Switchable RNA motifs for dynamic transcriptional control of RNA condensates

Abstract

RNA-driven phase separation is emerging as a promising approach for engineering biomolecular condensates with diverse functionalities. Condensates form thanks to weak yet specific RNA–RNA interactions established by design via complementary sequence domains. Here, we demonstrate how RNA condensates formed by star-shaped RNA motifs, or nanostars, can be dynamically controlled when the motifs include additional linear or branch-loop domains that facilitate access of regulatory RNA molecules to the nanostar interaction domains. We show that condensates dissolve in the presence of RNA “invaders” that occlude selected nanostar bonds and reduce the valency of the nanostars, preventing phase separation. We further demonstrate that the introduction of “anti-invader” strands, complementary to the invaders, makes it possible to restore condensate formation. An important aspect of our experiments is that we demonstrate these behaviors in one-pot reactions, where RNA nanostars, invaders, and anti-invaders are simultaneously transcribed in vitro using short DNA templates. Our results lay the groundwork for engineering RNA-based assemblies with tunable, reversible condensation, providing a promising toolkit for synthetic biology applications requiring responsive, self-organizing biomolecular materials.

Graphical Abstract

Graphical Abstract.

Introduction

Molecular condensation is a key biological process associated with the formation of membraneless organelles [1]. These organelles play diverse roles in cells, including pre-RNA processing, ribosome biogenesis, messenger RNA (mRNA) decay [2], DNA repair [3], and ubiquitylation [4, 5]. By compartmentalizing biomolecules, condensation helps isolate specific reactions from the cellular bulk phase, facilitating the concentration of intermediates and enhancing enzymatic synergies [6]. The emergence of spatially distinct phases is typically driven by weak multivalent interactions that form, break, and reform rapidly [1, 2, 7, 8], and overcome the entropic gain that would keep molecules well-mixed [9–11].

There is growing interest in biocondensation and in the development of artificial condensates, which could cluster enzymes, sequester toxic metabolites, or control transcription factors, allowing for fine-tuned metabolic regulation in cell factories [11–13]. Artificial condensates based on sequence-programmable biopolymers, such as proteins tagged with intrinsically disordered regions or low complexity domains, show promise as synthetic biology tools for condensate assembly [2, 10]. Similar progress in RNA-based condensates has also led to emerging technologies in this field. In particular, the discovery of condensation-prone RNA sequences linked to neurological diseases, such as long RNAs with expanded short repeats, has enabled the creation and functionalization of artificial RNA condensates [14, 15]. Recently, nanostructured RNA motifs, known as nanostars (NSs), have been designed to yield pure RNA condensates in vitro [16–18], taking inspiration from similar motifs in DNA nanotechnology [19–24]. To enable isothermal assembly under transcription conditions, RNA NSs are formed by a single RNA strand (120–300 nucleotide-long) that folds into a sequence of stem–loops, where NS interactions are mediated by loop–loop binding. Loop sequences, also known as kissing loops (KLs) can be systematically designed to obtain orthogonal NSs forming multiple noninteracting condensates. The modularity of this motif makes it possible to include protein binding motifs in the NS design [16, 17], bringing us closer to the goal of creating artificial membraneless organelles capable of reorganizing existing enzymes into co-localized clusters, without altering the total enzyme quantities.

Here, we demonstrate methods to advance the programmability of RNA condensates by introducing and characterizing RNA NSs whose capacity to form condensates can be modulated via hybridization with RNA inputs. We propose two alternative methods to achieve condensate dissolution and reformation through the transcription of RNA inputs. The first method involves the addition of a single-stranded overhang (toehold) to one of the NS stems [25] that becomes misfolded upon hybridization of the overhang with the RNA input. The second method involves the use of a branched KL, which makes it possible for RNA inputs to invade and thus occlude KL interactions [26, 27]. We demonstrate the feasibility of both methods in samples where NSs and inputs are transcribed simultaneously from DNA templates, creating a biologically relevant context. Our designs are immediately fungible in cell-free applications [28], where they could provide means to build dynamically responsive membraneless organelles, and may be ported to RNA condensates forming in living cells [29]. Fine control over the formation and dissolution of biocondensates represents a powerful tool that could be added to the synthetic biology methods used for metabolic engineering via phase separation of biomolecules.

Materials and methods

Sequence design

Three-armed single-stranded RNA NSs were designed on NUPACK by combining distinct 15 nt-long arm sequences with a KL (5′-GCGCGU-3′). Two of the three spacers between arms include two unpaired adenines, and the other is a nick, allowing the motif to have flexible configurations. Branched KL sequences were designed based on Sampedro Vallina’s work [26], by adapting the design to fit the NSs KL sequences using ROAD origami design software [30]. The NS and KL sequences were constrained, and the remaining unconstrained bKL sequence was designed using the perl script “batch_revolvr.pl” available at https://github.com/esa-lab/ROAD and as a web server at https://bion.au.dk/software/rnao-design/.

Cotranscriptional production of condensates

RNA strands were transcribed from DNA templates, including a T7 promoter, purchased from Integrated DNA Technologies. Lyophilized DNA was resuspended in nuclease-free water and DNA template was annealed in 1× TE/50 mM NaCl from 90°C to RT at −1°C/min. RNA strands were transcribed in vitro at 37°C using 7.5% (v/v) T7 polymerase from the AmpliScribe T7-Flash transcription kit (ASF3507, Biosearch Technologies), 40 mM of Tris–HCl, 10 mM of NaCl, 30 mM MgCl₂, 2 mM spermidine, 7.5 mM each NTP, 10 mM dithiothreitol (DTT), and 10 nM DNA template unless otherwise specified.

Fluorescence microscopy

Samples were stained with SYBR gold right before imaging using a Nikon Eclipse TI-E inverted microscope using a ×60 oil immersion objective. SYBR gold-stained samples were detected using the FITC channel (ex 455–485 nm/em 510–545 nm) with an exposure time of 100 ms.

Image processing and data visualization

All fluorescence images were processed in FIJI (ImageJ). Raw images were background subtracted, contrast-enhanced, and converted to a binary mask as described in detail in the Supplementary Information. For condensate number analysis, objects smaller than 6 px2 were considered noise and excluded. Condensate numbers were autogenerated by FIJI and recorded. To gather information on condensate size, we measured chord length distributions (CLDs) [31, 32] from the binary masks using a Python3 script based on PoreSpy, which relies on Scipy and Skimage. Starting from binary masks of epifluorescence images, chords were obtained by measuring the intersections between straight lines and regions corresponding to condensates. CLDs are useful to characterize the length scale of condensates, independent of their shape, which can range from spherical to aggregated network structures. Total area information was extracted from the binary masks using FIJI and a Python3 script. Faded area shows the standard deviation from the mean. Violin plots were made using the Superviolin Python Package [33] and combine data from three experimental replicates (FOV = 10) into one violin plot.

Data and code availability

Data and custom code for image analysis are available through public repositories, and links are provided in the Supplementary Information Materials and Methods.

Results and discussion

RNA condensates

We produced condensates from single-stranded RNA NSs whose design was inspired specifically by the “3sv2” NS variant described in Stewart et al. [16], which showed promising co-transcriptional condensation properties at a temperature close to 37°C. The NSs include three stem–loop motifs, or arms, interacting through KL domains [16, 17]. Arms are 15 bp long (Fig. 1A) and are spaced by two unpaired adenines to increase arm mobility thereby promoting condensation [34]. Each KL consists of nine nucleotides, incorporating a six-nucleotide interaction sequence flanked by three unpaired adenines, two upstream and one downstream of the interaction sequence (5′-AAGCGCGUA). The specific base-pairing interactions of the NSs’ palindromic KLs promote condensation (Fig. 1B), which occurs isothermally as RNA is transcribed in vitro [16]. We produced RNA NSs from linear DNA templates with a T7 promoter upstream of the NS sequence (Fig. 1B and Supplementary Table S1). Samples were stained with SYBR Gold prior to microscopy characterization (Fig. 1C and D). The addition of SYBR Gold has a minor influence on the condensate length scale statistics, and it is therefore not expected to have a significant impact on nucleation and fusion (Supplementary Fig. S0).

Figure 1.

(A) 2D schematic of a single-stranded RNA “nanostar.” (B) NSs are transcribed from DNA templates in vitro, and bind through palindromic KL interactions leading to the formation of RNA-rich condensates (C). (D) Example fluorescence microscopy image of RNA NS condensates stained with SybrGold. (E) In this study, we seek to achieve control over condensation and dissolution of the NS condensates by introducing distinct molecular inputs. (F) Our approach to control condensation is based on the addition of invader and anti-invader strands, that respectively prevent and restore the KL interactions and thus condensation; scale bar: 25 μm.

Our goal here is to engineer RNA NSs to achieve control over the formation and dissolution of condensates through other molecular inputs supplied during in vitro transcription (Fig. 1E). For this purpose, we used nucleic acid nanotechnology methods to regulate the NS valency [25]. Specifically, we used the mechanisms of strand invasion to sequester one or more of the KL domains through a complementary “invader” strand. If valency is reduced <3, the formation of condensates is suppressed [16]. We can restore the NS valency back to 3 by supplying an anti-invader molecule complementary to the invader (Fig. 1F). A challenge toward realizing these interactions in practice is posed by the fact that invasion must occur at the KL, and is disfavored when two KLs are bound. Typically, linear, single-stranded overhangs known as “toeholds” must be included to initiate and promote strand invasion and displacement [35, 36]. Because it is not possible to include toeholds near KLs in single-stranded NSs, we devised two alternative strategies to promote KL invasion.

It is important to note that RNA NSs are expected to form during transcription as proximal complementary domains fold into stable stems within milliseconds [37]. In this case, condensation is driven by KL interactions, rather than stem–stem binding [16]. In contrast, thermal annealing (which includes a melt phase) can foster stem–stem interactions [16]. To avoid this circumstance and possible confounding effects, here we consider the NS response exclusively under isothermal conditions: NSs and RNA molecules interacting with NSs are all transcribed from templates in the same reaction. We do not perform experiments with purified and annealed RNA components, as these steps would affect the NS secondary/tertiary structure formed during transcription.

RNA condensate dissolution through toehold-mediated strand invasion and displacement

Our first strategy to enable the control of RNA condensate formation and dissolution is to switch the conformation of one NS arm through toehold-mediated strand invasion (Fig. 2A). As the minimum NS valency required for condensation is 3 (Supplementary Fig. S1), we expected that disrupting a single arm and KL (bringing the valency down to 2) could either prevent the formation or cause the dissolution of condensates, as we previously demonstrated with DNA NSs [25]. For this purpose, we modified our RNA NSs to include a toehold domain at the 5′ end of the strand. To facilitate disruption of the KL, we moved the nick in the NS motif to be within the first arm (Fig. 2A and B), rather than being centered at the NS core like in previous work (Fig. 1A). We designed NS variants differing by toehold length, testing 6, 8, and 10 nucleotides (nt) long toeholds (Fig. 2B). As illustrated in Fig. 2A, we designed invader strands to be fully complementary to the toehold (blue domain), the arm (black), and part of the KL domain (red). The binding of the invader is predicted to unwind the first stem of the NS and impede KL–KL interactions with other NSs (Fig. 2A and Supplementary Fig. S2).

Figure 2.

Cotranscriptional condensation of toehold NSs and the dissolution via the toehold-mediated invasion. (A) Schematics of the toehold-mediated invasion. (B) 2D representation of the three different single-stranded RNA NS designs with a 6-, 8-, or 10-nt toehold. (C) Invader inhibits the condensate formation when supplied at the start of the transcription. Sample microscopy images were taken for 3 h of transcription and staining with SYBR Gold. (D) Violin SuperPlots of the CLD showing condensate growth from 1 to 3 h of transcription, with or without invader. (E) Schematics of the toehold-mediated invasion. The 10× invader DNA template was added 1 h after transcription of the 6-nt toehold NS. (F) Images were taken immediately after the addition of the invader and control invader (nonbinding sequence). (G) Plots show the progressive dissolution of the condensate, with the shaded area showing the sample standard deviation. Experiments were repeated three times, and during each experiment we collected and processed 10 FOVs for each time point. CLD violin superplots are in Supplementary Fig. S8; scale bars: 20 μm.

All NS designs including a toehold formed condensates isothermally during in vitro transcription using their respective DNA template (Fig. 2C). Representative microscopy images were all taken after 3 h of transcription incubation and processed to obtain CLDs of the condensates detected (Fig. 2D, left). CLDs are an expedient approach to gather information about the length scale of objects that present polydisperse, irregular shape or aggregation [16, 17]. We noted that condensates formed by the 8-nt toehold NS are generally smaller than the other designs. We estimated the cotranscription yield of the three RNA NSs using denaturing polyacrylamide gel electrophoresis. The 8-nt toehold NS exhibited considerably weaker bands compared to the other two NS constructs (Supplementary Fig. S3), which explains the smaller condensate size observed.

We first asked whether condensate formation could be prevented in the presence of invader. To test this, we transcribed RNA NS template and 10× invader template in the same, one-pot sample. We monitored our samples for 3 h and found no visible condensates, as shown in Fig. 2C and D, right. We verified correct RNA transcription using a native PAGE gel (Supplementary Fig. S4), confirming that the lack of visible condensates was not due to RNA degradation. RNA condensate formation was suppressed when a 1× invader template (10 nM) was supplied for the 6- and 8-nt designs, while the 10-nt design required at least a 4× invader template (Supplementary Fig. S4A and B). This requirement for the 10-nt design is likely due to insufficient invader transcription yield (Supplementary Fig. S4C). The concurrent transcription of NS template alongside a large excess (10×) of control invader (nonbinding) resulted in typical condensate formation (Supplementary Fig. S4B), suggesting that under our experimental conditions the NS production is not hindered by the addition of new templates that introduce competition for limited transcription resources (RNA polymerase and nucleotides). Two additional control experiments were performed to validate the condensate suppression mechanism expected to occur with the toehold system. In the first one, we transcribed our RNA NS without a toehold alongside its invader (Supplementary Fig. S5). In the second one, we transcribed the 8-nt toehold RNA NS with a control invader that is designed to not bind to the toehold and arm (Supplementary Fig. S6). Both experiments resulted in the formation of condensates, thereby confirming the expected mechanism in which the interaction of the invader and the toehold/arm prevents NSs from condensing.

Next, we tested the capacity of invaders to dissolve condensates that have already formed. To investigate this, we supplied the invader template to a sample in which RNA condensates were already formed isothermally by transcribing RNA NSs from their template for 1 h (Fig. 2E). The 1-h incubation mark for adding invader template was selected to ensure a comparable dissolution timescale: preliminary experiments showed that dissolution speed depends on the condensate size and large condensates can take many hours to dissolve, as one would reasonably expect. A significant decrease in the condensed mass and number of condensates was detected within 2 h from addition of invader template corresponding to the 6-nt toehold design, at a 10× concentration relative to RNA NS template (Fig. 2F and G, bottom). Condensates continued to grow in samples when 1× template was used (Supplementary Fig. S7) or when 10× of a template producing the control invader (noncomplementary to the target KL) was added 1 h into NS transcription (Fig. 2F, top). Notably, the number of condensates also decreased in the control experiments due to fusion events, as expected (Fig 2G, top and S8). Complete dissolution of small RNA condensates was observed for the 8-nt toehold NS condensates, while larger droplets exhibited a reduction in both size and number (Supplementary Fig. S9A). For the 10-nt toehold NS condensates, the number of condensates dropped significantly but never reached complete dissolution even after 4 h (Supplementary Fig. S9B). We also assessed NS constructs with the toehold placed at the central junction (called nick-in-core) with varying toehold lengths. While invader templates successfully suppressed condensate formation when added at the start of the co-transcription reaction, they failed to dissolve condensates that had already formed (Supplementary Figs S10 and S11).

Invasion of branched kissing loop

Our second strategy to enable the control of condensation and condensate dissolution takes advantage of the branched kissing loop motif (bKL) [26, 38]. The bKL design features, illustrated in Fig. 3A, include an additional 5 base pair stem with a 7-nucleotide apical loop upstream of the KL, depicted in blue (additional details are in Supplementary Fig. S12). We designed invader strands to be complementary to all nucleotides in the bKL interaction domain as well as most of the branched apical loop. We expect that this design promotes invasion as the 3′ end of the invader binds to the branch apical loop, which does not participate in the KL–KL interactions (Fig. 3A). We considered three NS variants differing by the number of bKL, as illustrated in Fig. 3B: in the first, we replaced a single KL with a bKL; in the second, we replaced two KLs with bKLs; in the third variant, all KLs were replaced by bKLs. The length of these RNA strands was validated using denaturing PAGE (Supplementary Fig. S13).

Figure 3.

Cotranscriptional condensation of bKL NSs and dissolution by invasion of the KL. (A) bKL invasion principle. (B) 2D representation of the three different single-stranded RNA NS designs carrying one, two, or three bKLs. (C) Effect of invader strand template (10×) cotranscription on the ability of different NS designs to condensate. Imaging after 3 h of transcription and staining with SYBR Gold. (D) Violin superplots of the CLD showing condensate growth with or without initial invader cotranscription. (E) Invasion was performed by adding the invader template (10×) to the 3-bKL design after 3 h of NS template transcription. Imaging after SYBR Gold staining (F), as well as condensate density and total condensate area metrics (G), shows growth or dissolution of condensates. The shaded areas show the sample standard deviation. In the violin superplots in panel (D), colored dots inside the distribution areas indicate the respective CL means of each replicate; dotted/solid lines connect data points as a guide to the eye. Experiments were repeated three times, and during each experiment we collected and processed 10 FOVs for each time point. CLD violin superplots are included in Supplementary Fig. S23; Scale bars: 20 μm.

We began by confirming that bKL NS variants still produce condensates during transcription (Fig. 3C and D). Interestingly, condensates emerged within 1 h of transcription from the 1- and 2-bKL variants, while 3-bKL condensates exhibited slower growth with a tendency to eventually fuse into large condensates (Supplementary Fig. S14). Not all the field of views (FOVs) captured in these samples displayed large fused-like condensate aggregates, but they contained the majority of the total condensate volume (Supplementary Fig. S15). This could be attributed to a possibly weaker interaction between bKLs compared to normal KLs. Although this hypothesis could not be validated by assessing condensate fluidity through a fluorescent recovery after photobleaching (FRAP) assay (Supplementary Fig. S16) showing a lack of recovery within the short observation window (10 min), it is supported by experiments measuring over several hours the speed of condensate fusion, which is significantly faster in the 3-bKL variant when compared to the fusion speed of condensates formed by 6 nt toehold NSs (Supplementary Fig. S17). This hypothesis is also corroborated by oxRNA simulations in Supplementary Fig. S18 that predict different conformations for the bKL arm [39]. In these simulations, bKL arms display a highly flexible structure, that either exhibits the KL stacked to the NS’s arm, or forms a bulge when the stem of the branched hairpin forms a stacking interaction with the stem of the arm instead. A control NS variant in which the interacting bKL sequences (red domains in Fig. 3A and B) were replaced by six adenosines (6A) displayed no condensation after 4 h of transcription, confirming that the branched stem–loop by itself cannot induce condensation (Supplementary Fig. S19).

Next, we tested whether condensate formation could be suppressed when invader template is transcribed together with NSs (note that, as all bKL are identical, a single invader can bind to all of them). Condensate formation was completely suppressed for 2- and 3-bKL designs when we added 10× invader DNA template to the transcription mix from the start of the reaction (Fig. 3C and D, right). In contrast, 1-bKL NSs still formed condensates, although at a slower rate and with significantly reduced size and average chord length (Fig. 3C and D, right). This suggests that invasion of an individual bKL is not sufficient to dissolve the condensates completely, and a significant fraction of the NSs maintains a valency of 3. Native PAGE of the RNA strands of bKL variants with and without RNA invader confirms that the disappearance of visible condensates is due to NS condensate dissolution and not the result of RNA degradation (Supplementary Fig. S20). Additional characterization experiments with different control invader variants confirm that only specific hybridization to the KL causes condensation inhibition (Supplementary Fig. S21). An invader that lacks the domain complementary to the branch loop (no-BL) can still interact with the NSs; this variant suppresses condensation of 3-bKL NSs when it is present from the beginning of the transcription (when condensates are not yet formed) (Supplementary Fig. S21C and E). The no-BL-invader variant works as a competing binding site for free NSs and reduces the fraction of KL–KL interactions, thus preventing NSs from binding to each other and suppressing condensation.

We then verified whether a condensate already formed using bKL NSs could be dissolved by the addition of the invader template. We focused exclusively on 2- and 3-bKL variants, whose condensation is suppressed when invader template is present in the mix from the beginning. In these new experiments, we first transcribed NSs alone, and then supplied the invader template (10×) to the transcription mix when condensates were expected to have formed. The 2-bKL condensates do not fully dissolve when the invader template is introduced after 1 or 2 h of transcription; instead, the condensates either stop growing or exhibit slower growth depending on when the invader template is added (Supplementary Fig. S22). In contrast, 3-bKL condensates completely dissolve when the invader template is added after 3 h of transcription (Fig. 3E), and no condensates are visible 225 min after the addition of the invader template (Fig. 3F and G, and Supplementary Fig. S23). The condensate number falls rapidly after the invader template is added (Fig 3G and Supplementary Fig. S23), yet the total condensate area decreases significantly only after 135 min. This suggests that small condensates dissolve first, whereas bigger condensates are slower to disassemble and may continue to grow. Naturally, the condensate number decreases also in the control experiment without invader template, due to condensate fusion events (Fig. 3F and G). Since control experiments show that an invader lacking the branched loop (BL) domain prevents the formation of condensates (Supplementary Fig. S21C), we asked whether this same variant could cause the dissolution of the 3-bKL condensates. Supplementary Figure S24 shows that condensate growth is unaffected by the addition of no-BL template after three hours of NS transcription; however, some large condensates forming after 3 h display a spongy appearance characterized by “bubbles” of low RNA density. Bubbling in DNA condensates has been observed as a result of catalytic DNA cleavage caused by restriction enzymes trapped in the dense phase. The influx of DNA fragments causes a pressure increase and growth of the bubble until it bursts [40, 41]. While further work is needed to clarify the mechanism in our case, bubbles may emerge if invader templates and RNA polymerases are trapped in the RNA-dense phase: this would increase the local invader concentration, which in turn would accelerate the rate at which NSs unbind from the condensate, forming a low-density vacuole containing inactive NSs. Continued production of invaders could cause the bubble to grow.

Dissolution and growth of condensates with transcribed invader and anti-invader strands

To reverse the capacity of invaders to prevent the formation of condensates and to stop their dissolution, we designed complementary RNA “anti-invader” strands to be used as an additional regulatory input. By sequestering and/or removing invaders bound to NSs, anti-invaders should promote regrowth of condensates. The anti-invader naturally sequesters free invaders, not bound to NSs. To facilitate the displacement of invaders bound to NSs, invaders were designed to include a 5′ end toehold region (Fig. 4A).

Figure 4.

Recovery of the condensate growth by anti-invasion after both toehold and bKL-mediated invasion. (A) Invasion/Anti-invasion cycle for the 6 nt toehold-in-arm system. (B and F) Experiment timeline with subsequent addition of NSs, invader and anti-invader templates. Effect of the addition of invader only or both invader and anti-invader on the condensation dynamics was monitored through imaging after SYBR Gold staining (C and G) and metrics like condensate density and total condensate area were extracted from the microscopy images (D and H). Shaded areas show the sample standard deviation. Experiments were repeated three times, and during each experiment we collected and processed 10 FOVs for each time point. CLD violin superplots are in Supplementary Fig. S31; scale bars: 20 μm.

Condensate formation is recovered when supplementing the NS and invader transcription mix with anti-invader template. This was demonstrated with the 6-nt toehold NSs variant, setting up an isothermal in vitro transcription reaction containing 10 nM NS DNA template, 1× invader template, and varying concentrations of anti-invader template (0×, 1×, 2×, and 5×) (Supplementary Fig. S25). Condensate growth was only observed when the anti-invader template was present. We then tested whether the sequential addition of invader and then anti-invader template could result in condensate dissolution and subsequent regrowth (Fig. 4A). In this experiment, NSs were transcribed for 1 h to produce condensates. We then supplied 10× invader template (Fig. 4B), which resulted in condensate dissolution, confirming earlier results (Fig. 2F). One hour after the addition of the invader template, we added anti-invader template at 100× level relative to the NS template (Fig. 4B). Example images shown in Fig. 4C shows that addition of anti-invader results in condensate regrowth (ii), in contrast with a control to which anti-invader template is not supplied (i). The length scale and number of RNA condensates drop following invasion and recover 3 h after addition of anti-invader template (Fig. 4D).

We then sought to demonstrate condensation recovery by anti-invasion after dissolution using the bKL NS systems. One immediate challenge arises from the fact that the KL design we adopted is palindromic, meaning that anti-invaders complementary to the invader also share complementarity with the KL domains. This became evident when we introduced a toehold at the 5′ end of the bKL invader sequence to facilitate its displacement from 3-bKL NSs and tested the effect of the addition of invader and anti-invader. The addition of 20× anti-invader template alone to the NSs (expected concentration to neutralize efficiently 10× invader) showed complete initial inhibition of the condensate growth (Supplementary Fig. S26). Similar behavior was not observed with the toehold-in-arm NS design, because the invader and anti-invader sequences only contain a fragment (3-nt) of the KL sequence.

In order to circumvent this design-specific challenge, we adopted a NS variant that includes two NSs with nonpalindromic, complementary KL (Supplementary Fig. S27A) [16]. Condensates can form only when both NSs A and B are transcribed together, and no condensates form when only one of the two is produced (Supplementary Fig. S27B and C). Hence, we reasoned that to dissolve condensates through invasion, the KL of only one NS of the two needs to be targeted. Based on this expectation, we modified only one of the NSs in this system and replaced each KL with a bKL, leaving the second NS unchanged. Denaturing PAGE of both NS A and B allowed for validation of the transcript sizes (Supplementary Fig. S28). In agreement with our intuition, we found that condensates could be dissolved in the presence of 10× invader template (Fig. 4H and Supplementary Fig. S29). We then verified that transcription of the anti-invader template together with the two templates transcribing NSs resulted in normal condensate formation (Supplementary Fig. S30). This supports that changing to a nonpalindromic KL prevented the anti-invader from acting as an invader. Finally, we showed that the sequential addition of invader and anti-invader templates causes dissolution of the condensates followed by regrowth ((ii) in Fig. 4E–H). Because the two-NS system yields condensates rapidly, we started the invasion 0.5 h after transcription began, once clear condensates were formed already (Fig. 4F). Since invasion appeared to affect the 2-NSs system faster when compared to the single-NS system, anti-invasion was initiated 0.5 h after invader addition (Fig. 4G and H, and Supplementary Fig. S31). Overall, condensate number and total condensate area clearly show that invader production causes dissolution, and anti-invader production causes regrowth of condensates during transcription at constant temperature, confirming that our approach can switch the state of NSs from inactive to active.

Conclusion

In this study, we successfully demonstrated two innovative systems for controlling the condensation of RNA NSs through RNA molecules that suppress or promote NS interactions by binding to the NSs in trans. Our approach involves the incorporation of toehold-mediated strand invasion and bKL motifs in the NS design, to facilitate binding. Both strategies allow for the tuning of RNA NS condensation in response to RNA inputs transcribed in the same pot as the NSs, showcasing their potential for precise manipulation in synthetic biology applications. We established that the addition of an invader strand template effectively dissolves RNA condensates, and this process can be reversed by introducing a complementary anti-invader strand template, resulting in the regrowth of condensates.

Notably, the suppression of condensate formation was particularly efficient with the toehold-in-arm system, because invasion of only one KL region was sufficient to cause condensate dissolution. In contrast, bKL designs require that all three arms of the NS include the bKL domain and can thus be invaded. Similarly, our results indicate that anti-invasion more effectively recovers condensate growth with the toehold-NS design. Following the addition of anti-invader template strands, both the density of condensates and the total area increased beyond pre-invasion values. This advantage stems from the reduced number of invaders needing to be replaced in the toehold design compared to the three bKLs in the alternative system, where the theoretical replacement count is three times higher. While this study primarily characterized a single bKL sequence design as a proof of concept, we believe that optimizing the sequence of the branched loop could further enhance invasion efficiency. For instance, increasing the length of the apical loop while maintaining some nucleotides free upstream of the invader binding site may facilitate the initial binding of the invader. Modifications to the overall NS design may also accelerate condensate formation and dissolution. For example, we noted how the introduction of bKL causes more rapid formation of large condensates when compared to designs with toehold, likely due to increased NS mobility.

Overall, our findings establish a foundation for versatile RNA-based systems that enable dynamic control over condensation. Because our RNA condensates form isothermally in commercial transcription kits, as well as in protein translation kits [16], they could be immediately useful as membraneless organelles for in vitro synthetic biology [28], and preliminary work by our group shows their promise as RNA organelles in mammalian cells. While achieving dynamic condensation in living cells is likely to be challenging, it may be achieved through further characterization and optimization of invasion efficiency. The modularity of RNA NSs could also allow us to combine the invasion domains presented here with protein recruitment domains, such as protein-binding aptamers previously characterized on RNA NSs (16), enabling dynamic control of protein aggregation. Dynamically controllable organelles could introduce temporal sequestration of other RNA molecules and RNA-binding proteins, achieving spatio-temporal control of translation and enzymatic function [11]. Finally, beyond opening new avenues for creating responsive biomolecular systems and materials [42], our bottom-up strategies for regulating condensate formation and dissolution may provide cues to understand the role of RNA in biological condensates [43].

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

The Regents of University of California has filed a patent application in the US Patent and Trademark Office which includes disclosure of inventions described in this manuscript, Provisional Application Serial No. 63/588,142, filed on 5 October 2023, and entitled: SINGLE STRANDED RNA MOTIFS FOR IN VITRO COTRANSCRIPTIONAL PRODUCTION OF ORTHOGONAL PHASE SEPARATED CONDENSATES.

Funding

This research was supported by the US National Science Foundation through awards BBSRC-NSF/BIO 2020039 and FMRG Bio 2134772 to E.F., by the Alfred Sloan Foundation through award G-2021-16831, and a Novo Nordisk Foundation Ascending Investigator grant (0060694) to E.S.A. Research reported in this publication was also partially supported by the National Institute of General Medical Sciences of the National Institutes of Health under award number R35GM155833 to E.F. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. E.F. also acknowledges funding from the UCLA Eli and Edythe Broad Center of Regenerative Medicine and Stem Cell Research Rose Hills Foundation Innovator Grant. Funding to pay the Open Access publication charges for this article was provided by the Alfred Sloan Foundation.

Data availability

Supplementary datasets are available on a Zenodo repository: https://zenodo.org/records/15010536. All scripts used for data analysis are available on GitHub, https://github.com/FrancoLabUCLA/Tang-Gobry-2025-NAR, and Zenodo, https://doi.org/10.5281/zenodo.15309506.