Riboswitches are regions of mRNA to which a metabolite binds in the absence of proteins, resoulting in alteration of transcription, translation or splicing. The most widespread forms of riboswitches are those responsive to TPP (thiamine pyrophosphate) the active form of vitamin B1, thiamine. TPP-riboswitches have been found in all bacterial genomes examined, and are the only ones found in eukaryotes. In each case, the riboswitch appears to regulate the expression of a gene involved in synthesis or uptake of the vitamin. Riboswitches offer an attractive target for chemical intervention, and identification of novel ligands would allow a detailed study on structure–activity relationships, as well as potential leads for the development of antimicrobial compounds. To this end, we have developed a medium-throughput methodology for screening libraries of small molecules using biophysical methods.
- chemical intervention
- fragment screening
- structure–activity relationship
- thiamine analogue
- thiamine pyrophosphate (TPP)
What are riboswitches?
Riboswitches are cis-acting regulatory sequences in mRNA that bind specific metabolites, causing alteration in the secondary structure of the transcript [1–3]. This conformational change either activates or more frequently represses gene expression, for instance, by blocking translation or causing premature termination of transcription. Thus gene expression is regulated directly by small molecules without mediation of a protein. Natural riboswitches responsive to over 15 different metabolites have been identified in bacteria [2,3], with ligands ranging from the simple amino acid glycine to adenosylcobalamin (coenzyme B12), one of the most complex primary metabolites (Figure 1A). The fact that the ligands are fundamental cellular metabolites, many of which are related to nucleic acids, has led to the proposal that riboswitches are a relic of the ancient RNA world. Nonetheless, they have also been found in eukaryotes, although to date only those responsive to TPP (thiamine pyrophosphate) (Figure 1A) have been reported, including in the ascomycete fungi Aspergillus nidulans  and Neurospora crassa , the higher plant Arabidopsis thaliana  and the green alga Chlamydomonas reinhardtii . In all organisms, riboswitches usually regulate genes that are involved in the biosynthesis, metabolism or transport of the cellular building block or coenzyme that acts as the ligand. Interestingly, before the first identification of these natural riboswitches in 2002 , artificial RNA sequences that bind small molecules were already known and had been used to control gene expression not only in bacteria but also in eukaryotes such as yeast and mammalian tissue culture cells, where natural riboswitches have so far not been identified [8,9].
Riboswitches comprise two regions: the aptamer, defined as the minimum region of the mRNA capable of recognizing the ligand, and the expression platform, which is responsible for the effect on gene expression (Figure 2A). Inspection of bacterial genome sequences has shown that the aptamers for each metabolite share a common secondary structure, and frequently a number of conserved residues. However, the expression platform controlled by similar aptamers differs considerably, even between different genes in the same organism. For example, the riboswitches that control the thiM and thiC genes of Escherichia coli are both TPP dependent, but whereas the thiM switch operates only by blocking initiation of translation, the thiC switch also causes transcription termination . In eukaryotes, the TPP riboswitches were identified by the fact that they contain the consensus sequence of the TPP-binding aptamer, but the expression platform appears to function by revealing a cryptic 5′-splice acceptor site within an intron [5,7]. Interestingly, in eukaryotes, the position of the riboswitches within the mRNA varies. In some cases they are at the 5′-end of the gene as in bacteria (e.g. in N. crassa  and in THI4 in C. reinhardtii ), but they have also been found in the middle of the coding region (THIC gene in C. reinhardtii ) and in the 3′-untranslated region (THIC gene in A. thaliana ).
Chemical intervention on riboswitches
The fact that riboswitches control vital biosynthetic processes, and naturally fold to bind small molecules with high selectivity and affinity (dissociation constants in the micro- to nanomolar range), makes them ideal targets for chemical intervention. The ability to modulate natural riboswitches with novel ligands would provide the means to probe the role of these regulatory sequences in metabolism. Moreover, there are several potential uses for novel riboswitch ligands. For example, there is interest in using riboswitches to regulate transgene expression in biotechnological applications, with both natural riboswitches such as TPP  and artificial constructs, for example, with those responsive to tetracycline . In addition, compounds that interfere with riboswitches offer a route to novel therapeutics, given the widespread distribution of these control sequences in bacteria, and the fact that they are often present in multiple copies within the cell. The antimicrobial compound pyrithiamine, an analogue of thiamine (Figure 1B), is thought to act by repressing the expression of genes with TPP-responsive riboswitches, thus switching off thiamine biosynthesis [4,11]. Similarly, several lysine analogues that bind to the riboswitch in the Bacillus subtilis lysC gene encoding aspartokinase II have been shown to have an antimicrobial effect .
We are interested in discovering novel ligands for TPP riboswitches, and to this end we have developed a set of biophysical methods to allow rapid screening of different compounds for their ability to bind to the aptamer, and a reporter gene assay to assess the effect on the expression platform [13,14]. Here, we describe the application of these methods to screen compounds such as thiamine analogues and a more chemically diverse fragment library. Fragment-based approaches have proven very successful in the discovery of enzyme inhibitors . Screening small molecules (<300 Da) allows a more effective exploration of chemical space, thereby increasing the probability of finding efficient ligands, and hence it is sufficient to screen libraries of a few hundred compounds rather than the hundreds of thousands of bigger ‘lead-like’ compounds. Once fragments are identified that bind to the riboswitch, they can be elaborated to make more potent ligands.
Structure–activity relationships of ligands binding to riboswitches
Riboswitches function within chemically complex environments. Therefore to regulate gene expression correctly, they must not only bind their ligands but also bind them selectively. The coenzyme B12 riboswitch from the btuB gene of E. coli is specific for the adenosyl form of the cofactor, and methylcobalamin binds much less efficiently . Moreover, modifications to the adenine group diminish binding, and a change in stereochemistry in ring C of the corrin ring eliminates binding . The adenine-responsive riboswitch from the ydhL gene of B. subtilis shows specificity for most parts of the adenine ring system but has a similar affinity for 2-aminopurine as for adenine (6-aminopurine, Kd=300 nM), and 2,6-diaminopurine is an even better ligand (Kd = 10 nM) . In contrast, guanine (2-amino-6-oxypurine) does not bind at all.
Likewise, TPP riboswitches can discriminate between the different forms of thiamine (Figure 1B) within the cell. The aptamer of the E. coli thiM riboswitch binds TPP with a Kd of 8 nM, whereas it binds TMP (thiamine monophosphate) and thiamine 100- and 200-fold less tightly respectively [13,19]. The E. coli thiC riboswitch appears to exhibit even greater discrimination, with a greater than 1000-fold difference in the apparent Kd of TPP compared with TMP and thiamine reported by Winkler et al. . This fine-tuning to favour binding of the pyrophosphate over the mono- and un-phosphorylated forms will be important in vivo, since only TPP has activity as a coenzyme.
Clues as to how this bias is achieved have been revealed by crystal structures of TPP riboswitches co-crystallized with TPP and analogues of TPP [20–23]. Structures of TPP bound to both the E. coli thiM aptamer [20,21] and the THIC riboswitch from the higher plant A. thaliana [22,23] have been solved. The three-dimensional structure can be described as Y-shaped  (Figure 2B), with the tail of the Y forming the switching helix P1, the formation of which causes the expression platform to be turned ‘off’. Stem P1 stacks coaxially with P2 and P3, which form one arm of the Y, and P4 stacks with P5 to form the other arm (Figure 2). The tops of the two arms fold inwards to form interactions (between L5 and P3). TPP is bound stretched out between the two arms, which are referred to as the pyrimidine- and the pyrophosphate-sensing helices respectively. The pyrimidine of TPP binds in J3-2 and is stacked between conserved residues and forms hydrogen bonds with the base and ribose of two further conserved residues (Figure 2A). The pyrophosphate in contrast forms few direct interactions to the RNA. Instead the phosphates are co-bound with Mg2+ ions (in the absence of Mg2+, no binding is observed and the RNA is much less compact) and most of the interactions are with the metal ions and hydrogen bonds with water molecules co-ordinated to the metal ions. As reported by Edwards and Ferré-D’Amaré , these magnesium- and water-mediated interactions seem to be important, since when TMP is bound instead of TPP many of the same bonds are conserved. Thiamine binds thiM with an affinity of 1–2 μM, but, as it lacks phosphate groups entirely, it may not be capable of triggering docking of the pyrophosphate-sensing helix, which is required for stabilizing the off conformation of the riboswitch.
In contrast with purine riboswitches which tend to contact all functional groups on the ligand , the thiazolium ring does not appear to be sensed by the TPP aptamer, as seen in crystal structures of the TPP aptamer bound with four synthetic compounds, pyrithiamine , pyrithiamine pyrophosphate, benfotiamine and oxythiamine pyrophosphate . As the C-2 position of the thiazole ring is the reactive position used to catalyse reactions in TPP-dependent enzymes, it has led to some speculation as to whether TPP riboswitches may once have acted as early coenzyme-dependent ribozymes . A simpler explanation could be that the riboswitch need not recognize the central ring as the other metabolites likely to be encountered can be resolved by sensing only the ends of TPP, and if binding also relied on the presence of the thiazole ring there would be less of a differential between TPP, TMP and thiamine. The apparent lack of sensitivity regarding the central part of the TPP ligand opens up interesting possibilities for creating analogues that bind to riboswitches but not TPP-dependent enzymes.
Methods to identify novel ligands
To facilitate screening of compounds for their ability to bind to the TPP riboswitch, we have developed a set of biophysical techniques that not only allow medium throughput but also provide information on the binding affinity [13,14]. The riboswitch aptamer can be produced in large amounts by in vitro transcription. This is then used in an equilibrium dialysis experiment in which the RNA and the natural radiolabelled ligand (in this case [3H]thiamine) are placed in two separate chambers separated by a dialysis membrane. The system is left to reach thermodynamic equilibrium (generally overnight), and then the radiolabel in the two chambers is measured. To screen for novel ligands, analogues or fragments are included in the chamber with [3H]thiamine, and then the extent of displacement of the radiolabel gives a measure of the affinity of the riboswitch aptamer for the test compound. We have used this method to screen a library of fragments in cocktails of five compounds, followed by individual testing of the components of any cocktail that exhibited greater than 30% displacement of [3H]thiamine. Figure 3 shows the result of screening three different examples of representative cocktails and their deconvolution. Cocktails 1 and 2 contained genuine hits, although interestingly the hit in cocktail 2 gave greater displacement on its own than in the cocktail, suggesting that there was interference from the other compounds, perhaps causing precipitation. This is a good reason for setting a relatively low threshold to choose cocktails to deconvolute . On the other hand, in the third cocktail, which showed 57% displacement, none of the individual fragments displaced more than 20% of the radioligand, and were not taken further. A total of 1300 fragments were tested and 17 hits were obtained . These were counter screened against the B. subtilis lysC riboswitch and, although some were capable of displacing [3H]lysine very efficiently, suggesting that they were non-specific RNA binders, 10 of the compounds did not bind to lysC aptamer, including the three compounds shown in Figure 3. Water-LOGSY (Ligand Observed via Gradient SpectroscopY) and T2 relaxation-edited NMR spectroscopy experiments were used to confirm binding, and the binding affinity was then measured by ITC (isothermal titration calorimetry). The Kd values were in the micromolar range, with one compound having a Kd of 22 μM, although this fragment was one of those that bound to lysC. The ligand efficiencies (free energy of binding per heavy atom)  of the best fragments were high (over 0.6), and comparable with very efficient fragments found in a more typical fragment screen against a protein target . These fragments would provide excellent starting points for synthetic elaboration into more potent binders; however for that to be done rationally, it would first be necessary to determine where and how the fragments bind to the riboswitch, most probably by X-ray crystallography.
By looking at analogues of thiamine we can start to see subtle effects in ligand recognition. For example, one of the two precursor moieties to thiamine, HMP [hydroxymethyl pyrimidine (2-methyl-4-amino-5-hydroxymethylpyrimidine)], showed no detectable binding to thiM. However, when the hydroxyl is replaced by an azide, this analogue has a Kd of 49 μM. This may suggest that the azide is picking up some additional binding interaction around the pyrimidine-binding site that favours its binding over HMP. That the thiazole does not contribute significantly to binding is borne out by the observation that modifying the thiazole ring significantly (e.g. as in the analogue benfotiamine) does not preclude binding. Also, exchanging the thiazolium ring for a methylpyridinium group as in pyrithiamine had only a marginal effect on binding, such that pyrithiamine binds with a comparable affinity to thiamine (6 μM compared with 1.5 μM). In contrast, TPP-dependent enzymes are very sensitive to changes in the thiazolium ring of TPP and analogues that have a neutral ring at this position tend to bind far more tightly (25000-fold in one instance) . This can be explained by the fact that TPP-dependent enzymes are not optimized for binding TPP but for stabilizing intermediates in the enzyme mechanism, and the key intermediates have neutral rings in this position.
Using a relatively simple and quantitative screening method based on biophysical techniques, we have identified a number of novel ligands that are specific for the TPP-responsive thiM riboswitch. These will be useful starting points for exploring the structure–activity relationships in more detail in terms of binding of the ligand to the aptamer, but it should be pointed out that none of the fragment ligands that bound well to the thiM aptamer affected the expression platform in an in vitro transcription–translation system , most likely reflecting the fact that they were unable to stabilize the P1–P1′ stem. Moreover, much remains to be established about riboswitch function in actual cellular processes. In particular, because most of what we know about riboswitches is as a result of in vitro experiments, this has given rise to a conceptually simple model for how riboswitches work, which may not be a true reflection of behaviour in vivo. As well as identifying possible therapeutic compounds, novel ligands for riboswitches offer the means to study the consequences of riboswitches on metabolism. Furthermore, in eukaryotes, in addition to its role in gene expression, RNA is essential for many cellular processes (e.g. replication and protein targeting), and many important plant and animal viruses have RNA genomes. This means that the ability to control RNA function with specific small molecules has enormous potential. In addition to addressing fundamental questions and opening up new aspects of biology, there are some very practical outcomes of this project, including the identification of specific small molecule modulators of RNA function that can be used as tools both in scientific research and in the biotechnology industry.
This work was supported by the Biotechnology and Biological Sciences Research Council.
NACON VIII: 8th International Meeting on Recognition Studies in Nucleic Acids: An Independent Meeting held at The Edge, University of Sheffield, Sheffield, U.K., 12–16 September 2010. Organized by Mike Blackburn, Mark Dickman, Jane Grasby, David Hornby, Chris Hunter, John Rafferty, Jim Thomas, David Williams and Nick Williams (Sheffield, U.K.).
Abbreviations: HMP, hydroxymethyl pyrimidine (2-methyl-4-amino-5-hydroxymethylpyrimidine); TMP, thiamine monophosphate; TPP, thiamine pyrophosphate
- © The Authors Journal compilation © 2011 Biochemical Society