Global Structure of the Hfq-Poly(A) Complex.

A C-terminal truncation mutant of Ec Hfq, lacking residues 70–102, was crystallized with the oligoribonucleotide A₁₅. This truncation, although reported to disrupt rpoS binding (37), has little effect on poly(A) binding affinity but moderately affects the stability of Ec Hfq (38). The Hfq-poly(A) structure was solved by molecular replacement using the apo Ec Hfq structure (PDB ID code 1hk9) (39) as the search model and refined to R_work and R_free values of 22.3% and 25.9%, respectively, at 2.40 Å resolution (Table S1). The asymmetric unit of the Hfq-poly(A) crystal contains 3 Hfq subunits and 9 adenosine nucleotides. The biologically relevant hexamer (6, 8) and an A₁₈ oligoribonucleotide are generated by a crystallographic 2-fold, resulting in individual nucleotide binding sites having 83.3% (15/18) occupancies. Regardless, the electron density of each nucleotide is clear (Fig. 1A). Each Hfq subunit contains an N-terminal alpha helix (α1, residues 8–18) followed by an extensively twisted 5-stranded β-sheet (β1, residues 22–26; β2, residues 29–39; β3, residues 43–47; β4, residues 51–55; β5, residues 61–64).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

The structure of Ec Hfq bound to poly(A) RNA. (A) Electron density maps for the RNA tripartite module. Composite F_o–F_c omit map densities are shown in blue (0.8σ) and 2F_o−F_c densities are shown in white (1σ). A stick model of the distal-side binding A₁₅ RNA is shown with carbons colored slate. Hfq is shown as a gray surface. (B) A ribbon diagram of the biologically relevant Hfq hexamer and A₁₈ oligonucleotide superimposed onto the apo Hfq hexamer (green). The Hfq trimers and A₉ oligonucleotides of the 2-fold related asymmetric units are gray and black, respectively. One of the 6 tripartite binding repeats is labeled A (adenosine site), R (purine nucleotide site), and E (entrance/exit site). (C) Side view. A stick model of the bound RNA on the distal side is shown with carbons colored white, with one tripartite binding unit colored bluish gray. A CPK model of AU₅G RNA with green carbons is docked in the proximal pore, as seen in the Sa-Hfq-AU₅G structure. Hfq is shown as a gray surface. Figures were produced with PyMol (64).

Superimposition of the refined apo and poly(A)-bound Ec Hfq hexamers reveals no significant structural changes (root mean square deviation = 0.50 Å) indicating the poly(A) binding site is preformed (Fig. 1B). The poly(A) RNA tract binds to the distal face of Hfq on which the phosphodiester backbone traces a circular, weaving pattern (Fig. 1 B and C). This circular binding conformation provides a ready explanation for data that showed covalently closed circularized A-tracts bound to Ec Hfq more tightly than their linear A-tract counterparts (40). Most of the RNA is solvent exposed and sits on Hfq much like a crown on the head of a monarch (Fig. 1C). None of the riboses take the C3′-endo conformation that is typically associated with A-RNA. Rather, 7 take the C2′-endo pucker, and the remaining 2 the closely related C1′-exo or C3′-exo conformation. Each adenine base is found in the anti conformation.

The Tripartite (A-R-E) RNA Binding Motif.

Hfq binds the poly(A) tail by sectioning the sequence into 6 trinucleotide repeats with the 5′-adenosine nucleotide binding to the adenosine specificity site (A site), the following nucleotide to the purine nucleotide selectivity site (R site; so named because of its likely ability to bind guanosine as well as adenosine), and the third nucleotide to a nondiscriminatory RNA entrance/exit site (E site; Fig. 1). Thus, the distal face of hexameric Hfq has the capacity to bind 18 nucleotides.

The specificity of Hfq for adenosine nucleotides is effected primarily by the A site, which is a surface-exposed groove composed of residues from β-strands 2 and 4 (Fig. 2A). Complementary hydrogen bonds between the peptide backbone amide and carbonyl oxygen groups of residue Gln-33 and the N7 and exocyclic N6 atoms of the base ensure adenine specificity and concomitant discrimination against guanine (Fig. 2B). Additional adenine-Hfq interactions include base stacking against the side chain of residue Leu-32 and a polar interaction between the N1 atom and the Nε amide of residue Gln-52. The 5′ phosphate group of the A site-bound adenosine nucleotide interacts with the peptide backbone amide of residue Lys-31. Although the closest approach of the side chain Nε group of residue Lys-31 to the RNA is ≈6 Å, substitution of Lys-31 with alanine results in a 100-fold loss of affinity for A₁₈ (35, 36) and likely reflects its importance in providing a positive electrostatic surface to complement the negatively charged RNA sugar phosphate backbone and in productive electrostatic steering of polynucleotides to the binding pockets of the distal face (41, 42). Pyrimidines could reciprocate the donor-acceptor hydrogen bonding pattern of the peptide backbone in the A site, but would require the base to be in higher-energy syn conformation, and for uridine, require a ≈3 Å shift. This shift would disrupt the phosphate backbone hydrogen bond to Lys-31 and a key ribose-protein hydrogen bond in the R site (vide infra). Both pyrimidines also display steric clash with the carbonyl oxygen of Lys-31 (Fig. S1) that clearly would contribute to a low distal-side binding affinity of the oligoribonucleotide C₆ (K_d = 6,200 nM), which is ≈39-fold weaker than A₆ (K_d = 160 nM), and the strong preference of AU₅G to bind to its high-affinity proximal side site (Table 1).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

The adenosine specificity site of Hfq. (A) The adenosine specificity pocket is labeled A and lies flat on the distal face of Hfq. The 3′-proximal purine nucleotide selectivity pocket (R) is shown with its adenine ring sandwiched in a crevice between 2 adjacent subunits (shown as magenta and cyan ribbons). The RNA is depicted as a gray worm with bases and sugars shown as gray filled structures. (B) The interaction between the A-site adenosine and Hfq. The interacting residues are labeled and shown as sticks or CPK models with carbons colored gray, oxygens red, and nitrogens blue. The adenosine nucleotide is shown as a stick model with carbons colored white, oxygens red, nitrogens blue, and phosphorus orange. Hydrogen bonds are depicted as black dashes with distances in Ångstrom.

The R site is a crevice that is formed between the β-sheets of neighboring subunits (Figs. 2A and 3A). The adenine ring inserts into the crevice and stacks against the side chains of residues Tyr-25, Leu-26′, Ile-30′, and Leu-32′ (where the prime denotes residues from the other subunit). Replacement of residue Tyr-25 with alanine results in a 100-fold decrease in binding affinity for A₁₈, and changing residue Ile-30 to alanine or aspartate lowers the affinity for A₂₇ nearly 10-fold, thereby confirming the importance of these interactions in poly(A) binding (35, 36). The exocyclic N6 of this adenine engages in a hydrogen bond to the Oε of residue Gln-52′, thereby linking A and R sites. The adenines bound in the A and R pockets are also connected by a nondiscriminating water-mediated hydrogen bond between the A-site N3 atom and R-site N7 atom. The exocyclic N6 of the R-site adenine is also proximal to the N3 atom of the A-site adenine (d = 3.6 Å) but is offset. The bases are not engaged in stacking interactions. Additional Hfq-R-site adenine hydrogen bonds are made between the adenine N1 and Oγ of residue Thr-61 and the N3 and Nδ of residue Asn-28′ via a water molecule (Fig. 3B). The R-site ribosyl 2′ hydroxyl group makes a hydrogen bond to the carbonyl oxygen of residue Gly-29, likely contributing to the preference of Hfq for RNA over DNA. Indeed, the Hfq dissociation constant for DNA-A₆ is 9,500 ± 2,800 nM, which is nearly 60-fold higher than that of A₆ (Table 1). In addition to its role in poly(A) binding, the R site is the most likely binding pocket for ATP and ADP as residue Tyr-25 has been implicated in such binding (43). However, ATP or ADP/AMP binding to the A site cannot be excluded as either a secondary or primary binding site.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

The purine nucleotide selectivity site of Hfq. (A) Top view showing all interacting residues as CPK. (B) Side view. Selected RNA interacting residues are labeled and shown with sticks with carbons colored gray, oxygens red, and nitrogens blue. The bound adenosine is shown as sticks with carbons colored white, oxygens red, nitrogens blue, and phosphorus orange. All other atoms in the pocket are shown as white spheres. Prime numbers represent residues from a neighboring Hfq subunit. Hydrogen bonds are depicted as black dashes with distances in Ångstrom.

The E site lies mostly above the surface of the distal face and makes no adenosine-protein interactions (Fig. 1 A and C). Thus, the exposed E site offers little in the way of sequence discrimination and likely represents the entrance and exit points of RNA tracts. Despite its lack of direct interaction with the protein, the 3 independent E-site nucleotides show well-conserved structures due to geometric constraints placed on nucleotide binding to the A and R sites.

Hfq-Poly(A) Binding Mechanism Is Novel.

The poly(A) binding mechanism of Ec Hfq differs completely from that which Sa Hfq uses to bind small U-rich RNAs (34). Indeed, Sa Hfq binds a U-rich oligoribonucleotide near a basic pore on the proximal face using 6 essentially identical nucleotide binding pockets (Fig. 1C and and Fig. S2). Similar binding by Ec Hfq to small U-rich oligonucleotides is expected in light of the strong sequence conservation of the Sa and Ec Hfq proximal sites (Fig. S2) and mutational studies on Ec Hfq that show only proximal side residues involved in the binding of U-rich and sRNA sequences (35, 36). Superimposing the hexamers of the Sa Hfq-AU₅G and Ec Hfq-A₁₈ complexes results in an rmsd of 0.84 Å and shows no steric clash between the proximally and distally bound RNAs. Combined, these findings provide structural evidence that a single Hfq hexamer could facilitate sRNA-mRNA annealing by simultaneously binding the sRNA to its proximal face and the mRNA to its distal face.

Intriguingly, despite belonging to the same superfamily and showing similar pinwheel-like binding of nucleotides to their respective proximal faces, the 2-sided RNA binding mechanisms of Ec Hfq and the Sm proteins of the U1 snRNP diverge with respect to their distal sides (44). Indeed, an RNA stem-loop structure extends above the Sm distal face and appears to have minimal interaction with this surface of the heptamer. Moreover, the RNA of the spliceosomal U1snRNP complex threads through the central pore of the Sm heptamer. Although Hfq might use similar pore threading with longer sRNAs, no evidence currently supports this possibility. The divergent distal face binding mechanisms of the Hfq and Sm proteins are likely a reflection of their different oligomer assembly mechanisms and differently evolved functions in RNA processes (45).

The poly(A) binding mechanism of Hfq also differs entirely from that used by the human poly(A) binding protein (hPABP), which instead uses RRM (RNA recognition motif) motifs, the β-sheets of which form a long trough to bind an extended, rather than circular, RNA (46). For the greater part, hPAPB uses side chain/base contacts to ensure specificity for adenines. One adenine is read by the peptide backbone, but beyond this there is little in common between the poly(A) binding mechanisms of Hfq and PABP. Indeed, even the peptide backbone readout of adenine differs between them; PABP contacts the N6 and N1 atoms, and Hfq, the N6 and N7 atoms. Thus, eukaryotes and prokaryotes exploit different mechanisms to bind RNA poly(A) tails.

The multitriplet purine binding mode of Hfq-poly(A) RNA does have some parallels with the RNA binding mechanism of TRAP, the Bacillus subtilis trp RNA binding attenuation protein, whereby 11 GAG/UAG triplets bind to the upper perimeter of that undecameric protein (47). Unlike Hfq-poly(A) binding, each GAG triplet binds to the edges of the β-sheet of TRAP. In addition, base-base stacking of the GAG triplet is an integral part of the binding mechanism, and each guanine is recognized specifically by engaging in one or more hydrogen bonds to proximal side chains. Neither of these is found in the Hfq-poly(A) complex. Like TRAP, the B. subtilis antiterminator protein HutP also binds UAG triplets, but again the RNA binding mechanism of this hexamer shows little resemblance to that used by Hfq (48). Indeed, each subunit of HutP uses residues from two loops to engage in side chain-purine ring hydrogen bonds to the adenosine and guanosine nucleotides. Moreover, the adenine and guanine bases of each UAG triplet are stacked. One similarity is found: like the adenosines bound to the E-site of Hfq, the uridines of the HutP-bound UAG triplets are not engaged in sequence-specifying contacts.

Hfq-poly(A) binding does bear a superficial resemblance to the RNA trinucleotide binding mechanism of the RsmA/CsrA (carbon storage regulator protein) family (49, 50). RsmA/CsrA binds as a dimer to 2 stem-loop structures that contain ANGGAN cores to regulate posttranscriptionally virulence factors, carbohydrate metabolism, biofilm production, and motility (51). The GGA triplets are highly conserved amongst sRNAs that are bound by RsmA/CsrA family members (52). The conserved GG pair of each bound oligoribonucleotide forms part of a 3-nucleotide loop that is recognized by a β-strand and loop via noncontiguous peptide backbone base-specific hydrogen bonds and an arginine side chain hydrogen bond to the N7 of the more 3′ guanine. Also different from the poly(A) binding mechanism of Hfq, these guanines are coplanar and form a hydrogen bond between the N2 amino group of one base to the N7 of the other. The conserved adenine of the GGA triplet is part of the RNA stem and read by the main-chain NH and CO group of β-strand residue Ile-3. Like Hfq binding to the A-site adenine, RsmA/CsrA utilizes the Hoogsteen edge of the base (Fig. 2B). Given the presence of at least 2 functionally important and distinct RNA binding surfaces, which can be filled simultaneously (8, 53), Hfq is qualitatively similar to the small RNA quality control protein Ro, which also uses separate surfaces to bind its multiple RNA ligands (54).

Hfq Binds Poly(A-R-N) and Poly(A-R-N-N′) Tracts.

At first glance, the A and R sites seem ideal for adenine base-only binding. However, modeling studies suggested that the R site could also accommodate a guanine base but neither pyrimidine, which are unable to reach deeply enough into the pocket to engage in direct interactions with Hfq (Fig. S3). The inability of pyrimidines to contact R-site residues would essentially preclude tight binding by [A-(pyrimidine)-N] tracts to the distal face. By contrast, guanosine binding to this site requires only a simple rotation of the Gln-52 side chain to make an Nε-O6 hydrogen bond. Such rotation is facilitated by the lack of any hydrogen bonds between the Nε and the rest of the protein. Moreover, the guanine exocyclic N2 would be able to make a hydrogen bond to the hydroxyl group of residue Ser-60. This finding implies that RNA binding to the distal face requires only 2 consecutive purines, of which only one must be an adenine. Given the stereochemical constraints of the A site, A-site binding adenosine nucleotides must be separated by at least 2 nucleotides and adjacent to either a 3′ adenosine or guanosine nucleotide. Hence, RNA with embedded triplets (5′-A-R-N-3′)_i, (where R is a purine nucleotide and N is any nucleotide) should bind the distal face of Hfq. In accord, Hfq binds (GGA)₉, which contains 8 successive AGG triplets, with a K_d = 16 nM—an only 10-fold lower affinity than Hfq binding to (AAA)₉ (Table 1 and Fig. S4). The less energetically favorable binding of guanosine to the R site originates in part by the required relocation of the Oε oxygen of residue Gln-52 to below the adenine ring in the A site. (GCA)₉, which contains 8 successive AGC triplets, binds Hfq with a K_d = 70 nM. The higher K_d of (GCA)₉ might result from its ability to form a stabile stem-loop structure (as predicted by mFold) (55) that would be unsuitable for binding the distal face and require an additional energy input to melt this RNA secondary structure. From the Ec Hfq-A₁₅ structure we also reasoned that an extra nucleotide added 3′ to the E-site nucleotide should not interfere with distal-side RNA binding. Remarkably, (GCCA)₉, in which a second C is inserted to form 8 consecutive AGCC quartets and disrupts the predicted RNA secondary structure of (GCA)₉, binds Hfq with ≈1.7-fold higher affinity (K_d = 42 nM) than (GCA)₉.

The effect of (A-R-N) tract length on affinity was also tested. (A-R-N)₂ tracts bind Hfq less well than longer tracts but still with considerable strength, whereby the affinity for Hfq drops only 5-fold for (GGA)₂, as compared with (GGA)₉, and ≈10-fold in comparison with A₆ (Table 1). The addition of a third G to the triplet, i.e., (GGG)₂ or G₆, results in a ≈170-fold loss of affinity as compared with (GGA)₂, and supports the premise that a guanine cannot bind the A site strongly. Indeed, even the single AAA triplet binds better than G₆ (Table 1). However, caution must be used in interpreting these latter data, as A₃ and A₆ might bind the proximal site as well as the distal site. Regardless, the significant gain in affinity of Hfq for longer A-R-N tracts can be attributed to an increase in the local concentration of these triplets, thus allowing the tract to zipper cooperatively into place, and the positive electrostatic nature of the distal face (56), which through electrostatic steering (41, 42) can attract longer nucleotides more effectively. Although the distal face can accommodate 6 (A-R-N) triplets, once 5 A-R-E sites are filled, the presence of additional triplets does not affect binding affinity (compare A₁₆ and A₂₇, the values of which parallel those reported previously) (refs. 35 and 40, and Table 1).

Functional Implications of Hfq Binding to Poly(A-R-N) Tracts.

The finding that Hfq binds poly(A-R-N) sequences with high affinity has wide-ranging functional consequences. Indeed, these data and the finding that Hfq inhibits PNPase (29), a component of the degradosome that adds AG-rich polynucleotide tails to mRNAs undergoing degradation, suggest that Hfq modulates PNPase activity by blocking access of this exoribonuclease to the 3′ end of the A-R-N tracts through its ability to bind such tracts to its distal face. Hfq has been shown already to protect poly(A) tails from exonucleolytic degradation by PNPase, and similar protection of A-R-N tracts is logical (28). The Hfq-poly(A) structure also suggests how Hfq changes PAP I from a distributive to a processive enzyme, but only after the addition of ≈20 adenosine nucleotides to the mRNA, i.e., the complete filling of the Hfq poly(A) binding site (27). Hfq binding to the A-tract would remove the growing poly(A) tail from the PAP I catalytic pocket, thereby aiding product dissociation and preventing product inhibition or backtracking. Intriguingly, in the presence of Hfq, PAP I produces poly(A) tails in vivo that are on average ≈20 nucleotides longer, again consistent with filling the 6 (A-R-N) binding sites on Hfq (29).

The broader biological significance of high-affinity Hfq-(A-R-N)_i tract binding can be seen by the recent identification of functionally important Hfq binding-A-R-N tracts in the upstream leader sequences of rpoS mRNA, including the AAYAA element, which contains an (AAN)₄ tract. The AAYAA element is critical for rpoS interaction with its regulatory sRNA, DsrA (57). Indeed, we found the AAYAA element (GGGAACAACAAGAAGUUA) binds truncated Ec Hfq very tightly (K_d = 7.7 nM). Given that the AAYAA element in the 5′ UTR of rpoS begins at nucleotide −370, a number of potential (A-R-N)_i tract-Hfq binding sites might exist in other sRNA-regulated messages that contain extended upstream regions. Moreover, the functional importance of such Hfq-binding (A-R-N)_i tracts within mRNA coding sequences (CDS) is likely, and supported by the recent finding that the Hfq-associated sRNA, MicC, targets the CDS of ompD mRNA and that the target site, which contains an AACAAA stretch, is bound (albeit weakly) by Hfq (58). Finally, a simple search of the E. coli genome for poly(A-R-N)_i sequences (http://genolist.pasteur.fr/Colibri/ using “Search Pattern”) revealed 69 unique (A-R-N)₅ tracts in the mRNA translation initiation regions (spanning nucleotides −100 to +1) with the average start site at position −44 and 14 (≈20%) overlapping the Shine-Dalgarno (SD) site (−23 to +1). A similar search for (A-R-N)₄ tracts yielded 394 unique examples with 72 (≈18%) overlapping the SD box and an average start site at nucleotide −39. Thus, Hfq is expected to have a more expansive role as a riboregulator than already described.