Adenine protonation enables cyclic-di-GMP binding to cyclic-GAMP sensing riboswitches.

In certain structural or functional contexts, RNA structures can contain protonated nucleotides. However, a direct role for stably protonated nucleotides in ligand binding and ligand recognition has not yet been demonstrated unambiguously. Previous X-ray structures of c-GAMP binding riboswitch aptamer domains in complex with their near-cognate ligand c-di-GMP suggest that an adenine of the riboswitch either forms two hydrogen bonds to a G nucleotide of the ligand in the unusual enol tautomeric form or that the adenine in its N1 protonated form binds the G nucleotide of the ligand in its canonical keto tautomeric state. By using NMR spectroscopy we demonstrate that the c-GAMP riboswitches bind c-di-GMP using a stably protonated adenine in the ligand binding pocket. Thereby, we provide novel insights into the putative biological functions of protonated nucleotides in RNA, which in this case influence the ligand selectivity in a riboswitch.

INTRODUCTION

The pKa values of the nucleotide building blocks in RNA and DNA are far away from neutrality (G, U ∼ 9.2, A ∼ 3.9, C ∼ 4.2) in unstructured nucleic acids (Saenger 1988). Watson–Crick base-pairing in double helical structural elements shifts these pKa values even further away from neutrality (Saenger 1988; Thaplyal and Bevilacqua 2014). Therefore, nucleotides in RNAs mostly adopt their canonical neutral protonation states and tautomeric forms. However, many RNAs incorporate non-Watson–Crick base pairs in their structures or adopt intricate tertiary structures. These structural environments can induce significant pKa shifts of RNA nucleotides toward neutrality leading to the occasional presence of nucleotides with altered protonation states in RNA structures. Very simple examples are A:C mismatches embedded in regular A-form double helical structural elements (e.g., Puglisi et al. 1990; Cai and Tinoco 1996; Huppler et al. 2002; Pechlaner et al. 2015). Adenine N1 protonation stabilizes these mismatches since two hydrogen bonds can be formed between A+ and C resulting in a base pair with the same geometry as the classical G:U wobble pair. The pKa for the A in such base pairs is shifted by ∼3 units and can be as high as 8.2 in a particularly stable helical context (Wilcox and Bevilacqua 2013a,b). Similarly, N1 protonated A's are able to stabilize G:A mismatches in Watson–Crick helical contexts (Pan et al. 1999) as well as the parallel double helices formed by poly(rA) (Gleghorn et al. 2016). Protonated C's occur for instance in base triples and triple helices. Both protonated C's and A's have been observed in frame-shifting pseudoknots (e.g., Cornish et al. 2005; Houck-Loomis et al. 2011; Wilcox and Bevilacqua 2013a) and pKa values between 6.2 and 8.2 have been reported for these nucleotides. At physiological pH values these pseudoknots therefore exist as conformational ensembles containing protonated and deprotonated species. Thus, A or C protonation might play a role in tuning conformational equilibria and thereby frameshifting efficiency. Of particular functional importance is the occurrence of nucleotides with shifted pKa values in the active site of ribozymes. The hammerhead, the hairpin, the VS, the twister and the pistol ribozyme all are reported to have a G with a pKa shifted toward neutrality in the active site (Wilcox et al. 2011; Kath-Schorr et al. 2012; Liu et al. 2014; Ren et al. 2014, 2016). There, it supposedly acts as the general base and activates the attacking 2′-OH group in the first step of the phosphodiester cleavage reaction. For both the hairpin and the VS ribozyme an N1 protonated A supposedly is the general acid in the cleavage reaction (Rupert et al. 2002; Wilson et al. 2010; Kath-Schorr et al. 2012). For the twister and the pistol ribozyme an N3 protonated A is suggested to play the same functional role (Ren et al. 2016; Wilson et al. 2016). A protonated C is used as the general acid in the HDV ribozyme (Gong et al. 2007). However, in all cases described so far, the shifted pKa is near neutrality (7.0), suggesting that at physiologically relevant pH values these RNAs exist in an equilibrium with significant populations of protonated and deprotonated species. These equilibria are most likely functionally important for switching between active and inactive conformations and/or by allowing the catalytic base of a ribozyme to function as the catalytic acid in the reverse reaction and vice versa. Recently, examples for nucleotides in stable RNA tertiary structures with very strongly shifted pKa values (>5 pH units) were reported where the protonation enables these nucleotides to function as stabilizing building blocks of such structural elements (Wilcox and Bevilacqua 2013b; Gottstein-Schmidtke et al. 2014; Wolter et al. 2017). What has not been reported so far at least to our knowledge, however, is an example where a protonated nucleotide plays a direct role in ligand binding and recognition by a functional RNA.

However, a putative example for such an RNA with a protonated nucleotide directly binding to a ligand is represented by the riboswitches binding to the cyclic dinucleotide 3′,3′-cyclic-GMP-AMP (c-GAMP, Fig. 1A). c-GAMP has been discovered recently as a regulatory cyclic dinucleotide in bacteria and not much is known about its associated signaling pathways (Davies et al. 2012). In Geobacter species it regulates the expression of numerous genes involved in exoelectrogenesis by binding to c-GAMP riboswitches (Kellenberger et al. 2015; Nelson et al. 2015). These riboswitches are surprisingly similar in terms of sequence, secondary and tertiary structure to a previously described riboswitch class binding to cyclic-di-GMP (Fig. 1B) called GEMM-I (Sudarsan et al. 2008). The c-GAMP binding riboswitches are therefore designated as members of the GEMM-Ib riboswitch class. A remarkable feature of the GEMM-I riboswitch aptamer domains is their generally high affinity for their cognate ligand c-di-GMP with KD values sometimes in the picomolar range (Sudarsan et al. 2008; Smith et al. 2009). The high affinity of these riboswitches is most likely the consequence of the rather low intracellular concentrations of c-di-GMP found to be in the high nanomolar to low micromolar range depending on the organism and growth conditions (Kader et al. 2006; Simm et al. 2009) and a kinetic control of their activity (Wickiser et al. 2005). Notably, at these low concentrations c-di-GMP does not yet form kinetically stable G-quadruplex-like and other oligomeric structures that would compete with its signaling functions (Gentner et al. 2012) despite the presence of high intracellular concentrations of potassium ions. The intracellular concentrations of c-GAMP have not yet been systematically quantified in different organisms and growth conditions. However, in Geobacter sulfurreducens c-GAMP can reach concentration levels comparable to those of c-di-GMP under certain conditions (Kellenberger et al. 2015).

FIGURE 1.

Ligand recognition by c-GAMP and c-di-GMP binding riboswitches. (A) Structure of 3′,3′-cyclic-GMP-AMP (c-GAMP, top) and secondary structure of a c-GAMP-binding riboswitch (bottom) from Geobacter sulfurreducens (Gs 1761, Ren et al. 2015). Nucleotides added to the terminus of the P1 stem in order to increase transcription efficiency and P1 stability in the NMR construct used in this study are boxed. The ligand is colored red. (B) Structure of c-di-GMP (top) and a c-di-GMP riboswitch from Vibrio cholerae (Vc2, Smith et al. 2009). Nucleotides added to the terminus of the P1 stem in order to increase transcription efficiency and an artificial stable UUCG tetraloop closing P3 in the NMR construct are boxed. The ligand is colored red. The position of the point mutation G20A that renders this riboswitch bispecific for c-GAMP and c-di-GMP is indicated by a shaded box. (C) Intermolecular base-pairing between the two ligand nucleotides Gα and Gβ and nucleotides of the c-di-GMP riboswitch as seen in the X-ray structure of the Vc2-RNA/c-di-GMP complex (PDB ID 3irw, Smith et al. 2009). (D) Intermolecular base-pairing between the two ligand nucleotides Aα and Gβ and nucleotides of the c-GAMP riboswitch as seen in the X-ray structure of the Gs1761-RNA/c-GAMP complex (PDB ID 4yaz, Ren et al. 2015). A14 here corresponds to G20 in the Vc2-riboswitch. (E) X-ray structure of the G20A-mutant of the Vc2 c-di-GMP-binding riboswitch bound to c-di-GMP (PDB ID 3mum, Smith et al. 2010). The ligand and riboswitch nucleotides involved directly in c-di-GMP binding are colored by atom type. The ligand-binding mode is shown as a close-up view. Nucleotides that recognize the ligand and the ligand itself are highlighted and shown as stick model. (F) Interaction between Gα of the ligand and nucleotide A20 as seen in E. In order to facilitate the description of possible hydrogen bonding interaction in this base pair the atom numbering for both the Ga and the A20 is shown below. (G) Two possible hydrogen-bonding patterns allowing the formation of two hydrogen bonds between Gα and A20. A20 could be protonated at N1 (top) with Gα adopting the standard keto tautomeric form or Gα could occur as the 6-enol tautomer with A20 in the neutral nonprotonated state (bottom).

GEMM-I riboswitches bind their ligand c-di-GMP in a bipartite binding site (Kulshina et al. 2009; Smith et al. 2009, 2010). The two guanine bases (designated Gα and Gβ) of the ligand are recognized differently (Fig. 1C). The Hoogsteen-edge of Gα is recognized by the Watson–Crick edge of G20 from the riboswitch (nucleotide numbering in the RNA corresponds to the Vibrio cholera Vc2 sequence presented in Smith et al. 2009). Two hydrogen bonds are formed between the C6 carbonyl group of Gα and the G20 amino group as well as between the N7 nitrogen of Gα and the G20 imino group. Gβ forms a standard Watson–Crick base pair with C92 of the riboswitch. In the c-GAMP-binding riboswitches G20 is replaced by an adenine nucleotide (A14 in the Geobacter sulfurreducens Gs1761 riboswitch, Fig. 1A). Aα of c-GAMP now binds to A14 of the GEMM-Ib riboswitch through a base-pairing interaction isosteric to the Gα:G base pair in the GEMM-I riboswitch (Fig. 1D, Ren et al. 2015, PDB ID 4yaz). The Aα amino group is hydrogen bonded to the N1 nitrogen of A14 and the Aα N7 nitrogen is hydrogen bonded to the A14 amino group (Fig. 1D). Surprisingly, the G20A mutation of the Vc2 GEMM-I riboswitch yields an RNA capable of binding to both c-di-GMP and to c-GAMP whereas the WT discriminates against c-GAMP (Smith et al. 2010; Kellenberger et al. 2013). Furthermore, a number of naturally occurring c-GAMP binding GEMM-Ib riboswitches are apparently bispecific for c-di-GMP and c-GAMP (Kellenberger et al. 2015; Nelson et al. 2015). X-ray structures of the Vc2 G20A mutant (Smith et al. 2010, PDB ID 3mum) and the Gs1761 c-GAMP riboswitch (Ren et al. 2015, PDB ID 4yb0) bound to c-di-GMP revealed a ligand recognition mode (Fig. 1E) with a hydrogen bond between the N7 nitrogen of Gα and the A20 amino group (Fig. 1F) similar to what was observed for c-GAMP binding (Fig. 1D). Furthermore, in both of these structures there is a short distance (2.8 Å in pdb 3mum, Smith et al. 2010 and 3.0 Å in pdb 4yb0, Ren et al. 2015) between the C6 carbonyl group of Gα and the N1 nitrogen of A20 (Fig. 1F). When assuming standard protonation patterns for the nucleotides, both of these functional groups are hydrogen bond acceptors carrying a negative partial charge. Therefore, the observed close contact should be energetically unfavorable. However, if as suggested previously (Smith et al. 2010; Ren et al. 2015) A20 is protonated at the N1 nitrogen, it could form a stabilizing hydrogen bonding interaction with the C6 carbonyl group of Gα (Fig. 1G, top). Alternatively, it was suggested (Ren et al. 2015) that Gα could adopt the enol tautomeric form with a hydroxyl group at C6. Then the C6 hydroxyl group could form a hydrogen bond to the N1 nitrogen of A20 (Fig. 1G, bottom). Both possibilities would be equally exciting because neither the direct participation of a protonated nucleotide in ligand binding in an RNA nor the stable induction of a rare tautomeric state for a standard nucleotide in an RNA structure have been demonstrated unequivocally. However, unusual tautomeric states in G:U mismatches are adopted transiently in low populations, as recently demonstrated by NMR spectroscopy (Kimsey et al. 2015). In addition, X-ray structures of near-cognate tRNAs bound to the decoding center of the ribosome and of mismatch-containing duplexes suggested that they play a role in miscoding (Demeshkina et al. 2012; Rozov et al. 2015, 2016; Rypniewski et al. 2016). However, X-ray structures with a resolution >1 Å do not allow the direct observation of hydrogen positions and thereby the unambiguous assignment of hydrogen bonding patterns in cases where the positions of the heavy atoms allow alternative hydrogen bonding patterns. For the complexes of c-di-GMP bound to Vc2 G20A and Gs1761 the resolution is 2.9 Å (pdb 3mum) and 2.1 Å (pdb 4yb0), respectively. On the other hand, solution NMR spectroscopy is a very useful tool for the direct unambiguous elucidation of hydrogen bonding patterns in nucleic acids (Dingley and Grzesiek 1998; Wöhnert et al. 1999; Duchardt-Ferner et al. 2011; Duchardt-Ferner and Wöhnert 2017). In particular, 13C and 15N chemical shifts are faithful reporters of changes in protonation states (Legault and Pardi 1994, 1997). In favorable cases, NMR signals for protons at the protonation site can be detected directly (e.g., Macaya et al. 1991; Brodsky et al. 1998; Nixon et al. 2002; Cash et al. 2013; Gottstein-Schmidtke et al. 2014; Wolter et al. 2017). Some of the NMR methods for the identification of protonation events are applicable even for larger RNAs in cases where initial structural information from other methods is already available as in the case of the c-GAMP riboswitches and their complexes with c-di-GMP and c-GAMP.

Here we set out to delineate the hydrogen bonding patterns between c-di-GMP and the c-GAMP riboswitches by NMR. We find that an N1 protonated adenine nucleotide of the riboswitch forms a hydrogen bond to the guanine base (Gα) of the ligand in its standard neutral imino tautomeric state. The adenine N1 protonation is induced by c-di-GMP binding. The protonated state of the riboswitch in the c-di-GMP complex is persistent even at pH values above 8.3 and most likely limits the specificity of the c-GAMP-binding riboswitches by allowing stable binding of c-di-GMP.

RESULTS AND DISCUSSION

The GEMM riboswitch variants with sizes of >80 nucleotides and their complex tertiary structures are challenging for NMR studies. In particular, NMR signal assignment strategies using the standard heteronuclear, multidimensional NMR approaches are no longer applicable for RNAs of this size with highly complex tertiary structures that are not amenable to “divide and conquer” approaches. Thus, our NMR studies had to rely on the available structural information from X-ray structures, the comparison of the NMR properties of different variants of the same riboswitch and chemical shift comparison in conjunction with base-type selective labeling. A particularly well-suited starting point for our investigations was therefore the G20A-mutant of the c-di-GMP binding GEMM-I riboswitch from V. cholerae (Vc2, Fig. 1B). There are X-ray structures available for the WT-Vc2 riboswitch bound to c-di-GMP (PDB ID 3mxh, 2.3 Å) as well as for the G20A-mutant bound to c-di-GMP (PDB ID 3mum, 2.9 Å) and c-GAMP (PDB ID 4yb1, 2.1 Å) revealing very similar overall structures (Smith et al. 2010; Ren et al. 2015). Therefore, the WT Vc2 riboswitch bound to c-di-GMP and the G20A-mutant bound to c-GAMP serve as reference states. Since they only differ by a single nucleotide from the G20A-mutant bound to c-di-GMP, they allow a meaningful comparison of their NMR spectra.

We initially characterized c-di-GMP and c-GAMP binding to the Vc2 WT riboswitch and the G20A mutant by 1D imino proton NMR spectra and ITC (Fig. 2). It should be noted that all NMR and ITC measurements reported in this study were carried out in buffers (25 or 50 mM BisTris pH 6.5, 5 mM magnesium acetate if not explicitly noted otherwise) completely lacking potassium ions. 1H- and 31P-NMR experiments showed that in this buffer c-di-GMP does not form G-quadruplex or oligomeric structures (Supplemental Fig. S1) at the c-di-GMP concentrations needed for ITC or NMR experiments. In contrast, the presence of 25 mM potassium phosphate or 25 mM potassium phosphate and 250 mM potassium chloride induced large changes in the NMR-spectra of free c-di-GMP (Supplemental Fig. S1). These changes indicate the formation of G-quadruplex and other oligomeric structures at the c-di-GMP concentrations required for the NMR and ITC experiments (>100 µM) in agreement with previous reports (Gentner et al. 2012). The formation of these structures would directly compete with c-di-GMP binding to RNA and thereby masking the effects of c-di-GMP binding in our experiments. Under in vivo conditions, however, these oligomeric c-di-GMP structures are apparently not relevant since the measured intracellular c-di-GMP concentrations are well below 100 µM (Kader et al. 2006; Simm et al. 2009). At such low concentrations of c-di-GMP no quadruplex formation and/or oligomerization was observed even in the presence of potassium ions (Gentner et al. 2012).

FIGURE 2.

Ligand binding to the Vc2-WT and the Vc2-G20A-mutant riboswitch. (A) Comparison of the imino proton spectra of the Vc2-WT (left) and the Vc2-G20A-mutant (right) riboswitch RNA in the ligand-free state (top), in the presence of 5 mM Mg2+ (middle), in the presence of 5 mM Mg2+ and c-di-GMP (middle) and in the presence of 5 mM Mg2+ and c-GAMP (bottom). Shaded areas highlight chemical shift regions with pronounced spectral changes. Note that the imino proton spectra of the G20A-mutant in the presence of c-di-GMP and c-GAMP are very similar to each other and to the spectra of the WT bound to c-di-GMP. In contrast, in the spectrum of the WT-riboswitch bound to c-GAMP imino proton signals are missing compared to those of the c-di-GMP complex and the mutant (red arrows) suggestive of incomplete folding. (B) ITC thermogram and fit for c-GAMP binding to the WT Vc2 riboswitch showing a KD of only 10 µM. c-di-GMP is bound with a KD in the picomolar range (Smith et al. 2009, 2010). (C) ITC thermograms and fits for c-GAMP (left) and c-di-GMP (right) binding to the G20A-mutant of the Vc2 riboswitch. C-GAMP is bound with a KD of 120 nM while c-di-GMP is bound with a KD of 53 nM. All ITC-experiments shown here were carried out in a buffer containing 50 mM Bis-Tris pH 6.5 and 5 mM magnesium acetate at 25°C.

Both the Vc2 WT and the G20A RNA showed significant changes in their imino proton NMR spectra upon ligand addition in the presence of 5 mM Mg2+. The appearance of novel imino proton signals upon ligand addition suggested ligand-induced folding of the RNA in all cases. Importantly, the spectral changes observed upon addition of c-di-GMP or c-GAMP to the G20A-mutant are similar to the spectral changes observed for the WT-RNA upon addition of c-di-GMP (Fig. 2A) in agreement with a very similar ligand binding mode in all three cases. In contrast, the spectra of the WT-RNA in the presence of c-GAMP show a reduced number of imino proton signals compared to the other complexes in line with incomplete folding of the RNA upon binding c-GAMP. The WT-RNA binds c-di-GMP with a picomolar KD (Sudarsan et al. 2008; Smith et al. 2009, 2010). ITC measurements with c-GAMP showed that the affinity of the WT-RNA for this ligand is much lower, revealing a KD of 10 µM (Fig. 2B). In contrast, the G20A-mutant binds tightly to both c-di-GMP and c-GAMP with KD values of 53 and 120 nM, respectively (Fig. 2B; Supplemental Table 1). Thus, as expected based on previous results, the G20A-mutant is bispecific and binds both ligands with similar affinity and a similar overall binding mode.

A very sensitive and well-established reporter for a putative protonation of adenines is their C2 chemical shift (Legault and Pardi 1994, 1997). The C2 chemical shifts for adenines are normally found in a range between 152 and 157 ppm. Upon protonation the C2 chemical shifts are observed ∼8–10 ppm upfield at 146–147 ppm. Thus, if c-di-GMP binding induces A20 protonation in the Vc2-G20A mutant, an H2C2 signal with an upfield C2 chemical shift should be found in a 13C-HSQC spectrum that would be absent in the spectra of both the c-di-GMP bound WT-RNA as well as the c-GAMP bound G20A-mutant RNA. In order to reduce spectral crowding 13C-HSQC-spectra were recorded for 13C,15N-adenine labeled RNAs and unlabeled ligand. Thus, all other nucleotides of the RNA and the ligand are spectroscopically silent. A comparison of the 13C-HSQC-spectra of 13C-adenine labeled G20A RNA in its free form and bound to c-di-GMP (Fig. 3A) reveals massive spectral differences. In particular, signal dispersion in the presence of c-di-GMP increases significantly in line with the expected ligand-induced folding of the RNA. Importantly, there is one signal present in the spectrum of the G20A-RNA/c-di-GMP complex with a chemical shift of 147.4 ppm in the 13C-dimension and 8.3 ppm in the 1H-dimension, respectively, which is not present in the free RNA and is a candidate for the H2C2 signal of a protonated adenine. In the 13C-HSQC-spectrum of the WT RNA containing a G at position 20 bound to c-di-GMP this signal is absent (Fig. 3B). This signal is also absent in the 13C-HSQC spectrum of the G20A mutant RNA bound to c-GAMP where the nonprotonated nucleotide A20 forms hydrogen bonds to the Hoogsteen-edge of Aα of the ligand (Fig. 3C). A second signal that is only found in the 13C-HSQC spectrum of the G20A-RNA/c-di-GMP complex but not in those of the other two complexes appears at a chemical shift of 7.5 ppm (1H) and ∼142 ppm (13C). Comparison with spectra of other RNAs containing protonated adenines (Legault and Pardi 1997; Wolter et al. 2017) suggests that this signal might correspond to the C8 of a protonated adenine. Thus, the comparison of the 13C-HSQC spectra for the G20A-RNA/c-di-GMP complex with those of the WT-RNA bound to the same ligand and the G20A-RNA bound to c-GAMP shows that the former contains a protonated adenine not occurring in the latter two complexes. Due to the sequential and structural similarities of the three systems, A20 is the only logical candidate for the protonated adenine. Importantly, the signal corresponding to the protonated A20 in the G20A-RNA/c-di-GMP complex is still observable with similar intensity and at the same position in 13C-HSQC spectra recorded at pH 8.3, suggesting a stable protonation even at elevated pH (Fig. 3D). Unfortunately, the absence of chemical shift, line widths or intensity changes for the A20 H2C2 resonance upon increasing the pH from 6.5 to 8.3 prevented us from determining the pKa for A20 protonation in the G20A-RNA/c-di-GMP complex by pH titrations. We avoided increasing pH values above pH 8.3 since under these conditions G and U nucleotides of the RNA, which are not in Watson–Crick base pairs, would start to become deprotonated to a significant amount (pKa ∼ 9.2). Since such nucleotides are found in the direct vicinity of the binding site (Smith et al. 2009; Ren et al. 2015), chemical shift changes potentially observable at higher pH values could no longer be attributed to A20 deprotonation alone. Thus, we can only conclude that the pKa for A20 deprotonation in the complex must be significantly larger than 8.3.

FIGURE 3.

Evidence for a protonated A20 in the complex of the G20A-mutant Vc2 riboswitch bound to c-di-GMP. (A) Overlay of 13C-HSQC spectra of 13C,15N-adenine labeled ligand-free G20A RNA (gray) and G20A RNA bound to c-di-GMP (black). Large spectral changes are observed as expected for ligand-induced RNA-folding. Importantly, a signal is observed with a 13C chemical shift of 147.4 ppm and a 1H chemical shift of 8.3 ppm, respectively (red circle)—the chemical shift range associated with C2 carbon nuclei of protonated adenine nucleotides. The typical chemical shift ranges for adenine carbon nuclei are indicated by bars on the left side of the spectrum. (B) Overlay of the 13C-HSQC-spectra of the G20A-mutant (black) and the WT-RNA (gray) both bound to c-di-GMP. The H2C2 signal indicative of adenine protonation occurs only in the G20A-mutant. (C) Overlay of the 13C-HSQC-spectra of the G20A-mutant RNA bound to c-di-GMP (black) or c-GAMP (gray). The H2C2 signal indicative of adenine protonation occurs only in the c-di-GMP complex. (D) 13C-HSQC spectrum recorded for the G20A-mutant bound to c-di-GMP recorded at pH 8.3. The H2C2 signal typical for the protonated adenine unique to this complex is still present at high intensity (red circle) and has the same 1H and 13C chemical shifts as in the spectrum recorded at pH 6.5.

The imino protons of protonated adenines are normally not directly observable at physiologically relevant pH due to fast exchange with the bulk solvent and/or rapid opening of base-pairing interactions. The only instances of NMR observable adenine imino protons in nucleic acids to our knowledge were reported under conditions of either low pH or for RNAs where the protonated nucleotide was an integral part of the tertiary structure (Macaya et al. 1991; Wolter et al. 2017). A 15N-HSQC-spectrum recorded at 10°C with 15N-adenine labeled G20A-mutant RNA bound to c-di-GMP showed a single adenine imino group resonance with chemical shifts of ∼13.7 (1H) and ∼152.1 (15N) ppm, respectively (Fig. 4A). This imino resonance corresponds to the imino group of the protonated A20 that serves as a hydrogen bond donor to the C6 carbonyl group of the Gα of the bound c-di-GMP. As expected, it is not observable in the G20A-RNA/c-GAMP complex (Fig. 4B) since in the c-GAMP complex the A20 adenine N1 nitrogen is not protonated and serves as a hydrogen bond acceptor for the Aα amino group of the c-GAMP ligand (Fig. 1D). In a two-bond 15N-HSQC spectrum the H2-proton at 8.3 ppm corresponding to the protonated A20 is connected to its N3 and the N1 nitrogen chemical shifts (Fig. 4C). The N1 chemical shift connected to the H2 proton of A20 is the same as the one observed in the standard 15N-HSQC and ∼60 ppm upfield from the N1 nitrogen chemical shifts of the unprotonated adenine nucleotides as expected. Furthermore, temperature-dependent 15N-HSQC-spectra show that the imino group signal is observable at temperatures up to 25°C (Fig. 4D). Thus, the protonation of A20 at N1 in the complex with c-di-GMP is very stable.

FIGURE 4.

A20 is stably protonated at the N1 nitrogen in the G20A-mutant riboswitch in complex with c-di-GMP. (A) A 2D-15N-HSQC spectrum recorded at 10°C for a sample containing 15N-adenine labeled G20A-RNA bound to unlabeled c-di-GMP shows a single imino group signal. (B) A comparison of 1D-15N-HSQC spectra at 10°C recorded with 256 scans in the c-di-GMP complex and 1024 scans in the c-GAMP complex show that while this imino proton signal is present in the c-di-GMP complex it is absent in the c-GAMP complex of the G20A mutant RNA. (C) A two-bond 2D-15N-HSQC spectrum links the H2 proton of A20 at 8.3 ppm to the N1 and N3 nitrogens of the same nucleotide. An inset shows the magnetization transfer pathway for this experiment. The A20 N1 nitrogen chemical shift of 152.3 ppm is ∼60 ppm upfield compared to the N1 chemical shifts for unprotonated adenines and is the same as the one observed for the imino group nitrogen in the standard 15N-HSQC experiment. Vertical bars on the right side of the spectrum indicate the typical 15N chemical shift ranges for different types of nitrogen nuclei in adenine nucleotides. (D) The imino proton signal of A20+ is observable up to 25°C in 1D-15N-HSQC-spectra at pH 6.5 suggesting a stable protonation.

In a recent example of an RNA containing a G:A+ interaction with a geometry resembling the one of the Gα:A20 interaction, hydrogen bonding between the G C6 carbonyl group and the N1 imino group of the protonated A lead to a pronounced downfield shift of the guanine C6 carbon compared to all other guanine C6 resonances (Wolter et al. 2017). In order to measure the C6 chemical shifts of c-di-GMP bound to either WT-RNA or the G20A-mutant, we enzymatically prepared and purified 13C-labeled c-di-GMP and measured 13C-1D-spectra of the labeled ligand bound to both RNAs in their unlabeled form (Fig. 5A). The C6 carbon of the c-di-GMP Gα bound to the G20A-mutant which interacts with the protonated A20 is shifted downfield compared to the Gα C6 when bound to the WT RNA, where it interacts with the neutral G20 (Figs. 5B, 1G,C). On the other hand, the Gβ C6 chemical shift is very similar in both complexes, in agreement with a similar binding mode for Gβ in both complexes.

FIGURE 5.

Adenine A20 N1 protonation induces a downfield shift of the Gα C6 carbon resonance. (A) Comparison of 1D-13C spectra of 13C,15N-labeled c-di-GMP alone (top) and in complex with Vc2 (middle) or Vc2 G20A mutant RNA (bottom). The bars on top indicate typical chemical shift ranges for guanine carbon nuclei. The spectral region typical for C5 nuclei in guanines (115–120 ppm) is not shown. (B) Zoom of the spectral region that is shaded in A to highlight the effect of the A20 protonation on the ligand C6 chemical shifts. Signals for c-di-GMP C6 carbon nuclei appear as doublets due to the C-C scalar coupling with the C5 carbons (88 Hz). Adenine N1 protonation in the Vc2-G20A mutant leads to a downfield shift of the c-di-GMP Gα C6 chemical shift compared to the Vc2-RNA (red) whereas the c-di-GMP Gβ C6 chemical shift is similar in both complexes (black). Natural abundance 13C-signals from the RNA are highlighted with an asterisk.

The X-ray structures of the Vc2 and Vc2 G20A-complexes with their ligands revealed that Mg2+-ions played an integral role for RNA ligand interactions. In particular, both phosphate groups of the ligand are coordinated by Mg2+-ions. The closest distance between a bound Mg2+ and the A20 protonation site is 7.4 Å. Furthermore, comparison of the NMR-spectra of the free RNA in the absence and the presence of Mg2+ revealed differences in the imino proton region that are suggestive of a certain degree of Mg2+-induced structural preorganization of the RNA. In other RNAs containing protonated nucleotides, it was demonstrated that protonation and Mg2+-binding are anti-cooperative (Huppler et al. 2002; Wilcox and Bevilacqua 2013b). In contrast, in the HDV ribozyme the simultaneous protonation of C41 in a base triple and the binding of a structural Mg2+ ion cooperatively promote ribozyme folding and cleavage activity (Nakano and Bevilacqua 2007). In order to further characterize the interplay between A20 protonation and Mg2+-binding in the Vc2 G20A riboswitch/c-di-GMP complex, we compared 1H-NMR spectra and measured the ligand affinity at pH 6.5 and 8.3 at different Mg2+ concentrations (Supplemental Fig. S2; Supplemental Table 2). At both pH values no ligand binding and no ligand-induced RNA folding was observable when the Mg2+ concentration was below ∼2 mM. At both pH values an increased Mg2+ concentration enhanced the affinity of the riboswitch for c-di-GMP. However, the c-di-GMP affinity is always higher at pH 6.5 than at pH 8.3. Thus, both a low pH favoring A20 protonation as well as high Mg2+-concentrations favoring Mg2+-binding contribute positively to c-di-GMP binding and RNA-folding in this system.

We next investigated binding of both c-GAMP and c-di-GMP to the Gs1761 riboswitch from Geobacter sulfurreducens (Kellenberger et al. 2015; Nelson et al. 2015). This riboswitch was described as being highly selective for c-GAMP. Based on in-line-probing experiments, Hammond and coworkers (Kellenberger et al. 2015) reported KDs of this riboswitch for c-GAMP and c-di-GMP of 0.53 and 660 nM, respectively. Using ITC, Patel and coworkers (Ren et al. 2015) measured KDs of 70 nM and 930 nM for c-GAMP and c-di-GMP, respectively, in a buffer containing 50 mM potassium acetate, pH 6.8, 100 mM KCl and 20 mM MgCl2 at an elevated temperature of 35°C probably in order to reduce c-di-GMP oligomerization. However, Patel and coworkers used an RNA sequence that accidentally differed by two point mutations from the WT-RNA (U72C, C73U) for X-ray structure determination and ITC-experiments (Fig. 1A; Supplemental Fig. S3). We initially used the same RNA sequence as Patel and coworkers with an additionally stabilized P1 stem (named Gs1761 throughout this paper) in our experiments (Fig. 1A). As expected from the earlier findings, our NMR titration experiments using 1D-1H-imino proton spectra showed that both ligands bound to the Gs1761 riboswitch in the presence of Mg2+ and caused very similar spectral changes (Fig. 6A). Overall, the general quality of the proton spectra for the Gs1761 riboswitch complexes with both ligands is lower compared to the Vc2-derived RNAs. ITC-experiments with our Gs1761 construct under buffer conditions chosen to suppress intermolecular G-quadruplex formation by c-di-GMP (50 mM Bis-Tris, pH 6.5, 5 mM Mg2+ acetate at 25°C) showed that Gs1761 bound c-GAMP only approximately threefold tighter than c-di-GMP with KD values of 290 nM and 710 nM, respectively (Fig. 6B). Restoring the two point mutations to the wild-type sequence (Gs1761 WT) did not change the affinities of the riboswitch for c-GAMP (236 nM) and c-di-GMP (589 nM) significantly (Supplemental Fig. S3; Supplemental Table 1). When Bis-Tris is replaced by 50 mM potassium phosphate in a buffer (pH 6.5) containing 5 mM Mg2+ at 25°C the Gs1761 riboswitch binds c-GAMP with a KD of 302 nM (Supplemental Fig. S4), which is very similar to the KD we measured in Bis-Tris buffer (292 nM). However, no binding of c-di-GMP to the RNA is detectable by ITC at 25°C (Supplemental Fig. S4), most likely due to the formation of competing G-quadruplex and oligomeric structures by c-di-GMP (Supplemental Fig. S1). Adding 250 mM potassium chloride to this buffer significantly enhances the affinity of the Gs1761 RNA for c-GAMP (KD = 33 nM, Supplemental Fig. S4) in agreement with the results of Patel and coworkers (Ren et al. 2015) but binding to c-di-GMP is again not detectable by ITC at 25°C (Supplemental Fig. S4).

FIGURE 6.

Ligand binding to the Gs1761 riboswitch and evidence for adenine protonation in the complex with c-di-GMP. (A) Imino proton spectra of the Gs1761 riboswitch RNA in the ligand-free state (top), in the presence of Mg2+ (middle), in the presence of Mg2+ and c-di-GMP (middle) and in the presence of Mg2+ and c-GAMP (bottom). Chemical shift regions with pronounced spectral changes are shaded in gray. (B) Representative ITC thermograms and fits for Gs1761 riboswitch RNA binding to c-di-GMP (left) and c-GAMP (right), respectively. (C) Overlay of 13C-HSQC spectra of 13C,15N-adenine labeled Gs1761 riboswitch RNA in its ligand-free state (gray) or bound to c-di-GMP (black). RNA-folding is induced by ligand binding as observed by large spectral changes. Interestingly, one signal appears with a 13C chemical shift of 147.7 ppm and a 1H chemical shift of 8.46 ppm (red circle)—the chemical shift range associated with C2 carbon nuclei of N1 protonated adenine nucleotides. (D) Overlay of 13C-HSQC spectra of 13C,15N-adenine labeled Gs1761 riboswitch RNA bound to c-di-GMP (black) or c-GAMP (gray). The signal indicative of the adenine protonation is absent in the c-GAMP bound complex.

The comparison of 13C-HSQC spectra for the 13C,15N-adenine labeled Gs1761 RNA either in its free form or in the presence of c-di-GMP showed the appearance of only one signal with the chemical shift characteristics for a H2C2-group of a protonated adenine with chemical shifts of ∼147.7 ppm (13C) and 8.46 ppm (1H, Fig. 6C). This signal is absent when the Gs1761 RNA is titrated with its cognate ligand c-GAMP (Fig. 6D). Thus, the appearance of this signal in the c-di-GMP complex therefore reports on the formation of the base-pairing interaction between Gα of the ligand and the protonated A14 (equivalent to A20 in the Vc2 G20A-mutant RNA) of the riboswitch.

Finally, we tested a c-GAMP riboswitch aptamer domain from Clostridium beijerinckii (Cbe 1–2) (Fig. 7A) for c-GAMP and c-di-GMP binding that was described as being bispecific for both ligands (Nelson et al. 2015). In agreement with the previous observations for the other c-GAMP binding riboswitch variants, both c-GAMP and c-di-GMP induced significant changes in the imino proton spectra of the RNA indicative of stable binding (Fig. 7B). ITC-experiments show that this RNA binds c-di-GMP with a lower KD (30 nM) than c-GAMP (463 nM) in 50 mM Bis-Tris buffer, pH 6.5, 5 mM Mg2+ acetate at 25°C (Fig. 7C). The comparison of 13C-HSQC spectra for the 13C,15N-adenine labeled Cbe 1–2 RNA either in its free form or in the presence of c-di-GMP again showed the appearance of one signal with the chemical shift characteristics for a H2C2-group of a protonated adenine (Fig. 7D). In analogy to the observations with the other riboswitches, this signal is absent when the Cbe 1–2 RNA is titrated with c-GAMP (Fig. 7E). Thus, Cbe 1–2 binds c-di-GMP via a protonated adenine residue as well. Protonation of the adenine nucleotide can therefore be regarded as a common feature for c-di-GMP binding in artificial riboswitch variants (Vc2 G20A), naturally occurring c-GAMP riboswitches with reportedly high specificity (Gs 1761) and those previously reported to be bispecific (Cbe 1–2). Thus, in these systems adenine protonation does not contribute to but probably rather limits the maximally achievable ligand selectivity since it supports c-di-GMP binding to c-GAMP riboswitches by allowing the formation of two hydrogen bonds between Ga and the relevant riboswitch adenine residue. The significant differences in ligand affinities measured for the Gs1761 constructs with a natural P1 stem used by Hammond and coworkers (Kellenberger et al. 2015) and for the Gs1761 variants used in this work and by Patel's group (Ren et al. 2015) with a strongly stabilized P1 stem suggest that the composition and stability of the P1 stem might contribute to ligand specificity by promoting c-GAMP binding. Furthermore, high potassium ion concentrations apparently also support high affinity c-GAMP binding. However, a more complete analysis of the structural determinants for ligand specificity in this riboswitch class is beyond the scope of this work.

FIGURE 7.

Ligand binding to the Clostridium beijerinckii (Cbe 1–2) riboswitch and evidence for adenine protonation in the complex with c-di-GMP. (A) Secondary structure of the Cbe 1–2 riboswitch aptamer domain. (B) Imino proton spectra of the Cbe 1–2 riboswitch RNA in the ligand-free state (top), in the presence of Mg2+ (middle), in the presence of Mg2+ and c-di-GMP (middle) and in the presence of Mg2+ and c-GAMP (bottom). (C) Representative ITC thermograms and fits for Cbe 1–2 riboswitch RNA binding to c-di-GMP (left) and c-GAMP (right), respectively. The resulting KD values are given for both complexes. (D) Overlay of 13C-HSQC spectra of 13C15N-adenine labeled Cbe 1–2 riboswitch RNA in its ligand-free state (gray) or bound to c-di-GMP (black). The signal at a 13C chemical shift of 147.3 ppm and a 1H chemical shift of 8.29 ppm (red circle) appears in the chemical shift range that is associated with C2 carbon nuclei of N1 protonated adenine nucleotides. (E) Overlay of 13C-HSQC spectra of 13C,15N–adenine labeled Cbe 1–2 riboswitch RNA bound to c-di-GMP (black) or c-GAMP (gray). The signal indicative of the adenine protonation is present only in the c-di-GMP bound complex.

Overall, the experiments described here clarify the hydrogen bonding patterns between the GEMM-Ib riboswitches with their near-cognate ligand c-di-GMP. Interatomic distances between heavy atoms observed in X-ray crystallographic studies suggested that either a protonated A nucleotide of the riboswitch forms two hydrogen bonds with Gα of the ligand in its canonical protonation state or that a standard A forms hydrogen bonds to the Gα of the ligand assuming a rare tautomeric state. NMR spectroscopy unambiguously shows a protonated A of the riboswitch when bound to c-di-GMP and thereby establishes the c-di-GMP/GEMM-Ib complex as the first example for an RNA where a protonated nucleotide is directly involved in ligand binding. Remarkably, the N1 protonated state of the A in this complex is persistent even at pH values well above neutrality. Thus, environmental pH changes apparently do not contribute to the modulation of ligand specificity of the GEMM-Ib riboswitch. Instead, adenine protonation might be an important contribution to the widespread bispecific binding behavior for c-GAMP and c-di-GMP observed for members of this riboswitch class.

MATERIALS AND METHODS

Templates for in vitro transcription

We used PCR generated double-stranded DNA fragments or linearized plasmid DNA as template for T7 in vitro transcription. Overlapping oligonucleotides encoding a modified version of the c-di-GMP sensing riboswitch from Vibrio cholera (Vc2, Smith et al. 2009) and a mutant of this riboswitch harboring a guanine to adenine mutation at position 20 (Vc2 G20A) were used to generate double-stranded DNA fragments using PCR. pUC18-derived plasmids encoding the template sequences for Vc2, Vc2 G20A and modified versions of the cGAMP sensing riboswitches from Geobacter sulfurreducens (Gs1761, Ren et al. 2015) and Clostridium beijerinckii (Cbe 1–2, Nelson et al. 2015) were generated by Gibson cloning (Gibson et al. 2009).

RNA preparation

All RNAs were synthesized by run-off in vitro transcription using T7 RNA polymerase as previously described (Stoldt et al. 1998). Unlabeled, 15N-adenine or 13C,15N-adenine labeled Vc2, Vc2 G20A, Gs1761, Gs1761WT as well as Cbe 1–2 were transcribed using SmaI linearized plasmid DNA or PCR generated double-stranded DNA fragments as templates and purified by denaturing PAGE as previously described (Duchardt-Ferner et al. 2010). 13C,15N- and 15N-labeled rNTPs were purchased commercially (Silantes).

RNAs were folded under conditions favoring monomeric hairpin structures in a low salt buffer (2.5 mM Bis-Tris, pH 6.5) by denaturing at 70°C for 5 min, rapid 10-fold dilution with the same ice-cold buffer and subsequent annealing on ice. They were exchanged into 25 mM Bis-Tris buffer, pH 6.5, 5 mM magnesium acetate for NMR spectroscopy using ultracentrifugation devices (VivaSpin 2, MWC 3 kDa) and multiple cycles of concentration and dilution with NMR buffer. The RNAs in the final NMR samples were monomeric and homogeneous as judged from a native gel (Supplemental Fig. S5).

Unlabeled c-di-GMP and c-GAMP were purchased commercially as sodium salts (Sigma Aldrich) and dissolved in NMR buffer.

Synthesis and purification of 13C,15N-labeled c-di-GMP

13C,15N-labeled c-di-GMP was enzymatically synthesized as described in Rao et al. (2009) using the stand-alone GGEEF domain of the Thermotoga maritima diguanylate cyclase (TM1788) harboring the R158A mutation, except that we used 13C,15N-labeled GTP as substrate for synthesis. Reactions were run at 45°C in TM1788 reaction buffer (50 mM Tris/HCl, pH 7.5, 250 mM NaCl, 20 mM MgCl2, 1 mM DTT). The progress of the reaction was monitored by 31P NMR and indicated complete conversion of GTP to c-di-GMP (Supplemental Fig. S6).

Magnesium pyrophosphate that precipitated during synthesis was separated by centrifugation. 13C,15N-labeled c-di-GMP was purified from the reaction by phenol–chloroform extraction of the supernatant followed by ethanol precipitation of the aqueous phase. The pellet was dissolved in water and the concentration was determined spectroscopically using 26,100 M−1 cm−1 as the extinction coefficient (ε260) for c-di-GMP. 1D-1H, 1D-13C, 1D-31P, and 2D-13C-HSQC-spectra of the product showed no impurities (Fig. 5A, top; Supplemental Fig. S6) and most importantly the absence of any remaining GTP. The molecular mass of the product was confirmed by high-resolution MALDI mass spectrometry.

NMR spectroscopy

NMR experiments were recorded on 600 MHz Bruker AV, 600 MHz Bruker AVIII-HD, and 950 MHz Bruker AVII NMR spectrometers, all equipped with 5 mm, z-axis gradient 1H[13C, 15N]-TCI cryogenic probes using standard pulse sequences (Fürtig et al. 2003). 1D imino proton spectra were recorded at 10°C using jump-return water suppression whereas 13C-HSQC-spectra were recorded at 25°C. 13C-1D spectra were recorded on an 800 MHz Bruker AV NMR spectrometer equipped with a 5 mm, z-axis gradient 13C,15N[1H]-TXO cryogenic probe at 25°C. Unless stated otherwise, samples were measured in 25 mM Bis-Tris (pH 6.5), 5 mM magnesium acetate containing 7.5% D2O. Spectra were processed and analyzed in TOPSPIN 3.2 (Bruker Biospin). RNA concentrations ranged from 100 µM for unlabeled RNA, 300 µM for 13C,15N-adenine labeled RNA to 900 µM for 15N-adenine labeled RNA. Samples with isotopically labeled RNA contained 1.2 equivalents of unlabeled c-di-GMP. The sample used for recording the 1D-13C-spectrum of 13C,15N-labeled c-di-GMP bound to the unlabeled G20A-riboswitch contained 400 µM c-di-GMP and 1.1 equivalents of RNA.

Isothermal titration calorimetry (ITC)

Unlabeled RNAs and both ligands (c-di-GMP, c-GAMP) were prepared in 50 mM Bis-Tris (pH 6.5) containing 5 mM magnesium acetate, if not indicated otherwise. Additional experiments to test for the influence of potassium ions on the binding affinities were carried out in buffer containing 25 mM potassium phosphate buffer, pH 6.5 and 5 mM magnesium acetate with or without 250 mM potassium chloride, respectively. ITC experiments at pH 8.3 were carried out in 50 mM Tris/HCl buffer containing 0, 2, 5, or 10 mM Mg2+. The ligand (200–450 µM) was injected into a solution of 20–45 µM RNA. All measurements were performed at 25°C using a MicroCal iTC200 instrument (Malvern Instruments). After an initial waiting time of 120 sec, the first injection of 0.2 µL was followed by 19 serial injections of 2 µL, separated by intervals of 180–1440 sec. For each experiment, the reference power was set to 11 µcal−1, stirring speed to 750 rpm and the high feedback mode was selected for experiments with c-di-GMP, while for c-GAMP experiments the low feedback mode was chosen. Three independent titrations were performed and the reported KD values are the average from these titrations. The thermograms were processed using Origin7.0 (OriginLab) assuming a one-site binding model. An overview of all binding parameters derived from the ITC measurements is given in Supplemental Tables 1 and 2.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.