Biological solution conditions and flanking sequence modulate LLPS of RNA G-quadruplex structures

Abstract

Guanine-rich regions of DNA or RNA can form structures with two or more consecutive G-quartets called G-quadruplexes (GQ). Recent studies reveal the potential for these structures to aggregate in vitro. Here, we report effects of in vivo concentrations of additives—amino acids, nucleotides, and crowding agents—on the structure and solution behavior of RNAs containing GQ-forming sequences. We found that cytosine nucleotides destabilize a model GQ structure at biological salt concentrations, while free amino acids and other nucleotides do not do so to a substantial degree. We also report that the tendency of folded GQs to form droplets or to aggregate depends on the nature of flanking sequence and the presence of additives. Notably, in the presence of biological amounts of polyamines, flanking regions on the 5′-end of the RNA drive more droplet-like phase separation, while flanking regions on the 3′-end, as well as both the 5′- and 3′-ends, induce more condensed, granular structures. Finally, we provide an example of a biological sequence in the presence of polyamines and show that crowders such as PEG and dextran can selectively cause its phase separation. These findings have implications for the participation of GQS in LLPS in vivo.

INTRODUCTION

G-quadruplexes (GQs) are highly stable G-rich RNA and DNA structures found in all domains of life and are characterized by G-repeat motifs (Bartas et al. 2019; Lombardi et al. 2019; Brázda et al. 2020). The standard GQ motif contains four or more repeats of two to four guanines followed by 1–7 nt, i.e., (G_2–4N_1–7)₄. Folding of GQs involves hydrogen bonding of both the Hoogsteen and Watson–Crick faces of guanines but can be challenged by interacting with C, A, and U (Supplemental Fig. 1). G-quadruplexes demonstrate a strong propensity to fold in vitro, especially in the presence of K⁺ ions (Bhattacharyya et al. 2016). Their detection remains challenging but possible in vivo. Recent technique development has allowed for direct detection of DNA and RNA GQ structures in vivo via antibodies (Biffi et al. 2013, 2014; Laguerre et al. 2016), as well as indirect detection via sequencing and chemical probing (Guo and Bartel 2016; Kwok et al. 2016; Shao et al. 2020). Antibody studies revealed localization of RNA GQ structures into cytoplasmic puncta suggesting that GQ sequences may be prone to liquid–liquid phase separation (LLPS) (Biffi et al. 2014).

Previous work from several labs, including our own, demonstrated that GQ sequences can aggregate or undergo LLPS in vitro (Fu et al. 2011; Fay et al. 2017; Zhang et al. 2019; Mimura et al. 2021; Williams et al. 2021). These in vitro conditions are often far from physiological, however, using either high salt or nonbiological crowders, or without metabolites. Therefore, expanding our understanding of the role that in vivo-like conditions play in GQ conformational equilibria may shed light on phase separation in vivo. Differences in aggregation versus LLPS could be due to intracellular conditions and metabolite concentrations as well as transcriptome context such as the identity and structure of flanking sequence. Such changes in conditions and context could lead to changes in RNA structure, altering phase properties.

Both biological and nonbiological small molecules have been shown to act as denaturants of RNA structure, especially at the secondary structural level (Lambert and Draper 2007). For instance, proline, urea, and sucrose show a high propensity to destabilize both secondary and tertiary structures in RNA (Lambert and Draper 2007). Additionally, functional studies demonstrate the importance of considering not just nucleotide content of the functional element for strength of base pairing, but also flanking nucleotide content and solution conditions (Woodson and Cech 1991; Serra and Turner 1995; Nikolcheva and Woodson 1999; Chadalavada et al. 2000; O'Toole et al. 2005; Tateishi-Karimata et al. 2014). Considering both in vivo metabolites and transcript context could deepen understanding of the behavior of individual RNAs. Moreover, intracellular conditions within organisms can change depending on growth media and stress. For example in Arabidopsis, proline concentration increases 30- to 50-fold under salinity stress conditions (Tack et al. 2020).

Herein, we investigate the effects of metabolites on a variety of GQ sequences (Supplemental Table 1). We also build upon our recent study of the effects of polyamines on GQs (Williams et al. 2021) by exploring effects in the presence of flanking nucleotides of different sequence and position. To do this, we carry out UV-detected thermal denaturation, circular dichroism (CD) spectroscopy, and confocal microscopy, as well as high-throughput microscopy and turbidity screens to determine the effects of added metabolites. Our findings show that location and identity of GQ flanking sequence can dictate morphology of phase separation of GQ structures in the presence of spermine. Additionally, we demonstrate that crowders in conjunction with spermine can change the propensity of phase separation of the biological sequence NRAS. Together, these results provide a better understanding of the driving forces for G-quadruplex LLPS.

RESULTS

Eukaryotic osmolytes have varied effects on model GQ stability

Cells are replete in metabolites that might affect GQ stability and phase separation. For example, our initial study determined that biological concentrations of 0.3 mM spermine produce amorphous aggregates for a model GQ sequence (Williams et al. 2021). Other metabolites such as amino acids and nucleotides can also interact with RNA (Yamagami et al. 2021). To determine if these metabolites could have general effects on GQ stability, we tested several amino acids as well as nucleotide mono- and triphosphates at their reported biological concentrations. To facilitate direct comparison to our previous experiments with polyamines spermine and spermidine, we used the same representative GQ sequence, (G₃A₂)₄.

Firstly, we examined the effect of six amino acids—arginine, aspartate, glutamine, glutamate, lysine, and proline—on the stability of (G₃A₂)₄ by UV-detected thermal denaturation. These particular amino acids were chosen due to their charge (both cationic and anionic) at physiological pH, and/or to their well-known potential to interact with nucleic acid structures (Treger and Westhof 2001; Corley et al. 2020). We used amino acid and salt concentrations physiologically relevant to yeast grown in YPD and YNB media (Park et al. 2016), both of which are common yeast media (values provided in Supplemental Table 2). As shown in Supplemental Figure 2, melting of G-quadruplexes monitored at 294 nm gave rise to inverted absorbance versus temperature melting curves, wherein the absorbance decreases with increasing temperature (Mergny et al. 1998). Changes in melting temperature support an effect of the given metabolite on thermostability, indicating potential interactions between the RNA and the metabolite. Decreasing signal at 294 nm with temperature supports unfolding of the GQ structure. However, none of the amino acids tested produced a significant change in the overall stability of the GQ sequence (Supplemental Fig. 2). To determine if higher, yet still biologically relevant, concentrations of glutamate could destabilize (G₃A₂)₄, we tested the E. coli concentration of 96 mM glutamate (Supplemental Table 2), which had 3.1 mM chelated Mg²⁺ and 0.5 mM free Mg²⁺ (Yamagami et al. 2018). Because of these high magnesium concentrations, we opted to not renature at high heat. To avoid RNA degradation, we cooled to room temperature before adding magnesium. However, we still found little change in stability of (G₃A₂)₄ in 96 mM glutamate (Supplemental Fig. 2). Apparently, amino acids do not substantially affect GQ stability at biological concentrations.

Nucleotides have the potential to base pair weakly with RNA and occur in relatively high abundance in vivo. For instance, the concentration of ATP is 1.9 mM in yeast and 9.6 mM in E. coli (Supplemental Table 2; Park et al. 2016). We therefore tested all non-G nucleotides—A, C, and U derivatives—to determine if they significantly affect the stability of the core GQ; guanine was eliminated because of its propensity to aggregate. In vivo concentrations of nucleotide triphosphates are significantly higher than those of the monophosphates (for example, 2.7 mM CTP vs 0.36 mM CMP in E. coli, see Supplemental Table 2), and were therefore expected to have the more significant effect on the folding stability of the GQ sequence. The nucleotide ATP at E. coli concentration of 9.6 mM provided little change in stability (Fig. 1), despite the ability to form a two-hydrogen bond base pair with guanine (Supplemental Fig. 1C). We also tested AMP, which is present at 0.28 mM in E. coli, and it also had no effect on the melting profile. We found that UTP, which is present at 8.3 mM, had a slight effect on melting of the core GQ, with absorbance below that of the no additive (Fig. 1). This could be because of the ability to form G•U wobbles (Supplemental Fig. 1D). To extend this idea, we tested the effect of CTP on melting of the core GQ. Strikingly, CTP had a distinct destabilizing effect on the melting signature of (G₃A₂)₄ (Fig. 1), which is all the more remarkable given its in vivo concentration is just 2.7 mM (versus 8.3 mM in UTP and 9.6 mM in ATP). Specifically, at 294 nm CTP provides more of a linear decreasing melting signature beginning at 20°C. This observation supports the notion that nucleotides can base pair with the guanines in the GQ to destabilize it. To expand on this, we focused on CTP and its derivatives.

FIGURE 1.

Thermal denaturation of (G₃A₂)₄ in the presence of nucleotide derivatives. We tested the effects of 0.28 mM AMP, 9.6 mM ATP, 8.3 mM UTP, 2.7 mM cytosine, 2.7 mM cytidine, 0.36 mM CMP, 2.7 mM CMP, and 2.7 mM CTP, and 2.7 mM pyrophosphate. All samples were in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5. Melts were monitored at 294 nm with a heating rate of 0.5°C/min and were internally normalized to the first data point at 20°C. All samples were buffer corrected. Legend is color-coded and arranged in the same order as the individual traces.

We first tested CMP, which has a relatively low in vivo concentration of 0.36 mM in E. coli. As shown in Figure 1 (light blue), this concentration did not significantly affect stability of the GQ via UV-detected thermal denaturation. To test if this was due to the triphosphate versus monophosphate, we repeated the experiment at 2.7 mM CMP and attained a similar result to 2.7 mM CTP, albeit with a slightly less pronounced melt (Fig. 1), supporting pairing with nucleotide and suggesting a role for phosphates in the interaction.

To test the importance of the sugar and phosphate in the destabilization of the GQ, we conducted melts on 2.7 mM cytosine (i.e., just the nucleobase with no phosphates or sugar). The curve was similar to that of the (G₃A₂)₄ core sequence, indicating that the base alone does not cause the same destabilization as CMP or CTP (Fig. 1). Therefore, there must be an important contribution from the phosphates and/or sugar in the destabilization of the core GQ structure. We then tested 2.7 mM cytidine (nucleobase and sugar, with no phosphates). Strikingly, this molecule produced a similar curve to CMP and CTP. This indicates that the sugar is important, perhaps in properly orienting the nucleobase for interactions with the GQ structure. Next, we tested the effect of pyrophosphate alone on the melts. No destabilization or change in melting signature as compared to salts alone was observed (Fig. 1). Apparently, the sugar and base are the most important features for destabilization of the core GQ structure, while the phosphates may assist with this phenomenon. These results further show that while the individual components of a nucleotide do not destabilize the structure, the sum of the parts has a dramatic effect on the stability in solution.

5′-flanking region can lead to droplet formation in the presence of spermine

Flanking sequence can influence RNA structure and function, for example in splicing and ribozyme cleavage (Woodson and Cech 1991; Cao and Woodson 1998; Nikolcheva and Woodson 1999; Chadalavada et al. 2000). In naturally occurring RNAs, GQs are found in the context of flanking sequence within a gene. We therefore set out to determine if 5′-, 3′-, or dual-flanking regions influence the properties of GQs. We first tested the contribution of four different 9-nt appendages to the 5′-end of the core (G₃A₂)₄ sequence (Supplemental Table 1; Williams et al. 2021). The 5′-flanking sequences were either homogeneous (A₉, C₉, or U₉) or a mixture of nine nucleotides. We wanted to address several questions related to conformational changes and LLPS with these flanking sequences: (1) Do flanking sequences alter the melting transitions? (2) Do the polyamines alter the melting transitions? (3) Do polyamines change the overall structure of the GQ as judged by CD spectroscopy? (4) If there is a resulting change in unfolding transitions and structure due to phase separation, can the product be visualized by confocal microscopy?

We first monitored absorbance as a function of temperature to look for a hypochromic transition. We previously reported that the core (G₃A₂)₄ sequence displays an inverted transition at ∼90°C; moreover, it has an additional transition between 20°C and 30°C in the presence of spermine (Williams et al. 2021). We reported that these properties are associated with formation of a GQ structure that forms aggregates in the presence of spermine that melt out at lower temperatures, as confirmed by microscopy (Williams et al. 2021).

To directly compare the melting transitions between this study and our previous study, we changed renaturation conditions used earlier to conditions reflecting those used in Williams et al. (2021). We plotted the first derivative of the melt data to determine relative melting temperatures of potentially multiple transitions. Similar melting behavior was observed for the 5′-mixed and 5′-A₉ flanking sequence (Fig. 2A,B, yellow and green, respectively). In the absence of polyamines, there is a downward melting transition at 294 nm near 85°C; in contrast, in the presence of both spermine and spermidine, there is a lower temperature melting transition between 20°C and 30°C, consistent with condensate formation of the core (G₃A₂)₄ sequence. As with the core sequence, the 20°C–30°C transition for 5′-A₉ and 5′-mixed is relatively insensitive to the presence of spermidine only but sensitive to the presence of spermine only (Supplemental Fig. 3). These results are consistent with phase separation from our recent study using the same technique (Williams et al. 2021). In our previous study, sequences that demonstrated aggregation had a similar early transition in the presence of spermine, but not spermidine, supporting the potential of the 5′-A₉ and the 5′-mixed sequences to phase separate as well.

FIGURE 2.

Thermal denaturation of 5′-flanking sequences. Experiments were completed in the (A) absence and (B) presence of both polyamines (0.3 mM spermine and 0.4 mM spermidine). All samples were in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5. Melts were monitored at 294 nm with a heating rate of 0.5°C/min. T_m is the temperature with a minimum in the first derivative. (G₃A₂)₄ is shown in orange as a reference point for the early melting transition in both polyamines.

Like the 5′-A₉ and 5′-mixed sequences, the 5′-U₉ sequence has an inverted melting profile with a high Tm in the absence of polyamines (Fig. 2A). But in contrast, it has no distinct early melting transition with both polyamines or spermine alone (Fig. 2B; Supplemental Fig. 3B), suggesting it still forms GQ structure but does not aggregate. A small-amplitude, low-temperature transition occurred in the presence of both polyamines (Fig. 2B) suggesting at most a slight tendency for LLPS for the 5′-U₉ sequence. The suppression of LLPS for the 5′-U₉ sequence may be because the Us can form wobble pairs with Gs in the GQ and Watson-Crick pairs with As in the loops (Supplemental Fig. 4A) consistent with the weak effect of UTP mentioned above.

In contrast to the 5′-A₉, 5′-U₉, and 5′-mixed flanking sequences, the melting profile of the 5′-C₉ sequence did not contain inverted melting signals at 294 nm (Fig. 2; Supplemental Fig. 3). This suggests that this sequence is forming a structure with conventional Watson–Crick base pairing rather than a GQ, consistent with the strong effect of CTP mentioned above. This is likely because of the potential for strong GC base pairs between the 5′-flank and the stretches of G's in the GQs. Such a structure is supported by a conventional Watson–Crick base paired structure with a predicted ΔG^o₃₇ of −10.0 kcal mol⁻¹ and Tm of 78.9°C (Supplemental Fig. 4B) calculated from mFold 2.3 utilizing default parameters (Zuker 2003). Taken together, the differences in melting profiles among the four different 5′-flanking sequences likely reflect differences in base pairing interactions with the GQ core (Supplemental Fig. 1): the C₉ sequence competes strongly with the GQ structure by forming strong CG Watson–Crick base pairs, the U₉ sequence competes weakly with the GQ structure by forming U•G wobble pairs and UA Watson–Crick pairs, and the A₉ and mixed sequences interact weakly if at all with the GQ structure.

Turbidity can indicate the presence of condensates in solution. In the case of GQ aggregates, which are small, we previously found observed turbidity at 350 nm is reflective of aggregates (Williams et al. 2021). To determine if the melting behavior between 20°C and 30°C might be due to aggregates, we conducted turbidity experiments for all 5′-flanking sequences at 350 nm in the presence of both polyamines. For 5′-C₉ and 5′-U₉, we found a turbidity of almost zero for all replicates (Fig. 3), supporting the lack of or weak condensate formation. In contrast, for 5′-A₉ and 5′-mixed sequences, high 350 nm turbidity (an average of 91% and 94%, respectively) was observed, consistent with formation of condensates for these sequences (Fig. 3). Results on these two sequences correlate with their low temperature melting transition in the presence of both polyamines at 294 nm (Fig. 2B).

FIGURE 3.

Turbidity experiments of (G₃A₂)₄ flanking sequences. Absorbance measured at 350 nm and calculated for turbidity using Equation 1. Readings were completed on three individually prepared samples, of 10 μM each, and mean value is shown with bar. Samples are in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5 and contain 0.3 mM spermine and 0.4 mM spermidine.

Our next goal was to determine if the 5′-flank and spermine change the structure of the core GQ. To assess this, we acquired CD spectra with and without spermine. We chose to focus on spermine-only because it was the polyamine necessary and sufficient to induce condensation (Supplemental Fig. 3). Generally speaking, parallel GQ structures have a positive spectral feature at ∼210 nm along with a maximum at ∼260 nm (Dapic et al. 2003; Kypr et al. 2009; Wang et al. 2019). The CD spectra for the 5′-mixed, 5′-A₉, and 5′-U₉ sequences had a positive feature at 210 nm in the absence of polyamines as well as a maximum at ∼265 nm, supporting a GQ structure (Supplemental Fig. 5A). In the 5′-A₉, the data were corrected for the strong negative signal seen in A₁₀ (Supplemental Fig. 6). In the presence of spermine, the spectral features of the 5′-mixed, A₉ and U₉ sequence are largely unchanged, consistent with GQ formation even in the presence of phase separation (Supplemental Fig. 5B).

In contrast, the 5′-C₉ sequence had a very strong negative feature at 210 nm in the absence of polyamines, consistent with duplex formation, as well as a slight red shift of the 265 nm maximum (Supplemental Fig. 5A). Results in the presence of spermine with the 5′-C₉ showed an even more prominent negative feature, moving from −4 to −6 mdeg, and a stronger red shift of the maximum (Supplemental Fig. 5B). These observations support the ability of C-rich 5′-flanks to base pair with core GQ, disrupting GQ formation, and leading to noninverted 294 nm melting curves, consistent with our findings from UV-detected thermal denaturation and turbidity experiments above.

Finally, we assessed if the 5′-flanking sequence-GQs form a product that can be visualized with microscopy similar to those recently reported for the core (G₃A₂)₄ sequence (Williams et al. 2021). We conducted confocal microscopy on all four 5′-flanking sequences: 5′-mixed, 5′-A₉, 5′-C₉, and 5′-U₉, initially in the absence of polyamines (Fig. 4, row 1). We observed nothing above background under any field in the microscopy experiments for any of the four sequences, consistent with the UV-detected thermal denaturation experiments, which revealed no early melting transition under these conditions (Fig. 2). We then conducted the microscopy experiments in the presence of spermine and spermidine, as well as spermine-alone for all four sequences. As expected, we found LLPS, but surprisingly we observed consistent round ∼2 µm condensates, both for the 5′-mixed and the 5′-A₉ sequences (Fig. 4, columns 1 and 2) rather than globular, amorphous aggregation seen for the previous core (G₃A₂)₄ sequence in Williams et al. 2021. In contrast, the 5′-C₉ sequence did not reveal LLPS under any conditions (Fig. 4, column 3), consistent with strong interaction of the 5′-flanking sequence with the core. The 5′-U₉ sequence showed only small and infrequent round droplets in the presence of both polyamines or spermine-alone (Fig. 4, column 4), consistent with its lower, yet still existent turbidity and intermediate interaction of the 5′-flanking sequence with the core (Figs. 3, 2B). These observations suggest that pyrimidine-rich 5′-flanking sequences may temper the extent of intermolecular, droplet-inducing interactions.

FIGURE 4.

Visualization of LLPS and aggregation for 5′-flanking GQs by confocal microscopy. Shown are 5′-mixed (column 1), 5′-A₉ (column 2), 5′-C₉ (column 3), and 5′-U₉ (column 4). Each row indicates a different condition: no polyamines (row 1), biological 0.4 mM spermidine and 0.3 mM spermine (row 2), and 0.3 mM spermine-only (row 3). All samples were in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5. Multiple 80 × 80 µm² fields were visualized, and representative images are shown. No LLPS was found in the absence of polyamines fields (row 1) or 5′-C₉ fields (column 3).

We next investigated the properties of the 5′-mixed and 5′-A₉ droplets to assess if the GQ structure is still folded inside of the droplets. GQ-specific fluorescent dyes such as Thioflavin T (ThT) can help determine if (1) the RNA is localized in the droplets, and (2) the RNA is folded into GQ structures (Xu et al. 2016; Mimura et al. 2021). We added ThT to our solutions containing droplets and visualized them using confocal microscopy. For both 5′-mixed and 5′-A₉ GQs, fluorescence localized within the droplets when overlaid with DIC images (Fig. 5). This observation suggests that GQ-folded structures are enriched within the droplets, consistent with our measurements of folding within the bulk of the solution via CD spectroscopy (Supplemental Fig. 5B, yellow and green lines), as well as the hypochromic signature from the UV thermal denaturation (Fig. 2B, yellow and green lines). However, it should be noted that there are reports in which non-GQ related, purine-rich samples fluoresce in the presence of ThT (Sugimoto et al. 2015), albeit with long chains of homo-polyA. While our data support the fold of GQ structures in droplets, further experimentation would be needed to confirm the fold. The difference in phase separation among the four 5′-flanking sequence GQs led us to ask if moving the flanking region to the 3′-end of the oligonucleotide would change the LLPS properties.

FIGURE 5.

Fluorescence microscopy of droplets in the presence of Thioflavin T. Fluorescence, DIC, and Merge of 5′-mixed and 5′-A₉ in both polyamines, 0.4 mM spermidine and 0.3 mM spermine. Samples are in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5.

Select 3′ flanking sequences form aggregates in the presence of polyamines

We next assessed the effect of sequence additions to the 3′-end of the core RNA. In the absence of polyamines, none of the 3′-flanking sequences showed completed melting curves even out to 95°C, and all had a single, high temperature transition (Fig. 6A, top). In the presence of both polyamines, 3′-mixed and 3′-A₉ showed some early melting signatures, while the 3′-C₉ and 3′-U₉ GQs showed a lack of responsiveness to both polyamines (Fig. 6B, top). This behavior was similar to their 5′-flanking counterparts (Fig. 2). Furthermore, all GQs showed no response to spermidine (Supplemental Fig. 7A), and both 3′-mixed and 3′-A₉ showed some response to spermine alone (Supplemental Fig. 7B).

FIGURE 6.

Thermal denaturation of 3′-flanking sequences in the absence and presence of polyamines. Sequences were renatured in the (A) absence or (B) presence of 0.4 mM spermidine and 0.3 mM spermine. T_m is the temperature with a minimum in the first derivative. Microscopy fields are shown with corresponding sequences. All samples are in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5. Melts were monitored at 294 nm with a heating rate of 0.5°C/min. (G₃A₂)₄ is shown in orange as a reference point for the early melting transition in both polyamines.

We then visualized possible phase separation of the 3′-flanking GQs utilizing confocal microscopy. No aggregates were found in the absence of polyamines (Fig. 6A, bottom). In the presence of both polyamines, aggregates were observed for the 3′-A₉ and 3′-mixed flanking sequences, which were only about half the diameter of their 5′-flanking counterparts (∼0.75–1 µm), and no aggregates were observed for 3′-C₉ and 3′-U₉ flanking sequences (Fig. 6B, bottom). The 3′-A₉ and 3′-mixed aggregates are reflective of the low temperature melting transition, which is associated with LLPS (Fig. 6B, top). Moreover, the condensates for 3′-A₉ and 3′-mixed were fewer in number than their 5′-counterparts (Fig. 4). The microscopy observations were supported by turbidity data (Fig. 3), which showed moderate levels for 3′-A₉ (average 36%) and no significant turbidity for the other three 3′-flanked sequences.

To determine if morphological differences of the 3′-flanked sequences in phase separation are associated with the differences in GQ folding, we conducted CD spectroscopy in the absence and presence of spermine (Fig. 7). In the absence of spermine, the 3′-mixed, 3′-A₉, and 3′-U₉ 3′-flanking sequences appear to form a significant amount of GQ structure, given the positive signature at 210 nm and the maximum near 265 nm (Dapic et al. 2003; Kypr et al. 2009; Wang et al. 2019). In contrast, 3′-C₉ has a 210 nm feature near zero and a red-shifted maximum (Fig. 7A). In the presence of 0.3 mM spermine, the signature did not change much for 3′-mixed, 3′-A₉, and 3′-U₉ 3′-flanking sequences, except for a modest decrease in the amplitude of the major peak in 3′-A₉ (Fig. 7B). In contrast, upon the addition of spermine the 3′-C₉ sequence showed a strong negative signature at ∼210 nm, supporting stabilization of the GC base pair hairpin conformation. Additionally, the maximum peak red-shifted even more upon addition of spermine (Fig. 7) further supporting a change in conformation toward hairpin. This change suggests that the conformation for 3′-C₉ is a mix of GQ and hairpin in the absence of polyamines, but primarily the hairpin conformation in spermine.

FIGURE 7.

CD spectra for 3′-flanking sequences in the absence and presence of spermine. (A) Absence and (B) presence of 0.3 mM spermine. 3′-mixed (yellow), 3′-A₉ (green), 3′-C₉ (blue), and 3′-U₉ (purple). All spectra are buffer corrected and the 3′-A₉ sequence is corrected for the spectrum of A₉. Samples are in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5.

Dual flanked GQ sequences present minor aggregation induced by both polyamines

After investigating effects of adding flanking regions to the 5′- or the 3′-ends of the core GQ sequence, we tested effects of adding flanking sequences to both sides of the core GQ at once, designated as “dual flanked” GQs, because this is the way most GQ sequences are found in biology. We completed UV-detected thermal denaturation on the dual flanked-mixed, dual flanked-A₉, and dual flanked-U₉. We omitted the cytosine flanking sequence for these experiments due to the lack of phase separation demonstrated by both single flanking sequences. In the absence of polyamines, all three sequences showed a single melting transition, which was present at ∼78°C to 80°C (Fig. 8A, top). These are a slightly lower T_ms than in either the 5′- or 3′-flanking sequences, suggestive of competition from additional flanking regions. In the presence of spermine and spermidine, only the dual flanked-A₉ (Fig. 8B, top, green line) showed a significant lower temperature (20–30°C) melting transition, with the dual flanked-mixed showing a small low temperature melting transition (Fig. 8B, top yellow line). Confocal microscopy experiments on these three sequences in the absence and presence of both polyamines showed a lack of phase separation without polyamines (Fig. 8A, bottom) and small aggregates in the presence of both polyamines for dual flanked-A₉ (0.5 µm in diameter) and even fewer aggregates for dual flanked-mixed (Fig. 8B, bottom) consistent with the relative amplitudes of their melting transitions (Fig. 8B, top) and turbidity (Fig. 3, center). No aggregates were found for the dual flanked-U₉ (Fig. 8B, bottom), which is also consistent with its absence of a low temperature melting transition (Fig. 8B, top) and no turbidity (Fig. 3, center), even in the presence of both polyamines.

FIGURE 8.

UV thermal denaturation and confocal microscopy of dual-flanked GQs. Sequences (A) without and (B) with both polyamines 0.4 mM spermidine and 0.3 mM spermine. T_m is the temperature with a minimum in the first derivative. Confocal images shown below corresponding melting plots. All samples were monitored at 294 nm and were in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5. (G₃A₂)₄ is shown in orange as a reference point for the early melting transition in both polyamines.

We also wanted to test whether the three dual flanked sequences form GQ structure. CD spectra showed a positive signature at 210 and 265 nm for dual flanked-mixed, dual flanked-A₉ (corrected for two A9 tails) and dual flanked-U₉ in the absence (Supplemental Fig. 8A) and presence (Supplemental Fig. 8B) of spermine consistent with formation of a GQ. There is little change between the absence and presence of spermine. The limited structure response to spermine and LLPS response could be due to the lower relative concentration of spermine compared to the length of the overall sequence.

Due to the possibility of uracil interacting with guanine by wobble pairing and interfering with GQ formation, especially with dual U₉ flanks, we tested if sequestering the U₉ flanking regions would increase the relative propensity of the sequence to aggregate. We mixed dual flanked-A₉ and dual flanked-U₉ together in a 1:1 ratio (10 µM total RNA) during the renaturation process to allow base pairing on either side of the GQ structure. We first completed UV-thermal denaturation which revealed a very significant negative early temperature transition at 294 nm that was less steep than the dual flanked-A₉ alone (Fig. 9A). This is due to GQ formation and not simply melting of A₉–U₉ base pairing as signified by the negative dA/dT at 294 nm. Confocal microscopy revealed phase separation in the presence of both polyamines or spermine alone (Fig. 9B), similar in size as to dual flanked-A₉ (Fig. 8B bottom). This result indicates that mixtures of flanking sequences have the potential to drive GQ LLPS when the GQ structure is conserved. Finally, turbidity data for the combined sequences (Fig. 9C) is reflective of similar levels to 10 µM dual flanked-A₉ alone (Fig. 3, middle), implying that sequestration of the flanking-U allows for phase separation equivalent to phase separation levels at higher dual flanked-A₉ RNA concentrations.

FIGURE 9.

Dual flanked-A₉ mixed with dual flanked-U₉ phase separate when combined. (A) UV-thermal denaturation monitored at 294 nm first derivative plot in the presence of both polyamines 0.4 mM spermidine and 0.3 mM spermine. T_m is the temperature with a minimum in the first derivative. (B) Representative confocal microscopy fields in various polyamine conditions. (C) Turbidity percentages of dual flanked-A₉ mixed with dual flanked-U₉ in both polyamines. All experiments were completed in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5.

NRAS demonstrates phase separation and aggregation depending on additives

We established behavior for model GQ oligonucleotides in biological amino acids, nucleotides, as well as polyamines. However, biological sequences contain variability in loop regions and flanking sequences that could change behavior in these biological conditions. To test this, the neuroblastoma RAS viral oncogene (NRAS) putative GQ was selected for further experimentation. This sequence has been shown to fold into GQ structures in vitro and in vivo (Kumari et al. 2007; Chen et al. 2016), and mutations to the GQ regions have been shown to influence gene expression (Kumari et al. 2007). We tested the biological NRAS sequence in a variety of individual conditions, as well as a series of piecewise combined conditions to better represent a cell.

Melting experiments in biological polyamines did not display a significant difference from biological salts alone suggesting that this sequence is not as likely to phase separate as the model sequences (Supplemental Fig. 9A). Consistent with this notion, CD spectroscopy in the presence and absence of 0.3 mM spermine showed similar signatures (Supplemental Fig. 9B). We then looked for solution conditions that might induce phase separation on the biological sequence. Addition of spermine to 0.5 and 0.9 mM also did not show phase separation; however, 1.5 mM spermine and above produced phase separation (Supplemental Fig. 10). We tested the effects of some of the small molecules. In biological salts, 16 mM arginine did not induce phase separation (Supplemental Fig. 11A), while 2.7 mM CTP did not disrupt phase separation caused by 3.0 mM spermine (Supplemental Fig. 11B).

Cells are crowded in nature by a large variety of biological macromolecules such as proteins, polysaccharides, and nucleic acids (Ellis 2001). Some crowders including PEG have been shown to facilitate phase separation of GQ structures (Zhang et al. 2019). We wanted to test PEG of various weights with NRAS to determine if it could induce phase separation. At 5% PEG 950-1050 and PEG 4000, small amounts of phase separation were observed (Supplemental Fig. 12, row 2, columns 1 and 2). However, larger PEG molecules did not produce the same effect (Supplemental Fig. 12, row 2, columns 3 and 4). In solutions with NRAS and 0.3 mM spermine, which alone did not induce phase separation (see above), all weights of PEG produced droplets (Supplemental Fig. 12, row 3). Increasing the concentration of spermine did not significantly change the propensity or morphology of PEG phase separation (Supplemental Fig. 12, row 4). This indicates a potential for cooperative and additive interactions between RNA, crowding reagents, and spermine, which lead to LLPS in the system.

For robustness, we wanted to test a secondary crowding reagent in similar conditions. Dextran 9000–11000 was added to NRAS and varying spermine concentrations. Control samples without dextran and relatively low concentrations of spermine showed minimum phase separation (Fig. 10, row 1). However, when we added 5% dextran to the samples with spermine, phase separation occurred (Fig. 10, row 2). In as little as 0.5 mM spermine and 5% dextran, phase separation occurred and droplets appeared mostly small and homogeneous. However, at 0.75 and 1.0 mM spermine and 5% dextran, larger and irregular clusters of phase separation occurred (Fig. 10, row 2). This demonstrates robustness of phase separation to crowder identity, and the ability of crowder to compound phase separation.

FIGURE 10.

NRAS in the presence of dextran and increasing spermine. Row 1 contains samples with increasing amount of spermine (SPM). Row 2 contains 5% dextran with increasing amounts of spermine. All samples were in the background of 10 mM HEPES (pH 7.5) and K₁₅₀N₁₀M_0.5.

DISCUSSION

In this study, we tested effects of metabolites and flanking sequence on the propensity of model and biological GQs for phase separation. Amino acids at biological concentrations did not affect thermostability, but CMP and CTP at biological concentrations destabilized GQ structure. These results demonstrate that formation of GQ can be suppressed by biological metabolites similar to formation of secondary structure (Lambert and Draper 2007).

We also showed that phase separation of GQ sequences can be diminished by certain 5′- and 3′-flanking regions of the GQ. Flanking sequences that can form a competing secondary structure with guanines (such as oligoC and oligoU) largely prevented phase separation of the RNA. The morphology of phase separation that did occur with the other flanking sequences depended on the position of the flanking region, with 5′-flanking regions inducing round, homogeneous structures, and 3′-flanking regions inducing smaller, amorphous structures. Additionally, GQs appear to be folded in the more droplet-like morphologies based on ThT fluorescence (Fig. 5) and CD spectroscopy (Supplemental Fig. 5), indicating a requirement for the GQ to be folded for phase separation.

Dual flanked-GQ structures had lower tendencies for phase separation in the presence of spermine than 5′-flanking- and 3′-flanking sequences. In addition, dual flanking sequence led to lower melting temperatures suggestive of interference with GQ formation. However, when potential interactions of flanking sequence with core GQ structure were limited by combining sequences with complementary flanking regions, phase separation was restored.

Finally, we demonstrated that the GQ-forming biological NRAS sequence phase separates in spermine when a crowding reagent, dextran or PEG, is added to the solution. The identity of the crowder seems to change the morphology of the phase separation somewhat, with PEG containing smaller, round, homogeneous moieties, and dextran having “sticky” morphologies that cluster together.

In summary, the studies herein show that GQ stability can be diminished by metabolites and pyrimidine flanking sequence. At the same time, A-rich flanking sequence allows GQ formation, and additional sequence that interacts with interfering sequence can restore GQ formation and even phase separation. Elucidating how cellular components change RNA structure and stability is important for understanding and targeting RNA function in vivo.

MATERIALS AND METHODS

Oligonucleotide preparation

Sequences were prepared as per Williams et al. (2021). Briefly, RNA was synthesized by Integrated DNA Technologies (IDT), spin dialyzed for desalting, and buffer exchanged into at least four volumes of 10 mM HEPES (pH 7.5). Concentrations were calculated via a Denovix nanodrop and then stored at −20°C.

UV thermal denaturation

Amino acid- and nucleotide-containing melts

Samples containing 10 µM RNA were incubated at 95°C for 2 min with all buffer and salt components (150 mM K⁺, 10 mM Na⁺, and 10 mM HEPES [pH 7.5], plus Glu-Mg or NTP-Mg as appropriate [Yamagami et al. 2018, 2019]). Samples were allowed to cool on the benchtop for 10 min then Mg²⁺ was added. This solution was allowed to equilibrate for another 10 min on the benchtop. Samples were melted on an OLIS-refurbished HP 8425 diode array spectrophotometer from 20°C to 95°C with a ramp rate of 0.5°C/min and scans taken every 0.5°C. Melting transitions for the NTPs were generally broad, so we normalized to the first data point and directly plotted the data without taking the derivative.

Spermine-containing melts

Preparation was as per Williams et al. (2021). Briefly, samples containing 10 µM RNA were incubated with all salts (150 mM K⁺, 10 mM Na⁺, 0.5 mM Mg⁺², abbreviated K₁₅₀N₁₀M_0.5, and 10 mM HEPES [pH 7.5]), as well as spermine and/or spermidine as appropriate, at 80°C for 2 min, cooled on ice for 1 min, and incubated at 37°C for 5 min. Samples were allowed to acclimate within the spectrophotometer for 5 min at 20°C and then melted as described above. These melting transitions were generally cooperative, so we took a first derivative followed by 11-point smoothing of dA₂₉₄/dT.

CD spectroscopy

Sample preparation was as per “spermine-containing melts.” All spectra were acquired at 25°C on a Jasco J1500 circular dichroism (CD) spectrometer, with the wavelength ranging from 200 to 350 nm. Samples were scanned at a speed of 50 nm/min. Data reported are the average of three scans, and all samples are buffer corrected in HEPES/salts mixture, as well as spermine as appropriate. For samples containing A₉-flanking regions, a correction factor of 9/10 the signature of A₁₀ was applied for each A₉-flank.

Confocal microscopy

Sample preparation was as per “spermine-containing melts.” RNA was visualized with a Leica TCS SP5 laser scanning confocal inverted microscope (LSCM). Using a 63× objective lens, at least three fields were obtained with a size of 82 × 82 µm². Samples with Thioflavin T (ThT) were prepared to a final concentration of 0.5 µM ThT and renatured in the same way. These samples were visualized on an Olympus Fluoview 1000 Confocal Microscope using a 405 nm laser at 30% power and a BS 20/80 filter. At least seven fields of 53 × 53 µm² were obtained for all samples.

Turbidity measurements

Sample preparation was as per “spermine-containing melts.” Absorbance at 350 nm was collected on the Olis spectrometer. Absorbance measurements were converted to turbidity as follows:

$$\mathrm{Turbidity}(\text{\%})=100-{10}^{\wedge }[2-(A_{350\text{nm, sample}}-A_{350\text{nm, buffer}})]$$ (1)

Turbidity experiments were completed in technical triplicate at 20°C.

High-throughput aggregation screen

Sample preparation was as per “spermine-containing melts.” After renaturation, samples were loaded in technical triplicate into a 96-well tray and analyzed using a Zeiss Colibri 7 microscope at room temperature. The wells were analyzed using a 40x air objective lens. Three 220 × 166 µm² images per sample were captured and representative subsections of the images are shown.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.

MEET THE FIRST AUTHOR

Allison Williams.

Meet the First Author(s) is a new editorial feature within RNA, in which the first author(s) of research-based papers in each issue have the opportunity to introduce themselves and their work to readers of RNA and the RNA research community. Allison Williams is the first author of this paper, “Biological solution conditions and flanking sequence modulate LLPS of RNA G-quadruplex structures.” Allison is a graduate student in the laboratory of Dr. Philip Bevilacqua at Pennsylvania State University. Her primary research focus is the influence of in vivo-like conditions on RNA structure.

What are the major results described in your paper and how do they impact this branch of the field?

There are two major results in this paper. The first is that common metabolites such as amino acids and most nucleotides do not destabilize G-quadruplex structures, while cytosine derivatives have the potential for destabilization. The second finding is that the location and sequence content of G-quadruplex flanking regions can modulate phase separation. These results are exciting because they give both a point of modulation for G-quadruplex stability and the potential to understand the selectivity of G-quadruplex phase separation.

What led you to study RNA or this aspect of RNA science?

I became interested in RNA research when I started to learn about gene regulation at Duquesne University for my freshman undergraduate biology class. From that moment on, I was completely hooked on learning more about how this “intermediate” molecule could have such interesting and dynamic roles in cellular regulation, as well as how it could change conformation and structure.

During the course of these experiments, were there any surprising results or particular difficulties that altered your thinking and subsequent focus?

During the course of this paper, we were very surprised to see a difference in phase separation morphology when we changed the A-rich and mixed flank from the 5′ to the 3′ end. Seeing droplets for the 5′ flanked sequences allowed for asking further questions about the G-quadruplex folding within the droplets, which we could not interrogate for aggregates.

What are some of the landmark moments that provoked your interest in science or your development as a scientist?

In late high school and early college, my favorite experiment to learn about was the discovery of the semiconservative method of DNA replication by Meselson and Stahl. The experiment was so elegant and simple, yet answered an incredibly complex question. It has inspired me throughout my time in the Bevilacqua laboratory to try to answer complex questions in a systematic way.

If you were able to give one piece of advice to your younger self, what would that be?

The advice that I would give to my younger self would be don't be afraid to try something, either in science or everyday life. In science, the worst that can happen is an experiment doesn't work. But, in just trying something, you could discover something new and unexpected.