed RBS, they did not clarify a central mechanistic question: how the 30S subunit subsequently gains access to the downstream RBS, which remains structurally occluded in predicted models of the timP mRNA. Thus, the molecular events that allow progression from pseudoknot formation to productive translation initiation at timP remain incompletely understood.

FASTBAC-Seq identifies single point loss-of-function (LOF) mutations in timP mRNA. (A) Schematic representation of the timP 5′UTR secondary structure in its active and inactive conformation. Yellow: the pseudoknot-forming sequences; gray: the TimR-binding site; blue: SD and start codon. (B) Schematic representation of the FASTBAC-Seq method. The timP gene from Salmonella enterica, flanked by either a constitutive promoter or a promoter-less sequence and homologous recombination sites (brown boxes), was amplified under error-prone conditions and recombined into the Escherichia coli chromosome. DNA was extracted from the resulting colonies, followed by polymerase chain reaction (PCR) amplification, high-throughput sequencing and statistical analysis to identify LOF mutations (see the “Materials and methods” section for details). (C) LOF point mutations identified by FASTBAC-Seq throughout the timP mRNA sequence. (D) LOF mutations identified at the start and stop codons of timP. (E) LOF mutations identified in the timP pseudoknot. N/A: not applicable. Green color indicates LOF mutations.

Saturation mutagenesis followed by phenotypic sorting and deep sequencing has been successfully applied to decipher posttranscriptional regulatory mechanisms in bacteria [21–23]. Functional AnalysiS of Toxin–Antitoxin Systems in BACteria by Deep Sequencing (FASTBAC-Seq) exploits the growth inhibitory activity of type I toxins to select for mutations that abolish toxicity [24–26]. In the current study, we used FASTBAC-Seq to uncover determinants of translation initiation at the timP mRNA. This strategy uncovered a new long-range RNA interaction that directly competes with formation of the SD-sequestering SL4. Based on the results obtained, we propose a new mechanistic model in which translation initiation at timP requires the formation of (i) a pseudoknot that likely facilitates 30S recruitment, and (ii) a long-range RNA interaction that liberates the RBS to enable initiation complex formation. Together, these findings identify an unexpected layer of structural control and expand the repertoire of translation initiation strategies in bacteria.

Material and methods

Growth conditions

Bacterial strains used in this study are listed in Supplementary Table S3. Bacteria were grown aerobically at 37°C, 200 rpm in LB medium supplemented with 0.2% glucose and/or antibiotics, if applicable. Chloramphenicol (12.5 µg/ml) was used as a selection marker, unless otherwise specified. Expression from the ParaBAD promoter was induced by addition of 0.2% L-arabinose at an OD_600_ of 0.3–0.5 for 30 min.

Cloning and strain construction

Plasmids, oligonucleotides, and synthetic DNA sequences constructed and/or used in this study are listed in Supplementary Tables S4–S6, respectively. Plasmids encoding timP mutants were constructed by PCR followed by DpnI (Thermo Scientific) treatment at 37°C for 1 h, heat inactivation at 85°C for 10 min, and ligation with T4 ligase (Thermo Fisher Scientific) at RT for 2 h or O/N, if applicable. Strains encoding chromosomal timP constructs under an arabinose inducible promoter were constructed by λ-red recombination, as previously described [27, 28]. First, expression of the λ -red recombination genes was induced at 42°C for 15 min, followed by transformation of the PCR-amplified timP fragments via electroporation. The λ-red plasmid was then removed by incubation at 42°C O/N, and the clones were selected using the counter-selection cassette.

FASTBAC-Seq experiment

The FASTBAC-Seq experiment was carried out as described [26] with minor modifications. A DNA sequence encoding timP mRNA from S. enterica serovar Typhimurium, preceded either by the PJ23106 promoter [29], or by an incomplete version of the native timP promoter encompassing only the −10 box, and flanked with homology regions for homologous recombination in E. coli, were synthesized by Eurofins (GLR11 and GLR12, respectively; Supplementary Table S6). These DNA sequences were amplified by 30 cycles of PCR with oligos LRO3 and LRO4 using Dream Taq DNA polymerase (Thermo Fisher Scientific) and purified. Hundred nanograms of purified DNA samples were further mutagenized using the GeneMorph II Random Mutagenesis kit (Mutazyme II DNA polymerase, Agilent) as per the manufacturer’s instructions with 30 PCR cycles. The PCR reactions were purified and 200 ng of the DreamTaq and the Mutazyme II PCR samples were recombined into the E. coli XTL632 chromosome [30] following the protocol described in [31]. Recombinant bacteria were selected on Tet/SacB counter-selection plates [31] at 42°C. Colonies were harvested from the plates, and DNA was extracted. The experiment, including the PCR and mutagenesis steps, was repeated three times. The number of collected colonies ranged from 13 000 to 116 000 per transformation (Supplementary Table S7). Extracted DNA was amplified using primers LRO394 and LRO395 to add Illumina adapters using the KAPA HiFi HotStart ReadyMix (Roche) or the Phusion Hot Start polymerase (Thermo Scientific). The amplicons were sequenced at the Bordeaux Transcriptome Genome Platform (PGBT, Cestas-Pierroton, France) using an Illumina NextSeq 2000 P1 instrument in paired-end mode 2× 300 nt (overlapping reads).

FASTBAC-Seq data analysis

Data were processed following the workflow described in [26] with minor modifications. Quality control and read assembly was performed as described previously [26]. Assembled reads were aligned to the timP mRNA reference sequences (GLR11 or GLR12; Supplementary Table S6) using BWA-MEM [62] with parameters -A 1 -B 4 -O 6 -E 1 -L 10,10, recovering between 1.5 and 2.4 million reads per sample (Supplementary Table S7).Reads matching the expected length (330 or 308 nt for GLR11 and GLR12, respectively) were extracted using BamTools v2.5.2 [32]. Sequences containing no substitutions in the promoter region and a single nucleotide substitution in the timP mRNA sequence were retrieved using utilities from the GenomicAlignments 1.38.2 [33], Rsamtools 2.18.0, and dplyr 1.1.4 packages in R 4.3.2 (https://www.R-project.org/; Supplementary Table S7). Differential analysis was performed to compare single substitution frequencies at each position between “no promoter” and “promoter” constructs using Trinity 2.15.1 [34] and the R package DESeq2 1.42.1 [35]. Samples amplified with DreamTaq and Mutazyme II were analyzed independently, and, in both cases, replicates amplified by KAPA and Phusion were pooled. “No promoter” and “promoter” biological replicates were paired. Substitutions were considered significantly over- or under-represented when the false discovery rate (FDR)-adjusted P-value (Padj) was ≤.05 and the absolute log_2_-fold change (|log2FC|) was ≥1. The last five positions at the 3′ end of the timP mRNA could not be included in the statistical analyses as there were no substitutions at these positions in any of the “no promoter” and “promoter” replicates. The DNA sequencing reads and results of the differential statistical analysis were deposited in the NCBI GEO (Gene Expression Omnibus) database under accession GSE313366.

Western blotting

After 0.2% arabinose induction for 30 min, bacterial cultures were pelleted and resuspended in Glycine–SDS–PAGE sample buffer supplemented with beta-mercaptoethanol. Samples were normalized to 0.05 OD_600_/µl, denatured at 95°C for 5 min, separated on Mini-PROTEAN^®^ TGX Stain-Free protein gels (Bio-Rad), and transferred to 0.2 μm pore size polyvinylidene difluoride membranes (Bio-Rad). When total protein load was used as loading control, a stain-free image of the gel was taken before transfer. The membranes were blocked overnight in 5% bovine serum albumin (Sigma–Aldrich) in Tris-buffered saline supplemented with 0.05% Tween 20 (Thermo Scientific). FLAG-tagged proteins were detected with a 1:10 000 diluted HRP-conjugated anti-FLAG mouse antibody (Sigma–Aldrich), and GroEL was detected by a 1:50 000 diluted HRP-conjugated anti-GroEL mouse antibody (Sigma–Aldrich).

RNA extraction

After 0.2% arabinose induction for 30 min, bacterial cultures were mixed with 0.2 volumes of stop solution (95% ethanol, 5% phenol) and immediately frozen in liquid nitrogen. Following centrifugation, bacterial pellets were resuspended in 1× TE buffer [10 mM Tris, pH 8.0, 1mM ethylenediaminetetraacetic acid (EDTA)], supplemented with 0.5 mg/ ml lysozyme. One per cent sodium dodecyl sulfate (SDS) was added in the suspension, and the samples were incubated at 64°C for 2 min. Sodium acetate pH 5.2 was added at final concentration of 300 mM, followed by addition of 1 vol of acidic phenol, and the samples were incubated for 6 min at 64°C. After incubation, samples were centrifuged, followed by chloroform extraction and ethanol precipitation. The extracted RNA was dissolved in sterile water and stored at −20°C. RNA integrity was determined by agarose gel electrophoresis.

Northern blotting

Total RNA samples (3.5 µg) were separated on a 6% polyacrylamide/8 M urea gel. Radioactively labeled pUC19 DNA/MspI (HpaII) Marker (Thermo Fischer) was used as a size ladder. The RNA was transferred to a Hybond-N+ nitrocellulose membrane (Amersham, Cytiva), 0.36 A, 17 V at 4°C for 2 h in Tris–borate–EDTA (TBE) buffer, followed by crosslinking by UV light exposure at 1200 mJ and prehybridization with church buffer (0.5 M sodium phosphate buffer, pH 7.2, 1 M EDTA, 7% SDS) for 45 min at 42°C. Radioactively labeled EHO-690 (targeting 5S rRNA) or EHO-1344 (targeting timP mRNA) was added to the hybridization buffer and incubated overnight at 42°C. The membrane was then washed in 2× saline-sodium citrate buffer/0.1% SDS, and exposed to a phosphor screen.

Toxicity assays

Bacterial cultures were pelleted and resuspended in 0.9% NaCl before and after induction with 0.2% arabinose. Ten-fold dilution series of bacterial cultures were prepared in 0.9% NaCl. From each dilution, 5 µl were spotted on LA plates supplemented with chloramphenicol (12.5 µg/ml) and incubated O/N at 37°C.

Sequence and structural conservation alignments

timP sequences of sites 1, 2, 3, and 4 of Serratia proteamaculans 568 (genome accession: CP000826.1), Rahnella aquatilis CIP 78.65 (genome accession: CP003244.1), Hafnia alvei FB1 (genome accession: CP009706.1), E. coli strain K-12 subst. MG1655 (genome accession: CP097884.1), S. enterica subsp. enterica serovar Typhimurium SL1344 (genome accession: FQ312003.1), Citrobacter amalonaticus Y19 (Sequence ID: CP011132.1), and Shigella flexneri 2002017 (genome accession: CP001383.1) were aligned using LocARNA with default parameters [36].

In vitro transcription, purification, dephosphorylation, and labeling of RNA

DNA templates carrying a T7 promoter were generated by PCR and served as templates for in vitro transcription using the MEGAscript kit (LifeTechnologies) or following the protocol used in the study by Uroda et al. [37]. In this case, Buffer 1 [37] was used as transcription buffer, and the RNA was eluted in sterile water after column separation as final step. Transcription reactions were separated by 5% polyacrylamide/8 M urea gel electrophoresis, followed by O/N gel extraction in elution buffer (10 mM magnesium acetate tetrahydrate, 500 mM ammonium acetate, 1 mM EDTA, 0.1% SDS) at 4°C, phenol-chloroform extraction, ethanol precipitation, and dissolved in sterile water. If appropriate, the in vitro transcribed RNAs were dephosphorylated by Calf Intestinal Alkaline Phosphatase (Thermo Fisher Scientific), purified by phenol-chloroform extraction, followed by ethanol precipitation, and radioactively labeled at the 5′-end using γ-^32^P-ATP and T4 PNK (Thermo Fisher).

Electromobility shift assay (EMSA)

Radioactively labeled RNA was denatured for 1 min at 95°C in sterile water, cooled on ice for 2 min, diluted in EMSA buffer (25 mM Tris–HCl, pH 7.4, 100 mM NaCl, 1 mM MgCl_2_), and renatured for 15 min at 37°C. For the rate experiments, 0.5 nM of labeled RNA was mixed with 500 nM unlabeled RNA and samples were taken at indicated time intervals. Samples were immediately separated in running native 6% polyacrylamide gel in 0.5x TBE buffer, at 200V, 4°C.

RNase H assays

Radioactively labeled RNA was denatured for 2 min at 95°C in 0.5× TE buffer, cooled on ice for 2 min, and refolded in TMN buffer (20 mM Tris–HCl, pH 7.4, 2 mM MgCl2, 100 mM NaCl) at 37°C for 15 min. The mRNA was annealed with 1.4 µM DNA probe for 5 min at 37°C. 2.5 U RNase H (New England Biolabs) were added to the reaction and incubated for 15 min at 37°C. The reaction was stopped by phenol/chloroform (25/24, v/v, pH 5) extraction followed by ethanol precipitation. The pellet was dissolved in 10 µl sterile water and 10 µl of loading denaturing buffer was added before separation by polyacrylamide denaturing gel electrophoresis. The cleavage fraction was obtained by quantification of the RNase H cleavage product and normalization to wild-type timP. Five replicates were performed for the wild-type timP, and the S2*-timP, and two replicates for the M2-timP, M2′-timP, and M2 + M2′-timP. Quantification of the bands of interest was performed in Image Lab Software (Bio-Rad). Statistical analysis was performed using two-tailed t-test.

RNA structure probing

Radioactively labeled timP mRNA, and cold TimR sRNA were denatured separately for 2 min at 95°C in 0.5× TE buffer, cooled on ice for 2 min, and refolded in TMN buffer at 37°C for 30 min. Subsequently, labeled timP mRNA was mixed with 200 nM cold TimR sRNA, and the mixture was incubated at 37°C for 5 min. 0.1 U RNase T1 (Thermo Fisher Scientific) was added to the reaction, and incubated for 30 s. The reactions were stopped by adding 5 µl 0.1 M EDTA and kept on ice, followed by phenol-chloroform extraction and ethanol precipitation. The samples were diluted in sterile water and Gel loading buffer II (Invitrogen), heated at 95°C, and separated on 8% polyacrylamide sequencing gels in 1× TBE buffer at 38 W. Single nucleotide ladders were prepared by subjecting denatured RNA to alkaline hydrolysis according to the manufacturer (Ambion). Single G-nucleotide ladders were prepared by RNase T1 cleavage of denatured RNA. Quantification was performed in Image Lab Software (Bio-Rad).

Results

FASTBAC-Seq identifies timP loss-of-function mutations

Our previous work suggested the timP mRNA as an attractive model for noncanonical translation initiation mechanisms. Importantly, we showed that in the context of a structurally sequestered RBS, an RNA pseudoknot is essential for initiation [20] (Fig. 1A). However, this did not explain how the presence of this structure would aid 30S initiation at the sequestered timP RBS. We therefore employed an unbiased method, FASTBAC-Seq [24, 26], to positively select single point mutations that abolish timP translation. A DNA fragment encompassing the timP gene from S. enterica preceded by a heterologous promoter was PCR-amplified under error-prone conditions to create a large library of timP mutants (Fig. 1B), which was recombined into the E. coli chromosome and plated on agar plates. Since high expression of TimP inhibits growth, colonies can only form if a PCR-induced mutation prevents TimP expression or renders the protein nontoxic. A promoter-less timP mutagenesis library displaying unaffected growth of all timP variants was used as control (Fig. 1B). Pools of tens of thousands of colonies (three replicates of each library) were subjected to amplicon sequencing and differential statistical analysis. This identified single substitutions that were significantly enriched FDR-adjusted P-value ≤.05, log_2_-fold change ≤−1 in the expressed over the nonexpressed libraries (Fig. 1B, Supplementary Fig. S1A, and Supplementary Table S1). This gave a total of 230 significantly enriched single point LOF substitutions (Fig. 1C and Supplementary Fig. S1B). PCR-based mutagenesis typically results in a bias for transitions over transversions [38–40]. Indeed, 71% of the mutations in the promoter-less library were transitions (Supplementary Fig. S1C). In contrast, among the LOF mutations, transversions were almost as frequent as transitions for both DNA polymerases used (Supplementary Fig. S1C), indicating a strong selective pressure for mutations that permitted growth under the assayed conditions. The LOF mutations mapped to all regions of the timP mRNA, including the 5′UTR, ORF, and 3′UTR (Fig. 1C, Supplementary Fig. S2, and Supplementary Table S1). As expected, all possible LOF mutations rendering the start or stop codon nonfunctional were enriched among the LOF mutations, while mutations preserving function (AUG to GUG, UAA to UGA/UAG) were not (Fig. 1D). LOF mutations in the ORF were dominated by missense and nonsense mutations (Supplementary Table S2), the majority of which were found in the putative signal peptide (first 20 amino acids). These mutations likely interfere with TimP targeting to the inner membrane [19]. In the 3′UTR, most LOF mutations interfered with base-pairing of the terminator stem (Supplementary Fig. S2), likely causing aberrant termination and mRNA destabilization [25, 26]. In the 5′UTR, several LOF mutations mapped to the pseudoknot, in line with its essentiality for translation initiation. Each of these mutations breaks or destabilizes a base-pair in the pseudoknot interaction site (Fig. 1E). Many LOF mutations were identified in stem-loops SL1, SL2, and SL3 (Supplementary Fig. S2), indicating that the integrity of these major structural elements is key for timP expression [20]. Mutations that destabilize SL1 and SL3 likely affects the ability to form the pseudoknot. SL2 harbors the binding site for TimR, which upon binding destabilizes SL2 and inhibits translation initiation. Hence, destabilizing LOF mutations in SL2 mimic TimR binding and should result in translation inhibition. In contrast to SL1–SL3, few destabilizing LOF mutations were found in SL4. This is expected since SL4 sequesters the SD sequence; mutations that destabilize SL4 should yield increased TimP synthesis and growth inhibition. Together, these results indicate that the mutagenesis strategy stringently enriched for mRNA variants with defects in mRNA stability, transcription termination, translation initiation, or protein function.

A long-range interaction renders the timP SD single-stranded

To obtain further insight into determinants for timP translation initiation, several 5′UTR LOF mutations were introduced into a plasmid encoding a timP-3xflag allele [19, 20]. Western blot analysis confirmed that seven out of eight tested LOF mutations abolished TimP protein synthesis, whereas timP-3xflag mRNA levels were only modestly affected (Fig. 2). Thus, the tested mutations primarily affect translation. Notably, a cluster of three adjacent nucleotides—C57, C58, and G59 (Supplementary Fig. S2)—exhibited strong enrichment for LOF mutations (Supplementary Table S1), tentatively suggesting that these positions participate in base-pairing interactions with other regions of the mRNA. Indeed, we identified a six-nucleotide region encompassing C57, C58, and G59 with full complementarity to the 5′-flank of SL4 (Fig. 3A). Since this putative interaction destabilizes SL4, it could facilitate 30S binding at the RBS by rendering the SD sequence single-stranded. For clarity, we denote these two putative interaction sequences as sites 2 and 4, and the two sequences forming the pseudoknot between SL1 and SL3 as sites 1 and 3 (Fig. 3A). To test whether the putative site 2–4 interaction affects TimP translation, we performed Western blot analysis with timP mutants predicted to destabilize this interaction. Introduction of mutations S2* (C57G) or S4* (G125C + C138G), designed to weaken the 2–4 interaction, abolished translation of timP-3xflag (Fig. 3A and B). Note that mutation S4* was designed to maintain the stability of SL4 itself (Fig. 3A). Strikingly, combining S2* and S4* to restore site 2–4 base-pairing restored translation (Fig. 3B), albeit to levels lower than on the wild-type mRNA. Congruent results were obtained by toxicity assays, where the loss of toxicity observed in the single S2* or S4* mutants was partially restored in the S2*+S4* double mutant (Fig. 3C). This indicates that the 2–4 interaction is required for translation at timP and functions to unblock the SD from SL4. To validate this proposal, a mutant (“open SD”) was constructed that breaks four G-C base-pairs in SL4 to render the SD sequence accessible to 30S loading (Fig. 3D). In contrast to the detrimental effect of the S2* mutation in the context of wild-type SL4 (Fig. 3B and E), it did not affect translation in the “open SD” construct (Fig. 3E). Thus, translation becomes independent of site 2 when the SD sequence is already unblocked. Together, these findings indicate that translation of timP mRNA requires two long-range interactions; formation of a pseudoknot through site 1–3 pairing, and an interaction between sites 2 and 4 that destabilizes SL4 to render the SD sequence accessible for initiation complex formation.

LOF mutations in the timP 5′UTR abolish translation. The two top panels show Western blot (WB) analysis of TimP3xFLAG expression in the indicated LOF mutants. The timP variants were expressed from an arabinose-inducible promoter on a plasmid. GroEL served as a loading control. The two bottom panels show Northern blot (NB) analysis of timP3xflag mRNA levels for the indicated LOF mutants. Probing of 5S rRNA served as loading control.

*An interaction between a previously predicted single-stranded region and SL4 is required for timP translation. (A) Secondary structure representation of the timP 5′UTR. Mutations designed to disrupt and restore the respective interactions are highlighted in red. (B) Western (WB) and Northern blots (NB) monitoring expression from the indicated timP3xflag constructs. The timP variants were expressed as in Fig. 2. An unspecific band served as loading control for Western blot. Probing of 5S rRNA served as loading control in the Northern blot. (C) Toxicity resulting from arabinose-induced overexpression of wild-type or mutant timP. (D) Secondary structure representation of timP SL4. Mutations predicted to destabilize SL4 are highlighted in red. (E) Western blot monitoring TimP3xFLAG expression. The tested timP variants were expressed from an arabinose-inducible promoter on the chromosome. GroEL served as loading control. (F) Quantification of heteroduplex formation between an antisense oligo and the SD sequence as a function of RNase H cleavage. Cleavage of the indicated mutants was normalized relative to wild type. Statistical analysis was performed using two-tailed t-test, ***P < .0005, **P < .005, P < .05, ns: nonsignificant.

Pseudoknot formation promotes liberation of the SD sequence

We previously showed that the site 1–3 pseudoknot strongly enhances translation even when SL4 is replaced by the unstructured RBS from the cspE gene [20]. Consistent with this, the pseudoknot-abolishing mutant M2 in site 1 (G33C + C34G, Fig. 3A) strongly reduced translation of the “open SD” construct. Combining the M2 mutation with the compensatory site 3 mutation M2′ (G102C + C103G) restored translation (Fig. 3E). However, since a pseudoknot-disrupting mutation permits residual translation both in the “open SD” (Fig. 3E), and more strongly so in the *cspE-RBS construct [20], this argues that timP translation can initiate independently of the pseudoknot when the RBS is accessible. Conversely, in constructs harboring wild-type SL4, pseudoknot-breaking mutations entirely abolish translation (e.g. mutant M2; Fig. 3E, [20]). This suggests that pseudoknot disruption somehow restricts formation of the site 2–4 interaction. This was tested by assessing the fraction of timP molecules with an accessible SD sequence. Specifically, timP mRNA was incubated with an oligodeoxynucleotide complementary to the 3′-flank of SL4, which harbors the SD sequence (Supplementary Fig. S3A), and binding of the oligo was assessed by RNase H cleavage of the resulting RNA-DNA hybrid. Introduction of the S2 mutation significantly reduced cleavage compared to wild type, compatible with the 2–4 interaction promoting SL4 destabilization (Fig. 3F). While the M2′ mutation strongly reduced RNase H cleavage (Fig. 3F), the M2 + M2′ double mutation restored cleavage to wild-type levels. This indicates that formation of the pseudoknot indeed stimulates liberation of the SD through site 2–4 interaction. However, only one of the pseudoknot-disrupting mutations, M2′, but not M2, affected RNase H cleavage. This suggests that while pseudoknot disruption itself has a negative impact on formation of site 2–4 interaction, the sequence of site 1 (M2 mutation), but not site 3 (M2′ mutation), counteracts this effect.

Site 1–2 interaction is a negative determinant for timP translation

The data in Fig. 3 indicated not only that the 1–3 pseudoknot stimulates formation of the site 2–4 interaction but also that the sequence of site 1 is critical for this effect. Interestingly, visual inspection revealed that site 2 is complementary to site 1 of the pseudoknot (Fig. 4A). We therefore hypothesized that the inactive conformation of timP is stabilized by the site 1–2 interaction, since this prevents formation of both the 1–3 interaction and site 2–4 pairing. Sequence covariation analysis lends strong support, with conserved base-pairing between sites 1–2, 1–3, and 2–4 among timP homologs (Fig. 4B and Supplementary Fig. S3B). To test this experimentally, we took advantage of the fact that TimR binds the active timP conformation with substantially higher affinity than the inactive conformation [20]. Binding of TimR to timP was assessed by in vitro structure probing. Protection from RNase T1 cleavage at position G76 within the SL2 loop served as a diagnostic marker for timP–TimR duplex formation (Fig. 4C, [20]). Disrupting the putative site 1–2 interaction by mutation S2* led to increased TimR-dependent protection at G76, compared to wild-type timP, indicative of enhanced TimR binding (Fig. 4C and D). Conversely, the pseudoknot-abolishing mutation M2′ strongly decreased TimR-dependent protection at G76. Moreover, the M2′ mutation increased cleavage in the SL3 loop due to abolished pseudoknot formation. Combining mutations S2* and M2′ resulted in increased protection compared to the M2′ mutation alone (Fig. 4C and D). These results were corroborated by assaying timP–TimR binding using gel shift assays (Supplementary Fig. S4A). Together, these findings strongly suggest that the site 1–2 interaction is a key feature of the inactive conformation of timP, and hence inhibits both TimR binding and timP translation. This is also congruent with M2 and M2′ mutations having different effects on SL4 accessibility in the RNase H experiment (Fig. 3F); the M2 mutation prevents site 1–2 interaction and thus allows site 2–4 pairing, while the M2′ mutation promotes site 1–2 interaction by abolishing site 1–3 pairing. To probe all three interactions simultaneously, a new site 1 point mutation (S1*; G33C) was employed to destabilize the site 1–2 and 1–3 interactions and simultaneously restore the site 1–2 interaction when combined with mutation S2* (Fig. 4A). Similarly, the new mutation in site 3 (S3*; C103G) should alone destabilize 1–3 interaction and restore binding when combined with S1* (Fig. 4A). Western blot analysis showed that each of the mutations S1*, S2*, and S3* alone abolished translation, while combining S1*+S3* or S2*+S4* restored translation, thereby confirming the essentiality of 1–3 and 2–4 pairing for translation initiation (Fig. 4E). In contrast, combining mutations that either lock the structure in the inactive form (S1*+S2*) or prevent formation of both sites 1–3 and 2–4 (S1*+S4* and S3*+S4*) resulted in abolished or very low TimP synthesis (Fig. 4E). Similarly, we did not manage to detect TimP_3xFLAG_ protein when we introduced site 2 mutations in the previously validated timP-M2+M2′ construct (Supplementary Fig. S4B). Combining all four mutations (S1+S2+S3*+S4*), i.e. restoring all three interactions, again permitted translation (Fig. 4E). Thus, timP mRNA can adopt a translationally inactive conformation determined by interaction between sites 1–2, and a mutually exclusive and translationally active conformation determined by interactions between sites 1–3 and 2–4, in which formation of site 2–4 depends on formation of site 1–3. Both 1–3 and 2–4 interactions are required for efficient translation, suggesting initial 30S recruitment via the pseudoknot, followed by initiation complex formation at the RBS.

Site 1–2 interaction determines the inactive conformation of timP 5′UTR. (A) Secondary structure representation of the timP 5′UTR. Mutations designed to disrupt and restore the respective interactions are highlighted in red. (B) Alignment of complementary regions of sites 1, 2, 3, and 4 in the indicated enterobacterial species, S. enterica serovar Typhimurium, Serratia proteamaculans, R. aquatilis. Nucleotides in red highlight changes compared to the Salmonella sequence. (C) RNase T1 cleavage of in vitro transcribed timP and indicated mutants in the presence or absence of TimR. (D) Quantification of RNase T1 cleavage at position G76 based on two independent experiments. For each lane, the band intensity at position G76 was first normalized to that of position G131. Then, for each timP variant, values were divided by the timP sample lacking TimR (control). Average values and standard deviation are shown. (E) Western blot monitoring TimP3xFLAG expression. The timP variants were expressed from an arabinose-inducible promoter on a plasmid. GroEL served as loading control.

Discussion

Type I toxins strongly and negatively impact bacterial cell growth. This activity has been shown to contribute to processes such as postsegregational killing [41, 42], abortive phage infection [43], and persister formation [44], but is expected to have adverse effects on bacterial survival under conditions favoring growth. Not surprisingly, type I toxin genes are tightly regulated to allow expression only under specific conditions. The well-characterized type I toxin gene tisB serves as an example to illustrate this tight control. Under nonstress conditions, tisB transcription is repressed by the SOS response master regulator LexA [45, 46]; the same applies to the toxin gene dinQ [47, 48]. Since transcriptional control is intrinsically noisy, low levels of bursty transcription in a repressed state may lead to inadvertent toxin synthesis. In the case of tisB, this is counteracted by two posttranscriptional mechanisms. First, primary tisB mRNA transcripts are translationally inactive, and translation only occurs after ribonucleolytic processing of the mRNA. The same mechanism applies to several other type I toxin genes [47, 49–53], and likely evolved to avoid cotranscriptional translation. Second, the processed tisB mRNA is targeted by the antitoxin sRNA IstR-1, which upon binding inhibits translation and induces RNase III-mediated cleavage of the sRNA–mRNA duplex [46, 49]. Together, the transcriptional and posttranscriptional regulation ensures tight control under conditions where toxin expression is unwanted. However, upon DNA damage, LexA inactivation results in levels of tisB mRNA high enough to override antitoxin repression [46]. The toxin is then produced to induce growth inhibition, and in the case of tisB, to increase bacterial survival upon antibiotics challenge [44, 45, 54].

Compared to tisB and other T1TAs mentioned above, the timP case is different in that its mRNA does not undergo enzymatic processing. Full-length timP mRNA is an efficient template in in vitro translation assays and is efficiently inhibited by TimR [19, 20]. This raised the question whether the translational activity of the nascent timP mRNA is controlled by other means than through ribonucleolytic cleavage. This was partially addressed in our previous study, showing that timP mRNA can adopt different conformations, of which only one is translationally active [20]. The hallmark of the active conformation is a pseudoknot structure in the 5′UTR, which also explains why translation is sensitive to 5′ truncations. In line with the processed and active form of tisB mRNA being the primary target for IstR-1 [46, 49, 55, 56], TimR binds with higher affinity to the active conformation of timP [20]. Together, this indicates that rather than ribonucleolytic processing, activation of the timP mRNA is through structural rearrangement.

A key feature that suppresses translation in toxin mRNAs is RBS sequestration in stable stem-loops. In the processed tisB mRNA, this hurdle is overcome by protein S1-dependent recruitment of 30S at an upstream standby site followed by movement of 30S/S1 along the mRNA until it reaches the RBS [49, 55–57]. Whether 30S access to the tisB RBS ultimately relies on intrinsic and transient unfolding of the RBS-sequestering stem-loop, or 30S/S1 has an active role in this process, is so far unclear. An essential feature of the tisB standby site is a pseudoknot structure, which forms specifically in the processed form of the mRNA and serves as a recruitment element for 30S/S1 [56]. Since the timP pseudoknot is only present in the active conformation of the mRNA, and is strictly required for translation initiation, this suggests that it is directly involved in recruitment of 30S/S1. Subsequently, however, 30S needs access to the RBS, which is sequestered in the stable SL4. We previously speculated that 30S may move along the timP mRNA and access the RBS as SL4 transiently unfolds, in a similar fashion as in tisB [55]. However, the data presented here rather indicate that 30S access to the RBS is made possible only through a structural rearrangement of the 5′UTR.

Specifically, the results of this study indicate that the timP 5′UTR can adopt two mutually exclusive conformations determined by base-pairing between four different sites (Fig. 4A). The interaction between sites 1 and 2 is, together with SL4, the hallmark of the inactive conformation, while the site 1–3 (pseudoknot) and 2–4 (SL4 anti-determinant) interactions define the active conformation, as follows. Mutations and compensatory mutations in the site 2–4 interaction abolished and restored timP translation, respectively (Fig. 3B). A mutation in site 2 reduced binding of a DNA oligo in SL4 (Fig. 3F), as expected if 2–4 interaction competes with SL4 formation. Upon artificial destabilization of SL4 (“open SD” construct), translation became independent on mutations in site 2 (Fig. 3E). Moreover, the 2–4 interaction is conserved in timP homologs (Fig. 4B and Supplementary Fig. S3B). Together, these findings indicate that the 2–4 interaction prevents SD sequestration in SL4 and promotes translation initiation at the RBS. If so, what is then the role of the pseudoknot (site 1–3 interaction)? First, in the “open SD” construct, as well as when the timP RBS was replaced by the unstructured cspE RBS [20], mutations abolishing pseudoknot formation strongly reduced translation levels, but did not fully abolish protein production (Fig. 3E). This suggests that the pseudoknot has an active role in initiation, probably acting as an enhancement element. We speculate that the pseudoknot attracts 30S to the mRNA to increase the local 30S concentration. In analogy with tisB, this may occur through protein S1, which binds RNA pseudoknots at high affinity [55, 56]. Second, formation of the pseudoknot also strongly affects opening of SL4. As shown in the RNase H assay (Fig. 3F), a pseudoknot mutation prevents a DNA oligo from binding to SL4, suggesting that the pseudoknot stimulates formation of the 2–4 interaction. A key to understanding the connection between the pseudoknot and the 2–4 interaction was the identification of an additional interaction between sites 1 and 2 (Fig. 4A). Since the 1–2 interaction prevents both the 1–3 (pseudoknot) and 2–4 interactions, formation of the site 1–3 interaction will release site 2 to interact with site 4, liberating the SD sequence. Additional evidence for this came from timP structure probing and timP–TimR binding assays, where mutations preventing site 1–2 pairing promoted TimR binding, and vice versa (Fig. 4 and Supplementary Fig. S4A). Taken together, we propose a model in which transcription yields timP mRNA molecules in an inactive conformation (Fig. 5). A structural rearrangement allows 30S binding at the pseudoknot and stimulates liberation of the RBS to allow initiation complex formation.

Schematic model for posttranscriptional regulation of the timPR T1TA system. The translationally inactive conformation of the timP 5′UTR is characterized by the interaction between sites 1 and 2, and SL4, which prevents 30S access to the SD. A structural rearrangement results in the formation of a pseudoknot structure between sites 1 and 3, which in turn allows site 2 to bind to site 4 to release the SD sequence from SL4. This permits 30S binding to the RBS and synthesis of the TimP toxin. Alternatively, the sRNA TimR binds on SL2, which abolishes pseudoknot formation and results in inhibited translation.

The presented findings raise a number of questions for future research to address. One important issue is when and how the switch between inactive and active conformations occurs. Since sites 1 and 2 are transcribed before sites 3 and 4, the 1–2 interaction could lock the native mRNA in the inactive conformation cotranscriptionally. This is reminiscent of the formation of early, metastable structures in the hok and aapA3 toxin mRNAs, which ensure sequestration of the SD sequence in a stable stem-loop [25, 58]. Understanding the temporal order of folding events during timP transcription would require single-molecule experiments. This could also shed further light on how TimR inhibits timP translation. We previously showed that TimR binding to the active conformation of timP prevents formation of the site 1–3 pseudoknot, but the details of this phenomenon were unclear. The structure probing experiment shown in Fig. 4C indicate protection of site 2 upon TimR binding to wild-type timP. However, this was not observed with mutants having the S2* mutation, where site 1–2 pairing is prohibited. Contrarily, the M2′ mutation led to decreased RNase T1 cleavage in site 2, consistent with this mutation favoring site 1–2 interaction. This suggests that TimR not only prevents pseudoknot formation but also favors site 1–2 pairing, maintaining the SL4 structure, and subsequent occlusion of the RBS. Another issue concerns the transition from an inactive to an active structure. Is it stochastic or induced? In our previous work, we observed transitions between timP structural conformations in vitro [20], suggesting that the mRNA can change structure without additional factors. Here, we show that the interaction between site 2–4 can form in pure timP mRNA in vitro (Fig. 3F). On the other hand, the thermodynamic stability of the mutually exclusive structures 2–4 and SL4 is very different, which SL4 being much more stable (ΔG of −11 kcal/mol versus −7 kcal/mol). This may indicate that a long-lived structural conformation including the site 2–4 interaction may require additional factors. Such factors may include 30S with S1, or S1 alone. In addition, it has not escaped our notice that the model suggested here, involving two mutually exclusive conformations, in some respects resembles translational riboswitches [59–61]. Therefore, it cannot yet be ruled out that, e.g. small ligands could trigger the transitions between timP conformations in vivo.

Supplementary Material

gkag240_Supplemental_Files