Acidic residues in the Hfq chaperone increase the selectivity of sRNA binding and annealing

Abstract

Hfq facilitates gene regulation by small non-coding RNAs (sRNAs), thereby affecting bacterial attributes such as biofilm formation and virulence. E. coli Hfq recognizes specific U-rich and AAN motifs in sRNAs and target mRNAs, after which an arginine patch on the rim promotes base pairing between their complementary sequences. In the cell, Hfq must discriminate between many similar RNAs. Here, we report that acidic amino acids lining the sRNA binding channel between the inner pore and rim of the Hfq hexamer contribute to the selectivity of Hfq’s chaperone activity. RNase footprinting, in vitro binding and stopped-flow fluorescence annealing assays showed that alanine substitution of D9, E18 or E37 strengthened RNA interactions with the rim of Hfq and increased annealing of non-specific or U-tailed RNA oligomers. Although the mutants were less able than WT Hfq to anneal sRNAs with wild type rpoS mRNA, the D9A mutation bypassed recruitment of Hfq to an (AAN)₄ motif in rpoS, both in vitro and in vivo. These results suggest that acidic residues normally modulate access of RNAs to the arginine patch. We propose that this selectivity limits indiscriminate target selection by E. coli Hfq, and enforces binding modes that favor genuine sRNA and mRNA pairs.

Graphical Abstract

INTRODUCTION

In bacteria, small RNAs (sRNAs) regulate gene expression during environmental stress conditions like cold shock, oxidative stress, osmotic shock and quorum sensing ^{1; 2; 3}. Transacting sRNAs activate or repress expression of certain genes by base pairing with the 5′-UTR of the target mRNAs ⁴, either masking or exposing the Shine-Dalgarno ribosome binding site, or recruiting RNase E for mRNA turnover ^{5; 6}. In Gram-negative bacteria, sRNAs typically act together with the RNA chaperone protein Hfq ^{6; 7}, which accelerates annealing of complementary RNA strands and stabilizes sRNA-mRNA duplexes ^{8; 9; 10; 11}. The observed regulation depends on the specificity of these complexes, as different sRNAs must compete for binding to Hfq and their proper mRNA target.

Bacterial Hfq is a member of the Sm/Lsm super-family that forms a homo-hexamer ring with two distinct single-stranded RNA binding surfaces ⁷ (Fig. 1a). The proximal face of Hfq binds to U-rich RNA sequences found in sRNAs while the distal face of Hfq binds to AAN motifs found in the 5′-UTR of mRNAs ^{12; 13; 14}. In addition, a patch of highly conserved arginine residues on the rim of Hfq binds U-rich regions of the mRNA and sRNA body ^{15; 16; 17} and is thought to be responsible for the RNA annealing activity of Hfq ¹⁸.

Fig. 1. Conserved acidic residues on the Hfq proximal surface.

(a) Conservation scores of 145 Hfq sequence clusters were rendered from blue (minimum) to magenta (maximum) on one subunit of the hexamer (chain A; 1HK9). Acidic residues (D9, E18, E37 and D40) are shown as sticks. (b) Schematic representation of the sRNA-rpoS-Hfq ternary complex, based on solution data ²⁰. (AAN)₄ (green) and U₆ (magenta) motifs bind to the distal and proximal faces of Hfq, respectively. The U₅ loop in rpoS (orange) interacts with the rim. The complementary regions of the sRNA and mRNA interact with charged residues at the rim, which are needed to initiate base pairing.

In our working model for the Hfq annealing cycle, sRNAs and mRNAs are recruited to the proximal and distal surfaces of Hfq through specific recognition of U and A-rich motifs. When Hfq forms a ternary complex with two complementary RNAs, interactions with the positively charged arginine patch accelerate base pairing between the anti-sense regions ¹⁸. In addition, simultaneous interactions with different RNA-binding surfaces of Hfq distort sRNA and mRNA structures, orienting them for anti-sense annealing ^{15; 19; 20}.

The arginine patch in Hfq is surrounded by four conserved acidic residues D9, E18, E37 and D40 in E. coli Hfq (Fig. 1a). Invariant D9 and D40 lie on either side of a channel that sRNAs use as a passage between the inner proximal face RNA binding site and the arginine patch ^{15; 17}, while the moderately conserved E18 and E37 flank the arginine patch on the distal rim of Hfq (Fig. 1a). Although negatively charged residues are not expected to interact with RNA, replacement of the almost universally conserved D9 with alanine impaired the ability of Hfq to support regulation of cirA by OmrA sRNA ²¹, chiP by ChiX sRNA and sdhCDAB by RyhB sRNA in E. coli ²². These results indicated that D9 is important for sRNA binding to Hfq, or for regulation of these target genes by sRNAs.

Here we show that removing D9 or other acidic residues makes the Hfq rim more accessible to unstructured RNAs, improving the ability of Hfq to act on RNAs lacking a specific Hfq binding site. Further in vitro and in vivo assays confirmed that mutating acidic residues overrides the requirement for AAN motifs which normally recruit Hfq to rpoS mRNA for riboregulation. The D9A mutation was even found to rescue Hfq alleles defective for RNA binding to the rim and distal face. The results presented here suggest that the acidic residues normally act to limit RNA binding to arginine patch, thereby increasing the selectivity of Hfq’s chaperone function.

RESULTS

Conserved acidic residues regulate the accessibility of the rim

Each of the four acidic side chains on the proximal face of Hfq was replaced with alanine (D9A, E18A, E37A and D40A) and the variants tested for RNA binding and annealing activity. The D40A mutant protein could not form a stable hexamer as previously reported ²² and was therefore not analyzed further. Equilibrium binding constants for oligoribonucleotides and sRNAs were measured using fluorescence anisotropy and electrophoretic gel mobility shift assays (EMSA), respectively (Table 1).

Table 1.

Equilibrium binding constant for RNA and Hfq

Hfq variants	KD1 (μM)		DsrAΔU6
	D16-FAM	DsrA sRNA
WT	0.067 ± 0.005	0.22 ± 0.015	1.31 ± 0.1
D9A	0.047 ± 0.002	0.46 ± 0.018	0.67 ± 0.022
E18A	0.024 ± 0.004	0.38 ± 0.02	ND
E37A	ND	0.51 ± 0.021	ND

Overall dissociation constants per Hfq monomer were measured by fluorescence anisotropy (D16-FAM) or native gel mobility shift (DsrA and DsrAΔU6) at 30 °C in TNK buffer. ND, not done. Values are the average and standard deviation from the mean of three independent experiments. The K_d for D16-FAM and WT Hfq is about half that reported for earlier preparations of the protein ¹⁸.

We compared RNAs that primarily bind the rim of Hfq, or that interact with both the inner pore and outer rim of the proximal surface. An unstructured 16 nt RNA derived from the body of DsrA ²³, D16, has six internal uridines and is thought to mainly interact with the proximal rim ¹⁸. Wild type (WT) E. coli Hfq binds FAM-labeled D16 (D16-FAM) with K_d = 67 nM (Hfq monomer) in TNK buffer at 30°C. The D9A and E18A variants bound D16-FAM more tightly than the WT protein, with dissociation constants of 47 nM and 24 nM, respectively.

Natural sRNAs end in 3′ uridines that tightly bind the inner proximal surface of Hfq, in addition to internal uridines and stem-loops that interact with the rim ^{24; 25}. The D9A and E18A mutations weakened overall binding of DsrA sRNA about twofold compared to WT Hfq (Table 1; Fig. S1). When the 3′ terminal uridines were removed (DsrAΔU₆), binding to WT Hfq became almost six-fold weaker (Table I). Thus, the WT Hfq-DsrA complex is stabilized more by interactions between the U-tail and the proximal face than the rim. By contrast, removing the U-tail had a much smaller effect on the overall affinity of DsrA for Hfq:D9A, raising the K_d less than twofold (Table I). As a result, the tail-less sRNA actually bound Hfq:D9A more tightly than it bound WT Hfq. We conclude that the D9A mutation increases RNA binding to the rim, perhaps owing to the increased net positive charge, while weakening the ability of structured sRNAs to simultaneously contact the rim and the inner pore of Hfq.

To directly test whether removing an acidic residue strengthens RNA interactions with the rim of Hfq, we monitored the ability of different RNAs to quench the fluorescence of a tryptophan introduced at residue 38 near the arginine patch (Fig. 2a; Fig S2a). This tryptophan substitution reports interactions with the rim of Hfq, as short RNAs that cannot bind the rim do not reduce the emission intensity ¹⁸. This mutation does not impair sRNA-mRNA annealing in vitro (Fig. S2b). All of the RNAs tested quenched the tryptophan fluorescence signal of Hfq:S38W by 15–20%. These RNAs were oligo C, an unstructured 16-mer that binds the Hfq rim non-specifically, oligo C-U₆ (CU), which interacts with the proximal face and rim, DsrA sRNA and ArcZ56 sRNA. The D9A and E18A mutations resulted in stronger quenching of the tryptophan signal by all of the RNAs tested (Fig. 2b), consistent with stronger binding of RNAs to the rim of Hfq when a negatively charged side chain is removed.

Fig. 2. Acidic side chains reduce RNA binding to the rim of Hfq.

(a) A unique tryptophan at position 38 monitors RNA binding to the rim arginine patch. ^{18; 42} (b) Percent quenching of tryptophan emission in the presence of 100 nM RNA in TNK buffer at 30 °C, for Hfq variants shown in the key. These variants had similar (±20%) intrinsic fluorescence in the absence of RNA (Fig. S2c). Values were corrected for the inner filter effect (Methods and Fig. S2). Error bars represent the standard deviation from the mean of three independent trials.

Acidic residues contribute to sRNA recognition

To ask whether the acidic residues influence how Hfq interacts with specific sRNA sequences, partial digestion with RNase I_f was used to probe the structure of ³²P-labeled DsrA with and without Hfq. WT Hfq protects single-stranded nt 25–28 and U’s 32–38 in the body of DsrA while enhancing cleavage of nt 51–53 in loop 2 (black labels in Fig. 3), as reported previously ^{8; 26}. Hfq:D9A protected DsrA much more strongly than WT Hfq (red labels in Fig. 3). The E18A and E37A Hfq mutants protected the same regions of DsrA, but to a degree more like that of WT Hfq (blue and green traces, Fig. 3b inset). Since the body of DsrA is thought to interact with the rim of Hfq ¹⁵, the footprinting results also indicate that the D9A mutation especially increases RNA binding to the rim of Hfq, even though the sRNA-Hfq complex is less stable overall owing to weaker interactions with the U₆ tail (Table 1). The loss of enhancements in stem-loop I and loops II and III further suggest that Hfq:D9A cannot interact with DsrA as precisely as the WT protein (Fig. 3c).

Fig. 3. Effect of acidic residues on DsrA recognition.

(a) Ribonuclease I_f footprinting of 5′-³²P-labeled DsrA at 30 °C with no Hfq or with 10 μM Hfq monomer as shown. Lanes T1, L and C represent control reactions in urea with RNase T1 (T1), alkaline hydrolysis (L) and no treatment (C). The D40A variant forms unstable hexamers and these data were not analyzed further. (b) Relative RNase I_f digestion of nucleotides in DsrA. Gray, (−) Hfq; black, WT Hfq; red, D9A; (inset): blue, E18A; dark cyan, E37A. Lines and error bars trace the mean and S.D. of four trials (Methods). (c) Summary of nuclease digestion pattern on the secondary structure of DsrA. Pointed (→) and flat (⊥) arrows indicate enhancement or protection, respectively, while the size indicates the degree of change in the digestion pattern (Black, WT; red, Hfq:D9A).

Acidic residues increase annealing of non-specific and U-rich RNAs

Since the arginine patch on the rim is essential for Hfq’s chaperone activity ¹⁸, we asked whether the loss of adjacent acidic residues could affect the annealing of complementary RNA strands. To compare the chaperone activity of each Hfq variant, the annealing kinetics of a molecular beacon and 16 nt target RNAs were measured by stopped-flow fluorescence (Fig. 4a). Wild type Hfq speeds up base pairing between a molecular beacon and unstructured RNAs, such as oligo C, which bind the Hfq rim non-specifically (Fig. S3) ²⁷. This acceleration reached a maximum at 300 nM Hfq, which equaled one Hfq hexamer per RNA pair (Fig. 4b, solid blue). Hfq acts most efficiently, however, when one of the RNA strands contains an A₁₈ tail that binds the distal face of Hfq, increasing the annealing rate 20 to 60 fold (Fig. 4b, solid black) ²⁸.

Fig. 4. Acidic residues help discriminate between A-rich and U-rich RNAs.

(a) Schematic of molecular beacon annealing assay. (b) Observed annealing rate constants (average of five trials) were plotted against Hfq concentration. Annealing rates were measured by stopped-flow fluorescence using 50 nM molecular beacon, 100 nM target RNA and 0–1500 nM Hfq in TNK at 30°C. WT Hfq, filled circles; Hfq:D9A, open circles. (c) Observed annealing rate constants for Hfq variants (300 nM). Black, A₁₈ tail; red, U₆ tail; blue, target sequence only. Results of 5 to 6 trials were averaged; errors bars represent the standard deviation of the mean.

When the A-tailed target RNA (oligo CA) was recruited to the distal face of Hfq, the D9A mutation reduced the annealing rate less than twofold (Fig. 4b; compare black open and closed circles). A subtle effect on the A-tailed target was expected because the D9 residue is located on the proximal face of Hfq. Far more surprising was that the D9A mutation greatly improved the annealing rate when a U-tailed target RNA (oligo CU) was recruited to the proximal face of Hfq (Fig. 4b; compare red open and closed circles). Similar increases in annealing rate were observed for oligo C, which lacks a specific Hfq binding site (Fig. 4b; compare blue open and closed circles). Parallel changes in the annealing kinetics were observed for the E18A and E37A variants (Fig. 4c), in that each of these acidic residue Hfq mutations reduced the preference for annealing A-tailed RNA substrates over U-tailed (CU) and untailed (C) substrates (Fig 4c; compare black bar to red and blue bars).

Loss of acidic residues makes sRNA-mRNA annealing less efficient

In contrast to unstructured oligonucleotides, natural sRNAs and mRNAs must make specific contacts with the proximal pore and the rim. We next tested whether the acidic residues affect the ability of Hfq to facilitate annealing of DsrA and ArcZ sRNAs to rpoS mRNA in vitro (Fig. 5a). As expected, DsrA and rpoS mRNA base paired with each other about 60 times faster with WT Hfq than without Hfq. All of the Hfq variants could facilitate sRNA annealing to rpoS mRNA, but the reaction became less efficient when an acidic side chain was mutated. For DsrA, the amplitude of the Hfq-dependent burst phase was lowered from 50% for WT Hfq to 33%, 43% and 34% for Hfq:D9A, E18A and E37A, respectively (Fig. 5b).

Fig. 5. Acidic side chains promote sRNA and mRNA annealing.

(a) Annealing kinetics of ³²P-labeled rpoS301 and 200 nM DsrA or ArcZ56 sRNA with or without 0.6 μM Hfq was measured by native 6% PAGE. (b) Formation of rpoS-sRNA-Hfq complexes versus time for DsrA (top) and ArcZ56 (bottom). Grey, no Hfq; black, WT Hfq; red, D9A; blue, E18A; dark cyan, E37A.

A 56 nt form of ArcZ sRNA (ArcZ56) requires Hfq to base pair with rpoS mRNA ^{29; 30}, unlike DsrA which slowly base pairs with rpoS in the absence of Hfq (grey lines, Fig. 5a, b). Mutations in the acidic residues did not change the initial burst, but they did diminish the amount of ArcZ56•Hfq•rpoS product formed after 4 minutes, from 80% in the presence of WT Hfq to only 64%, 47% and 55% with D9A, E18A and E37A Hfq, respectively. One explanation for these results is that tighter interactions with the rim make the ArcZ56-rpoS duplex less able to cycle off Hfq after base pairing has occurred, leading to refolding of rpoS and displacement of ArcZ56. Alternatively, the acidic residue mutants may not fold rpoS mRNA correctly for sRNA annealing, reducing the yield of sRNA-mRNA-Hfq complex for both DsrA and ArcZ56.

D9A mutation bypasses the requirement for rpoS AAN motifs

Up-regulation of rpoS by sRNAs depends on recruitment of Hfq to (AAN)₄, A₆ and U₅ motifs within the rpoS 5′ UTR (Fig. 6a), which are needed for Hfq to promote sRNA-mRNA base pairing in vivo and in vitro ^{20; 29; 31}. Mutagenesis of either the (AAN)₄ or U₅ motifs alone, or all three motifs together (ΔΔΔ), lowered the burst of DsrA annealing by WT Hfq from 67% with rpoS301 to 14% with ΔΔΔrpoS301 (Fig. 6b, white and grey bars). In contrast, mutagenesis of canonical Hfq binding motifs in rpoS mRNA had no effect on DsrA annealing by Hfq:D9A (Fig. 6b, red bars).

Fig. 6. D9A mutation bypasses an AAN motif in rpoS mRNA.

(a) Schematic representation of rpoS secondary structure and position of Hfq binding sites. (b) Fraction ³²P-labeled rpoS 301RNA bound to DsrA with or without Hfq after 30 s. Mutations in rpoS deleted the upstream AAN motif (Δ(AAN)4), the downstream U₅ loop (ΔU5), or both of these plus an A₆ sequence (ΔΔΔ). (c) E. coli cells (Table S1) containing a chromosomal copy of WT or mutant rpoS::lacZ fusion, with hfq⁺ or hfqD9A alleles in the chromosome for each, were grown overnight at 37 °C on MacConkey agar supplemented with 0.0001% arabinose. Strains used, listed clockwise from WT in Figure, were SG30214, SG30236 SG30233, SG30242, SG30277, SG30279, DJS3056, and DJS3059, all described in Table S1. Red colonies indicate up-regulation of lacZ expression by Hfq and chromosomally derived sRNAs (DsrA, ArcZ and RprA). See Fig. S4 and S5 for further data.

We next asked whether D9 affects the requirements for rpoS mRNA regulation by sRNAs in vivo. Over-expression of DsrA and ArcZ sRNAs from an IPTG-inducible promoter increased translation of an rpoS::lacZ fusion in the E. coli chromosome to a similar extent in strains containing wild type or mutant hfq alleles in the bacterial chromosome (Fig. S4). In fact, we observed slightly greater up-regulation of rpoS by ArcZ in strains expressing Hfq:D9A, E18A or E37A compared to WT Hfq. These results were consistent with previous studies, which found that D9A was more important for negative regulation than positive regulation of rpoS, and had little effect on the accumulation of ArcZ and DsrA sRNAs in E. coli ²². Because the rpoS-lacZ mRNA was strongly induced by arabinose in these experiments, however, deficiencies in Hfq binding to the mRNA may have been masked ²².

To test whether the D9A allele also bypassed the need for Hfq to bind the rpoS AAN motif in vivo, we assayed translation of a WT rpoS-lacZ fusion on MacConkey agar plates. In the absence of a plasmid (akin to the vector control in Fig. S4b), rpoS-lacZ expression depends on endogenous levels of DsrA, ArcZ and RprA ³⁰. Deletion of the (AAN)₄ motif alone, (AAN)₄ and A₆ (ΔΔ), or all three motifs (ΔΔΔ) lowered the expression of the rpoS fusion in the presence of WT Hfq (WT sectors in Fig. 6c). However, the deleterious effects of these rpoS mutations were suppressed when Hfq carried the D9A allele (D9A sectors in Fig. 6c), in agreement with the in vitro annealing results (Fig. 6b). The ability of Hfq:D9A to suppress deletion of the rpoS (AAN)₄ motif was also seen when mutants in the rim or distal site were combined with a deletion of the (AAN)₄ or deletion of all three motifs (Fig. S5). Thus, the D9A mutation not only compensates for weaker rim binding, in agreement with our tryptophan quenching data, but surprisingly bypasses the need to recruit Hfq to the rpoS mRNA via the upstream (AAN)₄ motif.

DISCUSSION

E. coli encodes about 100 different sRNAs that regulate up to 700 genes ^{32; 33}. Although not all sRNAs are expressed simultaneously, the Hfq chaperone must still discriminate between sRNAs and other transcripts, while promoting base pairing with a diversity of mRNA targets. This is largely accomplished by specific recognition of U-rich and A-rich sequence motifs present in natural substrates of Hfq. It has been recently recognized, however, that natural RNAs interact with multiple surfaces of Hfq in a manner that likely depends on the RNA secondary structure. These multi-lateral interactions trigger structural rearrangements needed for efficient sRNA-mRNA recognition. For example, single-stranded 3′ uridines anchor sRNAs to the inner pore of the proximal face, while the body of many sRNAs binds the rim of Hfq ^{15; 18}. Hfq restructures the rpoS mRNA leader even more dramatically by recognizing an upstream (AAN)₄ motif and the U₅ motif downstream of the sRNA binding site ²⁰. These interactions fold the rpoS mRNA 5′-UTR into a compact structure ²⁰, partially melting the inhibitory stem containing the sRNA target sequence ³⁴.

Our results suggest that acidic residues on the proximal face and rim act as “gatekeepers” that increase the selectivity of RNA interactions with arginine patch. Tryptophan quenching data showed that removal of an acidic side chain allows oligonucleotides and sRNAs to bind the rim of Hfq more tightly (Fig. 2). Similarly, Hfq:D9A protected internal residues of DsrA from ribonuclease digestion more strongly than WT Hfq protected them (Fig. 3). Mutations in acidic residues could improve RNA binding to the rim by simply making the rim more positively charged, or by changing the structure of Hfq to make the rim more accessible.

Tighter RNA binding to the rim appears to shift sRNA recognition from the inner proximal pore, which is highly selective for uridine, to the rim, which is less sequence-specific. Although the uridines at the 3′ end of DsrA are important for the stability of the complex with the WT protein, removing the U-tail had little effect on the stability of the Hfq:D9A-DsrA complex (Table 1). In addition, Hfq:D9A perturbed interactions with DsrA stem-loops II and III (Fig. 3). The rim of Hfq has been long reported to bind double-stranded RNA ^{35; 36}, and is likely important for folding structured RNAs around Hfq in a manner that promotes sRNA-mRNA annealing and gene regulation ^{17; 20}.

This shift toward stronger RNA interactions at the rim may explain why acidic residue mutations allow Hfq to bypass AAN motifs, which are normally required for Hfq to act on rpoS mRNA ^{29; 31} and other targets. In our annealing assays with unstructured oligonucleotides, Hfq:D9A was as active on non-specific and U-tailed RNAs as on A-tailed RNAs, while WT Hfq is far more active on RNAs containing an A-tail (Fig. 4). Similarly, increased non-specific binding of RNAs at the rim enables Hfq:D9A to form an alternative complex with rpoS which is modestly less active than the WT Hfq·rpoS complex but does not depend on interactions with the (AAN)₄ and A₆ motifs both in vitro and in vivo. Additionally, the D9A mutation overcame the deleterious effects of mutations on the rim (R16A) and distal face (Y25D and K31A) on rpoS up-regulation in the absence of the AAN motifs in vivo (Fig. S5), further indicating that it bypasses the need for specific recruitment of Hfq to AAN motifs. Together our results suggest that loss of a negatively charged side chain allows Hfq to act on RNAs that lack specific U or A-rich binding sites.

Although replacement of an acidic residue makes Hfq a less discriminating chaperone, this is apparently detrimental to the function of certain sRNA-mRNA pairs. The Hfq D9A allele was first isolated for its failure to support negative regulation of sdhC by RyhB sRNA and negative regulation of cirA by OmrB ²¹, and in subsequent studies D9A was more deleterious for negative than positive regulation ²². One possibility is that the mutant Hfq proteins fail to properly restructure both positive and negative mRNA targets for optimal base pairing with incoming sRNAs. These mutations could also weaken Hfq binding to mRNA sequence motifs needed for translational repression. Alternatively, the shift from recognition of the sRNA 3′ end to the sRNA body may allow Hfq to prematurely dissociate from sRNA-mRNA complexes, also weakening repression. Finally, the higher expression of rpoS::lacZ in the absence of sRNA overexpression suggests that these Hfq mutations may upset the competition between native sRNAs, elevating rpoS translation by basal levels of ArcZ, DsrA or RprA sRNAs. Regardless of the mechanism, we conclude that conserved negative charges surrounding the rim RNA binding site unexpectedly contribute to selective recognition of sRNAs and their targets by Hfq.

MATERIALS AND METHODS

Hfq purification

Wild type Hfq was overexpressed and purified from E. coli BL21 (DE3) cells using a Hi-Trap Co²⁺ column (GE Healthcare) and UNO S6 ion-exchange column (Bio-Rad) as previously described ³⁷. Alignment of 145 non-redundant (<90% identity) Hfq sequences showed that D9 and D40 are almost invariant while E18 and E37 are moderately conserved. Hfq mutants (D9A, E18A, E37A, D40A, WT-S38W, D9A-S38W, and E18A-S38W) were made by QuikChange (Stratagene) and purified using the same protocol.

Fluorescence binding assays

Binding constants for Hfq and D16-FAM RNA were measured by fluorescence anisotropy as described before ²³. RNA binding to the rim was measured by changes in the fluorescence emission of a single tryptophan Hfq-S38W as previously described ¹⁸. Contributions from the inner filter effect were determined by repeating the RNA titrations in buffer containing 1.25 M NaCl, which inhibits RNA binding. The percent quenching of the tryptophan fluorescence emission at 340 nm was calculated as 100% × {(I₀ − I)/I₀ − (I₀′ − I′)/I₀′}, in which I₀ is the initial tryptophan emission in the absence of RNA, I is the emission in the presence of RNA, and I₀′ and I′ are the initial tryptophan emission and the emission in the presence of RNA in 1.25 M NaCl respectively.

RNase I_f footprinting

5′-³²P-DsrA (0.1 μM) and DsrA–Hfq complexes (8 μl) containing 1 μM Hfq monomer were prepared as described ⁸ and incubated 30 min at 30°C. Each sample was treated with 2 μl 1.25 U/μl of RNase I_f for 1 min at 37°C. Reactions were stopped by the addition of 10 μl buffered phenol. After phenol-chloroform extraction and ethanol precipitation, RNA pellets were redissolved in 6 μl formamide loading dye (90% (v/v) formamide, TBE, 0.1% (w/v) bromophenol blue, 0.1% (w/v) xylene cyanol) and subsequently loaded on an 8% polyacrylamide sequencing gel. Sequence ladders were obtained by nuclease digestion under denaturing conditions as described ³⁸. Band intensities were integrated with SAFA ³⁹ and normalized to bands with constant intensity (nt 5–7) in different experiments.

Molecular beacon-target annealing kinetics

Reverse-phase HPLC purified RNA molecular beacon, 5′ FAM-GGUCCCCCACUCGA CUCACCACCGGACC-DABCYL (Trilink Biotechnologies), was used without further purification. Target RNAs (Dharmacon) were purified as described previously ²⁸. The target RNA sequences were oligo C, 5′GUGGUCAGUCGAGUGG3′; CU, oligo C plus U₆; CA, oligo C plus A₁₈. The annealing kinetics between the molecular beacon (50 nM) and target RNA (100 nM) in TNK buffer (10 mM Tris-HCl pH 7.5, 50 mM NaCl, 50 mM KCl) at 30°C was measured by an Applied Photophysics SX 18MV stopped-flow spectrometer as described previously ^{28; 40}. Reactions were performed with 0–1.5 μM Hfq monomer as shown in the figures.

Native gel mobility shift assays for rpoS RNA and sRNA

DsrA, ArcZ56 and rpoS301 RNA were transcribed in vitro ⁸. Binding affinities of Hfq variants with ³²P labeled DsrA at 30°C were measured by a native gel shift assay as previously described ⁸. The disappearance of labeled DsrA was plotted against Hfq monomer concentrations and subsequently fit to a Hill equation (θ = A [Hfq]ⁿ/([Hfq]ⁿ + K_dⁿ), in which θ is the fraction of DsrA bound to Hfq, A is the amplitude, K_d is the binding constant and n is the Hill co-efficient (Fig. S1). The association kinetics of ³²P-labeled rpoS mRNA 5′-UTR (rpoS301) with 200 nM sRNA in the presence of WT or mutant Hfq protein was measured by native gel mobility shift as described before ^{8; 29}. The fraction of counts in each lane corresponding to rpoS•sRNA duplex or rpoS•Hfq•sRNA ternary complex versus time was fit to a biphasic rate equation.

β-Galactosidase activity assay

A derivative of wild-type E.coli MG1655 in which the 5′-UTR of rpoS was fused to lacZ, downstream of a pBAD promoter, ²⁹ was further modified by P1 phage transduction ⁴¹ to introduce chromosomal Hfq mutations. The mutants were constructed as previously described ²² and confirmed by sequencing (see Supplementary Table S1). These strains were transformed with pLac, pDsrA and pArcZ plasmids ³⁰. Overnight cultures were diluted into fresh LB containing arabinose (0.2%), ampicillin (100 μg/mL), and IPTG (100 μM) for 6 hours. OD₆₀₀ and β-galactosidase activity were measured as described previously ²⁹. The specific activities were calculated by V_max/OD₆₀₀ and were the average of three independent measurements. For plate assays, E. coli strains listed in Table S1 were grown overnight at 37°C on MacConkey agar containing 0.0001% arabinose, if not stated otherwise.

Highlights.

• Hfq chaperones bacterial small non-coding RNAs during stress and pathogenesis

• E. coli Hfq normally selects cellular RNAs containing UUU or AAN sequence motifs

• Mutating acidic residues near the Hfq active site makes it less selective

• Asp 9 to Ala mutation bypasses recruitment of Hfq to AAN sites in mRNAs

• Acidic residues help Hfq discriminate between correct and non-specific targets