In vivo clustering of oskar mRNA is driven by RNA concentration, RNA binding proteins, and an RNA palindrome Authors: Ziqing Ye, Siran Tian, Ayse Ecer, Tatjana Trcek, Anita Corbett Published: 2026 ABSTRACT AbstractmRNA organization into clusters is observed in many cellular contexts, yet the features that govern this process in vivo remain poorly understood. Using super-resolution microscopy, single-mRNA imaging, and genetic perturbations, we investigated how mRNA concentration, the double-stranded RNA-binding protein Staufen, and intermolecular base-pairing driven by an RNA palindrome influence clustering of oskar mRNA in Drosophila embryos. We find that these factors collectively optimize oskar clustering by promoting its dimerization and subsequent oligomerization. Both processes depend on all three factors, although oligomerization is much more sensitive to their perturbation, indicating that the driving force for oskar oligomerization is partially distinct from that governing dimerization. Moreover, oskar palindrome is a potent driver of heterotypic mRNA clustering, further supporting its in vivo role in mediating intermolecular base pairing. Finally, computational analyses identified a subset of candidate mRNAs in the early embryo that are predicted to harbor oskar-like palindromes. Among these, eIF3a mRNA emerged as a potential candidate whose clustering may likewise be driven by intermolecular base pairing. These preliminary observations raise the possibility that mRNA clustering driven by palindrome-mediated intermolecular base pairing may be more widespread than previously appreciated and may represent an important mechanism for controlling mRNA spatial organization during Drosophila development. INTRODUCTION Spatial organization of messenger RNAs (mRNAs) is a widespread mechanism for controlling the spatial and temporal production of proteins within cells and is particularly impactful during organismal development, where localized mRNAs influence cell fate, morphology, and body plan formation across metazoans (Holt and Bullock, 2009). Beyond being targeted to defined cellular structures such as the endoplasmic reticulum or RNA granules, mRNAs can also organize with one another to form mRNA clusters (Little et al., 2015; Trcek et al., 2015; Aw et al., 2016; Lu et al., 2016; Pichon et al., 2016; Sharma et al., 2016; Cai et al., 2020; Chouaib et al., 2020; Gabryelska et al., 2022; Cardona et al., 2023; Klobucar et al., 2026). These clusters, which may contain only a few transcripts, arise through intermolecular interactions involving both RNA-binding proteins (RBPs) as well as the RNAs themselves (Tian et al., 2020). In this context, RNAs can serve as scaffolds that recruit RBPs to drive mRNA clustering, or they may directly promote clustering via RNA:RNA interactions (Wagner et al., 2001; Jambor et al., 2011; Jain and Vale, 2017; Van Treeck and Parker, 2018; Van Treeck et al., 2018; Elguindy and Mendell, 2021). Intermolecular base pairing between complementary sequences (CSs) represents a prominent mechanism of RNA-driven clustering (Eisenberg and Felsenfeld, 1967; Clever et al., 1996; Ferrandon et al., 1997; Jambor et al., 2011; Jain and Vale, 2017; Langdon et al., 2018; Van Treeck et al., 2018; Tauber et al., 2020; Tian et al., 2025). This mechanism is well characterized in RNA viruses where genome dimerization and higher-order assembly depend on defined CS elements. For example, viruses like Influenza A, Human immunodeficiency virus-1 (HIV-1), Murine leukemia virus (MLV), and the Flock House virus (FHV) rely on intermolecular base pairing to organize and package their RNA genomes (Berkowitz et al., 1996; De Tapia et al., 1998; Krishna and Schneemann, 1999; Kim and Tinoco, 2000; Paillart et al., 2004; Johnson and Telesnitsky, 2010; Dadonaite et al., 2019; Le Sage et al., 2020; Zhou and Routh, 2024), and disruption of these interactions blocks virion assembly (Berkowitz et al., 1996; De Tapia et al., 1998; Paillart et al., 2004; Johnson and Telesnitsky, 2010; Zhou and Routh, 2024). Dimerization-competent CSs are typically structurally exposed, which facilitates recognition among mRNAs and positions them for stable and persistent interactions (Clever et al., 1996; Paillart et al., 2004; Hussein et al., 2010; Johnson and Telesnitsky, 2010; Bou-Nader and Zhang, 2020). They are also often GC-rich, which thermodynamically enhances the stability of base pairing. For example, in HIV-1, a six-nucleotide GC-rich palindrome located at the apex of a stem-loop nucleates RNA dimerization (Marquet et al., 1994; Clever et al., 1996). Intermolecular base pairing also operates in eukaryotic mRNAs. In the filamentous fungus Ashbya gossypii, base pairing between BNI1 and SPA2 mRNAs, both essential for cell cycle progression and nuclear division, drives their organization within cells (Langdon et al., 2018). This organization is important for establishing compositional specificity of biomolecular condensates associated with polarized growth and nuclear divisions in the fungus (Langdon et al., 2018). Similarly, in Drosophila, complementary sequence interactions between the oskar (osk; Figure 1A, i and ii) and bicoid (bcd), two mRNAs that are critical for early embryogenesis, promote their respective enrichment at distinct embryonic locations (Ferrandon et al., 1997; Jambor et al., 2011; Jambor et al., 2014). Applying smFISH to examine osk mRNA clustering in early embryos. A, i and ii Schematic of an embryo depicting the location of localized osk mRNA (i) accumulated at the posterior pole stained with smFISH (ii). B Schematic illustrating dimerization of osk mediated by an RNA palindrome. The palindrome (CCGCGG) is exposed atop a long RNA stem (magenta), which facilitates intermolecular base pairing of osk mRNAs and enables its hitchhiking (Jambor et al., 2011; Jambor et al., 2014). C, i and ii. A diagram showing the position of five dsRBDs (green boxes) of the Drosophila Stau (Micklem et al., 2000) (i) and the position of Stau binding to the osk stem-loop structure (green boxes). The nucleotides of the osk palindrome are labeled in magenta(Jambor et al., 2014) (ii). The stem-loop structure comprised nucleotides 714–827, which is the conserved region within the dimerization-competent portion of the osk 3′UTR defined previously (Jambor et al., 2014). D Schematic of an embryo depicting an approximate location of a 3D region of interest (ROI) used to analyze clustering of unlocalized osk mRNA in Ei–iii. Ei–iii. Maximal projection of five consecutive optical sections of an ROI taken from the somatic region of the embryo (Ai), spaced 150 nm apart (i). Yellow ROI: zoomed-in image shown in (ii) with an accompanying heat map image showing the distribution of osk mRNA clusters of different sizes (iii). F, i–iii Maximal projection of five consecutive optical sections spaced 150 nm apart taken from an embryo stained with smFISH probes hybridizing to ccr4 mRNA (i). Yellow ROI: zoomed-in image (ii) with an accompanying heat map image (iii). G Number of fluorescent ccr4 or osk clusters per µm3 volume (density). Mean ± STDEV of five and seven 3D ROIs is plotted. Statistical significance: two-tailed t test. n.s.: not significant. H Shortest nm distance between neighboring ccr4 or osk clusters. Mean ± STDEV of 2999 and 2074 cc4 and osk clusters, respectively, were analyzed. Statistical significance: two-tailed t test. ****: P < 0.001. I Average size of osk mRNA clusters (black circle and error bar) and of 12 reference mRNAs (aret, ccr4, CG18446, eIF4G2, gcl, orb, Pi3K, p53, pum, shutdown, sra and tao1; magenta circles and error bars) plotted against their respective somatic concentrations. The data for the reference mRNAs were determined previously (Trcek et al., 2020). All differences in cluster sizes between osk (black circles) and reference mRNAs (magenta circles) are statistically significant, with the P values ranging from 0.01 (*) to 0.004 (**; two-tailed t test). In addition to RNA sequence features, RNA concentration and RBPs can also enhance mRNA clustering driven by CSs. For instance, dimerization of the MLV RNA is thought to occur during transcription, where local viral RNA concentration is high (Maurel and Mougel, 2010). Likewise, HIV-1 RNA dimerization frequently takes place upon RNA localization to the plasma membrane and requires the viral Gag protein (Weiss et al., 1984; Muriaux et al., 1996), while clustering of BNI1, SPA2, osk and bcd mRNA similarly depends on their respective RBPs (Ferrandon et al., 1997; Chekulaeva et al., 2006; Besse et al., 2009; Langdon et al., 2018; Bose et al., 2024). Together, these observations suggest that RNA concentration and RBPs are important for establishing efficient CS-dependent intermolecular base pairing to promote RNA clustering. However, how these factors enhance mRNA clustering in vivo has not been systematically investigated. Here, we dissect the contribution of mRNA concentration, RBPs, and intermolecular base pairing to the clustering of the Drosophila osk mRNA. osk’s ability to engage in intermolecular base pairing has been experimentally validated both in vitro and in vivo. Specifically, osk contains an exposed palindrome (CCGCGG) positioned at the apex of a stem within its 3′ untranslated region (UTR), which mediates intermolecular base pairing between osk mRNAs (Jambor et al., 2011; Jambor et al., 2014; Figure 1B). This palindrome enables one osk mRNA to “hitchhike” on another during its transport to the posterior pole (Jambor et al., 2011; Jambor et al., 2014; Figure 1B). Mutations that disrupt the palindrome abolish dimerization of osk 3′UTR in vitro and posterior localization of the mRNA in vivo (Jambor et al., 2011). Furthermore, trans compensatory mutations revealed that dimerization of osk occurs through direct intermolecular base pairing in vivo (Jambor et al., 2011), highlighting the palindrome as a critical cis-acting element required for osk mRNA localization. At the posterior pole, hundreds of osk mRNA copies oligomerize within founder granules (Little et al., 2015; Eichler et al., 2020; Trcek et al., 2020; Eichler et al., 2023), wherein osk palindrome remains engaged in intermolecular base pairing (Trcek et al., 2020), acting as an “interaction sticker” (Bose et al., 2024) that supports osk oligomerization in vivo. While distinct RBPs promote osk mRNA clustering (Chekulaeva et al., 2006; Besse et al., 2009; Bose et al., 2024), in this study we examined the contribution of the double-stranded RBP (dsRBP) Staufen (Stau) as a model to evaluate the contribution of RBPs to mRNA clustering. We focused on Stau for several key reasons. First, Stau is essential for osk mRNA localization to the posterior pole, and in its absence, osk fails to localize (Ephrussi et al., 1991; Kim-Ha et al., 1991; Mohr et al., 2021). Since intermolecular base pairing between osk mRNA also enhances osk localization to the posterior (Jambor et al., 2011), we reasoned that Stau may contribute to dimerization of osk as well. Second, Stau is a dsRBP (St Johnston et al., 1992) that specifically recognizes the stem structure that carries the palindrome within the osk 3′UTR (Figure 1Ci, green segments; Laver et al., 2013; Mohr et al., 2021). Mutations that disrupt base pairing within this stem abolish Stau binding, while compensatory mutations that restore base pairing also restore Stau binding (Mohr et al., 2021). Moreover, mutations of the first two cytosines in the palindrome disrupt osk dimerization in vitro and its localization in vivo (Jambor et al., 2011), but preserve Stau binding (Bose et al., 2024). Therefore, Stau association with the stem that carries the palindrome does not require prior dimerization. This may allow Stau to directly promote initial RNA dimerization steps. Lastly, Drosophila Stau contains five RNA-binding domains (Figure 1Cii; Ramos et al., 2000), suggesting that it may function as a molecular bridge, linking RNA monomers and promoting their assembly as well as stabilizing intermolecular base pairing interactions required for oligomerization (Ferrandon et al., 1997). Here, we demonstrate that RNA concentration, Staufen, and the RNA palindrome together promote osk mRNA clustering, with osk oligomerization being the most strongly affected by the perturbations to these three factors. Our work suggests that this multi-layered requirement likely increases the stringency of clustering, making it a tightly regulated and context-dependent process in vivo. RESULTS Applying smFISH to examine osk mRNA clustering in early embryos We combined structured illumination microscopy, a super-resolution approach, with 48 single-molecule fluorescent in situ hybridization (smFISH) probes labeling the osk sequence (Figure 1, D and E, i and ii; Supplemental Table S1, Materials and Methods section) to examine the effect of osk mRNA concentration, dsRBP Stau, and the RNA palindrome on clustering of unlocalized osk in early embryos. We examined unlocalized osk mRNA, specifically, to detect individual and clustered osk mRNAs, which would be more difficult to discern within founder granules at the posterior pole where localized osk exhibits enhanced clustering (Figure 1A, i and ii; Eichler et al., 2023). We quantified the osk cluster size, defined as the absolute number of osk mRNAs within each diffraction-limited smFISH spot, and osk mRNA concentration for each embryo, as we have done previously (Trcek et al., 2015; Trcek et al., 2017; Trcek et al., 2020) (Materials and Methods). Fitting the distribution of fluorescent puncta to a double-Gaussian curve (Supplemental Figure S1A, black circles and line) revealed an osk population with two mean fluorescence intensities: X0 = 154.3 ± 2.9 arbitrary unites (a.u.), and a second mean intensity that was twice the value of X0 (X1 = 277.2 ± 25.0; P < 0.0001, R2: 0.96; Supplemental Table S1, Materials and Methods section). Importantly, the intensity of the first Gaussian peak, which represented the population of single mRNAs, closely matched the value of 154.9 ± 4.6 a.u. determined in embryos with osk mRNA levels reduced via RNAi, whose expression was driven by a GAL4/UAS system under a strong maternal alpha-tubulin (matα) promoter (Curnutte et al., 2023; referred to as “RNAi/matα”; P < 0.0001, R2: 0.97; Supplemental Table S1). We have confirmed that the reduction of osk expression in these embryos was strong (Supplemental Figure S1B). Moreover, this value also closely matched the mean fluorescence intensity of 153.0 ± 4.6 a.u. recorded in embryos lacking Stau expression (referred to as “stau1/stau2”; P < 0.0001, R2: 0.95; Supplemental Table S1). We confirmed that the stau1/stau2 embryos failed to localize osk at the posterior pole (Supplemental Figure S1C) and did not hatch (Supplemental Figure S1D), consistent with previous observations (Ephrussi et al., 1991). Indeed, the average intensity of the second Gaussian was twice the intensity of the average intensity of the first Gaussian, which is consistent with the second Gaussian peak representing the mRNA dimers in the population (Supplemental Table S1). Therefore, our approach robustly and reproducibly detected single osk mRNAs across different genetic backgrounds in vivo, and enabled us to quantify the discrete number of osk mRNAs in oligomerized clusters. To assess the contribution of the osk palindrome to mRNA clustering in vivo, we analyzed the previously characterized engineered mRNAs in which the coding sequence (CDS) of eGFP was fused to the osk 3′UTR carrying either the WT osk CCGCGG palindrome (termed eGFP-WT) or mutated AAGCGG sequence (termed eGFP-AA; see also Figure 2, Diii and Eiii; Jambor et al., 2011). We used 32 smFISH probes targeting the eGFP sequence to detect these transgenes and differentiate them from endogenous osk mRNA (Supplemental Table S1). Fitting the distribution of eGFP-WT and eGFP-AA fluorescent puncta to a Gaussian model revealed a population corresponding to a single mRNA for each transgene with nearly identical mean fluorescence intensities of 51.7 ± 1.4 and 50.0 ± 1.6 a.u., respectively (P < 0.0001, R2: 0.95 and P < 0.0001, R2: 0.97 for eGFP-WT and eGFP-AA, respectively; Supplemental Table S1). This highlighted the robustness and reproducibility of single mRNA detection by smFISH in vivo. High osk mRNA concentration, Stau expression and the RNA palindrome drive osk clustering in vivo. A–Ei–iii. Maximal projection of five consecutive optical sections spaced 150 nm apart of a WT (Ai), RNAi/matα (Bi), stau1/stau2; osk± (Ci) embryos and those expressing eGFP-WT (Di) and eGFP-AA (Ei) transgenes. Corresponding heat maps depicting the spatial distribution of osk (Aii, Bii, and Cii), eGFP-WT (Dii), and eGFP-AA (Eii) mRNA clusters are shown. Schematic of the WT (Diii) and mutated (Eiii) exposed RNA sequence within the osk stem-loop. F Dependence of the average osk, eGFP-WT and eGFP-AA mRNA cluster size on its concentration (nM). Data represent mean ± STDEV from seven WT, four RNAi/matα, seven stau1/stau2; osk± embryos, 13 eGFP-WT and 12 eGFP-AA embryos. Statistical significance: two-tailed t-test. *: P = 0.01, ***: P = 0.002 and ****: P < 0.001. n.s.: not significant. G Distribution of osk mRNA across clusters of different sizes. Bars represent mean ± STDEV cluster size from seven WT, four RNAi/matα, and seven stau1/stau2;osk± embryos, 13 eGFP-WT and 12 eGFP-AA embryos. Blue numbers: the mean ± STDEV molarity of osk, eGFP-WT, and 12 eGFP-AA mRNA monomers (n = 7,7, 13, and 12 embryos, respectively) for each genotype is shown. To further validate the accuracy of our single-molecule quantification, we compared osk mRNA concentration measured by smFISH with the value extrapolated from a standard curve. This curve was generated by correlating nanomolar (nM) concentrations of six reference mRNAs, namely germ cell-less (gcl), tao1, ccr4, sra, pum, and eIF4G2, as measured by smFISH, with their respective expression levels determined by RNA-sequencing data obtained from Flybase (mod et al., 2010). These reference mRNAs were selected because we have previously shown that they predominantly appear as single molecules within the early embryo (Trcek et al., 2015; Trcek et al., 2020). Moreover, their lower expression levels minimized the potential for crowding, which could otherwise lead to undercounting and diminish the accuracy of mRNA concentration extrapolation. We observed a strong correlation between mRNA expression levels determined by smFISH and RNA-sequencing for the six reference mRNAs (R2 = 0.72; Supplemental Figure S1E), indicating that smFISH quantified mRNA concentrations across a range of gene expression levels, consistent with our previous findings (Trcek et al., 2015; Trcek et al., 2017). By fitting the relationship between RPKM values from RNA-seq (mod et al., 2010) and smFISH-derived concentrations with a linear regression, we estimated the somatic concentration of osk mRNA to be 1.59 nM (Supplemental Figure S1E, green circle). This prediction closely matched the 1.43 ± 0.2 nM concentration measured by smFISH (Supplemental Figure S1E, magenta circle and error bars). These results indicated that our approach detected single osk mRNAs even in the presence of oligomerized species, allowing reliable quantification of osk mRNA concentration and its cluster sizes in vivo. smFISH revealed that WT embryos formed a heterogeneous population of osk mRNA clusters, ranging from monomers to assemblies containing more than 11 osk transcripts (Figure 1E, i–iii). In contrast, ccr4, our control mRNA, predominantly formed monomers (Figure 1F, i–iii) even though ccr4 clusters exhibited similar densities (defined as the number of clusters per volume unit) and slightly decreased cluster-to-cluster distances compared with osk (Figure 1, G and H). In addition, while the fluorescent intensity of osk clusters increased with cluster size, their diameters, measured in nanometers using a line scan, did not (Supplemental Figure S1F, i and ii). Instead, these diameters were comparable to those measured for ccr4 clusters (Supplemental Figure S1G) and for 100 nm TertraSpeck microspheres (Supplemental Figure S1H), which were below the diffraction limit of light of our structured illumination microscope. Together, these results supported our conclusion that multiple osk mRNAs were organized within confined, diffraction-limited clusters within the early embryos. Finally, mRNA clustering was specific to osk, as other mRNAs that were expressed at a similar or higher concentration than osk did not exhibit comparable clustering behavior (Figure 1I). Therefore, RNA clustering was an intrinsic property of osk mRNA, rather than a phenomenon driven by a limited optical resolution of the microscope. Collectively, our results established robust and consistent detection of single osk mRNA using smFISH across distinct genetic backgrounds, providing strong confidence that the observed differences in osk cluster sizes reflect bona fide changes in cluster organization. High osk mRNA concentration and Stau expression increase the likelihood of osk clustering Applying smFISH to WT embryos, we recorded an average osk cluster size of 2.74 ± 1.00 mRNAs (Figure 2, A, i and ii, and F, black circle and error bars), with 70.2% of osk mRNA found in clusters that contain more than one osk mRNA (Figure 2G). Therefore, unlocalized osk dimerized and oligomerized in WT embryos, consistent with previous observations (Little et al., 2015; Trcek et al., 2015). Upon reducing the expression of osk by 2.6-fold in RNAi/matα embryos, osk cluster size was also reduced by 2.2-fold, a statistically significant difference (Figure 2, B, i and ii, and F, gray circle and error bars). Importantly, comparable osk concentrations yielded similar cluster sizes across different genetic backgrounds (Supplemental Figures S1B and S2, A and B), indicating that mRNA concentration rather than genetic background per se drove changes in osk clustering in RNAi/matα embryos. In stau1/stau2 embryos, generated by crossing flies carrying two distinct mutations in the stau gene, the average osk cluster size similarly decreased from 2.74 ± 1.00 mRNAs in WT early embryos to 1.61 ± 0.12 mRNAs in stau1/stau2 embryos, a statistically significant change (Supplemental Figure S3A). However, loss of Stau also led to a 64% increase in total osk mRNA levels (Supplemental Figure S3, A and B). The mechanism behind this increase is unclear; however, it may reflect the Stau-mediated mRNA decay recorded in mammalian cells (Gong and Maquat, 2011). Therefore, to evaluate the role of Stau in osk clustering without the accompanying effect of osk mRNA concentration change, we combined the stau1/stau2 mutations with the osk gene deficiency Df(3R)pXT103 (Ephrussi et al., 1991; hereafter “stau1/2;osk±”), which resulted in osk mRNA expression levels comparable to those recorded in WT embryos (Figure 2F, green circle and error bars, Supplemental Figure S3, A and B). Under these conditions, the average osk cluster size decreased by 55.5% from 2.74 ± 1.00 recorded in WT embryos to 1.22 ± 0.14 in stau1/2;osk± embryos, a statistically significant change (Figure 2, C, i and ii, and F). The distribution of osk cluster sizes paralleled that observed in RNAi/matα embryos (Figure 2G) despite osk concentration being approximately two-fold higher than the one recorded for RNAi/matα embryos (Figure 2F, green versus gray circle). These data indicated that high mRNA concentration and Stau expression both augmented osk clustering. Next, we examined how osk cluster size depended on RNA concentration and Stau expression. Focusing first on oligomers, we observed that the fraction of oligomerized osk sharply decreased by 7.8- to 8.7-fold from 51.3% in WT to 6.6% and 5.9 ± 5.4% in RNAi/matα and stau1/2;osk± embryos, respectively (Figure 2G). Similar dependence of cluster size distribution on osk concentration was recorded in embryos with different genetic backgrounds (Supplemental Figure S2B). Additionally, commercially available overexpression lines generated only a modest 2.7-fold upregulation of stau mRNA levels (Supplemental Figure S3E), limiting our ability to further examine whether changes in Stau protein levels augment osk clustering. Instead, we investigated stau1/stau2 embryos in which the proportion of dimerized osk doubled, from 18.9 ± 3.8% in WT early embryos to 40.7 ± 1.2% in stau1/stau2 embryos (Supplemental Figure S3, C, i and ii, and D). In the absence of Stau, even with 64% higher total osk mRNA levels (Supplemental Figure S3, A and B), there were fewer oligomers as measured by the 50% reduction of oligomers in stau1/stau2 embryos compared with WT (Supplemental Figure S3D). These results revealed that the transition of dimers into oligomers was severely impaired in the absence of Stau. Therefore, high RNA concentration and Stau both promoted osk mRNA oligomerization. We then examined osk monomers and dimers. Reduced mRNA levels doubled the fraction of osk monomers in RNAi/matα embryos (Figure 2, B, i and ii, and G), consistent with a decreased probability of intermolecular encounters and osk dimerization at lower transcript abundance. However, in stau1/2;osk± embryos, osk mRNA expression level was comparable to the one recorded in WT embryos (Figure 2F, green and black circles, Supplemental Figure S3, A and B), allowing us to compare changes in osk monomers and dimers directly. Specifically, we observed that the concentration of osk monomers doubled from 0.37 ± 0.05 nM to 0.69 ± 0.15 nM recorded for WT and stau1/2; osk± embryos, respectively (Figure 2G, blue numbers). A similar increase in monomer concentration was recorded in stau1/stau2 embryos, which exhibited increased osk expression relative to WT (Supplemental Figure S3, A–D). Given this accumulation of osk monomers, which suggested impaired dimer formation, our data suggest that dsRBP Stau promoted osk dimerization in vivo by approximately doubling its probability. Collectively, our data support a model in which dimers could still form in the absence of Stau, although at reduced frequencies compared with WT embryos, whereas their progression into larger osk mRNA clusters was much more severely impaired. These findings indicate that osk clustering is governed by both osk mRNA concentration and dsRBP Stau, with the higher-order oligomer formation being more acutely sensitive to perturbations in these factors than dimerization. This differential sensitivity suggests that dimerization and oligomerization rely on partially distinct combinations of intermolecular interactions, with oligomerization requiring additional interactions beyond those necessary for dimer formation. osk palindrome enhances mRNA dimerization and oligomerization Next, we examined the contribution of the osk palindrome to mRNA clustering. In vitro, dimerization of the RNA derived from osk 3′UTR depends on a palindrome presented within a stem-loop structure (Figure 1B; Jambor et al., 2011; Jambor et al., 2014) and mutations that disrupt this sequence also abolish osk localization to the posterior pole in vivo (Jambor et al., 2011). Additionally, substitution of the first two cytosines of the palindrome to adenosines increases the apparent dissociation constant (Kd) for osk 3′UTR dimerization by threefold, from 90 nM observed for WT osk 3′UTR carrying the CCGCGG palindrome to ∼270 nM observed for osk 3′UTR carrying the mutated AAGCGG variant (Jambor et al., 2011). To assess this role in vivo, we analyzed clustering of eGFP-WT, carrying the CCGCGG palindrome, and eGFP-AA carrying the mutated AAGCGG sequence (Jambor et al., 2011; Figure 2, D, i–iii, and E, i–iii). However, osk 3′UTR performs an essential noncoding function during oogenesis, and females that do not express osk do not lay eggs (Jenny et al., 2006; Kanke et al., 2015). Although the UAS-Gal4/nanos promoter system is active throughout oogenesis (Van Doren et al., 1998), it did not drive sufficient expression of eGFP-WT and eGFP-AA transgenes in oskA87/Df(3R)pXT103 females, which lacked endogenous osk RNA expression, to restore egg laying (Supplemental Figure S4, A and B). Instead, we expressed both eGFP transgenes using the UAS-Gal4/matα promoter system. Unlike the UAS-Gal4/nanos system, which is weaker but active throughout oogenesis, UAS-Gal4/matα restricts transgene expression to late oogenesis and early embryogenesis (Curnutte et al., 2023). As a result, the transgenes could not be analyzed independently of endogenous osk, necessitating that we examine their clustering alongside endogenous osk using spectrally distinct smFISH probes targeting their respective CDSs. Importantly, both eGFP-WT and eGFP-AA exhibited similar expression levels (Supplemental Figure S4, C and D), allowing us to compare the dimerization and oligomerization potential of eGFP-AA against eGFP-WT. We observed that while eGFP-WT mRNA localized at the posterior pole (fold enrichment of 1.96 ± 0.36, defined as the ratio of eGFP fluorescent levels between the posterior and outside, Supplemental Figures S4, E, i and ii, G, and H), eGFP-AA mRNA displayed minimal localization (fold enrichment of 1.20 ± 0.15, Supplemental Figure S4F, i and ii, G, and H). These data indicated that the intact palindrome is necessary for efficient localization of osk to the posterior, in agreement with published observations (Jambor et al., 2011). Clustering of eGFP-WT mirrored that of endogenous osk (Figure 2, A, i and ii, D, i and ii, and G): 79.4% of unlocalized eGFP-WT was found in clusters that contained more than one eGFP-WT mRNA (Figure 2G), yielding an average cluster size of 2.40 ± 0.58 mRNAs (Figure 2F, dark blue circle and error bars). This trend remained even after eGFP-WT clusters that associated with the endogenous osk mRNA were removed from the analysis (Supplemental Figure S4I; also refer to Figure 3). Therefore, like endogenous osk, unlocalized eGFP-WT mRNA efficiently dimerized and oligomerized. osk palindrome is a potent driver of heterotypic osk mRNA clustering. A Co-localization between endogenous osk and eGFP-WT or eGFP-AA, quantified with Pearson's correlation coefficient (PCC). Mean ± STDEV of PCC values from 13 eGFP-WT and 12 eGFP-AA embryos are shown. Statistical significance: two-tailed t test. ****: P < 0.001. B and C Representative images showing co-localization between endogenous osk (green) and eGFP-WT (C) or eGFP-AA (D) (magenta) with a 250 nm cutoff. Maximal projection of 5 consecutive optical sections of an ROI, spaced 150 nm apart are shown. D Pie charts show the composition of each eGFP cluster category based on its association with endogenous osk mRNA. For example, 20.6 ± 9.7% of eGFP-WT spots display the intensity expected for a single eGFP-WT mRNA (“monomers”; dark blue in the bar chart). Co-localization analysis using spot detection revealed that 90.4 ± 5.4% of these were true monomers (dark blue in the pie chart and legend), whereas 8.7 ± 4.4% and 1.0 ± 1.2% co-localized with endogenous osk dimers and oligomers, respectively (lighter blue shades in the pie chart and legend). Differences in heterotypic cluster size between eGFP-WT and eGFP-AA were evaluated using a two-tailed t test. E Fraction of eGFP-WT and eGFP-AA clusters with heterotypic composition per given eGFP-WT (black) and eGFP-AA (magenta) cluster size. For example, of the 63.1 ± 9.3% of eGFP-AA mRNA (magenta) identified as monomeric, 5.6 ± 2.0% were heterotically associating with endogenous osk, while 20.6 ± 9.7% of eGFP-WT mRNA identified as monomeric (black), 18.3 ± 8.6% were heterotipically associating with endogenous osk. Mean ± STDEV from 13 eGFP-WT and 12 eGFP-AA embryos is shown. In contrast, eGFP-AA showed impaired clustering, which mirrored that of osk in RNAi/matα embryos where the osk levels were reduced (Figure 2, B, i and ii, E–G). Only 36.9% of unlocalized eGFP-AA was found in clusters containing more than one mRNA (Figure 2G), yielding an average size of 1.31 ± 0.12 mRNAs, a statistically significant decrease compared with eGFP-WT (Figure 2F, light blue circle and error bars). Notably, this reduction occurred despite similar expression of eGFP-AA and eGFP-WT (1.4 ± 0.3 nM and 1.3 ± 0.4 nM, respectively; Figure 2F; Supplemental Figure S4, C and D). Consequently, the concentration of single eGFP-AA molecules nearly quadrupled, from 0.22 ± 0.05 nM for eGFP-WT mRNA to 0.85 ± 0.16 nM for eGFP-AA mRNA (Figure 2G, blue numbers). This trend remained after eGFP-AA clusters that associated with the endogenous osk mRNA were removed from the analysis (Supplemental Figure S4I; also refer to Figure 3). Notably, the magnitude of this decrease was comparable to the ∼three-fold reduction in dimerization potential for the osk 3′UTR carrying the mutated AAGCGG palindrome relative to the WT palindrome recorded in vitro (Jambor et al., 2011). These results revealed that eGFP-AA mRNA with a disrupted palindrome still dimerized, albeit nearly fourfold less efficiently than eGFP-WT containing a WT palindrome. Thus, the palindrome is required for efficient osk dimerization and subsequent oligomerization. Notably, although mutation of the palindrome reduced dimerization, higher-order oligomerization was affected much more severely, suggesting that oligomerization requires additional intermolecular interactions than those sufficient for dimerization. Furthermore, mutation of the palindrome reduced dimerization approximately twofold more strongly than loss of Stau (Figure 2G), indicating that Stau alone cannot promote dimerization as effectively as the palindrome itself. Consistent with this interpretation, Stau was unable to fully compensate for the disruption of the palindrome. Thus, while both factors contribute to osk dimerization, the palindrome appears to play the more prominent role. Taken together, these findings suggest that the palindrome and Stau make distinct, non-redundant contributions to osk dimerization, with efficient oligomerization requiring the contribution of both factors. The RNA palindrome is a potent driver of heterotypic osk mRNA clustering To further assess the role of the palindrome in the clustering of osk, we analyzed the ability of eGFP-WT and eGFP-AA transgenes to heterotypically dimerize and oligomerize with endogenous osk using spectrally distinct smFISH probes targeting their respective CDSs. Applying Pearson's Correlation Coefficient (PCC) analysis to evaluate co-localization, we found that eGFP-WT better co-localized with endogenous osk than eGFP-AA (Figure 3A), indicating that the palindrome enhances the likelihood for heterotypic mRNA clustering. Moreover, this difference in co-localization also further supported the previously reported in vivo role of the palindrome in mediating intermolecular base pairing (Jambor et al., 2011). Next, we analyzed the composition of eGFP-WT and eGFP-AA mRNA clusters. We quantified the number of eGFP monomers, dimers, and oligomers that co-localized with one, two, or more endogenous osk mRNAs. A 250-nm (nm) cutoff between the eGFP transgene and endogenous osk was used to define co-localization (Figure 3, B and C; Materials and Methods section; Supplemental Figure S5, A and B). This inclusive cutoff was based on our previous measurement showing that, in founder granules in embryos, osk forms clusters with a diameter of ∼100 nm (Trcek et al., 2020), which corresponds to a spacing between neighboring osk clusters of approximately 100–150 nm. We observed that of the 20.6 ± 9.7% of all eGFP-WT monomers (Figure 2G), 81.7 ± 8.6% were true monomers, with the remaining fraction co-localizing with one or more endogenous osk mRNAs (Figure 3D, blue pie chart). Similarly, of the 19.3 ± 7.0% and 60.1% of all eGFP-WT dimers and oligomers, respectively (Figure 2G), 49.9 ± 14.3% and 32.2 ± 13.0% formed homotypic clusters, respectively, while the remainder co-localized with endogenous osk mRNAs (Figure 3D, grey and orange pie chart, respectively). In contrast, of the 63.1 ± 9.3% of all eGFP-AA monomers (Figure 2G), 94.4 ± 2.0% were true monomers, with the rest co-localizing with endogenous osk (Figure 3D, blue pie chart). Moreover, of the 21.1 ± 1.8% and 15.7% of all eGFP-AA dimers and oligomers, respectively (Figure 2G), 76.4 ± 11.6% and 54.9 ± 16.6% dimers and oligomers formed homotypic clusters, respectively, showing a statistically significant reduction in heterotypic interaction compared with eGFP-WT (Figure 3D, grey and orange pie charts, respectively). Importantly, this trend persisted even when the co-localization threshold was reduced to 100 nm (Supplemental Figure S5, C and D), indicating that the observed differences were robust to spatial resolution. Therefore, osk palindrome promoted heterotypic association between endogenous osk and the eGFP chimera. Interestingly, the probability for heterotypic association applied differently to monomers, dimers, and oligomers. Specifically, even though monomers represented the majority (63.1 ± 9.3%) of eGFP-AA mRNAs, fewer than 6% formed heterotypic clusters with endogenous osk (Figure 3E). In contrast, while monomers represented only 20.6 ± 9.7% of eGFP-WT transcripts, they were 3.3-fold more likely than eGFP-AA mRNAs to form heterotypic clusters with endogenous osk (Figure 3E). Therefore, the palindrome was a potent enhancer of heterotypic dimerization. In contrast, the fraction of eGFP-AA mRNAs that formed heterotypic clusters with endogenous osk increased by more than four-fold to 23.5% and 45.1%, respectively, even though these clusters contained less than 40% of all eGFP-AA mRNA (Figure 3E). These findings suggested that heterotypic association of a single eGFP-AA with endogenous osk is limited but becomes substantially more likely once eGFP-AA dimers and oligomers have formed. We observed a similar, although more pronounced trend with the eGFP-WT mRNA (Figure 3E). These data suggested that the ability of osk monomers to form heterotypic interactions were strongly influenced by the osk palindrome, whereas heterotypic interactions involving dimers and oligomers were enhanced by additional interactions that may be driven by RBPs as well as other intermolecular base pairing interactions. This model is consistent with our observation that oligomerization of endogenous osk is mediated by multiple types of interactions, involving both RNAs and proteins (Figure 2). Together, our results demonstrated that the RNA palindrome not only enhanced dimerization and oligomerization of osk mRNA but also promoted its heterotypic organization with mRNAs derived from distinct genetic loci. RNA palindromes may spatially organize other mRNAs in the early Drosophila embryo Finally, we asked whether other Drosophila mRNAs may contain palindromes similar to osk. To address this, we identified all palindromes in transcripts of early embryos (NC 1 to 14) using an in-house script we described previously (Tian et al., 2025). While RNA motifs consisting of only a few nucleotides may contribute to intermolecular interactions in vivo, we restricted our analysis to palindromes that were at least six nucleotides long as well as at least 50% GC-rich. In addition to resembling the osk palindrome, this threshold was supported by our previous work, which demonstrated using non-denaturing agarose gel electrophoresis that in vitro-transcribed RNAs containing shorter palindromes or palindromes with less than 66% GC content failed to dimerize (Tian et al., 2025). Using these criteria, we analyzed 11657 non-localizing transcripts (termed “ubiquitous”) representing isoforms of 4304 genes (Supplemental Table S2). Because osk localizes to the posterior, we reasoned that a similar palindrome-driven dimerization mechanism may contribute to the localization of other posterior-enriched mRNAs. Therefore, we separately analyzed 211 transcripts representing isoforms of 59 mRNAs previously identified as germ granule-enriched (termed “posterior localized”; Rangan et al., 2009; Supplemental Table S2). We observed that the total number of palindromes per transcript scaled linearly with transcript length (Supplemental Figure S6A) and that palindromes were modestly abundant across transcripts regardless of posterior localization (Supplemental Figure S6B). Therefore, the presence of the palindrome was not predictive of an mRNA's ability to enrich at the posterior pole. GC-rich palindromes accounted for less than half of the total palindrome density, indicating that palindromes were more frequently AT-rich than GC-rich (Supplemental Figure S6B). Because the osk palindrome is both GC-rich and presented atop an RNA stem-loop (Figure 1B; Jambor et al., 2011; Jambor et al., 2014), we searched for GC-rich palindromes that resided within such RNA structures (Materials and Methods section). As the full secondary structure of the Drosophila transcriptome was not experimentally available, we approximated RNA structure using Vienna RNAfold. Due to the symmetrical nature of palindromes, we deemed a palindrome whose first half is embedded but second half exposed as not able to engage in intermolecular dimerization. This is because the base pairing partner of the second half is the first half. Therefore, to gain a more accurate understanding of how many palindromes may be able to dimerize, we considered the maximum number of consecutive nucleotides that were exposed in both palindrome halves, termed the “number of base-pairable positions” or “bp score” (Figure 4 Ai; Supplemental Table S2). RNA palindromes may spatially organize other mRNAs in the early Drosophila embryo. A, i and ii Schematic of two palindromes presented atop an RNA stem-loop with 6 or 0 base-pairable positions (Materials and Methods section) (i). Histogram of the distribution of all (gray), AT-rich (green) and GC-rich (magenta) palindromes with a given number of base-pairable positions identified within the 5′UTR, CDS and 3′UTR of ubiquitous and posterior localized transcripts expressed in the early Drosophila embryo (ii). Bi–Cii Maximal projection of five consecutive optical sections of an ROI, spaced 150 nm apart, of Pep mRNA (Bi) and eIF3a mRNA (Ci) with corresponding heat maps depicting the spatial distribution of Pep and eIF3a mRNA clusters (Bii and Cii). D Distribution of Pep and eIF3a mRNA across clusters of varying sizes. For each cluster category, mean ± STDEV of 14 embryos per mRNA is shown. Data for stau1/stau2 and stau1/stau2; osk± are from Supplemental Figure S3D and Figure 2G, respectively. E Dependence of average osk mRNA cluster size on its concentration (nM). For Pep and eIF3a mRNAs, data represent mean ± STDEV from 14 embryos. Data for stau1/stau2 and stau1/stau2; osk± are from Supplemental Figure S2A and Figure 2F, respectively. Statistical significance: two-tailed t test. ***: P = 0.001; ****: P < 0.001. n.s.: not significant. F A model showing how the RNA palindrome, elevated mRNA concentration, and Stau dsRBP promote osk dimerization and subsequent oligomerization. Importantly, Stau and the palindrome may function as distinct, nonredundant contributors to osk mRNA dimerization and oligomerization, or may have interdependent activities. Moreover, the palindrome also markedly increases the likelihood of heterotypic osk association (depicted as interacting black and orange RNAs), indicating that the palindrome is a potent driver of heterotypic osk mRNA organization. Dimerization and oligomerization may be further supported by additional RBPs (blue circles) and intermolecular base pairing interactions not studied here. We observed that exposed palindromes with at least six base-pairable positions were mostly AT-rich, while those resembling the GC-rich palindrome of osk were extremely rare (Figure 4Aii). These findings are consistent with our recent study in which we mapped GC-rich palindromes onto the predicted RNA structure of 3′UTRs of germ granule-localized mRNAs nanos, germ-cell-less and polar-granule-component derived from in vivo chemical probing. Although all three contained a few GC-rich palindromes, these sequences were embedded within RNA structures and therefore unavailable for intermolecular base pairing (Tian et al., 2025). Looking at both ubiquitous and posterior localized mRNAs and their isoforms, we identified 60 (1.4%) mRNAs representing 155 (1.3%) isoforms that contained exposed, GC-rich palindromes (Supplemental Table S3). The majority of these palindromes were found in the CDS (in 38 out of 60 mRNAs and 101 out of 155 isoforms), while the 3′UTR had the least (in 9 out of 60 mRNAs and 19 out of 155 isoforms; Figure 4Aii; Supplemental Table S3). While none of these mRNAs were simultaneously highly expressed and predicted Stau targets (Supplemental Table S3), two mRNAs with GC-rich palindromes found in the CDS, Pep and eIF3a, nevertheless stood out as their expression levels recorded by RNA-seq were the most similar to the one recorded for osk (Supplemental Table S3). smFISH analysis revealed that neither Pep nor eIF3a exhibited the strong clustering behavior characteristic of osk. Even so, 42.5 ± 6.8% and 11.3 ± 6.8% of eIF3a dimerized and oligomerized, respectively (Figure 4, B–D). The distribution of eIF3a cluster sizes was comparable to osk in stau1/2 embryos (Figure 4D), which was noteworthy given that eIF3a expression was approximately two-fold lower than that of osk in stau1/2 embryos (Figure 4E). In contrast, only 24.3 ± 6.0% and 2.8 ± 1.5% of Pep dimerized and oligomerized, respectively (Figure 4D), displaying a similar cluster size distribution observed for osk in stau1/2;osk± embryos (Figure 4, D and E). These data suggested that eIF3a mRNA may be a bona fide candidate transcript, whose spatial organization, although moderate, could be driven by intermolecular base pairing. Importantly, the early embryo may not provide an environment that would support abundant clustering of eIF3a as observed for osk. For example, eIF3a may require distinct dsRBPs, higher mRNA concentration, as well as additional intermolecular base pairing interactions for efficient clustering. Nevertheless, further experimental validation is needed to determine whether eIF3a represents a bona fide mRNA whose clustering is driven by palindrome-driven intermolecular base pairing in vivo. In summary, our transcriptome-wide characterization revealed that while not widespread, exposed GC-rich palindromes may not be unique to osk. Our findings raise a possibility that palindrome-driven mRNA clustering may represent a previously underappreciated mechanism that contributes to mRNA organization during early Drosophila embryogenesis. DISCUSSION Persistent intermolecular base pairing requires the involvement of high RNA concentration and RBPs Compared with in vitro studies using the isolated osk 3′UTR sequence (Jambor et al., 2011; Bose et al., 2024), the conditions that support intermolecular base pairing of osk RNA in vivo are more stringent and are driven by high RNA concentration, Stau, and the RNA palindrome. This principle closely mirrors RNA dimerization of HIV-1 (Moore and Hu, 2009). HIV-1 contains a GC-rich palindrome (“GCGCGC” or “GUGCAC” depending on the subtype; Brunel et al., 2002; Windbichler et al., 2003) exposed within the SL1 stem–loop (Moore and Hu, 2009) that promotes HIV-1 RNA dimerization in vitro (Sakuragi et al., 2016) and facilitates heterotypic RNA dimerization in cells (Moore et al., 2007). However, as with osk, the palindrome alone is insufficient for efficient HIV-1 RNA dimerization, and additional factors participate. The HIV-1 nucleocapsid protein 7 (NCp7), which is produced by processing of the larger polyprotein Gag (Weiss et al., 1984; Muriaux et al., 1996) binds to the stem-loop containing the GC-rich palindrome (Cruceanu et al., 2006; Abd El-Wahab et al., 2014), and stabilizes RNA dimers in vitro (Darlix et al., 1990; Feng et al., 1996; Yuan et al., 2003; Cruceanu et al., 2006) and in cells (Chen et al., 2016). Expression of Gag increases the population of dimeric RNAs by roughly three-fold (Chen et al., 2016), paralleling the effect of Stau on osk dimerization in embryos (Figure 2). Thus, NCp7 and Stau appear to act analogously as molecular “glues” that stabilize palindrome-initiated RNA-RNA interactions. Notably, Stau and the palindrome may function independently to promote osk clustering, or they may act interdependently. Future analysis of the combined Stau/palindrome mutations is required to distinguish between these mechanisms and clarify their respective contributions to osk clustering. Moreover, in vivo, dimerization of HIV-1 frequently occurs at the plasma membrane in the presence of the Gag protein (Chen et al., 2016), where membrane association likely elevates local RNA concentration. Indeed, many RNA viruses assemble at sites of high RNA density, including transcription sites, the plasma membrane, or spherule invaginations derived from vesicles formed by the mitochondrial outer membrane (Maurel and Mougel, 2010; Chen et al., 2016; Zhan et al., 2023). This reinforces our observation that exposed base-pairing sequences and elevated local RNA concentration work together with RNA-binding proteins to enable efficient and sustained intermolecular RNA interactions in vivo. Multivalency and higher-order mRNA clustering Despite their similarities, HIV-1 packages a defined RNA dimer into the virion, whereas osk transcripts assemble into clusters that may contain many mRNA copies (Figure 2A, i and ii; Little et al., 2015; Eichler et al., 2023). This difference likely reflects interactions promoted by NCp7 and Stau, other RBPs, as well as the RNA itself. NCp7 contains two zinc finger motifs that primarily bind the stem-loop structure responsible for HIV-1 RNA dimerization, with some binding also detectable with the adjacent stem-loop (Dannull et al., 1994; Muriaux et al., 1996; Mailler et al., 2016; Koma et al., 2025). In contrast, Stau contains five dsRBDs (Figure 1Cii) and binds the palindrome-containing stem with dsRBD 3 (Figure 3Ci; Micklem et al., 2000; Ramos et al., 2000; Mohr et al., 2021). This architecture suggests a model in which Stau stabilizes dimers through dsRBD3-mediated binding, while the remaining four dsRBDs could bridge additional osk transcripts, bringing them into proximity and enabling higher-order oligomerization. Importantly, mutations in the palindrome reduce but do not abolish osk dimerization (Figure 2). This is consistent with in vitro studies showing that disruption of the palindrome does not abolish dimer formation of the isolated osk 3′UTR on gels (Jambor et al., 2011; Bose et al., 2024). These findings suggest that additional RNA elements may promote intermolecular RNA:RNA interactions even in the absence of the osk palindrome, in line with recent in vitro RNA phase-separation experiments using the osk 3′UTR (Bose et al., 2024). Alternatively, the palindrome may mediate initial recognition, whereas other factors, including RBPs, stabilize the interaction. In this scenario, partial disruption of the palindrome would reduce, but not eliminate, dimerization. Similarly to the mutations of the palindrome, loss of Stau strongly reduced, but did not abolish, osk dimerization and oligomerization (Figure 2, D, i and ii, and E, i and ii), indicating that additional factors contribute to this process. Two RBPs, the polypyrimidine tract-binding (PTB) protein and translational repressor Bruno, contribute to intermolecular osk interactions within founder granules at the posterior pole (Chekulaeva et al., 2006; Besse et al., 2009; Bose et al., 2022; Bose et al., 2024), and they could compensate for the lost Stau in the oligomerization of unlocalized osk. Both proteins exhibit multivalent RNA-binding capacity (Chekulaeva et al., 2006; Snee et al., 2008; Besse et al., 2009; Reveal et al., 2011; Bose et al., 2022), and Bruno can also dimerize (Kim et al., 2015), potentially reinforcing osk oligomerization. Notably, Drosophila PTB contains four RNA recognition motifs capable of binding multiple PTB-binding sites within the osk 3′UTR (Besse et al., 2009). Similarly, Bruno harbors three RBDs, each capable of engaging multiple Bruno response elements in osk (Chekulaeva et al., 2006; Snee et al., 2008; Reveal et al., 2011; Bose et al., 2022). Given the multivalent RNA-binding capacity of both PTB and Bruno, together with Bruno's ability to dimerize (Kim et al., 2015) and the presence of multiple binding sites for these proteins within osk mRNA, collectively these proteins likely provide an additional interaction layer that reinforces oligomerization of unlocalized osk. Thus, osk clustering is supported by multivalency at both the RNA and protein levels. The palindrome as a determinant of compositional specificity Interestingly, we observed that eGFP-AA mRNAs were 3.3-fold less likely than eGFP-WT mRNAs to dimerize with endogenous osk (Figure 3E), despite maintaining the binding sequences for Stau, Bruno or PTB (Bose et al., 2024). This indicates that the palindrome itself is a potent driver of heterotypic mRNA:mRNA recognition. A similar principle operates in Ashbya gossypii, where base pairing between BNI1 and SPA2 mRNAs establishes compositional specificity of condensates associated with polarized growth (Langdon et al., 2018). By analogy, our data support a model in which palindrome-mediated base pairing drives initial recognition and dimerization, while RBPs predominantly stabilize and expand assemblies. Interestingly, a substantial fraction of reporter-derived oligomers remained homotypic (32.2 ± 13.0% for eGFP-WT and 54.9 ± 16.6% for eGFP-AA; Figure 3D), suggesting a possible demixing of oligomers as observed for mRNAs that enrich in Drosophila germ granules (Trcek et al., 2015; Trcek et al., 2020). Alternatively, dimerization and oligomerization may occur early in the osk mRNA life cycle. Both PTB and Bruno function in mRNA splicing in the nucleus and shuttle between the nucleus and cytoplasm (Wagner and Garcia-Blanco, 2001; Snee et al., 2008; Spletter et al., 2015). This raises the possibility that osk clustering could initiate co-transcriptionally, during or shortly after nascent transcript synthesis. This would mirror transcription-dependent dimerization proposed for the Murine leukemia virus (Maurel and Mougel, 2010). If clusters form co-transcriptionally at the site of synthesis, such assembly could ensure that endogenous osk and reporter alleles remain segregated once transported into the cytosol. RNA palindromes as a broader mechanism of mRNA organization Our computational analysis identified 60 candidate mRNAs expressed in the early embryo that were predicted to contain palindromes similar to the one found in osk (Supplemental Table S3). Subsequent smFISH analysis confirmed that one of these, eIF3a, also exhibited light clustering behavior comparable to osk in stau1/2 embryos (Figure 4D). Although none of these candidate mRNAs were both highly expressed and predicted Stau targets in early embryos, these parameters may change during later stages of development. Indeed, several transcripts reach expression levels comparable to or higher than osk at later developmental time points (Supplemental Table S3). Furthermore, efficient clustering of these mRNAs may require distinct double-stranded RNA-binding proteins (dsRBPs) as well as additional intermolecular base-pairing interactions. Notably, the Drosophila genome encodes at least 12 dsRBPs (DeSousa et al., 2003), any of which could contribute to the spatial organization of specific transcripts during development. Importantly, experimental validation is needed to determine whether these candidate mRNAs contain exposed palindromes that mediate mRNA clustering through intermolecular interactions in vivo. Additionally, limitations of our computational approach should also be considered. In particular, in silico RNA structure predictions may differ from structures measured by chemical probing in vivo (Rouskin et al., 2014). Consequently, our RNAfold-based search may both overestimate and underestimate the number of bona fide candidate mRNAs whose clustering could be driven by palindrome-mediated intermolecular base pairing. Indeed, our search failed to identify osk, despite evidence demonstrating that its exposed GC-rich palindrome mediates intermolecular interactions. Previous in vitro probing of the osk stem-loop structure revealed that in a subset of osk 3′UTR molecules, the first cytosine and last guanosine of the palindrome engage in intramolecular base pairing, leaving only the central four nucleotides available for intermolecular interactions (Jambor et al., 2011). Because RNAfold reports ensemble structural predictions across the RNA population, this partial pairing reduces the predicted exposure of the palindrome below the six-nucleotide threshold used in our analysis, thereby excluding osk from the candidate list. Despite these limitations, our study suggests that exposed GC-rich palindromes may be present in other maternally deposited mRNAs in the early Drosophila embryo. This may represent an important and previously underappreciated mechanism that contributes to mRNA organization and development of the early Drosophila embryo. More than 30% of maternally deposited transcripts exhibit spatial organization (Lecuyer et al., 2007), where they play critical roles in early developmental patterning (Holt and Bullock, 2009; Chiappetta et al., 2022). Thus, although exposed RNA palindromes may not be ubiquitous, they could nonetheless represent an important and previously underappreciated mechanism contributing to mRNA organization in the early Drosophila embryo, and potentially in other organisms. In conclusion, intermolecular base pairing has emerged as a potent mechanism of mRNA organization. We find that although RNA palindromes can robustly promote intermolecular interactions in vivo, the cellular conditions that support such base pairing extend well beyond simple sequence complementarity. Thus, in vivo, robust mRNA clustering requires cooperation between multiple RNA motifs and RNA-binding domains. Importantly, this may impose regulatory stringency on intermolecular base pairing in cells, making RNA clustering a controlled and context-dependent process rather than a promiscuous outcome of sequence complementarity, as is often observed in vitro. This increased stringency may be critical to prevent unintended interference with translation, as reported for condensates formed by in vitro–transcribed RNAs (Trussina et al., 2025). MATERIALS AND METHODS Fly lines Flies were maintained on cornmeal molasses/yeast medium at RT or 25°C using standard procedures (Trcek et al., 2017). The following fly lines were used: for imaging, embryos expressing Vasa:GFP transgene were used as “wild-type” (WT) flies (Trcek et al., 2015); for egg hatching experiments, w1118 (Bloomington; 5905) were used; matα-gal4; matα-gal4 (Curnutte et al., 2023); nosGal4VP16 (w+)/TM3 (Cinalli and Lehmann, 2013), osk RNAi (y[1] sc[*] v[1] sev[21]; P{y[+t7.7] v[+t1.8] = TRiP.GL01101}attP2; Bloomington; 36903); oskDef (Df(3R)pXT103/TM3 Sb, Ser (Lehmann and Nusslein-Volhard, 1986); yw; osk54 st ry ss/TM3 Sb (Lehmann and Nusslein-Volhard, 1991; Markussen et al., 1995); stau1 cn1 bw1/Cyo (Bloomington; 1507), cn1 stau2/Cyo (Bloomington; 86573; St Johnston et al., 1991); oskA87/ TM6B (gift from Elizabeth Gavis; Jenny et al., 2006); w-; egfp-oskar 3′UTR-WT/cyo (Jambor et al., 2011; termed eGFP-WT) and osk 3′UTR-AA/cyo (Jambor et al., 2011; termed eGFP-AA), (both gift from Anne Ephrussi's lab; Jambor et al., 2011); P{EP}stau[G2356] (w[*]; P{w[+mC] = EP}stau[G2356]/CyO, P{ry[+t7.2] = sevRas1.V12}FK1;; Bloomington; 26989) and P{EPg}Spn55B[HP25975] (w[*]; P{w[+mC] = EP}stau[G2356]/CyO, P{ry[+t7.2] = sevRas1.V12}FK1;; Bloomington; 22094). Fly crosses used to modulate osk mRNA concentration To characterize the dependence of osk mRNA cluster size on mRNA concentration, we examined embryos laid by females in which osk expression was downregulated with Gal4-responsive UAS RNAi using a weak nanos (nos) promoter (Trcek et al., 2020; termed “RNAi/nosp”) as well as in embryos laid by osk54/Df(3R)pXT103 (termed osk54/Df) and osk54/+ females; Supplemental Figures S1B and S2A). Here, osk54 carries a nonsense point mutation that prevents Osk translation (Ephrussi et al., 1991; Kim-Ha et al., 1991), triggering nonsense-mediated mRNA decay (NMD; Kurosaki et al., 2019), while Df(3R)pXT103 is a deficiency that deletes the osk gene (Ephrussi et al., 1991). In addition, we also examined embryos in NC 13-14 (termed “WT late”), in which osk mRNA levels have diminished due to decay (Eichler et al., 2020). We observed that RNAi/nosp and osk54/+ embryos exhibited similar mRNA expression levels to WT embryos and accordingly formed osk mRNA clusters of similar sizes, ranging from 2.69 ± 1.06 (osk54/+) to 2.84 ± 0.68 (RNAi/nosp; Supplemental Figures S1B and 2A). Additionally, osk54/Df and WT late embryos exhibited similar mRNA expression to RNAi/matα and accordingly formed osk mRNA clusters with lower sizes ranging from 1.22 ± 0.09 (RNAi/matα) to 1.66 ± 0.28 (osk54/Df; Supplemental Figures S1B and S2A). Egg hatching assays Approximately 20 stau1/stau2, their sibling (stau1or2/cyo), stau1/stau2;osk± and females were crossed with 10 W1118 young males and maintained at 25°C for three days on standard cornmeal/agar media supplemented with yeast powder. Afterward, the flies were caged, supplied with an apple juice plate containing a dollop of yeast paste, and incubated at 25°C for 24 h to lay eggs (Trcek et al., 2017). The apple juice plate was then collected, the yeast paste removed, and the number of eggs on it determined. The plate was then incubated at 25°C for an additional 48 h, after which the number of hatched eggs was counted. The hatching rate was determined by dividing the number of hatched eggs by the total number of eggs on the same plate. Egg laying assay 21 eGFP-WT/nos-Gal4;oskA87/Df(3R)pXT103 (e.i. eGFP-WT), eGFP-AA/nos-Gal4;oskA87/Df(3R)pXT103 (e.i. eGFP-AA), eGFP-WT/nos-Gal4;oskA87/TM6B or eGFP-WT/nos-Gal4; Df(3R)pXT103/MKRS (e.i. WT-SIB) and eGFP-AA/nos-Gal4;oskA87/TM6B or eGFP-AA/nos-Gal4; Df(3R)pXT103/MKRS (e.i. AA-SIB) three-to-four-day old females were crossed with 10 WT males an allowed to lay eggs on an apple juice plates for 24 h at 25°C supplemented with a dollop of yeast paste. The apple juice plate was then collected, the yeast paste removed, and the number of eggs on it determined. Embryo collection Unless noted otherwise, caged flies were allowed to lay eggs at room temperature (RT) on a fresh apple juice plate containing a dollop of fresh yeast paste for 1.5 h, after which the embryos were dechorionated, fixed in 4% paraformaldehyde and the vitelline membrane removed by methanol cracking. Embryos were then stored in 100% methanol at 4°C until further use (Trcek et al., 2015; Trcek et al., 2017). Single-molecule fluorescent in situ hybridization (smFISH) A total of 48 fluorescently labeled DNA probes were used to hybridize against osk, Pep, and eIF3a mRNA, and 32 were used for eGFP. This approach strongly increases signal-to-noise ratio and increases our detection sensitivity, as described previously (Trcek et al., 2015; Trcek et al., 2017; Trcek et al., 2020; Curnutte et al., 2023; Tian et al., 2025). Subsequent hybridization of mRNAs with smFISH probes was carried out as described before (Trcek et al., 2015; Trcek et al., 2017; Trcek et al., 2020). smFISH probe sequences against osk, ccr4, and eGFP were listed in Trcek et al. 2015. The following 48 smFISH probes against Pep were used (listed in the 5′ to 3′ direction): ccgatccgaacgacgaaaga, taaacactccgcctatccaa, aaacgattctgcggatttcc, gttgacctttgcgttattca, taccacggaaagccatgttg, caaaattgcggttgcggttc, gtccaccatagttgttgttg, caaggcgacatgttcatacc, gtcgcatgttgttaccaaac, gattgttggccaagttgatg, ggttcctgaacagattgttc, aagtcgagaagcgatggtgg, accacgttggttgcgatttc, gtatggagactccttggtag, tagaacatgtcgttcggcac, atgtgcttcttgcacagatg, ccttgatgtggttctcgaaa, cacgcatcatcagatgggtg, aggcgatagctctcctcaat, ctcctgacggatcatgttgg, gatcgacttgagctgctctg, atgcgcttcaagcggtcaaa, gaagttcaggtcgcacatgg, acttgcgatgggtcgagatg, cttcttcagctgcaaatgtc, tgttgcactcaatgcacttg, tcaatgcgagtggcgaactc, cggacagcagatgagtatcg, ctcttcttcggtgctgatgg, tcttgaccggcttcttcttg, tttgtctcgtcagcagcttc, gtcgttgaagtcaaggatgc, cagttgtacttaggtaggcg, ctccagcttggagatcagac, tgcacaccgagcactcatag, acctcggtgtcgaagaactt, cgtacgggagtgaatctcgg, agaagttgcggtgatgagtc, gatcttggtatcgcttgatt, tcaaccttgcgcttcttacg, cgatggatcgtacagctcac, tcagcattatcatccaccat, aggagtttggactggagctg, cggagatggagaggcgtctg, tagtagcgattgtagcgacc, ttacgaattccttcgacgtc, tcctgggcgtgaatgtaaat, atctgttctttcttcatcct. The following 48 smFISH probes against eIF3a were used (listed in the 5′ to 3′ direction): gttgaggaacattcaagcgc, agattgcgaaagtcctcaga, ttcagtgggttgaaatccac, tagactgaatgcgcttgcac, ccgttctcaatgaagtccac, atgtaaggagtgagcagagc, cattatagtcacgtccttca, cctgagagatctgacgaatc, gcaagagacgctggaactta, atgttgcaaaaactggccag, accagcagtttttccaattc, atctgcatgtcattgtgacg, tgttcttctggtggtcaata, gtcagatcagtgccgaaata, cgagggcatcgattgcaaag, aaccaactgagagcggatct, ccgtgtcagaacagtagaca, ggttcgggtagacgattgaa, aatggtttaccatctggttg, tcgcgatccttgatctcatg, gtcgctgaagaatacgctgg, tttctgtcctcgatgatctt, cacgagcattattctgcttc, ttgcgctctctttcttcttg, gcttggatttcgttctgatg, ttctccttgagactcttctc, cgtctgcgagatttgctgaa, cagcttggaaagcatcttct, tccaacttcttaatgccttc, cgatttgagtttcgactgca, gcgttcgaagtagtcgattt, aaaagcggtatctcctccag, cttttcggccaagtattttt, cagaactccttgtccttaac, aatggcattctcgatgcgtg, ggatacatgcgcttgagacg, aagggcctcaaggaattcat, cttttcgacgtagagggagg, agagcagcctcaaacttttt, aataatgcggtcggctaagc, ctctttctcgcgaaggaagg, tgtgccaagcgaatctcttc, ttttcgtattgggcacgacg, ttccttgattttacgctcag, ttagatgcgcttggacggtc, tgtgaatctggctctcgtcg, cacttgtcatcgcgcgattg, catcaatgcggaaatcgcgg. Probes were coupled to a fluorescent dye using an in-house labeling protocol as we described previously (Tian et al., 2025). mRNA fold enrichment To measure mRNA fold enrichment, defined as the ratio of the mRNA fluorescence between the posterior and outside, the total fluorescence intensity of a 3D ROI cropped from the two respective locations was determined. The ROI were of the same dimensions. The analysis was performed using ImageJ, as we have done previously (Tian et al., 2025). Microscopy Images were acquired in three dimensions (3D) with a 150 nm Z step with a vt-instant Structured Illumination Microscope (vt-iSIM; BioVision Technologies). The microscope was equipped with the 405 nm 100 mW, 488 nm 150 mW, 561 nm 150 mW, 642 nm 100 mW, and 445 nm 75 mW lasers, two ORCA-Fusion sCMOS cameras and the Leica HC PL APO 63x/1.30 GLYC CORR CS2, HC PL APO 63x/1.40 OIL CS2 and HC PL APO 100x/1.47 OIL CORR TIRF objectives as described before (Curnutte et al., 2023). Line scan analysis to determine the dimensions and fluorescence intensity profiles Line scans were performed as described previously (Trcek et al., 2015). Briefly, for each cluster size, mRNA or bead, 20 line scans were obtained in ImageJ, and mean ± STDEV were plotted. 100 nm TetraSpeck microspheres (Thermo Scientific, T7279) were used. Quantifying nearest cluster-to-cluster distance To quantify local osk WT and ccr4 mRNA density, we performed a nearest-neighbor distance analysis on detected cluster coordinates. For each mRNA cluster, the Euclidean distance to its closest neighboring cluster was calculated in three dimensions. Coordinates were converted from pixel units to nanometers using voxel dimensions of 64.5 nm in x and y and 150 nm in z before distance computation. Nearest-neighbor distances were computed independently for each cluster without enforcing one-to-one assignment, allowing the same neighboring cluster to be identified for multiple reference clusters where applicable. mRNA concentration and cluster size measurements using smFISH Somatic mRNA concentrations were quantified as described previously (Trcek et al., 2015). In short, a 3D region of interest (ROI) with known spatial dimensions was cropped from the image and the absolute number of smFISH-labeled mRNAs within the ROI determined using Airlocalize (Lionnet et al., 2011; Trcek et al., 2017), allowing calculation of the nanomolar (nM) mRNA concentration within the ROI, as we have done previously (Trcek et al., 2015; Trcek et al., 2017). To determine the average fluorescence intensity of a single smFISH-labeled osk mRNA, we initially focused on data obtained from WT embryos. We plotted the integrated fluorescence intensities of all detected spots as a histogram and fitted it to a single-peak (1) and two-peak (2) Gaussian curve using Sigmaplot, as we have done previously (Trcek et al., 2017; Trcek et al., 2020): As the fluorescence intensity scales linearly with the number of fluorescently labeled probes within a diffraction-limited spot (Trcek et al., 2017), we inferred the mRNA counts by normalizing the cumulative fluorescence intensity of each smFISH-labeled spot against the fluorescence intensity of a single osk molecule (i.e., X0 value), following established protocols (Trcek et al., 2015; Trcek et al., 2017; Trcek et al., 2020). The two-peak Gaussian curve provided a significantly better fit to our data than the single-peak Gaussian model (R2 of 0.96 vs. 0.75, respectively; Supplemental Table S1), which displayed a significant deviation from the measured data (Supplemental Figure S1A). This was anticipated as a considerable proportion of osk mRNAs formed clusters. For osk mRNA analyzed in WT embryos, this analysis generated mean ± STDEV fluorescence intensity of osk clusters containing one (X0) or two (X1) mRNAs (Supplemental Table S1), derived from fits in Supplemental Figure S1A. Fitting the distribution of osk fluorescent puncta for WT embryos to a two-component Gaussian curve revealed two osk populations, with mean fluorescence intensities of 154.3 ± 2.9 and 277.2 ± 25.0 a.u., which differed by approximately two-fold (P < 0.0001, R2:0.96, Supplemental Table S1). We interpreted these as representing clusters containing one and two osk mRNAs, respectively. Accordingly, for all subsequent analyses, we fitted the distribution of integrated fluorescence intensities of all detected spots to a two-peak Gaussian model. To estimate an absolute number of mRNAs within each fluorescent spot, we normalized the integrated fluorescence intensity of each smFISH-labeled spot to that of a single osk mRNA molecule (X0 value), as we have done previously (Trcek et al., 2015; Trcek et al., 2017; Trcek et al., 2020). To determine the size of mRNA clusters (defined as the number of mRNAs per cluster), we analyzed a 3D ROI as described above. Afterward, we determined the number of mRNAs in each cluster by normalizing its fluorescent intensity to the intensity of a single smFISH-labeled mRNA (Trcek et al., 2015; Trcek et al., 2017). Finally, to plot the spatial distribution of osk mRNA clusters as a heatmap, we used Sigmaplot as we have done previously (Trcek et al., 2017; Trcek et al., 2020). Notably, our interpretation of changes in osk mRNA clustering relies on fluorescence-intensity distributions. This approach provides greater phenotypic resolution when assessing differences between monomers and higher-order mRNA clusters, as they exhibit the largest difference in integrated fluorescence intensity. Quantifying total embryonic mRNA levels using qRT-PCR Caged flies were provided with a fresh apple juice plate with a dollop of fresh yeast paste and allowed to lay eggs for 1.5 h, except for the WT late condition, which was aged for an additional hour. After dechorionation, eggs were washed with 1XPBS, and 200 µl TRIzol reagent (Invitrogen; 15596026) was added. Samples were homogenized using a pellet pestle motor (Kimble), followed by addition of 800 µl TRIzol. Samples were stored at −20°C until RNA extraction. Total RNA was purified using the Zymo RNA Clean & Concentrator kit (Zymo; R1018) according to the instructions. cDNA synthesis and qRT-PCR were performed as previously described (Tian et al., 2025). osk, stau, and Gapdh2 primer sequences were described previously (Curnutte et al., 2023). The primer sequences are provided in the Supplemental Table S4. For analysis, Act5C or Gapdh2 was used as a control to calculate the relative transcript levels, which was . Co-localization analysis using spot detection To quantify co-localization between endogenous osk mRNA and eGFP chimera (WT or AA) transcripts, images were acquired in three dimensions (3D) using an iSIM microscope with a 100x oil objective and subsequently corrected for chromatic aberration using Huygens (Scientific Volume Imaging), as done before (Trcek et al., 2015; Trcek et al., 2020). smFISH was performed using Quasar 670-labeled probes to detect endogenous osk and CalFl590-labeled probes to detect the eGFP chimera (WT or AA) transcripts (Trcek et al., 2015). Spot detection for each fluorescence channel was carried out independently using an algorithm called Airlocalize (Trcek et al., 2015; Trcek et al., 2017). After thresholding, the algorithm produced 3D coordinates (x, y, and z) of all detected spots in pixel units, which were then converted to nanometers based on imaging parameters. Pairwise distances between endogenous osk and eGFP chimera (WT or AA) transcript spots were calculated using a nearest-neighbor greedy-matching approach, ensuring a one-to-one assignment between spots (Trcek et al., 2015; Trcek et al., 2020). Co-localized pairs were defined using distance cutoffs of 100 nm and 250 nm. Both cutoffs yielded consistent trends with the 250 nm cutoff data presented in the main text (Figure 3, B–D). Custom Python scripts, mRNA Spot Matcher and mRNA Colocalization Visualizer, were developed to identify and visualize co-localized spots, respectively. Both codes with detailed documentation are available in the X GitHub repository: https://github.com/ayseecer/mRNA_Spot_Matcher_and_Colocalization_Visualizer Identification of GC-rich palindromes in the Drosophila transcriptome GC-rich palindromes are identified from fasta sequences downloaded from Flybase using an in-house script. Briefly, the fasta sequences of early Drosophila embryo genes are downloaded from Flybase in bulk, and GC-rich palindromes throughout each fasta sequence are identified as previously described (Tian et al., 2025). ViennaRNA RNAfold was used to generate predicted secondary structures of each mRNA sequence. A full list of the identified palindromes is included in Supplemental Table S2. The code for identifying GC-rich palindromes in bulk and for scoring each palindrome's intermolecular base pairing propensity (“bp score”) is available in the GitHub repository: https://github.com/AnneyYeZiqing/pfind-bulk-palindrome-finder Supporting information ACKNOWLEDGMENTS We would like to thank Drs. Sarah Woodson and Andrew Gordus for providing comments on our work. This research was supported in part by grant no. NSF PHY-2309135 to the Kavli Institute for Theoretical Physics (KITP). This research was supported by the NIGMS R35GM142737 grant awarded to T.T. Abbreviations used: double stranded RNA binding protein messenger RNAs oskar protein oskar gene single molecule fluorescent in situ hybridization staufen gene staufen protein. Footnotes This article was published online ahead of print in MBoC in Press (http://www.molbiolcell.org/cgi/doi/10.1091/mbc.E26-04-0176) on August 12 2026. REFERENCES Supplementary Materials