The long noncoding RNA mimi scaffolds neuronal granules to maintain nervous system maturity.

RNA binding proteins and messenger RNAs (mRNAs) assemble into ribonucleoprotein granules that regulate mRNA trafficking, local translation, and turnover. The dysregulation of RNA-protein condensation disturbs synaptic plasticity and neuron survival and has been widely associated with human neurological disease. Neuronal granules are thought to condense around particular proteins that dictate the identity and composition of each granule type. Here, we show in Drosophila that a previously uncharacterized long noncoding RNA, mimi, is required to scaffold large neuronal granules in the adult nervous system. Neuronal ELAV-like proteins directly bind mimi and mediate granule assembly, while Staufen maintains condensate integrity. mimi granules contain mRNAs and proteins involved in synaptic processes; granule loss in mimi mutant flies impairs nervous system maturity and neuropeptide-mediated signaling and causes phenotypes of neurodegeneration. Our work reports an architectural RNA for a neuronal granule and provides a handle to interrogate functions of a condensate independently of those of its constituent proteins.

INTRODUCTION

Cells partition their content as a strategy to coordinate the function of biomolecules in space and time. Ribonucleoprotein (RNP) granules are membrane-less cellular compartments composed of RNA and RNA binding proteins (RBPs) that interact with each other through multivalent interactions and dynamically exchange with the surrounding cellular milieu (). In the nucleus, granules are often scaffolded by long noncoding RNAs (lncRNAs), termed architectural RNAs (arcRNAs). In contrast, cytoplasmic granules are typically characterized by the presence of specific proteins. Cytoplasmic compartmentalization via biomolecular condensation is evolutionarily ancient and supports the maintenance of cellular function in a broad range of tissues; however, diseases linked to RNP granule homeostasis are predominantly associated with the nervous system. Hypo-assembly of RNP granule components in RBP loss-of-function models has been linked to neurodevelopmental, neurodegenerative, and neuropsychiatric disorders, while granule hyper-assembly and maturation into aggregates notoriously cause neurodegenerative diseases (, ).

The members of the highly conserved Staufen family of RBPs () compartmentalize the neuronal cytoplasm by forming two distinct types of granules: large, mostly immobile condensates associated with the rough endoplasmatic reticulum membrane and small granules that actively transport mRNAs from the soma to dendrites (–). Drosophila and mouse staufen mutants display severe neurological defects, including abnormal dendritic arborization, reduced locomotor activity, and memory deficits (–). The role of each type of Staufen condensate and the respective contribution to these phenotypes are not understood. Another well-known family of granule-associated RBPs comprises the highly conserved neuronal ELAV (embryonic lethal abnormal vision)–like proteins (nELAV proteins), which serve as early markers of neuronal identity across model organisms (). Mammalian and Drosophila cytoplasmic nELAVs play important roles in neuronal development, synaptogenesis, and synaptic plasticity/excitability (–). nELAV-deficient animals display a wide range of neurological phenotypes including epileptic seizures, impaired behavior, and age-dependent progressive motor deficits (–). Cytoplasmic nELAV proteins are found in different types of RNP granules, including neuronal granules (), stress granules (), and pathogenic insoluble inclusions (). How the cellular functions of nELAV proteins are tied to their condensation in the neuronal cytoplasm has not been established.

Contrasting with the prevailing model that cytoplasmic granules are scaffolded by proteins, in this study, we describe the first arcRNA for neuronal granules: the previously uncharacterized lncRNA mimi. mimi is undetectable outside of condensates, suggesting that its main cellular function is to support RNP granules. In addition, we find that Staufen, nELAV proteins, and mimi are all essential for those granules, demonstrating specific and nonredundant interactions at the core of multicomponent condensates. Our findings show how the molecular functions of nELAV proteins and Staufen converge in mimi condensates in the adult brain to regulate neuronal signaling and behavior.

RESULTS

FNE and RBP9 are required for specific neuronal granules

We investigated whether FNE and RBP9, the two cytoplasmic Drosophila nELAV proteins, are involved in neuronal granule formation or maintenance. We assessed the morphology and abundance of different types of RNP granules in the brain of adult flies lacking FNE (Δfne), RBP9 (Δrbp9), or both (ΔfneΔrbp9). Granules marked with Drosophila Imp (), Trailer Hitch (Tral), or Fragile X mental retardation protein (FRMP) () were not detectably affected in neurons of ΔfneΔrbp9 flies (fig. S1A). However, large Staufen (Stau) granules were drastically and specifically depleted (Fig. 1A). Small Stau granules were preserved, and Stau protein levels were not affected (Fig. 1, A and B), indicating that Stau underwent redistribution from granules to the neuronal cytoplasm. Moreover, large Stau granules were not depleted in individual Δfne or Δrbp9 mutants (fig. S1B). Together, our results show that FNE and RBP9 act redundantly in forming or preserving large Stau granules.

Fig. 1.

FNE and RBP9 are required for large Stau granules in neurons.

(A) Confocal imaging of neurons in the midbrain of control (w) and ΔfneΔrbp9 adult flies. The number of large Stau granules is indicated as a percentage of granules per cell found in control brains. Cells scored: n = 550 (control) and n = 645 (ΔfneΔrbp9). DAPI, 4′,6-diamidino-2-phenylindole. (B) Western blot comparing Stau protein expression in adult fly heads in the indicated genotypes. ELAV serves as a loading control. (C and D) Confocal imaging (C) and quantification (D) of the association of large Stau granules with tag-FNE and tag-RBP9. Granules typically contained Stau only (arrowhead) or all three proteins (arrow). Granules scored: n = 376. Scale bars, 4 μm (magnified images, 1 μm).

Next, to detect FNE and RBP9 by immunofluorescence (IF), we N-terminally tagged the endogenous fne () and rbp9 loci with FLAG-V5 and FLAG-MYC, respectively (fig. S1, C and D). Imaging of MYC, V5, and Stau in adult brains revealed that roughly half of large Stau granules include tag-FNE and tag-RBP9. Given that tag-RBP9 is expressed at levels similar to tag-FNE (fig. S1E), it is possible that Stau granules contain comparable levels of RBP9 and FNE. Notably, Stau granules contained either both or neither of the two proteins (Fig. 1, C and D, and fig. S1F). Together, our data show that FNE and RBP9 colocalize in Stau granules and indicate the existence of a possible intermediate molecule through which the three RBPs interact.

FNE and RBP9 regulate a lncRNA specific to the adult brain: mimi

Because FNE and RBP9 are required for Stau granule formation but do not appear to be systematically associated with them, we considered the possibility that FNE and RBP9 regulate a granule constituent RNA. To test this possibility, we performed protein immunoprecipitation (IP) from head tissue of flies coexpressing FLAG-V5-FNE and FLAG-MYC-RBP9, using antibodies directed against V5 and MYC, in the presence and absence of ribonuclease (RNase). We found that tag-FNE and tag-RBP9 interact in a predominantly RNA-dependent manner (fig. S1G). Therefore, we set out to identify an RNA coregulated by FNE and RBP9 in granules. Total RNA sequencing (RNA-seq) on adult fly heads revealed the lncRNA CR31451 as the top down-regulated transcript in ΔfneΔrbp9 (Fig. 2A). We set out to further characterize CR31451 and named it “mimi” (French for “cute”). mimi is a polyadenylated, ≈1-kb lncRNA (fig. S2, A and B). Notably, mimi-RC (thereafter referred to simply as mimi) is exclusively restricted to the adult nervous system and constitutes one of the most highly expressed polyadenylated RNAs in the Drosophila head (Fig. 2A and fig. S2, C and D). In ΔfneΔrbp9 mutants, the selective depletion of mimi (Fig. 2B and fig. S2E) concurrently with the specific loss of large Stau condensates (Fig. 1A), raises the possibility that mimi constitutes an architectural component of those granules.

Fig. 2.

mimi is an arcRNA for large Stau granules.

(A) Differential RNA expression in ΔfneΔrbp9 mutants compared to control flies (w) represented as a function of mean expression levels. Dark blue represents: |log2 fold change| > 0.5 with P < 0.01 and base mean > 10. (B) RNA-seq tracks for mimi and flanking gene regions in the indicated genotypes. The genomic deletion in Δmimi flies is indicated. Range: 0 to 1864 reads per kilobase million (RPKM). (C and D) Confocal imaging of Stau IF combined with mimi RNA in situ hybridization. Quantifications represent large Stau granules that contain mimi in wild-type neurons (w) (granules scored: n = 386) (C) and large Stau granules found in Δmimi brains compared to control (w) [cells scored: n = 429 (control) and n = 445 (Δmimi)] (D). Scale bars, 4 μm. (E) Quantification of mimi expression and granule number in Δmimi mutants (elav-Gal4; Δmimi) and Δmimi mutants in which mimi was expressed from a transgene (elav-Gal4 > UAS-mimi; Δmimi). Reverse transcription quantitative polymerase chain reaction (RT-qPCR) signal was normalized to RpL32. Expression and granule number in wild-type flies are set to 100%. Error bars represent ±SD of three biological replicates. Cells scored for granule quantification: n = 731 (control) and n = 623 (rescue). *P = 0.032 (two-sided t test).

mimi is an arcRNA for large Stau granules

To qualify as an arcRNA for an RNP granule, a lncRNA needs to fulfill two requirements: (i) enrichment in specific biomolecular condensates and (ii) disintegration of the condensates and dispersion of protein markers upon its removal. First, we determined mimi subcellular localization. In adult Drosophila brains, mimi was exclusively restricted to large Stau condensates and absent from small Stau granules and from the cytoplasm (Fig. 2C and fig. S2, F to H). Therefore, mimi is a specific, stable component of large Stau condensates. Second, we generated a fly specifically lacking mimi (Fig. 2B). Notably, adult Δmimi brains were completely devoid of large Stau granules. Loss of mimi caused a redistribution of Stau into the neuronal cytoplasm but did not alter total Stau protein levels (Fig. 2D and fig. S2I). tag-FNE and tag-RBP9 were up-regulated but expressed in the expected cytoplasmic pattern (fig. S2, J and K). Next, we investigated whether reintroduction of mimi RNA into Δmimi brains can restore mimi granule formation. Because mimi RNA is very highly expressed endogenously (Fig. 2A), our genetic rescue experiment used a UAS-mimi transgene under the control of a strong, pan-neuronal, elav-GAL4 driver. Nonetheless, we were able to reach only ≈10% of wild-type mimi RNA levels in Δmimi flies, which restored mimi granules to 9% of their wild-type levels (Fig. 2E). This partial rescue confirms that granule loss in Δmimi flies is caused by the absence of mimi and suggests that the lncRNA constitutes a limiting factor for granule formation. In conclusion, mimi is the first described arcRNA for neuronal granules: “mimi granules,” large Stau-containing cytoplasmic condensates characteristic of neurons of the adult nervous system.

mimi is directly bound by FNE, RBP9, and Stau in vivo

We tested whether permanent (Stau) and dynamic (FNE and RBP9) mimi granule components directly bind mimi. First, we identified mimi-bound proteins by applying RNA antisense purification and mass spectrometry (RAP-MS) () to adult fly heads (Fig. 3A and fig. S3, A and B). We identified 14 proteins specifically enriched in the cross-linked sample, among which all three Drosophila nELAV proteins (ELAV, FNE, and RBP9) were top hits (Fig. 3B). While FNE and RBP9 interact with mimi in cytoplasmic granules, the exclusively nuclear protein ELAV does not colocalize with mimi granules (fig. S3C) and likely binds mimi before the RNA exits the nucleus; alternatively, the identified nuclear proteins may constitute contaminants of the sample. Stau could not be detected in any sample, including the input, likely due to the low sensitivity of shotgun proteomics. To assess whether Stau directly binds mimi and confirm results for FNE and RBP9, we performed RNA immunoprecipitation with ultraviolet (UV) cross-linking (xRIP) on the head tissue of flies expressing, from the endogenous locus, tag-Stau, tag-FNE, or tag-RBP9 (Fig. 3C and figs. S1, C and D, and S3, D and E). We observed high and specific enrichment of mimi in the protein-bound RNA fraction for all three RBPs (Fig. 3D). Together, our results demonstrate that Stau and mimi directly interact to constitute mimi granules. FNE and RBP9 bind mimi directly and regulate the formation and/or maintenance of mimi granules.

Fig. 3.

mimi is directly bound by FNE, RBP9, and Stau in vivo.

(A) RAP-MS workflow. mimi RNA was hybridized to biotinylated DNA probes and captured on streptavidin beads. UV–cross-linked protein-RNA complexes were analyzed by shotgun proteomics. (B) Proteins recovered by mimi RAP-MS. In blue, significantly enriched or depleted proteins compared to the non–cross-linked control (P < 0.05, two-sided limma t test adjusted with the Benjamini-Yekutieli method). nELAV family proteins are underlined. (C) xRIP-qPCR workflow. tag-FNE, tag-RBP9, and tag-Stau proteins were captured on anti-FLAG beads. UV–cross-linked protein-RNA complexes were analyzed by RT-qPCR. (D) RT-qPCR quantification of RNA levels of the indicated transcripts in FLAG-xRIP samples from flies expressing tag-FNE, tag-RBP9, or tag-Stau. The RNA fold enrichment in the xRIP-tagged sample over input was normalized to enrichment in the untagged control (w) for each genotype. Error bars represent ±SD of at least three biological replicates for each genotype. (E) Consensus secondary structure predicted for mimi RNA based on multiple sequence-structure alignment. Paired nucleotides with increased covariance are in orange. Putative Stau, FNE, and RBP9 binding is indicated as a dotted line box and blue shapes, respectively.

To better understand the RNA-protein interactions at the source of mimi granules, we combined mimi secondary structure prediction with a covariance analysis that describes how distinct nucleotides have coevolved to maintain structural elements (fig. S3F) and generated a structural model of mimi (Fig. 3E). Because nELAV protein family members can interchangeably act on the same RNA target sequences (), and ELAV directly and specifically binds mimi (Fig. 3B and fig. S3G), we consider mimi regions identified in ELAV iCLIP (individual-nucleotide resolution UV crosslinking and immunoprecipitation) () to be putative FNE and/or RBP9 binding sites. Notably, ELAV binding predominantly occurs at predicted loop regions (Fig. 3E), consistent with ELAV’s known binding to single-stranded RNA (). Stau reportedly interacts with RNAs through well-described secondary structures (, ). In the mimi structural model, one region stood out by its increased covariance among homologs, thereby possibly representing a secondary structure of conserved functional relevance. Notably, this region is a stereotypical Stau binding stem-loop (Fig. 3E), which we propose constitutes the main interaction site for Stau.

Granule assembly and maturation depend on mimi interaction with FNE, RBP9, and Stau

The formation of mature RNP granules is preceded by a local increase in concentration of proteins and associated RNAs (). In fly brains, we were able to visualize structures that likely represent a precursor state of mimi granules: cytoplasmic foci in which Stau and mimi signal colocalize in a concentric pattern (Fig. 4A). Notably, tag-FNE and tag-RBP9 were found in these less condensed foci significantly more frequently than in mature granules (Fig. 4B and fig. S4A), suggesting that FNE and RBP9 are required for granule assembly rather than maintenance. To test whether the role of nELAV proteins in neuronal granule regulation is conserved in mammals, we cultured cells from the subventricular zone of adult mice and visualized nELAV/Hu proteins and Stau1 in differentiated neurons (fig. S4B). We found that roughly one-third of large Stau1 granules contained HuB, HuC, and HuD, respectively (fig. S4C), suggesting a potential role for mammalian nELAV proteins in Stau1 granule assembly. We propose the following model: Stau and nELAV proteins directly bind to mimi and instruct the assembly of granules. mimi and Stau constitute the core of mature granules, whereas nELAV proteins interact in a dynamic fashion. In mimi mutant flies, loss of the interactions between Stau, FNE, RBP9, and mimi prevents granule formation (Fig. 4C).

Fig. 4.

Granule assembly and maturation depend on mimi interaction with FNE, RBP9, and Stau.

(A) Visualization of mimi and Stau accumulation, and quantification of signal overlap, in precursor structures of cytoplasmic condensates in neurons of wild-type flies. mimi signal is overexposed for visibility. Scale bars, 4 μm (magnified images, 1 μm). (B) Quantification of the association of diffuse (assembling) and condensed (mature) Stau foci with tag-FNE and tag-RBP9. Data for mature granules are reproduced from Fig. 1. ***P = 3.7 × 10−4 (two-sided Fisher’s exact test); assembling granules scored: n = 72 (A) and n = 63 (B). (C) Model of mimi granule assembly. Local interactions of mimi with FNE, RBP9, and Stau are essential to initiate granule assembly. Interactions with FNE and RBP9 decrease with granule maturation, whereas Stau and mimi remain constitutive granule components. (D) Confocal imaging of neurons in the midbrain of stau-deficient adult flies and quantification of granule number. Cells scored: n = 429 (w control, not shown) and n = 487 (stau-deficient). Scale bars, 4 μm. (E) RT-qPCR quantification of stau and mimi RNA levels in control (w) and stau-deficient adult heads. Signal was normalized to RpL32. Error bars represent ±SD of four biological replicates.

Consistent with Stau, FNE, and RBP9 expression in both developing and mature neurons and their broad localization pattern throughout the cytoplasm, all three proteins are required for multiple aspects of neuron physiology (, , ); so far, it has not been possible to characterize adult-specific roles or to uncouple granule-dependent from granule-independent functions of the three RBPs. In contrast, mimi has the potential to represent the unique link between FNE, RBP9, and Stau functions in the condensates of mature nervous system. To address how exclusive mimi is to those condensates, we investigated whether the lncRNA can exist outside of granules. We used a combination of two alleles to obtain a fly in which stau levels were strongly reduced (fig. S4D). This eliminated 85% of mimi granules (Fig. 4D) and decreased mimi RNA levels by 85% (Fig. 4E). Residual mimi RNAs were exclusively found in granules marked with residual Stau protein (Fig. 4D and fig. S4E). tag-FNE and tag-RBP9 proteins were up-regulated (fig. S4F), indicating that, although necessary (Fig. 2), the two RBPs are not sufficient for mimi expression. Although we cannot exclude a possible effect of FNE, RBP9, or Stau loss on mimi transcription or nuclear export, our data, together with the depletion of mimi in an independent model of granule loss (ΔfneΔrbp9 mutant; Fig. 1A), strongly suggest that mimi can only subsist when incorporated into granules, where it is possibly protected from cytosolic microRNAs or RBPs by the phase boundary. Because mimi is required for mimi granules, Δmimi mutant flies represent a unique animal model of condensate loss and give us the opportunity to study the composition and physiological function of mimi granules in vivo, independently of its constituent proteins.

mimi granules maintain nervous system maturity

To study the consequences of mimi granule loss on the neuronal transcriptome and proteome, we performed total RNA sequencing and MS analyses on Δmimi heads (Fig. 5, A and B). For roughly half of all significantly deregulated mRNAs, the corresponding protein product was also affected, usually in the same direction (fig. S5A), revealing the broad impact of mimi granules on neuronal gene expression.

Fig. 5.

mimi granules regulate nervous system maturity and synaptic signaling.

(A and B) Quantification of mRNAs (A) and proteins (B) differentially expressed in Δmimi adult fly heads compared to control (w). Shown are mRNAs with |log2 fold change| > 0.5, P < 0.05 (two-sided t test), and base mean > 10 and proteins with |log2 fold change| > 0.5 and P < 0.05 (blue). Nonsignificant proteins are in gray. The horizontal line indicates the cutoff P < 0.05 (two-sided t test). mRNAs represented in (D) are colored accordingly. Genes colored in (A) that were detected in (B) are named. (C) Early embryonic genes are preferentially up-regulated in Δmimi heads, whereas adult-specific genes are down-regulated. The heatmap represents the developmental expression of genes up-regulated or down-regulated in Δmimi heads (vertical panel, left). Developmental expression data are from modENCODE (). (D) Top 3 enriched biological processes (P < 0.05; one-sided EASE score adjusted using the Benjamini-Hochberg method) in up-regulated (top) and down-regulated (bottom) mRNAs in Δmimi heads. (E) Workflow to purify neuronal granules including (wild type) or excluding (Δmimi) mimi granules. (F) Proteins identified as mimi granule components based on (E) (selected). Pink shading, components validated by IF; green shading, components involved in synaptic signaling. Shab belongs to both categories.

Most genes down-regulated in Δmimi heads were adult-expressed genes and enriched for synaptic signaling processes characteristic of mature neurons. Moreover, genes normally expressed in early embryonic development, and genes involved in cell proliferation, were enriched among up-regulated mRNAs (Fig. 5, C and D, and table S1). This indicates that mimi granules, which form exclusively in adult neurons, as soon as mimi is expressed (figs. S2C and S5B), play a role in maintaining the mature state of the nervous system.

mimi granules regulate synaptic signaling

Upon the loss of mimi granules, genes involved in neuropeptide-mediated signaling were specifically down-regulated at both mRNA and protein levels. Virtually, all neuropeptides act on G protein–coupled receptors (GPCRs) (). Consistently, the components of GPCR signaling pathways were also affected (Fig. 5, A, B, and D, and table S1), showing that mimi granules regulate neuropeptide signaling.

We hypothesized that mimi granules contribute to synaptic signaling by harboring signaling molecules. To investigate this possibility, we applied differential centrifugation [adapted from (, )] on lysate from adult fly heads and performed total RNA-seq and shotgun proteomics on the biochemical fractions (Fig. 5E). We obtained fractions enriched in specific RBPs and components of RNP complexes, including mimi (fig. S5, C and D, and table S2). To identify the specific components of mimi granules, we searched for RNAs and proteins enriched in the granule fraction of control flies compared to Δmimi flies. We identified 352 mRNAs and 297 proteins associated with mimi granules (table S3), including, as expected for mimi granules, FNE and RBP9. We confirmed the granule localization of several newly identified components by IF: Atox1, Shab, USP14, and BicD localization to mimi granules ranged from ~20 to 70% (fig. S5E). Hence, our analysis accurately detects granule components.

The composition of mimi granules was rather heterogeneous; Gene Ontology (GO) analysis revealed a modest but significant (P < 0.05) enrichment in regulators of neuronal signal transduction and behavior (fig. S5, F and G, and table S3). Notably, we found specific proteins involved in synaptic transmission and memory, including Shaker cognate b and Hyperkinetic (subunits of potassium channels involved in neurotransmitter release) (, ), Dunce and Rutabaga (two classic memory-related proteins) (, ), and the more recently identified Klingon (Fig. 5F) (). Together, our results show that mimi granules are hubs for regulation of RNAs and proteins implicated in neuropeptide signaling, with a potential role in learning and memory.

Evidence for a function of mimi granules in mRNA storage and distribution

To address the molecular effect of mimi granules on their resident RNAs and proteins, we asked whether granule components were affected in mimi mutants. Granule-associated proteins and RNAs (fig. S6, A to C), but not proteins encoded by granule-associated RNAs (fig. S6, D and E), were significantly more affected by granule loss compared to nongranule components, indicating that mimi granules contribute to RNA and protein homeostasis but are unlikely to directly affect translation. However, the global effect of granule loss on the cellular levels of most granule-resident proteins and mRNAs was mild (fig. S6, B, C, and E), suggesting that mimi granules do not play a major role in protein/RNA stabilization or degradation. We hypothesize that, instead, mimi granules could serve as storage and distribution centers for neuronal RNAs and proteins. Notably, mimi granules harbor an assortment of molecules involved in synaptic processes (Fig. 5 and fig. S5, F and G). The regulation of long-distance transport to neuronal projections by pre-assembling and/or stabilizing synaptic RNP transport complexes constitutes a possible molecular function of mimi granules.

Loss of mimi granules impairs life span and locomotion in aging flies

In both flies and mammals, neuropeptides are crucial regulators of multiple behavioral aspects including learning and memory (, ). Because in mimi mutants we can uncouple the granule-associated role of RBPs from their other functions in the cytoplasm, we asked whether memory is mediated through mimi granules. In an independent, large RNA interference screen study, knockdown of the gene CG31451 (mimi) was found to cause a reduction in memory performance without affecting the fly morphology or activity (). We performed an olfactory learning assay that assesses immediate aversive memory. Δmimi flies did not perform significantly worse than control flies; as the flies aged, memory worsening was notable but not statistically significant (Fig. 6A), potentially due to the low learning score seen in old flies that precludes proper analysis. Our results could not formally prove, but do not exclude, a role of mimi granules in learning or memory in aged flies.

Fig. 6.

Loss of mimi granules impairs life span and locomotion.

(A) Analysis of aversive memory in Δmimi flies compared to the heterozygous control (w/Δmimi). Learned odor avoidance was tested immediately after aversive conditioning. n.s.P = 0.96 (young) and n.s.P = 0.33 (aged) (two-sided t test). n.s., not significant. (B) Age-dependent reduction in climbing performance in Δmimi flies compared to the heterozygous control (w/Δmimi). The percentage of flies climbing above 3 cm within 5 s after startling was scored. One-hundred flies were assessed in 10 biological replicates and two technical replicates for each genotype and time point. Whiskers show the minimum and maximum values. From left to right, n.s.P = 0.078, ***P = 5.5 × 10−4, *P = 0.014, ****P = 2 × 10−8, and ***P = 5.5 × 10−4 (one-way analysis of variance adjusted with Holm-Šídák method). (C) Life-span shortening in Δmimi flies compared to the heterozygous control (w/Δmimi). Percentage of live flies is represented as a function of time after eclosion. Flies scored: n = 80 (ΔfneΔrbp9), n = 300 (Δmimi), and n = 300 (control). Shading represents ±SD of four (ΔfneΔrbp9) or six (Δmimi and control) independent replicates. Dotted lines represent half-lives. **P = 0.0022 (two-sided Kolmogorov-Smirnov test).

Negative geotaxis is an inherent aspect of Drosophila locomotor behavior, which can be measured in a climbing assay, an established approach to study motor function in fly models of neurological disease (). In our negative geotaxis assay, Δmimi flies showed reduced climbing performance. Notably, the loss of performance compared to age-matched controls was exacerbated as the flies aged (Fig. 6B). Moreover, we found that the life span of mimi mutants was significantly shortened, by ~20%, compared to control flies (Fig. 6C). The decreased longevity and age-dependent degradation of motor functions, in combination, are typical indicators of neurodegeneration (, ). Together, our results show that mimi granules play an important role in neuronal health and performing vital behaviors.

DISCUSSION

Cytoplasmic granules are typically thought to be scaffolded by proteins. So far, only one arcRNA, NORAD (noncoding RNA activated by DNA damage), has been described to drive the formation of cytoplasmic condensates: NORAD-Pumilio bodies in nonneuronal cells (). Our study identified the first arcRNA for neuronal granules—mimi—a constitutive and essential component of mimi biomolecular condensates. The stoichiometry of RBP/RNA interactions is central to phase separation and granule composition, suggesting that many condensates lack a specific scaffolding RNA or protein (). Consistent with this idea, multiple types of neuronal granules undergo dynamic remodeling and contain varying and partially overlapping sets of proteins and RNAs (). Although mimi granules share these properties, they distinguish themselves from other neuronal condensates by the presence of a unique identifier, mimi. It is therefore conceivable that many neuronal granules, or more generally cytoplasmic condensates, may be architecturally dependent on yet-to-be-discovered arcRNAs. We searched for a functional homolog of mimi in mammals using sequence and structure conservation and nELAV () as well as Staufen () binding data. So far, we were unable to identify a lncRNA that accumulates in large, Stau1/2-marked granules in mouse or rat neurons. However, the localization of nELAV proteins to the large Stau granules seems to be conserved in mammalian neurons (fig. S4). We propose that strategies to distinguish small from large Staufen granules could have evolved independently: In flies, the sole ortholog Staufen requires not only nELAV proteins but also an arcRNA to form large granules. In mammals, granule identity may be achieved by the preferential accumulation of Stau1 and nELAVs in large granules, while the Stau2 paralog scaffolds small granules.

Our study showed that mimi expression and mimi granule formation are mutually conditional: In Δmimi flies, the role of the condensate per se can be uncoupled from the individual roles of granule proteins. mimi granules mark the transition to the adult nervous system, and their loss caused a shift in gene expression reminiscent of immature cell states. mimi granules represent functional nodes at which FNE, RBP9, and Stau interact to maintain the mature neuronal state and perform important neuronal functions such as synaptic signaling. The mechanism of action of mimi granules is still not entirely clear. In agreement with our finding that mimi granules contain mRNAs and proteins associated with synaptic signaling, mimi granules may represent transport intermediates, storing mRNAs destined for travel to neuronal projections via small Stau granules. In addition, compartmentalization of neuronal RBPs may help avoid aberrant function of mislocalized granule-resident proteins or mRNAs, thereby promoting the full deployment of adult neuronal functions.

The functional links between disease-causing mutations and altered granule homeostasis are not well known (). We provide direct evidence that granule hypo-assembly drives neurological dysfunction. mimi granule loss results in disrupted expression of neuropeptide-GPCR–mediated signaling components, life-span shortening, and age-dependent decline of motor functions. Neuropeptides are implicated in the pathophysiology of psychiatric disorders such as depression, anxiety, and addiction, as well as of neurodegenerative diseases, particularly those associated with cognitive decline (, ). Neuropeptides and GPCRs, through which neuropeptides signal, constitute one of the most widely studied therapeutic targets for the treatment of neurological disorders (). We speculate that compartmentalization of the neuronal cytoplasm through mimi granules constitutes a broadly conserved strategy to regulate neuropeptide signaling; studying the underlying mechanisms in mammalian systems will be useful to gain a better understanding of human neurological disease.

MATERIALS AND METHODS

Experimental model

Experiments in this study used adult male and female Drosophila melanogaster. Flies were raised at 25°C. w, elav-Gal4, green fluorescent protein–marked balancer chromosomes, deficiency lines for rbp9 and stau, and the allele stau were obtained from the Bloomington Drosophila stock center (5905, 64349, 458, 4559, 6662, 6663, 8038, 24987, and 10742, respectively). Null alleles for fne were obtained from M. Soller () [Df(1)fneΔ] and M.-L. Samson () (fne). The null allele rbp9 () was obtained from M. Soller. Flies denoted as control are of the genotype w (Figs. 1 to 5 and figs. S1 to S4) or fne and rbp9 (fig. S2K) or w/Δmimi (Fig. 6). Flies denoted as wild type are of the genotype w. Flies denoted as Δfne are of the genotype Df(1)fneΔ/fne (females) and Df(1)fneΔ or fne (males) and Δrbp9 flies are of the genotype rbp9 (males and females). Flies denoted as ΔfneΔrbp9 are of the genotype Df(1)fneΔ/fne (females) and Df(1)fneΔ; rbp9 or fne (males). Flies denoted as stau deficient are of the genotype stau Flies denoted as tag-FNE are of the genotype fne () and express an endogenously, N-terminally FLAG-V5–tagged FNE protein. CRISPR-Cas9 genome editing followed the procedure described in (). To generate the mimi null mutant Δmimi, two guide RNAs (TGGTAGTACCATGAGGCGTG and AGTGTTAATTGTAAGATCCC) targeted the CR31451 gene region, generating a 1.2-kb deletion beginning 83–base pair (bp) upstream of the annotated transcription start site (Fig. 2B). To generate tag-RBP9 flies, rbp9 (expressing an endogenously, N-terminally FLAG-MYC–tagged RBP9 protein), a guide RNA (AGCGTTCGCAAGATGGTCGA) targeted rbp9 and genome editing used a 3142-bp homology donor (sequence in table S4). To generate the tag-Stau fly strain, stau (expressing an endogenously, N-terminally FLAG-V5–tagged Staufen protein), a guide RNA (AGCACAACGTTCATGCCGCC) targeted stau and genome editing used a 1440-nucleotide gBlock (Integrated DNA Technologies) as a homology donor (sequence in table S4). To construct the UAS-mimi transgene, mimi-RC was amplified from fly genomic DNA using the primers (ACCAGGAGCAGTTGAGTATC and CCTGGGATCTTACAATTAACA) and cloned Eco RI/Not I into pUASt-attB (). Constructs were injected, and transgenic flies were generated using targeted integration. All embryo injections were performed by Bestgene Inc. C57BL/6J female mice were bred in-house in the animal facility at the Max Planck Institute of Immunobiology and Epigenetics (MPI-IE) in individually ventilated cages and euthanized by cervical dislocation at 8 weeks of age according to German guidelines. Animal procedures were performed according to the protocols approved by the German authorities and the Regierungspräsidium Freiburg [the sacrificing of animals for scientific purposes according to §4 (3) of the German Animal Protection Act].

Antibodies and protein detection

For Western blots, rabbit anti-histone H3 (Abcam, Ab1791; RRID:AB_302613), rabbit anti-Staufen (), and peroxidase-conjugated mouse anti-FLAG (Sigma-Aldrich, A8592; RRID:AB_439702) were used at concentrations 1:10,000, 1:2000, and 1:10,000, respectively. Secondary peroxidase-conjugated anti-rabbit antibody (Cell Signaling Technology, #7074; RRID:AB_2099233) was used at 1:2000. For immunohistochemistry, detection was carried out with primary antibodies at concentrations 1:2000 [rabbit anti-Staufen; mouse anti-FMRP; Abcam, ab10299; RRID:AB_297038; rabbit anti-Tral ()], 1:1000 [mouse anti-Cnx99A; Developmental Studies Hybridoma Bank (DSHB), catalog no. Cnx99A 6-2-1; RRID:AB_2722011], 1:100 (mouse anti-MYC; Invitrogen MA1-21316-D550; RRID:AB_2536993; mouse anti-V5; Invitrogen, 37-7500-A488; RRID:AB_2610630; rabbit anti-Staufen1; Abcam, ab73478; RRID:AB_1641030), 1:200 [rabbit anti-Imp ()], 1:500 [guinea pig anti-MAP2 (microtubule-associated protein 2); Synaptic Systems, 188 004; RRID:AB_2138181], and 5 μg/ml (mouse anti-Atox1; DSHB, catalog no. MMC-Atox1-2E6; RRID:AB_2618262; mouse anti-Shab; DSHB, catalog no. K89/34; RRID:AB_2877280; mouse anti-USP14; DSHB, catalog no. AFFN-USP14-9H6; mouse anti-BicD; DSHB, catalog no. anti-Bicaudal-D 1B11; RRID:AB_528102). Mouse anti-HuB (Proteintech, 67097-1-Ig; RRID:AB_2882402), mouse anti-HuC (Santa Cruz Biotechnology, sc-515624), and mouse anti-HuD (Santa Cruz Biotechnology, sc-28299; RRID:AB_627765) antibodies were used at 5 μg/ml. Fluorophore-conjugated secondary antibodies (Invitrogen) were used at concentrations 1:500 (Drosophila brains) or 1:750 (mouse primary cultures).

Immunofluorescence

Drosophila adult or larval brains were dissected in phosphate-buffered saline (PBS) and fixed in 4% paraformaldehyde (PFA) in 0.3% PBS–Triton X-100 for 20 min at room temperature. Tissue was permeabilized and rehydrated in 0.3% PBS–Triton X-100 and blocked in blocking solution [5% bovine serum albumin (BSA; Sigma-Aldrich, A9647) and 0.3% PBS–Triton X-100]. Embryos were dechorionated before fixation, then devitellinized, and rehydrated before blocking. Incubation with primary antibodies was carried out in blocking solution for 36 hours at 4°C. Brains were rinsed with 0.3% PBS–Triton X-100, washed three times for 20 min with 0.3% PBS–Triton X-100, and incubated with secondary antibodies in blocking solution for 2.5 hours at room temperature. Brains were rinsed and then washed three times for 20 min with 0.3% PBS–Triton X-100, counterstained with 4′,6-diamidino-2-phenylindole (DAPI) and mounted with VECTASHIELD antifade mounting medium (Vector Laboratories, H-1000). Mouse primary cells were fixed in 4% PFA in PBS, permeabilized with 0.2% PBS–Triton X-100, and blocked with 3% BSA for 1.5 hours at room temperature. Incubation with primary antibodies was carried out overnight at 4°C followed by secondary antibodies for 1 hour at room temperature. Cells were counterstained with DAPI and mounted with VECTASHIELD antifade mounting medium. Where appropriate, specimens with swapped fluorophores or single-stained controls were prepared to ensure minimal signal bleed-through with the current fluorescence filter settings. All images were acquired with a Zeiss LSM 880 confocal microscope with Fast Airyscan in a sequential scanning mode.

Image analysis

Image analysis used Fiji software () with Java 8u181. ZEN (black edition) Imaging Software (Zeiss) was used for image processing. Colocalization between granule components was scored manually by counting the n-number of granules or cells showing signal overlap.

Western blot sample preparation

Adult flies were decapitated, and heads were homogenized in 1× PBS supplemented with protease inhibitor cocktail (Roche, 11873580001). Samples were spun down at 10,000g for 3 min to remove the debris. Head lysates were mixed with 4× NuPAGE sample buffer (Invitrogen, NP0007) supplemented with 0.2 M dithiothreitol (DTT) and boiled for 5 min at 95°C.

RNA purification and RNA-seq

For sequencing of whole head tissue, flies were decapitated and heads were homogenized in QIAzol Lysis Reagent (QIAGEN, 79306). For sequencing of purified neuronal granules, RNA was isolated using TRIzol LS Reagent (Ambion, 10296028). For 3′-seq, RNA samples were prepared as described in the xRIP–quantitative polymerase chain reaction (xRIP-qPCR) input preparation protocol [see “Cross-linking RNA IP followed by reverse transcription q-PCR (xRIP-qPCR)” in Materials and Methods for details]. For all experiments, the total RNA was extracted according to the manufacturer’s instructions and RNA integrity was analyzed using the 2100 Bioanalyzer (Agilent Technologies). Libraries for total RNA-seq were prepared with 725 ng (Fig. 2, A and B) or 100 ng (Fig. 5A) of total RNA using the TruSeq Stranded total RNA (Gold) (Illumina) according to the manufacturer’s instructions. 3′-seq libraries were prepared with 10 ng of total RNA using the QuantSeq 3′-Seq Library Prep Kit REV (Lexogen) according to the manufacturer’s instructions. Paired-end sequencing was performed using the NovaSeq 6000 platform (Illumina) and 51-bp reads (Fig. 2, A and B), 151-bp reads (fig. S2A), or 101-bp reads (Fig. 5A).

Poly(A) selection

One microgram of purified total Drosophila head RNA was used. Poly(A) enrichment was performed with either Oligo d(T)25 magnetic beads [New England Biolabs (NEB), S1419S] or Oligo dT beads included in the TruSeq Stranded mRNA Library Prep Kit (Illumina, 20020594) according to the manufacturer’s instructions.

RNA antisense purification

RNase-free, high-performance liquid chromatography (HPLC)–purified, 5′biotin-conjugated DNA probes with Internal Spacer 18, targeting mimi-RC, were obtained from Integrated DNA Technologies (sequences in table S4). Head powder was prepared from w flies. For each sample, 700 mg of head powder was UV-irradiated six times in a Bio-Link BLX 312 crosslinker at 300 mJ/cm2. Non–cross-linked powder was prepared as a control. Head powder was homogenized in 6 ml of lysis buffer [50 mM tris (pH 7), 500 mM LiCl, 10 mM EDTA, 5 mM DTT, and 2% lithium dodecyl sulfate (LDS)]. Unless specified otherwise, all buffers were supplemented with protease inhibitor cocktail (Roche, 11873580001) and RiboLock RNase inhibitor (Thermo Fisher Scientific, EO0384). Homogenized samples were incubated at 65°C for 2 min at 1100 rpm. Samples were transferred to room temperature for 3 min. After centrifugation (13,750g, 4°C, 10 min), the top layer and the pellet were discarded. The lower layer containing soluble cellular material was carefully extracted and precleared with high-capacity streptavidin agarose resin (Thermo Fisher Scientific, 20357) at 4°C for 45 min. For hybridization, 1.6 nM biotin-conjugated probes (two sets of probes with two different probes each) were used. Probes were hybridized for 2 min at 65°C and 1100 rpm and then gradually cooled down to 20°C. Samples were diluted 1:1 in lysis buffer without LDS and incubated for 1 hour at 4°C with MyOne Streptavidin C1 Dynabeads (Invitrogen, 65002). Beads were washed once in modified lysis buffer (supplemented with 1% LDS and 15 mM DTT) for 5 min at 4°C, twice with CHIRP (chromatin isolation via RNA precipitation) buffer (2× SSC, 0.5% SDS, and 15 mM DTT) for 5 min at room temperature, once in urea buffer (1 M urea, 0.5 mM DTT, and no protease or RNase inhibitors) for 5 min at room temperature, and once in ammonium bicarbonate (ABC) buffer (200 mM ABC, 1 mM EDTA, and 15 mM DTT) for 30 min at 37°C and 600 rpm and quickly rinsed with 200 mM ABC with 0.01% RapiGest SF surfactant (Waters). Proteins were eluted in 66 μl of elution buffer [50 mM ABC, 0.5 mM MgCl2, and 0.01% RapiGest, supplemented with 0.05 U of RNaseI (Thermo Fisher Scientific, EN0601), 0.05 μg of RNaseA (Thermo Fisher Scientific, EN0531), and 13 U of RNaseH (NEB, M0297S)] for 1 hour at 37°C and 900 rpm. The eluates were treated with 0.25 U of Benzonase (Millipore, 70664) for 1 hour at 37°C and 900 rpm and adjusted to 10 mM tris-HCl (pH 7.9) and 0.1% SDS before paramagnetic bead–based single pot, solid phase–enhanced sample preparation (SP3) purification as described in (). Peptides were eluted by addition of ultra HPLC grade water (Pierce) in two steps and a final step using 0.1% trifluoroacetic acid pooling the eluates. Pooled eluates were concentrated in vacuo and resuspended in 0.1% formic acid before nano–liquid chromatography–tandem MS (nanoLC-MS).

Co-IP of FNE and RBP9

Flies (5 g) of the genotype fne were anaesthetized and blended with five short pulses in a kitchen blender (Philips ProBlend 6) in 400 ml of PBS. The blended mixture was applied onto a stack of sieves with the grid sizes of 710, 425, and 355 μm and subjected to separation with pressurized water. Head tissue was collected on a 100-μm mesh and briefly spun to remove excess liquid. All subsequent steps were performed at 4°C. Heads were homogenized with 12 gentle strokes in a KONTES Tissue Grinder (VWR) (loose pestle) in 3 ml of ice-cold lysis buffer [50 mM Hepes-KOH (pH 7.5), 120 mM NaCl, 2 mM MgCl2, and 0.1% NP-40 with protease inhibitor cocktail (Roche, 11873580001)]. After centrifugation through a 40-μm strainer at 400g for 2 min, the supernatant was collected and centrifuged at 12,000g for 5 min. The middle layer was carefully transferred to a new tube, avoiding the pellet and the top layer. Centrifugation was repeated, and the resulting middle layer (input) was split into 330 μl of samples. Samples were either supplemented with 40 μg of RNaseA (Thermo Fisher Scientific, EN0531) and 25 U of Benzonase (Millipore, 70664) (RNase+) or with RiboLock RNase inhibitor (Thermo Fisher Scientific, EO0384) (RNase−). Samples were precleared with 50 μl of Pierce control agarose resin (Thermo Fisher Scientific, 26150) for 1 hour. Samples were loaded onto 40 μl of prewashed V5-trap magnetic particles (ChromoTek, catalog no. v5td) or MYC-trap magnetic agarose (ChromoTek, catalog no. ytma) and incubated for 1.5 hours. As a control, one RNase− sample was loaded onto 40 μl of anti-hemagglutinin beads (Thermo Fisher Scientific, 88836). After IP, beads were washed three times for 5 min with 1 ml of ice-cold lysis buffer. After the third wash, samples were transferred into a new tube, the wash buffer was removed, and beads were resuspended in 30 μl of 1× NuPAGE sample buffer (Invitrogen, NP0007) supplemented with 0.2 M DTT and boiled for 5 min at 95°C. Eluted proteins were separated on a Western blot together with the input sample.

Cross-linking RNA IP followed by reverse transcription qPCR (xRIP-qPCR)

One-hundred milligrams of UV–cross-linked head powder (6 × 300 mJ/cm2) was homogenized in 1 ml of lysis buffer [200 mM NaCl, 50 mM Hepes-KOH (pH 7.9), 1 mM EDTA, 0.5 mM EGTA, and 0.5 mM DTT] containing 0.2% Triton X-100. Unless specified otherwise, all buffers were supplemented with a protease inhibitor cocktail (Roche, 11873580001) and RiboLock RNase inhibitor (Thermo Fisher Scientific, EO0384). After homogenization, detergents were added to a final concentration of 0.5% SDS and 0.5% Na-deoxycholate, and the samples were incubated 5 min on ice. The homogenate was centrifuged twice at 15,000g for 10 min at 4°C. Ionic detergents were quenched with 1% NP-40. The processed lysate (750 μl; input) was incubated with 40 μl of anti-FLAG M2 magnetic beads (Invitrogen, M8823) for 1.5 hours at 4°C. Beads were rinsed with lysis buffer containing 0.1% SDS and 0.1% Na-deoxycholate and washed three times for 5 min at 4°C with lysis buffer containing 0.1% SDS, 0.1% Na-deoxycholate, and 1% Triton X-100 and twice with lithium chloride buffer [350 mM LiCl, 50 mM Hepes-KOH (pH 7.5), 1 mM EDTA, 1% NP-40, and 0.7% Na-deoxycholate] for 5 min at 4°C. Cross-linked protein-RNA complexes were eluted with 120 μl of elution buffer [10 mM tris-HCl (pH 7.4), 150 mM NaCl, 0.1% Triton X-100, and no protease inhibitor] containing FLAG peptide (0.2 mg/ml; Sigma-Aldrich, F3290) for 1 hour at 4°C. Eluates and corresponding inputs were subjected to proteinase K (Ambion, AM2546) treatment for 30 min at 50°C and 1100 rpm. RNA was purified with TRIzol LS Reagent (Ambion, 10296028) according to the manufacturer’s instructions. Equal volumes of total RNA from input and IP samples were used for quantitative reverse transcription PCR (RT-qPCR).

Sequential IF–fluorescence in situ hybridization

Stellaris fluorescence in situ hybridization (FISH) probes with TAMRA dye targeting the mimi-RC coding sequence were used (BioCat GmbH). IF-FISH was performed as described in () with the following modifications. Brains were dissected in PBS; 0.5% PBS–Triton X-100 was used for the preparation of all buffers. Blocking solution contained 2.5% normal goat serum (Abcam, ab138478), 2.5% normal donkey serum (Abcam, ab138579), and BSA (2 mg/ml; Sigma-Aldrich, A9647). Before probe hybridization, brains were incubated overnight with primary antibodies diluted in blocking solution in the presence of Recombinant RNasin Ribonuclease Inhibitor (Promega, N2511). Brains were rinsed three times, washed three times for 15 min with 0.5% PBS–Triton X-100, and incubated for 2.5 hours with fluorescent secondary antibodies diluted in blocking solution. Brains were again rinsed three times and washed three times for 15 min with 0.5% PBS–Triton X-100. Probes were used at a concentration of 5 μM. Hybridization buffer was supplemented with Recombinant RNasin Ribonuclease Inhibitor. Brains were washed two additional times for 10 min with 2× SSC before they were mounted with VECTASHIELD antifade mounting medium (Vector Laboratories, H-1000) and imaged after 24 hours. Where appropriate, specimens with swapped fluorophores or single-stained controls were prepared to ensure minimal signal bleed-through with the current fluorescence filter settings. Images were acquired with a Zeiss LSM 880 confocal microscope with Fast Airyscan in a sequential scanning mode.

Reverse transcription qPCR

Unless specified otherwise, 500 ng of total RNA was used for RT-qPCR. RT used the iScript gDNA Clear cDNA Synthesis Kit (Bio-Rad). RT-qPCR was performed in a LightCycler 480 II instrument using FastStart SYBR Green Master (Roche). RT-qPCR primer sequences are listed in table S4.

Purification of native neuronal granules

Nine grams of flies was anesthetized and blended with five short pulses in a kitchen blender (Philips ProBlend 6) in 400 ml of PBS. The blended mixture was applied onto a stack of sieves with the grid sizes of 710, 425, and 355 μm and subjected to separation with pressurized water. Head tissue was collected on a 100-μm mesh and briefly spun to remove excess liquid. All steps of the protocol including centrifugations were performed on ice or at 4°C, and all buffers were supplemented with protease inhibitor cocktail (Roche, 11873580001) and RiboLock RNase inhibitor (Thermo Fisher Scientific, EO0384). In a KONTES Tissue Grinder (VWR), 20 strokes with the loose pestle were applied to the sample in 5 ml of lysis buffer [50 mM tris-HCl (pH 7.4), 1 mM EDTA, 150 mM NaCl, and 0.2% Triton X-100], followed by centrifugation through a 100-μm strainer (100g, 2 min). The flow-through was processed with 20 strokes using the tight pestle, passed 10 times through a 27-gauge needle, and poured through a 40-μm strainer to remove cuticular debris. The flow-through (tissue lysate) was centrifuged at 1000g for 5 min. The resulting supernatant was split in two and centrifuged again at 10,000g for 15 min. The pellet (granules 1) was either collected directly or washed by resuspension in 2 ml of 0.3% PBS–Triton X-100/protease inhibitor cocktail/RiboLock followed by centrifugation at 10,000g for 10 min (pellet = granules 2). For RNA analysis, both granule fractions were resuspended in lysis buffer without protease inhibitor and subjected to proteinase K (Ambion, AM2546) treatment in the presence of 0.5% SDS for 30 min at 50°C and 1100 rpm. Proteinase K treatment was also applied to tissue lysate samples. For proteome analysis, only the fraction granules 2 was resuspended in radioimmunoprecipitation assay buffer [50 mM tris-HCl (pH 8), 150 mM NaCl, 1% NP-40, 0.5% Na-deoxycholate, 0.1% SDS, and 5 mM TCEP], homogenized for 1 min with a mechanical Pellet Pestle Motor (KONTES), incubated for 15 min on ice, and analyzed by nanoLC-MS together with the tissue lysate sample, subsequent to SP3-assisted sample preparation.

RNA-seq analysis

RNA-seq data from this paper and publicly available data (, ) were analyzed using the RNA-seq module from the snakePipes package (). The Drosophila annotation from the Ensembl release 96 was used as a reference for mapping and read counting. Differential expression analysis was done using DESeq2 (). Differentially expressed genes were defined on the basis of an absolute log2 fold change cutoff of 0.5, base mean > 10, and adjusted P < 0.01 (Fig. 2A) or adjusted P < 0.05 (Fig. 5A). To identify RNAs specific for mimi granules, all genes with base mean > 10 were considered and the log2 fold change in granules versus tissue lysate fraction was calculated. To classify an RNA as mimi granules component, the following criteria were applied: (i) the difference of the log2 fold changes in granules versus tissue lysate fraction for wild type versus Δmimi had to be >0.5 or (ii) the log2 fold change in granules versus tissue lysate fraction had to be >0 and significant (P < 0.05) only in wild-type condition. For the final classification as mimi granule component, an RNA had to fulfill condition (i) or (ii) in the independent analysis of both granule fractions. For 3′-seq analysis, reads were trimmed to remove poly(A) stretches using fastp () and mapped to the dm6 genome using STAR v2.6.1b () with default parameters except “--sjdbOverhang 74 --limitBAMsortRAM 60000000000 --alignIntronMax 1.” To remove the signal that likely originates from internal priming, all poly(A) sites that overlap a strand-specific blacklist region containing all genomic positions with more than 70% As in a 10-bp upstream window were removed. Regions with high A density but within 250 bp of annotated transcription end sites were not blacklisted. Remaining single base pair poly(A) sites from all samples with a minimum coverage of 5 reads per sample were clustered together so that sites within 15 bp were merged to one poly(A) cluster.

Mass spectrometric acquisition

Nanoscale LC-MS analysis was done either on a Q Exactive Plus mass spectrometer coupled to an EASY-nLC 1200 nUHPLC (RAP) or a Q Exactive MS system interlinked to an EASY-nLC1000 nUHPLC (both Thermo Fisher Scientific) as described in () with modifications detailed below. For the RAP experiment, samples were injected twice (60-min nLC-MS method). The gradient was 5 min, 10%; 40 min, 40%; 4 min, 80%; at the flow rate 250 nl/min. This was followed by a “wash out step” from 80% B buffer at 5 min and a 5-min inverse gradient from 80 to 2% B buffer (flow rate of 450 nl/min). In case of the mimi granule study, tissue lysate and granule samples were measured in seven technical replicates using a 120-min nLC-MS method. The gradient was 5% at 0 min, 8% at 5 min, 35% at 90 min, 45% at 10 min, and 80% at 7 min (flow rate of 300 nl/min). This was followed by 5 min inverse gradient from 80 to 2% B buffer (flow rate of 500 nl/min). All measurements were carried out in data-dependent mode using the “sensitive method” ().

MS data analysis

MaxQuant version 1.6.14.0 using standard parameters was used to identify peptides and final protein identification role-up [both at 1% false discovery rate (FDR)]. MS raw data were searched simultaneously with the target-decoy standard settings against the Uniprot D. melanogaster database (Uniprot_reviewed+Trembl including canonical isoforms; downloaded on 5 August 2020) and an in-house curated FASTA file containing commonly observed contaminant proteins. The mimi granule study was further analyzed using intensity-based absolute quantification (iBAQ) values. Each protein from a given protein group was assigned a FlyBase ID and included in the analysis as an individual entry. After the analysis, duplicated FlyBase entries were removed. For each of the four conditions (wild-type tissue lysate, wild-type granules, Δmimi tissue lysate, and Δmimi granules), if a single replicate’s iBAQ value was reported missing, then the missing value was imputed on the basis of a nearest neighbor calculation. If two or three of three replicate iBAQ values were missing in a given condition, then these values were instead imputed with a low basement value of 8 to facilitate subsequent processing. Two subsequent filters were applied to remove unreliable proteins from the analysis before assessing for differential expression. A protein was removed if, (i) in wild-type samples, any granules-tissue lysate replicate pair showed missing values in both conditions or (ii) if, in either wild-type or Δmimi samples, a replicate where an iBAQ value was imputed on the basis of the nearest neighbor calculation showed an opposite trend in the pairwise comparison granules versus tissue lysate compared to the replicate with actual reported iBAQ value. To identify candidate mimi granule proteins, the relative log2 difference in differential expression between granules and tissue lysate samples in the wild-type relative to the Δmimi condition was calculated. iBAQ values were mean normalized using the R package caret () before differential expression was assessed. In cases where the expression was observed in both wild-type and Δmimi conditions, log2 fold change in the granules versus tissue lysate fraction in wild type had to be greater by at least 4.8 than the log2 fold change in Δmimi. In cases where no expression was observed in Δmimi samples, only a positive log2 fold change expression in granules versus tissue lysate was required in wild type. Proteins that met either of these criteria were considered granule associated. Because some proteins were identified on the basis of <2 unique peptides (as classified by MaxQuant), for proteins shown in Fig. 5F and fig. S5C and proteins indicated as such in tables S2 and S3, we manually aligned (BLASTp search against the D. melanogaster whole genome) all identified peptides, which resulted in substantially higher number of peptides that are indeed unique for a given gene. To investigate the effect of mimi granule loss on the global proteome, differential protein expression between wild-type and Δmimi tissue lysate samples was assessed. Missing iBAQ values were imputed using the nearest neighbor method and mean-normalized using the R package caret. A two-sample t test (two-tailed, equal variance) was run to assess for significant (P < 0.05) differential expression between wild-type and Δmimi samples (|log2 fold change| > 0.5).

For mimi-RAP experiments, the intersample relative abundance was determined using the MaxQuant MaxLFQ algorithm with enabling the match between runs option (matching time window of 0.5 min). Downstream analysis was carried out using an in-house modified R script using the DEP package as base (). Briefly, contaminants, reverse, and only identified by site entries were filtered out. At least two valid quantitation values in any group (bait or control) were required. Data were vsn (variance-stabilizing normalization)–transformed, and missing values were imputed by drawing values from a normal distribution (width of 0.5, downshift of 1.8). Statistical analysis was done using limma (with trend = TRUE), and the obtained P values were corrected for multiple hypotheses by Benjamini-Yekutieli. Differentially enriched proteins were classified by having an adjusted P ≤ 0.05 and a |log2 fold change| > 0.5 (UV–cross-linked sample versus non–cross-linked control).

GO enrichment analysis

GO enrichment analysis was performed using DAVID version 6.8 online server. Genes differentially expressed in Δmimi versus control and RNAs classified as mimi granule components were queried against the background of all RNAs expressed at base mean > 10. Proteins classified as granule components were queried against the background of proteins detected in either granule or tissue lysate fraction. Significant GO terms were denoted on the basis of P < 0.05 calculated using modified Fisher’s exact test (EASE score) with (Fig. 5D) or without (fig. S5, F and G) Benjamini-Hochberg adjustment for FDR correction.

Behavior assays

Survival assay

Flies eclosed within a 2-day window were collected and housed at a density of 25 males and 25 females per bottle. Flies were raised at 25°C and transferred into bottles with fresh food every 2 to 3 days. The number of dead and surviving flies was counted at each transfer.

Climbing assay

The climbing assay was performed as described in () with minor modifications. Flies enclosed within a 2-day window were collected and housed at a density of five males and five females per vial. Once a week, starting 3 to 5 days after eclosion, animals were anesthetized on a CO2 pad and transferred to an empty vial. After a 20-min recovery period, flies were tapped down to the bottom and the number of flies climbing above the 3 cm mark within the first 5 s after the tap was scored. For each of the 10 biological replicates, two taps separated by 1-min recovery period were performed.

Memory assay

All Drosophila strains were reared at 25°C and 40 to 50% humidity on a standard cornmeal agar food in 12-hour light:12-hour dark cycle. Three- to 5-day-old adult flies were used. Eighty to 100 flies were placed in a 25-ml vial containing standard food and a 20 mm–by–60 mm piece of filter paper for 14 to 22 hours before behavioral experiments. Odors used in all experiments were 4-methylcyclohexanol (MCH) and 3-octanol (OCT) diluted in mineral oil. An odor dilution of approximately 1:103 was used for all experiments (specifically, 7 ml of OCT or 17 ml of MCH in 8 ml of mineral oil). Experiments were performed at 23°C and 55 to 65% relative humidity. Aversive olfactory conditioning in the T-maze was conducted as previously described (, ). Groups of flies were exposed to the first odor for 1 min [the conditioned stimulus+ (CS+)] paired with 12 90-V electric shock pulses with 5-s intervals. Following 45 s of clean air, a second odor (the CS−) was presented for 1 min without shock. Memory was subsequently assessed by testing flies for their odor preference between the CS− and the CS+ in a T-maze (2 min). The performance index was calculated as the number of flies in the CS+ arm minus the number in the CS− arm divided by the total number of flies. MCH and OCT were alternately used as CS+ or CS−, and a single sample, or n, represents the average performance score from two reciprocally trained groups. Statistical analyses were performed in GraphPad Prism version 9 (two-sided unpaired t test).

Secondary structure modeling

The taxonomic neighborhood of D. melanogaster was screened to find potential homolog sequences. BLASTN (BLAST+ 2.12.0) () was run against the refseq_genomic database with search results restricted to organism of the Diptera order [National Center for Biotechnology Information (NCBI) taxonomy ID: 7147]. Fourteen candidate sequences for significant hits with full query coverage were extracted. The sequence-structure alignment tool LocARNA version 2.0.0RC8 () was used in iterative mode to filter candidate sequences for those with compatible sequence and secondary structure. Nine hits with negative alignment scores or mapping to unassembled scaffold regions were discarded. The multiple sequence-structure alignment of the remaining sequences was preprocessed with SelectSequencesFromMSA version 1.0.5 () predicted by RNAz (2.1) () to be of structural alignment quality and shows insignificant coding potential with RNAcode version 0.2 (). While some covariant base pairs, representing compensatory mutations to conserve the functional secondary structure, are present, their number is limited due to the phylogenetic scope of the detected potential homologs. RNApuzzler () and RNAalifold (), both integrated in the ViennaRNA suite version 2.4.13 (), were used to create a visualization of the consensus sequence and secondary structure and of the multiple sequence alignment itself.

Mouse primary cultures

Eight-week-old female C57BL/6J mice were used. The subventricular zone was dissected, and single-cell suspensions were prepared and cultured as previously described (). Cells were differentiated for 7 days in Dulbecco’s modified Eagle’s medium/F12 medium without glutamine (Gibco, 21331-020) containing 1× B27 supplement (Gibco, 17504-044), 8 mM Hepes (Gibco, 15630-056), 1× GlutaMAX I (Gibco, 35050-038), and antibiotics in 37°C and 5% CO2.

Data visualization

The R package ggplot2 or GraphPad Prism version 9 was used for data visualization. Any additional information required to analyze the data reported in this paper is available from V.H. upon request.

Quantification and statistical analysis

Statistical parameters and tests are reported in the respective figure legends. Statistical tests were performed using DESeq2 (), limma (), caret (), Fisher’s test in R, and GraphPad Prism version 9 for MacOS and t test in Microsoft Excel.

Sample details

In Fig. 1 (A and C), fig. S1 (A, B, and F), Fig. 2 (C and D), figs. S2 (G and K) and S3 (C and E), Fig. 4 (A and D), and figs. S4 (A, B, C, and E) and S5 (B and E), confocal images (IF and IF-FISH) were acquired from three brains (or embryos) per genotype in at least two independent experiments (flies) or from cells isolated from multiple mouse brains. One representative image is shown. Quantification of colocalization between different granule components was scored on the basis of the n-number of granules indicated in the respective figure legend. In Fig. 2 (A and B), the total RNA-seq was performed on 20 adult Drosophila heads (1-day-old flies, 10 males and 10 females) in five independent replicates (n = 5). In Fig. 3B, mimi RNA was antisense-purified from thousands of adult Drosophila heads in four replicates per condition. In Fig. 3D, tag-FNE, tag-RBP9, and tag-Stau proteins were purified from thousands of adult Drosophila heads in three (FNE and RBP9) or four (Stau) replicates per genotype. In Fig. 5 (A, C, and D) and fig. S5 (D and G), for sequencing of neuronal granule components, thousands of adult Drosophila heads were used. Tissue lysate samples were prepared and sequenced in three replicates per genotype. Neuronal granules were purified and sequenced in six replicates per genotype (three biological replicates and two granule fractions each). In Fig. 5 (B and F) and fig. S5 (C and F), for LC-MS analysis of neuronal granule components, thousands of adult Drosophila heads were used. Both tissue lysates and neuronal granule fractions were analyzed in three independent biological replicates per genotype. In Fig. 6A, memory assays were performed on 200 flies per genotype per replicate. In Fig. 6B, climbing assays were performed on 100 flies per genotype at time point. In Fig. 6C, survival assays were performed on 300 flies per genotype (control and Δmimi) or 80 flies (ΔfneΔrbp9).