IntroductionRNA granules are a type of biomolecular condensates that are enriched with proteins and RNAs and inextricably linked with post-transcriptional gene regulation1,2. RNA granules have been observed across species3,4,5,6,7,8, and often appear structured, with their proteins9,10,11,12,13,14,15 and RNAs3,16,17,18,19,20 condensing into clusters, which are composed of multiple macromolecules that assemble through intermolecular interactions, in vivo and in vitro.Protein conformation and protein-protein interactions within protein clusters have been extensively studied21,22,23,24. Depending on the species, some proteins, like Poly(A)-binding protein (Pab1) and globular proteins, partially or completely unfold during condensation21,24. Others, such as Fused in Sarcoma and androgen receptor activation domain, transition to a more compact and folded state22,25. Regardless of their conformations, both in vitro and in vivo, these clusters require multivalent interactions driven by charge-charge, cation-π, π-π, dipole-dipole and hydrophobic interactions, facilitated by diverse amino acid sequence compositions23,26,27.However, our understanding of RNA-RNA interactions and RNA conformation within RNA clusters remains limited and is mainly derived from in vitro studies. In these studies, in vitro transcribed RNAs self-assemble into visible clusters upon the addition of salts and crowding reagents. Notably, clustering occurs in the absence of other cellular components16,19,20,28,29, demonstrating the potential of RNAs to interact with each other. Like clustering of proteins, RNA clustering requires interaction multivalency30 and may involve sequence complementarity. For example, in the filamentous fungus Ashbya gossypii, the co-clustering of BNI1 and SPA2 mRNAs is facilitated by base pairing of exposed complementary sequences (“zipcodes”) shared between the two transcripts located in five distinct RNA regions28.To facilitate RNA clustering, intermolecular base pairing may require melting of the RNA structure. Notably, repeat RNAs that cause human neurodegenerative disorders31,32,33 contain an array of exposed and repeated GC-rich sequences that induce RNA condensation in vitro and in vivo16. Computer simulations of repeat RNAs revealed that in the process of condensation, the hairpin structures formed by GC-rich repeats transition into an unfolded state34. This unwinding facilitates intermolecular base pairing, increases the multivalency of interactions and results in the formation of an extended interaction network. Similarly, guanidine riboswitches29 melt their secondary structures to augment intermolecular base pairing in silico or in vitro. Finally, melting of the secondary structures of antisense non-coding RNAs (ncRNAs) is thought to enable intermolecular base pairing with their sense mRNA partners, thereby facilitating enrichment of ncRNAs in stress granules20. These studies illustrate that multivalent intermolecular base pairing could be required for the formation of RNA clusters.In vivo, mRNAs critical for Drosophila development enrich in germ granules located at the posterior of the developing embryo, where they are post-transcriptionally regulated35. Within these granules, the mRNAs form clusters that contain multiple transcripts derived from the same gene. In contrast, outside of the granules, these mRNAs are mostly single transcripts17,36. Notably, different RNA clusters do not mix with each other, and their interactions are instead homotypic in nature17,18,37. The mRNA concentration within these clusters is remarkably high (5–15 µM)36 suggesting dense packing of mRNAs, and potential involvement of intermolecular base pairing in their formation.Our previous study using chimeric experiments and stochastic super-resolution microscopy (STORM) on germ granule mRNAs failed to detect dependence of RNA clustering on a particular RNA zipcode36. These findings differ from observations of BNI1 and SPA2 mRNAs, as well as the Drosophila oskar (osk), bicoid (bcd) mRNAs and human immunodeficiency virus (HIV) genomes, whose clustering relies on specific, often GC-rich complementary sequences (CSs) that are present as exposed loops on top of stems36,38,39,40,41,42,43,44,45. This structural arrangement facilitates recognition among mRNAs, leading to stable intermolecular base pairing and RNA clustering36,38,39,40,41,42,43,44,45. These results highlight the distinctive nature of intermolecular base pairing in Drosophila germ granules, setting it apart from the mechanisms observed in BNI1, SPA2, osk, bcd mRNAs and HIV genomes.In this study, we characterize the type and prevalence of intermolecular interactions that occur within RNA clusters in Drosophila germ granules. While these interactions undoubtedly involve protein-protein, protein-RNA and RNA-RNA interactions, here we focus specifically on intermolecular base pairing. Importantly, the mechanisms resulting in the specificity for homotypic RNA clustering are unclear. However, we speculate that base pairing interactions that help build RNA clusters may also aid in specifying their composition.Our work revealed that in vitro, within a protein-free environment, intermolecular base paring in RNA clusters does not require sequence complementarity of 6 or more consecutive nucleotides, which we refer to as extended complementarity. Furthermore, the foldedness of mRNAs within and outside of germ granules is similar. Finally, our in silico data predicted that the tendency of RNAs to engage in intermolecular interactions is independent of high sequence complementarity or significant unwinding of secondary structure. This crucial result could explain the lack of dependence of RNA clustering on a particular RNA sequence in germ granules we reported previously36. Finally, engineered germ granule mRNAs, with exposed GC-rich CSs, presented within stem loop structures, induce persistent base pairing in vitro and enhanced intermolecular interactions in vivo but also disrupt normal fly development. Notably, while flies expressing nos with stem loop structures exhibited major developmental defects, those with exposed GC-rich CSs within the stem loops displayed exacerbated phenotypes. Our study, from multiple perspectives, reveals that germ granule mRNAs employ RNA folding as a mechanism to fend off potential detrimental effects of stem loops and exposed GC-rich CSs, while facilitating multivalent intermolecular interactions and RNA clustering. Therefore, RNA folding may not only mediate the nature and prevalence of intermolecular interactions within RNA clusters but also ensure optimal gene functionality within the germ granules.ResultsStable dimerization of shu-top and shu-bottom RNAs relies on extensive sequence complementarity in vitroTo begin characterizing the type and prevalence of intermolecular base pairing in RNA clusters, we first examined the type of base pairing that is required for stable RNA dimerization in vitro. To this end, we used the split broccoli system coupled with the chemical reporter DFHBI-1T46. This system involves top and bottom RNAs, each carrying half of the broccoli aptamer. When these two RNA segments dimerize, they form a stable broccoli structure detectable by fluorescence emitted by DFHBI-1T intercalated between base-paired top and bottom RNAs (Fig. 1a)46. We reasoned that the observation of a strong fluorescent signal generated by the broccoli system will report on intermolecular base pairing driven by extensive sequence complementarity during dimerization.Fig. 1: RNA clustering can occur without extensive sequence complementarity or significant RNA structural melting in vitro.a Schematic and predicted structures of shu-top and shu-bottom RNAs. shu-top and shu-bottom RNAs contain split-broccoli sequences top (blue) and bottom (orange), and their intermolecular base pairing is recognized by DFHBI-1T (green or gray star) (see Supplementary Data S1 for sequences). The predicted structures were drawn by FORNA48. b In vitro dimerization of shu-top and shu-bottom RNAs. Top panel: RNA; bottom panel: DFHBI-1T signal. The ladder was run on a different gel. c Schematics of RNA clustering of shu-top and shu-bottom RNAs that is independent (i) or dependent (ii) of base pairing generated by extensive sequence complementarity or significant structural melting. The two models are distinguished by the absence (gray circle (i)) or presence (green circle (ii)) of DFHBI-1T fluorescence. d, e In vitro RNA clusters (magenta) formed of only shu-top or shu-bottom RNAs (negative controls), co-folded (positive control) or separately folded shu-top or shu-bottom RNAs (e) in the presence of DFHBI-1T (green). The DFHBI-1T images in (d, e) were normalized using the parameters of the co-folded DFHBI-1T image (see also Supplementary Fig. S1d for images normalized to the DFHBI-1T signal of shu-top). Initial RNA concentrations: 3.2 μM (negative controls), 1.6 μM for each RNA in co-folding and separate folding. Scale bars: 7.5 μm. f Normalized intensity of DFHBI-1T by RNA intensities (Supplementary Fig. S1a–c) from outside (gray) or inside (green) in RNA clusters per condition. n = 30 RNA clusters for each condition. Data: Mean ± SEM. n.s.: not significant. p-values (unpaired t test; two-tailed): 0.30 (shu-top), 0.47 (shu-bottom), 0.30 (co-folding) and 0.089 (separate folding). See also Supplementary Fig. S1 and Supplementary Data S1. Source data are provided as a Source Data file.Full size imageTo place the broccoli aptamer in the sequence context of a germ granule mRNA, we inserted top and bottom RNAs into the 3′UTR of the Drosophila shutdown (shu) RNA, creating shu-top and shu-bottom (Fig. 1a and Supplementary Data S1). While shu accumulates in the germ granules, it does not form clusters within them36. In addition, shu-top and shu-bottom did not form homodimers in vitro (Fig. 1b), allowing us to study the formation of the broccoli aptamer without inference from the shu 3′UTR sequence.RNAcofold47 predicted the correct broccoli structure when shu-top and shu-bottom were co-folded, while RNAfold47 predicted that top and bottom segments base paired intramolecularly when they were folded separately (Fig. 1a, the predicted structures were drawn using FORNA48). To validate the RNAcofold and RNAfold predictions that shu-top and shu-bottom form a correct broccoli structure detectable by DFHBI-1T, we used non-denaturing RNA gel electrophoresis following established protocols39,41. To this end, we heat-denatured and then co-folded shu-top and shu-bottom RNAs and observed a dimerized broccoli RNA accompanied by a strong DFHBI-1T fluorescence on the gel (Fig. 1b). In contrast, when we mixed shu-top and shu-bottom RNAs that were folded separately, the two RNAs did not dimerize and showed no DFHBI-1T fluorescence signal. Notably, the absence of DFHBI-1T fluorescence was similar to the one recorded for separately folded shu-top and shu-bottom, which were loaded individually on the gel as negative controls (Fig. 1b). Lack of a visible dimer band in samples containing only shu-top and shu-bottom RNAs, or when the two RNAs were folded separately and afterwards mixed, revealed that even though shu-top and shu-bottom RNAs shared extensive sequence complementarity, folded shu-top and shu-bottom RNAs lacked the potential for stable dimerization even at high RNA concentrations (3.2 μM) (Fig. 1b).RNA clustering of shu-top and shu-bottom can occur without extensive sequence complementarity in vitroWe then examined if clustering of shu-top and shu-bottom relies on base pairing among complementary sequences, accompanied by unfolding of the RNA structure as observed for repeat RNAs16,34. To this end, we fluorescently labeled shu-top and shu-bottom RNAs and adapted existing in vitro RNA clustering protocols20,28. We hypothesized that if RNA clustering of shu-top and shu-bottom does not require intermolecular base pairing generated by extensive sequence complementarity, then the levels of DFHBI-1T fluorescence would be similar inside and outside the shu-top and shu-bottom clusters. This result would indicate that the RNA clusters do not promote the formation of additional broccoli structures (Fig. 1ci). However, if clustering of shu-top and shu-bottom RNAs relied on the intermolecular base pairing driven by extensive sequence complementarity, then we expect to observe a much higher DFHBI-1T fluorescence inside the clusters compared to outside (Fig. 1cii).We first estimated the DFHBI-1T to shu-top/shu-bottom RNA dimer ratio to be approximately 30:1 in our reactions. This ratio ensured a large excess of the fluorescent reporter compared to the dimerized RNA and saturated detection of broccoli structures49 (“Methods”). Therefore, the intensity of the DFHBI-1T fluorescence should be scaled with the prevalence of broccoli structure.To establish the baseline for the DFHBI-1T fluorescence inside and outside of RNA clusters, we first measured DFHBI-1T fluorescence in RNA clusters formed only by shu-top or shu-bottom, our negative controls (Fig. 1d, green signal). Importantly, the raw DFHBI-1T fluorescence outside (Supplementary Fig. S1a) and inside (Supplementary Fig. S1b) of RNA clusters for both RNAs was minimal compared to the positive control (see below) (Fig. 1d and Supplementary Fig. S1d). After normalizing the fluorescence to their respective RNA intensity inside and outside the clusters (Fig. 1d, magenta signal, Supplementary Fig. S1c; “Methods”), we did not observe significant difference in DFHBI-1T fluorescence between inside and outside the clusters for either negative control (Fig. 1f). This suggests that the increased DFHBI-1T intensity in RNA clusters formed by the two negative controls was primarily due to the higher RNA concentration within clusters, leading to an increased background signal.As a positive control, we induced clustering of shu-top and shu-bottom RNAs that were first co-folded and generated stable RNA dimers (Fig. 1b). As anticipated, we observed a 30- to 89-fold increase in DFHBI-1T fluorescence outside the clusters, and a 64- to 80-fold increase inside the clusters, compared to shu-top or shu-bottom, respectively (Supplementary Fig. S1a, b). However, after normalization to the RNA intensity (Fig. 1d and Supplementary Fig. S1c), we observed no significant difference in DFHBI-1T levels between inside and outside of clusters (Fig. 1f). Therefore, the increased RNA concentration within the observable clusters generated by co-folded shu-top and shu-bottom RNAs did not stimulate the formation of additional broccoli dimers.Finally, we examined DFHBI-1T levels in clusters formed by mixed but separately folded shu-top and shu-bottom RNAs. After normalization to the RNA intensity (Fig. 1d and Supplementary Fig. S1c), we observed no significant difference between the DFHBI-1T levels inside and outside the clusters (Fig. 1f), which also remained similar to the ones recorded for the negative controls. Thus, these results revealed that no additional broccoli structure formed between shu-top and shu-bottom when the two separately folded RNAs clustered. Moreover, we observed no significant differences in the size (Supplementary Fig. S1e) or morphology of RNA clusters (Fig. 1d, e) generated by co-folded and separately folded RNAs. However, based on the increase in RNA intensity, co-folded shu-top and shu-bottom RNAs formed denser RNA clusters (Supplementary Fig. S1c). Collectively, our data suggest that in vitro and in a protein-free environment, the clustering of shu-top and shu-bottom RNAs does not require intermolecular base pairing driven by extensive sequence complementarity nor significant RNA unfolding to expose sequences to accommodate such interactions.The foldedness of 3′ untranslated regions of germ granule mRNAs is similar inside and outside of germ granulesGiven our result with shu-top and shu-bottom RNAs, we reasoned that the absence of stable intermolecular interactions between mRNAs in RNA clusters in germ granules we recorded previously36, may also be due to RNA folding. Such folding may be independent of specific structures formed upon mRNA localization to germ granules. To test this hypothesis, we examined the base accessibility to interactions of germ granule mRNAs using dimethyl sulfate mutational profiling with sequencing (DMS-MaPseq) and probed the folded state of an RNA inside and outside of germ granules. DMS preferentially modifies accessible and unpaired adenosines and cytosines, which creates a mutational profile during reverse transcription50,51.Exposure to DMS fragments RNA, making it susceptible to loss during subsequent germ granule purification. In addition, the vitelline membrane surrounding the embryos prevents the delivery of DMS. To circumvent these issues, we probed RNA with DMS using isolated germ granules. We purified germ granules marked by a fluorescently tagged core germ granule protein Vasa (Vasa:GFP) from embryo lysates, as done previously52,53. This approach yielded a pellet with germ granule-bound mRNAs as well as a soluble fraction that contained mRNAs outside the germ granules (Supplementary Fig. S1fi, ii). Importantly, germ granules resisted treatment with 10% DMS (Supplementary Fig. S1fii). Moreover, germ granule mRNAs nanos (nos), polar granule component (pgc), and germ-cell-less (gcl) exhibited 114, 117, and 88-fold enrichment in the pellet relative to the soluble fraction, respectively (Supplementary Fig. S1gi–iii), indicating that we were probing the structure of germ granule-bound mRNAs rather than the structure of mRNAs outside the granules.We specifically focused on the 3′ untranslated regions (UTRs) of nos, pgc and gcl because these regions drive the localization of nos, pgc and gcl to germ plasm54,55. Furthermore, since germ granule mRNAs translate while associated with germ granules54,56,57,58,59,60, we anticipated that the RNA folding and the intermolecular interactions would be minimally perturbed by ribosomes within the 3′UTRs.Focusing first on the 3′UTR of unlocalized nos, we recapitulated the previously characterized structure of its translational control element (TCE) (Fig. 2a)61,62,63. This result, together with the fact that DMS efficiently recapitulates the structure of the yeast 18S rRNA to 94% accuracy in vivo51, demonstrated the effectiveness of the DMS-MaPseq in probing the base accessibility of germ granule mRNAs in a proteinaceous environment in vivo.Fig. 2: The foldedness of nos 3′UTR is similar inside and outside of germ granules.a Predicted secondary structure of nos TCE outside the germ granules with a mapped normalized DMS reactivity generated by DMS-MaPseq. The normalized DMS reactivities were based on an average of two biological replicates (see Supplementary Fig. S1h, i for the correlation coefficients between the two replicates). b An example of the predicted secondary structure of nos 3′UTR outside the germ granules with mapped reactivity by DMS-MaPseq (an average of two replicates is shown). Blue dashed box: nos TCE. c Correlation of the normalized DMS reactivities between inside (y-axis) and outside of germ granules (x-axis). Each purple dot represents the DMS reactivity of the same probed base on nos 3′UTR recorded inside (y-axis) and outside of germ granules (x-axis). d The ratio of the normalized DMS reactivity between inside and outside of the germ granules for the same probed bases on nos 3′UTR. Black dotted line: no change in DMS reactivity. Blue dots and lines: ratio of DMS reactivities between inside and outside of germ granules for the same probed bases. Orange dashed lines represent thresholds with a p-value