An exon-biased biophysical approach and NMR spectroscopy define the secondary structure of a conserved helical element within the HOTAIR long non-coding RNA

Abstract

HOTAIR is a large, multi-exon spliced non-coding RNA proposed to function as a molecular scaffold and competes with chromatin to bind to histone modification enzymes. Previous sequence analysis and biochemical experiments identified potential conserved regions and characterized the full length HOTAIR secondary structure. Here, we examine the thermodynamic folding properties and structural propensity of the individual exonic regions of HOTAIR using an array of biophysical methods and NMR spectroscopy. We demonstrate that different exons of HOTAIR contain variable degrees of heterogeneity, and identify one exonic region, exon 4, that adopts a stable and compact fold under low magnesium concentrations. Close agreement of NMR spectroscopy and chemical probing unambiguously confirm conserved base pair interactions within the structural element, termed helix 10 of exon 4, located within domain I of human HOTAIR. This combined exon-biased and integrated biophysical approach introduces a new strategy to examine conformational heterogeneity in lncRNAs and emphasizes NMR as a key method to validate base pair interactions and corroborate large RNA secondary structures.

Graphical Abstract

Introduction:

Long noncoding RNA (lncRNA) are dynamic modulators of gene expression and can function as molecular scaffolds, associating with chromatin modifying complexes near genomic loci to influence chromatin structure and gene expression (Spitale et al. 2011; Rinn and Chang 2012; Marchese et al. 2017). In terms of gene architecture and sequence similarity, most lncRNAs do not have identifiable homologs but there are thousands of human lncRNAs that have defined homologs and share similar expression levels in vertebrate genomes. Comparative transcriptome studies show that some lncRNAs exhibit a varying degree of sequence conservation across short nucleotide stretches that reside within exonic regions, suggesting that conserved functions require only short sequence or structural regions that can be tolerated within the syntenic architecture (Hezroni et al. 2015; Quinn et al. 2016; Ulitsky 2016).

In addition, many homologous lncRNAs are alternatively spliced at levels approaching that of mRNAs and are thought to have rapidly evolved to acquire a functional importance (Schorderet and Duboule 2011; Haerty and Ponting 2015; Ulitsky 2016). During evolution, alternative splicing and exon number globally increase, while exon length decreases (Koralewski and Krutovsky 2011; Haerty and Ponting 2015; Lin et al. 2016). This notion is best illustrated in the analysis of tertiary-quaternary structure in protein coding genes, where protein domain boundaries can correlate with exon boundaries (Richardson 1981; Liu and Grigoriev 2004). These generalities suggest that exon boundaries can stabilize or destabilize RNA elements and demarcate regions within large RNAs that contain a defined tertiary fold.

We sought to explore the correlation of gene sequence elements and domain boundaries within the context of lncRNAs and developed an RNA secondary structure determination strategy termed Exon-Biased structure probing. This method assumes that a secondary structure fold can be contained within an exonic region of the lncRNA. As an example, we analyzed the heterogeneity and folding properties of individual exons of the human HOX transcript antisense RNA (hHOTAIR) using native gel electrophoresis, thermal melting, and analytical ultracentrifugation. In addition, we interrogated the most stable and homogenously folded hHOTAIR exon in vitro using chemical probing and NMR spectroscopy.

HOTAIR is a classic example of a lncRNA involved in silencing specific homeotic genes in embryonic stem cells and whose overexpression is associated with tumor metastasis and poor prognosis (Rinn et al. 2007; Gupta et al. 2010). Human HOTAIR primarily contains six exons and was identified to function in trans, influencing the transcriptional repression of a distant chromosomal domain (Rinn et al. 2007; Gupta et al. 2010; Tsai et al. 2010). Based upon the UCSC genome browser, targeted RNA-sequencing of hHOTAIR has revealed the existence of at least six isoforms, with alternative splicing events generating an additional 16 different isoforms (Kent et al. 2002). Although the physiological regulation of hHOTAIR alternative splicing is unknown, it is possible that different splice isoforms could impact the tertiary fold of the RNA, potential RNA-protein interactions, and the extent of transcriptional repression.

We selected hHOTAIR using an exon-biased mapping approach because its RNA secondary structure has been extensively studied in vitro (Kertesz et al. 2010; He et al. 2011; Wu et al. 2013; Somarowthu et al. 2015; Portoso et al. 2017; Spokoini-Stern et al. 2020). The full-length hHOTAIR has been defined using multiple chemical probing strategies, identifying four large domains (Domain I-IV) that contain specific structured regions that are proposed to contain high sequence co-variation (Somarowthu et al. 2015). Domain I has the highest degree of covariation support when compared to other regions of hHOTAIR, yet some R-scape and power covariation studies suggest that hHOTAIR does not contain any evolutionarily conserved RNA structures (Somarowthu et al. 2015; Rivas et al. 2017; Rivas et al. 2020). In this paper, we show that at least one region of hHOTAIR can adopt a homogenous tertiary fold and represents a structural domain that is preserved within an exonic boundary of hHOTAIR, suggestive of an evolutionarily conserved RNA structure. In addition, we propose that this exon-focused biophysical approach, when combined with hybrid structural bioinformatic studies, may serve as a generalizable strategy to examine the evolution of conserved lncRNA secondary structure within mammalian genomes.

Materials and methods

2.1. HOTAIR RNA synthesis and purification

Double stranded (ds) DNA templates for each HOTAIR exon were generated from the pLZRS-HOTAIR plasmid (H. Chang, Addgene Plasmid ID #26110) (Gupta et al. 2010) by polymerase chain-reaction (PCR) with primer pairs as listed in Table S1. Forward primers contain a T7 RNA Polymerase binding site to initiate in vitro transcription reactions (Milligan et al. 1987). PCR reactions were purified via spin-column (Qiagen), eluted in RNAse-free water, and quantified by UV absorbance at 260 nm.

In vitro transcription reactions contained 1 μM template DNA derived from PCR products of exon boundaries (Figure S1), 40 mM Tris pH 8.0, 38 mM MgCl₂, 10 mM dithiothreitol (DTT), 1 mM spermidine, 0.01% (v/v) Triton X-100, and 5 mM each rNTP (A, C, G, U), 0.4 U/μL RNasin (Promega). After addition of T7 RNA Polymerase to a final concentration of 0.05 mg/mL, reactions were incubated at 37 °C for 4 hours. DNA template was removed by the addition of 5 U RNase-free DNase I (New England Biolabs) and continued incubation at 37 °C for 30min. For native gel analysis, an aliquot was removed at this step and buffer exchanged via a Micro Bio-Spin P-6 column (Bio-Rad Laboratories) equilibrated in 10 mM HEPES pH 7.4, 100 mM KCl, and 5 mM MgCl₂. The remainder of the transcription reaction was ethanol precipitated and air-dried overnight. RNA pellets were resuspended in denaturing loading buffer containing 8M urea, 1x TBE (90 mM tris-borate, 2 mM EDTA, pH 8.3) and xylene cyanol with bromophenol blue, and electrophoresed on 6% polyacrylamide (29:1) gels containing 7 M urea and 1x TBE. Gel slices containing correctly-sized RNAs were excised under UV shadow and passively eluted overnight into 300mM sodium acetate buffer (pH 5.5) at 4 °C and filtered (0.45 μm) to remove polyacrylamide. RNAs were ethanol precipitated and stored dry at −20°C until use.

2.2. RNA Folding

RNAs were resuspended in RNase-free water and concentrations measured by UV absorbance at 260 nm. RNAs (typically between 1–5 μM, depending on application) were distributed to thin-walled PCR tubes, placed in a thermal-cycler with a heated lid and incubated at 95 °C for 2 minutes, 55 °C for 5 minutes, and 30 °C for 15 minutes. At the 55 °C step, HEPES pH 7.4 (10 mM final) and KCl (100 mM final) were added from concentrated stocks, along with concentrations of 0–40 mM MgCl₂ for various folding and thermostability assay. Folded RNAs were chilled on ice for 30 minutes prior to use.

2.3. Native gel analysis

RNAs corresponding to the first five exons of a 2148 nt hHOTAIR construct were prepared (Figure S1, Addgene Plasmid ID #26110) (Gupta et al. 2010). An aliquot representing non-denatured RNA (Chillón et al. 2015; Somarowthu et al. 2015) was reserved and the remainder of the reaction purified via denaturing polyacrylamide gel electrophoresis (PAGE). Non-denatured transcripts and RNAs were refolded at increasing magnesium concentrations (0–5 mM MgCl₂) and were analyzed on a non-denaturing polyacrylamide gel to assay conformational heterogeneity (Woodson and Koculi 2009). The selected Mg²⁺ concentrations reflect conditions where helix assembly and conformational changes typically occur in RNAs (<150 nts) and represent physiological conditions (0.5 – 3.0 mM free Mg²⁺) (Draper et al. 2005; Woodson 2010).

10 pmol folded RNA/lane was electrophoresed on 8% polyacrylamide at 100 V for 60 minutes in THE (34 mM Tris, 66 mM HEPES, 0.1 mM EDTA, pH 7.4) running buffer and visualized by ethidium bromide staining. Each of the five exonic RNAs of HOTAIR were subject to the identical procedure. The RNAs were taken directly from an in vitro transcription reaction, desalted via a Bio-Spin P6 spin columns (Bio-Rad), and incubated at room temperature (10 min) with 34 mM Tris, 66 mM HEPES, 0.1 mM EDTA, pH 7.4 at increasing amounts of MgCl₂ concentrations (0, 0.1, 0.2, 0.4, 0.8, 1.6, 3.2, 5.0 mM).

2.4. Thermal melting

Thermal stability studies were conducted using a Varian Cary 100 Bio spectrophotometer operated at 260 nm. Folded HOTAIR exon 4 RNA at 1 μM in 10 mM HEPES pH 7.4, 100 mM KCl, and at MgCl₂ concentrations (0, 5, 20 mM) was heated from 15–95 °C at a rate of 1 °C/min while absorbance data was collected at 260 nm in 1 °C increments. This identical experiment was repeated for the folded helix 10 RNA with 0 and 5 mM MgCl₂ concentrations. Two scans were taken for each RNA and the melting temperatures (T_m) were calculated from first order derivatives (δA/δT) of the absorbance versus temperature profiles using GraphPad Prism 6 for Windows.

2.5. Analytical Ultracentrifugation

Purified RNA was diluted with 10 mM HEPES pH 7.4, 100 mM KCl, and MgCl₂ (ranging from 0, 5, and 20, and 40 mM MgCl₂) to an A₂₆₀ range between 0.5–0.8. Samples were run for 16 hours in an Optima XLI ultracentrifuge equipped with a four-hole An-60 Ti rotor at 48 000 rpm at 4 °C. Samples were loaded into double-sector cells (path length of 1.2 cm) with charcoal-filled Epon centerpieces and sapphire windows. Data were fit to a continuous c(s) distribution model using SedFit (version 12.0) (Brown and Schuck 2006) using every seven scans from a total of 360 scans. The folded RNA was able to sediment as a single homogenous species with a sedimentation coefficient of approximately 3.5 (Table S2).

2.6. RNA chemical modification for SHAPE and primer extension

Selective 2’ hydroxyl acylation chemical probing reactions were performed as previously described (Wilkinson et al. 2006). 10 pmol of folded HOTAIR Exon 4 SHAPE construct RNA (Figure S3A) in 95 μL 25 mM Tris pH 8.0, 100 mM KCl, and 10 mM MgCl₂ was treated with 5 μL of anhydrous dimethylsulfoxide (DMSO) (control, −), 65 mM (3.25 mM final, +) N-methylisatoic anhydride in DMSO (NMIA, Life Technologies), or 130 mM (6.5 mM final, ++) NMIA in DMSO for five NMIA half-lives at room temperature (~2.5 hr). RNAs were precipitated by addition of 4 μL 5 M NaCl, 1μL 20 mg/mL glycogen, 2 μL 100 mM EDTA pH 8.0, and 350 μL of ice-cold ethanol. Air-dried RNA pellets were stored at −20 °C until use.

RNA pellets were resuspended in 10 μL 10 mM Tris-HCl pH 8.0 giving a final concentration of 1 μM. RNA (1 pmol) in 10 μL 10 mM Tris-HCl pH 8.0 was used per extension reaction; (−) RNA was also used to generate sequencing ladders. All subsequent steps and optimization conditions, including a primer binding site, low magnesium ion concentrations, and short extension times, were performed exactly as described (Wilkinson et al. 2006). After the reactions were quenched by RNA hydrolysis, reactions were stored at 4 °C until use. Four lanes in a sequencing gel in addition to the SHAPE experiment provided high resolved chemical probing data for nucleotides 322–397 of HOTAIR.

2.7. Sequencing and structure mapping

Each primer extension reaction (5 μL) was loaded onto a 0.75 mm × 31 cm × 38.5 cm sequencing gel consisting of 10% polyacrylamide (19:1), 1x TBE, and 7 M urea. Gels were run at 45 W for 5 hours. The gel was transferred to Whatman 3MM paper, covered with plastic wrap, and exposed to a K-screen (Bio-Rad) for one hour. The phosphorimaging screen was subsequently scanned using Pharos FX Plus Molecular Imager (Bio-Rad) and the Semi-Automated Fragment Analysis(Das et al. 2005) (SAFA) package was used to analyze the data. Approximately 75 nucleotides of hHOTAIR exon 4 RNA were well resolved (nucleotides 322–397), quantified, and the resulting pseudo-free energies for these nucleotides were used as restraints for RNAStructure secondary structure prediction (Bellaousov et al. 2013). SHAPE chemical probing experiments were performed in triplicate and a representative sequencing gel and SHAPE experiment is shown in Figure S2.

2.8. NMR analysis

Three RNA constructs of the helix 10 hHOTAIR RNA were prepared comprising a 67-nucleotide helix 10 construct (327–394 nucleotides), a truncated lower helix (327–337 and 384–394 nucleotides HOTAIR regions), and a truncated upper helix (352–370 nucleotides). Each RNA construct of helix 10 was desalted on gravity gel-filtration Micro Bio-Spin P-6 columns (Bio-Rad Laboratories) and then lyophilized using a Labconco^™ freeze dry chamber. The dried RNAs were folded and buffer exchanged and the folding conditions were established from the native gel analysis. The NMR buffer for all RNAs was 50 mM potassium phosphate (pH 6), 100 mM potassium chloride, and 5 mM MgCl₂ in 5% D₂O. NMR spectra were obtained on a 900 MHz Bruker AVIII HD spectrometer at NMRFAM (University of Wisconsin-Madison).

One-dimensional exchangeable proton spectra were acquired with a 1–1 spin-echo water suppression pulse for all samples. Two-dimensional spectra were acquired with a WATERGATE pulse with flipback or with a 1–1 spin-echo water suppression (noesyph11). Imino proton resonances were assigned by reference to 2D ¹H-¹H NOESY experiments (150 ms mixing times) at 285 K. All spectra were processed with Top-Spin 4.0 (Bruker, BioSpin, Germany) and analyzed with NMRFAM-Sparky (Lee et al. 2015). The 2D NOESY identifies uridines (N3-H3 imino) within A-U pairs or guanosines (N1-H1 imino) within G-C pairs. It is also well-established that non-canonical guanosines and uridines within a G-U base pair result in characteristic upfield shifts (10.5–12.4 ppm δ ¹H) and exhibit diagnostic NOE patterns between the G-H1 and the U-H3 (Fürtig et al. 2003). This G-U hydrogen bonding cross-peak pattern served as a starting point to assign the helix 10 imino resonances of hHOTAIR.

Results

To probe the general interplay between splicing and the folding properties of lncRNAs, we sought to characterize the conformational heterogeneity of individual, isolated hHOTAIR exonic transcripts and identify potential regions that contain a highly stable secondary structure. HOTAIR exons 1 (60 nt) and 2 (126 nt) displayed significant heterogeneity, regardless of treatment (Figure 1). In particular, exon 1 exhibited a marked propensity to dimerize in both non-denatured (‘native’) and refolded states, especially at higher magnesium conditions. Exons 3 (102 nt) and 5 (57 nt) were also heterogeneous in their native state but homogenous after refolding. The observed structural heterogeneity of exonic transcripts 1,2,3 and 5 can be partially explained by examining the exon-exon boundaries and inter-exonic base pairs within the context of the Domain I HOTAIR secondary structure (Somarowthu et al. 2015) (Figure S1).

Figure 1: Native gel analysis of HOTAIR exons 1–5.

N = non-denatured RNA sample taken directly from an in vitro transcription reaction. The five RNAs correspond to the exonic RNA sequences of hHOTAIR (exons 1–5) and were subject to non-denaturing PAGE (8% acrylamide:bisacrylamide 19:1) at 100 V for 60 minutes in 34 mM Tris (pH 7.4), 66 mM HEPES (pH 7.4), 100 mM KCl, and at various MgCl₂ concentrations. 10 picomoles of RNA were loaded per lane, and bands were visualized by ethidium bromide staining.

In contrast, Exon 4 (124 nt) was homogenous and refractory towards treatment, suggesting that exon 4 either has no intrinsic secondary structure (i.e. single-stranded) or can adopt a defined fold under different conditions. Notably, the construct choice was informed only by exon boundaries of hHOTAIR, yet exon 4 exhibits the highest degree of sequence conservation between human and mouse HOTAIR (Schorderet and Duboule 2011; Somarowthu et al. 2015). Exon 4 is the only exon of hHOTAIR that contains a defined substructure (helix 10) with extensive intra-exon base pairing (Figure S1B). In addition, its primary sequence is embedded within a single, much larger exon in mouse HOTAIR, suggesting that acquisition of new splice sites during lncRNA evolution could be influenced by structured sequences (Schorderet and Duboule 2011; Somarowthu et al. 2015). Taken together, these native gel observations prompted our selection of hHOTAIR exon 4 for structural characterization.

Sedimentation velocity analytical ultracentrifugation (SV-AUC) and thermal melting data support native PAGE analysis and demonstrate that exon 4 adopts a homogenous and stable structure (Figure 2). Modest compaction of the exon 4 structure in the presence of magnesium was observed during SV-AUC (Figure 2A and Table S2). This observation is consistent with previous results with full-length HOTAIR and RNA folding (Somarowthu et al. 2015), where peak broadening and a decrease in UV absorbance at increasing MgCl₂ concentrations is considered a characteristic of increased RNA folding (Chillón et al. 2015). In all cases, exon 4 sediments as a single species, supporting native gel analysis and demonstrating that the exon 4 fold is refractory to both non-denaturing and denaturing purification approaches. The homogeneity of the sample based upon continuous sedimentation coefficient distribution values, ranging from 3.29 – 3.99 c(s) at increasing metal concentrations (0 – 40 mM MgCl₂), indicate a monodisperse RNA molecule that exhibits a stable RNA folding pathway that is independent of the purification approach (Table S2). This is in contrast to full-length hHOTAIR, which is deleteriously affected by denaturing purification (Somarowthu et al. 2015).

Figure 2: Folding and thermostability of HOTAIR exon 4 RNA.

(A) Sedimentation velocity analytical ultracentrifugation (SV-AUC) profiles of HOTAIR exon 4 RNA, with a graph of the sedimentation coefficient (S) and the continuous c(s) distribution model. RNA in the presence of 10 mM HEPES pH 7.4, 100 mM KCl at 25 °C was prepared with 0 mM (circle), 5 mM (square), 20 mM (triangle), and 40 mM (inverted triangle) MgCl₂. Sedimentation coefficient data are reported in Table S2. (B) Normalized absorbance versus temperature for exon 4 RNA containing 0 mM (black), 5 mM (black dash), and 20 mM (grey dash) MgCl₂. (C) First derivative plots of δA/δT versus temperature for exon 4 RNA at various magnesium ion concentrations, as described in B. Two magnesium dependent melting transitions were identified.

Melting curves of folded RNA samples in increasing magnesium concentrations show a change in hyperchromicity during heating, suggesting that HOTAIR exon 4 RNA contains a structural element that is stabilized by the presence of magnesium (Figure 2B). Peaks in the calculated first derivative plot of each melting curve represent two distinct thermodynamic transitions at 0 mM MgCl₂ (T_m¹ = 45 °C, T_m² = 78 °C) (Figure 2C). In the presence of magnesium, a known stabilizer of RNA tertiary structure (Serra et al. 2002; Lambert and Draper 2007), the T_m¹ shifts to ~60 °C. This magnesium dependent change in stability suggests that exon 4 contains tertiary contacts. The second transition in the presence of magnesium likely also occurs at 85–90 °C, but is beyond the limit of detection. These SV-AUC and UV-melting data indicate that the hHOTAIR exon 4 adopts a stable thermodynamic structure in the presence of 5 mM MgCl₂.

Given the stability of HOTAIR exon 4 in vitro, we used SHAPE to gather backbone-based nucleotide-resolution data on the exon 4 RNA. High-confidence experimental SHAPE data (Figure 3, Figure S2–S3) was obtained between construct positions 322 and 397 (Figure S3A), corresponding to helix 10 of full-length HOTAIR (Somarowthu et al. 2015). Data were quantified and normalized with the SAFA software package (Das et al. 2005) (Figure S3B), and used as restraints in RNAStructure (Bellaousov et al. 2013) to generate a secondary structure model (Figure 3). The experimentally derived model and per-nucleotide SHAPE reactivities for helix 10 correlate well with published data (Somarowthu et al. 2015; Portoso et al. 2017) with only minor deviations (Figure S5), corroborating observations that exon 4 contains a highly stable, independently folded structure. Further, hHOTAIR exon 4 contains three splice acceptor sites (Kent et al. 2002; Schorderet and Duboule 2011) that all appear to preserve helix 10 integrity but not helix 9 (Figure 3), suggesting that helix 10 is likely present in all hHOTAIR splice isoforms.

Figure 3: HOTAIR exon 4 secondary structure derived from SHAPE experimental data.

The SHAPE reactivity profile is depicted as colored nucleotides, based upon extent of relative reactivity (low (black), mild (blue), substantial (orange), and high (red)). The chemical probing experiment defined HOTAIR regions 322–397. The alternate splice acceptor sites (α, β, and γ) are indicated with asterisks and grey nucleotides flanking Helix 10 were based on previous secondary structure studies, as defined in Supplemental Figure 5.

To compare and validate this SHAPE data, we sought to define the base pair interactions using NMR spectroscopy. 2D ¹H-¹H NOESY correlates RNA imino protons through space (within 5.0 Å) and represents a useful approach to analyze HOTAIR secondary structure because the presence of non-canonical G-U base pairs can be detected via a diagnostic NOE cross-peak connectivity pattern. Previous chemical probing experiments revealed that up to four G-U wobble base pair interactions occur within helix 10 of HOTAIR (Somarowthu et al. 2015). Thus, the G-U non-canonical pairs should provide a starting point in the NMR assignment of the large RNA molecule. Three different RNA constructs were prepared under identical folding conditions with the goal of defining the RNA base pair interactions. NMR spectra of the lower helical RNA construct (Figure 4A), containing 327–337 and 384–394 nts, linked via a GUAA tertraloop, show near identical NOE connectivity patterns to the full-length 67-nucleotide helix 10 RNA (327–394 nts) (Figure 4B). This is especially apparent in the non-canonical region of the imino spectra, where distinct G-U interactions are observed, corresponding to assigned U330-G391, U333-G388, and G334-U387 wobble pairs. This result confirms that the lower helical secondary structure is preserved within the full-length helix 10 RNA (Figure 4).

Figure 4: NMR spectroscopy defines the helix 10 secondary structure of hHOTAIR RNA.

(A) Imino 2D NOESY region identifies formation of non-canonical G-U and G:A base pairs. NOE connectivities are noted (dashed lines) and the G-U and G-A region is boxed (magenta). The truncated, lower region of helix 10 contains 327–337 and 384–394 nucleotides and the secondary structure of the 30 nucleotide RNA construct is shown. (B) Imino 2D NOESY of the 67-nucleotide Helix 10 RNA. Assigned resonances are indicated in the 1D imino inset, along with unassigned U-A (asterisk) and G-C (dashed) resonances. NOE connectivities present in (A) are identified in the spectrum and boxed (black) within the secondary structure. Non-canonical G-U and G-A base pairs are noted (magenta box). Non-native nucleotides (grey) are identified and included for efficient in vitro transcription or to stabilize the stem-loop region. In A and B, the secondary structures are colored according to the SHAPE reactivity profile, as defined in Figure 3.

In contrast, comparison of the full-length helix 10 (327–394 nts) with an upper helical RNA region (352–370 nts) reveals that no G-U base pairs form within the upper stem of helix 10. Comparative NMR analyses reveal that only the formation of G-C and A-U base pairings are consistent between the full-length and upper stem of helix 10 (Figure S4). The lack of a G358-U364 wobble pair within the upper stem of helix 10 is also consistent with our SHAPE-based structure prediction, and represents the only change when compared to the previously determined HOTAIR secondary structure (Somarowthu et al. 2015) (Figure S5). This NMR data is also consistent with the derived Boltzmann-probability average positional entropy values of the helix 10 RNA, where a putative G358-U364 wobble pair is poorly predicted and exhibits the highest degree of base pair uncertainty across the RNA (Figure S7). Although it is possible a G358-U364 pair could form transiently and is not observable in the timescale of the NMR experiment, we propose that the NMR data fully supports the SHAPE-derived helix 10 secondary structure under these experimental conditions (Figure 3).

Based upon the time scale of imino proton exchange in the 2D NOESY experiment, it was not possible to unambiguously assign NOE connectivities corresponding to the internal regions of the helix 10 RNA (nucleotides 345–352 and 370–379). This suggests a large and highly dynamic internal bulge region, consistent with SHAPE chemical probing results (Figure 3, S3C). To validate and compare the folding properties of helix 10 and the full-length exon 4, UV-melt was performed, showing that both helix 10 and exon 4 RNAs have comparable magnesium dependent melting transitions (Figure S6). Taken together, the exon-biased approach and NMR results suggest only minor secondary structure changes when compared to the work of Somarowthu et. al, demonstrating that helix 10 stably forms within exon 4, within the defined Domain I region, and within the full-length hHOTAIR.

Discussion

The characterization of functionally relevant lncRNA-mediated biological mechanisms requires multiple experimental biophysical and bioinformatic approaches that probe the structure and thermodynamic properties of the RNA target (Butcher and Pyle 2011; Pyle 2014; Chu et al. 2015; Yao et al. 2017; Jones and Sattler 2019). Although there still remains limited and conflicting mechanistic insight into the structure-function relationships of HOTAIR (Tsai et al. 2010; Wu et al. 2013; Somarowthu et al. 2015; Meredith et al. 2016; Portoso et al. 2017), there is a steadily growing number of rigorous biochemical studies ascribing specific structural mechanisms to lncRNA function (Li et al. 2010; Wang and Chang 2011; Eidem et al. 2016; Quinn et al. 2016; Uroda et al. 2019). HOTAIR, much like other lncRNAs, has multiple isoforms and is predicted to assume alternative conformational states that will influence its overall RNA architecture. Akin to mRNAs in protein coding genes that can give rise to structured protein domains, it has been proposed that there is intrinsic structure within some lncRNA exonic boundaries (Uszczynska-Ratajczak et al. 2018). Thus, it is possible to envision that a stable RNA conformation would arise within the splice boundary of an exon.

We sought to examine whether the HOTAIR lncRNA contains any discernable structure within its individual exonic regions, as any putative helical elements may suggest that there is evolutionary pressure to preserve an RNA helical element (Shepard and Hertel 2008; Ulitsky 2016; Uszczynska-Ratajczak et al. 2018). Our work introduces an exon-biased multidisciplinary biophysical method to probe lncRNA secondary structure. We observe structural heterogeneity of hHOTAIR exonic transcripts 1,2,3 and 5 that correlates well with exon-exon boundary locations within the context of the full-length secondary structure (Somarowthu et al. 2015) (Figures 1,S1). In addition, we identify that the most homogenous, well-folded exonic region of HOTAIR is exon 4, a highly conserved RNA element that is predicted to occur in both human and mouse HOTAIR and is likely preserved among all splice isoforms of hHOTAIR. The mapping of high confidence exon-exon boundaries on the Domain I of hHOTAIR reveal that helix 7 and 10 substructures, previously identified to contain covariation, are both well preserved among all splice site locations (Somarowthu et al. 2015) (Figure S1B).

It was previously proposed that helix 10 of HOTAIR contained at least six conserved base pairs, several consistent half-flips (AU to GU; GC to GU), and substantial base pair covariation (Somarowthu et al. 2015; Tavares et al. 2019). Nonetheless, the Rfam database and some R-scape statistical analysis studies have called into question the degree of base pair covariation and the correlation between sequence alignments and putative evolutionarily conserved structures within HOTAIR (Rivas et al. 2017; Rivas et al. 2020). While sequence analysis and covariance modeling represent a powerful means to distinguish whether a lncRNA contains an evolutionarily conserved structure, there are some inherent challenges and limitations in relying solely on covariation statistical significance as the criteria for identifying putative conserved mammalian lncRNA architectures. For example, the limitation of available mammalian sequences, incomplete splicing information of an RNAs, as well as the thermodynamic propensity of homologous sequences to adopt a conserved fold, can all potentially influence an RNA secondary structure (Tavares et al. 2019). Incorporating these additional parameters along with the inclusion of transcriptome-wide experimental data sets (Lucks et al. 2011; Watters et al. 2016; Zubradt et al. 2017; Smola and Weeks 2018) may help to improve global predictive modeling and the identification of evolutionarily conserved lncRNA structures.

In the case of HOTAIR, the lncRNA structure has been extensively characterized in vitro under different buffer conditions and probed via multiple experimental approaches that include: SHAPE, dimethyl sulfate (DMS) modification, parallel analysis of RNA structure (PARS), terbium chloride probing, Cryo-EM, and atomic force microscopy (Kertesz et al. 2010; Somarowthu et al. 2015; Spokoini-Stern et al. 2020). All of these methods provide insight into HOTAIR RNA structure prediction based upon chemical reactivity but fail to directly inform upon hydrogen bonding base pair geometry. In general, SHAPE data can improve secondary structure predictions to approximately 90% accuracy, but reactivity at the 2’ hydroxyl of the RNA is not always predictive of whether a base is paired or unpaired (Kutchko and Laederach 2017). This caveat holds true even for base-specific quantitative dimethyl sulfate (DMS) modifications, which represent a powerful in vitro and cell-based technology but does not, by itself, directly report on base pair orientation.

The inclusion of NMR spectroscopy with chemical probing allows for the definition of bona-fide base pairing interactions within large RNA secondary structures and reports on the local dynamics of the RNA backbone. The identified homonuclear NOEs of the imino hydrogen bonding environment in the lower stem and full-length helix 10, along with SHAPE reactivity data, all corroborate an accurate secondary structure of the full-length 67 nucleotide region of helix 10. This NMR-based divide and conquer approach has been implemented in the study of many RNAs to date (Keane et al. 2015; Barnwal et al. 2017; Zhang and Keane 2019), though only a few studies have begun to incorporate and compare NMR with probing methods in a direct comparative manner (Chen et al. 2015; Kotar et al. 2020). A recent NMR spectroscopy-DMS probing study highlights the impact of this hybrid approach in the secondary structure determination of 15 conserved RNA elements within the SARS-CoV-2 genome, thus providing a highly accurate template for potential drug targets to combat Covid-19 (Wacker et al. 2020).

In our study, NMR data of the individual lower or upper stem regions were compared to NMR and SHAPE data of a dynamic 67-nucleotide RNA, helping to validate the formation of G-U base pairs within the lower stem of helix 10 and minimally refine the secondary structure within the upper stem of the RNA helix when compared to the determined full length secondary structure (Somarowthu et al. 2015). It should be noted that, despite completely different RNAs (exon 4 vs. full length hHOTAIR), different conditions (10 vs. 25 mM Mg²⁺), and different SHAPE reagents (NMIA vs. 1M7), only subtle structural differences were observed between the experimentally derived RNA secondary structures (Figure S5). The NMR-SHAPE based approach allows for an additional base pair validation step that is consistent with Boltzmann probability derived average positional entropy values across helix 10, where all lower helix G-U pairs (U330-G391, U333-G388, and G334-U387) possess well ordered nucleotide positions and the upper stem G358-U364 pair exhibits poorly defined positional entropy and high uncertainty (Figure S7). Future studies that implement NMR-based strategies with chemical reactivity experiments will allow for improved atomic level insight and a direct comparison of the temporal conformations of the RNA.

Conclusions

We combined an exon-biased method with chemical probing and NMR to determine that HOTAIR can form independent structural domains with varying degrees of heterogeneity and thermodynamic stability. Although this method may be applied to large ncRNAs with defined splice boundaries, the exon-biased structure determination of HOTAIR with NMR spectroscopy and chemical probing does validate that the conserved helix 10 serves as a core helical element within the 5’ region of HOTAIR (Somarowthu et al. 2015). A key limitation of this strategy is that it is initially biased towards lncRNA substructures that contain intra-exon base pairings and overlooks extensive inter-exonic interactions that can form within large mammalian RNAs. A bioinformatic-based map of predicted intra- versus inter-exon base pair conservation among mammalian lncRNAs is lacking at the moment but would significantly bolster the general applicability of this exon-biased approach. Dissecting the interplay of base pairing and single-strandedness near exon-exon boundaries and within exonic regions will be a challenge, but should inform upon the potential regulatory roles, splice site selection preferences, and protein binding properties of multi-spliced transcripts (pre-lncRNA and pre-mRNAs). With future, detailed mapping and functional annotation of the mammalian transcriptome, a combined NMR-chemical probing approach will help to improve the accuracy of large RNA secondary structures and will provide a comprehensive strategy to monitor the transient conformational modalities that exist within large RNA architectures.

Highlights:

• Structure determination is challenging for lncRNAs that contain multiple splice isoforms.

• An exon-biased biophysical approach that includes NMR and chemical probing was developed to identify homogenous conformations within the exons of HOTAIR.

• HOTAIR exon 4 contains a well-defined helical conformation, previously identified as a conserved RNA structure.

• Combined NMR with chemical probing methods provide accuracy in the secondary structure determination of large RNA molecules

Abbreviations:

hHOTAIR

human Homeobox transcript antisense RNA

NMR

nuclear magnetic resonance

R-scape

RNA significant covariation above phylogenetic expectation

lncRNA

long non-coding ribonucleic acid