Structure of Essential RNA Regulatory Elements in the West Nile Virus 3′-Terminal Stem Loop

Abstract

Flaviviruses, such as West Nile and Dengue Virus, pose a significant and growing threat to global health. Central to the flavivirus life cycle are highly structured 5’- and 3’-untranslated regions (UTRs), which harbor conserved cis-acting RNA elements critical for viral replication and host adaptation. Despite their essential roles, detailed molecular insights into these RNA elements have been limited. By employing nuclear magnetic resonance (NMR) spectroscopy in conjunction with SAXS experiments, we determined the three-dimensional structure of the West Nile Virus (WNV) 3’-terminal stem-loop core, a highly conserved element critical for viral genome cyclization and replication. Single nucleotide mutations at several sites within this RNA abolish the ability of the virus to replicate. These critical sites are located within a short 18-nucleotide hairpin stem, a substructure notable for its conformational flexibility, while the adjoining main stem-loop adopts a well-defined extended helix interrupted by three non-Watson-Crick pairs. This study enhances our understanding of several metastable RNA structures that play key roles in regulating the flavivirus lifecycle, and thereby also opens up potential new avenues for the development of antivirals targeting these conserved RNA structures. In particular, the structure we observe suggests that the plastic junction between the small hairpin and the tail of the longer stem-loop could provide a binding pocket for small molecules, for example potentially stabilizing the RNA in a conformation which hinders the conformational rearrangements critical for viral replication.

Graphical Abstract

1. Introduction

Flaviviruses such as Dengue (DENV), West Nile (WNV), and Yellow Fever (YFV) are responsible for over 400 million human infections annually.¹ Symptoms of these viral infections can range from mild fever to severe hemorrhagic conditions.^2,3 WNV, in particular, is known for potentially serious neurological manifestations such as encephalitis and meningitis. Prevalent in tropical regions, WNV is expanding its reach due to the shifting habitat of its mosquito vector, as a result of changes in climate and global trading networks.⁴ Currently, there are no effective treatments or vaccines against most flaviviruses,⁵ making it imperative to better understand their lifecycles at the molecular and structural level, in order to identify new targets for therapeutic intervention.

WNV, like all flaviviruses, is a single-stranded positive-sense RNA virus. Its genomic RNA is approximately 11,000 nucleotides long, containing a single open reading frame (ORF) that encodes a large polyprotein. This polyprotein is cleaved into three structural proteins (capsid (C), envelope (E) and pre-membrane/membrane (prM)), as well as seven nonstructural viral proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5), as shown in Fig. 1A. The ORF is flanked by a short 5’-untranslated region (about 100 nucleotides in length) and a longer 3’-untranslated region (approximately 600 nucleotides). These untranslated regions harbor conserved RNA structures that function as promoters or repressors to regulate viral replication, translation and host adaptation.^6,7 They also contain inverted complementary sequences, which form long-range RNA-RNA interactions that cyclize the viral genome, a critical step in the flavivirus lifecycle.^8–10

Figure 1. Genomic organization of Flaviviruses.

(A) Schematic representation of the West Nile Virus genome; secondary structures are shown for the 5’- and 3’-UTRs. (B) Secondary structures of the 3’-UTRs of three flaviviruses, highlighting their structural similarities. The Flavivirus 3’-terminal stem-loop is essential for viral replication and highly conserved in secondary structure across species, despite lower sequence conservation. (C) Secondary structure of WNV_3’BSL, the construct studied in the present work, comprising the base of the 3’-UTR terminal stem-loop (sequences were renumbered for the ease of reading), with reported binding sites for cellular proteins indicated. eEF1a (eukaryotic translation elongation factor 1 alpha) and AUF1 (AU-rich element RNA-binding protein 1) are host factors that assist the viral multiplication process and interact with this RNA element.

The WNV 3’-UTR consists of three domains: Domain I, immediately following the stop codon, is the most variable part of the 3’-UTR and contains RNA structures that resist degradation by the host cellular machinery;^11,12 Domain II, moderately conserved, features duplicated dumbbell-shaped (DB) structures that form pseudoknots;¹³ Domain III, the most conserved region, forms a short ~18-nucleotide stem-loop (short hairpin; ‘sHP’) linked to a larger stem-loop of about 80 nucleotides. This characteristic secondary structure of the 3’-terminus, as revealed by SHAPE probing and functional studies,^14–17 is indispensable for viral replication and highly conserved across the Flavivirus family (Fig. 1B). Alterations within it significantly impact viral fitness; for example, single nucleotide substitutions in the sHP can fully abrogate viral replication.^18,19 The 3D structures of these essential regulatory elements within the WNV genome remain largely unexplored. Our work aims to elucidate the structure and dynamic basis of WNV 3’-terminal stem-loop function using nuclear magnetic resonance (NMR) spectroscopy.

The predicted secondary structure of the WNV3’-terminal RNA is shown in Fig. 1B. This domain is notable for containing major and minor binding sites for eEF1A,¹⁹ a cellular translation factor crucial for polyprotein synthesis. The major eEF1A binding site is localized to the C10961-A10964 region on the 5’-side of the terminal stem-loop, contributing approximately 60% of the binding activity, while the minor binding site has been mapped to the conserved hexanucleotide loop (G10935-A10940) of the sHP. Recently, another cellular protein, AUF1 (also known as heterogeneous ribonucleoprotein D or hnRNPD) was reported to bind at the double bulged region of the larger stem-loop and act as an RNA chaperone²⁰ to aid structural rearrangements required to initiate viral replication.

The complete WNV 3’-terminus domain encompasses more than 100 nucleotides, making its structural determination a formidable task for NMR. Therefore, in this study, we focused our investigation on the stem-loop base of the 3’-terminus (WNV_3’BSL, Fig. 1C), the domain core containing the identified binding sites for cellular proteins AUF1 and eEF1a. WNV_3’BSL also includes the sHP, which is crucial for viral replication and has been proposed to interact with the remaining portion of the 3’-terminal stem-loop.²¹ We have characterized the global fold and uncovered local structural details such as the presence of several non-Watson-Crick pairs in WNV_3’BSL, offering insight into features that promote specific binding of various host factors. In particular, we observe the sHP and the lower part of the larger adjacent stem-loop to exhibit a metastable structure, perhaps providing the plasticity necessary for the transition of the RNA genome between linear and circular conformations, thereby mediating the balance of viral replication and translation in the host cells.^22,23 Overall, our work sheds light on the structural attributes of the WNV 3’-terminus, expanding upon our previous work on RNA structures within the flavivirus genome 5’-UTR.^24,25 The resulting enhanced understanding of molecular mechanisms underpinning the WNV lifecycle could facilitate the development of new RNA-targeted antiviral therapies.

2. Results

2.1. Secondary structure and verification of the construct design

The WNV_3’BSL construct, whose residues are renumbered to facilitate the discussion, is 68nts in length and despite its reduced size, remains challenging for study by NMR. Therefore, we employed a divide-and-conquer strategy, in which the conformation of this larger RNA is reconstructed from those of smaller and independently folded structural elements. This strategy works for many RNAs because RNA structures are generally modular, such that their constituent fragments frequently fold the same way both within and outside of the context of the complete parental RNA^24–26. The resulting experimental data provide good support to this approach, as described below.

We thus divided WNV_3’BSL into two smaller fragments, dubbed WNV_SL (corresponding to the longer terminal stem-loop), and WNV_JCT (corresponding to the small hairpin, junction, and lower section of the longer stem-loop. Fig. 2A). Comparison of NMR spectra of the different constructs to illustrate and validate the divide-and-conquer strategy is presented below. The GACA sequence preceding the 5’end of wild type WNV 3’-terminus was mutated to GG in WNV_3’BSL to allow in vitro transcription. For the same reason, the terminal A19-U68 base pair of WNV_3’BSL was changed to G19-C68 in the WNV_SL sequence. Residue A66 is not conserved among all viral isolates but is present in WNV strain E1–1, which was used for most functional studies. We therefore included it, though its conservation pattern suggests it is not important for viral replication.^19,27 In addition, a UUCG tetraloop²⁸ was introduced beyond the eEF1A and AUF1 binding sites to cap the WNV_SL stem-loop (Fig.1C and 2A) and reduce aggregation. Similarly, the U32-A54 base pair in the parental RNA was modified to C32-G54 in the WNV_JCT fragment, to maintain its predicted secondary structure. For each construct, we recorded Nuclear Overhauser Effect (NOESY) and ¹H-¹⁵N Heteronuclear Single Quantum Coherence (HSQC) experiments in 90% H₂O/10% D₂O first, to establish the RNA secondary structure, then recorded a series of 2D NOESY experiments in D₂O to assign the non-exchangeable protons. All NMR spectra were obtained at pH 6.5; to investigate potential protonation events for mismatched base pairs, some experiments were also repeated at pH 5.8 and 5.0.

Figure 2. WNV constructs investigated using the Divide and Conquer Strategy.

(A) Secondary structures of constructs used for the structural determination of WNV_3’BSL. (B) (right) NOESY spectrum of WNV_SL with dashed lines showing sequential NOEs involving imino protons, which report on the RNA secondary structure. (left) Assignment of Uracil H-N3 and Guanine H-N1 resonances in the ¹H-¹⁵N HSQC spectrum recorded in phosphate buffer (pH 6.4), at 288K. (C) Correspondence of imino resonances for WNV_3’BSL (black) and the two smaller constructs, WNV_SL (blue) and WNV_JCT (red). The little red lines highlight peaks from WNV_JCT fragment; the assignments and similarity in the spectra confirm the robustness of the divide and conquer strategy.

Spectra of WNV_SL reveal a well-folded RNA presenting a single dominant conformation. We were able to assign all of the observed NH peaks in the H₂O NOESY spectrum by tracing sequential NOE connectivity and obtained further confirmation of the assignments from the ¹⁵N HSQC experiments, which separate G’s from U’s based on the ¹⁵N shift (Fig. 2B). Most of the imino protons expected from the predicted secondary structures were observed, except for those of residues located at the end of helical stretches (G19, G20, which are expected to undergo rapid exchange with solvent), and of residue U25, which is flanked by an internal loop and UG mismatch. The predicted U24-G62 wobble pair was identified unambiguously from the strong cross peaks between G62H1 and U24H3 within the non-canonical base-pairing range (10–12 ppm)²⁹. Thus, the NMR data fully support the predicted secondary structure of WNV_SL.

The NMR spectra of WNV_JCT and WNV_3’BSL, however, reveal the presence of significant conformational flexibility, accompanied by the observation of unexpected G-U imino protons (Fig. 2C) that were not predicted in the SHAPE-derived secondary structure. The observed motions are most evident within the short stem-loop at the 5’-end of the RNA (nts G1-G18) and will be discussed in greater detail in results section 2.3. Nevertheless, comparing the spectra of WNV_SL and WNV_JCT to that of WNV_3’BSL reveals a high similarity in both chemical shifts and NOE patterns (Fig. 2C, Supplementary Fig. S1). Because chemical shifts are very sensitive to local structures, we conclude that the structures of WNV_SL and WNV_JCT faithfully represent their corresponding structures in the complete RNA.

2.2. Structure of the major host factor binding domain within the WNV RNA

2.2.1. NMR spectral analysis and three-dimensional structure of WNV_SL

Partially deuterated RNA samples were prepared to mitigate the extensive overlap in the region of the ribose protons and reduce cross-relaxation that would compromise linewidths. Partially deuterated (D3′, D4′, D5′/D5′′ and D5, but H6/H8, H1′, H2′) WNV_SL samples^24,30,31 reduced the spectral overlap and sharpened the linewidths (Supplementary Fig. S2), allowing unambiguous assignments of the remaining protons using standard double helical NOE walks.³² Those assignments are crucial for establishing sequential base-stacking patterns. 3D NOESY spectra were then collected for a ¹³C, ¹⁵N labeled sample to identify the ribose H3′, H4′ and H5′/H5′′ protons and further populate the assignment table. Once assignments were substantially complete (100% of the non-exchangeable aromatic protons, namely H5, H6, H8 and Ade H2, 100% of H1’ and H2’, 88% of imino protons, 45% of H3’, 36% of H4’, and 30% of H5’/H5” were assigned), inter-proton distance restraints were derived from the NOE peak intensities as described in Materials and Methods.

A set of 200 refined structures were then calculated for WNV_SL using the Xplor-NIH software^33,34 package (for details refer to Materials and Methods; The complete NMR and structural statistics are shown in Supplementary Table S1.) On average, 20 experimental NOE interactions were identified for each nucleotide of WNV_SL, including 4 long range restraints per residue, permitting us to obtain well-defined local and global folds. The ensemble of the 10 lowest energy structures, all with no NOE violations greater than 0.3 Å, are shown in Fig. 3A. The independently calculated models converge well, with a heavy atom RMSD of only 0.95 Å.

Figure 3. Solution NMR structure of WNV_SL, and overlay with the corresponding SAXS envelope.

(A) Ensemble of the 10 lowest-energy structures of WNV_SL, with the lowest energy structure used to represent the ensemble in image (B). Residues deviating from canonical A-form conformation are identified in magenta and mismatches are highlighted in blue. (C) SAXS envelope of WNV_SL, superposed with the lowest energy structure of the NMR ensemble; residue G53 is highlighted in blue.

The overall shape of the WNV_SL NMR model was further supported by Small-Angle X-ray Scattering (SAXS) experiments (1D scattering profile and analysis of SAXS data are shown in Supplementary Fig. S3.) The ab initio-modeled³⁵ SAXS envelope resembles a drumstick with a maximal distance distribution (D_max) of ~76 Å. This dimension agrees well with the NMR ensemble; the SAXS envelope also reproduced the slight bend in the structure induced by the bulged-in G53 nucleotide (Fig. 3C, highlighted in blue). To cross-validate our NMR structure of WNV_SL, the theoretical scattering curve was back-calculated from the lowest energy NMR model³⁶ and gave good agreement to the experimental scattering data (χ² =7.06).

The NMR ensemble and the lowest energy structure of WNV_SL present an extended helix, interrupted by several bulges and mismatches which are highlighted in blue in Fig. 3B. While the WNV genomic sequence for this region is not well-conserved, these structural features are conserved in West Nile Virus isolates and other flaviviruses (Fig. 1B). All of the sequential NOEs expected from the sugar protons (mainly H2’, H1’) to the bases (H6/H8) were observed (examples are shown in Supplementary Fig. S4). Additionally, all of the H2s of Adenines were assigned and showed distinguishable cross-peaks with two H1’ protons, sequentially and across the strand, as expected for A-U base pairs in A-form helices (Fig. 4A and 4C). The observation of these peaks confirmed the presence of the U25-A61 base pair, although the corresponding U25 imino proton was not observed in the H₂O spectra due to fast exchange with solvent, revealing the dynamic nature of the C26-C60 mismatch. All sequential aromatic-aromatic H6/H6, H6/H8 and H8/H8 NOEs were identified for WNV_SL (Supplementary Fig. S5), as were most H6/H8 and H5 NOEs expected for stacked bases, and all sequential H1’/H1’, H5/H1’ interactions (data not shown). These observations demonstrate the retention of helical base stacking throughout the bulges and internal loops, supporting the possible formation of hydrogen-bonded mismatched pairs in those regions. Indeed, in addition to the predicted GU wobble pair, several other non-Watson-Crick pairs (Fig. 4B) were discovered within the WNV 3’-terminus stem upon further investigation.

Figure 4. Characteristic helical Adenine H2 NOE patterns.

(A) The characteristic H2 to H1’ cross-peaks observed for WNV_SL Adenines involved in A-form helical stacking and base-pairing; here explicitly indicated only for well-isolated signals in the spectrum. (B) The predicted secondary structure of WNV_SL, with observed non-Watson-Crick base pairs highlighted by red dashed lines. (C) The signal observed for A57-H2 was substantially downfield-shifted, suggesting the possibility of non-canonical base pairing, and in particular the possible formation of a protonated A⁺C base pair.

2.2.2. Mismatched base pairs

Notably, the spectra indicate that residue A57 is protonated at the N1 position, leading to formation of a protonated C29-A⁺57 mismatched base pair in the lower helix of WNV_SL, stabilized by two hydrogen bonds between C29N3-A57⁺H61 and C29O2-A57⁺H1 (Fig. 5B). This possibility was first suggested by the chemical shift of A57 H2, which was significantly downfield compared to all other Adenine H2s (Fig. 4C), consistent with previous reports of protonated adenine nucleotides.^37,38 Additionally, the C2 peak of A⁺57 was significantly upfield-shifted and broad at pH 6.5, but sharpened increasingly as the environment was made more acidic (Fig. 5C).

Figure 5. NMR Analysis for CA⁺ mismatch.

Close-up view of the C29 and A⁺57 mismatch, taken from ensemble of 10 structures (A) and as viewed from above in the lowest energy structure (B). This CA⁺ mismatch is stablized by two hydrogen bonds, indicated here by red dashes. (C) ¹H-¹³C HMQC spectra of WNV_SL collected at pH 6.4 (red), 5.8 (green) and 5 (blue), focusing on the C2 region. Assignments refer to the WNV_SL spectrum at pH 6.4; the C2-H2 peak for loop nucleotide A37 was only observed when it was protonated at pH = 5.

While the A57⁺H1 proton was not observed in the imino region, probably due to rapid exchange with solvent, the addition of hydrogen bonding restraints for C29-A⁺57 in structure calculations did not produce clashes with existing NOE assignments, and the 10 lowest energy structures converged well within this region (Fig. 5A). While the pKa of A57 is below physiological pH, this base pair would be partially protonated as predicted by simple acid-base chemistry. The C29-A⁺57 base pair could then be recognized by cellular factors such as AUF1. This suggestion is consistent with previous reports that replacing the C29-A57 apposition with C29-G57 or G29-A57 is tolerated, but the flexible C26-C60 loop is required for efficient viral replication.^19,22

The canonical U32-A54 base pair was found to coexist in equilibrium with a U32-G53 wobble pair, with the latter being favored at lower pH, as shown in Supplementary Fig. S6. Notably, the N3H3 peak of U32 from the U32-G53 pair is only detected at pH 5.8 (Supplementary Fig. S7), and no NOE contacts are observed for the U32H3 proton, presumably because of rapid exchange with solvent. Likewise, the N1H1 peak of G53 was distinctly weak (Fig. 2B) and inconsistently present in the ¹H-¹⁵N HSQC experiments. Upon studying the lowest-energy structures for the G53 bulge, it was evident that the H3 and O2 of U32 could potentially form hydrogen bonds with the O6 and H1 of G53, even when planarity and hydrogen bonding restraints for a base pairing of U32 with A54 had been applied (Fig. 6A). Moreover, G53 is not a conserved nucleotide across West Nile Virus variants, with some instances of mutation to U53,³⁹ which would maintain the bulged-in and inconsistently base-paired structure we find for WNV_SL. The temperature and pH sensitivity of the GU wobble pair we observe here underscore the adaptability of non-Watson-Crick pairs in biological contexts, which might modulate protein binding.^40,41

Figure 6. Examination of the GA and distorted GU mismatches.

(A) A close-up view of the single bulged-in nucleotide G53, as seen in the lowest-energy NMR structure. The possible hydrogen bonding or electrostatic interactions which would promote formation of a distorted U32-G53 wobble pair are indicated by red dashes. (B) Left: A close-up view of the A37-G48 mismatch, as seen in the lowest energy NMR structure. The cross-strand C1’-C1’ distances are represented by black dashed lines. Right: Examination of the GA mismatch reveals the likely formation of two hydrogen bonds: A37H61-G48O6 and A37N1-G48H1.

Direct evidence of hydrogen bonding was not observed for the C26:C60 or A37:G48 mismatches. Because the major eEF1A binding site has been mapped to residues C34-A35-C36-A37 of the WNV 3’ stem-loop, the region containing the A37-G48 mismatch was examined more closely in the lowest energy NMR structures (Figure 6B.) Here the structure converges well, with the bases of A37 and G48 stacked between the neighboring G49-C36 and A38-U47 base pairs (Supplementary Fig. S8). Typically, in A-form helices, the cross-strand C1’-C1’ distances average about 10.5 Å. However, the A37-G48 mismatch introduces a local irregularity, characterized by a wider groove with a C1’-C1’ distance of 11.8 Å. This deviation may play a role in facilitating specific recognition of this region by eEF1A. Looking at the A37-G48 mismatch from above, the H61 and N1 of A37 and O6 and H1 of G48 point at each other and are in close proximity (about 2 Å), strongly suggesting formation of the corresponding hydrogen bonds, even without direct NMR evidence. To provide additional evidence supporting formation of the non-Watson-Crick base pairs, NOEs observed for the A37-G48 and C29-A⁺57 appositions are labeled in Supplementary Fig. S9. However, we notice that there are not significant deviations for these pairs, relative to regular W-C pairs, with the exceptions of NOEs involving H2 of protonated Adenines which have been noted in Figure 4. The imino protons of non-canonical base pairs were not observed, due to rapid exchange with the solvent.

We also assessed RNA dynamics by examining the sugar conformation through Total Correlation spectroscopy (TOCSY) experiments.^25,32 For a 3’-endo sugar pucker, as observed in A-form RNA helices, the H1’-H2’ coupling is very small, thus no peaks are observed in the anomeric region of TOCSY spectra. The 2’-endo conformation observed in single stranded loops and in B-form DNA helices gives rise to strong peaks for both H1’-H2’ and H1’-H3’, but not to the H4’ because the H3’-H4’ coupling is very small. If dynamic equilibrium is present, the H1’ has TOCSY correlations to H2’, H3’ and H4’, as well, due to averaging of the scalar couplings. Based on this analysis, most nucleotides in WNV_SL adopt a 3’-endo sugar pucker, as expected for A-form helical RNA. Besides the loop nucleotides U42 and C43, which are known to occupy a 2’-endo conformation within the UUCG tetraloop,²⁸ the sugar puckers of G20 and A54 adopt a mixture of 3’- and 2’-endo conformations. Since the C2’-endo sugar pucker extends the phosphate backbone⁴², this arrangement might allow the unusual bulge conformation to form, while minimizing backbone distortions for nucleotides A66 and G53.

2.3. Dynamic structure of the tail and lower stems of WNV_3’BSL

2.3.1. The short hairpin samples multiple conformations

Transferring the assignments of nucleotides U22 to U68 from WNV_SL to WNV_3’BSL was straightforward because of the spectral match of these two RNAs in both peak resonances and NOE patterns (Fig. 2C, Supplementary Fig. S10). Mismatched pairs of WNV_SL were faithfully reproduced in spectra for the stem of WNV_3’BSL. However, very few NOEs were observed in the spectrum of a deuterated WNV_3’BSL sample for nucleotides within the sHP. Furthermore, inspection of the H₂O NOESY spectrum of WNV_3’BSL reveals two sets of resonances for the single predicted G6-U15 wobble pair. The same pattern was observed for the WNV_JCT fragment (Fig. 2C), suggesting that this 18-nucleotide side arm stem loop adopts two conformations in equilibrium with each other. This was surprising, since the stem-loop was predicted to be independently and stably folded.

This dynamic complexity, coupled with the slower tumbling and more severe spectral overlap of the 68nt long RNA, makes NOE identification very difficult. Thus, we focused on the smaller WNV_JCT fragment, again following the divide-and-conquer strategy, to increase the number of restraints for structure determination and establish the nature of this dynamic exchange. The spectral assignments were completed for the major conformation of WNV_JCT, which is populated more than 75% of the time based on the intensity of the G6 imino peak; only protons from this conformation gave traceable NOEs. The pattern of NOEs observed for the major conformation is consistent with its predicted secondary structure (Fig. 7, Supplementary Fig. S11). Distance and torsion angle restraints for nucleotides G1 to A21 were therefore extracted from data on the WNV_JCT segment and added to the WNV_SL region of WNV_3’BSL to generate hybrid NOE constraint tables^24,43 for structure calculations with Xplor-NIH. The NMR and resulting structural statistics for WNV_3’BSL are shown in Supplementary Table S1.

Figure 7. Sequential NOE walk for WNV_JCT.

The H2’ to H6/H8 NOE helical walk, labelled from G1 to G20, for the deuterated RNA WNV_JCT construct. The spectral quality is not high due to the dynamic nature of this segment containing the short hairpin arm. The characteristic H1’ to H6/H8 sequential pattern is shown in Supplementary Fig. S11.

2.3.2. The small hairpin sHP is a flexible side arm attached to the main stem-loop structure

The NMR models of WNV_3’BSL converge well for nucleotides C23 to G63 (RMSD = 1.36 Å), but the hairpin side arm is poorly defined (Fig. 8A, Supplementary Table S1). This lack of structural definition around the helical junction is consistent with our expectations, considering that NOE signals were detected only for nucleotides C3 and G18 to A19, but not among other nucleotides located where the helices come together. Consequently, the orientation of the small hairpin (C3-G18) in relation to the larger stem-loop WNV_SL (A19-U68) is determined only by a few experimental constraints. We also evaluated the alternative G1-U68 and G2-C67 base pairs in the tandem junction, which might contribute to the complexity of motions around the junction. Modeling the WNV_3’BSL structure imposing artificial G1-U68, G2-C67 base pairing restraints, however, suggested that residue U68 and C67 preferably position their bases parallel to that of A19 and G20 (Supplementary Fig. S12), in order to satisfy the NOE connections. Also considering the fact that NOE patterns for residue C67 and U68 of WNV_3’BSL match well with that of WNV_SL, the construct without the side arm, we concluded that the terminal U68, C67 most likely pair with A19, G20 and that these potential interactions don’t add to the structural flexibility.

Figure 8. Superposition of lowest-energy NMR structures on SAXS envelope for WNV_3’BSL.

(A) Ensemble of the 10 lowest energy structures of WNV_3’BSL. Structures of the flexible small hairpin are shown with high transparency to favor visualization, except for the average structure which is shown explicitly in a stick and ribbon representation. (B) SAXS envelope model for WNV_3’BSL with Dmax=100 Å, superposed with the average structure of the NMR ensemble.

SAXS experiments were thus conducted to generate low-resolution models of WNV_3’BSL to assess the position of the small hairpin relative to the rest of the structure. The scattering profiles of WNV_3’BSL indicated aggregation at concentrations above 0.5 mg/ml, but interpretable data were obtained at lower concentrations (Supplementary Fig. S13). The optimal maximal distance distribution (Dmax) was found to range between 85 Å and 100 Å. Ab initio-modeled SAXS envelopes, with varying Dmax, consistently suggested a spoon-like shape for WNV_3’BSL, in which part of the structure appears to sample a large conformational space, with variance in structural length and overall dimensions arising from one end of the envelope. This envelope aligns well with the ensemble of NMR-derived structures for WNV_3’BSL, underscoring the plasticity of the short side arm. The best fit between the mean structure of the NMR ensemble and its corresponding SAXS envelope is shown in Fig. 8B.

The structures of nucleotides within the short hairpin, C3 to U22, align reasonably well (RMSD = 1.74 Å), but only locally (Supplementary Fig. S14), i.e. within the short stem-loop. When examining the hexanucleotide apical loop (G8 to A13), H1’ and H2’ to H6 and H8 NOE peaks were observed for each of these nucleotides, although at low intensity (Fig. 7, Supplementary Fig. S11), consistent with dynamic equilibrium. Additionally, structures of the apical loop region suggest that it is substantially flexible, unlike the well-converged helix region (Fig. 9). This conformational plasticity is not unexpected, as the loop sequence shares similarities with a GNRA tetra-loop motif (where R represents a purine and N any nucleotide). Such motifs are known for their structural variability,^44,45 enabling different RNA-RNA and RNA-protein interactions to be mediated by this motif. The noted motion for the apical loop (G8-A13) and junction of WNV_3’BSL was further supported by previous SHAPE results.¹⁴ Specifically, the residues which display high reactivity in SHAPE also show high levels of dynamics in the NMR analysis (Supplementary Fig. S15).

Figure 9. Conformational dynamics of the WNV_JCT apical loop.

A close-up view of the WNV_JCT apical loop (A) and of WNV_JCT helix (B), with the former showing considerable flexibility in contrast to the well-converged helical region.

A previous in-vitro study²¹ of WNV_JCT postulated a tertiary interaction between nucleotides A11-A13 and U25-U27 (highlighted by red dashes in Supplementary Fig. S16), but this hypothesis is not supported by our experimental data. The proton chemical shifts of nucleotides U25-U27 in the WNV_SL fragment closely match those observed in the WNV_3’BSL spectrum (Fig. 2, Supplementary Fig. S10), indicating that they exist in very similar chemical environments. This similarity implies these nucleotides do not form a pseudoknot with the hairpin sequence, as was proposed. Moreover, it can clearly be seen in the SAXS envelope (Fig. 8B) that the small hairpin loop is located far from the terminal stem loop. This conclusion was supported by a recent work modeling the 3D shape of long non-coding flavivirus sgRNAs using SAXS profiles,¹⁴ although their reconstruction of WNV3’-terminal RNA yielded an even more elongated rod-like envelope. Earlier work¹⁹ reported 1D NMR studies on an RNA element similar to our WNV_3’BSL construct. Although the primary conclusion of that study was the absence of some expected base pairs, it also reported conformational flexibility of the apical loop (G8-A13) sequence and the C26-C60 bulge, as well as the absence of the proposed pseudo-knotted interaction. Since Mg²⁺ is known to stabilize RNA structures and facilitate tertiary interactions,^46–48 we introduced up to 6 equivalents of MgCl₂ to both NMR samples (Supplementary Fig. S16), a condition that had never been investigated, and conducted TOCSY experiments to further probe the possible formation of a pseudoknot. However, the observed chemical shift changes for the H5 and H6 protons of all nucleotides were less than 0.1 ppm, consistent with small differences in conformation but providing no evidence of pseudoknot formation between the two subdomains of WNV_3’BSL, even in the presence of a large excess of Mg²⁺.

To assess the reliability of our RNA structure results, we ran simulations using Rosetta FARFAR (Fragment Assembly of RNA with Full Atom Refinement). As its name implies, this method synergizes the approach of fragment assembly with the accuracy of full-atom refinement, enabling more precise and adaptable modeling of RNA structures.⁴⁹ An ensemble of the 10 lowest-energy FARFAR conformers, predicted based on the WNV_3’BSL sequence and its established secondary structure, is depicted in Supplementary Fig. S17. Mirroring our NMR ensemble, the modeled WNV_3’BSL structures exhibit good convergence (RMSD = 2.38 Å) in the terminal stem, while the sHP side arm remains disordered. Notably, the modeled WNV_3’BSL RNA appears more compact than its experimentally determined counterpart and shows several differences in specific structural details. For example, in many FARFAR models, including the lowest energy structure, G53 is looped out, which is in contrast with our experimental findings. The C29 and A57 mismatches point away from each other in most models, reflecting weakness in predicting non-Watson-Crick pairs by current RNA prediction tools.⁵⁰ The conformation of the well-characterized UUCG tetraloop was also not consistently reproduced. Discrepancies in local structural details such as these, however, likely reflect the limited availability of RNA structures in the training-set that contain non-canonical elements such as these. Therefore, more experimentally-determined structures are still needed to better predict and understand the characteristics of energetically favorable RNA folds.^50–52

3. Discussion

The single-stranded RNA genomes of flaviviruses serve dual functions: as templates for synthesizing complementary minus-strand RNAs (3’–5’ direction) and as viral mRNA for protein synthesis (5’−3’ direction). These processes must be rigorously regulated to ensure the progression of viral infection within host cells. Throughout their life cycles, flavivirus genomes modify their secondary and tertiary structures to perform necessary functions. It is well-documented that flavivirus replication involves the cyclization of the genome, mediated by long-range RNA-RNA interactions through complementary sequences in the 5’- and 3’-UTR regions (Supplementary Fig. S18). Specifically, the 5’-UTR of the WNV genome contains conserved subdomains, such as stem-loop A (SLA) and stem-loop B (SLB), with SLB harboring a sequence that complements the C3-G28 sequence in WNV_3’BSL. As depicted in Fig. S18, RNA circularization partially opens the 3’SL helix, precisely where structural flexibility has been observed in the NMR ensemble discussed in this work. Cyclization enables elements required for replication, yet positioned far apart in the linear genome, to interact in the cyclized form. For instance, the RNA-dependent RNA polymerase (RdRp) initially binds to SLA at the WNV 5’-end, but subsequently transfers to the 3’-terminal end upon genome circularization to initiate the RNA multiplication.⁵³ Moreover, genomic cyclization is reported to inhibit viral translation initiation.⁵⁴ The findings that eEF1A binds to multiple sites on the 3’-terminus of the WNV genome suggests a role for this protein to facilitate RNA-RNA interactions between the 5’-end of the RNA and nucleotides within the bottom helix of WNV_3’BSL.⁷ Because the eEF1A elongation factor was also found to co-immunoprecipitate with viral replication complex (RC), it was proposed to promote interactions between the 3’end of WNV genome and the RC apparatus, perhaps to optimally orient the RNA template for minus-strand synthesis.²⁷

While genome cyclization is indispensable for viral replication, the linear form is translation-competent.^7,54 Single nucleotide substitutions (e.g. at G12) in the sHP apical loop completely abrogate viral RNA replication,¹⁹ as do mutations in WNV_SL that either stabilize the bottom helices or prematurely expose the single stranded nucleotides,^19,22 demonstrating that a subtle balance between different structures of WNV_3’BSL is required.^23,55 Following infection and several rounds of translation, genome cyclization “shuts-off” the process and facilitates exponential RNA synthesis.^7,56 The observed conformational plasticity in the tail and lower part of WNV_3’BSL could potentially regulate the stability of circularization, controlling the balance between the circular and linear forms of the viral genome, so that the translational “shut-off” is not permanent if circularization occurs before sufficient production of viral proteins in the host cells. The mechanism by which the genome’s 5’ and 3’ ends are brought together and the initiation of base pairing between the complementary sequences are still not clear. However, it is plausible to suggest that small molecules interacting with this structure could modify the efficiency or timing of genome cyclization, thus affecting viral replication. The structure we have determined suggests that the junction between the small hairpin and the tail of the longer stem-loop could provide a binding pocket for small molecules,^57–59 and that targeting this region could stabilize the RNA conformation and in turn hinder the conformational rearrangement or breathing of the structure, in the same way that stabilizing nucleotide substitutions have been shown to do this.¹⁹

In summary, we have established the three-dimensional structure of the West Nile Virus 3’-terminal core (WNV_3’BSL), a fairly large RNA element for NMR studies, utilizing the “divide-and-conquer strategy”. This structure likely represents a common conformation conserved across the flavivirus family, and is required for viral replication. Several non-Watson-Crick pairs were discovered within the helical structure, providing valuable insights for interpreting existing biochemical data obtained from mutational studies of flaviviruses.^15,19,22 This research marks an initial step toward broadening our structural analysis across other conserved flavivirus domains. Furthermore, it is increasingly acknowledged that the dynamic bulge/loop regions of conserved functional RNAs can be of great interest for rational drug design.^57–61 Our report opens avenues for identifying small molecules that target these metastable RNA structures, offering the potential for innovative antiviral strategies against flavivirus-related diseases.

Materials and Methods:

RNA preparation

All RNA samples (as shown in Figure 2A) were transcribed in vitro using an in-house purified T7 RNA polymerase and synthetic DNA templates containing the T7 promoter, following established standard methods.^62–64 The templates were procured from Integrated DNA Technologies. Labeled RNA samples were synthesized using either partially deuterated rNTPs (deuteration of H5, H3’, H4’, H5’ and H5” protons), or fully ¹³C-¹⁵N labeled rNTPs, which were purchased from Cambridge Isotope Laboratories.

The resulting RNA stock solutions were purified according to standard methods (20% denaturing polyacrylamide gel electrophoresis (PAGE), followed by electroelution, ethanol precipitation and extensive dialysis.)⁶⁵ UUCG Tetraloops were introduced in place of existing loops in the sequences of the three RNA sequences studied here to generate well folded constructs for NMR studies, as done previously.^24,25,28 Tetraloops are known to improve the spectral quality considerably by stabilizing the secondary structure and reducing aggregation at the high concentrations required for NMR.^29,32

NMR spectroscopy

Most RNA samples for NMR investigations were prepared in 20mM potassium phosphate buffer (pH = 6.4 or 5.8) containing 0.001 mM EDTA and 10um DSA as internal standard. Sodium Acetate buffer was used for RNA samples with pH = 5. The final concentrations of RNA samples used for NMR experiments were between 0.7mM and 1.2mM. All samples were annealed by heating at 90–95 °C for 3 minutes, followed by immediate snap cooling on ice for 10–20 minutes, prior to NMR data recording. To study exchangeable protons, RNA samples were dissolved in 95% H2O buffer/5% D2O, while for examination of non-exchangeable protons, RNA samples were lyophilized and re-dissolved in 99.99% D2O.

All NMR experiments were carried out at 310K (D2O samples) and 288K (95%H2O+5%D2O samples) on Bruker Avance 800 or/and 600 MHz NMR spectrometers equipped with 1H/13C/15N triple resonance (TCI) cryoprobes and triple-axis gradients. A few experiments were also recorded on a Bruker Avance III 700 MHz with a room temperature BBO probe. The 1D ¹H spectra were recorded using excitation sculpting for water suppression. 2D homonuclear total correlation spectroscopy (TOCSY) spectra were recorded with mixing times of 80 ms using MLEV17 mixing. All 2D Nuclear Overhauser Enhancement Spectroscopy (NOESY) spectra were recorded at different mixing times (100, 200, and 300 ms) to permit more accurate quantitative evaluation of internuclear distances by comparing peak intensities for pairs of protons with fixed distances and to facilitate spectral assignments. Spectra for ²H labeled samples were collected in the same manner. 3D ¹³C-NOESY−HSQC, 2D ¹H-¹⁵N and ¹H-¹³C HSQC/HMQC spectra were recorded as well to assign the sugar protons with confidence and to further confirm the assignments, when needed. All NMR data were processed with Bruker TopSpin (3.6.3) and visualized/analyzed in NMRFAM-SPARKY.⁶⁶ Assignments of RNA spectra were guided by predicted RNA chemical shift values, and based on well-established double-helical sequential NOE patterns.^29,67

Residual dipolar couplings (RDCs) experiments can provide information about global features of RNA structure and are generally complementary to NOE-derived information, which only provide local information. However, we did not collect RDC’s for this project because, in the presence of conformational flexibility, RDC’s would be averaged differently for different segments of the structure, requiring different orientation tensors for each rigid sub-structure. Thus, RDCs would require a very extensive analysis and be subject to data over-interpretation, and might in the end not add much to the structural analysis. While SAXS is less rich in information, it is a more robust structure validation tool.

Small-angle X-ray scattering

RNA samples for SAXS were prepared similarly to samples made for NMR studies, but at lower concentrations (1–4 mg/ml), and dissolved in 20 mM Tris, 100 mM NaCl, and 0.1 mM EDTA (pH 6.5). SAXS experiments were recorded on a BioSAXS-2000 SAXS instrument at Argonne National Laboratory. Scattering data was processed by RAW⁶⁸ and the Particle distance distribution functions P(r) were calculated using GNOM⁶⁹ and used for low resolution initial shape constructions with DAMMIF.³⁵ A total of 15 models were generated with DAMMIF using the ATSAS online server.⁷⁰ The representative models were processed with a suite of software (DAMSEL, DAMSUP, DAMAVER and DAMFILT) for comparison and fitting to the NMR structure.⁷¹ The theoretical scattering curve was back-calculated from the NMR model by the RAW built-in method DENSS PAD2SAS.⁷²

Experimental restraints and structure calculation

Interproton distance restraints were derived from NOE peak assignments in 2D NOESY spectra and were sorted into strong (2.5 ±0.7 Å), medium (3.5 ±1.2 Å), and weak (4.5 ±1.5 Å) bins, based on peak intensities relative to fixed reference distances (e.g., H5–H6= 2.5 Å, H3′–H6/H8=3.5 Å). Base-pair planarity and hydrogen-bonding constraints were applied for unambiguously assigned base pairs, as identified by 2D NOESY NH protons. Dihedral restraints were only included for base-paired nucleotides that reside in helices and conform to A-form helical structures, as established from the pattern of NOE backbone walks. Backbone and ribose dihedral angle restraints were determined by qualitative analysis of the through-bond 2D TOCSY experiments. Sugar puckers were constrained to C3’-endo for all ribose sugars in the stem which showed strong NH cross peaks in the water NOESY but without appreciable H1’−H2’ cross-peaks in 2D 1H-1H TOCSY spectra. Residues with C2’-endo pucker were identified by strong H1’−H2’and H1’−H3’ correlations in TOCSY spectra. The remaining backbone torsion angles were left un-restrained due to lack of sufficient information or conformational flexibility. Restraints for overlapped regions were obtained from the 3D NOESY−HSQC spectrum recorded at 100 ms mixing time acquired at 800 MHz. A total of 1367 NMR restraints for WNV_SL (an average of 27.4 restraints per nt) and 1664 NMR restraints for WNV_3’BSL (an average of 24.5 restraints per nt) were used in the final calculation steps.

A set of 200 refined structures for each NMR construct of WNV were calculated using Xplor-NIH following a simulated annealing protocol.^33,34 Refinement included NOE and hydrogen bond distance constraints, and dihedral torsion angle restraints. Starting from randomized coordinates, high-temperature torsion angle dynamics were performed, initially folding the structure using NOE and van der Waals terms. After this, base-pair planarity and hydrogen-bonding restraints were introduced. The system was gradually cooled from 3,500 to 298 K, with incremental force constant increases for various terms. Final refinement involved two Powell minimization steps. The 10 lowest-energy structures were selected from 200 calculations following NOESY data and predicted A-type helical base-pair values.

Highlights:

1. The high-resolution three-dimensional NMR structure of the 68 nucleotides West Nile Virus 3’ terminal stem-loop presents a long stem-loop punctuated by non-Watson-Crick base pairs and connected to a flexible smaller loop by a stacked helical junction.

2. This structure is very likely to represent a functional element conserved across the flavivirus family.

3. The small hairpin at the 5’-end of the structure, where even single nucleotide mutations abolish viral replication, is conformationally flexible and forms a metastable junction with the main RNA stem-loop

4. The junction between the small hairpin and the tail of the longer stem-loop provides a binding pocket for small molecules to stabilize the structure and interfere with the conformational rearrangements required for viral replication.

5. The NMR analysis, supported by the SAXS envelop, confirms the absence of putative tertiary interactions between the small hairpin and terminal stem loop.

Data Availability: Accession numbers

The coordinates of WNV_SL construct and WNV_3’BSL are deposited in the Protein Data Bank (PDB) (https://www.rcsb.org/) under the accession codes (PDB ID) 9BCI and 9BLM. Their chemical shifts have been deposited in the BMRB with ID 31165 and 31171, respectively.