Cooperation between Magnesium and Metabolite Controls Collapse of the SAM-I Riboswitch

Abstract

The S-adenosylmethionine (SAM)-I riboswitch is a noncoding RNA that regulates the transcription termination process in response to metabolite (SAM) binding. The aptamer portion of the riboswitch may adopt an open or closed state depending on the presence of metabolite. Although the transition between the open and closed states is critical for the switching process, its atomistic details are not well understood. Using atomistic simulations, we calculate the effect of SAM and magnesium ions on the folding free energy landscape of the SAM-I riboswitch. These molecular simulation results are consistent with our previous wetlab experiments and aid in interpreting the SHAPE probing measurements. Here, molecular dynamics simulations explicitly identify target RNA motifs sensitive to magnesium ions and SAM. In the simulations, we observe that, whereas the metabolite mostly stabilizes the P1 and P3 helices, magnesium serves an important role in stabilizing a pseudoknot interaction between the P2 and P4 helices, even at high metabolite concentrations. The pseudoknot stabilization by magnesium, in combination with P1 stabilization by SAM, explains the requirement of both SAM and magnesium to form the fully collapsed metabolite-bound closed state of the SAM-I riboswitch. In the absence of SAM, frequent open-to-closed conformational transitions of the pseudoknot occur, akin to breathing. These pseudoknot fluctuations disrupt the binding site by facilitating fluctuations in the 5′-end of helix P1. Magnesium biases the landscape toward a collapsed state (preorganization) by coordinating pseudoknot and 5′-P1 fluctuations. The cooperation between SAM and magnesium in stabilizing important tertiary interactions elucidates their functional significance in transcription regulation.

Introduction

The discovery of noncoding RNAs and their versatile functions has revolutionized the modern concept of gene regulation and many other cellular processes. Their functional impact ranges from the level of transcription, RNA processing, RNA stability, translation initiation, and genome reorganization to the regulatory control of pathogenicity (1, 2, 3). The S-adenosylmethionine (SAM)-I riboswitch is a noncoding RNA element, located in the 5′-untranslated region of messenger RNA in Gram-positive bacteria (4). It binds to small metabolites, such as SAM, and terminates the transcription of several genes that regulate Methionine and Sulfur biosynthesis. SAM-I contains two partially overlapping domains: the aptamer and the expression platform. These two domains compete for a common RNA segment which, in the aptamer, forms the 3′ side of the helix P1 and P4 stems in the presence of metabolite (4).

SAM-I is the first discovered riboswitch in the SAM-family and has gained the most attention so far (3, 5, 6). Since the x-ray crystal structure of the SAM-bound aptamer domain has been solved, a new direction of investigation of metabolite sensing has emerged. Yet, little is known about how the metabolite is able to control the aptamer domain of such riboswitches. In the aptamer domain, P1, P4 and P2, P3 are pairs of coaxially stacked helices linked by a fourway junction (4, 7, 8). Apart from these helices, there are two well-characterized tertiary interactions, a kink turn and a pseudoknot that support these coaxial stacks (Fig. 1, A and B). The diverse secondary and tertiary structural components make this riboswitch system and its control over the gene transcription process an interesting target for investigation.

Figure 1.

Crystal structure of metabolite-bound SAM-I riboswitch and its SHAPE reactivity analysis. (A) Given here is the secondary structure and (B) (i) tertiary structure of SAM-I riboswitch in metabolite (SAM)-bound state (PDB: 2GIS). Note the pseudoknot (PK) and the kink-turn (KT) tertiary motifs in the structure. (ii) Shown here is a zoomed-in view of the binding site. One can observe the triple-base interactions and other hydrogen-bonded interactions with U-shaped SAM (with red colored base). (C) Given here is a comparison between normalized SHAPE reactivity calculations from our structure-based GEM simulations of the apo-form of SAM-I, without Mg²⁺ ions and in the presence of 2.0 mM Mg²⁺. (D) Shown here is a comparison between normalized SHAPE reactivity measurements of the metabolite-bound state of SAM-I, without Mg²⁺ ions and in the presence of 2.0 mM Mg²⁺. (E) The SHAPE reactivity differences between 0 and 2 mM Mg²⁺ are plotted for the apo-form of SAM-I and for the metabolite-bound form of SAM-I. The difference for each nucleotide helps to identify the most Mg²⁺-affected regions. Here PK@P2 represents part of the pseudoknot region (corresponding to nucleotide residues 25–28), which is close to P2-helix and PK@P4 represents another part of the pseudoknot (corresponding to nucleotide residues 65–68), which is close to P4-helix.

RNA tertiary structures of the SAM-I aptamer and their implications for the global RNA fold have been studied by x-ray crystallography, NMR, single molecule spectroscopy, chemical genetics, phylogenetic analysis, and different computational approaches (9, 10, 11, 12). In the crystal structure (Fig. 1), determined from the Thermoanaerobacter tengcongensis riboswitch, one can observe that the close juxtaposition of the P1 and P3 helices makes a tight binding site for the SAM metabolite (4). A zoomed-in view of the binding site is shown in Fig. 1. The binding core is further stabilized by additional interactions, such as an extensive hydrogen bonding network near the binding site and other SAM-mediated interactions. The recent crystal structure of the aptamer domain from Bacillus subtilis yitJ revealed a set of tertiary interactions that zip up the binding site in a SAM-independent fashion (8). Beyond structural differences in the peripheral region of these two SAM-I riboswitches, their binding sites are quite similar. The pocket is zipped up by a set of conserved ribose-zipper interactions between nucleotide (nt) positions 112–114 of the P1 helix and positions 77–80 of P3 in B. subtilis yitJ (8). This invokes similar possibilities of zipper formation between nt-88 (P1) and nt-58 (P3), nt-89 (P1) and nt-47 (P3), nt-89 (P1) and nt-57 (P3), and nt-90 (P1) and nt-56 (P3) in the crystal structure. Some additional zipper formations are also plausible above the methionine group of SAM (4, 13). Such extended zipper formation between the 3′-P1 and the P3 helix suggests that the aptamer secures the 3′-end of the P1 helix by metabolite binding, as this is the switching element that forms the antiterminator. It appears that formation of this ribose zipper has two advantages: 1) it can moderately secure the binding site from ligand escape and 2) it can protect the switching element as an alternative strategy in a SAM-independent manner. The securing of the binding site also suggests that SAM-I could adopt a preorganized structure that resembles the metabolite-bound conformation (14, 15). To facilitate the tertiary interactions in the binding pocket, SAM folds into a U-shape form. This U-shape is involved in a base-triple formation including A45, U57, and the central adenine ring of SAM (Fig. 1). The methionine moiety of SAM forms a series of hydrogen bonds with the G11-C44-G58 base-triple between junction J1/2 and helix P3. SAM also consists of a ribose sugar and a positively charged sulfonium group, which is close to two conserved basepairs in the minor groove of the P1 helix, A6-U88 and U7-A87 (4). The positively charged sulfonium group of SAM is probably important for its recognition as it is connected with the RNA through an electrostatic interaction (4, 13, 15). This electrostatic interaction could provide additional stability to the P1 helix in the presence of SAM.

To understand the structural response upon metabolite binding, in our earlier study, we used all-atom structure-based model (SBM) simulations to characterize the folding of the aptamer domain of this riboswitch (16), focusing on the secondary structure folds. Our results showed that the folding of the nonlocal P1 helix is the rate-limiting step in aptamer formation. Comparative gel electrophoresis and fluorescence resonance energy transfer experiments have also found that the stability of the metabolite binding site depends on the formation of the P1 helix and its close juxtaposition with the P3 helix (11). A temperature-dependent selective 2′-hydroxyl acylation analyzed by primer extension (SHAPE) experiment studied the thermal stability of different segments of the SAM-I riboswitch, showing that many tertiary interactions are more stable compared to the secondary structure elements, such as the P1 and the P4 helices (8). The study revealed the hierarchical melting pathway of the B. subtilis yitJ S-box riboswitch. Early small angle x-ray scattering (SAXS) experiments investigated the structural detail of the ligand-free state of the SAM-I riboswitch (17). The experimental SAXS profile for the metabolite-free riboswitch results in a noticeable decrease in the radius of gyration of the aptamer domain. On the other hand, the SHAPE reactivity measurements with fixed temperature but varying Mg²⁺ concentrations found the signature of a clear population shift from an open to a closed state upon ligand binding. This transition involves structural modifications over the pseudoknot and the fourway junction region, even in the presence of SAM. All these experimental and computational investigations suggest that there is a dynamic coupling among the tertiary structural elements, SAM, and magnesium ions that together determine the riboswitch function, which is not fully understood from the structural perspective.

To study the interplay between RNA structure and Mg²⁺ ions at an atomic resolution, we previously performed ten 2-μs explicit solvent molecular dynamics simulations of the SAM-I riboswitch at varying ion concentrations (18). To explore its energy landscape with the associated ion-atmospheric effect of Mg²⁺, recently we developed an electrostatic model capable of describing the ion-atmosphere surrounding the RNA with quantitative accuracy (19). The model treats the Mg²⁺ ions explicitly to take into account the direct ion-ion correlations. After the experimental buffer setup, KCl is treated implicitly. This implicit treatment of KCl is included in the model using generalized Manning counterion condensation theory (19, 20). We implemented this electrostatic effect as a form of potential within the coarse-grained structure-based model (21, 22). We have earlier tested this model against experimental measurements of the number of excess Mg²⁺ ions, associated with the RNA, which determines the Mg²⁺-RNA interaction free energy (19). To extend the applicability of our model, we made each Manning condensation variable a dynamic quantity that evolves with the corresponding atomic coordinates. The dynamic version of our structure-based generalized electrostatic model (GEM) has been found useful in predicting conformational ensembles of the SAM-II riboswitch in its metabolite-free form, and the salt concentration dependence of the ensembles (23). We observed a dynamic exchange between a partially closed triplex and an open state of the apo-SAM-II riboswitch at low Mg²⁺ concentrations. This dynamic transition and its Mg²⁺ concentration dependence are found in excellent agreement with recent single molecule fluorescence, SAXS, ¹³C, and ¹H chemical exchange saturation and size-exclusion chromatography experiments (23, 24, 25, 26).

Magnesium also plays a key role in the SAM-I riboswitch. Previous SHAPE chemical probing experiments performed in our wetlab demonstrated that magnesium and SAM separately cannot drive the collapse transition to completion (27). Instead, both magnesium and SAM are required to form the fully collapsed metabolite-bound closed state of the SAM-I riboswitch. In this article, we perform atomistic simulations to interpret these experimental data. Specifically, we use simulations to investigate the effect of SAM and magnesium ions on the folding free energy landscape of the SAM-I riboswitch. We addressed the importance of the presence of magnesium ions in the preorganization event and the presence of both magnesium ions and SAM in the postmetabolite-binding closure of this riboswitch. In the simulations, SAM stabilizes the terminal helix P1 by connecting it with helix P3. Here, the effect of the small, dynamic Mg²⁺ ion alone is insufficient to stabilize the highly flexible 5′-end of the P1 helix. Conversely, Mg²⁺ plays an important role in stabilizing the pseudoknot region, which is quite distant from the SAM binding site. The combined effects of SAM and Mg²⁺ produce the fully collapsed state. Furthermore, recent temperature-dependent SHAPE data have shed light on the hierarchical folding pathway of this riboswitch, indicating relatively lower stability of the P1 and P4 helices compared to other secondary and tertiary interactions (8). In addition to the flexible P1 helix, our simulations reveal the highly flexible behavior of the P4 helix. In the simulations, helix P4, being a small helix, can spontaneously fold and unfold to facilitate formation of the other elements in the RNA fold.

Materials and Methods

Potential for structure-based GEM simulation

We have implemented ion interactions as a form of potential, within the coarse-grained SBM (21, 22), following Hayes et al. (19). The functional form of the Hamiltonian used for our structure-based GEM simulations consists of three terms as shown (19):

$$H=H_{\text{SBM}}+H_{\text{Excl}}+H_{\text{Elec}},$$ (1)

where H_SBM contains harmonic potentials that restrain the bonds (r), angles (θ), and improper/planar dihedrals (χ). Proper/flexible dihedral angles (ϕ) are treated with cosine terms (28), as shown in Eq. 2. A Lennard-Jones’ 6–12 potential form is used to treat the nonlocal pair-interactions present in the native crystal structure whereas all other, nonnative pair-interactions present are repulsive:

$$\begin{array}{c}{H}_{\text{SBM}}=\underset{\text{Local}}{\underbrace{\sum _i^{\text{bonds}}\frac{k_r}{2}{\left(r_i-r_{ni}\right)}^2+\sum _i^{\text{angles}}\frac{k_{\theta }}{2}{\left({\theta }_i-{\theta }_{ni}\right)}^2+\sum _i^{\begin{array}{l}\text{improper}\\ \text{or}\text{planner}\end{array}}\frac{k_{\chi }}{2}{\left({\chi }_i-{\chi }_{ni}\right)}^2+\sum _i^{\begin{array}{l}\text{proper}\\ \text{dihedral}\end{array}}{k}_{\varphi }{F}_D\left({\varphi }_i-{\varphi }_{ni}\right)}}\\ \underset{\text{Nonlocal}}{\underbrace{+\sum _{ij}^{\text{contacts}}{k}_c\left({\left(\frac{{\sigma }_{ij}}{r_{ij}}\right)}^{12}-2{\left(\frac{{\sigma }_{ij}}{r_{ij}}\right)}^6\right)+\sum _{ij}^{\begin{array}{l}\text{RNA}\text{-}\text{RNA}\\ \text{non}\text{-}\text{contacts}\end{array}}{k}_{\text{NC}}{\left(\frac{{\sigma }_{\text{NC}}}{r_{ij}}\right)}^{12}}},\end{array}$$ (2)

where

$$F_D\left(\varphi \right)=\left[1-\text{cos}\left(\varphi \right)\right]+\frac{1}{2}\left[1-\text{cos}\left(3\varphi \right)\right].$$ (3)

The geometric parameters (r_ni, θ_ni, χ_ni, ϕ_ni, σ_ij) are obtained from the crystal structure. Following our previous setup of SBM potential for proteins, the ratio of total contact energy to total dihedral energy is maintained, as ${\sum }_i^{\text{contacts}}{k}_c/{\sum }_j^{\text{proper}\text{dihedral}}{k}_{\varphi }=2$ (28, 29). All the interaction coefficients used in this potential are given in Table 1.

Table 1.

Values of Parameter Set Used for Current Structure-Based GEM Simulation of SAM-I Riboswitch

Reduced parameter set
τR	2 ps
μR	15 amu
εR	3.75 kJ/mol
TR	3.75 K
SBM potential
kr	2 × 104εR/nm2
kθ	40 εR/rad2
kχ	10 εR/rad2
σNC	0.17 nm
kNC	1.33 kBT
Excluded volume potential
σRNA-Mg2+	0.34 nm
σMg2+-Mg2+	0.56 nm
Electrostatic potential
σμ	0.7 nm
ση	0.34 nm
kHole	104kBT/P(0,σn)2
kW	1 kBT

The excluded volume effect for the explicit hexa-hydrated Mg²⁺ (30) ion is included in the potential, H_Excl, as shown:

$$H_{\text{Excl}}={\sum _{ij}^{\begin{array}{c}\text{Mg}\text{-}\text{RNA}\\ \text{Mg}\text{-}\text{Mg}\end{array}}{\epsilon }_{ij}\left(\frac{{\sigma }_{ij}}{r_{ij}}\right)}^{12}.$$ (4)

The electrostatic ion potential, H_Elec, consists of four free energy terms: G_E, G_Mix, G_Hole, and G_Rest.

The competition between the electrostatic energy and the mixing entropy determines the classical Manning counterion condensation effect (20, 31). Apart from the condensed charges, the electrostatic free energy also takes into account the Debye-Hückel screening charges. The removal of redundant screening charges from the excluded volume of RNA has generalized Manning condensation model extending its applicability for other polyionic systems. To model the heterogeneous charge distribution, three types of charge distribution have been included: 1) a Gaussian mixing charge (P(r,σ_μ)) to address the mixing free energy, 2) a Gaussian hole charge (P(r,σ_η)) to estimate the implicit charges that need to be subtracted from the excluded volume of RNA, and 3) a point charge (σ₀) for each particle (19). Screening charges are included only in an effective volume region considering Gaussian occupancy, (V_μ,i − V_η,i) = V_eff,i. Thus, the total electrostatic free energy is expressed as:

$$G_E=\frac{1}{2}\underset{\text{point}\text{charges,}\text{condensed}\text{charges}}{\underbrace{\sum _{ij}\sum _{xy}{q}_{x,i}{q}_{y,j}\varphi \left(r_{ij},{\left({\sigma }_x^2+{\sigma }_y^2\right)}^{1/2}\right)}}+\underset{\text{screening}\text{charges}}{\underbrace{\sum _i\sum _s{z}_s{c}_s\left(1-\frac{z_s{\mathrm{\Phi }}_{\mu }\left(\overrightarrow{r_i}\right)}{k_{\text{B}}T}\right)V_{\text{eff},i}{\mathrm{\Phi }}_{\mu }}}\left(\overrightarrow{r_i}\right),$$ (5)

where

$${\Phi }_x\left(\overrightarrow{r_i}\right)=\sum _j\sum _n{q}_{y,j}\varphi \left(\left(\overrightarrow{r}-{\overrightarrow{r}}_j\right),{\left({\sigma }_x^2+{\sigma }_y^2\right)}^{1/2}\right).$$ (6)

The indices x and y ensure the pairwise Debye-Hückel interactions (ϕ) between any of these charge types. The value r_ij is the distance between particles i and j. The value z_s is the charge number and c_s is the concentration of a specific ion, s. For particle $i$, the condensed charges are $q_{\mu ,i}={\sum }_s{z}_s{\mu }_{is}$ and $q_{\eta ,i}={\sum }_s{z}_s{\eta }_{is}$.

Mixing free energy within this effective volume is given in

$$G_{\text{Mix}}=\sum _i\sum _s{k}_{\text{B}}T{n}_{\text{Mix},is}{V}_{\text{eff},i}\text{ln}\left(\frac{n_{\text{Mix},is}}{e{c}_s}\right)$$ (7)

where

$$n_{\text{Mix},is}=c_s\left(1-\frac{z_s{\mathrm{\Phi }}_{\mu }\left(\overrightarrow{r_i}\right)}{k_{\text{B}}T}\right)+n_{\mu ,s}\left(\overrightarrow{r_i}\right),$$ (8)

and e is Euler’s number.

Two other restraining potentials, G_Hole and G_Rest, are applied to maintain the charges within their restricted volume region and are shown in Eqs. 9 and 11, respectively:

$$G_{\text{Hole},is}=\frac{1}{2}{k}_{\text{Hole}}\sum _i\sum _s{n}_{\text{Hole},is}^2,$$ (9)

where

$$n_{\text{Hole},is}=c_s\left(1-\frac{z_s{\mathrm{\Phi }}_{\mu }\left(\overrightarrow{r_i}\right)}{k_{B}T}\right)+n_{\mu ,s}\left(\overrightarrow{r_i}\right)+n_{\eta ,s}\left(\overrightarrow{r_i}\right)$$ (10)

and

$$G_{\mathrm{Re}\mathrm{st}}=\frac{1}{2}{k}_w\sum _i\sum _s{\left({\mu }_{is}-n_{\mu ,s}\left(\overrightarrow{r_i}\right)V_{\mu ,i}\right)}^2+\frac{1}{2}{k}_w\sum _i\sum _s{\left({\eta }_{is}-{\eta }_{\eta ,s}\left(\overrightarrow{r_i}\right)V_{\eta ,i}\right)}^2.$$ (11)

The parameter set we used for current simulation is given in Table 1. The related parameter set and its calibration are also available in our early studies (19, 23).

Equilibrium simulation detail

All atomic coordinates together with μ_is and η_is are evolved with Langevin dynamics with a time step of 0.001 τ_R. For rapid sampling, we used underdamped conditions. The mass of each explicit particle of 1 μ_R with a drag coefficient of 1 τ_R⁻¹ is used. Condensation coordinates μ_is and η_is are given a mass of 15 μ_R nm² with a drag coefficient of 0.05 τ_R⁻¹ nm². Effective charges were taken for each phosphate group as −1, each magnesium ion as +2, and each SAM molecule has, effectively, +1 charge. In SAM, a positively charged sulfonium group, an amino group, and a negatively charged carboxylate group, are important for SAM recognition (32). Temperature was chosen such that both folded and partially unfolded states were accessible at physiological Mg²⁺ concentration ([Mg²⁺] ∼ 2 mM) (T = 90 T_R). To prepare different Mg²⁺ compositions we set up a large cubic box of length 75 nm. The number of Mg²⁺/SAM molecules included in that box determines the overall concentration of the corresponding solute. Periodic boundary conditions were applied. Each simulation was propagated for a total of 20,000,000 time steps.

Free energy calculation

To calculate the free energy landscape, we used the umbrella sampling method (33). We first generated a series of initial structures at our desired Q_Global values. To ensure equilibration, we reinitialized Mg²⁺ distributions at each umbrella window. Each window ran for 10,000,000 steps. Umbrella sampling has been performed slightly below the folding temperature to cover the whole landscape including the sparsely populated preorganized ensemble of SAM-I. Every repeat used a total of 40 windows along the reaction coordinate, ensuring their substantial overlap in the conformational space. Finally, the weighted histogram analysis method is used to calculate the unbiased free energy profile (34). Four repeats were performed to check the consistency of the results with errors estimated. For the simulations, in the presence of SAM, the binding strength between the metabolite and the riboswitch was slightly tuned to ensure that we can obtain multiple binding-unbinding events. More detail about contact counting potential, generation of initial configurations, and error estimations are given in the Supporting Material.

Covariance and principal component analysis

To analyze the dynamic correlation between the PK and the P1 helix, we have calculated the covariance map (35) derived from the atomic position fluctuation of each backbone P-atom (of the backbone phosphate group) along the trajectory started from the closed state. The covariance matrix is written as

$$C_{ij}=\langle \left(x_i-\langle x_i\rangle \right)\rangle \langle \left(x_j-\langle x_j\rangle \right)\rangle ,$$ (12)

where x_i, x_j are Cartesian coordinates of the selected atoms. The covariance parameters were calculated for the backbone P-atoms of all residues after aligning the backbone of the closed state, which is taken as the reference state. To analyze the i, j pairwise correlations, we have estimated the sum of the xx, yy, and zz covariances of each P-atom, representing each residue.

Apart from the backbone, we have also analyzed the symmetric covariance matrix for all RNA atomic positions, after aligning them to the closed state. This covariance matrix has been diagonalized with an orthonormal transformation matrix and the corresponding eigenvalues were calculated. The trajectories of all RNA atoms are then projected on the eigenvectors to obtain the principle components. The essential dynamics were captured by the first principal component eigenvector (36).

SHAPE reactivity calculations

Chemical probing, including SHAPE and other emerging smart reagent probing approaches (such as RING-MaP experiments), have been used to gain a deeper understanding of RNA structure (37). In the experiment, the SHAPE reactivity measures the conformational flexibility, which is probed by the selective acylation at the 2′-hydroxyl position and then analyzed by primer extension (38). SHAPE reactivity indicates how much a region or a nucleotide is conformationally flexible. To theoretically address the conformational flexibility for each residue, we monitor the related fluctuations of the angle, 〈O2′ – P – O5′〉. The angle involves O2′ of the 2′-hydroxyl group, P of the phosphate group, and O5′ of ribose sugar (〈O2′ – P – O5′〉) (39). The corresponding angle (θ_i) fluctuations encode the stability of each nucleotide (i) by the following free energy change as

$$\mathrm{\Delta }G\left(i\right)=k_{\text{B}}T\text{ln}\frac{\langle {\theta }_i^2\rangle -{\langle {\theta }_i\rangle }^2}{\frac{1}{N}\sum _{i=1}^N\left(\langle {\theta }_i^2\rangle -{\langle {\theta }_i\rangle }^2\right)}.$$ (13)

Deigan et al. (40) proposed a pseudo-energy function that relates the stability of a nucleotide (i) and its normalized SHAPE reactivity. We used the same formalism, as ΔG(i) = a [ln{SHAPE reactivity (i)} +1] +b, where the slope a = 2.6 kcal/mol and the intercept b = −0.8 kcal/mol were used as general parameters observed for large RNA folding.

Results

Molecular simulations of SAM and magnesium-induced aptamer collapse are consistent with SHAPE chemical probing experiments

Previously, from our experimental SHAPE probing data, we obtained the indications of coupling among the metabolite, SAM, and Mg²⁺ ions in the SAM-I riboswitch (27). We found that the metabolite-induced collapse mostly involves stabilization of the four-helix junction, the P1 helix, and the pseudoknot region. In a recent study, we have calculated a pseudo-energy function that has proven useful in regenerating the experimental data (39). To study the conformational mechanism of metabolite binding, in this study we used the same pseudo-energy function to evaluate the normalized SHAPE reactivity from our structure-based GEM simulations in a full ionic environment (40).

The computed normalized SHAPE reactivity profile for T. tengcongensis yitJ and our early experimental results for T. tengcongensis metF closely match with each other. The Mg²⁺ ion-induced collapse for the metabolite-free (Fig. 1 C) and for the metabolite-bound states (Fig. 1 D) is reflected in the corresponding SHAPE reactivity calculations. The estimated errors for our theoretically computed SHAPE reactivity values for each residue are rather small (see Fig S1). To better extract the Mg²⁺ effect, we have plotted the SHAPE reactivity difference between 0 and 2 mM Mg²⁺ for each nucleotide. The positive difference in SHAPE reactivity between 0 and 2 mM Mg²⁺ clearly identifies the regions that lose their mobility with the addition of Mg²⁺ ions. The affected regions are highlighted with circles in Fig. 1 E. The difference plot captures the effect of Mg²⁺ on P1, P3, and the pseudoknot regions, both in the presence and in the absence of SAM. Note that each of these elements contains highly exposed phosphate groups.

Effect of metabolite and magnesium ions on the open-closed dynamics of pseudoknot and nonlocal P1 helix

The SHAPE reactivity data help us to pinpoint the regions that are most affected by the Mg²⁺ ions (27). Therefore, we analyzed our equilibrium trajectories especially focusing on the dynamics of P1 and pseudoknot regions. Trajectory plots of the fraction of native contacts in the pseudoknot (PK) region extracted from our structure-based GEM equilibrium simulations of the metabolite-free SAM-I riboswitch show a conformational transition between a PK-closed and a major PK-open state (Fig. 2 A). With the addition of 2.0 mM Mg²⁺, the system visits the open state less frequently, as summarized in the corresponding contact histograms, P(Q_PK) (Fig. 2 B). The pseudoknot shows the same open-closed transition in the presence of metabolite, but without Mg²⁺ (Fig. 2, C and D). The presence of both metabolite and Mg²⁺ helps the PK to preserve its closed state. The nonlocal P1 helix also shows similar behavior to the dynamics of the PK in the metabolite-free state of the riboswitch (Fig. 2, E and F). These data indicate that whereas the metabolite appears more effective in securing the nonlocal P1 helix, the presence of both Mg²⁺ and SAM could help in securing the pseudoknot. Our early SHAPE wetlab measurements have shown that the effect of Mg²⁺ and SAM separately cannot drive the folding transition to completion. The detailed mechanism involving the PK and P1, as exhibited in our simulations, requires further experimental validation.

Figure 2.

Magnesium-sensitive dynamics of the pseudoknot and the nonlocal P1 helix in SAM-I riboswitch. (A) Conformational dynamics of the pseudoknot, in the absence of SAM, in terms of its fraction of native contacts (Q_PK) extracted from simulations, indicates more frequent open-closed transitions in the absence of Mg²⁺ than that in the presence of 2.0 mM Mg²⁺. (B) The fraction of native contact population histograms at these two Mg²⁺ conditions shows that a major pseudoknot-closed state prevails with the addition of 2.0 mM Mg²⁺. (C) Same as (A), but done in the presence of SAM. (D) Same as (B), but in the presence of SAM, it indicates that both SAM and Mg²⁺ ions are required to form the pseudoknot. (E) Conformational dynamics of the nonlocal P1-helix, in the absence of SAM, in terms of its fraction of native contacts (Q_P1) extracted from simulations indicates more frequent unfolding in the absence of Mg²⁺ than that in the presence of 2.0 mM Mg²⁺. (F) The fraction of native contact population histograms at these two Mg²⁺ conditions indicates 2.0 mM Mg²⁺ is able to only partially fold P1. (G) Same as (E), but in the presence of SAM. (H) Same as (F), but in the presence of SAM, which indicates that the effect of SAM is crucial for the helix P1 to form.

Thermodynamic characterization of magnesium effects and magnesium-metabolite cooperation

Our previous SHAPE experiment and this structure-based GEM equilibrium simulation data reveal that both SAM and Mg²⁺ ions are essential to facilitate the structural collapse (27). Here we attempt to understand and compare the isolated effect of Mg²⁺ ions in the absence of SAM and the integrated effect, combining both Mg²⁺ ions and SAM on the energy landscape of this riboswitch. We calculate the folding free energy profile in the presence of 2 mM Mg²⁺ without SAM and in the presence of 2 mM Mg²⁺ with SAM. In the absence of metabolite, the free energy profile as a function of the global fraction of native contacts (Q_Global) reveals three distinct minima from lower Q_Global to higher Q_Global, defined as O (open state), PO (partially open state), and PC (partially closed preorganized state) (Fig. 3 A). The free energy profile in the presence of explicit SAM has a distinct bound-state minimum, reflecting the metabolite-induced stabilization of the closed conformation (C), which appears at higher Q_Global than the Mg²⁺-induced partially closed state. We have also calculated the fraction of regional contacts, Q_Regional along Q_Global (Fig. 3 B). Here each secondary/tertiary region (such as P1, P2, P3, P4, PK, P1–P3 zipper) has been individually distinguished by the total number of pairwise contacts involved in each segment. During this analysis, in the fraction of regional contacts, we have only considered total basepairing contacts (excluding local base-stacking contacts). Thus during the unfolding and the folding of a particular secondary structure, it spans from 0 to 1, respectively. The plot shows the sequence of folding events, illuminating each secondary and tertiary structure formation event along the free energy profile, where formation of the pseudoknot and partial formation of the nonlocal P1 helix appear to be the rate-limiting step. The sequence of folding directly helps to characterize each minimum in the free energy profile. The representative structures corresponding to four distinct minima ((i) O, (ii) PO, (iii) PC, and (iv) C) are shown in Fig. 3 C. Fig. 3 A suggests that even in the absence of SAM, this riboswitch can attain a well-ordered preorganized state (PC). Although this preorganized state resembles the SAM-bound closed conformation (C), several contacts involving the P1 helix and fourway junction are missing. Recent temperature-dependent SHAPE data have shed light on the hierarchical folding pathway of this riboswitch, indicating relatively lower stability of P1 and P4 helices compared to other secondary and tertiary interactions (8). In addition to the flexible P1 helix, our simulations reveal the highly flexible behavior of the P4 helix. It seems that P4, being a small helix, can spontaneously fold and unfold to facilitate formation of the other elements in the RNA fold. Such backtracking of the P4 helix has also been found in a recent study of transition-rate-limited free folding of the SAM-I riboswitch (41).

Figure 3.

Folding free energy landscape of the SAM-I riboswitch shows how Mg²⁺ ions preorganize the riboswitch in the absence of SAM. (A) The free energy landscape as a function of the fraction of global native contacts (Q_Global) of SAM-I near physiological Mg²⁺ concentration ([Mg²⁺] = 2.0 mM), in the absence of SAM. The system explores three distinct barrier-separated minima in the free energy landscape, namely, O, PO, and PC. Estimated errors are highlighted with blue bars for the SAM-bound profile and with red bars for the profile in the absence of SAM. (B) The order of secondary structure formation is shown as a function of Q_Global, as measured by the fraction of nonlocal regional contacts (Q_Regional). (C) Representative structures are given. The representative structure corresponding to each minimum of the energy landscape is designated as follows: (i) an open state (O) showing initial secondary structure and formation of helices P2 and P4; (ii) a partially open state (PO) where the connection between helix P3 and the 3′-P1 strand prepares the binding platform; and (iii) a partially closed state (PC) that shows the formation of a pseudoknot and partial P1-helix. The metabolite-bound closed state minimum has been characterized from the free energy profile of the folding transition of SAM-bound RNA, which indicates that the metabolite-free partially closed (PC) conformation resembles the metabolite-bound closed state (C, shown as (iv)).

In the presence of a very high concentration of SAM (1 mM), the preorganization of the binding site follows the same pathway of folding as it does in the absence of SAM. The initial formation of P2 and P4 helices (O state) precedes the preorganization of the binding site by the closure of 3′-P1 and the P3-helix (PO state) (Fig. 4 A). This close juxtaposition of the P3 helix to the 3′-P1 strand prepares the binding platform, which is occupied by a SAM molecule as shown in Fig. 4 B. The rate of binding is highly dependent on the concentration of the metabolite. It is interesting to note that at high concentration, a metabolite can associate any early stage prebinding conformation once the binding site forms. However, such flexible conformations may not guarantee the stabilization of the bound metabolite. In the SAM-I riboswitch, the postligand-binding closure event initiated by the tertiary pseudoknot and nonlocal P1-helix interactions determines the stabilization of the binding site and prevents metabolite escape. The presence of metabolite favors the formation of P1 by reducing the barrier for the process, where the dynamic 5′-P1 strand associates with its relatively stable partner 3′-P1 strand. The premetabolite binding folding events are similar both in the absence and the presence of metabolite. However, the postmetabolite binding event shows that, depending on the association of the P1 helix with SAM, the order of PK and the rest of nonlocal P1 helix formation is altered. This suggests that there is a dynamic interplay between the tertiary PK region and the P1 helix.

Figure 4.

Folding free energy landscape of the SAM-I riboswitch, in the presence of 2.0 mM Mg²⁺ and 1.0 mM SAM. (A) Given here is the free energy landscape of SAM-I, as a function of the fraction of global native contacts (Q_Global), near physiological Mg²⁺ concentration ([Mg²⁺] = 2.0 mM) and in the presence of 1.0 mM SAM. Estimated errors are highlighted with blue bars in the profile. The system essentially explores two distinct barrier-separated minima on the free energy landscape: (i) a flat open state basin at lower Q_Global consists of an open state (O) and a partially open (PO) state; and (ii) a deep bound-state well consisting of a partially closed (PC) and a metabolite-bound closed (C) state. (B) The order of secondary structure formation as a function of Q_Global, as measured by the fraction of nonlocal regional contacts (Q_Regional), follows the same preorganization pathway (through O and PO) as observed in Fig. 3C. The sequence of postmetabolite-binding events shows partial formation of P1-helix earlier than the pseudoknot.

Dynamic correlation between pseudoknot and nonlocal P1 helix during preorganization and postmetabolite-binding closure

To understand the interplay between the P1 helix and the PK motif, we have performed covariance analysis. This method is often used to explore the correlated motion between different distant regions/domains in a biomolecule. We analyzed the pairwise covariance map derived from the fluctuation of backbone P-atoms of the riboswitch (details of the calculations are described in the Materials and Methods) at four different buffer conditions, as shown in Fig. 5 A. We find two pairs of crosspeaks with higher intensity corresponding to PK and P1-helix regions. The first pair of crosspeaks (highlighted by a circle) appears near the P1-helix (extended to the J1/2 junction region) and the PK@P2 regions. Another pair of crosspeaks (highlighted by an ellipse) reflects dynamical correlations between distant PK@P2 and PK@P4 regions. Such coherently correlated PK-P1 motions might be essential for the preorganization that might again disappear when the collapsed functional state is reached in the presence of both SAM and physiological Mg²⁺ concentration, as we obtained in Fig. 5 Aiv. We have also characterized their most important motions from a principal component analysis of SAM-I riboswitch simulations, both in the presence and the absence of SAM. To account for the effect of flexibility, we have performed the analysis at no Mg²⁺ conditions to gain sufficient conformational flexibility and to explore metabolite-induced stabilization. In the absence of SAM, we observe how the PK unknots and influences the 5′-P1 helix to unwrap the binding site (Fig. 5 B). Such correlated open-closed dynamics appears to form the basis of the preorganization of the SAM-I riboswitch in the absence of its metabolite. Although we found a dynamic correlation between the motion of the PK and the P1 helix in the absence of SAM, this correlation disappears in the presence of SAM (Fig. 5 C) as we had similarly obtained during the covariance analysis. SAM plays a dominant role in stabilizing the P1 helix, which, in turn, secures stable SAM binding.

Figure 5.

Correlated dynamics of the pseudoknot and the nonlocal P1 helix drive preorganization. (A) Residue pairwise covariance map is derived from backbone P-fluctuation at four different buffer conditions 1) (−) Mg²⁺, (−) SAM; 2) 2.0 mM Mg²⁺, (−) SAM; 3) (−) Mg²⁺, (+) SAM; and 4) (+) Mg²⁺, (+) SAM. The crosspeak highlighted by the dotted circle is due to the correlated motion between PK and P1-helix regions. At low to no Mg²⁺ and SAM concentrations, long-range PK motions are highlighted by dotted ellipse. (B) Given here are representative snapshots extracted from the first PCA eigenvector of the simulation trajectories of SAM-I riboswitch in the absence of SAM. The correlated dynamics of the pseudoknot and the nonlocal P1 helix forms the basis for preorganization. The sequence of a large opening motion is shown here that triggers the opening of 5′-P1. (C) The presence of SAM turns off the correlated motion between PK and 5′-P1. (D) The resulting mechanism of stable metabolite binding is summarized with the events of preorganization and the postmetabolite-binding closure. In the absence of metabolite, whereas Mg²⁺ favors the preorganization with the partial closure of P1 and pseudoknot, metabolite preferentially promotes full P1 helix formation. The underlying cooperation between Mg²⁺ and SAM is mostly reflected in a stable pseudoknot formation, which SAM alone cannot reach.

Discussion

Comparative gel electrophoresis, fluorescence resonance energy transfer, SAXS, temperature-dependent SHAPE experiments, and our earlier structure-based simulations have attempted to characterize the hierarchical folding events of the SAM-I riboswitch (8, 11, 17). All these studies and our current thermodynamic analysis consistently found that the stability of the nonlocal P1 helix is a key determinant factor, because a part of P1 (3′-P1) is the switching segment between the aptamer and the expression platform. The importance of nonlocal P1 helix stabilization in connection with stable metabolite-binding for the adenine riboswitch has also been reported in NMR, fluorescence, and atomic force spectroscopy studies (42, 43). The role of metabolite in stabilizing the nonlocal P1 helix, and thus its effects in terminating the transcription of many downstream genes, has been recognized earlier (44, 45). An integrated effect of both metabolite and magnesium ions, especially affecting the open-closed dynamics of the pseudoknot and the P1-helix in the aptamer domain, has been found in this study as shown in Fig. 5 D. Our SHAPE reactivity analysis earlier identified that the presence of Mg²⁺ ions is indeed essential, along with the presence of SAM to reach the metabolite-bound closed state (27). Our present molecular dynamics simulation study has explicitly identified the target RNA motifs sensitive to Mg²⁺ ions and SAM. We found that the U-shaped SAM essentially stabilizes the P1-helix by connecting it with the P3-helix, where the effect of the small, dynamic Mg²⁺ alone is insufficient to stabilize the highly flexible 5′-end of the P1-helix. On the other hand, Mg²⁺ plays an important role in stabilizing the pseudoknot region where the metabolite, SAM, cannot reach. Our covariance matrix and principal component analysis have identified important correlated dynamics of the pseudoknot and the P1 helix where the pseudoknot and the dynamic 5′-P1 end wrap the binding site to protect the metabolite. Whenever the PK and P1 unwrap, they expose the binding site, opening the fourway junction.

Before metabolite binding, the prebinding interactions preorganize the riboswitch so that the aptamer can attain a closed, boundlike conformation, although sparsely populated, even in the absence of metabolite. During preorganization, the apo-aptamer shows a dynamic equilibrium between a P1-PK open state and a P1-PK closed state. Although the rate of metabolite binding depends on the binding affinity of metabolite, the preorganized conformation is expected to facilitate rapid detection of the metabolite with high selectivity when the metabolite concentration is low. At higher metabolite concentration, the diffusion search time for the binding pocket can be lowered compared to the folding time. Therefore, metabolite binding, at high concentration, can induce a different folding pathway as we have observed in Figs. 3 B and 4 B. The spontaneous backtracking motion of the P4 helix seems to be an important dynamic event that helps to overcome geometric frustration in the structure (41).

The role of Mg²⁺ ions in the preorganization has also been recognized in other riboswitches, such as the SAM-II, pre-Q, adenine, and Thi-box riboswitches (23, 24, 43, 46, 47). In a recent study, we have found that a sufficiently high concentration of Mg²⁺ is required to stabilize the preorganized state in the energy landscape. We studied the mode of interaction where Mg²⁺ ions bridge a number of adjacent phosphate groups together to favor collapse. It is important to note that during the transcription process the metabolite finds a brief opportunity to bind in the aptamer fold and terminate transcription. If the rate of transcription is faster than the rate of folding, then this may not be the biological pathway for transcription termination. Thus, the presence of Mg²⁺ is important in accelerating fast folding by lowering the barrier of preorganization.

Whereas the presence of Mg²⁺ ions is important for preorganizing the aptamer, it also plays an essential role in the postmetabolite-binding closure (Fig. 5 D). Our present simulation results, together with our earlier experimental SHAPE measurements (Fig. 3 in (27)), highlight the effect of Mg²⁺ ions on the pseudoknot region and partly on the 5′-P1 helix. The characterization of preorganization, postmetabolite binding closure, and the Mg²⁺ ion sensitivity of these events advances our understanding of the biological function of this mRNA element. Moreover, finding the relevant regions for preorganization will be useful to promote in vitro transcription attenuation. In this study we also draw attention to the functional significance of the important tertiary structural elements, such as the pseudoknot region and the zipper connection in between the 3′-P1 segment and the P3 helix, as observed in the crystal structure of the SAM-I riboswitch in both T. tengcongensis and B. subtilis yitJ (4, 8).

Editor: Tamar Schlick.