An RNA thermometer dictates production of a secreted bacterial toxin

Frequent transitions of bacterial pathogens between their warm-blooded host and external reservoirs are accompanied by abrupt temperature shifts. A temperature of 37°C serves as reliable signal for ingestion by a mammalian host, which induces a major reprogramming of bacterial gene expression and metabolism. Enteric Yersiniae are Gram-negative pathogens accountable for self-limiting gastrointestinal infections. Among the temperature-regulated virulence genes of Yersinia pseudotuberculosis is cnfY coding for the cytotoxic necrotizing factor (CNFY), a multifunctional secreted toxin that modulates the host’s innate immune system and contributes to the decision between acute infection and persistence. We report that the major determinant of temperature-regulated cnfY expression is a thermo-labile RNA structure in the 5’-untranslated region (5’-UTR). Various translational gene fusions demonstrated that this region faithfully regulates translation initiation regardless of the transcription start site, promoter or reporter strain. RNA structure probing revealed a labile stem-loop structure, in which the ribosome binding site is partially occluded at 25°C but liberated at 37°C. Consistent with translational control in bacteria, toeprinting (primer extension inhibition) experiments in vitro showed increased ribosome binding at elevated temperature. Point mutations locking the 5’-UTR in its 25°C structure impaired opening of the stem loop, ribosome access and translation initiation at 37°C. To assess the in vivo relevance of temperature control, we used a mouse infection model. Y. pseudotuberculosis strains carrying stabilized RNA thermometer variants upstream of cnfY were avirulent and attenuated in their ability to disseminate into mesenteric lymph nodes and spleen. We conclude with a model, in which the RNA thermometer acts as translational roadblock in a two-layered regulatory cascade that tightly controls provision of the CNFY toxin during acute infection. Similar RNA structures upstream of various cnfY homologs suggest that RNA thermosensors dictate the production of secreted toxins in a wide range of pathogens.


Introduction
Throughout their infection cycle, bacterial pathogens encounter changing nutritional and physical conditions. To withstand these potentially harmful cues, pathogens constantly monitor multiple external and internal parameters. Temperature is among the most important signals upon entry into or exit from the host and induces profound changes in gene expression and metabolism [1]. Bacterial pathogens have evolved diverse signal transduction cascades that adjust gene expression as soon as they are confronted with elevated temperatures (37˚C) in the warm-blooded host. Such a shift in temperature has an impact on almost any biomolecule. For example, it influences the activity or conformation of proteins and their interaction with membrane lipids. Moreover, it can affect the spatial arrangement of DNA and the structure of RNA [1][2][3][4]. The latter is particularly relevant in RNA thermometers (RNATs) that rearrange their structure in a temperature-dependent manner and thereby control translation efficiency. They are found in the 5'-untranslated region (5'-UTR) or intercistronic regions (ICRs) of many bacterial mRNAs [5,6]. At low temperatures, a secondary structure occludes the Shine-Dalgarno (SD) sequence and/or the translation initiation codon leading to inhibition of ribosome binding and, ultimately, inhibition of translation. An increase in ambient temperature provokes a gradual melting of the secondary structure, which liberates the ribosome-binding site (RBS) enabling translation initiation. Because of their instantaneous and reversible zipper-like control mechanism, RNATs are well-suited to adjust expression of virulence genes in bacterial pathogens during their infection cycle [7,8].
Besides Yersinia pestis and Yersinia enterocolitica, the Gram-negative γ-proteobacterium Yersinia pseudotuberculosis is one of the three pathogenic species from the genus Yersinia. While infection with Y. pestis, the causative agent of plague, leads to a grave and rapidly progressing febrile illness with severe mortality rates, the enteric foodborne pathogens Y. pseudotuberculosis and Y. enterocolitica cause self-limiting gastrointestinal diseases such as enteritis, diarrhea, and mesenteric lymphadenitis [9,10]. To infect target tissues and to survive in them, pathogenic Yersinia species possess an armory of virulence factors. Among these are the type three secretion system (T3SS), the Yersinia outer proteins (Yop effectors), and the adhesin YadA [11], whose genes are tightly controlled in response to extracellular calcium (Ca 2+ ) concentration [12,13] and temperature [14]. Transcription of these virulence genes is activated by the low calcium response regulator (LcrF) which underlies an extensive thermoregulation cascade in Y. pestis and Y. pseudotuberculosis [15,16]. An RNAT in the yscW-lcrF intercistronic region plays a key role in this process as it mediates translational control of lcrF expression [15].
Previously, we experimentally mapped the structures of more than 1,750 Y. pseudotuberculosis RNAs by applying the parallel analysis of RNA structure (PARS) method [17] at three physiologically important temperatures. We were particularly interested in Shine-Dalgarno (SD) regions that unfold upon temperature elevation and identified at least 16 new RNATs controlling translation of genes from different functional categories [18]. A virulence-related thermometer candidate was found in the 5'-UTR of cnfY encoding the cytotoxic necrotizing factor (CNF Y ). In contrast to many other cell-targeting Yersinia virulence factors, CNF Y is not delivered via the T3SS but through extracellular membrane vesicles [19]. Once inside the host cell, this single-chain A-B toxin modulates the immune response by constitutively activating Rho GTPases and enhancing delivery of Yop effector proteins into target cells [20][21][22]. CNF Y activity ultimately leads to the induction of acute inflammatory responses and to the formation of necrotic areas in infected tissues [20]. Related CNFs were identified in uropathogenic Escherichia coli (UPEC) strains (CNF1-3) as well as in Y. pseudotuberculosis strains YPIII and IP2666 [23,24].
In this study, we characterized the thermoregulatory features of the cnfY RNAT. We show that it forms a new type of structure, which is thermo-labile and allows increased ribosome binding at body temperature. Point mutations that strengthen the anti-SD/SD interaction prevent CNF Y production at 37˚C. Y. pseudotuberculosis strains carrying such stabilized RNAT variants were avirulent in a mouse model demonstrating the importance of precise temperature sensing for successful host-microbe interactions.

Ethics statement
Animal work was performed in strict accordance with the European Health Recommendations of the Federation of Laboratory Animal Science Associations (FELASA). The protocol was approved by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit: animal licensing committee permission no. 33.19-42502-04-18/3034. Animals were treated with appropriate care and all efforts were made to minimize suffering.

Bacterial strains and plasmids
Bacterial strains used in this study are listed in S1 Table. Cells were grown in LB medium or BHI medium (Sigma-Aldrich, St. Louis, USA) at indicated temperatures. If required, antibiotics were added to the media at following final concentrations: ampicillin, 100 μg/ml; gentamycin, 10 μg/ml; kanamycin, 50 μg/ml. For the induction of the P BAD promoter the medium was supplemented with L-arabinose to a final concentration of 0.01% (w/v; E. coli) or 0.1% (w/v; Y. pseudotuberculosis). To generate YP216, the kanamycin resistance of strain YP147 was removed as described previously [25].

Plasmid construction
All utilized oligonucleotides and plasmids are summarized in S2 or S3 Tables, respectively. Enzymes for cloning were obtained from Thermo Scientific (St. Leon-Rot, Germany). PCR, DNA manipulations, DNA restrictions, and transformations were performed according to standard protocols [26]. Point mutations were generated by site-directed mutagenesis according to the instruction manual of the QuikChange mutagenesis kit (Agilent Technologies, Santa Clara, USA). All constructed plasmids were confirmed by restriction analysis and automated sequencing (Eurofins Genomics, Ebersberg, Germany).

Reporter gene activity assays
For β-galactosidase activity assays, E. coli DH5α or Y. pseudotuberculosis YPIII cells carrying the cnfY 5'-UTR:bgaB fusion plasmids were grown overnight in LB with ampicillin at 25˚C. Before being inoculated with an overnight culture (OD 600 = 0.2), LB media supplemented with ampicillin was pre-warmed to 25˚C. After growth to an OD 600 = 0.5, transcription was induced with 0.01% w/v (E. coli) or 0.1% w/v (Y. pseudotuberculosis) L-arabinose. The culture was split up and shifted to pre warmed 100 ml flasks (temperatures indicated in the respective figure). The cultures were incubated for 30 min and 400 μl samples were subsequently taken for βgalactosidase assay. The β-galactosidase assay was carried out as described previously [18,27]. Standard deviations were calculated from three technical replicates.
The gfp reporter gene assay was performed as follows: E. coli DH5α or Y. pseudotuberculosis YPIII cultures carrying the wild type of the mutated cnfY 5'-UTR:gfp fusion plasmids were grown and treated as described above. Afterwards, 2 ml samples were taken for Western blot analysis and 4 ml samples were taken for total RNA isolation and subsequent Northern blot analysis.

Northern blot analysis
Northern blot analysis was performed as described previously [28]. For detection of gfp transcript a 286 bp fragment was PCR-amplified from pBAD2-gfp. RNA probes were derived from in vitro transcription with digoxigenin (DIG) labeled nucleotides (Roche, Mannheim, Germany).

Quantitative real-time PCR (qRT-PCR)
Samples for comparative qRT-PCR and Western blot analysis of cnfY expression were taken from the same cultures. Transcript abundance was determined from three independent cultures measured in duplicate. The non-thermoregulated reference genes gyrB and nuoB were used for normalization [29]. After isolation, 10 μg total RNA were treated with DNase (DNAfree Kit, Invitrogen, Carlsbad, United States) followed by cDNA synthesis (iScript cDNA synthesis Kit, Bio-Rad, Munich, Germany) according to the manufactures protocols (1 μg RNA per reaction). Two μl of diluted cDNA (1:10) were mixed with 0.25 μl of each primer (10 ρM), 5 μl of iTaq Universal SYBR Green Supermix, and 2.5 μl sterile water (Carl Roth GmbH, Karlsruhe, Germany). Amplification and detection were performed in a CFX Connect Real-Time System (Bio-Rad, Hercules, United States) under conditions optimized for each gene target. For analysis of each gene target, a six-point standard curve was included to ensure satisfactory amplification efficiency and ascertain that all samples fall within the linear range of the standard curve. Relative cnfY transcript amounts were calculated using the ΔΔCt method [30] with experimental Ct values normalized to that of nuoB and gyrB. Primer efficiencies calculated by the CFX Maestro software were as follows: gyrB: 102.9%, nuoB: 102.7%, and cnfY: 98.4%.

In vitro transcription
RNAs for structure probing and primer extension inhibition experiments were synthesized in vitro by run-off transcription with T7 RNA polymerase (Thermo Scientific, St. Leon-Rot, Germany) from EcoRV-linearized plasmids (listed in S3 Table) as previously described [18].

Enzymatic structure probing
Structure probing of the 5'-UTR and 80 nt of cnfY was performed with in vitro transcribed RNA using pBO4465 or pBO4466 as template. The in vitro transcribed RNA was purified and dephosphorylated using CIP enzyme (Calf intestinal phosphatase, Thermo Scientific, Waltham, USA). The RNA was labeled with [ 32 P] at the 5 0 end as described elsewhere [31]. Partial digestions of radiolabeled RNA with ribonuclease T1 (0.0025 U) (Thermo Scientific, Waltham, USA) and T2 (0.056 U) (MoBiTec, Göttingen, Germany) were performed according to [32] at 25, 37, and 42˚C. For digestion with RNase T1 and RNase T2 a 5 x TN buffer (100 mM Tris acetate, pH 7, 500 mM NaCl) was used. An alkaline hydrolysis ladder was prepared as described before [31]. The T1-ladder was generated by using 30000 cpm labeled RNA. The RNA was heated with 1 μl sequencing buffer (provided with RNase T1) at 90˚C. Afterwards, the RNA was incubated with the enzyme at 37˚C for 5 min.

Primer extension and Primer extension inhibition analysis (toeprinting)
The cnfY transcriptional start site was mapped by primer extension as described previously [33]. Primer cnfY_RO_rv was used for reverse transcription of RNA isolated from bacteria that were grown at 37˚C. The DNA sequence reaction was performed with the same primer using the Thermo Sequenase cycle sequencing kit (USB, now Affymetrix) and plasmid pBO4465 as template.

Measurement of cnfY expression by bioluminescence
Y. pseudotuberculosis YPIII harboring the fusion plasmids pBO6500, pBO6501, or pBO6502 were diluted 1:50 in fresh LB from overnight cultures and grown to late exponential phase (4 h) at 25˚C or 37˚C. Reporter fusions (transcriptional and translational fusions to luxCDABE) emitting bioluminescence were measured in non-permeabilized cells with a Varioskan Flash (Thermo Scientific, St. Leon-Rot, Germany) using the SkanIt software (Thermo Scientific, St. Leon-Rot, Germany) or with a CLARIOstar plus reader (BMG Labtech, Ortenberg, Germany). The data are given as relative light units (RLU/OD 600 ) from three independent cultures measured in triplicate.

Mouse infection studies
Female BALB/c mice (6-and 8-week-old) were acquired from Janvier (Saint Berthevin Cedex, France) and housed under specific pathogen-free conditions in accordance with FELASA recommendations in the animal facility of the Helmholtz Centre for Infection Research, Braunschweig.
For survival assays, mice were infected orally via food pellets inoculated with approximately 2x10 9 bacteria of Y. pseudotuberculosis strains YPIII with the empty vector pJNS11 and YP147 (cnfY::KanR), harboring either the empty vector pJNS11 or the cnfY complementation vector pJNS10 (pP cnfY :cnfY + ) with either the wild type, the R1, or the R1+2 variant of the cnfY RNAT. Prior to infection, bacterial cells were washed three times and resuspended in sterile PBS. The infected mice were monitored daily for 14 days to determine the survival rate, the body weight and health status. Bacteria used for organ burden experiments were grown over night in LB medium at 25˚C, washed and resuspended in sterile PBS. Groups of 5 animals were infected orally (via inoculated food pellets) with approximately 2x10 8 bacteria of Y. pseudotuberculosis strains YPIII with the empty vector pJNS11 and YP147 (cnfY::Kan R ), harboring either the empty vector pJNS11 or the cnfY complementation vector pJNS10 (pP cnfY :cnfY + ) including the wild type, the R1, or the R1+2 variant of the cnfY RNAT. Five days after infection, mice were euthanized by CO 2 asphyxiation. The small intestine, the mesenteric lymph nodes, the liver, and the spleen were isolated and the Peyer's patches were removed from the small intestine. Subsequently, whole organs were weighed and homogenized in PBS at 30000 rpm for 20 sec using a Polytron PT 2100 homogenizer (Kinematica, Switzerland). To determine the bacterial load of the organs, serial dilutions of the homogenates were plated on LB plates with 0.5 μg/ml triclosan. The colony forming units (cfu) were counted and are given as cfu per g organ/tissue.

The cnfY transcript harbors a temperature-sensitive RNA element
Comparative RNAseq-based transcriptome profiling of Y. pseudotuberculosis grown at 25˚C and 37˚C, representing environmental and host body temperature, respectively, revealed an about 7-fold induction of cnfY transcripts at elevated temperature indicating transcriptional thermoregulation of cnfY expression [29]. To determine the corresponding protein levels of this toxin, we performed Western blot analysis using an antibody against CNF Y after growing Y. pseudotuberculosis YPIII at 25˚C or 37˚C (S1A Fig). Substantially elevated levels of CNF Y were detected at 37˚C compared to 25˚C regardless of whether cells were grown to exponential or early stationary growth phase. To distinguish between transcriptional and translational control, we performed qRT-PCR and Western blot analyses using simultaneously collected samples from exponentially grown cells. We found a four-fold increase of cnfY transcript and 10-fold increase of CNF Y protein ( Fig 1A; representative Western blot in S1B Fig). This finding strongly suggests an additional layer of thermoregulation at post-transcriptional level.
Our previous global analysis of the Y. pseudotuberculosis RNA structurome [18] had provided evidence for translational temperature regulation through unfolding of the cnfY 5'-UTR from 25˚C to 37˚C (drop in PARS values in particular around the SD and anti-SD regions; Fig  1B). The PARS-derived secondary structure of the cnfY 5'-UTR constitutes a stem-loop structure including the SD sequence (5'-AAGGAG-3') that partially pairs with a 5'-CAUCAGU-3' region ( Fig 1C). The intramolecular interaction region contains a bulged nucleotide (A29) and an internal loop (AG32-33, AG73-74) in the anti-SD/SD region, which are typical characteristics of temperature-sensitive RNA structures able to melt at elevated temperatures. The AUG translation start codon is part of a 9-nt loop, which separates the stem-loop in a short basal and a long terminal part. The overall sequence and structure is unrelated to the fourU-type RNAT upstream of Yersinia lcrF and does not resemble any of the known RNAT families.
Preliminary experiments in E. coli suggested that the cnfY 5'-UTR is a translational control element [18]. To characterize the regulatory features further, we established a new protocol for measuring bgaB-derived β-galactosidase activity in Y. pseudotuberculosis at 25˚C and 37˚C. The well-studied lcrF RNAT served as positive control [15]. The cnfY RNAT showed a clear induction of bgaB activity from 25 to 37˚C comparable to the positive control and regardless of whether being tested in E. coli or Y. pseudotuberculosis ( Fig 1D). Overall, these results show that cnfY expression is under combined transcriptional [29] and translational control.

Thermosensing by the cnfY RNAT is independent of the transcription start site
Based on transcriptional profiling data, we previously assumed a transcriptional start site (TSS) of cnfY, which is located 82 nt upstream of the AUG start codon (TSS A; Figs S2A and 2A). The corresponding 5'-UTR is able to confer thermal control in a reporter gene assay ( [18] and Fig 1D). Two observations made us reassess the TSS. First, two additional peaks in an earlier transcriptome profile by Nuss et al. ( [29] and S2A Fig) and secondly, a striking almost identical duplication further upstream of the RNAT in the Y. pseudotuberculosis YPIII genome ( S2A Fig and Fig 2A). A transcript starting from TSS A covers a full-length copy of the RNAT. The duplicated RNAT sequence further upstream of TSS A deviates in one nucleotide (T26C).
To map the 5'-end(s) of the cnfY mRNA, we performed primer extension experiments on total RNA isolated from Y. pseudotuberculosis YPIII grown at 37˚C. No signal was detected around position -82 relative to the AUG codon (TSS A , Fig 2A and 2B). Instead, we found a weak signal 183 nt (TSS B) and a strong signal 266 nt (TSS C) upstream of the start codon, respectively. The TSS C signal corresponds to the first of three signals visible in transcriptome profiling data of Y. pseudotuberculosis YPIII grown to exponential phase at 37˚C [29] (S2A Fig). To test whether the long 5'-UTR is capable of regulating reporter gene expression in response to temperature, we compared the activities of P BAD promoter-controlled translational bgaB fusions comprising the long TSS C or the short TSS A-derived 5'-UTRs followed by 30 nt of the coding region of cnfY fused to bgaB [18]. Both the long and short 5'-UTR conferred Y. pseudotuberculosis YPIII was grown to exponential growth phase (OD 600 = 0.5) at 25˚C or 37˚C and samples were harvested for subsequent qRT-PCR and Western blot analysis. Temperature-dependent transcription was measured by qRT-PCR. Data were analyzed by the ΔΔCt method [30] and cnfY transcription was normalized to that of reference genes nuoB and gyrB. Amounts of cnfY transcript relative to levels detected at 25˚C are shown. CNF Y signals from Western blot analysis were quantified by integrated density quantification using AlphaEaseFC software. The mean density values and standard deviations from three independent experiments were calculated and normalized by the respective mean density value measured at 25˚C. A representative Western blot is displayed in S1A Fig. (B) PARS profiles of the cnfY RNAT (-82 nt to +30 nt from AUG) at 25˚C and 37˚C. The potential SD sequence, its pairing sequence (anti-SD), and the AUG are marked. (C) PARS-derived secondary structure of cnfY (-82 nt to +30 nt from AUG) at 25˚C [18]. The potential SD region and the translation start codon (AUG) are highlighted in black and orange, respectively. (D) The RNA control element in the cnfY 5'-UTR confers temperature-dependent reporter gene expression. To test for temperature-dependent translational control, the cnfY RNAT was translationally fused to bgaB under control of the P BAD promoter (pBO3192). The yscW-lcrF intercistronic region (lcrF RNAT) served as positive control (pBO3146). A schematic representation of the reporter gene fusions is displayed. E. coli DH5α and Y. pseudotuberculosis YPIII cells harboring the corresponding plasmid (technical triplicates per each construct) were grown to an OD 600 = 0.5 at 25˚C. Afterwards, transcription was induced with 0.01% (w/v) (E. coli) or 0.1% (w/v) (Y. pseudotuberculosis) L-arabinose, respectively. The cultures were split immediately: One half remained at 25˚C while the other was transferred into pre-warmed flasks at 37˚C. After 30 min incubation, samples were taken for subsequent β-galactosidase assay. The displayed results represent the mean similar temperature-regulated expression (Fig 2C) suggesting that the TSS A-derived short 5'-UTR of cnfY with a single RNAT copy is sufficient for temperature regulation. From here on we will refer to this 5'-UTR (-82 nt from AUG) as cnfY RNAT.
Sequence comparison of the cnfY locus revealed duplicated thermometer sequences in other Y. pseudotuberculosis strains and in Y. pestis (S2B Fig). Interestingly, the nucleotide substitution (T26C) present in the distal copy of the thermometer in Y. pseudotuberculosis YPIII was also found in the downstream copy in several Y. pseudotuberculosis strains and in Y. pestis (S2B Fig). In the secondary RNA structure, nucleotide 26 pairs with G79 ( Fig 1C) and the U to C exchange should strengthen this interaction. To analyze the impact of this single-nucleotide polymorphism (SNP) on translational control, a translational bgaB fusion of the U26C variant was constructed ("Y. pestis" RNAT in Fig 2C) and found to respond to increasing temperature like the Y. pseudotuberculosis YPIII thermometer demonstrating that the SNP-containing natural variants retained their regulatory potential.

Point mutations impair thermometer functionality
The naturally occurring SNPs in the cnfY 5'-UTR motivated us to construct RNAT variants with increased stability. The bulged A29 residue and the internal loop (AG32-33, AG73-74) in the antiSD/SD region are destabilizing elements ( Fig 3A) and most likely critical for temperature-dependent regulation of cnfY. Using site-directed mutagenesis, we previously constructed the repressive (stabilizing) mutations R1 (AG32-33CU) and R2 (ΔA29) into the cnfY RNAT, which partially inhibited expression in the β-galactosidase assay in E. coli [18]. To achieve full repression, we combined both mutations to obtain the R1+2 (ΔA29, AG32-33CU) variant of the cnfY RNAT with a perfectly paired SD region (Fig 3A). The R1, R2, and R1+2 variants fused to bgaB were analyzed at 25˚C and 37˚C in E. coli and Y. pseudotuberculosis (Fig 3B). RNAT variants with mutations R1 (AG32-33CU) and R2 (ΔA29) exhibited reduced levels of reporter gene activity compared to the wild-type thermometer. The combined mutation R1+2 (ΔA29, AG32-33CU) impaired temperature sensing and resulted in an almost complete loss of β-galactosidase activity even at 37˚C. Due to the similar responses in the heterologous E. coli host and in Y. pseudotuberculosis, we conducted all subsequent experiments in Yersinia only.

The cnfY thermosensor controls translation in a temperature-dependent manner
To determine whether the cnfY RNAT mediates thermoregulation at translational rather than at transcriptional level of gene expression, we used a previously established gfp-based reporter gene system [35], in which the cnfY RNAT downstream of the P BAD promotor is translationally fused to gfp (Fig 4A). We measured gfp transcript levels by Northern blots with an anti-gfp RNA probe and GFP protein by Western blots with a GFP-specific antibody after growing Y. pseudotuberculosis YPIII cultures at 25˚C or after a shift to 37˚C. A translational lcrF RNAT: gfp fusion served as positive control (Fig 4B). While reporter gene transcript levels remained nearly unaffected by temperature (Fig 4C), protein levels increased substantially after a shift to 37˚C (Fig 4C). In accordance with the results of the β-galactosidase activity assay (Fig 1D), these results show that the cnfY RNAT confers temperature-dependent translational control to its downstream gene. activities from three independent experiments (biological replicates). Mean standard deviations are indicated as error bars. https://doi.org/10.1371/journal.ppat.1008184.g001 An RNA thermometer dictates production of a secreted bacterial toxin  [18] and in this study are marked by arrows. Additionally, the sequence of the intergenic region between YPK_2616 (coding for a putative transposase) and cnfY is displayed. The underlined sequences (orange line) represent the duplicated RNAT sequences. The sequence variation T26C observed in the duplicated sequences is marked by asterisks. (B) Identification of cnfY TSSs by primer extension. Total RNA was isolated from a Y. pseudotuberculosis YPIII culture grown at 37˚C to exponential (OD 600 = 0.5) and stationary growth phase (OD 600 = 1.5). Primer extension was performed using a radiolabeled primer (binding is shown in A). The cDNA products were separated via polyacrylamide gel electrophoresis. The experiment was conducted in three biological replicates. (C) Y. pseudotuberculosis cnfY 5'-UTRs (+30 nt of the coding region; pBO3192, pBO6527) and the Y. pestis cnfY RNAT (-82 nt to +30 nt from AUG; pBO6523) were translationally fused to bgaB and tested for β-galactosidase activity. The lcrF RNAT was used as positive control (pBO3146). E. coli DH5α cells harboring the corresponding plasmid (technical triplicate per each construct) were grown to an OD 600 = 0.5 at 25˚C. Afterwards, transcription was induced with 0.01% (w/v) L-arabinose and the culture was split. One half remained at 25˚C while the other was transferred into pre-warmed flasks (37˚C). After 30 min of incubation, samples were taken for subsequent βgalactosidase assay. The displayed results represent the mean activities from three independent experiments (biological replicates). Mean standard deviations are indicated as error bars. https://doi.org/10.1371/journal.ppat.1008184.g002 An RNA thermometer dictates production of a secreted bacterial toxin We also verified that the repressive mutations affected translation rather than transcription and determined RNA and protein levels by the pBAD2-gfp reporter system [35] as described above. In all cases, the transcript levels were similar at 25 and 37˚C whereas the protein amounts of the WT, R1 and R2 fusions increased at 37˚C (Fig 4D and 4E). Consistent with previous reports, the transcript levels of poorly translated repressive RNAT variants was low presumably because they are "naked" and not protected by translating ribosomes [28,36]. As observed with the bgaB fusions (Fig 3B), the R1 and R2 constructs retained some temperature responsiveness (Fig 4D and 4E). More specifically, stabilization of the SD region in R1 ( Fig  3A) resulted in stronger translational repression compared to deletion of the bulged A in the anti-SD region in R2. The R1+R2 fusion combining both mutations was fully repressed.

Thermosensing by the cnfY RNAT is independent of the promoter
The results above demonstrate that the cnfY RNAT mediates translational thermoregulation to various reporter genes under control of the arabinose-inducible P BAD promoter (Figs 1 to 4). In a final set of reporter gene fusion experiments, we measured bioluminescence of transcriptional and translational luxCDABE fusions downstream of the native cnfY promoter in plasmids pJNS02 [20] or pBO6500), respectively (Fig 5A). The reporter gene fusions include the promoter region as well as 6 nt (transcriptional fusion) or 30 nt (translational fusion) of the cnfY coding region. The established lux reporter gene system allows monitoring of gene expression in living cells by measuring bioluminescence [37]. Increased transcription at 37˚C is consistent (Fig 5B) with previous results [29]. Two results suggest an important contribution of the RNAT to overall cnfY expression. First, lower bioluminescence at 25˚C in the translational fusion supports an inhibitory function of the secondary RNA structure at low temperatures even when transcript is present. Second, higher temperature-dependent induction of the translational fusion is consistent with translational induction at host body temperature.  The wild type (pBO4481) and the mutated cnfY RNATs (R1: pBO6505, R2: pBO6506, R1+2: pBO6507) were translationally fused to gfp and tested for differential GFP synthesis and gfp transcription in Y. pseudotuberculosis YPIII as described in Fig 3. Each experiment was performed at least in We also used translational luxCDABE fusions to confirm that the repressive RNAT mutations R1 and R1+2 affect translation of cnfY expressed from its natural promoter. This will later become relevant in the infection model (see Fig 8). We grew Y. pseudotuberculosis cells harboring the translational fusion plasmid with the wildtype RNAT at 25˚C and 37˚C and measured substantially higher bioluminescence at the higher temperature with the wild type triplicate. (E) Comparison of transcriptional and translational control by the cnfY RNAT variants measured by densitometrical analysis. Signals from Northern and Western blot analysis were quantified by integrated density quantification using AlphaEaseFC software. The mean density values and its standard deviations (three independent experiments) were calculated and normalized by the respective mean density value measured for the wild type thermometer at 25˚C. https://doi.org/10.1371/journal.ppat.1008184.g004  [20]) and a translational fusion (pBO6500) of the cnfY promoter region (P cnfY ) to the luxCDABE reporter gene operon. (B) Y. pseudotuberculosis YPIII cultures harboring these plasmids were grown at 25˚C and 37˚C. Samples for bioluminescence measurement (described in Material and Methods) were taken after 4 hours. The data are depicted as relative light units (RLU/OD 600 ) from three independent cultures measured on a CLARIOstar plus reader (BMG Labtech, Ortenberg, Germany). (C) Y. pseudotuberculosis YPIII cultures harboring the translational fusion including the wildtype (pBO65000), the R1 (pBO6501), or the R1+2 (pBO6502) RNAT variant were grown at 25˚C and 37˚C. Samples for bioluminescence measurement (described in Material and Methods) were taken after 4 hours. The data are depicted as relative light units (RLU/OD 600 ) from three independent cultures measured on a Varioskan Flash (Thermo Scientific, St. Leon-Rot, Germany). Results from bioluminescence measurements of variant R1+2 are additionally shown as inlet on a different scale to reflect the low signal intensity.  [18] at 25˚C. Nucleotides cleaved by RNase T1/T2 at all three temperatures or exclusively at 37 and 42˚C are indicated. Nucleotides of the SD region and the AUG start codon are colored in black and orange, respectively. (C) Cleavage signals of the SD sequence (nucleotides G74, G75, and G77 cleaved by RNase T1) were quantified by integrated density quantification using AlphaEaseFC software. The mean density RNAT (Fig 5C). Bioluminescence was strongly reduced in cells harboring the R1 variant and several orders of magnitudes lower in cells carrying the R1+2 variant consistent with the notion that translation is blocked due to the stable RNAT structure. All data described above demonstrate that the cnfY thermometer is functional in foreign and native genetic contexts and that it is a major determinant of thermoregulation of cnfY expression.

The cnfY RNAT melts gradually at elevated temperatures
To address whether the secondary structure of the cnfY RNAT melts open with increasing temperature, we performed enzymatic structure probing on in vitro transcribed radiolabeled RNA. The probing experiments were carried out at 25, 37 and 42˚C using RNase T1 and RNase T2, which preferentially cleave 3' of unpaired guanines and unpaired adenines, respectively. Due to these properties both RNases are ideal to monitor unfolding of the purine-rich SD region. The cleavage products were separated on a denaturing 8% polyacrylamide gel ( Fig  6A). Noticeable cleavage by RNase T1 or RNase T2 at 25˚C suggests that nucleotides 73 to 77 of the SD region are in a partially single-stranded conformation. This is in good agreement with the PARS data-derived secondary structure of the cnfY RNAT, where A73 and G74 are located in an internal loop (Fig 6B). Notably, the SD sequence is temperature-sensitive as the cleavage intensities increased from 25˚C to 37/42˚C ( Fig 6C) and nucleotides 71-72 and 78-79 flanking the SD sequence were only accessible at 37 and 42˚C. Partially deviant from the PARS-derived secondary structure are the AUG start codon and the subsequent adenosine residues, which are poorly accessible at 25˚C. This result suggests tertiary interactions at low temperature that are released at 37 and 42˚C.
The cleavage pattern of RNase T1 and T2 in the hairpin up-and downstream of the translation initiation region largely reflects the PARS-constrained secondary structure (Fig 6B). Nucleotides 14 and 15, which are part of or in close proximity to an interior loop, are accessible to RNase T2 at 25˚C. The base of the entire hairpin structure is thermo-labile since guanines at position 11, 96, and 97 are solely accessible to RNase T1 at 37 and 42˚C. The same is true for the top of the structure, which is in a rather loose conformation since nucleotides 40, 45, 46, and 47 are already accessible to RNases T1 and T2 at 25˚C. Guanines 57, 59, and 64, however, are responsible for structure formation at 25˚C and become prone to nucleolytic attack only at 37˚C. Overall, the structure probing results show that multiple irregularities (bulged residues, internal loops) throughout the needle-like structure of the cnfY 5'-UTR contribute to its temperature-sensitive nature, which allows opening after a rather mild transition from environmental to host body temperature. Fully consistent with impaired translation in reporter gene assays, probing of the R1+2 variant (ΔA29, AG32-33CT) revealed a structure that is overall similar to the wild-type structure, but in which the RBS is completely heat-resistant and inaccessible to RNases T1 and T2 at 37 and 42˚C (Fig 6A).
To assess whether melting of the RNAT in vivo results in a gradual zipper-like or a sudden switch-like response, we assayed β-galactosidase activity of the bgaB fusion along a temperature range from 25 to 42˚C. The wild-type cnfY thermometer induced a gradual increase of reporter gene activity (Fig 6D), indicating zipper-like opening of the RNAT structure. In line ratio (37˚C/25˚C) and its standard deviation was calculated from three independent experiments (biological replicate). (D) The wild type (WT; pBO4481) and R1+2 cnfY RNAT (pBO6507) were translationally fused to bgaB and reporter gene activity was measured as follows. E. coli DH5α cells harboring the corresponding plasmid (technical triplicate per each construct) were grown to an OD 600 = 0.5 at 25˚C. Afterwards, transcription was induced with 0.01% (w/v) L-arabinose and the culture was split. One sixth remained at 25˚C while the other sixths were transferred into pre-warmed flasks at the indicated temperatures. After 30 min of incubation, samples were taken for subsequent β-galactosidase assay. The displayed results represent the mean activities from three independent experiments. Mean standard deviations are indicated as error bars. https://doi.org/10.1371/journal.ppat.1008184.g006 An RNA thermometer dictates production of a secreted bacterial toxin with the structure probing results (Fig 6A), the R1+2 fusion was completely unresponsive even at 42˚C. These results provide evidence that the RNAT adopts a labile secondary structure at moderate environmental temperatures that reduces accessibility to the RBS to prevent cnfY translation when it is not necessary.

Ribosome binding to the SD region of the cnfY RNAT is temperaturecontrolled
The mode of action of most RNATs relies on temperature-dependent ribosome binding to the SD sequence, which results from a zipper-like melting process [5]. To address whether the cnfY leader sequence follows this regulatory principle we performed toeprinting experiments [34]. After annealing of a cnfY-specific primer to the cnfY RNAT including 80 nt of the coding region, the RNA was incubated with tRNA fMet and 30S ribosomal subunits at 25, 37 and 42˚C. Following primer extension, the reverse transcripts were separated on an 8% denaturing polyacrylamide gel. Toeprinting signals caused by ribosomes acting as a roadblock for reverse transcription appeared at +11 to +16 from the translation start codon only in presence of 30S ribosomal subunits (Fig 7A). Although residual ternary complex formation between mRNA, tRNA fMet and ribosome occurred at 25˚C, this process was increased almost 3-fold at elevated temperatures ( Fig 7B). Consistent with the absence of RNA melting in vitro and translation initiation in vivo, no toeprinting signal was detected with the R1+2 variant (ΔA29, AG32-33CT) of the cnfY thermometer. An additional signal between the AUG codon and SD sequence appeared, which increased in the presence of 30S ribosomal subunits. Such cDNA products are commonly observed with repressed RNAT structures, for example of the Salmonella enterica groES [38] and the Y. pseudotuberculosis lcrF RNATs [15]. They might result from inhibition or drop-off of the reverse transcriptase by the stable secondary structure. How the 30S subunit favors such premature reverse transcription products is unclear. Altogether, the results strongly suggest a temperature-controlled formation of a ternary complex in absence of any additional bacterial factors.

The cnfY RNAT controls provision of the CNF Y toxin and is critical for virulence
A recent study demonstrated the pivotal role of the CNF Y toxin in Y. pseudotuberculosis pathogenesis [20]. A deletion of cnfY rendered Y. pseudotuberculosis avirulent and this phenotype was restored in presence of the cnfY complementation vector pJNS10. This plasmid harbors the complete cnfY promoter region followed by the full-length cnfY open reading frame ( Fig  8A). We inserted the repressive mutations R1 and R1+2 into pJNS10 (pJNS10-R1 and pJNS10-R1+2) and investigated their influence on CNF Y synthesis via immunodetection. Y. pseudotuberculosis ΔcnfY (YP216) cells harboring the respective complementation vectors were grown to exponential growth phase at 25˚C and 37˚C, and samples were taken for Western blot analysis using a CNF Y -specific antibody. Both, the R1 as well as the R1+2 mutation prevented cnfY translation at 37˚C as no CNF Y was detectable (Fig 8B).
Next, we investigated the relevance of the RNAT in pathogenesis by measuring the survival of BALB/c mice (n = 10 per strain) orally infected with the Y. pseudotuberculosis wild type carrying the empty vector pJNS11 or a cnfY-deficient strain (cnfY::Kan R , YP147) harboring either pJNS11 or the cnfY-complementation vector (pJNS10) with the wild type, the R1, or R1+2 RNAT variant. The survival rate of infected mice was monitored for two weeks (Fig 8C). Mice infected with the wild type strain showed clear signs of infection (e.g. lethargy and rough fur) and reached humane endpoints (cut-off criteria based on health scores). Consistent with previous results [20], infection with the cnfY deletion strain did not result in the death of any mice.
In presence of the cnfY complementation vector, however, the cnfY mutant was as virulent as the wild type. Mice infected with cnfY deficient mutants harboring pJNS10-R1 exhibited a survival rate of 80% post infection indicating that stabilization of the cnfY RNAT reduces virulence potential. The cnfY::Kan R strain with the R1+2 RNAT was avirulent like the cnfY deletion mutant showing that proper temperature measurement is indispensable for host infection.
Bodyweight measurements of infected mice essentially reflected the survival results. The weight of mice infected with the wild type strain or the cnfY mutant complemented with plasmid-borne cnfY severely declined after two days of infection (Fig 8D). Mice infected with the cnfY::Kan R mutant carrying the empty vector lost little weight and regained it within a week. Mice infected with Y. pseudotuberculosis mutants carrying stabilized cnfY RNATs showed intermediate phenotypes. They lost weight in the first six days but recovered and regained it thereafter. Consistent with the obtained survival rates, the strain carrying the most stable R1+2 RNAT variant least affected the fitness of the mice. Primer extension inhibition of the wild type (pBO4465) and R1+2 cnfY RNAT (pBO4466) in presence (+) or absence (−) of 30S ribosomal subunits was conducted as described in material and methods. The resulting signals correspond to the full-length reverse transcription (RT) product and the termination product (toe print; corresponding to nucleotides +11 to +16 from AUG). ACGT indicate the corresponding DNA sequencing reactions. The SD sequence as well as the ATG codon are labelled. The experiment was performed at least in triplicate (three biological replicates). (B) Toe printing signals (nucleotides +11 to +16 from AUG codon) were quantified by integrated density quantification using AlphaEaseFC software. The mean density ratio (25˚C/37˚C) and its standard deviation was calculated from three independent experiments. https://doi.org/10.1371/journal.ppat.1008184.g007 An RNA thermometer dictates production of a secreted bacterial toxin Finally, we tested whether stabilization of the cnfY thermosensor affects colonization of Y. pseudotuberculosis within Peyer's patches, mesenteric lymph nodes (MLNs), liver, or spleen. Mice (n = 10 per strain) were orally infected with 2x10 8 cells of Y. pseudotuberculosis YPIII carrying the empty vector or with the cnfY deletion mutant carrying cnfY complementation vectors with the wild type, R1 or R1+2 RNAT. The number of bacteria in infected organs was determined by plating of organ homogenates five days post infection (Fig 9). No difference in  [20]. Transcriptional start sites (TSS) identified in a previous study [18] and in this study (see Fig 2) are marked by arrows. (B) Y. pseudotuberculosis ΔcnfY (YP216) cells without vector or harboring vector pJNS10 (cnfY complementation vector; [20]), pJNS10 R1 (cnfY complementation harboring mutation R1; pBO6503), pJNS10 R1+2 (cnfY complementation vector harboring mutation R1+2; pBO6504), and pJNS11 (empty vector) were grown to exponential growth phase (OD 600 = 0.5). Afterwards, samples were taken for SDS-PAGE as well as Western blot analysis with an antibody against CNF Y . The experiment was performed at least in triplicate. (C) Survival and (D) body weight of BALB/c mice (n = 10/strain) was monitored for up to 14 days after oral infection via food pellets with 2x10 9 CFU of either Y. pseudotuberculosis YPIII harboring the empty vector pJNS11 (black line + black dots) or the cnfY mutant YP147 (cnfY::Kan R ) harboring the empty vector pJNS11 (grey line + grey squares) or the cnfY complementation vector pJNS10 (pP cnfY :cnfY + ) including either the wild type (grey line + grey asterisk), the R1 (blue line + blue triangles), or the R1+2 (red line + red crosses) variant of the cnfY RNAT. Displayed data represent results from two individual experiments (biological replicates). https://doi.org/10.1371/journal.ppat.1008184.g008 An RNA thermometer dictates production of a secreted bacterial toxin colonization of the Peyer's patches and the liver were observed between the five Y. pseudotuberculosis strains. In contrast, significantly reduced numbers of bacteria were recovered from MLNs and spleen when mice were infected with the cnfY mutant carrying either an empty vector or the complementation plasmid with the stabilized R1 or R1+R2 thermometers. The latter strain showed the same poor recovery from infected spleen, liver, and MLNs as observed for the cnfY deletion mutant. Taken together, the results clearly demonstrate that RNAT-controlled synthesis of CNF Y is crucial for the virulence of Y. pseudotuberculosis.

Discussion
Precise spatio-temporal control of virulence gene expression determines the outcome of bacterial infections. Temperature is one of the consistent parameters that correlate with the stage of invasion during the infection of a warm-blooded host [1]. In contrast to transcriptional control mechanisms that proceed through several levels of gene expression, RNATs are one step ahead and directly feed the temperature signal into already existing transcripts and modulate the efficiency of translation initiation [7,8]. Accordingly, several pathogens control expression of their master virulence regulators by RNATs, such as PrfA, ToxT and LcrF in Listeria monocytogenes, Vibrio cholerae and Yersinia species, respectively [15,16,39,40]. Thereby, entire virulence regulons are recruited under temperature control. In addition, individual genes responsible for specific virulence traits can be under direct control of RNATs. Examples are the capsule biosynthesis gene cssA in Neisseria meningitidis [41] or the heme uptake genes shuA and shuT in Shigella dysenteriae [42,43].
Here we report a detailed structure-function analysis of an RNAT controlling synthesis of the secreted toxin CNF Y in Y. pseudotuberculosis, which we had identified in a global RNA structurome study [18]. The approx. 100 nt long RNA structure, which reaches more than 20 nt into the coding region contains several remarkable features. Three bulged residues, a tworesidue bulge, a large bulge of 10 nt and four short internal loops (Fig 6B) contribute to an overall labile structure ideally suited to calibrate translation in the physiological temperature range like a rheostat. These multiple irregularities in the hairpin structure might provide breathing points also allowing for the observed residual translation even at low temperatures. As one might expect from the many unpaired nucleotides throughout the cnfY hairpin, the entire structure is engaged in the melting process. The overall architecture of the cnfY RNAT is unique and does not conform to known classes of RNA-based thermosensors, such as the fourU or ROSE elements [5]. The partially paired translation initiation region within an extensive hairpin structure is reminiscent of the prfA RNA thermosensor from L. monocytogenes [39]. However, in contrast to the sudden switch-like response around 37˚C in the prfA thermometer [41], the temperature-induced opening of the cnfY thermosensor was gradual as in many other RNATs.
Another interesting observation relates to the genomic duplication of and polymorphisms in the cnfY thermometer, which is remarkably conserved among various Y. pseudotuberculosis strains and Y. pestis. Owing to a potential transposase gene (YPK_2616) upstream of cnfY, it was postulated that the cnfY gene was acquired via horizontal gene transfer [23]. During reevaluation of the transcription start site(s) of the cnfY gene, we discovered a major start site upstream of the previously annotated start site. The resulting alternative transcript (TSS C, Fig  2A) contains a duplicated RNAT sequence and is equally functional as the cnfY RNAT when present in single copy. A common SNP (U26C) that slightly increases the thermodynamic stability of the secondary structure is common in the upstream copy. The presence of this substitution in the downstream copies in other Y. pseudotuberculosis strains (IP32593 or PB1/+) and in Y. pestis did not abrogate thermoregulation (Fig 2C) suggesting an evolutionary pressure to maintain the functionality of the cnfY RNAT. Despite having functional thermosensors, several Y. pseudotuberculosis strains and Y. pestis do not produce CNF Y because they encode a truncated cnfY pseudogene encoding a non-functional toxin [23]. Our results suggest that the ancestral and still intact cnfY genes in these organisms have been under temperature control.
The inspection of the 5'-UTRs of a variety of genes coding for CNF family toxins in bacteria other than Yersinia suggests that RNAT-mediated thermoregulation of GTPase-activating toxins is quite common. Although the nucleotide sequences upstream of cnf1 from UPEC, dnt from Bordetella pertussis, and vopC from Vibrio paraheamolyticus, which code for the cytotoxic necrotizing factor 1 (CNF1), the dermonecrotizing toxin (DNT), and the CNF-like toxin VopC, respectively [44][45][46], do not exhibit any sequence conservation among each other or to Y. pseudotuberculosis cnfY (S3A Fig), all of them are predicted to fold into cnfY RNAT-like secondary structures (S3B Fig). Partially or completely occluded SD sequences and/or start codons in extensive stem-loop structures strongly suggest temperature-driven control of translation initiation in all these pathogens.
The combined results from this and previous studies (data from transcriptome studies summarized in S4 Table) show that thermoregulation of cnfY is a two-layered process comprised of transcriptional and translational control (Fig 10). Notably, even transcription of the cnfY pseudogene (YPTB1468) in Y. pseudotuberculosis IP32953 is temperature-dependent [47]. It is currently unknown how cnfY transcription is regulated. It could possibly be due to a temperature-sensitive regulator or the temperature-regulated synthesis of a transcription factor. Transcriptional profiling of Y. pseudotuberculosis YPIII at 25˚C and 37˚C suggested an involvement of the cAMP receptor protein (CRP) because temperature-dependent transcription of cnfY was lost in a Δcrp mutant [29]. The CRP protein is a global transcriptional regulator that, in interplay with the signaling nucleotide cAMP, steers expression of genes associated with sugar utilization [48,49]. Moreover, this regulator was shown to be crucial for virulence of pathogenic Yersinia species [50][51][52], but also other enteropathogens such as UPEC strains [53], Klebsiella pneumonia [54,55], Proteus mirabilis [56], or Salmonella enterica [57].
Although temperature-regulated transcription plays a role in the provision of the CNF Y toxin, our study shows that translational control by the cnfY thermometer accounts for a large proportion in this concerted process. Thermoregulation was maintained when transcription was driven from a foreign arabinose-inducible promoter suggesting that pre-existing cnfY  Table) and translation is hindered by a stem-loop structure in the 5'-UTR of cnfY. Within the host (37˚C), transcription is enhanced and the RNAT structure melts open, allowing translation of the cnfY transcript. During the infection process, the CNF Y toxin is secreted via outer membrane vesicles to the target host cells. Upon reaching the host target cell, the toxin is taken up by endocytosis and released into the cytosol due to a low pH-dependent mechanism. CNF Y then activates small Rho GTPases (RhoA, Rac1, and Cdc42) leading to increased pore formation of the T3SS and boosts Yop effector protein delivery into target cells. Additionally, caspase-1 is activated and triggers secretion of cytokines resulting in cell death due to inflammatory pyroptosis [20,21]. https://doi.org/10.1371/journal.ppat.1008184.g010 An RNA thermometer dictates production of a secreted bacterial toxin transcripts are only efficiently translated when body temperature has been reached. Additional upregulation of transcription probably helps boost the toxin levels inside the host and to integrate other host-specific signals than temperature. A complex interplay between transcriptional and translational temperature control of gene expression has also been reported for synthesis of the virulence regulator LcrF from Y. pseudotuberculosis. Thermal induction involves temperature-regulated proteolysis of the YmoA transcription factor, which represses transcription of the yscW-lcrF operon at low temperatures, and an intergenic RNAT upstream of lcrF [15]. A most recent study demonstrated that the RNA-binding protein CsrA is involved in cell-contact mediated expression of lcrF [58]. CsrA binds to two sites within the lcrF transcript, one within and one downstream of the SD sequence, which destabilizes the RNAT structure and overcomes translational repression at low temperature (25˚C). A dual layer of transcriptional and translational control is also quite common in heat shock gene regulation. Here, many heat-induced genes are under transcriptional control, for example by the alternative heat shock sigma factor σ 32 , plus RNAT-mediated translational control [27,33,59,60].
The CNF Y toxin is one of the important virulence determinants predominantly expressed during the early stage of infection [61]. In this study, we show how crucial individual nucleotides in the untranslated region of a virulence gene are by applying an in vivo infection model. Animal models have previously only been used to demonstrate the importance of RNATs upstream of genes coding for virulence master regulators, namely LcrF and ToxT of Y. pseudotuberculosis and V. cholerae, respectively [15,40]. Other studies have been restricted to structure-function analysis and in vitro models [41][42][43][62][63][64][65]. Here, we demonstrated that converting the loose RNAT structure into a perfectly matched zipper repressed synthesis of CNF Y within the mouse, which made Y. pseudotuberculosis avirulent. This result reinforces the notion that translational control by the RNAT is the dominant mechanism determining cnfY expression in the warm-blooded host. The CNF Y toxin is supposed to amplify crucial virulence factor functions securing fine-tuning of Y. pseudotuberculosis pathogenicity [20]. Activity is known to be crucial for virulence as deletion of cnfY results in reduced mortality rates of infected mice [20]. Moreover, loss of CNF Y reduces inflammatory responses in infected host tissues and reprograms the pathogen's transcriptional response towards a persistent life style [66]. The finding that stabilization of the cnfY thermosensor reduced the pathogen's potential to colonize MLNs and spleens of infected mice is consistent with the observation that absence of CNF Y provokes a significant increase of infiltrating macrophages, monocytes and natural killer cells in the spleen allowing a rapid killing of the pathogen [20].
Unlike many other Yersinia virulence factors, CNF Y is not delivered via the T3SS but secreted via outer membrane vesicles (OMVs) [19,20,67]. The CNF-1 toxin from UPEC strains also takes the OMV route [67-69]. Also unlike many other virulence genes in Yersinia, the cnfY gene is not under temperature control by LcrF. Instead, the horizontally acquired gene is equipped with its own RNAT in order to produce the toxin once the bacterium is inside the warm-blooded host. We find it intriguing that similar RNAT-like structures are likely to exist upstream of various other toxin genes (see above). This finding supports the emerging view that RNA-mediated signal perception is common in pathogenic bacteria and that such RNAbased sensors might be attractive drug targets [70]. Interestingly, a magnesium-binding site in the Salmonella fourU thermometer could be targeted by Hexaammincobalt(III)-chlorid [71] and triptycene-based small molecule binding to a central three-way junction in the E. coli rpoH thermosensor were able to stabilize the structure [72]. These findings promise that antimicrobial compounds interfering with the functionality of RNA thermometers can be developed.
Supporting information S1  Fig. Expression of cnfY is thermoregulated. (A) Detection of CNF Y protein levels at 25˚C and 37˚C via Western blot analysis. Y. pseudotuberculosis YPIII cells were grown in LB medium to exponential (OD 600 = 0.5) and early stationary growth phase (OD 600 = 1.5) and samples were taken for SDS-PAGE and Western blot analysis using a CNF Y -specific antibody. (B) Representative, original Western blot for CNF Y quantification at 25˚C and 37˚C used for data analysis displayed in Fig 1A. Samples were taken from a Y. pseudotuberculosis YPIII culture grown in LB medium to exponential growth phase (OD 600 = 0.5).  [10] visualized via the Artemis genome browser [12]. Different potential TSSs (for details see Fig 1A) are indicated by arrows. (B) Schematic representation of the cnfY upstream region (559 nt) including 30 nt of the cnfY coding region and 29 nt of the upstream transposase gene found in Y. pseudotuberculosis strains YPIII (GenBank accession: CP009792), IP2666 pIB1 (CP032566), IP31758 (CP000720), IP32953 (CP009712), and PB1/+ (CP009780) or in Y. pestis strains CO92 (CP009973) and Pestoides F (CP00668). The cnfY RNAT sequence and its upstream duplication are marked in orange. Sequence insertions are marked in blue, whereas asterisks mark nucleotide exchanges within the RNAT sequences (relative to Y. pseudotuberculosis YPIII). Broken lines indicate sequence deletions relative to the Y. pseudotuberculosis YPIII genome. Overall sequence identities (relative to Y. pseudotuberculosis YPIII) are displayed under each species name. (TIF) S3 Fig. Sequence and structure conservation of leader regions upstream of CNF encoding genes. (A) Multiple alignment of sequences located upstream of genes coding for CNFs or related toxins. Displayed is the alignment of sequences upstream of cnfY from Y. pseudotuberculosis YPIII, cnf1 from E. coli O18:K1:H7 UTI89, dnt from Bordetella pertussis J262, and vopC from Vibrio parahaemolyticus BB22OP. The multiple sequence alignment was calculated with Clustal Omega (https://www.ebi.ac.uk/Tools/msa/clustalo/) and visualized via jalview application [13]. (B) Secondary structures of the cnfY RNAT (-82 nt; [6]) from Y. pseudotuberculosis YPIII (ΔG˚= -13.19; [ΔG˚] = kcal � mol -1 ) and the upstream regions (-100 nt and +30 nt from AUG) of cnf1 from E. coli O18:K1:H7 UTI89 (ΔG˚= -28.20), dnt from B. pertussis J262 (ΔG˚= -61.93) and vopC from V. parahaemolyticus BB22OP (ΔG˚= -23.05) are displayed. Structure of the cnfY RNAT originates from [6]. The remaining structures were predicted via RNAfold [14] with temperature set to 25˚C. The proposed SD sequences and AUG start codon are depicted in black and orange, respectively. (TIF) S1 References. References for supporting information.
(DOCX) 67. Blumenthal B, Hoffmann C, Aktories K, Backert S, Schmidt G. The cytotoxic necrotizing factors from Yersinia pseudotuberculosis and from Escherichia coli bind to different cellular receptors but take the same route to the cytosol. Infect Immun. 2007; 75 (7)