|Home | About | Journals | Submit | Contact Us | Français|
The ZTP riboswitch is a widespread family of regulatory RNAs that upregulate de novo purine synthesis in response to increased intracellular levels of ZTP or ZMP (AICAR). As an important intermediate in purine biosynthesis, ZMP also serves as a proxy for the concentration of 10-formyltetrahydrofolate, a key component of one carbon metabolism. Here we report the structure of the ZTP riboswitch bound to ZMP at a resolution of 1.80 Å. The RNA contains two subdomains brought together through a long-range pseudoknot further stabilized through helix-helix packing. ZMP is bound at the subdomain interface of the RNA through a set of interactions with the ligand's base, ribose sugar and phosphate moieties. Unique to nucleobase recognition by RNAs, the Z base is inner sphere coordinated to a magnesium cation bound by two backbone phosphates. This interaction, along with steric hindrance by the backbone, imparts specificity over related analogs such as ATP/AMP.
One-carbon metabolism is an essential component of the biosynthesis of almost all macromolecules in normal cellular homeostasis and whose hyperactivation is increasingly appreciated as a major factor in oncogenesis (Amelio et al., 2014; Locasale, 2013). The central cofactor in these diverse biochemical processes is tetrahydrofolate (THF), a carrier of one-carbon units acquired from the glycine cleavage system. These carbon units are either directly transferred to substrates by the appropriate THF derivative or transferred to other one-carbon carriers such as vitamin B12 or S-adenosylmethionine (SAM). Regulation of the THF pool in the cell is critical for maintaining flow of biosynthetic intermediates through a number of pathways and the reduced folate pool is an important indicator of cellular nutrient status (Amelio et al., 2014; Locasale, 2013). One critical pathway part of this integrated network is purine biosynthesis, in which one of the last steps requires transfer of a formyl group from THF to 5-aminoimidazole-4-carboxamide ribonucleotide (AICAR or ZMP; Figure 1A) to form 5-formamidoimidazole-4-carboxamide (FAICAR) that is transformed into inosine monophosphate (Zhang et al., 2008). As this is the only obligate THF-requiring step in purine biosynthesis, the buildup of ZMP serves as a potentially important proxy for depletion of 10-formyl-THF. ZTP was hypothesized to be an alarmone signal for deficiency of this one carbon carrier (Bochner and Ames, 1982), however until recently, there has not been an identifiable receptor for ZMP/ZTP that directly senses and regulates the intracellular 10-formylTHF pool (Ducker and Rabinowitz, 2015).
In many bacteria, a variety of aspects of metabolism are genetically regulated through an RNA-based element called a riboswitch. These elements, generally found in the 5’-leader of mRNAs, directly sense a specific small molecule through a highly structured aptamer domain, which typically communicates its occupancy status to a downstream secondary structural switch that instructs the expression machinery (reviewed in (Garst et al., 2011; Roth and Breaker, 2009)). Many riboswitches are observed regulating aspects of one-carbon metabolism. Regulation of S-adenosylmethionine biosynthesis can be achieved through one of at least six distinct classes of SAM binding RNAs, as well as the buildup of its toxic product, S-adenosylhomocysteine (Batey, 2011). A riboswitch that productively binds methylcobalamin has also been identified, but it is likely monitoring the total intracellular cobalamin pool rather than the fraction dedicated to one-carbon transfer (Johnson et al., 2012; Weinberg et al., 2010). Finally, the intracellular reduced folate pool can be monitored by the THF riboswitch (Ames et al., 2010). However, this RNA does not discriminate between the various forms of reduced folic acid, and thus cannot sense depletion of activated THF from the general reduced folate pool (Trausch and Batey, 2014; Trausch et al., 2011). Thus, many of these riboswitches, while sensing one-carbon carriers, do not obtain a global perspective on the status of carbon flux through the system as would be achieved by a sensor of ZMP/ZTP buildup due to insufficient 10-formylTHF (Ducker and Rabinowitz, 2015; Kim et al., 2015).
Recently, the Breaker laboratory identified and validated a riboswitch capable of monitoring the intracellular 10-formylTHF pool via ZMP/ZTP concentrations (Kim et al., 2015). Originally identified as the “pfl” motif in a comparative genomics-based approach to find structured noncoding RNA sequences, this motif generally precedes genes involved in purine and formylTHF biosynthesis (Weinberg et al., 2010). Despite this strong indicator of possible effectors, this motif resisted initial attempts to validate its ligand (Meyer et al., 2011). A subsequent study that surveyed a series of pfl riboswitch candidates revealed a subset of sequences able to bind ZTP, ZMP and Z ribonucleoside (Kim et al., 2015). This RNA, renamed the ZTP riboswitch, discriminates by at least 1000-fold against related compounds, including inosine monophosphate (IMP), which differs from ZMP by a single carbon unit, and AMP. This is important as adenosine and guanosine nucleotides have high intracellular concentrations and could therefore be potent competitors of Z nucleotides.
To understand the structural basis for Z recognition and its discrimination against chemically related nucleosides and nucleotides, we solved the crystal structure of a ZTP riboswitch in complex with ZMP. The two subdomains of this RNA (P1/P2 and P3), separated by a flexible linker region, pack together via the predicted pseudoknot. Conserved residues outside of the pseudoknot form further tertiary interactions that mediate helical packing. This formation of these tertiary interactions creates a binding pocket where ZMP is specifically recognized through hydrogen bonding and van der Waals interactions. Unique to nucleobase recognition by RNA, the carboxamide oxygen on the Z base makes an inner-sphere coordination with a magnesium ion held in place by two backbone phosphates and is likely essential for discrimination against chemically similar metabolites.
RNA sequences were chosen from the pfl Rfam alignment (accession RF01750) (Griffiths-Jones et al., 2005; Nawrocki et al., 2015), and screened for the ability to bind ZMP and crystallize. Structural analysis focused on an aptamer from Actinomyces odontolyticus (Aod), a bacterium associated with human bacteremia (Figure S1) (Cone et al., 2003). This switch regulates expression of serine hydroxylmethyltransferase, a gene responsible for generating 5,10-methyleneTHF. To facilitate crystallization, we altered the sequence in L2 from the wild type GCCA tetraloop to GAAA. This is a conservative mutation as only the first purine and last adenosine in L2 are phylogenetically conserved, and GNRA type tetraloops are observed in a number of sequences in the pfl alignment (Kim et al., 2015; Weinberg et al., 2010). In addition, the 11 nucleotide J-P1/P3 linker was truncated to two uridines. An in-line probing assay revealed this linker is highly susceptible to cleavage in both the absence and presence of ligand, indicating this sequence is conformationally flexible and not directly involved in ligand binding (Kim et al., 2015). As the length of P1 is not conserved, this sequence was also shortened by two base pairs. Finally, two unpaired adenosines were added to the 3’-end of the RNA, which has been used in the past to facilitate crystallization of other riboswitch aptamers (Reyes et al., 2009). The sequences of the wild type and crystallized RNAs are given in Supplemental Table 1.
To validate that these changes do not significantly alter the affinity of the RNA for ZMP, binding was tested by isothermal titration calorimetry (ITC). The wild type Aod sequence (Aod(wt)) binds ZMP in this assay, but with a stoichiometry consistent with the majority of the RNA misfolded (Table 1). A previous report indicated that a number of sequences identified as pfl riboswitches were found to be inactive--proposed to be due the inability of a subset of sequences to efficiently refold under in vitro transcription conditions, as monitored by in-line probing (Kim et al., 2015). Consistent with this observation are the binding characteristics of other sequences that we tested, including those from Clostridium bartlettii (Cba), Fusobacterium varium (Fva) and Dorea longicatenaI (Dlo) whose titrations indicate an appreciable fraction of the RNA is misfolded (Table 1). While the RNA sequence used for crystallization (Aod(xtal)) displayed a binding affinity similar to other sequences tested, it notably had an observed stoichiometry consistent with all of the RNA folded and active, assuming a 1:1 ZMP:RNA binding ratio.
Crystals of the aptamer domain of the Aod pfl riboswitch aptamer domain (Aod(xtal)) were grown in the presence of ZMP that diffracted x-rays to 2.5 Å resolution. The phase problem was solved using data collected on a rotating copper anode home x-ray source using a combination of anomalous diffraction from bound iridium hexammine (Keel et al., 2007) and molecular replacement with three model helices (Robertson et al., 2010). An initial model was built and refined, which was used for refinement against a dataset extending to 1.8 Å resolution obtained using a synchrotron x-ray source (x-ray data and refinement statistics are in Table 2). The final model, including all residues of the RNA and a single molecule of ZMP had good final geometry with an Rxtal of 21.4% and Rfree of 24.9%.
The observed secondary structure of the RNA is consistent with that predicted from covariation analysis, including the P1-P3 helices and a pseudoknot (Figure 1B) (Kim et al., 2015; Meyer et al., 2011; Weinberg et al., 2010). However, the observed pseudoknot is intermolecular, formed through base pairing between nucleotides in J1/2 and L3 within two neighboring RNA molecules in the crystal. Within each monomer, the two uridines of J-P1/P3 form base pairs with the two adenosines added to the 3’-end to create a small helix that promotes coaxial stacking of the P1 and P3 helices (Figure S2A). Thus, the RNA forms an elongated structure with the P2 stem-loop at one end and P3 stem-loop at the other end. In the crystal, the monomer forms a head-to-tail dimer with another molecule (Figure S2B). This interaction is further supported by additional long-range interactions between adjacent molecules mediated by conserved nucleotides within P1 and P3.
Several lines of evidence strongly support that one half of the head-to-tail dimer reflects the structure of the biologically relevant monomer. The biological unit includes the P1-P2 region from one strand and P3 from the second (Figure 1C and D, Figure S2C). This type of dimer has been observed in the CrPV IRES (Pfingsten et al., 2006) and an engineered dimer of the P4-P6 tetraloop-tetraloop receptor (Davis et al., 2005); in each case the head-to-tail half reflects the biologically relevant structure. A single molecule of ZMP is observed in the proposed biological monomer, sitting between P1-P2 and P3 elements. Notably, the tertiary architecture is reconcilable with previously published structural probing data (Kim et al., 2015) as ZMP dependent protections cluster around the binding pocket and tertiary contacts (Figure 2A). Highly conserved nucleotides are clustered around the ligand binding site and key tertiary interactions that hold the structure together, further supporting the proposed biological monomer (Figure 2B) (Kim et al., 2015; Weinberg et al., 2010). All further discussion of the ZMP riboswitch will be in the light of the biological monomer.
The ZMP riboswitch adopts an overall architecture dictated by side-by-side organization of the P1-P2 and P3 subdomains (Figure 1). The P1 helix is extended by formation of four additional nonWatson-Crick and Watson-Crick base pairs not predicted by covariation analysis, which coaxially stacks upon P2. The P3 helix is positioned parallel to the P1-P2 coaxial stack, which is primarily enforced by the formation of a pseudoknot helix (PK) between nucleotides in the 5’-side joining strand between P1 and P2 (J1/2) and the loop of P3 (L3). This creates an infrequently observed HLIN type pseudoknot (Han, 2003). A single molecule of ZMP is observed in the core of this structure with most interactions mediated by insertion between nucleotides in L3 supported by additional interactions with the ribose-phosphate backbone of P2.
In the crystal structure, the 3’-end of the P1 helix contains an additional unanticipated five base pairs, many of which play structural roles in mediating P1-P3 packing (Kim et al., 2015). The first two pairs are tandem “sheared” sugar-edge/Hoogsteen edge G•A base pairs that form a well-established tandem G•A module (G8•A38, A9•G37) (Cruz and Westhof, 2011; Gautheret et al., 1994). Notably, Cruz and Westhof correctly predicted the presence of this module in pfl using a simple descriptor of this module comprising the tandem G•A pairs flanked by Watson-Crick pairs in a survey of Rfam alignments (Cruz and Westhof, 2011). The most conserved nucleotides in this region of the RNA form a C10-G36 base pair followed by a wobble U11-G35 base pair. Mediating the P1-P2 coaxial stack is an unusual one hydrogen bond sugar edge-sugar edge A19•G34 pair.
The tandem G•A module mediates extensive side-by-side helical packing interactions with P3 (Figure 3A). The adenines of each G•A pair (A9 and A38) interact with the minor groove of the C49-G61 and C50-G60 base pairs in P3 (Figure 3B). A38 forms an imperfect “type I” A-minor triple with the C49-G61 base pair (Nissen et al., 2001) while the A9 forms a highly oblique minor triple with the two G-C pairs, similar to that observed in the glmS ribozyme (Cochrane et al., 2007; Klein and Ferré-D'Amaré, 2006). A further set of interactions is formed by the highly conserved C10-G36 base pair to the G51-G59 pair that forms part of the ZMP binding pocket (see below). This packing arrangement is similar to that observed for other A-minor triple mediated interactions such as that mediated by a loop E module in the 5S rRNA that docks with the 23S rRNA (Nissen et al., 2001) and intermolecular docking in crystals of small RNA duplexes harboring the tandem G•A module (Jang et al., 2004). Thus, this structure suggests that many of the examples of conserved tandem G•A modules found by Cruz and Westfhof in diverse ncRNA families serve to mediating helix-helix packing.
The second element of interaction between the two subdomains is a pseudoknot between nucleotides in J1/2 (nt 12-15,17) and P3 (nt 53-57). Sequence and covariation analysis of the ZMP riboswitch predicts that this interaction is made up of four to five contiguous bases in J1/2 and L3 to form a Watson-Crick base paired helix (Kim et al., 2015; Weinberg et al., 2010). The Aod aptamer domain's pseudoknot contains the expected four contiguous Watson-Crick base pairs between nucleotides 12-15 and 54-57 (Figure 1B). A fifth unusual pair is formed by extrusion of A16 from the helix and the sugar edge of A17 pairing with the Watson-Crick face of C53; this helix is capped by stacking of the unpaired C18 with A17. In line probing of three representative members of the pfl family by the Breaker group as a function of ZMP concentration reveals a substantial degree of protection of almost all nucleotides in the pseudoknot as well as those in the G•A module (Kim et al., 2015). Coupled with the structure, these data indicate that ZMP binding stabilizes intersubdomain interactions, a key component of the regulatory switch (Kim et al., 2015).
An unexpected feature of this RNA is interactions between L2 and the core of the RNA. Within the ZMP class of riboswitches, the length of P2 is highly conserved at either four or five base pairs capped by either a terminal loop or interrupted by an internal loop followed by another helix (P2b) of variable length (Kim et al., 2015; Weinberg et al., 2010). The 5’ and 3’ nucleotides of the loop are conserved as a purine and adenosine, respectively (>90%). In the RNA sequence used for crystallization, the wild type GCCA loop was altered to GAAA—a conservative change considering the poor conservation of sequence between these purines and that the GAAA tetraloop is observed in biological sequences. In the crystal structure, the GAAA tetraloop is observed in a highly unusual conformation. The canonical GAAA loop conformation is such that the first and fourth nucleotides form a sheared G•A pair, while the second and third adenosines are stacked upon the forth (Heus and Pardi, 1991). This conformation is found both in the isolated tetraloop as well as engaged with a wide variety of docked states (Cate et al., 1996; Pley et al., 1994). However, in the ZMP riboswitch, the first two nucleotides (G25 and A26) are stacked with the closing G24-C29 base pair, while the other two are splayed out from the loop and stacked on one another (Figure 4A). The fourth adenosine (A28) forms a type-I A-minor triple interaction (Nissen et al., 2001) with the G12-C57 Waston-Crick pair of the pseudoknot. This interaction, and its proximity to the binding pocket, justifies the previously unclear conservation of this nucleotide. While this disruption of a GAAA tetraloop architecture is unprecedented amongst known GAAA interactions with RNA, it is has been observed in complex with proteins, such as in the complex between the sarcin homolog restrictocin bound to RNAs that mimic the sarcin/ricin loop of rat 23S rRNA (Figure 4B) (Yang et al., 2001). This interaction expands the means by which GAAA tetraloops mediate tertiary RNA-RNA interactions.
The binding site for ZMP is primarily composed of four nucleotides in the P3 loop. As discussed above, the central five nucleotides of L3 are involved in base-base interactions with J1/2 to form the pseudoknot while the other four (two on each on the 5’- and the 3’-sides of L3) remain free of helical elements to form the core of the binding pocket (G51, C52, U58, and G59). Importantly, three of these nucleotides (G51, U58 and G59) are nearly invariant in the pfl family sequence alignment (Kim et al., 2015; Nawrocki et al., 2015; Weinberg et al., 2010). The base of ZMP, 5-aminoimidazole-4-carboxamide, forms a two hydrogen bond pairing interaction with U58, which is structurally analogous to an A•U Hoogsteen edge-Watson-Crick edge pair (Figure 5A). The Z base is stacked between the highly conserved G12 and G59 residues. Further Z base specific interactions are between its 5-amino group and N7 of G51 in L3 and O5’ of U11 in P1. O5’ also provides the basis for discrimination against chemically related IMP and AMP, which differ from ZMP sterically by the presence of C2. Superimposition of adenine on ZMP reveals that O5’ clashes with C2 (Figure S3), such that the purine rings of AMP and IMP would be excluded by the binding pocket, in agreement with the ~1000-fold selectivity for ZMP/ZTP over AMP and >10,000-fold selectivity against IMP observed by the Breaker group (Kim et al., 2015).
The ribosyl moiety of ZMP is recognized through a combination of hydrogen bonding interactions and van der Waals packing with nucleotides in L3. Both hydroxyl groups are engaged in hydrogen bonding interactions with the RNA: the 2’-OH of ZMP interacts with N7 of G12 of the first base pair in the pseudoknot and the 3’-OH of ZMP interacts with the 2’-OH of G51 in L3 (Figure 5A, B). The ribose sugar further forms van der s interactions with G51, which is in the atypical syn conformation. This interaction is supported by a one-hydrogen bonding interaction between G51 and G59 in a highly buckled pairing arrangement.
The ZTP riboswitch binds Z bases containing a phosphate (ZMP/ZTP) with higher affinity than the ribonucleoside alone (Kim et al., 2015). Interactions between the phosphate group of ZMP and RNA are mediated by a magnesium ion that forms a single inner sphere coordination with O2 of C52 (Figure 5A, B). Across phylogeny, the identity of this nucleotide is variable, indicating that recruitment of this ion is not achieved identically in all ZMP variants. Based upon this structure, it could be predicted that a purine at this position could move the magnesium away form the sugar and the α-phosphate to better recognize the β- and γ-phosphates of ZTP but this hypothesis is countered by previous reports of constructs with an A (C. bartlettii) or U (H. arsenicoxydans) at this position show no discrimination between ZMP and ZTP (Kim et al., 2015). Instead, members of the ZTP riboswitch family most likely only interrogate for the presence of the α-phosphate. A second direct interaction occurs between one of the phosphate oxygens and N6 of C57 of the first base pair of the pseudoknot. The identity of this base pair is >97% conserved across phylogeny and so this interaction is most probably the primary means of phosphate recognition by the ZMP riboswitch.
The most unanticipated aspect of ZMP recognition is metal mediated recognition of the oxygen of the 4-carboxamide group of the Z base. Positioned 2.2 Å away from this oxygen is a Mg2+ that forms two other inner sphere coordinations to non-bridging phosphate oxygens of U11 and C29. The octahedral coordination geometry and interatomic distances (2.1-2.2 Å) are consistent with the identity of this metal ion as a specifically bound magnesium. To investigate the role of divalent cations in ZMP recognition, we measured the binding of ZMP and Z-riboside to the riboswitch aptamer in the presence of a series of divalent ions (Figure 5C and 5D; Supplemental Table 2). While both ZMP and Z-riboside bound with low micromolar affinity to the aptamer in the presence of magnesium, the only other divalent cation supporting binding is manganese. ZMP displayed a 53-fold lower affinity for the RNA in Mn2+ as compared to magnesium, while Z-riboside showed an ~12-fold difference. Since Z-riboside does not have a 5’-phosphate, the ~12-fold difference is ascribed to the divalent ion adjacent to the 4-carboxamide group. The additional ~4-fold effect observed in ZMP is likely due to the divalent interacting with the phosphate group. The moderate effect by the magnesium adjacent to the phosphate is consistent with the phosphate group's weak contribution to overall binding affinity to the RNA. Further, this metal ion is only weakly chelated in comparison to magnesium ions observed adjacent to the phosphate groups of thiamine pyrophosphate (Edwards and Ferre-D'Amare, 2006; Serganov et al., 2006; Thore et al., 2006) and flavin mononucleotide (Serganov et al., 2009) in complex with their riboswitches. Thus the magnesium ion adjacent to the phosphate group of ZMP likely behaves more as diffusely bound cation rather than as specifically bound. To ensure that the effect of these metals was due to the specific sites within the binding pocket and not due to RNA misfolding, these conditions were also surveyed using native gel electrophoresis mobility shift assay (EMSA). All divalent cations amenable to the EMSA assay promoted the same shift in mobility due to RNA structure formation and dimerization as Mg2+ in the absence of ZMP, while only the magnesium conditions saw an additional mobility supershift in the presence of ZMP (Figure S3). Together, these data reveal a specifically bound magnesium is an essential constituent of the binding pocket.
The structure of the ZMP/ZTP binding riboswitch significantly furthers understanding of interactions of cellular nucleotides with RNA by revealing several novel aspects of recognition of these compounds. Over the past decade, a number of structures of riboswitch aptamers in complex with nucleobases (guanine, adenine and pre-Q1), nucleosides (2’-deoxyguanosine) and derivatives (e.g., S-adenosylmethionine, coenzyme B12, and cyclic di-nucleotide second messengers) have yielded an increasingly clear picture of how RNA recognizes these compounds (reviewed in (Peselis and Serganov, 2014)). Invariably, the nucleobase or nucleotide moiety is recognized as part of a base pair or triple that is further stabilized through π-stacking interactions between the ligand and RNA ((Batey, 2012)). The structure of the ZTP riboswitch, the first of a nucleotide-selective RNA, reveals novel features of recognition. Unlike the “ATP” aptamer (Sassanfar and Szostak, 1993), which appears to show little selectivity for AMP over adenosine and whose structures do not reveal how the RNA would specifically recognize the α-phosphate group (Dieckmann et al., 1996; Jiang et al., 1996), the ZTP riboswitch has a clear preference for nucleotides over the corresponding nucleoside or nucleobase (Kim et al., 2015), which is supported by the crystal structure. Like other riboswitches that recognize phosphate groups such as TPP and FMN, the ZMP riboswitch uses a bound divalent cation to bridge the phosphate group and RNA, as well as a single direct contact that is likely non-essential.
Unique to nucleobase recognition, the ZMP riboswitch uses a metal ion to promote high affinity and specificity of binding. The magnesium ion is held in place by inner sphere coordinations to two backbone phosphates, positioning the cation to form an inner sphere coordinate bond with the electronegative carboxamide oxygen of the Z-base (Figure 4). The most analogous means of ligand recognition is found within the SAM-I and SAM-II riboswitches in which the positively charged sulfonium group of SAM is recognized by carbonyl groups in the major groove of uridine bases, and in the case of SAM-I has been shown to impart selectivity for SAM over S-adenosylhomocysteine (SAH) (Batey, 2011). While the direct Z-base interaction with magnesium is unique to known modes of nucleobase recognition by RNA, it is common for RNAs to use carbonyl groups in the major groove to recruit metal ions or as part of metal ion binding pockets (Auffinger et al., 2011).
The tertiary structure of the binding pocket further sheds light on the ZMP analog screening done by the Breaker group (Kim et al., 2015). ZTP, ZMP, and cyclic-ZMP all bind with high affinity, consistent with the phosphate(s) directed away from the RNA and into solution. Loss of the α-phosphate results in a moderate loss in affinity, while further loss of the ribose (Z-base only) yields a significant (10-fold) reduction in affinity. Other analogs disrupting the hydrogen bonding around the Z base are not tolerated. This makes it that much more surprising that any affinity is observed with ATP and other adenine derivatives. It is possible that coordination of the carboxyl oxygen in ZMP is substituted by coordinating with N1 of the adenine base. Superimposition of adenine into the binding pocket reveals a clash between C2 of adenine and the non-bridging phosphate of U11. A static interpretation of the structure is incompatible with ATP binding but a subtle structural shift could be possible. There is no evidence that ATP is an effector of the ZMP riboswitch. Together, these data from the Breaker group further support the observed mode of ZMP binding in the crystal structure.
Along these lines, the ZTP riboswitch may be a promising target of for the development of antibiotics. De novo purine synthesis is important for production of DNA and RNA precursors making this biosynthetic pathway a key pharmacological target to reduce the growth of rapidly growing cells (Christopherson et al., 2002; Legraverend and Grierson, 2006), and has been successfully targeted with antibiotic and anticancer compounds (Gonen and Assaraf, 2012; Shoaib Ahmad Shah et al., 2013). A small molecule that prevents the two subdomains (P1/P2 and P3) from productively associating might inhibit expression of enzymes required for continued rapid growth such as formate-tetrahydrofolate ligase. The structural understanding presented here represents the starting point for computational design and screening of molecular inhibitors.
Another emerging application of riboswitch aptamers is the in vivo sensing of small molecules. For example, the SAM-I riboswitch aptamer domain was attached to a fluorophore-binding module to create a robust sensor of SAM in E. coli, revealing substantial heterogeneity in the intracellular concentration of this important metabolite (Paige et al., 2012). In addition, the use of similar sensors using aptamers that bind second messenger signaling molecules have enabled the discovery of new biochemical pathways (Kellenberger et al., 2015). While it is difficult to conceive of how the monomeric ZTP riboswitch can be easily attached to modular fluorophore-binding aptamers to create a genetically encodable sensor of ZMP/ZTP, the observed crystallographic dimer may provide a solution to this problem. Circular permutation of a linked dimer would enable attachment through a variant that contains a stem-loop in J1/2, such as observed in the C. bartlettii ZTP riboswitch variant. Given the importance of one-carbon metabolism in oncogenesis, such a sensor could be a valuable tool for monitoring this process or redox state associated with THF pools.
Riboswitches are a major means of controlling gene expression in bacteria by exploiting a number of regulatory mechanisms including transcriptional and translational regulation, mRNA degradation and synthesis of antisense RNAs. Currently, over 25 validated classes of riboswitches have been identified and validated that specifically bind effectors spanning a broad spectrum of small molecules from ions to protein cofactors. Beyond serving as an important model system for understanding RNA structure and function, riboswitches are being increasingly explored as potential targets of antimicrobial therapeutics, exploited as biosensors of small molecules in the cellular environment and as parts of synthetic biological devices. The pfl motif, which specifically was recently discovered to recognize ZMP or ZTP, represents a widely dispersed and ancient class of riboswitches. This study reveals the first structure of an RNA that selectively recognizes a nucleotide and yields new insights into how recurrent motifs such as the tandem G•A pair and GNRA tetraloop are used to create higher-order RNA architecture. Most interestingly, selectivity for the Z-base over adenine or hypoxanthine is in part achieved through metal-mediated recognition, unprecedented amongst known RNA-nucleobase interactions. The structure of the ZTP-riboswitch can serve as the basis for structure-guided small molecule targeting of this riboswitch as well as design of novel ZMP/ZTP biosensors.
DNA templates for transcription of RNAs used for biochemical and structural experiments were created using recursive PCR (Prodromou and Pearl, 1992). To minimize nontemplated addition of nucleotides at the 3’-end of the transcript by T7 RNA polymerase, the template contained 2’-O-methyl groups on the two 5’ bases on the reverse strand (Kao et al., 1999). RNA was transcribed and purified using previously published methods (Reyes et al., 2009). Briefly, T7 RNA polymerase was used to transcribe RNA for 2 hours at 37 °C in a reaction containing 1.25 mL of 10X transcription buffer (300 mM Tris-HCl pH 8.0, 100 mM DTT, 5 mg/mL spermidine, 0.1% Triton X-100), 32 mM MgCl2, 4 mM each NTP, 8 mM DTT, 1.2 mL DNA template from PCR (~1 μM stock template), 4 units of inorganic pyrophosphatase, 50 μL 10 mg/mL T7 RNA polymerase for a total reaction volume of 12.5 mL. The reaction was ethanol precipitated, RNA pelleted by centrifugation, solubilized in 0.5X T.E. buffer (5 mM Tris-HCl, pH 8.0, 0.5 mM EDTA), and electrophoresed on a 12 % denaturing (8 M urea) 29:1 polyacrylamide gel. The band containing product RNA was identified by UV shadowing, the gel crushed, and passively eluted into 0.5X T.E. buffer. The solution was concentrated and buffer exchanged into 0.5X T.E. buffer. This stock of RNA is stored at concentrations between 0.5 and 1 mM at −20 °C until use.
RNA was crystallized using the hanging drop vapor diffusion method. An RNA solution (300 μM RNA, 1 mM ZMP (Sigma-Aldrich)) was heated to 65 °C for two minutes and immediately cooled on ice for ten minutes. One microliter was added to a siliconized glass slide and mixed with 1 μL of mother liquor (29-31 % PEG 8K, 150-170 mM NH4OAc, 5 mM Mg(OAc)2, 50 mM Na-cacodylate pH 7.0). The drop was equilibrated against 500 μL mother liquor at 30 °C. Elongated bipyramidal cry stals formed within three days reaching a maximal size of 0.4 m. For phase determination, the crystals were soaked for 10-30 minutes in mother liquor supplemented with 30 mM iridium hexamine. Crystals were scooped and flash frozen in liquid nitrogen. Upon mounting, many crystals saw an improvement in spot shape and resolution using a crystal annealing protocol in which the cryostream was blocked for 7 seconds and refreezing through rapidly reintroducing the cryostream (Harp et al., 1999; Yeh and Hol, 1998).
Initial screening and data collection was performed with a Rigaku R-Axis IV image plate system using CuKα radiation. The resulting diffraction data, extending to 2.5 Å resolution was indexed, integrated, and scaled using iMoslfm (Battye et al., 2011). These data revealed modest anomalous signal from bound iridium but efforts to find heavy atom sites using SOLVE as implemented PHENIX (McCoy et al., 2007) were unsuccessful. To supplement the anomalous signal we employed methods of molecular replacement for de novo phasing developed by the Scott lab (Robertson et al., 2010). The A-form helices of the RNA predicted from covariation analysis (P1, P2/L2, P3 and PK) were modeled with MC-Sym (Parisien and Major, 2008). Phaser-MR was used to independently search and place models corresponding to P1, P2/L2, and PK with translation function Z-score (TFZ) of 6.4. Although P1 and PK displayed clear electron density, P2 did not and was removed from the search model. It was discovered later that the placement of PK was incorrect and corresponded to P2. For the next round of molecular replacement, the search model of P1 and PK was locked and a search for P2 without L2 done, resulting in a TFZ score of 5.5. Although density around each of the placed helices was clear, it failed to pull out any unmodeled density and was unsuitable for solving the phase problem alone. The resulting three helices were combining into a single search model and used to bootstrap traditional SAD phasing. Phaser-EP was now successful in finding four iridium hexamine sites with a figure of merit of 0.359. The solution was exported to PHENIX AutoBuild where the majority of the molecule was built. At this stage, the density was sufficiently clear to manually build and assign every base as well as the ZMP within Coot (Emsley and Cowtan, 2004). The model was eventually refined using phenix.refine to a Rwork and Rfree of 0.25 and 0.27, respectively.
A higher resolution data set from a different crystal was collected at the Advanced Light Source (Beamline 5.0.2). Some radiation damage was observed after ~110 frames at a 1 second shot time. Despite this, a high quality data set extending to 1.8 Å resolution was collected. The test set of the previous dataset was matched to this dataset and extended to the highest resolution before using the final model derived from home source data was used in molecular replacement against the synchrotron data. Refinement of the model against the synchrotron data enabled a better solvent model, particularly with respect to magnesium ions, and resulted in an Rwork and Rfree of 0.21 and 0.25, respectively. Electron density maps are presented in Figure S5. The atomic coordinates and associated structure factors are deposited in the RSCB Protein Data Bank under accession codes 4XW7 and 4XWF.
RNA used for ITC was dialyzed overnight at 4 °C against a buffer containing 10 mM K-HEPES pH 8.0, 135 mM KCl, 15 mM NaCl, and 10 mM divalent chloride salt. The concentration of RNA was determined by UV spectroscopy with an extinction coefficient calculated using Integrated DNA Technologies Oligo Analyzer tool. ZMP and Z-ribonucleoside (Sigma-Aldrich) were dissolved in dialysis buffer and the concentration was determined similarly using an extinction coefficient of 12,600 M−1cm−1 at 269 nm. The concentration of RNA for each experiment was chosen to achieve an optimal c-value (Turnbull and Daranas, 2003). For most titrations this was approximately 100 μM; the ligand concentration was 10-fold higher than RNA concentration to achieve saturation. Each titration consisted of 21 injections where the first was 0.2 μL and the rest of which were 2.0 μL (Gilbert and Batey, 2009). The titrations were integrated and fit using Origins 7.0 (MicroCal) using a single-site binding equation. Example thermograms are presented in Figure S5. Data presented represents the average and standard deviation of three independent titrations.
For each lane, 50 μL containing 5 μM RNA, 0 or 1 mM ZMP, 10 mM divalent chloride salt or 1 mM EDTA, and 20% (v/v) loading dye (30% (v/v) glycerol, 0.25 (w/v) bromophenol blue, 0.25% (w/v) xylene cyanol FF) was incubated at 4 °C for 30 min. An 8% native 29:1 polyacrylamide gel containing 1X TBE and 10 mM divalent chloride salt was run in a buffer containing 1X TBE and 10 mM divalent chloride salt. The gel that did not contain a divalent ion contained 1 mM EDTA pH 8.0 in both the gel and running buffer. Gels were loaded with 10 μL of sample and run at 250 volts while maintaining a temperature of 4 °C. RNA was resolved using ethidium bromide. MnCl2 and CoCl2 were not amenable to this assay as they undergo redox reactions under an electric potential.
The ZTP riboswitch aptamer folds into a compact pseudoknot
A GAAA tetraloop docks with the core aptamer disrupting its canonical structure
The ZTP nucleobase directly interacts with a magnesium ion bound by the RNA
The structure explains weak selectivity for ZTP/ZMP over Z-riboside
This work was supported by a grant from the National Institutes of Health (R01 GM073850) to R.T.B. We would like to thank David McKay for his support with aspects of the crystallographic analysis and to Jacob Polaski for fruitful discussions.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
J.J.T. and R.T.B. conceived of the project, designed experiments and wrote the manuscript. J.G.M. and M.M.M. preformed some calorimetric experiments and screening of RNA-ZMP complexes in crystallization trials. J.J.T. performed all of the crystallographic analysis and EMSA assays, along with some calorimetric assays. All authors reviewed the paper.
Supplemental Information includes five figures and two tables and can be found with this article at http://XXXX
Competing financial interests
The authors declare no completing financial interests.