The lncRNA Gas5 is an activity-responsive scaffold for mediating cAMP-dependent synaptic plasticity

Abstract

Changes in the transcriptome are critical in shaping the structural plasticity of neurons, which underpins learning and long-term memory storage. Here, we explored the effect of two opposing, plasticity-associated pathways—cAMP second-messenger signaling and metabotropic glutamate receptor (mGluR1 and mGluR5) signaling— on the transcriptome in hippocampal neurons, and how these pathways operate in distinct and coordinated manners to induce structural changes. Integration of transcriptome data and molecular pathway analysis identified central “hub” genes that were rapidly induced by cAMP and/or mGluR1/5 in hippocampal neurons. These included the long non-coding RNA (lncRNA) Gas5, whose expression was induced specifically by cAMP and which was targeted to dendrites by the kinesin motor protein KIF1A. In the dendrites, Gas5 interacted with various proteins and coding and non-coding RNAs associated with synaptic function and plasticity, and these interactions were altered by cAMP signaling. Notably, Gas5 interacted with the microRNA miR-26a-5p and sequestered it from several of its mRNA targets associated with neuronal function and whose translation was induced by cAMP. Gas5 was critical for excitatory synaptic transmission induced by cAMP, but not those induced by mGluR1/5. Furthermore, Gas5 deficiency impaired dendritic branching and synapse morphology, and Gas5 abundance was decreased in the hippocampus of a mouse model of Alzheimer’s disease. Together, these provide insight into the transcriptional networks involved in synaptic plasticity and a lncRNA interactome that mediates dendritically localized regulation of excitatory synaptic transmission and neuronal architecture.

Introduction

The transcriptional networks that emerge from signal transduction pathways are pivotal in determining the neuronal response to activation. These networks lead to specific alterations in neuronal architecture and connectivity, with lasting effects on cognitive processes such as learning and memory [1–3]. Broadly, these pathways can be classified as excitatory or inhibitory, reflecting whether the synaptic connectivity of a neuron is ultimately enhanced or diminished during various forms of learning [4–6]. Given that a neuron’s long-term synaptic capabilities are fundamentally shaped by its transcriptional networks, it is crucial to understand how these networks influence the genes involved in excitatory and inhibitory signaling, and how these genes interact to attenuate synaptic plasticity. Therefore, gaining insight into the specific components of these transcriptional networks and their roles in synaptic plasticity is essential.

Recent advances in high-throughput sequencing technologies and functional profiling methods have expanded our understanding of transcriptional networks by revealing a vast array of noncoding RNAs, ranging from small types [microRNA (miRNA), small nuclear RNA (snRNA), small nucleolar RNA (snoRNA)] to large types, such as long noncoding RNA (lncRNA) and pseudogenes. These discoveries have provided deeper insight into the complexity of transcriptional regulation, introducing noncoding RNAs as integral elements of these networks. Many of these noncoding RNAs have been found to play functional and mechanistic roles in neurobiological processes associated with learning, memory, and disorders leading to memory impairment [7, 8]. Consequently, multiple miRNAs and lncRNAs have been identified as crucial regulators of gene expression, with the ability to enhance or suppress the expression of protein-coding genes [9–11]. Thus, when mapping gene networks influenced by excitatory and inhibitory signaling, it is imperative to consider the intricate interplay between coding and noncoding RNAs.

In this study, we examined the transcriptional responses in hippocampal neurons to two distinct signaling pathways relevant to learning and long-term memory storage. Specifically, we utilized potent activators of excitatory and inhibitory signaling: Forskolin—which activates adenylyl cyclase to initiate the cAMP signaling pathway, a cascade well-known for its association with long-term potentiation and enhanced learning—and dihydroxyphenylglycine (DHPG), , a potent agonist of group I metabotropic glutamate receptors (mGluR1 and mGluR5) that induce long-term depression of synaptic transmission [12, 13]. We analyzed the differential regulation of mRNAs, lncRNAs, and miRNAs under cAMP and mGluR1/5 signaling conditions, providing insights into the complex interactions between coding and noncoding elements within the transcriptomes.

We further explored the relationships between lncRNA and mRNA cis-expression patterns, constructed co-expression networks of differentially expressed genes to identify key players in these pathways, and conducted a meta-analysis of relevant in vivo datasets. By integrating this data, we uncovered multiple regulatory RNA interaction networks under cAMP and mGluR1/5 conditions. We then functionally interrogated one of these networks by silencing the central lncRNA, Gas5, revealing that this lncRNA supports the expression of cAMP-mediated RNAs by directly interfering with the brain-enriched miR-26a-5p. Transcriptomic and proteomic profiling of Gas5 complexes suggested that Gas5 acts as a scaffold for various RNAs and protein partners involved in synaptic function, suggesting that Gas5 constrains excitatory synaptic transmission, dendritic arborization, and synapse morphology.

Results

cAMP and mGluR1/5 signaling in hippocampal neurons elicit unique transcriptomic signatures

To determine whether different learning-related signaling pathways generate distinct or overlapping transcriptional changes in hippocampal neurons, we analyzed the coding and noncoding transcriptomes induced by cAMP and mGluR1/5 signaling. Specifically, we cultured primary neurons derived from the hippocampi of CD1 pups. At DIV 18, these neurons were treated with forskolin or DHPG for 30 minutes to activate cAMP or mGluR1/5 signaling, respectively, with DMSO serving as a control. After treatment, RNA was extracted from the neurons and subjected to both total and small RNA sequencing (RNA-seq) from the same samples (Fig. 1A).

Figure 1: cAMP and mGluR1/5 signaling in hippocampal neurons generate distinct molecular signatures.

(A) Schematic demonstrating the experimental paradigm of RNA preparation from mouse primary hippocampal neurons (representative picture is shown with Map2 staining) and sequencing. (B) Principal component analysis of total RNA-seq TPM values for all treatment groups. FSK, forskolin. (C and D) Volcano plot indicating DEseq-identified significant genes in neurons treated with forskolin vs DMSO (C) and in those treated with DHPG vs DMSO (D). N=4-6 per group from two experiments. p_adj < 0.05 criteria were used to control FDR at 5%. (E to G) Venn diagram showing the number of overlapping genes (E) and miRNAs (F and G) from the seq data described in (A to D) in the indicated comparisons. p <0.05, DEseq2. (H and I) Network plots of top 5 DHPG-upregulated (H) and FSK-upregulated (I) miRNAs and their putative targets, as determined by starBase and TargetScan. Blue indicates the miRNA name; green and red indicate the targets are significantly upregulated or downregulated, respectively, in the total RNA-seq for each group.

Principal component analysis of the batch-corrected data revealed that samples clustered closely according to their pharmacological treatment (Fig. 1B). To validate our RNA-seq findings, we selected well-characterized immediate early genes (IEGs) for qRT-PCR. Both RNA-seq and qRT-PCR confirmed forskolin- and DHPG-dependent changes in the expression of IEGs such as Arc, Egr2, and cFos (fig. S1, A to C; data file S1). In contrast, Syt4 expression remained unchanged between groups in both RNA-seq and qRT-PCR analyses (fig. S1D), highlighting the reproducibility of gene expression data across different methods.

Given the overlap in validated IEGs between cAMP and mGluR1/5 signaling (fig. S1, B to D), we hypothesized that these pathways might attenuate largely common gene sets but with differing magnitudes. Differential expression analysis revealed that forskolin treatment resulted in 1,438 differentially expressed genes compared to DMSO (2.89% of all annotated genes) (Fig. 1C, and data file S1), whereas DHPG treatment led to 239 differentially expressed genes (0.48% of all genes) (Fig. 1D, and data file S1). Intriguingly, a comparison between forskolin and DHPG treatments identified 5,857 differentially expressed genes (11.6% of all genes) (Fig. 1E), of which only 14 genes were upregulated by forskolin but downregulated by DHPG. Furthermore, when examining genes upregulated or downregulated relative to DMSO, only 12 and 9 genes overlapped, respectively (Fig. 1E).

Gene ontology analysis indicated that genes predominantly attenuated by forskolin were associated with synaptic function, RNA processing, and catabolic processes (fig. S1, E and G; data file S1), whereas genes attenuated by DHPG are involved in cilium assembly, microtubule formation, and respiratory development (fig. S1, F and H, and data file S1). Collectively, these findings suggest that cAMP and mGluR1/5 signaling pathways do not attenuate largely common gene sets but instead elicit unique transcriptomic signatures.

To further explore the relevance of cAMP signaling in cultured hippocampal neurons, we performed a meta-analysis of RNA-seq data from hippocampal slices undergoing forskolin-induced LTP at 30, 60, and 120 minutes [14] (fig. S2A). Comparing differentially expressed genes between the LTP and our forskolin datasets, we observed an increasing overlap over time: 14 genes at 30 minutes LTP (fig. S2B), 22 genes at 60 minutes LTP (fig. S2C), and 67 genes at 120 minutes LTP (fig. S2D). GO analysis revealed that these 67 overlapping genes are involved in RNA polymerase II-associated transcription factor activity, suggesting that early cAMP signaling activates a positive feedback loop of gene expression (fig. S2E).

cAMP and mGluR1/5 signaling attenuate miRNA-targeting of specific mRNAs

To gain a comprehensive understanding of the transcriptional networks attenuated by cAMP and mGluR1/5 signaling, we next focused on the expression of microRNAs (miRNAs). miRNAs are small, endogenous non-coding RNAs that attenuate post-transcriptional processes by directly interacting with mRNAs, leading to their degradation via the RISC complex [15]. These miRNAs have been implicated in various signaling processes in the brain, as well as in cognitive functions [16, 17]. We hypothesized that specific miRNAs could mediate the transcriptional dynamics induced by cAMP and mGluR1/5 signaling. To test this, we profiled differential miRNA expression from the same RNA preparations used in our previous analyses.

Our results revealed that forskolin treatment led to the upregulation of 32 miRNAs and the downregulation of 12 miRNAs (Fig. 1F, and data file S2). Similarly, DHPG treatment resulted in the upregulation of 46 miRNAs and the downregulation of 57 miRNAs (Fig. 1G, and data file S2). To further elucidate the interactions between the most differentially upregulated miRNAs and their respective targets, we employed a stringent approach by integrating two databases starBase and TargetScan, [18] and [19]. We identified miRNA-mRNA interactions only when the targets overlapped between these databases. Using Cytoscape, we then constructed interaction networks to visualize these relationships. Among the top 5 miRNAs upregulated by forskolin treatment, we identified 63 upregulated and 26 downregulated mRNA targets (Fig. 1I). In contrast, among the top 5 miRNAs upregulated by DHPG treatment, we found 20 downregulated mRNA targets and 1 upregulated mRNA target (Fig. 1H). The fact that many of these miRNAs share overlapping targets suggests a complex interplay between these miRNAs and their targets, attenuated by forskolin- and DHPG-induced signaling.

cAMP signaling activates multiple lncRNA and mRNA cis expression patterns

To further elucidate the role of long noncoding RNAs (lncRNAs) in our transcriptional network, we examined their expression patterns alongside those of mRNAs. The mammalian genome encodes thousands of lncRNAs, which are not translated but serve essential regulatory functions in cellular processes [20, 21]. These lncRNAs have been linked to synaptic plasticity and learning processes [22]. Many lncRNAs are localized in the nucleus, where they function as transcriptional regulators [23]. We hypothesized that expression patterns of lncRNAs might correlate with those of mRNAs located near their transcriptional loci (cis relationships) in the signaling pathways we studied. Using the BiomaRt Bioconductor package and UCSC Genome Browser, we cross-referenced lncRNAs with mRNAs within 100 kb. Although no cis relationships were identified for mGluR1/5 signaling, our analysis of the cAMP dataset revealed multiple cis relationships, including bi-directional, antisense, convergent, sense intronic, and intergenic co-transcription (fig. S3, A to E). Additionally, we observed two types of interfering cis patterns: intergenic and antisense interference (fig. S3, F and G). These results highlight the complex interactions within the transcriptional networks attenuated by forskolin and DHPG.

cAMP-PKA signaling-dependent genes are co-regulated by their biological function

Given the distinct transcriptional signatures elicited by cAMP and mGluR1/5 signaling, we next investigated the interactions of these genes within co-transcriptional frameworks. To achieve this, we used the multiscale-embedded gene co-expression network analysis (MEGENA) algorithm [24]. This method enables the identification of overlapping gene modules by performing planarity testing on RNA-seq data. MEGENA has been previously utilized to profile brain cell types [25] and aging brain tissue [26], revealing meaningful co-expression patterns and identifying novel gene targets.

For the cAMP-regulated genes, we analyzed data from 880 upregulated genes. Of these, 592 genes passed planarity testing, resulting in the identification of 32 gene modules (fig. S4, A and D; data file S3). One of the most statistically significant modules, C1_8, contained two hub genes, Eif3j2 and Atp11b (fig. S4D), which encode a translation initiation factor and an ATPase, respectively. The module with the largest number of cAMP-regulated genes from the planar network, C1_5, featured two hub genes—Zfp182 and Zcchc7—encoding Zinc finger proteins that have not been previously studied in the context of cAMP signaling. The C1_5 network comprised a mix of mRNAs, lncRNAs, and pseudogenes, suggesting an intricate interplay between coding and noncoding elements within this module (fig. S4B, and data file S3). Notably, gene ontology analysis of the C1_5 module revealed that the genes in this network are primarily involved in RNA processing and transport. This indicates that cAMP-induced gene products regulating these functions are coordinated in a precise manner (fig. S4C, and data file S3). Given the importance of RNA processing and transport in learning-related signaling pathways [27, 28], these findings suggest that cAMP signaling orchestrates the temporal expression of genes involved in these processes. The central role of two previously uncharacterized Zinc finger proteins within this network highlights their potential importance to these biological functions.

For mGluR1/5 signaling, we analyzed data from 41 upregulated genes, of which 35 passed the planarity testing, resulting in two distinct modules (fig. S4E). In module C1_2, the central hub genes Emd and Copg2 encode a cytoskeletal binding protein and a Golgi-to-ER transport protein, respectively. Module C1_3 featured Gm26493 as its central hub gene, which codes for a snoRNA previously unassociated with mGluR1/5 signaling (fig. S4F).

lncRNAs compete with miRNAs for the same target mRNAs under cAMP and mGluR1/5 signaling

We observed that most mRNA targets for the top 5 miRNAs upregulated by forskolin were also upregulated (Fig. 1I and fig. S8A). This suggests that mRNAs, which would typically be downregulated by these miRNAs, might be rescued by cAMP signaling. It is well established that cytoplasmic lncRNAs can act as competing endogenous RNAs (ceRNAs), thereby sequestering miRNAs away from their target mRNAs [29]. To investigate this potential mechanism, we profiled differentially expressed (DE) lncRNAs and their associated miRNA targets using the databases starBase and TargetScan. We identified 65 lncRNAs differentially expressed under cAMP conditions that bind to miRNAs with similar expression patterns. Under mGluR1/5 signaling, we found 8 such lncRNAs, with 2 lncRNAs common to both signaling pathways (Fig. 2A). We constructed miRNA/lncRNA/mRNA interaction networks for two of the DHPG-regulated lncRNAs, Gm38182 and Pvt1 (Fig. 2, B and C). In these networks, most mRNA targets of the miRNAs were downregulated, suggesting that these lncRNAs may not function as ceRNAs under mGluR1/5 signaling. Additionally, we analyzed the lncRNA MIRG, which was upregulated in response to both cAMP and mGluR1/5 signaling. We identified one putative miRNA interactor, miR-3086-5p, which was significantly upregulated in both pathways (Fig. 2D). This miRNA had 16 known mRNA targets that were also upregulated by cAMP signaling, indicating that although MIRG is upregulated in both signaling contexts, it may only act as a ceRNA in cAMP signaling.

Figure 2: lncRNAs compete with miRNAs for the same target mRNAs under cAMP and mGluR1/5 signaling.

(A) Venn diagrams indicating differentially expressed lncRNAs in forskolin- or DHPG-treated mouse primary hippocampal neurons that target differentially expressed miRNAs in their respective datasets (as determined by genes detected in both starBase and TargetScan databases) (B and C) lncRNA/miRNA/mRNA network for Gm38182 (B) and Pvt1 (C) under mGluR1/5 signaling. (D and E) lncRNA/miRNA/mRNA network for Mirg (D) and Gas5 (E) under cAMP signaling. Network analysis and visualization were performed using Cytoscape; red colored genes are downregulated in the respective treatment group, and green colored genes are upregulated. Up and down arrows indicate the direction of regulation of the associated miRNA or lncRNA in the respective treatment group.

Further analysis of the 63 lncRNAs exclusively upregulated by cAMP signaling revealed the lncRNA Gas5, which is robustly upregulated under cAMP signaling (fig. S8B). Cross referencing Gas5 with the starBase and TargetScan databases unearthed two miRNAs that bind this lncRNA: miR-26a-5p [30] and miR-9-3p [31], both of which are enriched in the brain and involved in neuronal and cognitive functions. Notably, Gas5 has been experimentally shown to bind to miR-26a-5p [32]. Similar to Mirg, we observed that most mRNAs targeted by both Gas5 and these miRNAs were also upregulated in response to cAMP signaling (Fig. 2E and fig. S8A). This suggests that Gas5 may act as a ceRNA in hippocampal neurons during cAMP signaling. To further elucidate the transcriptional regulation of Gas5 by cAMP, we examined its expression over time following forskolin treatment. We treated neurons with forskolin for 30 minutes, 1 hour, 3 hours, and 6 hours and found that Gas5 expression peaked at 3 hours before returning to near basal levels at 6 hours (Fig. 3, A and B; data file S3). These results indicate that Gas5 may function as an immediate-early response lncRNA in hippocampal neurons.

Figure 3: lncRNA Gas5 is activated by cAMP signaling and localized to dendrites.

(A) DESeq2-generated normalized counts of Gas5 expression in forskolin (FSK)- and DMSO-treated mouse primary hippocampal neurons. (B) qRT-PCR time course of forskolin-dependent Gas5 expression. N=4-6 neurons per group from three independent experiments. *p <0.05 compared to vehicle, by one-way ANOVA followed by Dunnett’s test. (C) DIV 16 hippocampal neurons were fixed and processed for combined FISH and ICC using Quasar fluorophore-labelled redundant probes. Cells were immunostained with actin to label processes and DAPI to visualize the nucleus. N=12-15 neurons per group from two independent experiments. Scale bars: 10 μm, inset 3 μm. (D) Scatter plots quantifying Gas5 puncta distribution in nucleus, cell body, and distal processes. (E) Scatter plots quantifying Gas5 puncta distance from the cell body in distal processes. N=11-15 neurons per group from 3 independent experiments. (F) Experimental schematic for KIF1A KD in primary neurons followed by qRT-PCR for Gas5 in synaptoneurosome or combined ICC-FISH. (G) Bar graph showing relative expression of Gas5 in synaptoneurosomes (n=4 per group from two independent experiments) after knockdown of the indicated KIF. Data were analyzed by unpaired student’s t-test, *p<0.05. Error bars represent SEM. (H) DIV 16 hippocampal neurons were fixed and processed for combined FISH and ICC after KIF1A knockdown. Cells were immunostained with actin to label processes and DAPI to visualize the nucleus. N=12-15 neurons per group from two independent experiments. Scale bar: 10 mm, inset 3 mm. (I) Scatter plots quantifying Gas5 puncta distribution in the cell body and distal processes of neurons described in (H). Data are mean ± SEM from n=12-15 neurons per group from two independent experiments, analyzed by two-way ANOVA followed by Dunnett’s test, **p<0.01

Dendritic localization of Gas5 is mediated by the molecular motor protein KIF1A

To assess the function of Gas5, we examined its subcellular localization. Fluorescence in situ hybridization (FISH) analysis using DIG-labeled riboprobes, combined with counterstaining for MAP2 (a dendrite marker), revealed that Gas5 is present in both the cell body and dendrites (fig. S6C). The dendritic localization suggests a potential role in modulating synaptic function. This finding is consistent with recent research demonstrating Gas5 localization to dendritic spines following fear conditioning [67]. To further elucidate Gas5’s subcellular distribution, we designed 25 redundant fluorescent probes spanning the Gas5 lncRNA. Combined FISH and immunocytochemistry (ICC) showed that Gas5 is distributed across the nucleus, cell body, and distal processes as globular puncta (Fig. 3C). Quantitative analysis of the puncta distribution revealed that Gas5 is enriched in the cell body, compared to distal processes and the nucleus (Fig. 3D). Notably, Gas5 puncta varied in distance from the cell body, with most located within 1000 μm but some extending as far as 4500 μm (Fig. 3E). This observed dendritic localization prompted us to investigate the mechanisms underlying Gas5’s subcellular distribution.

Our previous work and that of other groups have highlighted the role of specific kinesin family proteins (KIFs) in mediating RNA localization in neurons [51, 65, 66]. Notably, we have shown that the localization of lncRNA ADEPTR depends on KIF2A [51] and that SLAMR lncRNA requires KIF5C for its localization [70]. Therefore, we investigated whether KIFs might mediate the dendritic localization of Gas5. We used SMARTpool siRNAs to knock down a subset of KIFs in primary hippocampal neurons and assessed Gas5 abundance in synaptoneurosomes. Among the four KIFs tested (KIF1A, KIF2A, KIF5C, and KIF16B), KIF1A, KIF2A, and KIF5C siRNAs elicited significant knockdown levels of their target (fig. S7F and data file S8) compared to scrambled control (siNC) after 48 hours of transfection. Subsequent analysis of Gas5 levels in isolated synaptoneurosomes revealed that knockdown of KIF1A led to a significant reduction in Gas5 abundance (Fig. 3, F and G; data file S8). To confirm the role of KIF1A in Gas5 localization, we performed FISH analysis, which demonstrated that KIF1A knockdown significantly decreased Gas5 puncta in both the cell body and dendrites (Fig. 3I and data file S3). Collectively, these results indicate that KIF1A is crucial for the proper subcellular localization of Gas5.

Gas5 interacts with miR-26a-5p and is required for cAMP-dependent stability of its targets

We next investigated whether the expression of miR-26a-5p and miR-9-3p, along with their target mRNAs, depends on Gas5 expression. To explore this, we utilized locked nucleic acid (LNA) gapmers designed to specifically target and degrade Gas5 via RNAse H. We transfected these gapmers, either individually or in combination, along with a scrambled control, into neurons for 72 hours. We then assessed cAMP-dependent gene expression changes. Our data confirmed that the gapmers effectively suppressed the cAMP-induced increase in Gas5 expression (Fig. 4A and data file S3).

Figure 4: Gas5 interacts with RNAs and proteins involved in synapse function.

(A) Mouse primary hippocampal neurons were transfected with scrambled or one of three different gapmers targeting Gas5 to test their efficacy. Data are from N=3 wells per group from two independent experiments, analyzed by one-way ANOVA followed by Dunnett’s test, *p<0.05. (B) Primary hippocampal neurons were transfected with scrambled or one of three different gapmers targeting Gas5 for 72 hours, then treated with forskolin for 30 min prior to RNA isolation and qRT-PCR. Data are from n=3 wells per group from three independent experiments, analyzed by one-way ANOVA followed by Dunnett’s test, *p<0.05. (C) Primary hippocampal neurons were treated with forskolin for 30 min prior to RNA isolation and qRT-PCR to quantify levels of mature miR-26a-5p by stem loop qRT-PCR. Data are from N=5 wells per group from two independent experiments, analyzed by paired student’s t-test, **p<0.01. (D) Primary hippocampal neurons were transfected with mature miR-26a-5p mimics for 48 hours prior to RNA isolation and qRT-PCR to assay the levels of Gas5. Data are from N=6 wells per group from three independent experiments, analyzed by paired student’s t-test; ns, not significant. (E) Schematic representing the miR-26a binding sequence and Gas5 binding region (WT and MUT) in the design of the luciferase assay. (F) Bar graph showing relative luciferase activity in Gas5-WT and Gas5-MUT constructs with miR-26a or a negative miRNA mimic. Data are mean ± SEM, N=8 wells per group from two independent experiments, analyzed by two-way ANOVA followed by Tukey’s multiple comparison test, ***p<0.005; ns, not significant. (G) Euclidean distance matrix for the sense and antisense RNA-seq probe groups used for Gas5 pulldown experiments. (H) Volcano plot indicating RNAs that are enriched in the sense-probe group, relative to those in the antisense-probe group, described in (G). P_adj < 0.05 was used as a threshold for statistical significance, DEseq2, N = 3 wells per group from two independent experiments. (I) GO analysis (Biological Processes) of the differentially expressed genes in (H), input as Entrez Gene IDs. X-axis indicates the number of genes in each cluster. P_adj < 0.05 was used as a cutoff for inclusion in the figures, as determined by ClusterProfiler R package. (J) Volcano plot indicating miRNAs that are enriched in the sense-probe group, relative to those in the antisense-group, described in (G). P < 0.05 was used as a threshold for statistical significance, DEseq2, N = 3 wells per group from two independent experiments. (K) Log₂-transformed normalized spectral counts of sense and antisense probe groups. Blue indicates sense probe–enriched proteins. N = 3 wells per group from two independent experiments; p<0.05 by pairwise t-test with Benjamini-Hochberg correction. (L) GO analysis (Biological Processes) of the significantly differential genes in (K). X-axis and P_adj as described in (I).

We then focused on two sets of three genes upregulated by forskolin that are exclusive targets of miR-26a-5p (Klf4, Plp1, Ugt8a) or are shared targets of both miR-26a-5p and miR-9-3p (Hspa4, Cops2, Kcnh7). Gas5 knockdown led to a suppression of the cAMP-induced increase in two out of the three miR-26a-5p-only targets: Plp1 and Ugt8a (Fig. 4B and data file S3). In contrast, the expression of the three targets that were common to both miRNAs remained unaffected by Gas5 knockdown. These findings suggest that Gas5 is essential for the cAMP-dependent enhancement of miR-26a-5p targets. We also observed that Gas5 knockdown reduced the expression of the IEG Arc, which was not associated with any cAMP-regulated miRNAs in our dataset, suggesting that Gas5 may stabilize Arc expression through a miRNA-independent mechanism.

The abundance of transcripts (lncRNA, mRNA, and miRNA) is a key determinant of ceRNA functions. Reanalysis of our RNA-seq data revealed that the basal and activated transcript levels (TPM) of Gas5 are comparable to, or even exceed, those of cFos mRNA (fig. S8A). Notably, basal Gas5 levels were found to be more abundant than several well-studied lncRNAs, previously identified as ceRNAs, including Meg3, Neat1, Ube3a, Uchl1os, and Paupar (29, 86,97). Independent validation experiments further confirmed that the fold change of basal Gas5 (normalized to 18S rRNA) exceeds that of Arc mRNA (fig. S8B).

To further understand Gas5 regulation of miRNAs, we measured mature miR-26a-5p levels in hippocampal neurons 30 minutes after forskolin treatment. We observed an upregulation of miR-26a-5p (Fig. 4C, data file S3). To explore miRNA regulation further, we transiently overexpressed mature miR-26a-5p mimics in neurons and measured Gas5 levels 48 hours after transfection. Despite confirming the overexpression of miR-26a-5p by qPCR, we found no significant change in Gas5 levels (Fig. 4D). This result aligns with the hypothesis that Gas5 acts as a miRNA sponge, sequestering miRNAs from their targets.

To validate direct miRNA-lncRNA interactions, we performed a luciferase assay. We cloned the region in wild-type Gas5 to which mature miR-26a binds (Fig. 4E) into the pGL3 vector and co-expressed it with miR-26a mimics. A vector encoding a mutated sequence of the Gas5 region served as a control. The difference in luminescence between the wild-type and mutant vectors in the absence of miR-26a-5p was statistically insignificant, and only that of the wild-type Gas5–encoded vector was repressed by the miR-26a-5p mimic (Fig. 4F, data file S3), validating the interaction.

Characterization of Gas5-interacting partners in hippocampal neurons

To gain mechanistic insight into Gas5 function, we conducted comprehensive analyses of the Gas5 interactome using RNA sequencing (RNAseq), miRNA sequencing (miRNAseq), and proteomics. We performed a pulldown of Gas5 from mature primary hippocampal neurons using a biotin-labeled bait of the full-length Gas5 transcript [33]. An antisense probe served as a negative control. RNA isolated from the streptavidin-biotin complexes underwent total and small RNA sequencing.

Hierarchical clustering of the total RNA samples, based on Euclidean distances, revealed that the three Sense replicates clustered closely together, distinct from the three Antisense replicates (Fig. 4G). Differential expression analysis identified 610 RNAs specifically enriched with Gas5 (Fig. 4H, data file S4). Gene biotype analysis confirmed that these RNAs include a diverse array of coding and noncoding RNAs. Notably, gene ontology analysis indicated that these RNAs are predominantly associated with synapse organization (such as Camk2a, Syn1, Dlgap3), cell junctions (such as Rapgef1, Plec, Ajuba), and transport (such as Kif5c, Kif5a, Kif1b) (Fig. 4I, fig. S5A, and data file S4). Small RNA sequencing of these complexes revealed eight miRNAs specifically enriched with Gas5 (Fig. 4J, data file S4). However, these miRNAs did not overlap with our predicted Gas5 targets (Fig. 2E).

The predominance of RNAs associated with Gas5 in processes distal to the soma, such as synapse organization and cell junctions, coupled with the relatively few interacting partners, suggests that Gas5 functions within a larger ribonucleoprotein complex (RNP) [34] rather than through an antisense interaction. To identify associated proteins, we pulled down Gas5 (or the antisense probe control), isolated protein complexes, and subjected them to LC-MS/MS. Differential analysis revealed 58 proteins specifically enriched in the Gas5 complex compared to the antisense control (Fig. 4K, data file S4). These proteins were primarily involved in synapse organization (such as GSK-3β, TUBB5, APOE), apoptosis (such as TPT1, PDIA3, CLU), and viral processes (such as HSPA8, RAB7, P4HB) (Fig. 4L, fig. S5B, and data file S4).

To validate these findings, we compared our data with a previous proteomics study of Gas5 pulldown in hippocampal tissue (Wang et al., 2021) [33]. We found an overlap of eight proteins involved in RNA binding/splicing/translation, phosphorylation, and cytoskeletal structuring (fig. S5C). Analysis of MS spectra confirmed the reliability of these data (fig. S5D). For further validation, we focused on glycogen synthase kinase-3 beta (GSK-3β), a protein enriched in the Gas5 pulldown complex (fig. S3). We validated this interaction by pulldown experiments using a Gas5 probe following cAMP stimulation (Fig. 5A). Primary hippocampal neurons were treated with forskolin or DMSO for 30 minutes before lysis and pulldown. Results showed that Gas5 associated with both total and phosphorylated GSK-3β, with forskolin treatment leading to an increased phosphorylated-to-total GSK-3β ratio (Fig. 5, B–E).

Figure 5. GSK-3β interacts with Gas5.

(A) Experimental schematic for Gas5 Pulldown followed by western blot. (B to E) Immunoblotting analysis for total (B and C) and phosphorylated (Ser⁹; C and D) GSK-3β in Gas5 pulldown lysates (sense and antisense) from mouse primary hippocampal neurons treated with either forskolin (Fsk+) or DMSO (Fsk-) for 30 min. Data are from N=6 wells per group from three independent experiments, analyzed by two-way ANOVA followed by Sidak’s (C) or Tukey’s (D) multiple comparisons test, *p<0.05 and **p<0.01; ns, not significant. (F) DIV 16 hippocampal neurons were fixed and processed for combined FISH and ICC using Quasar fluorophore-labelled Gas5 redundant probes (or no probe control). Cells were immunostained for GSK-3β, and with DAPI to visualize the nucleus. Representative SIM single slice images. Scale bar: 10 μm, inset, 3 μm. (G) Scatter plot showing Pearson’s and Mander’s coefficients between Gas5 and GSK-3β in (F). N=12-15 neurons per group from 3 independent experiments. (H) Bar graph showing relative fold change in GSK-3β-encoding mRNA levels after overexpression of miR-26a or a scrambled control. N=5 wells per group from two independent experiments, paired student’s t-test, *p<0.05. (I) Bar graph showing relative fold change in GSK-3β-encoding RNA levels after Gas5 knockdown relative to the negative control. N=5 wells per group from two independent experiments, paired student’s t-test, **p<0.01). (J to L) Bar graph showing relative fold change of Gas5 (J). miR-26a-5p (K) and GSK-3β RNA (L) expression in 5X-FAD mice vs that in wild-type mice. Data are mean ± SEM from n=3-4 mice per group from two independent experiments, analyzed by unpaired student’s t-test, *p<0.05 and **p<0.01.

To understand the Gas5–GSK-3β interaction further, we performed super-resolution imaging (SIM) following FISH and immunocytochemistry (ICC). Representative images (Fig. 5E) showed that Gas5 and GSK-3β are in proximity in the cell body and colocalize in distinct entities in distal processes (Fig. 5E). Pearson’s Correlation and Mander’s Overlap analysis indicated a strong association between Gas5 and GSK-3β (Fig. 5G). Colocalization features were further supported by line intensity profiles (fig. S5D).

In silico analysis using open-source algorithms [Encyclopedia of RNA Interactomes (ENCORI)] suggested that mRNA encoding the kinase GSK-3β is a target of miR-26a-5p (fig. S6A). Experimental validation showed that transient overexpression of miR-26a-5p resulted in a significant reduction in GSK-3β at the mRNA level (Fig. 5H and data file S5). Additionally, transient knockdown of Gas5 with gapmers also led to a downregulation of GSK-3β-encoding mRNA expression (Fig. 5I). Consistent with these observations, our RNA-seq data showed comparable levels of Gas5 and GSK3B mRNA (fig. S8A). Given the role of GSK-3β in synaptic plasticity and neurodegenerative diseases [62, 63], Gas5 may modulate these processes through its interaction with GSK-3β.

Expression of Gas5 and its interactors are impaired in a mouse model of Alzheimer’s disease

Deregulation of non-coding RNA networks, particularly miRNAs and lncRNAs, has been well-documented in Alzheimer’s disease (AD) [56]. The implications of altered Gas5 levels in AD remain an active area of investigation [57–58]. Increased levels of miR-26a-5p have been observed in both mouse models and patient samples of AD [59–60]. Given our finding that Gas5 is enriched following cAMP signaling, which is disrupted in neurodegenerative diseases [61], we investigated the expression levels of Gas5 and its interactome in a mouse model of Alzheimer’s disease. For this purpose, we utilized 5XFAD mice, which harbor human APP and PSEN1 transgenes with five AD-associated mutations: Swedish (K670N/M671L) and Florida (I716V) mutations in APP, and M146L and L286V mutations in PSEN1. Age-matched controls were also included. We measured the levels of Gas5, GSK-3β, and miR-26a-5p in hippocampal tissues using qRT-PCR. Gas5 abundance was comparable to that of miR-26a-5p (normalized to U6 snRNA) and GSK3β–encoding mRNA, in wild-type mice, a key criterion to satisfy our ceRNA hypothesis (fig. S8C).

Our analysis revealed a significant decrease in Gas5 levels in AD mice compared to controls (Fig. 5J, fig. S8C, and data file S5). Conversely, we observed an upregulation of miR-26a-5p in the AD mice (Fig. 5K). Notably, GSK-3β levels were also upregulated in the AD mice, suggesting additional layers of complexity in the regulation of GSK-3β in AD (Fig. 5L).

Activation of cAMP signaling attenuates the interactions between Gas5 and its associated RNPs

Given Gas5’s interactions with various mRNAs and non-coding RNAs in hippocampal neurons, we investigated whether these interactions are influenced by cAMP signaling. We hypothesized that Gas5 forms a multi-ribonucleoprotein (RNP) complex comprising several RNAs and proteins in response to cAMP signaling. To identify critical regions in Gas5 that mediate these interactions and to assess whether cAMP signaling modifies the composition of its associated RNPs, we conducted RNase protection experiments on Gas5-RNP complexes. We synthesized a biotin-tagged full-length 558 bp Gas5 transcript and performed a pulldown assay in paraformaldehyde (PFA)-fixed primary hippocampal neurons treated with forskolin or DMSO for 30 minutes prior to fixation. Following RNAse-mediated digestion of the Gas5 pulldown complexes, we extracted the bound RNA fragments through ultracentrifugation under a sucrose cushion and subjected these fragments to RNA sequencing (Fig. 6A).

Figure 6: cAMP-induced changes in the Gas5 interactome.

(A) Schematic illustrating Gas5 pulldown strategy from mouse primary hippocampal neuron using sense probes in cross-linked condition, protected fragment purification by sucrose gradient ultracentrifugation after RNase treatment and identifying fragments by RNA-seq. (B) IGV pileups highlighting predominant Gas5 transcripts in forskolin (Fsk)- and DMSO-treated samples. (C) Volcano plots depicting significantly enriched transcripts in the Gas5 pulldowns from Fsk-treated samples compared to DMSO-treated controls. N=3 biological replicates per group from two independent experiments; *p<0.05 DEseq2. (D) Enriched sequences and IGV pileups of transcripts from DMSO- and Fsk-treated samples after Gas5 pulldown. (E) Bar graph showing relative fold change of indicated genes after Gas5 pulldown in Fsk- vs DMSO-treated neurons. N=5 neurons per group, paired student’s t-test, *p<0.05 and **p<0.01. Error bars represent SEM. (F) Cartoon illustration of Gas5 interactome dynamics induced by cAMP. (G) Cartoon model depicting the Gas5 network. Identified LncRNA are in red, pseudogenes in blue, mRNA in black, miRNA in green, and regulators of Gas5 expression in purple).

Unexpectedly, forskolin treatment led to a significant increase in ribonuclease accessibility of Gas5 (Fig. 6B) and a decrease in the enrichment of several mRNAs (Fig. 6C). Examination of the enriched mRNAs using Integrated Genome Viewer (IGV) revealed specific fragments of Myh9 and Jun mRNAs that were exclusively enriched under DMSO conditions but sequestered following forskolin treatment (Fig. 6D). Conversely, expression of the pseudogene Obox-ps27 was significantly enriched under forskolin conditions, with a predominance of a specific fragment (Fig. 6D). These findings were independently confirmed by qPCR analysis (Fig. 6E, and data file S6). Together, these results suggest that cAMP signaling induces dynamic changes in ribonuclease sensitivity of Gas5 and its interactome, thereby modulating its sequestering function in dendrites (Fig. 6F, with a cartoon model of some of the specific observations here in Fig. 6G).

Gas5 mediates cAMP but not mGLUR1/5 induced changes in synaptic transmission and network activity

The interaction of Gas5 with several components known to influence synaptic plasticity, such as GSK-3β, suggests its potential role in modulating cAMP-induced changes in synaptic transmission. To investigate this, we conducted high-density microelectrode array (HD-MEA) measurements to assess cAMP-induced changes in synaptic transmission and network activity following Gas5 loss of function in hippocampal neurons. We also examined the specificity of Gas5’s effects by measuring changes in synaptic transmission induced by DHPG, a known mGluR1/5 agonist, under the same experimental conditions.

At day in vitro (DIV) 16, hippocampal neurons plated on HD-MEA Accura chips (3Brain AG) were transfected with either Gas5-specific or scrambled control gapmers. After 48 hours, neurons were treated with forskolin or DHPG to induce cAMP or mGluR1/5 signaling, respectively, or with DMSO as a control (Fig. 7A). Electrical activity was recorded from multiple chips before pharmacological treatments to establish baseline health and spontaneous activity of the neurons (movie S1). The MEA chips’ baseline was set with a cut-off at 50 μV to minimize noise, and spike detection was set at ≥100 μV. An electrode was deemed active if it exhibited >10 spikes per minute, and a network burst was defined as activity from at least 30 electrodes with >10 spikes per electrode at an inter-spike interval of <100 ms. Network burst frequency was calculated as the total number of bursts divided by the duration of the analysis

Figure 7: Gas5 expression is necessary for cAMP-dependent, but not mGLUR1/5-dependent, changes in synaptic transmission.

(A) Schematic demonstrating the experimental paradigm of Gas5 KD in mouse primary hippocampal neurons followed by pharmacological treatments before Multi Electrode Array (MEA) recordings and analysis. (B to E) Representative trend charts (B and D) depicting Mean Firing Rates (MFR) (spikes/sec) in negative control/ Gas5 knockdown neurons with/ without forskolin (X axis-time in sec; Y axis, spike numbers). Bar graphs (C and E) with data points quantifying MFR in negative control/ Gas5 knockdown neurons with/without forskolin. N=23-24 neurons per group from three independent experiments, analyzed by mixed model ANOVA followed by Tukey’s post hoc test, ****p<0.001. (F) Violin plot depicting MFR values as in (C and E). N=23-24 neurons per group from 3 different experiments, analyzed by mixed model ANOVA followed by Tukey’s post hoc test.,**p<0.01 and ****p<0.001. (G) Representative raw traces from 4 neurons depicting amplitude ranges in negative control/ Gas5 knockdown neurons with forskolin. (X axis, amplitude in μV; Y axis, time in msec). (H and I) Violin plot depicting peak-to-peak amplitude values (μV; H) Inter Spike Interval values (msec; I) in negative control and Gas5-knockdown neurons with/without forskolin. N=23-24 neurons per group, analyzed by mixed model ANOVA followed by Tukey’s post hoc test, ***p<0.005 and ****p<0.001. (J to L) Representative trend charts depicting Mean Firing Rates (MFR) (spikes/sec) in negative control and Gas5-knockdown neurons with/without DHPG (J and L; X axis, time in sec; Y axis, spike numbers), quantified in violin plots in (L). N=23-24 neurons per group, analyzed by mixed model ANOVA followed by Tukey’s post hoc test, ****p<0.001; ns, not significant. Error bars represent SEM for all graphs.

Representative trend charts (Fig. 7, B and D) from MEA recordings showed that mean firing rates peaked at 28-29 spikes/sec on average, with the majority exceeding 15 spikes/sec. Gas5 knockdown led to noticeable differences in spike behavior even before pharmacological treatments (Fig. 7B, data file S7, and movies S2 to S4). Specifically, Gas5 knockdown resulted in decreased intensity and frequency of spike firing, with a more pronounced effect following forskolin application. Forskolin induced a strong, uniform response across neurons in the control group, characterized by denser and more frequent spike trains (Fig. 7B and movie S3). In contrast, Gas5-knockdown neurons exhibited less uniform spike trains and reduced spike density (Fig. 7D, movie S5). Quantitative analysis of the data (Fig. 7C) revealed that basal mean firing rate did not significantly differ from MFR before forskolin treatment but was significantly lower compared to MFR after forskolin treatment. In the Gas5-knockdown group (Fig. 7E), the basal mean firing rate differed from that before forskolin treatment, and both were significantly lower than that after forskolin treatment. These results indicate that knocking down Gas5 impaired action potential density under both spontaneous and cAMP-evoked conditions (Fig. 7F).

We also observed that Gas5 knockdown affected the peak-to-peak amplitude of neuronal firing. Representative traces from four neurons (Fig. 7G) showed that forskolin elicited strong responses (≥ ± 500 μV) in both Gas5-knockdown and control conditions, but quantitative analysis (Fig. 7H) revealed that the amplitude was significantly lower in the Gas5-knockdown group compared to controls, both before and after forskolin treatment. A similar trend was observed in the inter-spike interval (Fig. 7I), where the control group had shorter refractory durations both before and after forskolin compared to the Gas5-knockdown group. These results suggest that loss of functional Gas5 impairs both the density of spike firing and the strength of action potentials in response to cAMP signaling.

We then assessed whether Gas5 also mediates mGluR1/5 signaling-induced changes in synaptic transmission. DHPG treatment, which activates mGluR1/5, led to a reduction in network activity. Representative trend charts (Fig. 7, J and K) showed that whereas the mean firing rates were strong before DHPG treatment, they were almost abolished after DHPG in both the NC and Gas5-knockdown groups (movies S6 and S7). Further analysis revealed significant differences in mean firing rates between the groups without DHPG, but no significant differences in response to DHPG (Fig. 7L). DHPG treatment also did not affect the peak-to-peak amplitude of neurons in the Gas5-knockdown group compared to controls (fig. S7C). These results suggest that although DHPG significantly reduces basal synaptic transmission, Gas5 loss does not alter DHPG-induced effects.

Finally, we assessed whether Gas5 loss of function affects neuronal network phenotypes induced by forskolin or DHPG. Raster plots of entire MEAs showed that NC conditions exhibited regular patterns of network bursts, which became denser and extended upon forskolin treatment (fig. S7A). In contrast, Gas5 knockdown resulted in significantly reduced and less dense network bursts (fig. S7A). Further analysis revealed that network burst frequency was significantly higher in the control group before and after forskolin treatment compared to the Gas5-knockdown group (fig. S7D, data file S7). A similar trend was observed in the network burst duration (fig. S7E). DHPG-induced signaling, on the other hand, significantly diminished neuronal population bursts in both NC and Gas5-knockdown neurons (fig. S7B). These results indicate that Gas5 lncRNA selectively mediates cAMP signaling-induced changes in synaptic transmission and neuronal network activity.

Gas5 loss-of-function impairs dendritic arborization and spine density.

Our molecular and electrophysiological data indicate that Gas5 plays a crucial role in maintaining neuronal morphology, specifically in dendritic arborization and spine density. To evaluate the impact of Gas5 loss-of-function on these parameters, we investigated dendritic arborization and spine morphology in hippocampal neurons. Knockdown of Gas5 using gapmers resulted in a significant reduction in the number of tertiary dendritic branches (60 to 110 μm) compared to control neurons (Fig. 8, A and B, data file S8). However, there was no significant difference observed in secondary branching (20 to 50 μm). Additionally, spine density and the percentage of thin and mushroom spines were significantly reduced in Gas5-knockdown neurons (Fig. 8, C to E, and data file S8). These findings suggest that proper Gas5 function is essential for maintaining dendritic arborization and spine morphology in hippocampal neurons.

Figure 8: Gas5 loss-of-function impairs tertiary dendritic arborization and spine density (A and B).

Representative confocal maximal intensity projection images to assess the morphology of mouse primary hippocampal neurons (scale bar, 30 μm), with quantitative analysis of dendritic morphology changes in Gas5-knockdown (KD) vs control neurons using Sholl analysis of intersections per 10-μm step size (B). N=15-18 neurons per group from three independent experiments, unpaired student’s t-test, **p<0.01. (C to E) Representative confocal maximal intensity projection images showing area analyzed for spine morphology (red rectangle), each with a digitally enlarged image in the inset for spine details (C; full image scale bar, 30 μm; dendrite inset scale bar, 2 μm), and quantitative analysis of the effect of Gas5 knockdown (KD) on the spine density and maturation (D and E). N=16-25 neurons per group, from three independent experiments, analyzed by paired (D) and unpaired (E) student’s t-tests, *p<0.05 and **p<0.01. Error bars represent SEM. (F) Experimental schematic for total RNA-seq after Gas5 knockdown (KD) or negative control (NC) in primary hippocampal neurons. (G) Volcano plot indicating DEseq2-identified significant genes in the neurons described in (F); p_adj < 0.05 significance by DEseq2, from N = 4 wells per group from four independent experiments. (H) GO analysis (Biological Processes) of the significantly differential genes from (G). Size of node indicates gene number, and color code indicates −log₁₀ FDR values.

To further understand the molecular basis of these morphological changes, we examined the transcriptional consequences of Gas5 loss of function. We hypothesized that Gas5 knockdown might affect the expression of genes crucial for neuronal morphology. Following Gas5 knockdown using gapmers, we performed total RNA sequencing (Fig. 8F). Biotype analysis of differentially expressed transcripts (fig. S5E) revealed that of the 219 perturbed genes, the majority (142 genes, or 64.8%) were protein-coding genes, followed by processed pseudogenes (43 genes, or 19.6%) and lncRNAs (34 genes, or 15.5%). Several significantly downregulated genes included early response genes, such as Egr1, Egr2, and Arc, as well as the translation factor-encoding gene Eif3e (Fig. 8G and data file S8). Additionally, genes critical for neuronal physiology, such as Rheb and Cftr, were also downregulated. Several lncRNAs were deregulated in the Gas5-knockdown set, including Hotairm1, Macrod2os1, and Gas5 itself. Notably, a substantial subset of altered genes included pseudogenes. To validate the sequencing data, we designed gene-specific primers for 5 candidate genes: Arc, Egr1, Rheb, Cftr and Hotairm1. Subsequent qPCR data showed that indeed the expression of Arc, Egr1, Rheb and Cftr were downregulated in the validation sets, whereas that of Hotairm1 was increased (fig S7G and data file S8). Gene ontology analysis revealed that these RNAs are involved in various processes, such as axon guidance (including Cdk5, Epha6, Rock2, Sema3e, Lrig, and Srgap3), the cAMP signaling pathway ( Hcn4, Abcc4, and Myl9), regulation of the actin cytoskeleton ( Arpc4, Egfr, and Itga11), and are associated with neurodegenerative diseases, like ALS and Alzheimer’s Disease (Frat1, Casp12, Adam10, Rtn3, and Tuba1b ) (Fig. 8H and data file S8).

Discussion

In this study, we explored how coding and noncoding transcriptomes were influenced by cAMP and mGluR1/5 signaling—two pathways with opposing effects on synaptic transmission and neuronal morphology. Our analyses identified the lncRNA Gas5 as a key mediator of cAMP-induced changes that impact excitatory synaptic transmission and neuronal structure.

To examine pathway-specific transcriptome effects, we created a web-based tool (https://egrinman.shinyapps.io/FSK_DHPG_Gene_Expression/) for visualizing lncRNA, mRNA, and miRNA expression. Though both forskolin (cAMP) and DHPG (mGluR1/5) stimulation produced hundreds of differentially expressed genes, only 14 overlapped. cAMP-influenced genes were associated with RNA processing, synaptic organization and protein degradation, suggestive of post-transcriptional and translational regulation, whereas mGluR1/5-modulated genes were associated with cell assembly and cytoskeletal dynamics. To identify co-regulated genes, we applied multiscale network analysis and identified pathway-specific gene modules. In the cAMP dataset, one module (“C1_5”) was enriched for RNA splicing and transport functions and included two novel zinc finger proteins, Zfp182 and Zcchc7, suggesting roles in RNA processing. In contrast, gene modules in the mGluR1/5 pathway showed weaker associations with initial pathway targets, likely due to cell-type heterogeneity in hippocampal cultures.

We next analyzed interactions among differentially expressed miRNAs, lncRNAs, and mRNAs. In the DHPG dataset, upregulated miRNAs correlated with decreased target mRNAs. However, in the cAMP dataset, many miRNA targets were paradoxically upregulated, suggesting a competing endogenous RNA (ceRNA) mechanism. Normally, miRNAs mediate transcript repression [40], but near-perfect complementarity can trigger miRNA degradation [41]. Bioinformatic and functional assays confirmed that Gas5 acts as a ceRNA for miR-26a-5p, preserving expression of targets Plp1 and Ugt8a, genes involved in myelination—suggesting this regulatory interaction may also occur in oligodendrocytes. Abundance of ceRNA components is key. Our RNA-seq data revealed Gas5 levels that were comparable to, or exceeded, those of GSK3B and cFos. It also surpassed well-known lncRNAs like Meg3, Neat1, and Ube3a. Independent qPCR validation confirmed that Gas5 levels exceeded those of Arc, and that its abundance was on par with those of miR-26a-5p and GSK3B in both hippocampal neurons and in AD-model mouse tissue. Further evidence supports Gas5’s ceRNA function via single-site interactions. Prior studies showed Gas5 sponging various miRNAs to regulate gene expression in kidney injury [88], coronary artery disease [89], and atherosclerosis [90]. Similarly, other lncRNAs like SCAMP1 [91], NORAD [92], and KCNQ1OT1 [93] were shown to sequester miR-26a-5p in cancer models. These studies support the idea that miR-26a-5p is a central node in lncRNA-mediated gene regulation. Prompted by this, we analyzed the Gas5 interactome and identified 611 enriched RNAs involved in synapse organization, transport, and cell adhesion—key processes in local translation at synapses [46, 47]. Eight miRNAs interacted with Gas5, but none matched the predicted targets in databases, suggesting direct mRNA binding. Prior studies have shown that Gas5 associates with RNA-binding proteins like AGO2 [45] and EIF4E [43]. Our proteomic analysis revealed 58 interacting proteins involved in synapse organization and apoptosis. Notably, the mRNA encoding GSK-3β—a key plasticity regulator—was among them, with cAMP stimulation enhancing its interaction with Gas5. Functional experiments confirmed this relationship, supporting a regulatory axis involving Gas5, miR-26a-5p, and GSK-3β.

Eight of Gas5’s interacting partners in the hippocampus identified here overlapped with findings by Wang et al. [33], suggesting Gas5 is part of a larger ribonucleoprotein (RNP) complex. These dynamic, non-membrane-bound assemblies regulate synaptic plasticity, and their disruption may contribute to neurodegeneration. Given Gas5’s interactions and dendritic localization, it likely functions within RNPs at synapses.

Given these findings, Gas5 appears to play a substantial role in gene regulation, synaptic transmission, and plasticity—processes that are notably impaired in neurodegenerative diseases such as AD [61]. In our AD-model 5XFAD mice, Gas5 levels were reduced, whereas those of miR-26a-5p were increased. Unexpectedly, GSK3B levels were increased, suggesting its regulation in AD may be miRNA-independent in the mouse hippocampus [71–72].

To pinpoint interaction regions in Gas5, we performed RNAse accessibility mapping. In resting neurons, Gas5 was largely RNAse-protected, indicating RNA or protein interaction. After cAMP stimulation, parts became RNAse-sensitive, suggesting dynamic remodeling of its interactome to support synaptic transmission. These structural changes extended to Gas5’s RNA interactome, which also showed cAMP-dependent shifts in RNAse sensitivity, suggesting Gas5 serves as a tunable scaffold for dendritic RNP complexes. Such tunability may involve cAMP-induced modifications of associated proteins or RNAs.

Because of the strong transcriptomic signal, we next asked whether Gas5 functionally mediates synaptic changes. In MEA experiments, Gas5 knockdown impaired forskolin-induced increases in excitatory synaptic activity, supporting a direct role in synaptic transmission. Though other lncRNAs (such as Gm12371) have been implicated in forskolin responses, their specificity is unclear. Here, Gas5 emerged as a pathway-specific mediator.

Given the tight link between morphology and synaptic plasticity [64], we also evaluated structural effects. Knocking down Gas5 led to reduced dendritic arborization. Whereas control neurons maintained a pyramidal structure, Gas5-deficient neurons showed elongated morphology with diminished tertiary branches and reduced spine density, particularly of thin and mushroom spines. These morphological changes coincided with downregulated expression of immediate early genes, ERG complex components, and synapse-related genes [54,73], emphasizing Gas5’s role in maintaining dendritic structure and signaling. Because Gas5 is expressed in the nucleus, soma, and dendrites, further research is needed to map compartment-specific roles.

Together, these findings suggest that dendritically localized Gas5 is critical for excitatory synaptic function and structure—processes central to learning and memory. This is consistent with studies showing that lncRNAs play key roles in learning and long-term memory [67–70]. Liau et al. [67] found that Gas5 was enriched at synapses and interacted with RBPs to regulate RNA trafficking, and that its knockdown impaired fear extinction memory. Meier et al. [69] observed that hippocampal Gas5 levels negatively correlated with anxiety-related behaviors and were increased in response to stress. Similarly, Banerjee et al. [74] showed that maternal separation stress upregulated Gas5 expression, leading to spatial memory deficits.

Our study provides insights into the alterations in gene expression, of both coding and noncoding RNAs, associated with cAMP and mGluR1/5 signaling in hippocampal neurons. Additionally, the findings shed light on the selective recruitment of lncRNAs, specifically of Gas5 and the dynamic modification of its interactome for mediating cAMP signaling. Understanding the mechanisms governing lncRNA interactome stability and compartment-specific functions is essential for deciphering the molecular foundations of lncRNA function in dendrites.

Materials and Methods

Animals:

CD1 pregnant mice (Charles River Laboratories) were housed individually on a light dark cycle (12 hour/ 12 hour) with ad libitum access to food and water. Experiments were performed during the light part of the diurnal cycle. Housing, animal care and experimental procedures were consistent with the Guide for the Care and Use of Laboratory Animals and approved by the Institutional Animal Care and Use Committee of The Wertheim UF Scripps Institute.

Primary hippocampal neuron culture:

Hippocampi were dissected from E18 CD1 mouse pups and plated on poly-D-lysine coated plates in Neurobasal media supplemented with Glutamax, Penstrep, and 5% FBS. Density of neurons plated was determined by each experimental requirement. Four hours after plating, media was replaced with feeding media consisting of Neurobasal media supplemented with Glutamax, Penstrep, and 2% B27 (Invitrogen), and half of the media was replaced every 4 days until the time of experiments at 37°C with 5% CO₂.

Quantitative real-time PCR:

RNA was isolated using Trizol and reversed transcribed into cDNA using qScript cDNA SuperMix (Quanta Bio) and/or TaqMan MicroRNA Reverse Transcription Kit (Thermofisher). qRT-PCR was performed in 384-well plates using SYBR Green master mix or TaqMan Fast Advanced Master Mix (Thermofisher) for detection in the ABI 7900 thermal cycler (Applied Biosystems). dCT values were obtained using Ct values for 18s for normalization, followed by relative quantification using 2^−ΔCT method. For qRT-PCR following synaptoneurosome preparations, 6 well pooled neurons were lysed in SynPER reagent and the synaptoneurosome pellet was isolated as per manufacturer’s instructions. RNA from these pellets was isolated using TRIzol followed by Zymo Micro RNA kit. All primer sequences are listed in the supplement (data file S1).

Total and small RNA-seq and analysis:

Total RNA and small RNA sequencing was performed following the protocol of The Wertheim UF Scripps Institute Genomics Core. Total RNA samples were treated with DNase I (NEB, Ipswich, MA) and purified with PureLink RNA Micro kit (Invitrogen, Carlsbad, CA). Subsequently, samples were quantified using a Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA) and run on Agilent 2100 Bioanalyzer RNA nano chips (Agilent Technologies, Santa Clara, CA) for quality assessment. All RNA samples were excellent quality with RNA Integrity Number (RIN) > 8.5. Total RNA-seq libraries were prepared using Illumina TruSeq total RNA stranded kit (Illumina, San Diego, CA) with 300ng of input RNA per sample, following the manufacturer’s protocol. Small RNA-seq libraries were prepared using Illumina TruSeq small RNA kit (Illumina, San Diego, CA) with 1ug of input RNA. Final libraries were validated on Bioanalyzer DNA chips and pooled in equimolar ratios. The small RNA pool was size selected on 6% polyacrylamide gel to recover products between 145bp and 160bp. Purified libraries were loaded onto a NextSeq 500 flow cell (Illumina, San Diego, CA) at 1.8pM final concentration and sequenced using 2 x 40bp paired-end chemistry. On average we generated 22-27 million reads per sample for total RNA-seq samples and around 4-14 million reads per sample for small RNA-seq samples. Fastq files were first assessed for quality using fastqc [75] and trimmed using Trim Galore on The Wertheim UF Scripps Institute high-performance computer cluster. Fastq reads were then aligned to the Mus musculus transcriptome using Salmon v1.10.1 [76], also using The Wertheim UF Scripps Institute high-performance computing cluster, to generate raw counts and TPM data based on Ensembl Transcript IDs. Data was then imported into R using tximport [77] Bioconductor package. Hierarchical clustering using euclidean distances, as well as Principal Component Analysis were performed using R base/factoextra package and plotted with ggplot or pheatmap plotting packages. Differential expression analysis was performed using DEseq2 [78]. A p-adj value of 0.05 was used to set the FDR to 5% in total RNASeq, while such FDR cut of was not applied and a p-value cut off of 0.05 was used for small RNA sequencing, given the lower number of miRNA genes available for multiple-testing correction. Some volcano plots were plotted without the non-DEGs, and y-axis normalized to Sq Root −log10 (p-value or padj), for visibility purposes of displaying the relevant data in a limited amount of space. GO pathway analysis was performed using the clusterProfiler [79] package. Gene biotypes were ascertained using the biomaRt [80] package.

Multiscale network analysis:

MEGENA [81] networks were constructed from genes upregulated in FSK vs DMSO or DHPG vs DMSO, as determined by DEseq2, using the R package MEGENA. All TPM values of the upregulated genes were tested for planarity to grow a Planar Filtered Network (PFN). Multiscale clustering analysis was conducted with the resulting PFN to identify co-expression modules. We then searched for the most significant modules (as determined by MEGENA) with the largest numbers of genes, which led to the discovery of module C1_5. Plots were generated using Cytosape [82] with the edge-weighted force directed layout.

Uncovering cis lncRNA/mRNA pairs:

Significantly attenuated (p_adj <0.05) lncRNAs and mRNAs were used for the analysis in both the forskolin and DHPG datasets. lncRNAs were first subset by biotype using biomaRt, then queried against every significant mRNA within 100kb up and downstream of the lncRNA TSS using biomaRt. All hits from this query for individually confirmed using the UCSC Genome Browser to find the precise cis relationship.

Constructing miRNA/lncRNA/mRNA interaction networks:

Once miRNAs of interest were identified, we set out to identify all predicted targets of these transcripts. Using two commonly used databases for miRNA target prediction (TargetScan and starBase) we identified targets pertaining to our miRNA of interest and cross-referenced with the statistically significant genes identified in the total RNA-seq dataset. These included mRNAs, processed transcripts, pseudogenes, and lincRNAs, according to biotype designations found previously. Cytoscape was used to visualize the networks of miRNA-target interactions identified by the two databases. In the significance networks red and green represent down- and up-regulated targets, respectively.

Transfection of plasmid constructs, LNA gapmers, and siRNAs:

LNA Gapmers made by Exiqon (now QIAGEN) targeting Gas5 and SMARTPool siRNAs (Dharmacon) targeting Kif1a, Kif1a, Kif2a, Kif5c, Kif16b and siNC were transfected using Lipofectamine RNAiMax (Invitrogen). Briefly, Gapmers or siRNAs were introduced to primary hippocampal neurons (12 to 14 days in vitro (DIV)) using Lipofectamine RNAiMAX (Invitrogen) according to manufacturer’s guidelines.

RNA FISH and immunofluorescence:

Two 300-400 bp fragments of Gas5 cDNA were sub cloned into PCRII TOPO vectors and in vitro transcribed into DIG labeled probes for fluorescence in situ hybridization (FISH). Anti-DIG fab fragment antibody (Roche) was used at a concentration of 1:4000, and Tyramide signal amplification (Akoya Biosciences) was used using manufacturer instructions. One of the fragments, targeting the 3’ end of Gas5, was more efficient at producing a robust signal, and therefore was used for all subsequent studies. Additionally, 25 20-mer redundant probes targeting Gas5 were designed using Stellaris designer software and ordered as a pool (conjugated to Quasar570 dye), Gas5 probe sequences are listed in the supplement (data file S1). FISH protocol was performed as per the manufacturer’s instruction. Immunofluorescence was performed in conjunction to or separate from FISH, according to experiment and samples were imaged on the Zeiss LSM 880 confocal microscope at Max Planck Florida Institute light microscopy facility and Olympus IX81 at the UF Scripps Biomedical Research Imaging core (antibodies used were- GSK-3β (27C10-CST), β-Actin (8H10D10-CST) and Alexa Fluor conjugated secondary antibodies). Neurons were plated at a low density (80,000 cells per well of a 24-well plate) to allow for imaging of single neurons at a time, with minimal crossover from other neurons. Super resolution imaging was performed using the Scripps Florida Imaging core housed Zeiss Elyra PS.1 scope (Carl Zess, Germany), at a resolution of 1028 by 1028 pixels, using an oil immersion Zeiss 63x/ 1.4 NA Plan apochromatic objective, in the SIM mode. Images were acquired in the 405, 488 and 561 channels using 5 pattern rotations with 3 translational shifts 20-25 z stacks were captured for each image in 3 channels (SIM grating) and postprocessed to show maximum intensity projection images or single slice images (SIM reconstruction using Zen Black software).

Proteomics and analysis:

Proteins were separated by SDS-PAGE, washed in water, and fixed overnight (10% glacial acetic acid, 30% ethanol) on an end-to-end shaker. For mass spectrometry, lanes corresponding to antisense and sense probe groups were cut out and in-gel digested with trypsin (Pierce Biotechnology, Rockford, IL) for 3 hours at 37°C using ProteaseMax^™ Surfactant trypsin enhancer following reduction and alkylation with dithiothreitol and iodoacetamide, respectively, according to the manufacturer’s instructions (Promega Corporation, Madison, WI). LC-MS/MS analysis of extracted peptides was subsequently carried out using an Orbitrap Fusion Tribrid mass spectrometer, following a 2-mg capacity ZipTip (Millipore, Billerica, MA) C18 sample clean-up according to the manufacturer’s instructions. Peptides were eluted from an EASY PepMapTM RSLC C18 column (2 mm, 100Å, 75 mm x 50cm, Thermo Scientific, San Jose, CA) into the mass spectrometer using a gradient of 5-25% solvent B (80/20 acetonitrile/water, 0.1% formic acid) in 90 min, followed by 25-44% solvent B in 30 min, 44-80% solvent B in 0.10 min, a 10 min-hold of 80% solvent B, a return to 5% solvent B in 3 min, and finally with another 3-minute hold of 5% solvent B. The gradient was then extended for the purpose of cleaning the column by increasing solvent B to 98% in 3 min, a 98% solvent B hold for 10 minutes, a return to 5% solvent B in 3 minutes, a 5% solvent B fold for 3 min, an increase of solvent B to 98% in 3 min, a 98% solvent B hold for 10 min, a return to 5% solvent B in 3 minutes and a 5% solvent B hold for 3 min and finally, another increase to 98% solvent B in 3 min and a hold of 98% solvent B for 10 min. All flow rates were 250 nL/min delivered using an nEasy-LC1000 nano liquid chromatography system (Thermo Fisher Scientific, San Jose, CA). Solvent A consisted of 0.1% formic acid. Ions were created with an EASY Spray source (Thermo Scientific, San Jose, CA) held at 45°C using a voltage of 2.3kV. Data dependent scanning was performed by the Xcalibur v 4.0.27.10 software using a survey scan at 120, 000 resolution in the Orbitrap analyzer scanning mass/charge (m/z) 350-2000 followed by higher-energy collisional dissociation (HCD) tandem mass spectrometry (MS/MS) at a normalized collision energy of 30% of the most intense ions at maximum speed, at an automatic gain control of 1.0E4. Precursor ions were selected by the monoisotopic precursor selection (MIPS) setting to peptide and MS/MS was performed on charged species of 2-8 charges at a resolution of 30,000. Dynamic exclusion was set to exclude ions once within a 25 second window. All scan events occurred within a 2-second specified cycle time. Tandem mass spectra were searched against the mouse proteome protein sequences from Uniprot (UP000000589) downloaded on March 03, 2020, and common contaminant proteins available with Proteome Discoverer v 2.5.0.400. At the time of the search the UP000000589 contained 54,348 sequences and the contaminant protein database contained an additional 298 sequences. All MS/MS spectra were searched using Thermo Proteome Discoverer 2.5.0.400 (Thermo Fisher Scientific, San Jose, CA) considering fully tryptic peptides with up to 2 missed cleavage sites. Variable modifications considered during the search included methionine oxidation (15.995 Da), and asparagine and qlutamine deamidation (0.984 Da). Cysteine carbamidomethylation (57.021 Da) was considered as a static modification. Proteins were identified at 99% confidence with XCorr score cut-offs [83] as determined by a reversed database search. The Minora Feature Detector node which detects chromatographic peaks and features as also used in Proteome Discoverer at default settings. The protein and peptide identification results were also visualized with Scaffold v 5.0.0 (Proteome Software Inc., Portland OR), a program that relies on various search engine results (Sequest, X!Tandem, MASCOT) and which uses Bayesian statistics to reliably identify more spectra [84]. Proteins were accepted that passed a minimum of two peptides identified at 1% peptide and protein FDR, within scaffold. Pairwise t-test with Benjamini-Hochberg correction was used and p<0.05 was used to determine statistical significance of normalized spectral counts. The mass spectrometry analysis was performed at The Herbert Wertheim UF Scripps Institute for Biomedical Innovation & Technology, Mass Spectrometry and Proteomics Core Facility (RRID:SCR_023576).

Western blotting (immunoblotting):

For the Western blot analysis, to validate Gas5 protein interaction from the pulldown proteomics study, primary hippocampal neurons were treated with forskolin or DMSO as stated above. Following crosslinking and pulldown with either Gas5 sense, antisense, or control IgG probes, as above, neurons were lysed with RIPA buffer (50 mM Tris HCl, 150 mM NaCl, 1.0% (v/v) NP-40, 0.5% (w/v) sodium deoxycholate, 1.0 mM EDTA, 0.1% (w/v) SDS, and 0.01% (w/v) sodium azide at a pH of 7.4), and the protein concentration was determined using a BCA kit. 25-30 μg of protein was used for Western analysis. The antibodies used were GSK-3β [27C10, Cell Signaling Technology (CST)], phospho-GSK-3β (Ser⁹) (5B3, CST), and β-actin (8H10D10, CST). The target proteins were detected using anti-rabbit HRP-linked (#5127, CST) or anti-mouse HRP-linked (#7076, CST) secondary antibodies at a 1:5000 dilution and then visualized by chemiluminescence (Amersham Biosciences, Piscataway, NJ). The blots were analyzed by ImageJ.

Luciferase reporter assay:

The wild-type and a mutant 3′UTR sequence of mouse Gas5, which contained the potential miR-26a-5p binding sites, were inserted into pGL3 Basic Luciferase vector (Promega, USA) using the XbaI restriction sites, just downstream of the Luc gene. The plasmids were named Gas5-WT and Gas5-MUT, and the sequences were as follows for Gas5-WT: 5’gtgagaactgcaaatgcttaaccgggaacctactccagaatacatgatgatctcacacaacttgaactctctcactgattacttgatgatagtaaaagatctgatgttctgtgttttaacagttaccatttaagttaaaattgtagaaaagtgtttaacagctaccttctgttggttgttgcag-3’, and Gas5-MUT: 5’gtgagaactgcaaatgcttaaccgggaacctactccagaatacatgatgatctcacacaacttgacatctctccagttaattcgttatgatagtaaaagatctgatgttctgtgttttaacagttaccatttaagttaaaattgtagaaaagtgtttaacagctaccttctgttggttgttgcag-3’. Cells were co-transfected with either Gas5-WT or Gas5-MUT or an “empty” pGL3 with miR-26a-5p mimic or negative control, respectively, using Lipofectamine 2000 (Thermo Fisher Scientific, USA) according to the manufacturer’s protocol. Luciferase activity was measured after 48 hours using the Bright-Glo luciferase reporter assay kit (Promega, Madison, WI, USA), also according to manufacturer’s instructions. In brief, prior to the assay, transfected cells in 96-well format were equilibrated to room temperature. To each well, 100 μL of Bright-Glo reagent (to 100 μL of growth medium) was added and mixed well. Cell lysis was allowed to proceed for 5 min and then luminescence was measured in a PHERAstar FSX microplate reader. Normalized firefly luciferase activity for each construct was compared to that of the pGL3 empty Vector control. For each transfection, luciferase activity was averaged from eight biological replicates.

Electrophysiological recordings and analysis:

Primary mouse hippocampal neurons were plated on the CMOS (complementary metal-oxide-semiconductor)-based HD-MEA, BioChip HD-MEA Accura, from 3Brain AG, which consists of 4096 recording channels (in a 64 × 64 grid; 3.8 mm x 3.8 mm, 60 μm pitch) (each microelectrode is 21 μm x 21 μm). Accura chips were sterilized in 70% ethanol for 20 min and rinsed thoroughly with sterilized double-distilled water, then left to air dry under a laminar hood. Chips were then pre-conditioned with neurobasal medium overnight at 37 °C (to increase hydrophilicity) and then pre-coated with poly-d-lysine (50 μg/mL) overnight at 37 °C, rinsed thoroughly with double-distilled water and used for neuron seeding. After the cells were dissociated, the concentration was diluted to 1000 cells/μL and 90,000 to 110,000 cells were seeded on the HD-MEA chip with a 100 μL droplet. Cells were left to settle for 3 to 4 hours before adding 1.5 mL of neurobasal feeding medium. The cultures were incubated at 37 °C, 5% CO₂ and half of the medium was changed every 3 to 4 days. Multiple recording sessions of 5 min were performed at different days with a HD-CMOS technology microelectrode array of 4096 microelectrodes (BioCam X, 3Brain) sampled at 17.8 KHz/electrode and analyzed with the integrated Brainwave 5 software application.

Neuronal morphology assessments:

Mouse primary hippocampal neurons were transfected with either Gas5-targeting gapmers or negative controls and after 72 hours of images of dendrites were collected at 37°C in the light microscopy facility at UF Scripps Biomedical Research, using a confocal microscope (FV1000; Olympus; Apo N 60X/1.49 Oil) in Hibernate-E (Brainbits). Z-stack images were acquired using Fluoview1000 (64 bit) software (Olympus) and converted into a maximum projection intensity image in FIJI (ImageJ, NIH). Dendritic arbor was quantified via the Sholl analysis plugin in FIJI. The center of soma is considered as the midpoint and origin of the concentric radii was set from that point to the longest axis of soma. The parameters set for analysis were: starting radius 20 μm, ending radius 100 μm, radius step size 10 μm. The maximum value of sampled intersections reflecting the highest number of processes/branches in the arbor was calculated and the number of intersections plotted against distance from the soma center in μm. Data was analyzed using two-way ANOVA. Spine morphology was analyzed using a custom MATLAB [85] script developed in the light microscopy facility at the Max Planck Florida Institute. By using a geometric approach, this software automatically detects and quantifies the structure of dendritic spines from the selected secondary branch (100 μm length) in the Z-stack confocal image. The software assigns the detected spines to one of the three morphological categories (thin, stubby or mushroom) based on the difference in structural components of the spines, meaning the head, neck and shaft. Student’s t tests were carried out to evaluate the statistical difference amongst the groups.

RNA pull-down assay with proteomic, transcriptomic and protected fragments analysis:

The full-length sense Gas5 RNA sequence was cloned into a pCR-II-TOPO vector from a template (Thermo Fisher Scientific), linearized and 5’ end biotin-labeled (Roche) by in vitro transcription using a SP6 RNA polymerase enzyme for the sense strand and T7 for antisense strand that was used as a control. Mouse primary neurons, seeded in 6 well plates, with a density of 150K per well, were grown until DIV 17. Then, they were treated with either forskolin (50 μM), DHPG (50 μM), or DMSO control. 30 minutes after treatment, each sample was fixed with 4% PFA for 15 min, followed by 3 washes in chilled 1X PBS, then lysed in 600 μl of Lysis Buffer I [Tris HCl (50mM), NaCl (150mM), EDTA (0.5M), NP-40 (10%), protease inhibitor cocktail (100x), phosphatase inhibitor 2 (100x), phosphatase inhibitor 3 (100x), RNAse inhibitor (50ul), BSA (100x), and DTT (1mM)]. Lysed samples were then centrifuged at 1200xg for 20 min at 4°C and the pellet discarded. Meanwhile, streptavidin magnetic beads were washed with this buffer twice (x 2 min each) and blocked with yeast tRNA and glycogen (0.2 mg/ml) overnight. During centrifugation time, probes were linearized at 72°C for 15 min and placed on ice. Then 2 μl of the probes were added to the lysate (approximately 500 μl for each sample per condition), and nutated for 2 hours at 4°C (during this time, the blocking from beads was removed by washing them once with lysis buffer). After incubation with the probes, 10 μl of beads were added to each sample and incubated for an additional hour at 4°C. After incubation, the lysate was removed, and the beads were washed 3 x 2 min with lysis buffer I, then another 3 x 2 min washes in high stringency buffer (Lysis buffer II – no BSA, NaCl – 600mM). Then, 40 μl of LB buffer was added to isolate the protein (95°C for 10 min) or diluted in trizol and incubated for 40 min in rotation. Then, the RNA from the samples were isolated using the Direct-zol RNA miniprep kit (Zymo) according to the manufacturer’s instructions. The RNA was processed for RNA-seq and analyzed by the Wertheim UF Scripps Institute Genomics Core.

RNA-seq analysis of RNAse-protected fragments:

Protected fragments study was carried out following the pull-down protocol described for the RNA-protein interaction. After RNAse A incubation (30ul of Tris Buffer with 0.15% SDS. and 1ul of RNAse A and incubated at 37°C for an hour, the samples were diluted in PBS. The protected fragments were purified and concentrated using ultracentrifugation at 70,000 RPM for 4 hours using a sucrose cushion. The supernatant was removed, and the RNA fragments pellet was diluted in Trizol reagent for their purification using the trizol-chloroform RNA purification method. The samples were sequenced by the Wertheim UF Scripps Institute Genomics core using the protocol stated above.

Statistical Analysis:

Statistical analysis was performed in R Studio version 1.2.1335. and Prism 8. Data are represented by the mean and error bars represent SEM. All outcome variables in non-omics experiments were assumed to be normally distributed and were analyzed using unpaired two-tailed t-tests, one-way ANOVA, or two-way ANOVA, for designs with no repeated measures. Planned post hoc tests were chosen based on the procedures recommended by the GraphPad Prism, and the tests are noted in the legends and supplemental materials. Paired t-test and mixed model ANOVA were used for designs where the same sample was measured at different time points or different treatments. N represents the number of independent samples for each experiment, unless stated otherwise. A p-value cutoff was 0.05 and were marked in the graphs accordingly.

Data and materials availability:

The RNA sequencing data are deposited to NCBI Gene Expression Omnibus with the accession number GSE161071 (Fig. 1) and GSE180503 (Fig. 7). The mass spectrometry proteomics data are deposited to the ProteomeXchange Consortium via the PRIDE [55] partner repository with the dataset identifier PXD027453 and 10.6019/PXD027453. All other data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additionally, the data will be made available from the authors upon reasonable request.