Multiple factors dictate target selection by Hfq‐binding small RNAs

We began by searching for mRNAs that potentially base pair with Spot 42. Using TargetRNA with a standard parameter set, we generated a list of the 10 top‐scoring mRNAs containing putative Spot 42 targeting sites within 45 nucleotides upstream and 25 nucleotides downstream of annotated start codons (Table I). This search yielded one known target, galK, in line with the original identification of this target based on its extensive complementarity to Spot 42 (Møller et al, 2002). To assess whether any of the other nine genes are regulated by Spot 42, we fused the annotated 5′ end (or at least 200 nucleotides upstream of the start codon for genes encoded in operons) through the ∼14th codon of each gene to a lacZ reporter. Overexpression of Spot 42 led to repression of two of the 10 reporter fusions (galK, puuE) beyond what was observed for an empty plasmid (>1.2‐fold) (Table I). Negligible repression of the other eight reporters by Spot 42 suggests that these genes are not targets, although the possibility exists that generation of the lacZ fusions compromised regulation. These results indicate that TargetRNA can identify genes regulated by Spot 42, albeit with low accuracy.

In our previous characterization of Spot 42, we found that genes repressed following Spot 42 overexpression were predicted to base pair with three regions (I–III) of Spot 42 (Figure 1A) (Beisel and Storz, 2011a). These same regions were predicted to base pair with the galK (regions II and III) and puuE (region III) mRNAs, where mutational analysis of the puuE fusion confirmed that region III is critical for regulation (Figure 1B). Secondary structure prediction and in vitro structural probing suggested that these three regions of Spot 42 are unstructured (Møller et al, 2002). The unstructured regions of sRNAs generally may be responsible for target regulation, as a recent bioinformatics analysis showed that the conserved, unstructured regions of Hfq‐binding sRNAs tend to contribute to base‐pairing with target mRNAs (Peer and Margalit, 2011). We hypothesized that utilizing these three unstructured regions of Spot 42 rather than the full‐length sRNA would improve the accuracy of target prediction.

We repeated the target search using the unstructured regions of Spot 42 as the sole input into TargetRNA. We then considered the five top‐scoring genes for each unstructured region (Table II). This list partially overlapped with the list using full‐length Spot 42, where galK and usg were within the top five for region II. Of the 15 genes in Table II, three (nanC, srlA, galK) previously were shown to be direct targets of Spot 42 and one (nanT) was shown to be regulated by Spot 42 with no evidence for direct base‐pairing (Møller et al, 2002; Beisel and Storz, 2011a).

We generated lacZ translational fusions with the 15 top‐scoring genes and again performed β‐galactosidase assays to assess repression by Spot 42. Ten of the 15 gene fusions listed in Table II were repressed following Spot 42 overexpression (compared to 2/10 fusions generated based on predictions with full‐length Spot 42). The fusions showed varying basal levels of expression, which may reflect differences in mRNA levels and/or translation. We conducted mutational analysis on three of the regulated fusions (ascF, nanT, fucP) to determine whether the predicted base‐pairing interactions are responsible for the observed regulation (Figure 1B–E). Mutations in the region implicated in base‐pairing disrupted repression while compensatory mutations restored regulation, confirming that Spot 42 base pairs with these fusions through the predicted interactions. We additionally evaluated whether endogenous expression of Spot 42 can alter mRNA levels of these new target genes. Quantitative real‐time PCR analysis was performed on WT and Δspf cells grown in M9 minimal media supplemented with glucose to induce Spot 42 expression. Among the genes tested, two of the five regulated as lacZ fusions (glpF, paaK) were significantly upregulated in the Δspf strain. The other three genes may be regulated at the level of translation or are not measurably regulated by Spot 42 under the conditions tested. In contrast, all three of the genes not regulated as lacZ fusions (usg, moeA, entB) were not upregulated in the Δspf strain (Supplementary Figure S1). Together, these results demonstrate that focussing on the unstructured regions of sRNAs can improve the prediction of direct targets.

Hfq‐binding sRNAs display large differences in the strength of regulation of target mRNAs, even for targets regulated by the same sRNA. One explanation for the variation in regulatory strength is the extent of base‐pairing, where more extensive base‐pairing is thought to lead to increased regulation (Mitarai et al, 2007, 2009). We tested how increasing the extent of base‐pairing affects regulation of three weakly regulated targets: gltA (region I), srlA (region II), and fucP (region III). Specifically, we inserted up to six nucleotides in each lacZ fusion either upstream or downstream of the targeting site to extend the predicted base‐pairing (Figure 2A–C). The inserted nucleotides extended base‐pairing through either the remainder of the unstructured region or into the structured region of Spot 42.

We found that extending base‐pairing through the remainder of the unstructured region substantially improved regulation (gltA_L, srlA_R, fucP_R). In contrast, extending base‐pairing into the structured region did not improve regulation (gltA_R, fucP_L). For srlA, extending base‐pairing into the structured region of Spot 42 (srlA_L) improved regulation less than what was observed when base‐pairing was extended through the remainder of the unstructured region (srlA_R), although interpretation of this result is complicated by the necessity of having the start codon interrupt the extended pairing. For all constructs, the measured strength of regulation did not correlate with the predicted increase in free energy (Supplementary Figure S2), suggesting that regulation by Hfq‐binding sRNAs is kinetically driven rather than thermodynamically driven. Overall, these results indicate that (i) the unstructured regions of Spot 42 are critical for target regulation and (ii) extending base‐pairing through these regions and not the structured regions of Spot 42 can improve the strength of repression.

Generally, individual unstructured regions of Hfq‐binding sRNAs are involved in base‐pairing interactions. However, for sRNAs with multiple unstructured regions, more than one region could be involved in base‐pairing with individual mRNAs. To assess whether Spot 42 employs multiple unstructured regions to regulate individual targets, we employed TargetRNA and the folding algorithm NUPACK to identify genes containing more than one putative Spot 42 targeting site. We maintained two criteria: (i) TargetRNA predicts that two unstructured regions of Spot 42 each form at least six base pairs with the target mRNA and (ii) the folding algorithm NUPACK predicts the same base‐pairing interactions as TargetRNA. Using this approach, we identified four target mRNAs that have the potential to base pair with two unstructured regions of Spot 42: nanC (regions I and III), galK (regions II and III), sthA (regions I and III), and ascF (regions I and III) (Figure 3A). Mutational analysis of Spot 42 and the nanC, sthA, and ascF fusions in this work and previous work supported multi‐site pairing, as mutations in individual base‐pairing sites partially reduced repression while mutations in both sites (one site mutated in Spot 42 and the other site mutated in the target fusion) eliminated regulation (Figures 1C and 3A) (Beisel and Storz, 2011a). In most cases (e.g., nanC), one site predominantly contributed to regulation. We observed that deletion of hfq greatly compromised repression of the most strongly regulated target, nanC, suggesting that Hfq is required even when multiple sRNA targeting sites are present (Supplementary Figure S3).

To assess whether Spot 42 can base pair with these targets through two regions, we performed in vitro structural probing with RNase T1, lead, and RNase V1 on Spot 42 complexed with the nanC mRNA. The altered cleavage patterns in the presence of unlabelled nanC mRNA supported base‐pairing between regions I and III of Spot 42 and the nanC mRNA (Figure 3B). Altered cleavage also was observed outside of regions I and III, which may be attributed to more extended base‐pairing and/or Spot 42 undergoing conformational changes upon pairing with the nanC mRNA. The cleavage pattern of radiolabelled nanC mRNA incubated with unlabelled Spot 42 similarly supported Spot 42 base‐pairing with the two predicted targeting sites (Figure 3C). These results indicate that multiple unstructured regions of Spot 42 can base pair with multiple targeting sites in a particular mRNA.

Mutational analysis of the nanC, sthA, and ascF fusions demonstrated that the presence of an additional targeting site improved the strength of regulation. We thus asked whether regulation could be strengthened in single‐site targets by introducing additional targeting sites for Spot 42. To address this, we focussed on the srlA and fucP fusions that only base pair with regions II and III of Spot 42, respectively (Figure 1E) (Beisel and Storz, 2011a). For the srlA fusion, we inserted 11 nucleotides that are complementary to region III of Spot 42 (srlA+III) upstream of the original targeting site (Figure 4A). For the fucP fusion, we mutated 11 nucleotides to be complementary to region I of Spot 42 (fucP+I) downstream of the original targeting site (Figure 4B). For both fusions, introduction of the additional targeting site substantially improved regulation from 2.8‐ to 27‐fold for srlA and from 4.7‐ to 13‐fold for fucP, an effect that was compromised in an hfq‐deletion strain (srlA+III; Supplementary Figure S3). Spot 42 variants containing mutations in either base‐pairing region reduced but did not eliminate repression (Figure 4), suggesting that both regions contribute to target regulation. Mutations in both base‐pairing regions (one in Spot 42 and the other in the fusion) either eliminated (fucP‐III+I) or greatly reduced (srlA‐II+III) repression. The residual repression of srlA‐II+III by pSpot42‐III may be attributed to a persisting potential for base‐pairing even after mutations were introduced into regions II and III. These results demonstrate that base‐pairing through multiple unstructured regions of an sRNA can improve the strength of target regulation.

Spot 42 expression had negligible effects on four of the gene fusions (usg, moeA, lon, entB) listed in Table II despite strong basal expression and putative base‐pairing near the ribosome‐binding site of each fusion. We sought to determine why Spot 42 did not have an effect on these target fusions and whether regulation could be activated.

One potential barrier to regulation was insufficient Hfq binding to the fusion mRNA. We began with the usg fusion, which lacks a recognizable binding site for the distal side of Hfq (Supplementary Table S1). The usg mRNA also was not enriched following co‐immunoprecipitation of Escherichia coli mRNAs bound to Hfq (A Zhang, unpublished data) (Table I). To introduce Hfq binding, we inserted the 5′ end of srlA containing a putative binding site for the distal side of Hfq immediately upstream of the ribosome‐binding site in the usg fusion (Figure 5A). The resulting srlA–usg mRNA was modestly enriched following co‐immunoprecipitation of E. coli mRNAs bound to Hfq and showed increased binding to Hfq in vitro similar to that observed for srlA (Figure 5C; Supplementary Figure S4). Insertion of the 5′ end of srlA also imparted 3.7‐fold repression of the srlA–usg fusion by Spot 42 (Figure 5D) that was lost in an hfq‐deletion strain (Supplementary Figure S3). The predicted base‐pairing interactions were responsible for regulation of the srlA–usg fusion, as a mutation in the implicated region of Spot 42 disrupted repression while a compensatory mutation in the fusion restored regulation (Figure 5B and D). These results indicate that mRNAs repressed by Hfq‐binding sRNAs require a binding site for the distal side of Hfq.

Next, we assessed why Spot 42 had a negligible effect on the moeA fusion. Similar to usg, the moeA fusion lacks a recognizable binding site for the distal side of Hfq (Supplementary Table S1) and the moeA mRNA was not enriched following co‐immunoprecipitation of Hfq‐bound mRNAs (Table II). However, unlike the usg fusion, introduction of the 5′ end of srlA immediately upstream of the predicted Spot 42 targeting site (Figure 6A) did not impart regulation by Spot 42 (Figure 6D). Secondary structure predictions revealed that two different sequences within the 5′ untranslated region of moeA may mask the putative targeting site (Figure 6B). Masking the sRNA targeting site previously was shown to reduce and even eliminate regulation of the sodB mRNA by the sRNA RyhB (Hao et al, 2011).

We hypothesized that the moeA fusion fails to be regulated by Spot 42 for two reasons: lack of an Hfq‐binding site and occlusion of the targeting site. To assess whether targeting site occlusion prevents regulation of the moeA fusion containing an Hfq‐binding site (srlA–moeA), we introduced two modifications predicted to free the targeting site: mutation of two nucleotides downstream of the targeting site (srlA–moeA1) or placement of lacZ immediately downstream of the start codon (srlA–moeA2) (Figure 6A and B). In conjunction with the Hfq‐binding site, the two modifications imparted repression by Spot 42 either individually or when combined (srlA–moeA1,2) that was lost in an hfq‐deletion strain (srlA–moeA1,2; Supplementary Figure S3). Furthermore, mutations in the unstructured region of Spot 42 implicated in base‐pairing eliminated repression, supporting the predicted base‐pairing interactions (Figure 6C and D). Relieving occlusion of the targeting site in the moeA fusion lacking an Hfq‐binding site (moeA2) did not impart regulation (Supplementary Figure S5A and B). Regulation of the lon fusion by Spot 42 appears to be hindered by a similar structural barrier, as introduction of the 5′ end of srlA did not impart regulation by Spot 42 and NUPACK predicted extensive base‐pairing between the 5′ end of srlA and the putative Spot 42 targeting site (Supplementary Figure S5C–E). We thus conclude that occlusion of the targeting site can prevent sRNA‐based repression.

Finally, we investigated why Spot 42 had a negligible effect on the entB fusion. Unlike usg and moeA, the lack of an Hfq‐binding site does not appear to be the culprit: the entB fusion contains two putative binding sites for the distal side of Hfq (Supplementary Table S1) and the entB mRNA was strongly enriched following co‐immunoprecipitation of Hfq‐bound mRNAs (Table I). In addition, the lack of regulation likely is not due to occlusion of the Spot 42 targeting site, as NUPACK predicts that the Spot 42 targeting site is not as structured as the srlA–moeA and the srlA–lon fusions (Figure 6B; Supplementary Figure S5D and G). Thus, we predicted that the entB fusion lacks an additional factor important for regulation by Spot 42.

We first focussed on the putative upstream Hfq‐binding site (Figure 6E). If this is the principal site of Hfq binding, then Hfq stimulation of Spot 42 pairing could be impeded by sequences upstream of this site or by the large stretch of 31 nucleotides separating this putative Hfq‐binding site and the Spot 42 targeting site. We tested these possibilities by placing the transcriptional start site 18 nucleotides upstream of this putative Hfq‐binding site or shortening the distance between this site and the Spot 42 targeting site to 12 nucleotides or 4 nucleotides (Supplementary Figure S5F and G). Overexpression of Spot 42 had a negligible effect on the expression of all three constructs (Supplementary Figure S5H), suggesting that the putative upstream Hfq‐binding site is not involved in target regulation. What prevents this site from contributing to regulation by Spot 42 remains unclear, although other factors important for Hfq binding (e.g., the presence of an adjacent hairpin) and function may remain to be elucidated.

We next focussed on the putative downstream Hfq‐binding site (Figure 6E), which directly overlaps with the Spot 42 targeting site. If Hfq must bind the mRNA for sRNA‐based regulation to occur, then Hfq binding could be preventing base‐pairing between Spot 42 and the entB mRNA. We tested this hypothesis by introducing either a separate Hfq‐binding site or a separate Spot 42 targeting site. To introduce a separate Hfq‐binding site, we introduced the 5′ of srlA immediately upstream of the predicted Spot 42 targeting site, which replaced the 5′ end of entB and the putative upstream Hfq‐binding site (srlA–entB; Figure 6E). To introduce a separate Spot 42 targeting site, we mutated the first 11 nucleotides downstream of the start codon to be complementary to region I of Spot 42 (entB+I; Figure 6E). The resulting fusions showed 4.2‐ and 1.7‐fold repression by Spot 42, respectively, which was disrupted when the implicated region of pairing in Spot 42 was mutated (Figure 6F and G). In addition, regulation of the srlA–entB fusion by Spot 42 was disrupted when hfq was deleted (Supplementary Figure S3). These observations suggest that, while Hfq is required for Spot 42 to associate with the entB mRNA, the Hfq‐binding and sRNA targeting sites cannot be overlapping.

In total, our results suggest that mRNAs containing a putative sRNA targeting site require additional factors to undergo sRNA‐based repression, including an unstructured sRNA targeting site and a non‐overlapping Hfq‐binding site.

We searched for targets of Spot 42 using the sRNA target prediction algorithm TargetRNA (snowwhite.wellesley.edu/targetRNA/). See Supplementary data for a description of the parameter set employed and the identification of multiple targeting sites in individual mRNAs.

Secondary structure predictions were performed using the folding algorithms NUPACK (http://www.nupack.org) (Zadeh et al, 2011) and mfold (http://mfold.rna.albany.edu/?q=mfold/RNA-Folding-Form) (Zuker, 2003) using default parameters. The two structures of the srlA–moeA fusion reported in Figure 6B represent the minimal free energy structure and the most stable suboptimal structure predicted by NUPACK and mfold.

Oligonucleotides, plasmids, and strains used in this study are listed in Supplementary Table S2. Plasmids pBRplac, pSpot42, pSpot42‐I, pSpot42‐II, and pSpot42‐III were reported previously (Beisel and Storz, 2011a). Plasmids pSpot42‐II′ and pSpot42‐III′ were constructed using the Gibson assembly method (Gibson, 2011). See Supplementary data for a detailed description of the procedure.

All strains were derived from E. coli strain K‐12 substrain MG1655. MG1655 Δspf was generated by P1 transduction of Δspf::kan^R from NM525 Δspf::kan^R (Beisel and Storz, 2011a) and excision of the kan^R resistance cassette with plasmid pCP20 (Cherepanov and Wackernagel, 1995). The lacZ fusions were generated in PM1205 Δspf::kan^R as described in detail previously (Mandin and Gottesman, 2009). Briefly, each gene was amplified by PCR from the MG1655 genome with flanking ends complementary to the P_BAD promoter (5′ end) and the lacZ coding region (3′ end). The amplified DNA template was recombined in place of the cat‐sacB cassette using mini‐λ‐mediated recombination (Court et al, 2003). Desired recombinants were identified by selective growth on M9 sucrose plates. Mutant constructs were generated by amplifying each gene in two fragments by PCR from the associated PM1205 strain. The two fragments were assembled by PCR to generate the final DNA template for recombination into PM1205 Δspf::kan^R. Transductions were confirmed by PCR and all recombination events were confirmed by sequencing.

All strains were grown by shaking at 250 r.p.m. at 37°C unless noted otherwise. Strains were grown in Luria‐Bertani (LB) media (1% bacto‐tryptone, 0.5% yeast extract, and 1% NaCl) or M9 minimal media (1 × M9 salts, 10 μg/ml thiamine, 2 mM MgSO₄, and 0.1 mM CaCl₂) supplemented with 0.2% glucose. Cell density was determined by measuring A₆₀₀ using an Ultrospec 3000 UV/Vis spectrophotometer (Pharmacia Biotech).

β‐Galactosidase assays were performed as described previously (Beisel and Storz, 2011a). Three separate colonies were grown overnight in LB, back‐diluted to A₆₀₀=0.01 in the same media and grown to A₆₀₀=∼0.1. l‐arabinose and IPTG then were added to each culture to final concentrations of 0.2% and 1 mM, respectively, as indicated. Cells were assayed for β‐galactosidase activity (Miller, 1977) after an additional 1 h of growth when cultures attained A₆₀₀=0.4–0.6. The A₆₀₀ and A₄₂₀ of the cultures were measured using an Ultrospec 3000 UV/Vis spectrophotometer (Pharmacia Biotech).

Template DNA for T7 transcription was amplified from the genomic DNA of the corresponding PM1205 strain by PCR with primers containing the T7 promoter (GTTTTTTTTAATACGACTCACTATAGG). T7 transcription was conducted using the MegaShortscript T7 Transcription Kit (Ambion) according to the manufacturer's instructions. Transcribed RNAs were checked for complete synthesis. The 5′ ends of RNAs were ³²P‐radiolabelled as described previously, after which the RNAs were gel purified (Sittka et al, 2007).

In vitro structural probing was performed in 10 μl reactions similarly to previous work (Sharma et al, 2007). Radiolabelled RNA (∼0.2 pmol) was denatured at 95°C for 1 min and chilled on ice for 5 min, followed by the addition of 1 × structure buffer (Ambion), 0.1 μg/μl yeast RNA, and the indicated concentration of unlabelled RNA. Concentrations of unlabelled RNAs were selected based on gel shift assay results (Supplementary Figure S6). Following incubation at 37°C for 1 h, 2 μl of RNase T1 (0.01 U/μl; Ambion) or 2 μl RNase V1 (0.0001 U/μl; Ambion) was added to the corresponding reaction and incubated for 6 min. For the lead cleavage reactions, 2 μl of fresh lead(II) acetate (25 mM) was added and the samples were incubated for 1.5 min at 37°C. Reactions were stopped with the addition of 20 μl of Inactivation/Precipitation buffer (Ambion) and vortexing. Reactions were precipitated out of solution at −80°C for 15 min, spun down and washed with 70% EtOH, and finally RNA pellets suspended in 7 μl loading buffer II (95% formamide, 18 mM EDTA, 0.025% SDS, 0.2% xylene cyanol, 0.2% bromophenol blue; Ambion) and placed on ice.

To generate the RNase T1 ladder, radiolabelled RNA (∼0.4 pmol) was combined with 1 × Sequencing Buffer (Ambion), denatured at 95°C for 1 min, and incubated with RNase T1 (1 μl, 0.1 U/μl) at 37°C for 5 min. For the hydroxyl ladder, radiolabelled RNA (∼0.4 pmol) was combined with 1 × Alkaline Hydrolysis Buffer (Ambion) and incubated at 90°C for 5 min. Both ladder reactions were stopped with the addition of 12 μl of loading buffer II. All samples were denatured at 95°C for 3 min and 3 μl resolved on a 6% denaturing polyacrylamide/7 M urea gel in 1 × TBE. Gels were dried and exposed to BioMax XAR film (Kodak).

Hfq co‐immunoprecipitation was performed similarly to previous work (Zhang et al, 1998). Briefly, cultures were grown to mid‐log phase and incubated in 0.2% l‐arabinose for 1 h as described in the β‐galactosidase assays. Cultures then were pelleted, resuspended in lysis buffer (20 mM Tris–HCl/pH 8.0, 150 mM KCl, 1 mM MgCl₂, and 1 mM DTT, 0.2 U RNaseOUT (Ambion)), and lysed by vortexing with glass beads for 10 min. Cell lysates then were used to extract total RNA or immunoprecipitate Hfq. To immunoprecipitate Hfq, 200 μl cell lysate was combined with 24 mg of Protein A Sepharose CL‐4B beads (Amersham Biosciences) complexed with 20 μl of α‐Hfq serum, 200 μl of Net2 Buffer (50 mM Tris–HCl/pH 7.4, 150 mM NaCl, and 0.05% Triton X‐100), and 1 μl RNaseOUT. The mixture was incubated at 4°C for 2 h with rotation then washed 5 × with 1.5 ml Net2 Buffer. Following the washes, the beads were combined with 400 μl of Net2 Buffer, 50 μl of 3 M NaOAc, 5 μl of 10% SDS, and 600 μl of Phenol:Chloroform:Isoamyl Alcohol (Ambion) and RNA was ethanol precipitated. Total RNA (2 μg) or co‐immunoprecipitated RNA (0.2 μg) then was used for primer extension assay. Primer extension analysis was performed as described previously (Zhang et al, 1998). The products (5 μl) were resolved by denaturing PAGE. Gels were dried and either exposed to BioMax XAR film or to a phosphor screen and quantified using a phosphorimager.

See Supplementary data for detailed descriptions of plasmid construction, sRNA target predictions, enrichment scores calculated from the Hfq co‐immunoprecipitation data, quantitative real‐time PCR, gel shift assays for RNA hybridization and Hfq binding, primer extension analysis, and free energy calculations.

Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).