Structural basis for gene regulation by a thiamine pyrophosphate-sensing riboswitch

Abstract

Riboswitches are metabolite-sensing RNAs, typically located in the non-coding portions of messenger RNAs, that control the synthesis of metabolite-related proteins^1–3. Here we describe a 2.05 Å crystal structure of a riboswitch domain from the Escherichia coli thiM mRNA⁴ that responds to the coenzyme thiamine pyrophosphate (TPP). TPP is an active form of vitamin B₁, an essential participant in many protein-catalysed reactions⁵. Organisms from all three domains of life^6–9, including bacteria, plants and fungi, use TPP-sensing riboswitches to control genes responsible for importing or synthesizing thiamine and its phosphorylated derivatives, making this riboswitch class the most widely distributed member of the metabolite-sensing RNA regulatory system. The structure reveals a complex folded RNA in which one subdomain forms an intercalation pocket for the 4-amino-5-hydroxymethyl-2-methyl-pyrimidine moiety of TPP, whereas another subdomain forms a wider pocket that uses bivalent metal ions and water molecules to make bridging contacts to the pyrophosphate moiety of the ligand. The two pockets are positioned to function as a molecular measuring device that recognizes TPP in an extended conformation. The central thiazole moiety is not recognized by the RNA, which explains why the antimicrobial compound pyrithiamine pyrophosphate targets this riboswitch and downregulates the expression of thiamine metabolic genes. Both the natural ligand and its drug-like analogue stabilize secondary and tertiary structure elements that are harnessed by the riboswitch to modulate the synthesis of the proteins coded by the mRNA. In addition, this structure provides insight into how folded RNAs can form precision binding pockets that rival those formed by protein genetic factors.

More than 2% of the genes in some species are regulated by riboswitches^4,10–13, whose representative classes compete in number with known metabolite-sensing regulatory proteins¹⁴. On metabolite docking, the sensing domain of the riboswitch shows conformational stabilization, which typically alters the base-pairing arrangements in the adjoining expression platform carrying gene-expression signals^1–3. Riboswitches have a remarkable affinity for their cognate ligands and can discriminate against even closely related analogues, as shown by biochemical experiments^1,13,15 and X-ray structures of purine-sensing riboswitches bound to hypoxanthine¹⁶, guanine¹⁷ and adenine¹⁷. TPP and pyrithiamine pyrophosphate (PTPP; Fig. 1a) are chemically more diverse than purines and pose several additional challenges for negatively charged RNA receptors, namely their large size, conformational flexibility and, most importantly, the presence of negatively charged phosphate groups¹⁸.

Figure 1. Structural models of a TPP riboswitch and its ligands.

a, Chemical structures of the natural metabolite TPP and the antimicrobial compound PTPP. b, c, Crystal structure of the TPP-bound sensing domain, showing front (b) and top (c) views. The RNA is in a stick-and-ribbon representation, with bound TPP in red. Stems, loops and junctions are colour coded. d, Schematic depiction of the RNA tertiary fold observed in the structure. Tertiary contacts formed by hydrogen bonds (w, water-mediated bonds) between bases and stacking interactions are represented by thin and thick dashed lines, respectively. Red shading shows nucleotides conserved in more than 97% of sequences (J. E. Barrick and R.R.B., unpublished observations). Encircled M notations represent Mg²⁺ ions.

We crystallized an 80-nucleotide TPP-sensing domain from the thiM mRNA⁴. This RNA conforms well with the consensus sequence and secondary structure model for TPP riboswitches that was established by previous phylogenetic^6–9 and biochemical^4,6,19 analyses (Supplementary Fig. S1). The structure (Fig. 1b–d, Supplementary Fig. S2 and Supplementary Table S1) reveals that the conserved nucleotides and secondary structure elements common to all TPP riboswitches form a complex tertiary architecture consisting of two parallel helical domains (P2/J3-2/P3/L3 and J2-4/P4/P5/L5) connected to a helix (P1) by means of a three-way junction.

Unlike most other ligand–RNA interactions that exploit the helical grooves of RNA, TPP is positioned perpendicular to the two main helical domains and uses separate pockets formed by each helical segment to grip the ends of the compound in an extended conformation (Fig. 2). In addition to the interdomain bridge formed by the ligand, the riboswitch uses tertiary contacts between L5 and P3, and between J2-4, P2 and P4, to assist in stabilizing the global fold.

Figure 2. Structure and interactions in the TPP-binding pocket.

a, Stereo view of the central region of the complex containing bound TPP. b, View of TPP, coordinated Mg²⁺ ions (magenta) and water (blue spheres) in the binding pocket. c, Details of the interactions between the HMP ring and RNA. d, Hydrogen bonding between Mg²⁺ ions and RNA.

The J3-2 region (Fig. 1d) is essential in the recognition of the 4-amino-5-hydroxymethyl-2-methylpyrimidine (HMP) ring. The ring intercalates between G42 and A43, and forms hydrogen bonds between its polar functionalities, previously shown to be critical for molecular recognition⁴, and the G40 base and the 2-OH′ of G19 (Fig. 2b, c). This binding arrangement is held by a T-loop-like turn²⁰ formed by the conserved U39-G40-A41-G42-A43 segment, closed by a reverse Hoogsteen U39·A43 pair (Fig. 3a). The backbone is extended at both G40-A41 and G42-A43, resulting in the mutual intercalation and insertion of the HMP ring between these interacting segments. Further, the fold is stabilized by continuous stacking of A41, G42, HMP, A43 and G21, and by the formation of (G19·A47)·G42 and (G18-C48)·A41 triples (Fig. 2a and Supplementary Fig. S3).

Figure 3. Tertiary interactions defining TPP riboswitch structure and accessibility to the binding pocket.

a, Interaction between J3/2 and P2, mediated by the HMP ring. b, Stabilization of the J2-4 junction by two stacked tetrads (in space-filling representation). c, Interactions between L5 and P3 mediated by three K⁺ ions (red spheres). d, Surface representation of RNA and accessibility to the TPP-binding pocket. TPP is depicted in a stick and mesh representation.

TPP riboswitches were the first of several classes found to make productive interactions with negatively charged phosphate groups^4,12,15. The pyrophosphate group of TPP is bound in a spacious pocket by a pair of hexa-coordinated Mg²⁺ ions (Mg1 and Mg2) with octahedral ligation geometry that are positioned by direct and water-mediated hydrogen bonds with RNA (Fig. 2d and Supplementary Fig. S1). The terminal phosphate of TPP is coordinated to both Mg1 and Mg2, whereas the thiazole-linked phosphate is coordinated to Mg2. Mg1 is directly coordinated to the O⁶ carbonyls of the conserved G60 and G78 that are located in the region previously implicated in pyrophosphate recognition⁴ (Fig. 2b, d, and Supplementary Fig. S2). In addition, the conserved C77 and G78 form hydrogen bonds to the oxygen atoms of the terminal phosphate of TPP.

An analysis of 72 structures of proteins bound to TPP or its analogues from the Protein Data Bank indicates that proteins typically use charged amino acids to position Mg²⁺, Ca²⁺ or Mn²⁺ ions in a site equivalent to Mg2. However, the Mg1 ion is unique to TPP riboswitches. A bivalent cation at this location allows TPP to reach into the pyrophosphate-binding pocket, thereby stabilizing tertiary interactions important for gene regulation and corroborating the requirement of Mg²⁺ for TPP binding in eukaryotic²¹ and bacterial (Supplementary Fig. S4) TPP riboswitches. One could consider the ligand as TPP with two bound Mg²⁺ ions, which overcomes the ligand’s strongly negative electrostatic character. Other riboswitches could use similar simple structural arrangements in combination with cations and water to bind phosphorylated ligands selectively, which would make RNA surprisingly adept at binding negatively charged ligands. The dual bivalent cation arrangement is also distinct from that found in ribosomal RNAs that bind non-phosphorylated antibiotics, in which some contacts with RNA are mediated by single cations^22–24.

Nucleotides distal to the ligand-binding sites, conserved in sequence or secondary structure, are important in forming the overall architecture of the TPP-sensing domain. J2-4 (Fig. 1d) is stabilized through the formation of a pair of stacked tetrads, where highly conserved A56·G83 and A53·A84 non-canonical pairs (Supplementary Fig. S5a, b) are aligned in the minor groove of adjacent G·C pairs within P2, forming type I A-minor motifs²⁵. Continuous stacking is observed between P1 and P2, and between non-canonical pairs in J2-4 and P4 (Fig. 3b). This structural arrangement is expected to be stabilized by ligand binding, and reinforces previous suggestions^4,12,13 that stabilization of P1 stems in riboswitches is important for the control of gene expression.

Another key tertiary interface involves multiple base–base and base–backbone interactions between nucleotides of L5 and P3 (Fig. 3c), thereby locking into register side-by-side arrangements of P4/P5 and P2/J3-2/P3 segments (Fig. 1b). Three adjacent K⁺ ions additionally stabilize the interactions (Fig. 3c). It should be noted that nucleotides of the L5/P3 interface are poorly conserved, and other TPP riboswitches could be stabilized by alternative tertiary contacts spanning this region.

A surface view of the complex shows that the thiazole and diphosphate moieties of TPP are visible, whereas the intercalated HMP ring is buried (Fig. 3d). The deep insertion of the HMP ring most probably implies a conformational transition within the riboswitch upon binding TPP. Support for this interpretation comes from in-line probing experiments, which have identified numerous internucleotide linkages that become conformationally restricted on TPP binding⁴. The structure shows that these regions either contact the ligand directly or participate in critical contacts that form the global structure of the sensing domain.

To assess global conformational changes brought about by ligand binding, we conducted nuclease V1 (helix-specific) and nuclease T2 (single-strand-specific) partial digests of the free and ligand-bound riboswitches. In the absence of TPP these nucleases cleave regions that participate in TPP binding, such as nucleotides (nt) 39–44 (J3-2) and nt 59–61 (part of P4/P5), as well as some regions that are not directly involved in formation of the TPP-binding pocket, such as nt 67–72 (L2) and nt 84–86 (Fig. 4 and Supplementary Figs S6 and S7). The addition of TPP decreases nuclease cleavage in regions that are important for both ligand binding and RNA folding (Fig. 4a, c). These results indicate that TPP not only stabilizes the two sections of the binding pocket by bridging the J3-2 and P4/P5 regions but also promotes the formation and/or stabilization of distal tertiary contacts within the J2-4/P2/P4 region and between L5 and P3.

Figure 4. Structural probing of the TPP riboswitch and implications for TPP-mediated gene repression.

a, Representative RNase V1 and T2 cleavage patterns for the thiM riboswitch (nt 1–166). 5′ ³²P-labelled RNAs were treated by nucleases in the absence (−) or presence (+) of one, three and ten equivalents of TPP as described in Supplementary Methods and ref. 30. ⁻OH and Bc stand for ladders prepared by partial digestion with alkali or B. cereus RNase, respectively. The major cleavage protections and enhancements in the presence of TPP are labelled with triangles and stars, respectively. b, Primer extension analysis in the absence and presence of TPP or TMP. RT pauses are indicated by arrows. c, Summary of structure probing experiments. Major TPP protections against V1 (red) and T2 (magenta) RNases are shown. The RT pauses are indicated by arrows. d, Typical mechanisms of TPP-specific gene repression. Top: translation initiation regulation (thiM genes). Bottom: transcription termination regulation (thiC genes). Complementary sequences and alternate base-pairing are shown in blue. SD sequence and initiation codon are shaded green. TPP and Mg²⁺ ions are depicted in red and magenta, respectively. e, Diagram of the OFF state of the E. coli thiM riboswitch.

Tertiary contacts associated with the TPP-stabilized fold of the riboswitch are also apparent in primer extension assays using reverse transcriptase (RT). In the presence of various analogues, including thiamine monophosphate (TMP), RT does not alter its pattern of pausing compared with that observed in the absence of ligand (Fig. 4b). This probably reflects the inability of TMP to span the two ligand-binding portions of the sensing domain fully and to stabilize the global fold, consistent with the progressive loss of binding affinity for thiamine ligands that carry one or no phosphates⁴. However, in the presence of TPP, RT pauses at two distinct positions (Fig. 4b). One pause at G86 occurs close to a pair of stacked tetrads anchoring tertiary contacts involving J2-4, P2 and P4, indicating that this conserved structural element becomes stabilized on ligand binding (Fig. 4c). A second RT pause occurs at C74 adjacent to the tertiary contact between L5 and P3, which seems to be sufficiently strong despite the disruption of base pairing in P1, P4 and P5 during primer extension.

The composite TPP-binding pocket can be defined as a molecular ruler in which only a ligand of proper length can stabilize tertiary interactions and modulate gene expression. This structural characteristic allows the riboswitch to achieve up to about 3,000-fold discrimination against shorter TMP and thiamine, and to provide feedback gene regulation only in the presence of biologically active TPP (ref. 4). In contrast to such unprecedented discrimination of related ligands by the RNA distance ruler, purine riboswitches have developed a completely different approach to distinguishing between related ligands. Adenine-specific and guanine-specific riboswitches use a single tight pocket to position their ligands precisely for Watson–Crick pairing with the discriminatory nucleotide, uracil or cytosine, respectively, thus making a single nucleotide the key element governing riboswitch regulation^16,17.

Two predominant modes of TPP-dependent gene regulation, translation inhibition and transcription termination, are illustrated in Fig. 4d. Both require an over-threshold concentration of TPP in the cell for interaction with the riboswitch^4,9,11,19. As evident from our biochemical experiments, binding of TPP to the riboswitch induces conformational rearrangements leading to stabilization of the overall RNA fold and, most importantly, to stabilization of conserved tertiary contacts adjacent to P1, as defined in the crystal structure (Fig. 4e). These interactions, in turn, stabilize P1 and promote folding of the expression platform, either to a hairpin that sequesters the Shine–Dalgarno (SD) sequence or to a terminator hairpin (OFF state); this results in the failure of translation initiation or premature transcription termination, respectively (Fig. 4d). Without TPP the riboswitch adopts alternative conformations, forming an anti-terminator hairpin or opening the SD sequence for ribosome binding (ON state).

Recent research¹⁹ has revealed that the antimicrobial compound PTPP (Fig. 1a) binds to bacterial and fungal TPP riboswitches and can turn off the expression of critical biosynthetic genes. The structure indicates that the loss of PTPP activity, associated with drug-resisting mutations in TPP riboswitches, might be due to the disruption of key tertiary contacts made by tetrads (C50G, A84G, G86A) and J3-2 elements (A47Δ) (Fig. 3a, b). PTPP carries a pyridine ring in place of the thiazole ring of TPP (Fig. 1a). Because the TPP riboswitch does not make any substantive contacts with the ligand in this chemically distinctive region, our structure-based model reveals why PTPP functions as a mimic of the natural ligand (Supplementary Fig. S8). Given the important functional role of riboswitches in numerous microorganisms and the fact that riboswitches have not yet been detected in the human genome, structures of TPP and other riboswitch classes should enable researchers to employ rational drug discovery strategies to create novel classes of antibacterial and antifungal compounds that target riboswitches^19,26,27.

METHODS

Crystallization

The TPP–riboswitch complex was prepared by mixing RNA transcribed in vitro and TPP in a buffer containing 50 mM potassium acetate pH 6.9 and 5 mM MgCl₂. Crystals were grown by hanging-drop vapour diffusion. The complex solution and reservoir solution (28% w/v poly(ethylene glycol) 4000, 100 mM sodium acetate pH 4.8 and 200 mM ammonium acetate) were mixed in a 1:1 ratio and incubated at 20°C. For soaking, crystals were washed with stabilizing solution lacking MgCl₂ and then incubated in the presence of 3 mM Os(NH₃)₆ for two days.

Structure determination

Native and Os(NH₃)₆ multiple anomalous diffraction (MAD) data were collected at beamline X25 at the Brookhaven National Synchrotron Light Source. Data were processed with the HKL2000 suite of programs (HKL Research). The structure was determined by using MAD osmium data and SOLVE/RESOLVE²⁸. The RNA model was built using TURBO-FRODO (〈http://afmb.cnrs-mrs.fr/rubrique113.html〉) and refined with REFMAC²⁹ using a native data set (Supplementary Figs S9–S11). TPP and cations were added to the model on the basis of analysis of 2F _o − F _c and F _o − F _c electron density maps. Na⁺, K⁺ and hydrated Mg²⁺ were modelled on the basis of the number of coordination bonds, their distances and their coordination geometry. RNA residue 55 has a partial electron density.