An atlas of Hfq‐bound transcripts reveals 3′ UTRs as a genomic reservoir of regulatory small RNAs

A dynamic landscape of Hfq‐associated sRNAs

We profiled Hfq‐associated transcripts in Salmonella using Solexa sequencing of RNA enriched by co‐immunoprecipitation (coIP) with the chromosomally encoded, epitope‐tagged Hfq protein (Sittka et al, 2008). Samples were collected at several time points along the growth curve of a shaking culture within 1 day of inoculation (Figure 1A), and covered the exponential phase of growth (OD₆₀₀ of 0.15 or 0.5), the early stationary phase (OD₆₀₀ of 2) when the Salmonella invasion genes are transiently activated, and the stationary phase at four subsequent time points including overnight culture.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Dynamic sRNA profiles of Hfq over bacterial growth. (A) Growth curve of Salmonella grown for 14 h in LB at 37°C, 220 r.p.m. Time points when culture samples were withdrawn for Hfq co‐immunoprecipitation are indicated. (B) Distribution of reads matching experimentally validated sRNAs in Hfq‐coIP cDNA libraries at several stages of growth. Percentage indicates the reads of a given sRNA compared to all sRNAs in a cDNA library. The relative amount of reads and enrichment factors for individual sRNAs are listed in Supplementary Table S3. ON: overnight culture.

Figure 1B depicts the individual sRNA species enriched by pull‐down with Hfq protein at the selected seven points of growth, as a proportion of the experimentally validated sRNAs of Salmonella annotated in this (Supplementary Table S1) and previous studies (Padalon‐Brauch et al, 2008; Sittka et al, 2008; Sittka et al, 2009; Kröger et al, 2012). We observed a dramatic change in the profiles along the axis of growth, with some sRNAs dominating in individual growth phases. For example, the catabolite repression‐associated Spot42 (Spf) sRNA prevails in fast growing cells, whereas a different set of sRNAs including RprA, SdsR and RybB become prominent partners of Hfq as bacteria progress into the stationary phase. There are only two abundant sRNAs—ArcZ (activator of σS and repressor of serine uptake and oxidative stress related functions) and ChiX (repressor of chitoporin synthesis)—which significantly occupy Hfq throughout growth, and both are well‐known examples of growth phase‐independent sRNA expression (Argaman et al, 2001; Vogel et al, 2003; Figueroa‐Bossi et al, 2009; Papenfort et al, 2009; Mandin and Gottesman, 2010; Rasmussen et al, 2009). For other well‐studied sRNAs, Hfq occupancy was also in excellent agreement with previously determined expression profiles in Salmonella. For example, GcvB accumulates in fast growing cells only (Argaman et al, 2001; Sharma et al, 2007), whereas SdsR and RybB are not expressed until stationary phase when their transcription is activated by the alternative σ^S and σ^E factors, respectively (Vogel et al, 2003; Papenfort et al, 2006; Fröhlich et al, 2012).

A distinct set of sRNAs mark the transition from exponential to stationary phase, as exemplified by the invasion gene‐associated InvR sRNA of Salmonella pathogenicity island 1 (SPI‐1). The SPI‐1 locus is transiently transcribed as aerobic cultures reach an OD₆₀₀ of 2 (Pfeiffer et al, 2007), and it is this cDNA library that contains the vast majority of InvR reads. Other prominent sRNAs in the transition phase are RybD, STnc440 and DapZ; the latter was here renamed from candidate STnc820 due to a genetic association with dapB (see below).

Regarding protein‐coding transcripts, a total of 3517 mRNAs were detected (represented by ⩾10 reads) at the different stages of growth, and 1253 of them were enriched at least three‐fold by coIP with Hfq (Supplementary Table S2), which expands the putative Hfq regulon to more than a quarter of all Salmonella genes. All in all, the new Hfq profiles reveal that the well‐studied changes of primary gene expression over growth are accompanied by consistent and dynamic changes of RNA binding to Hfq, the major hub of post‐transcriptional control.

Hfq‐bound 3′ regions of mRNA genes accumulate as discrete sRNAs

While many of the abundantly recovered sRNAs are transcribed from free‐standing genes, e.g., ArcZ, ChiX, InvR, RprA, RybB, SdsR and Spot42, we noticed that libraries from all growth phases contained ample cDNA reads that overlapped in sense with the 3′ UTR of mRNAs, e.g., DapZ, STnc840, STnc850 and STnc870 (Figures 1B and 2; Supplementary Table S1). Note that for simplicity, we refer to 3′ UTR as either the RNA or DNA region that follows the coding sequence of the respective gene down to the transcription terminator. In other words, 3′ UTR can denote either the transcribed 3′ part of the mRNA or its corresponding DNA in the genome, depending on context. To test whether these cDNAs represent discrete RNA species, rather than premature termination products of cDNA synthesis at the 3′ end of mRNA, we selected candidate regions with both a high cDNA count and a predicted ρ‐independent terminator for northern blot analysis (Supplementary Table S1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Expression analysis of 3′ UTR‐derived sRNAs. Total RNA was prepared from wild‐type Salmonella grown in LB at the time points indicated in Figure 1A, and subjected to northern blot analysis. (*) denotes detection of associated full‐length mRNAs in the cases of DapZ, STnc850, STnc870 and STnc2090. The position of sRNAs, the name and the length of flanking genes, as well as the length (bp) of the intergenic regions are shown in the schematic presentations below the blots. All sRNAs identified are in close proximity to, or even partially overlap with upstream genes. Hybridization probes are listed in Supplementary Table S8.

Of 22 candidates tested, 8 yielded discrete hybridization signals (Figure 2) that agreed well with the respective transcript sizes predicted by Solexa sequencing (Supplementary Table S1). We observed excellent correlations between the northern blot signals of individual sRNAs and their relative coverage in the cDNA libraries. For example, DapZ and STnc840 whose expression sharply peaks at OD₆₀₀ of 2 also show highest recovery in the corresponding cDNA library (compare Figures 2 and 1B, or Supplementary Table S3). Similarly, the northern blots confirmed that STnc850 and STnc870 strongly accumulate in late stationary phase, as predicted by the fact that these sRNAs together constitute 10–25% of all reads in the corresponding cDNA libraries. In several cases, we also detected the mRNA of the same locus, with expression patterns that either matched (STnc850/ycfJ or STnc2090/yfiA) or deviated from (DapZ/dapB or STnc870/cpxP) the sRNA in its 3′ region. We consistently observed much stronger sRNA signals than mRNA signals, supporting the notion that these 3′‐derived sRNAs may accumulate to fulfill an mRNA‐independent function.

An Hfq‐dependent sRNA from the 3′ region of the dapB gene

For proof‐of‐principle that 3′ UTR‐derived sRNAs are functional regulators, we focussed on the ∼80‐nt DapZ sRNA that coincided with the 3′ UTR of dapB (Figure 2). The dapB gene encodes dihydrodipicolinate reductase, an essential protein that catalyses the second step of lysine biosynthesis (Bouvier et al, 2008a). Intriguingly, whereas the coding sequence of dapB is highly conserved in γ‐proteobacteria (Supplementary Figure S1), the 3′ UTR is not, except for the ρ‐independent terminator (Figure 3A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Promoter and sequence analysis of DapZ sRNA. (A) Sequence alignment of the dapB 3′ coding sequence (CDS, grey box) and 3′ UTR of related enterobacterial species. The Salmonella dapZ sequence and its homologous sequences in other species are shown in bold. The arrow denotes the +1 site of dapZ in Salmonella. Putative promoter motifs within the dapB CDS and the ρ‐independent terminator sequence are indicated. The conserved GU‐rich motif R1 (boxed) is found in most species but absent in E. coli and Shigella. ST: Salmonella typhimurium; SB: S. bongori; ET: Enterobacter spp.; CN: Cronobacter spp.; SE; Serratia spp.; PA: Pantoea spp.; KP: Klebsiella pneumonia; YP: Yersinia pestis; ER: Erwinia spp.; CR: Citrobacter rodentium; SF: Shigella flexneri; EC: E. coli; Con: consensus sequence. Below left: DapZ GU‐rich motif R1 (boxed) displays high similarity to GcvB R1. Below right: The secondary structure of DapZ sRNA (the GU‐rich motif R1 is boxed) predicted by Mfold and validated by structure probing (Supplementary Figure S11). (B) The DNA sequence downstream of the dapB start codon down to the ρ‐independent terminator was cloned into a high‐copy plasmid, and the expression of DapZ in wild‐type Salmonella at OD₆₀₀ of 2 was determined by northern blot. (*) denotes read‐through to the rrnB terminator encoded on the plasmid. (C) Identification of the primary transcription start site of dapZ by 5′ RACE. PCR products were analysed on a 4% agarose gel. The arrow indicates the band corresponding to the primary transcript, which is enriched by RNA pre‐treatment with TAP. The appearance of a weaker RACE signal for the full‐length DapZ sRNA in the ‘− TAP’ lane may be attributed to the activity of RppH or a related pyrophosphohydrolases (Deana et al, 2008), which converts to 5′PPP to 5′P ends in vivo. The shorter RACE product in the TAP− lane represents a processing intermediate at nucleotide U12 of DapZ (unpublished results), and as expected, is not found in the TAP+ reaction. (D) Schematic drawing showing that the DapZ sRNA is transcribed from the 3′ UTR of the dapB gene and enriched at early stationary phase in the Hfq coIP library.

We considered that DapZ may be either a stable intermediate of mRNA decay, as suggested by its mutually exclusive accumulation with the full‐length dapB transcript (Figure 2; Supplementary Figure S2), or the product of a hidden sRNA gene. Three experimental results confirmed the latter scenario, i.e., that DapZ is an independently transcribed sRNA, encoded within the 3′ region of dapB. First, the presence of a 5′ truncated, promoterless dapB gene on a multi‐copy plasmid (pdapZ) caused overexpression of DapZ (Figure 3B), indicating that DapZ synthesis did not require transcription of dapB. Second, using a specialized 5′ RACE protocol we detected a transcriptional start site (TSS) immediately downstream of the dapB stop codon that coincided with the 5′ flank of reads in the cDNA libraries (Figure 3C), and was preceded by well‐conserved putative −10 and −35 boxes in the dapB CDS (Figure 3A). Assuming that DapZ transcription terminates at the dapB terminator, this TSS perfectly matches the observed sRNA size on the northern blot (Figure 2). Third, the main DapZ transcript is resistant to treatment with a 5′ monophosphate‐dependent exonuclease (Kröger et al, 2012), which is consistent with DapZ being a primary transcript that possesses a 5′ triphosphate.

A comparison of RNA abundance and stability between Salmonella wild‐type and an isogenic Δhfq mutant in early stationary phase confirmed that DapZ is an abundant Hfq‐dependent sRNA. At the height of its expression, DapZ accumulates to ∼100 copies/cell, but the lack of Hfq reduces the in vivo copy number to less than one (Supplementary Figure S3). The RNA half‐life was reduced from ∼2.5 min to <30 s in the Δhfq strain, suggesting that as with many other sRNAs (Andrade et al, 2012), Hfq protects DapZ from degradation. Note that we also observed shorter, processed DapZ species in both the RACE and northern blot experiments. However, these partially or fully lack the seed region of DapZ (as defined below) and so are likely non‐functional decay intermediates of the sRNA. Altogether, these results established that DapZ is an Hfq‐dependent sRNA that shares the 3′ region and terminator of the dapB mRNA gene (Figure 3D). Remarkably, both its proximity to the dapB reading frame and its low degree of sequence conservation would prevent faithful prediction of this sRNA by biocomputational methods.

The master regulator of virulence HilD regulates dapZ

Although their genes partly overlap, the dapB mRNA and DapZ sRNA are clearly differentially expressed (Figure 2; Supplementary Figure S2). Expression of dapB, which according to E. coli data may be regulated by lysine and the LysR‐type regulator ArgP (Bouvier et al, 2008a), was highest during exponential growth. By contrast, DapZ sharply accumulated in the transition to the stationary phase, matching the profile of the SPI‐1‐associated InvR sRNA (Figure 1B; Pfeiffer et al, 2007), which raised the intriguing possibility that DapZ was co‐activated with the SPI‐1 invasion genes. This hypothesis was supported by northern blot analysis of Salmonella strains with genomic deletions of the major virulence islands. Figure 4A shows that a ΔSPI‐1 mutation diminished DapZ expression, whereas a ΔSPI‐2 mutation, affecting the genes for intracellular survival, did not affect DapZ expression.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

DapZ is activated by the Salmonella‐specific virulence regulator HilD. (A) Salmonella wild‐type as well as several deletion mutant strains were grown in LB to an OD₆₀₀ of 2, and total RNA was probed for DapZ expression by northern blot. 5S rRNA served as loading control. (B) Salmonella wild‐type, ΔhilD and ΔSPI‐1 strains were transformed with an arabinose‐inducible pBAD control plasmid, pBAD‐hilD or pBAD‐hilA (ΔSPI‐1 strain) and cultivated in LB to OD₆₀₀ of 1.0. Expression from pBAD plasmids was induced by addition of 0.2% l‐arabinose (final conc.) for ∼45 min and DapZ levels were determined by northern blot. (C) Western blot analysis of GFP reporters in which dapZ homologues from several related enterobacteria were fused to a promoterless gfp gene: Salmonella typhimurium (ST), Yersinia pestis (YP), Klebsiella pneumonia (KP), Citrobacter rodentium (CR) and E. coli (EC). P_ompC and P_hilA are the control promoter regions known to be non‐responsive or responsive, respectively, to HilD. E. coli co‐transformed with the indicated GFP‐reporter plasmids as well as the pBAD‐hilD (hilD) plasmid were grown in LB for 2 h in presence of 0.0004% l‐arabinose after reaching OD₆₀₀ of 0.5. Probing for GroEL served as loading control.

Next, to identify the responsible transcription factor, we investigated DapZ expression in strains with individual disruptions of the hilA, hilC, hilD, invF and rtsAB genes, which encode transcriptional activators that control the SPI‐1 genes in a hierarchical manner (Ellermeier and Slauch, 2007). These experiments suggested HilD, which acts at the very top of the SPI‐1 cascade, as the cognate activator of DapZ expression (Figure 4A). The regulation is also observed with a reporter gene fusion to the dapZ promoter, arguing that HilD indeed regulates DapZ at the level of transcription (Supplementary Figure S4).

Ectopic expression of HilD restored DapZ levels in Salmonella ΔhilD or ΔSPI‐1 strains (Figure 4B). Again, this was specific to HilD, since the downstream acting HilA factor could not restore DapZ levels in these mutant backgrounds. Most importantly, however, HilD protein also activated the Salmonella dapZ promoter after transfer to E. coli, an organism that lacks SPI‐1 and all of its associated transcription factors (Figure 4C). Although a molecular interaction remains to be proven, this trans‐complementation strongly argues that HilD activates the dapZ promoter directly.

The results of additional trans‐complementation experiments in E. coli predict that the control of the widely conserved dapZ gene by the Salmonella‐specific factor HilD evolved very recently, i.e., after Salmonella diverged from the E. coli lineage. That is, of five selected enterobacterial dapZ promoters tested in this assay (Figure 4C), the Salmonella promoter was the only one to respond to HilD, but did so as strongly as the hilA promoter, a positive control and well‐established direct target of HilD (Schechter and Lee, 2001). By contrast, the other four dapZ promoters were invariably insensitive to HilD expression, and either showed intermediate (Citrobacter, Klebsiella and Yersinia) or very high basal expression (E. coli). Thus, closely related enterobacteria may use different factors to control the dapZ promoter.

DapZ sRNA regulates major ABC transporters with a GcvB‐like seed domain

Hfq‐associated sRNAs typically regulate gene expression by base pairing with mRNAs. To identify targets of DapZ in Salmonella, we took a pulse‐expression approach (Masse et al, 2005; Papenfort et al, 2006) analysing global mRNA level changes after a transient overexpression of DapZ from an inducible plasmid. Within 10 min of induction, DapZ downregulated the conserved dpp and opp operons, which encode major ABC transporters, and the yahO and STM1513 genes of unknown function ⩾2‐fold (Figure 5A). This rapid downregulation suggests that DapZ regulates these mRNAs directly.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

DapZ is a repressor of the opp and dpp operons. (A) Microarray analysis of genes affected by pulse expression of DapZ compared to pBAD control vector in Salmonella. Salmonella dapZ mutants containing a pBAD control vector or pBAD‐DapZ plasmid were grown in LB until OD₆₀₀ of 1.5 and then 0.2% l‐arabinose was added to both cultures for 10 min to induce DapZ expression (Supplementary Figure S12). Global transcriptome changes were scored on Salmonella‐specific microarrays. Genes which show ⩾2‐fold change (P‐value <0.1) are marked in red. (B) Heat map analysis of microarray results of genes regulated by pulse expression of wild‐type DapZ and DapZ‐ΔR1. All the experimentally validated direct targets of the Salmonella GcvB sRNA (Sharma et al, 2007, 2011) are shown, and the fold‐change values are listed in Supplementary Table S5. (C) Endogenous DapZ represses oppA and dppA protein synthesis in Salmonella at OD₆₀₀ of 2. Translational lacZ fusions were constructed in the Salmonella chromosome by fusing lacZ to the 17th codon of oppA or the 10th codon of dppA. β‐Galactosidase activity in wild‐type, ΔdapZ, ΔgcvB and ΔdapZΔgcvB double mutant Salmonella was determined in triplicates.

Intriguingly, the DapZ candidate targets appeared to be a subset of the many targets of the widely conserved Hfq‐associated sRNA, GcvB (Figure 5B). Moreover, the putative DapZ targets overlapped specifically with those mRNAs that GcvB recognizes by its G/U‐rich seed domain (Sharma et al, 2011), which suggested that DapZ may select targets by a similar mechanism. Indeed, inspection of the DapZ sequence revealed a single‐stranded GUGAUGUGGUU (nucleotides 11–21) stretch that is conserved in enterobacteria except for E. coli and Shigella (Figure 3A). In analogy to its counterpart in GcvB, this G/U‐rich stretch will be referred to as domain R1 of DapZ (Figure 3A). By repeating the pulse‐expression experiment with an R1 mutant of DapZ (internal deletion of nucleotides G₁₃A₁₄U₁₅G₁₆), we observed that this domain was indeed essential for the repression of dpp and opp operons (Figure 5B).

GcvB represses the dpp and opp operon mRNAs through base‐paring interactions with the first cistrons, dppA and oppA, respectively (Sharma et al, 2007; Pulvermacher et al, 2008). Using translational lacZ fusions to the chromosomal dppA and oppA genes of Salmonella, we found that DapZ also exerts negative regulation at the 5′ end of these operons. A ΔdapZ mutation activated the oppA::lacZ reporter, both alone and further with a ΔgcvB mutation. The dppA::lacZ reporter was also upregulated upon deletion of dapZ, although this required prior inactivation of the gcvB gene (Figure 5C). Perhaps when GcvB is present, it occupies the dppA target site owing to its more stable binding relative to DapZ (Sharma et al, 2007), so the dppA target is only bound and suppressed by DapZ when its competitor GcvB RNA is absent. Northern blot analysis showed that the two sRNAs did not influence each other's expression (Supplementary Figure S5), suggesting that DapZ is able to regulate dppA and oppA independently of GcvB. Likewise, the expression of dapB was not significantly affected by the ΔdapZ mutation (Supplementary Figure S6) or DapZ overexpression (Supplementary Figure S7), ruling out the possibility of indirect effects through an impaired metabolic function of DapB. Altogether, these experiments showed that the major ABC transporters are regulated independently of protein output from the dapBZ locus, and that the DapZ sRNA resembled GcvB such that it repressed some shared targets, likely via a similar seed domain.

Evidence for similar seed pairing by DapZ and GcvB

To elucidate how DapZ recognizes the dppA and oppA targets, we subjected in vitro transcribed DapZ and the 5′ mRNA regions to structure probing with single strand‐specific RNase T1 or lead(II), alone and after mixing sRNA and target. Preliminary gel‐shift assays confirmed that DapZ formed complexes with the oppA and dppA RNA fragments (Supplementary Figure S8). Probing of the 5′ labelled DapZ RNA showed that presence of either the dppA and oppA RNAs protected the G/U‐rich R1 domain of the sRNA from cleavage, which was most pronounced with lead(II) backbone cleavage (Figure 6A). Reciprocally, probing of the targets identified DapZ‐induced protection of C/A‐rich sites around the start codon of oppA, or upstream of the Shine‐Dalgarno sequence in dppA, as well as a structural rearrangement in dppA (Figure 6B and C). In other words, although the target interactions of DapZ and GcvB slightly differ with respect to helix length and continuity (Figure 6E), the two sRNAs essentially recognize the same regions in each of the two mRNAs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

DapZ targets C/A‐rich sites in oppA and dppA. Identification of duplex formation sites by in vitro secondary structure probing using 5′ end‐labelled DapZ sRNA (A), oppA (B) and dppA (C) mRNA leaders. Radio‐labelled RNA (∼5 nM) was subjected to RNase T1 or lead (II) cleavage in absence or presence of unlabelled 100 nM (+), 500 nM (++) oppA and 20 nM (+), 100 nM (++) dppA in (A), 100 nM (+), 500 nM (++) DapZ in (B), and 20 nM (+), 100 nM (++) DapZ in (C). C: control RNA, T1: RNase T1 ladder, OH: alkaline ladder. The G residues are labelled relative to the translation start site in oppA and dppA mRNA leaders. The regions protected by duplex formation with cold RNA are marked with red square brackets. The DapZ R1 region is indicated with a blue bar. (*) denotes a structure rearrangement in dppA, which we tentatively exclude as another targeting site, because the GAGUAUUUCCUU nucleotides (+3 to +14 of dppA) in question have no obvious complementarity with DapZ. Thus, there might be further structural rearrangement of the mRNA upon DapZ binding, which would also explain the difference in migration of dppA leader RNA in native gels (Supplementary Figure S8). (D) Validation of the base‐pair interactions using translational oppA::gfp and dppA::gfp reporter gene fusions by compensatory base‐pair exchange in vivo. Salmonella strains containing both a gfp reporter plasmid and a vector overexpressing DapZ were grown overnight in LB and analysed by flow cytometry. Overexpression of a ∼50 nt nonsense RNA was used as control (pJV300). (E) RNA duplexes formed between DapZ sRNA and the dppA or oppA leaders. Nucleotides in bold in oppA and dppA were previously shown to be involved in binding to GcvB sRNA (Sharma et al, 2007); the GcvB‐dppA and GcvB‐oppA interactions are shown for comparison below. Point mutations introduced for compensatory base‐pair exchange experiments are indicated. The ribosome binding site and the start codon are marked in orange.

To prove that the predicted RNA duplexes guided DapZ‐mediated mRNA repression in vivo, we introduced compensatory point mutations in DapZ and the dppA or oppA mRNAs (Figure 6D and E). DapZ was constitutively expressed from a plasmid in a ΔdapZΔgcvB strain, and target regulation was monitored using translational dppA::gfp or oppA::gfp fusions, respectively, on compatible plasmids (Sharma et al, 2007; Urban and Vogel, 2007). Flow cytometry‐based measurements of GFP fluorescence showed that DapZ repressed the oppA::gfp and dppA::gfp reporters 6‐fold and 2.3‐fold, respectively. Repression was abolished with DapZ‐M1, a mutant sRNA with a single G₁₃→C change in the G/U‐rich seed (Figure 6D and E). Likewise, M1′ variants of the dppA::gfp and oppA::gfp reporters containing the opposite C→G point mutation were insensitive to wild‐type DapZ. However, restoration of the predicted RNA interactions by combining DapZ‐M1 with the M1′ reporters of dppA or oppA fully restored repression, which validates that DapZ employs its G/U‐rich seed to repress dppA and oppA by direct base pairing in vivo.

Discovery of sRNAs via Hfq profiling

Biocomputational searches have traditionally focussed on the discovery of free‐standing sRNA genes in the IGRs of bacterial chromosomes, and relied much on the conservation of transcription elements including the 3′ terminal ρ‐independent terminators (Vogel and Sharma, 2005; Backofen and Hess, 2010). However, 3′ UTR‐derived sRNAs such as DapZ pose a challenge to in‐silico prediction owing to their close proximity or even overlap with mRNA sequences. The dapZ gene indeed escaped detection in biocomputational sRNA screens (Argaman et al, 2001; Rivas and Eddy, 2001; Wassarman et al, 2001; Chen et al, 2002; Livny et al, 2006; Pfeiffer et al, 2007), which we attribute to both its poorly conserved primary sequence and proximity to the dapB reading frame. Interestingly, DapZ was also not recognized in cDNA cloning‐based screens (Vogel et al, 2003; Kawano et al, 2005) or earlier analyses of Hfq‐bound transcripts by tiling arrays (Zhang et al, 2003) or deep sequencing (Sittka et al, 2008). By contrast, our present study builds upon deep profiling of Hfq ligands over growth and dedicated inspection of UTR‐derived transcripts (Figure 1B). In addition, since Hfq is a limiting factor in vivo (Lease and Woodson, 2004; Fender et al, 2010; Moon and Gottesman, 2011), the high recovery of DapZ (∼12% of reads in the transition phase library) was a strong predictor of a physiological function. That is, other sRNAs that dominate the Hfq profile at this condition are known to be functional: ChiX and InvR repress porin synthesis (Pfeiffer et al, 2007; Figueroa‐Bossi et al, 2009; Rasmussen et al, 2009), and ArcZ and RprA regulate the rpoS mRNA and additional targets (Papenfort et al, 2009; Mandin and Gottesman, 2010). By the same token, prominent Hfq‐binding sRNAs from 3′ UTRs which accumulate in different growth phases may similarly turn out as bona fide regulators of gene expression.

The general trends of sRNA expression are well reflected by our Hfq profiling over growth, as clearly seen for the abundant marker sRNAs such as GcvB, DapZ/InvR and RprA/SdsR/RybB, which accumulate in early, middle or late growth stages, respectively (Figure 1B). Copy numbers are available for several Salmonella sRNAs, and often match the relative recovery of a given sRNA. For instance, both DapZ and InvR accumulate to ∼100 copies/cell in early stationary phase (Supplementary Figure S3; Pfeiffer et al, 2007), and are equally represented in the corresponding library (Figure 1B). Under the same condition, ArcZ and SdsR are present in ∼20 copies/cell, respectively (Papenfort et al, 2009; Fröhlich et al, 2012), and irrespective of deviations from expected numbers, these sRNAs generally exhibit lower recovery than DapZ and InvR (Figure 1B). Nonetheless, many factors may influence the recovery rate, ranging from non‐linear binding to Hfq to sRNA‐specific biases during cDNA preparation and sequencing. Thus, low abundance cannot be used to rule out physiological activity. For example, although SgrS RNA constitutes <0.1% of all reads in early stationary phase, it strongly regulates the mRNA of the secreted virulence factor SopD under this condition (Papenfort et al, 2012).

Our investigation into the narrow expression timing of DapZ uncovered a recently evolved transcriptional control by HilD (Figure 4). HilD is the master transcriptional activator of the SPI‐1 invasion genes, and additional targets outside SPI‐1 are typically among those ∼25% of Salmonella genes that were horizontally acquired since Salmonella and E. coli diverged from a common ancestor (Porwollik and McClelland, 2003; Ellermeier and Slauch, 2004). Importantly, HilD activation of the dapZ promoter within dapB provides Salmonella with a selective repressor of major ABC transporter synthesis under invasion conditions, as compared to recruiting the GcvB sRNA whose activity would impact amino‐acid uptake and synthesis in a much broader fashion. In more general terms, HilD and DapZ constitute a novel paradigm of how the core and accessory parts of the Salmonella genome are intermeshed by Hfq‐dependent regulation at the RNA level. Contrasting previously reported sRNA‐mediated mRNA control across the conserved and virulence regions (Pfeiffer et al, 2007; Papenfort et al, 2012), this study reveals novel cross‐wiring wherein a horizontally acquired transcription factor (HilD) employs a conserved Hfq‐dependent sRNA (DapZ) to regulate Salmonella core gene (dpp and opp) expression. Considering that Salmonella expresses >140 sRNAs (Pfeiffer et al, 2007; Padalon‐Brauch et al, 2008; Sittka et al, 2008; Kröger et al, 2012), additional Hfq‐dependent regulation connecting the two parts of the Salmonella genome will surely be discovered.

We present the most comprehensive profiling of Hfq‐associated RNAs for any organism to date. This atlas of Hfq targets must now be expanded by the integration of more growth and stress conditions, and genome‐wide maps of transcription start and RNA polymerase binding sites (Cho et al, 2009; Mooney et al, 2009; Kröger et al, 2012), to unravel the full scope of the Hfq network.

3′ UTR‐derived sRNAs and dual output from mRNA loci

Most mRNAs and sRNAs are encoded by loci with single output function, but exceptions from simplicity have been known. Some small genes give rise to a dual‐function mRNA which both serves as the template for synthesis of a short peptide and acts as an antisense regulator on other mRNAs in trans (Wadler and Vanderpool, 2007; Sonnleitner et al, 2008; Gimpel et al, 2010; Romby and Charpentier, 2010). In addition, earlier studies in E. coli detected abundant sRNAs from UTRs (Tjaden et al, 2002; Vogel et al, 2003; Zhang et al, 2003; Kawano et al, 2005). These observations were conceptualized in a model of ‘parallel transcriptional output’ wherein some protein‐coding genes may produce both an mRNA template for translation and a regulatory RNA, by dual use of either the promoter or terminator (Vogel et al, 2003). While this model recently received support by the discovery of 5′ UTR (riboswitch)‐derived sRNAs in Listeria monocytogenes that regulate a virulence factor in trans (Loh et al, 2009), DapZ represents the first example of dual output via the 3′ UTR of a widely conserved gene.

3′ UTR‐derived sRNAs may be generated by two major biogenesis pathways (Figure 7). The first is exemplified by DapZ, which is transcribed from an mRNA‐internal promoter and thus expressed and functions independently of the host mRNA gene despite a partial overlap in sequence. In the other pathway, the sRNA is generated by mRNA processing in the 3′ region. Inspection of 5′ end status in available dRNA‐seq data sets wherein primary and processed transcripts can be differentiated (Sharma et al, 2010; Kröger et al, 2012) supports a notion that the 3′ derived sRNAs shown in Figure 2 represent examples of both pathways: Aside from DapZ, the STnc860 sRNA (RyeF) is a primary transcript, with its promoter being located ∼260 bp upstream of the cutC stop codon (Supplementary Figure S9). RyeF was originally discovered by Hfq coIP in E. coli (Zhang et al, 2003) but it was previously not annotated in Salmonella due to poorly conserved flanking genes and lack of expression in early stationary phase (Figure 2; Sittka et al, 2008, 2009). Overall, these sRNAs with mRNA‐internal promoters bolster recent findings that transcription start sites within the coding sequences of E. coli and Salmonella may be common (Kawano et al, 2005; Mendoza‐Vargas et al, 2009; Kröger et al, 2012).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Biogenesis of sRNAs from bacterial UTRs. A 3′ UTR‐derived small RNA can be either transcribed from its own promoter in the upstream coding sequence, or generated by internal processing of the associated mRNA. The common denominator is the shared use of the ρ‐independent terminator of the mRNA. Hfq plays a seminal role in either pathway such that it facilitates the base pairing of the 3′ UTR‐derived sRNA with trans‐encoded target mRNA(s), but it may also participate in recruiting a nuclease (such as RNase E) to the 3′ end of the mRNA in the case of processing.

Processing in the mRNA 3′ region likely underlies the generation of STnc840, STnc870 and STnc2090, because these sRNAs seem to carry a 5′ monophosphate end (according to dRNA‐seq data by Kröger et al, 2012) and lack obvious motifs for RNAP binding in the respective upstream DNA regions. The processing must come with a cost given that these cleavages collectively occur in the coding sequence and so render the respective mRNA inactive for further rounds of translation. The RybD sRNA, which was originally detected at the 3′ end of the conserved sucABCD operon in E. coli (Zhang et al, 2003), also belongs to this category. Salmonella RybD is prominent in the early stationary phase (Figure 1B) and likely generated by mRNA cleavage 12 nucleotides upstream of the sucD stop codon (Supplementary Figure S10). A promising candidate for the responsible nuclease is RNase E, the major endoribonuclease in Gram‐negative bacteria (Belasco, 2010; Bouvier and Carpousis, 2011) which can cleave in the 3′ regions of mRNAs by either processive action from the 5′ end (Mackie, 1998) or direct internal entry (Kime et al, 2010). Of note, RNase E was proposed to process MicX sRNA from a long 3′ UTR in Vibrio cholerae (Davis and Waldor, 2007). RNase E and Hfq can form a complex (Morita et al, 2005; Ikeda et al, 2011), which could then be guided to the 3′ end of mRNAs by the recently discovered propensity of Hfq to bind to ρ‐independent terminators (Otaka et al, 2011; Sauer and Weichenrieder, 2011; Ishikawa et al, 2012). While experiments are in progress to map Hfq sites and RNase E‐dependent cleavage in the new 3′ derived sRNAs, we note that many Hfq‐associated mRNAs show enrichment in the putative terminator region. Of the predicted 770 ρ‐independent terminators of Salmonella mRNA genes, 291 were significantly enriched by coIP with Hfq (Supplementary Table S4), which illustrates a vast array of candidate regions for new 3′ UTR‐derived sRNAs.

Functional classification of sRNA by seed

An exciting finding of this paper is the discovery of a GcvB‐like seed region in DapZ, with potential ramifications for sRNA ontology. Except for E. coli and Shigella where the R1 region is mutated, all predicted DapZ RNAs contain this ∼15 nt long G/U‐rich seed which, according to our probing experiments, targets the same C/A‐rich sites in dppA and oppA as does the longer R1 domain of GcvB (Figure 6; (Sharma et al, 2007). Inferring from the proven inhibition of ribosome loading by GcvB at these sites (Sharma et al, 2007), we posit that DapZ represses the dppA and oppA mRNAs primarily at the level of translation initiation.

The concept of bacterial seed domains is recent (Storz et al, 2011; Vogel and Luisi, 2011), and arose through observations that the nucleotides involved in binding to target mRNAs are usually conserved (Vanderpool and Gottesman, 2004; Udekwu et al, 2005; Sharma et al, 2007) and that seed regions as small as 13 nucleotides retain their function upon transplantation to unrelated sRNAs (Bouvier et al, 2008b; Pfeiffer et al, 2009; Papenfort et al, 2010). The R1 domains of DapZ and GcvB strongly support the concept of independent seed regions in sRNAs, and may even constitute a case of convergent evolution. The coinciding binding regions of DapZ and GcvB in dppA and oppA may indicate that these mRNAs are constrained with respect to where sRNAs can operate effectively, e.g., in terms of target site accessibility and the presence of an Hfq site (Busch et al, 2008; Link et al, 2009; Peer and Margalit, 2011; Beisel et al, 2012). This observation adds to recent reports of clustered sRNA sites in the rpoS (Mandin and Gottesman, 2010), csgD (Holmqvist et al, 2010; Jorgensen et al, 2012; Mika et al, 2012; Thomason et al, 2012), ompD (Balbontin et al, 2010; Gogol et al, 2011; Fröhlich et al, 2012) and sdhC mRNAs (Desnoyers and Masse, 2012).

We presently tend to classify Hfq‐associated sRNAs according to their cognate transcription factors (Corcoran et al, 2011). As more sRNAs are discovered, one may want to consider an alternative nomenclature that is based on the seed. Accordingly, DapZ would be a GcvB‐like sRNA, that is similar to referring to HilD as an AraC‐like transcription factor because it carries the same DNA‐binding domain as the ancestral AraC protein (Schechter et al, 1999). Note, however, that while ‘domain’ in a transcription factor denotes a structurally recognizable, self‐folding entity of a protein that confers a specific function for DNA binding, the seed region is defined primarily by the specific mRNA interactions it bestows on an sRNA. A nomenclature that follows the seed would seem particularly appropriate for sRNAs like DapZ whose Salmonella‐specific control by HilD rejects the cognate transcription factor as a useful classifier.

Bacterial strains, media and growth conditions

Salmonella enterica serovar Typhimurium strain SL1344 (JVS‐1574) was used as wild‐type strain and for mutant construction. The complete list of bacterial strains used in this study is provided in Supplementary Table S6. Bacteria were grown in Luria Bertani (LB) medium at 37°C at 220 r.p.m. When appropriate, 100 μg/ml ampicillin, 50 μg/ml kanamycin or 20 μg/ml chloramphenicol (final concentrations) were added to the medium, or used in agar plates.

Strain construction

Chromosomal mutagenesis of Salmonella SL1344 was performed as previously described (Datsenko and Wanner, 2000). To construct the dapZ deletion strain (JVS‐9207), 3′ UTR of dapB was replaced by the ‘scar’ sequence while the ρ‐independent terminator was kept intact. Wild‐type Salmonella containing pKD46 was electroporated with 300–500 ng DNA amplified from pKD4 with oligonucleotides JVO‐7698/‐5641 (see Supplementary Table S8 for sequences of deoxyoligonucleotides). The kanamycin resistance cassette was eliminated using the FLP helper plasmid pCP20 at 42°C (Datsenko and Wanner, 2000). The same strategy was applied constructing the ΔrtsAB strain using primer pair JVO‐5604/‐5605 (verified by PCR with JVO‐5606/‐5607).

For construction of translational oppA‐17aa‐lacZ (JVS‐8992) and dppA‐10aa‐lacZ (JVS‐8996) fusions in the chromosome, the 3′ part of the oppA or dppA coding sequences were first replaced by kanamycin resistance cassette amplified with oligonucleotides JVO‐7322/‐7323, or JVO‐7324/‐7325 from pKD13. The resulting mutants were ‘healed’ by pCP20, and then transformed with pCE40 to generate translational lacZ fusions (Ellermeier et al, 2002). The insertion of lacZ gene was verified by colony PCR using oligonucleotides pMC874‐lac and JVO‐0421(oppA) or JVO‐0423(dppA). All mutations were transduced into fresh wild‐type or desired Salmonella background using phage P22 (Sternberg and Maurer, 1991).

Plasmid construction

A complete list of plasmids used in this study can be found in Supplementary Table S7. In order to make the pdapZ (pYC20) construct, dapZ and ∼800 bp upstream sequence (until the start codon of dapB CDS) was amplified using oligonucleotides JVO‐5373/‐5374, digested with XbaI and XhoI and ligated plasmid pZE12‐luc digested with the same enzymes. To construct an l‐arabinose inducible dapZ construct (pBAD‐dapZ, pYC39), the dapZ gene amplified by oligonucleotides JVO‐5646/‐5374 was inserted into pBAD‐His‐myc (Invitrogen) backbone which was amplified by oligonucleotides JVO‐0900/‐0901. To generate the constitutive dapZ overexpression plasmid (pYC40‐2) driven by the P_LlacO promoter, the very same insert was cloned into pZE12‐luc backbone, which was prepared by PCR amplification with oligonucleotides PlacB and PlacD with Phusion DNA polymerase (Finnzymes, Finland).

The 4 bp (GATG) in dapZ R1 region was deleted from pYC39 by overlapping PCR with oligonucleotides JVO‐8992/‐8993. After DpnI digest, the PCR product was transformed into E. coli TOP10 for generating plasmid P_BAD‐dapZ‐ΔR1 (pYC108). Likewise, the point mutation in dapZ was introduced in pYC40‐2 by overlapping PCR with JVO‐7197/‐7198 resulting in plasmid dapZ‐M1(pYC73‐2); the compensatory point mutation in oppA was introduced in pJL19‐1 by overlapping PCR with JVO‐7199/‐7200 resulting in plasmid oppA‐M1’(pYC74). The compensatory point mutation in dppA was introduced in pJL18‐1 with JVO‐7201/‐7202 resulting in plasmid dppA‐M1’ (pYC75‐3).

Transcriptional P_dapZ–gfp fusion plasmids were constructed by cloning DNA sequences containing putative promoter elements into plasmid pAS0046 via AatII/NheI sites as previously described (Pfeiffer et al, 2007). DNA sequences containing putative dapZ promoters in enterobacteria species (∼110 to +5 bp relative to the +1 site of Salmonella dapZ) were amplified from genomic DNA of wild‐type Salmonella with JVO‐7635/‐7636, E. coli K‐12 with JVO‐7637/‐7638, Yersinia pestis KUMA with JVO‐7639/‐7640, Klebsiella pneumonia MGH78578 with JVO‐7641/‐7642 and Citrobacter rodentium with JVO‐7643/‐7644.

coIP of Hfq and deep sequencing analysis

Wild‐type Salmonella SL1344 and the 3 × FLAG‐tagged Hfq strain (hfq^FLAG, JVS‐1338) were grown in LB overnight for about 16 h (220 r.p.m., 37°C). Cells equivalent to an OD of 50 were collected for coIP, and another 5 ml overnight culture was washed by PBS twice, diluted 1:200 to 1 l fresh LB and grown for 9 h after reaching OD₆₀₀ of 2. At several time points during growth (Figure 1A), a 50 OD culture of wild‐type and hfq^FLAG strain were collected by centrifugation (4000, r.p.m., 4°C) and subjected to coIP according to the protocol previously described by Pfeiffer et al (2007) and Sittka et al (2008), with slight modifications. Briefly, bacteria were resuspended in 0.8 ml of lysis buffer (20 mM Tris pH 8, 150 mM KCl, 1 mM MgCl₂, 1 mM DTT), and disrupted with 0.8 ml glass beads (Roth, 0.1 mm diameter) by rigorous vortexing (30 s burst followed by 30 s chill on ice) for 5 min. The cleared lysate was incubated with 35 μl anti‐FLAG antibody (Sigma; #F1804) at 4°C for 30 min and incubated with 75 μl of Protein A sepharose (Sigma; P‐6649‐5ML) for another 30 min. After five washes in lysis buffer, the sepharose was subjected to RNA and protein preparation by Phenol:Chloroform:Isopropanol extraction. After DNase I digestion, the RNA was used to construct cDNA libraries by Vertis Biotechnologie AG (Munich, Germany), and sequenced on a Solexa GAIIx machine. Raw cDNA reads were quality trimmed (cutoff Phread score of 20) and poly‐A clipped. Solexa reads (⩾20 nt) were mapped to the Salmonella SL1344 genome (http://www.sanger.ac.uk/Projects/Salmonella/) using segemehl with a minimal accuracy of 95% (Hoffmann et al, 2009). The per nucleotide coverage was calculated, normalized by the number of total number of mapped of reads for each library and visualized in the Integrated Genome Browser (Nicol et al, 2009). Gene‐wise quantity analysis was performed by counting the number of reads which overlap at least for 10 nt with genes annotations in the set recently generated by Kröger et al (2012). The sequencing data have been deposited in the GEO database (accession no. GSE38884). To calculate the enrichment of RNAs, the number of reads from Hfq coIP library was divided by that from control coIP library, followed by normalization to the total number of mapped reads.

RNA isolation and northern hybridization

RNA isolation and northern hybridization experiments were performed as previously described (Pfeiffer et al, 2007; Sittka et al, 2008). Briefly, samples were collected with addition of 0.2 vol/vol of STOP solution (95% ethanol, 5% phenol) before the RNA was prepared with Trizol reagent (Invitrogen). 5–10 μg total RNA was denatured for 5 min at 95°C in RNA loading buffer (95% formamide, 0.1% xylene cyanole, 0.1% bromophenol blue and 10 mM EDTA), separated on 7 M urea/ 6% polyacrylamide gels, and transferred onto Hybond‐XL membranes (GE Healthcare) by electroblotting (50 V) for 1 h at 4°C. Oligos were 5′ end‐labelled with γ‐³²P by PNK (Fermentas). All oligonucleotides used for detection of DapZ, newly identified 3′ UTR sRNAs (Figure 2) and 5S rRNA are listed in Supplementary Table S8. The 5′ end‐labelled oligos were hybridized to membranes overnight at 42°C, before washing with 5 × SSC/0.1% SDS, 1 × SSC/0.1%SDS and 0.5 × SSC/0.1% SDS for 15 min each. Signals were visualized on a phosphoimager (Typhoon FLA 7000, GE Healthcare) and quantified using the AIDA software (Raytest).

5′ RACE

Primary transcription start sites were determined with a modified 5′ RACE protocol (Bensing et al, 1996) as described in detail by Pfeiffer et al (2007). Briefly, 12 μg total RNA was split in two parts while only one was treated with Tobacco Acid Pyrophosphatase (TAP; Epicentre) at 37°C for 30 min. Subsequently, RNA was ligated to a 5′ RNA oligonucleotide adaptor using T4 RNA ligase at 17°C overnight. Following purification, half of the ligated RNA was used for reverse transcription with a random hexamer oligonucleotide primer mix using Superscript III RT (200 U final; Invitrogen) with the following programme: 25°C for 5 min, 50°C for 60 min, 70°C for 15 min. Oligonucleotides JVO‐4661 and JVO‐0367 (antisense to the RNA linker) were used to amplify the 5′ end DapZ by PCR with Taq polymerase (New England Biolabs), following 35 cycles of: 95°C for 20 s, 56°C for 20 s, 72°C for 20 s. The PCR products were separated on a 4% agarose gel and the sequence of the TAP‐specific band was determined by Sanger sequencing.

Microarray analysis

Microarray analysis of pBAD‐induced sRNAs expression has been described previously (Papenfort et al, 2006, 2009, 2012). In brief, wild‐type Salmonella was transformed with plasmid pKP8‐35 (pBAD control), pYC39 (pBAD‐DapZ) or pYC108 (pBAD‐DapZ‐ΔR1), and grown in LB until OD₆₀₀ of 1.5. The expression of sRNA was induced for 10 min with 0.2% l‐arabinose, and total RNA was prepared for microarray analysis with Trizol reagent as described above. Differential expression was considered when genes displayed ⩾2‐fold changes in all replicates and were statistically significantly different (Student's t‐test). Statistical analysis, data visualization and data mining were analysed using GeneSpring 7.3 (Agilent). The microarray data have been deposited in the GEO database (accession no. GSE38523).

Western blot and antibodies

Bacteria culture were collected by centrifugation for 2 min at 16 100 g at 4°C, and pellets were resuspended in 1 × protein loading buffer to a final concentration of 0.01 OD/μl. After incubation for 5 min at 95°C, 0.05 OD equivalents of samples were separated on 12% SDS–PAGE. GFP fusion and GroEL proteins were detected as described in Urban and Vogel (2007). HilD was detected with a polyclonal anti‐HilD antibody.

β‐galactosidase and GFP reporter assays

β‐Galactosidase activity was determined with ortho‐Nitrophenyl‐β‐galactoside (ONPG) as substrate and the CHCl3‐SDS permeabilization procedure (Miller, 1972). To assay GFP reporters, bacteria were grown in LB in presence of ampicillin and chloramphenicol overnight. Bacterial cells corresponding to 1 OD were pelleted and fixed with 4% paraformaldehyde. The GFP fluorescence intensity was quantified by flow cytometry with FACS Calibur (BD Bioscience). All experiments were performed in triplicates.

In‐vitro structure mapping and foot‐printing

DNA templates carrying a T7 promoter sequence for in‐vitro transcription were generated by PCR. Primers and sequences of the T7 transcripts have been deposited in Supplementary Table S7. RNA was in vitro transcribed and quality checked as described (Sharma et al, 2007; Sittka et al, 2007). The protocol for 5′ end labelling of RNA has been described previously (Papenfort et al, 2006).

Secondary structure probing and mapping of RNA duplexes were conducted with ∼0.1 pmol 5′ end‐labelled RNA in 10 μl reactions. RNA was denatured for 1 min at 95°C and chilled on ice for 5 min, upon which 1 μg of yeast tRNA and 10 × structure buffer (0.1 M Tris at pH 7.0, 1 M KCl, 0.1 M MgCl₂; Ambion), and unlabelled mRNA/sRNA were added to anneal at 37°C for 10 min. Thereafter, 2 μl of RNase T1 (0.05 U/μl; Ambion, #AM2283) or 2 μl of a fresh solution of lead(II) acetate (25 mM; Fluka #15319) wsd added and incubated for additional 3 or 1 min at 37°C, respectively. Reactions were stopped by adding 12 μl cold loading buffer II (95% Formamide; 18 mM EDTA; 0.025% SDS, Xylene Cyanol and Bromophenol Blue; Ambion). RNase T1 ladders were obtained by incubating labelled RNA (∼0.2 pmol) in 1 × sequencing buffer (Ambion) for 1 min at 95°C. Subsequently, 1 μl of RNase T1 (0.1 U/μl) was added, and incubation was continued for 5 min at 37°C. The OH ladder was generated by incubation of 0.2 pmol of labelled RNA for 5 min in alkaline hydrolysis buffer (Ambion) at 95°C. Samples were denatured for 3 min at 95°C prior to separation on 6% polyacrylamide/7 M urea sequencing gels in 1 × TBE. Gels were dried and analysed using PhosphorImager FLA‐7000 and AIDA software.