RNA aggregation is enhanced in multi-component biomolecular condensates
To study the emergent role of homotypic RNA-RNA interactions within heterotypic protein-nucleic acid condensates, we utilized a model condensate system amenable to quantitative experimental interrogation. This model system consists of an RNA binding motif-inspired multivalent disordered polypeptide [RGG; sequence: (RGRGG)5] and a single-stranded nucleic acid [poly-thymine DNA, d(T)40]. Phase separation in RGG-d(T)40 mixtures is driven by obligate cation-π and electrostatic protein-nucleic acid interactions8. Furthermore, the material and physical properties of RGG-nucleic acid condensates are fully tunable via peptide and nucleic acid sequence and length8, 47, providing a robust means to tune the condensate microenvironment. Introducing additional components, such as client RNAs featuring specific primary sequence and secondary structures, to this model condensate allows us to systematically probe the effects of compositional complexity and RNA-driven changes on condensate physical properties (Fig. 1a). We hypothesize that the introduction of RNAs capable of forming homotypic intra- and inter-molecular contacts within RGG-d(T)40 condensates would result in one of two outcomes: the formation of a homogenous ternary condensate where the RNA fully mixes with the two primary condensate-forming components, RGG and d(T)40, or a multiphasic condensate with RNA preferentially partitioning into one of the two co-existing phases. A possible third outcome would be a homogenous ternary condensate undergoing age-dependent transformation into a multiphasic condensate due to RNA demixing (Fig. 1a). In our first set of experiments, we put this idea to the test by utilizing the naturally occurring telomeric repeat-containing RNA48, 49 [TERRA, sequence: (UUAGGG)n] (Fig. 1b), which is known to form G-quadruplex (GQ) structure50, 51, 52.
Without multivalent cofactors such as Mg2+ ions, (TERRA)10 alone remains fully soluble over a period of 24 hours at room temperature (Fig. 1c). When introduced as a client to RGG-d(T)40 condensates, (TERRA)10 preferentially partitions in the dense phase and remains homogeneously distributed at initial time points (Supplementary Fig. 1). Using a nucleic acid-binding dye (SYTO-13), we next tracked whether intra-condensate spatial distribution of TERRA changes with condensate age. Remarkably, we find that TERRA undergoes age-dependent demixing inside RGG-d(T)40 condensates leading to the formation of RNA clusters (Fig. 1d). As a result, a distinct multiphasic architecture emerges that is defined by an RNA-rich inner core surrounded by an RNA-deplete shell. Utilizing fluorescently labeled anti-sense TERRA, we confirmed that the inner SYTO-13-positive sub-phases in aged condensates are indeed formed by TERRA clusters (Supplementary Fig. 1). The outer phase of aged condensates displays liquid-like dynamics such as fusion with the outer phase of nearby condensates, which is not observed for the RNA-rich core (Fig. 1d; Supplementary Video 1). To elucidate the molecular compositions of these distinct phases, we employ a pairwise imaging approach and identify the localization of each component. The TERRA-enriched clusters within the condensate are significantly depleted of RGG (probed by Alexa594-labeled RGG) and d(T)40 [probed by Cy5-labeled d(T)40] (Fig. 1e-h). Line profiles illustrate the anti-correlation between TERRA and either of these two components (Fig. 1f, h). Moreover, the RNA clusters are shown to be quite irregular in morphology (Fig. 1i). Replacing d(T)40 with r(U)40 as a condensate forming component in the ternary system did not alter the age-dependent clustering of TERRA (Supplementary Fig. 2). However, we do not observe this time-dependent appearance of RNA clusters in the dense phase of binary condensates composed of TERRA and RGG (Supplementary Fig. 3). These observations suggest that age-dependent RNA demixing is an emergent property of the ternary condensates. Overall, these results demonstrate that RNA partitioning into multi-component biomolecular condensates can drive age-dependent homotypic RNA clustering (Fig. 1j). Next, we attempt to dissect which features of the RNA molecules enable this emergent behavior.
Intra-condensate RNA aggregation is driven by RNA percolation transition
In the absence of proteins, RNA can undergo reversible temperature-dependent phase separation with a secondary percolation transition43 (Fig. 2a). The percolation transition, which is manifested by homotypic RNA-RNA interactions and sensitively depends on the RNA sequence and secondary structure43, dynamically arrests RNA in the dense phase and makes RNA phase separation irreversible. Therefore, RNAs with a strong percolation propensity tend to form irreversible condensates in contrast to RNAs with a weak or no percolation propensity (Fig. 2a). We hypothesized that the demixing of (TERRA)10 in the dense phase of the multicomponent protein-nucleic acid condensates (Fig. 1; Supplementary Fig. 2) stems from a strong percolation propensity of the RNA. To test this idea, we performed temperature-controlled microscopy of (TERRA)10. We observed that (TERRA)10 [1.0 mg/ml RNA in 50 mM HEPES (pH 7.5), 6.25 mM Mg+ 2] remains homogenous at 20°C but undergoes an irreversible phase transition with a lower cloud point temperature (LCPT) of T = 60.2 ± 1.3°C upon heating. The irreversibility of the (TERRA)10 condensates upon cooling to 20°C suggests the formation of a strongly percolated RNA network in the dense phase (Fig. 2b, c; Supplementary Video 2). Using a series of Mg2+ titrations, we mapped the state diagram of (TERRA)10 and observed that the phase separation coupled to percolation behavior of (TERRA)10 is extremely sensitive to Mg2+ concentration with a very narrow range of Mg2+ concentrations (5.75 to 6.5 mM) where thermo-responsive phase separation is experimentally observable (Fig. 2c). At Mg2+concentrations higher than 6.5 mM, we found that (TERRA)10 always exists as percolated clusters even at the lowest temperature tested, 2.0°C (Fig. 2c). At temperatures above percolation temperature (Tprc), (TERRA)10 clusters underwent shape relaxation into energetically favorable spherical condensates that persist after cooling, thereby irreversibly trapping the RNA in the condensed state (Fig. 2d; Supplementary Video 3). These results therefore reveal a strong percolation propensity of (TERRA)10, which is likely to drive its age-dependent aggregation within RGG-d(T)40 condensates.
An important molecular feature of RNA percolation is the contribution of purine-mediated base-pairing and stacking interactions43. It is also known that G-rich sequences of RNA can form aggregates in solution26. Many of these sequences are also known to form GQ structures with high stability in the presence of monovalent ions53. We reasoned that the propensity of TERRA to form GQ structures50, 52 may contribute to homotypic RNA clustering via percolation. If true, this can be modulated by TERRA sequence perturbations. To test this idea, we employed a mutant (TERRA)10 sequence [(mut-TERRA)10] with a G-to-U substitution e.g., (UUAGUG)10, which is expected to disrupt the stability of the GQ state54, 55. We performed temperature-controlled microscopy experiments with (mut-TERRA)10 samples and observed that the LCST transition of (mut-TERRA)10 requires a substantially higher concentration of Mg2+ ions (Fig. 2b, c vs. 2e, f). At 25 mM Mg2+, (mut-TERRA)10 undergoes reversible phase separation upon heating with an LCPT of 28.8 ± 0.42°C (Fig. 2e; Supplementary Video 4). We note that under the same condition, wild-type (WT) (TERRA)10 would exist as percolated clusters at all experimentally accessible temperatures. Percolation of (mut-TERRA)10 was not observed at any conditions tested (Fig. 2f) demonstrating that the G-to-U substitution attenuates the intermolecular RNA-RNA interactions between TERRA chains. Consistent with the absence of RNA percolation in RNA-only condensates, we find that (mut-TERRA)10 does not form age-dependent RNA aggregates in RGG-d(T)40 condensates (Fig. 2g; Supplementary Fig. 4, 5). This data suggests that intra-condensate RNA percolation is RNA sequence- and structure-specific.
Does the intra-condensate aggregation of TERRA, as reported in Fig. 1d-i, proceed via a transition from intra-molecular to inter-molecular GQ structures? We tested this idea by utilizing a GQ-selective fluorescent probe Thioflavin T (ThT), which has been previously shown to display fluorescence activation upon binding to GQ structures with high specificity56, 57, 58. At an early time-point after the preparation of RGG-d(T)40 condensates containing (TERRA)10, ThT fluorescence was observed to be uniformly distributed throughout the condensate suggesting that TERRA is likely forming intramolecular GQs within the homogenous condensate (Fig. 2h). Importantly, RGG-d(T)40 condensates with (TERRA)10 clusters, formed upon aging, displayed ThT positivity, with fluorescence intensities of the clusters being ~ 4-fold higher than nascent homogeneous condensates. This observation indicates that the TERRA in these aggregates is also likely to form GQs (Fig. 2h). In the case of the (mut-TERRA)10, there was no detectable ThT signal under identical imaging conditions at all time points (Fig. 2h). This is consistent with the previous reports that this particular G-to-U substitution disrupts the GQ structure of TERRA54, 55. Negative control of RGG-d(T)40 and RGG-d(T)40 condensates containing a non-percolating RNA43, poly-uridylic acid (polyU), did not show any measurable ThT signal under identical experimental conditions (Supplementary Fig. 6). Finally, intra-condensate (TERRA)10 clusters were observed ubiquitously in the presence of a wide variety of monovalent salts (Supplementary Fig. 7) that differentially impact the stability of RNA GQ structures, suggesting observed RNA percolation may directly stem from the clustering propensity of purine-rich RNAs43.
Timescale of RNA clustering depends on the TERRA repeat number
The number of guanine tracts (G-tracts) plays important roles in monomeric RNA GQ and multimeric RNA structure formation59, 60, 61. We reasoned that time-dependent TERRA aggregation in RGG-d(T)40 condensates (Fig. 1d) can be tuned by the ability of TERRA to form multi-molecular GQ structures, which is expected to depend on the number of repeat units of TERRA59, 62. To test this idea, we titrated the number of TERRA repeat units, [UUAGGG]n, where n = 1, 4, 6, and 10; far below the number of repeat units transcribed in cells, 100 to 9000 nucleotides49, 63. We observed that all RNAs that could form intramolecular GQs, e.g., n = 4, 6, and 10, showed age-dependent intra-condensate cluster formation with a corresponding timescale correlated inversely with the number of repeats (Fig. 3a). (TERRA)10 forms microscale aggregates the fastest with a timescale of ~ 2 hours, whereas (TERRA)4 form clusters at substantially slower rate (~ 8 hours) and the clusters are less distinct (Fig. 3a). To quantify the relative intra-condensate cluster sizes as a function of time, we employed spatial autocorrelation analysis (SAC) of the confocal fluorescence images of condensates (Fig. 3a insets, white-dotted boxes; Supplementary Fig. 8). In the absence of clusters larger than the detection limit of the microscope, which is the case for (TERRA)6 at 15 minutes after sample preparation, SAC returns the size of spatial fluctuations, 0.28 ± 0.01 µm, which closely corresponds to the image resolution (1 pixel = 0.2196 µm). However, at 18 hours time point when the RNA clusters in the condensate are detectable, SAC reveals the characteristic size of the RNA clusters being 0.89 ± 0.04 µm (Fig. 3b). Employing SAC, we find that the size of RNA clusters increases, and the timescale of RNA cluster formation decreases with increasing number of repeat units of TERRA (Fig. 3c, d). Simultaneously, we see increased negative correlation between the intensities of TERRA and the primary components, RGG and d(T)40, as a function of condensate age (Fig. 3e, f; Supplementary Fig. 9, 10). As clusters become more pronounced with the aging of condensates, the anti-correlation of intensities becomes more apparent (Fig. 3e, f; Supplementary Fig. 9, 10). These data suggest that percolated TERRA clusters demix from the RGG and d(T)40 in a time-dependent manner. Together with the data presented in the previous section on (mut-TERRA)10 failing to form homotypic intra-condensate clusters, these data suggest that repeat number and RNA sequence variations are molecular regulators of age-dependent RNA percolation within biomolecular condensates.
Repeat expanded RNAs form intra-condensate clusters in a length-dependent manner
Intracellular RNA aggregation has been widely implicated in several diseases primarily in the category of trinucleotide repeat expansion disorders such as Huntington’s disease (CAG repeat), Fragile-X syndrome (CGG repeat), and myotonic dystrophy (CUG repeat)37, 64, 65, 66, 67, 68, 69. One characteristic of these disorders is that pathology occurs at a number of repeats beyond what is found in healthy individuals. Further, the threshold number of repeats associated with pathological outcomes varies with the repeat RNA sequence. For example, the number of r(CAG) repeats required for disease onset is ~ 15 to 20 repeats less than that required for r(CUG) (Fig. 4a). This implies that the cellular machinery can better tolerate r(CUG) repeats than it can with r(CAG) repeats64, 65, 66, 67, 68.
Motivated by our results of enhanced TERRA clustering as a function of increasing repeat number, we interrogated the clustering propensity of two repeat RNAs, r(CAG)n and r(CUG)n, with repeat lengths higher and lower than the pathological limit using our model condensate system. r(CAG) repeat lengths lower than 21 are found in healthy individuals, however, repeat lengths exceeding this number are considered to be at the intermediate-high risk scale of being most likely to attain diseases such as spinocerebellar ataxia and Huntington’s disease (Fig. 4a). We observed that r(CAG)20 stays predominantly homogeneous in the dense phase of RGG-d(T)40 condensates at all ages till 24 hours, the end time of our experiments. The mean RNA cluster size is estimated to be 0.28 ± 0.01 µm, which closely corresponds to the image resolution (1 pixel = 0.2196 µm) (Fig. 4b, c; Supplementary Fig. 11). Strikingly, r(CAG)31, which corresponds to a pathological number of repeats, spontaneously demixes into fractal-like clusters prior to sample imaging (15 minutes post-preparation) (Fig. 4b, c; Supplementary Fig. 11). In these samples, r(CAG)31 clusters are extensively percolated with cluster sizes that are nearly 4-fold larger (3.94 ± 0.14 µm) than those formed by TERRA at the same concentration (Fig. 4c). Reduced r(CAG)31 concentration reduces the intra-condensate cluster size, but the RNA aggregation timescale appears to be independent of r(CAG)31 concentration (Supplementary Fig. 12).
We next examined intra-condensate percolation of r(CUG)n for n = 31, which represents a healthy number of repeats. Similar to r(CAG)20, we did not observe intra-condensate r(CUG)31 clustering. The estimated cluster size is 0.40 ± 0.02 µm, which closely corresponds to the image resolution (1 pixel = 0.2196 µm) (Fig. 4d, e; Supplementary Fig. 11). In stark contrast, r(CUG)47 forms microscale clusters within 15 minutes after sample preparation (Fig. 4d, e; Supplementary Fig. 11). We also tested intra-condensate clustering behavior of a hexanucleotide repeat expanded RNA, r(GGGGCC)n, that is associated with ALS and frontotemporal dementia (FTD)59, 70, 71. We observed spontaneous r(GGGGCC)5 clustering in RGG-d(T)40 within 15 minutes after sample preparation (Supplementary Fig. 13). Importantly, r(GGGGCC)5 clusters showed ThT positivity but not r(CAG)20 or r(CAG)31 clusters (Supplementary Fig. 14). This is consistent with the ability of r(GGGGCC)5 to form GQ structures59. Overall, the distinct differences in the intra-condensate percolation behavior of each of these RNAs and TERRA of various lengths suggest a heuristic framework where a sequence and length-specific activation energy barrier dictates the timescale of intra-condensate RNA clustering (Fig. 4f, left). In this picture, r(CAG)20, r(CUG)31, and (mut-TERRA)10 molecules have a substantially higher energy barrier to undergo intra-condensate percolation as compared to (CAG)31, r(CUG)47, and r(GGGGCC)5, whereas (TERRA)10 features an intermediate activation energy barrier (Fig. 4f, right).
Intra-condensate RNA clustering drives a liquid-to-solid phase transition
What are the material properties of condensates containing percolated RNA clusters? Time-lapse microscopy suggests that the shell phase of the aged condensates containing TERRA clusters can undergo fusion, signifying liquid-like properties (Fig. 1d, Supplementary Video 1). However, these condensates are in contact with the glass surface, which can influence their fusion kinetics substantially72. To examine the dynamical behavior of (TERRA)10 containing RGG-d(T)40 condensates quantitatively as a function of age, we employed controlled condensate fusion using optical tweezers73, 74. We trapped two condensates using a dual-trap optical tweezer and initiated condensate fusion while recording the fluorescence of FAM-(TERRA)4 and Cy5-d(T)40 simultaneously. At 20 minutes after the condensate preparation, when both TERRA and d(T)40 are uniformly distributed throughout the droplets, they undergo fusion with a fusion relaxation time (τ) of 18.40 ± 0.02 ms/µm. The components were homogeneously mixed after the condensate coalescence was completed (Fig. 5a; Supplementary Video 5). However, for condensates aged for 150 minutes, distinct RNA clusters were visible within condensates (Fig. 5a; Supplementary Video 6). These aged condensates containing RNA clusters were still able to fuse indicating that the condensate shell is dynamic and behaves as a terminally viscous liquid. Force relaxation analysis showed that fusion of the outer shell of aged condensates occurs more rapidly with a fusion relaxation time of 3.87 ± 0.06 ms/µm (Fig. 5a). However, the irregular RNA clusters within the condensate remained demixed at all conditions, indicating that they are in a different material state than RGG-d(T)40 rich phase (Fig. 5a). We tested this directly by designing a condensate dissolution assay, which would preferentially dissolve the non-percolated components (Supplementary Fig. 15a). We observed that addition of 0.5 µl of 5 M NaCl is sufficient to rapidly dissolve condensates formed by RGG-d(T)40 (Supplementary Fig. 15b). However, NaCl addition to aged RGG-d(T)40 condensates containing TERRA clusters showed only partial dissolution where the RGG-d(T)40 rich shell phase dissolved immediately but the RNA clusters persisted (Fig. 5b; Supplementary Videos 7, 8). Notably, freshly prepared (TERRA)10 containing RGG-d(T)40 condensates lacking RNA clusters (15 minutes post preparation) were observed to dissolve completely by NaCl treatment (Supplementary Fig. 15c; Supplementary Video 9). We also probed the translational mobility of each component in the condensate as a function of age by fluorescence recovery after photobleaching (FRAP; Fig. 5c). We chose (TERRA)4 as a representative RNA for these measurements which forms intra-condensate clusters by 8 hours (Fig. 3a, c). FRAP analysis reveals that at early time points before the onset of visible RNA clusters (< 4 hours), all three components are relatively dynamic (Fig. 5c; Supplementary Fig. 16; Supplementary Videos 10, 11, 12). However, FRAP traces of TERRA, but not RGG and d(T)40, show a progressive drop in recovery with condensate age (Fig. 5c; Supplementary Fig. 16; Supplementary Videos 13, 14, 15).
The reduced translational mobility of TERRA could result from the formation of dynamically arrested intra-condensate RNA networks, driving a liquid-to-solid phase transition. To directly probe the material properties of condensates, we employed video particle tracking (VPT) nanorheology using 200 nm fluorescently labeled beads8 (Fig. 5d). The mean squared displacement (MSD) profiles of the probe particles have distinct characteristics for Newtonian liquids, viscoelastic fluids with terminal viscous behavior, and Kelvin-Voigt solids with terminal elastic behavior12, 75 (Fig. 5d, right). VPT measurements reveal that freshly prepared RGG-d(T)40 condensates containing (TERRA)10 display material properties similar to a Maxwell fluid9 with a terminal viscosity of 35.2 ± 4.85 Pa.s (Fig. 5e-h; Supplementary Video 16). Upon physical aging, the probe particles within condensates are caged as evidenced by a narrower spread of the particle trajectory (Fig. 5f). Correspondingly, the ensemble-averaged MSD profiles showed dramatic differences at times beyond the emergence of RNA clusters, e.g., t > 60 minutes (Fig. 5g). We find that this is an emergent property of the (TERRA)10 containing RGG-d(T)40 condensate system since the two-component condensate system composed of RGG and d(T)40 does not show time-dependent changes in MSD profiles (Supplementary Fig. 17). Further measurements of MSD profiles in RGG-d(T)40 condensates containing (TERRA)10 revealed a dramatic arrest of the probe particles at time points, t = 30 and 45 minutes, at which the microscale RNA clusters were not clearly visible (Fig. 5g; Supplementary Fig. 18; Supplementary Video 17). The dynamical slowdown of probe particles prior to the emergence of microscale RNA aggregates may indicate the formation of nanoscale pre-percolation clusters of TERRA76. Upon further aging, the ensemble-averaged MSDs show a plateauing behavior at longer lag times indicating a terminally elastic response reminiscent of Kelvin-Voigt solids12 (Fig. 5g; observation time ≥ 150 minutes). This may stem from the caging of beads due to the onset of percolated RNA clusters (Fig. 5f). The estimated terminal viscosity of the condensates at 90 minutes is 489.3 ± 69.3 Pa.s (Fig. 5h), which is an order of magnitude higher than condensates at 15 minutes. At 150 minutes after sample preparation, the bead motions were completely arrested, which is a characteristic property of terminally solid material12 (Supplementary Video 18).
We further performed nano-rheology measurements with (mut-TERRA)10 (Fig. 2e, f), which lacks the ability to undergo intra-condensate percolation (Fig. 2g; Supplementary Fig. 4, 5). We observed that (observation time ~ 30 minutes) RGG-d(T)40 condensates containing (mut-TERRA)10 behave as a Maxwell fluid with viscosity 21.5 ± 3.1 Pa.s, which is slightly lower than WT (TERRA)10 (Fig. 5i, j; Supplementary Fig. 19). Importantly, no discernable changes in viscoelastic properties was observed for these condensates over the same period of time (4 hours; Fig. 5i, j; Supplementary Fig. 19) where the WT TERRA containing condensates undergo complete dynamical arrest. These results show that selectively inhibiting the percolation ability of the RNA through sequence perturbations abrogates RNA percolation-driven condensate dynamical arrest.
RNA-binding protein G3BP1 inhibits intra-condensate RNA clustering
In cells, RNP granules are comprised of diverse RNA and protein species with an assortment of sequence-specific and non-specific interactions77, 78, 79, 80. Based on our results on age-dependent intra-condensate RNA clustering, we reasoned that introducing additional components in these condensates that can compete with homotypic RNA-RNA interactions may inhibit RNA cluster formation. We tested this idea first by employing an anti-sense oligonucleotide [ASO; sequence: r(CCCUAA)] targeting (TERRA)10. ASO-treated condensates did not show any signs of RNA clusters even after 24 hours in our fluorescence microscopy experiments (Fig. 6a, b; Supplementary Fig. 20). VPT-based nanorheology confirmed the absence of nanoscale clusters (Fig. 6c; Supplementary Fig. 21) as no substantial change in condensate viscosity was observed during 24 hours of aging (Fig. 5f). These results suggest that selectively targeting homotypic RNA-RNA interactions can buffer against RNA clustering-mediated liquid-to-solid phase transition. Could such buffering effects be also imparted by multivalent RBPs with broad specificity for a diverse repertoire of RNA? We tested this idea utilizing G3BP1, which is a core scaffolding RBP of cytoplasmic stress granules81, 82, 83. G3BP1 binds to RNAs through a folded RNA recognition motif and disordered Arg/Gly-rich domain (Fig. 6d). In the presence of 10 µM G3BP1, we observed that (TERRA)10 containing RGG-d(T)40 condensates do not show the emergence of RNA clusters after aging (within 24 hours). All components remain homogenous and the cluster sizes derived from SAC analysis show no substantial change in their spatial distribution (0.14 ± 0.01 µm at 15 minutes and 0.14 ± 0.01 µm at 24 hours; 1 pixel = 0.09765 µm) (Fig. 6e, f; Supplementary Fig. 23).
Earlier, our nano-rheology experiments on (TERRA)10 containing RGG-d(T)40 condensates (Fig. 5g, h) revealed signs of dynamical arrest and a substantial increase in viscosity even before microscale RNA clusters were visible. To quantify the effects of G3BP1 on the rheology of these condensates, we probed the time-dependent changes in condensate material properties using VPT. MSD profiles and viscosity measurements reveal that despite the absence of microscale RNA clusters, G3BP1 containing condensates undergo a progressive dynamical slowdown (Fig. 6g; Supplementary Fig. 24). There is a concomitant increase in viscosity from 42.3 ± 9.0 Pa.s for condensates at 30 minutes to 154 ± 22.3 Pa.s within 3 hours of condensate preparation (Supplementary Fig. 24). However, condensate aging is substantially slower in the presence of G3BP1 (Fig. 6h). Therefore, although microscale RNA clusters
are disfavored in the presence of G3BP1, our rheology measurements suggest that there may be nanoscale RNA clustering that contributes to increased viscoelasticity of aged condensates. Nonetheless, G3BP1 can increase the activation energy barrier of intra-condensate (TERRA)10 clustering.
We further tested the generalizability of the observed buffering effect of G3BP1 with GC-rich repeat RNAs, r(CAG)31 and r(CUG)47, that spontaneously form microscale clusters in RGG-d(T)40 condensates (Fig. 4). In the presence of G3BP1, both of these RNAs formed homogeneous condensates (Fig. 6i-l; Supplementary Fig. 23). Furthermore, in the case of (CAG)31, no microscale clusters were observed even at 24 hours age where the cluster size = 0.39 ± 0.02 µm, close to the detection limit (1 pixel = 0.2196 µm) (Fig. 6i, j; Supplementary Fig. 23). In the case of r(CUG)47 we observed the formation of a few RNA foci at 24 hours (Fig. 6k, l; Supplementary Fig. 23). However, the total fraction of RNA, determined by intensity-based analysis, in these clusters were only ~ 6.65 ± 2.31% (Supplementary Fig. 25), whereas the same RNA was present almost exclusively (~ 100%) in the intra-condensate clusters in the absence of G3BP1 (Fig. 4d). Together, these results signify that G3BP1 can frustrate84 RNA-RNA homotypic interactions in these condensates thereby disfavoring RNA clustering and preserving intra-condensate solubility of repeats RNAs.