RNA-RNA interactions coordinate GABA metabolism and signaling

Abstract

Production, signaling, and degradation of the inhibitory neurotransmitter GABA requires tight regulation for proper nervous system function, yet the gene regulatory mechanisms underlying this control remain poorly understood. Here, we identify a post-transcriptional mechanism that coordinates expression of the GABA synthetase Gad1 with its vesicular transporter VGAT in Drosophila. While Gad1 is transcribed specifically in GABAergic neurons, VGAT mRNA is transcribed broadly across the nervous system, but is selectively translated in GABAergic neurons. We show that this specificity depends on CG14989, a putative long non-coding RNA located adjacent to the Gad1 gene locus that shares its GABAergic expression pattern. CG14989 RNA contains sequences complementary to three predicted miR-7 sites in the VGAT 3′UTR, and ectopic expression experiments revealed that it is sufficient for VGAT translation in non-GABAergic neurons, consistent with antagonism of miR-7-mediated repression. Furthermore, complementary regions to CG14989 are present in multiple other genes related to metabolism and signaling of GABA, suggesting a broader role for this mechanism. Together, our findings suggest that CG14989 functions as a regulatory hub that coordinates the molecular identity of GABAergic neurons.

Introduction

Coordinating the expression of multiple components in a molecular pathway is an essential feature of biological systems. In the nervous system, this principle is particularly evident in neurotransmitter pathways, where the synthesis, packaging, release, reception, and degradation of signaling molecules must be tightly balanced. Disruptions in this balance can contribute to a range of neurological disorders[1–3].

Much of what we understand about such coordination results from analyses of transcriptional and chromatin regulation. For instance, in mammals, the acetylcholine-synthesizing enzyme ChAT and its vesicular transporter VAChT are regulated by shared transcriptional inputs at distinct promoters, ensuring their co-expression in cholinergic neurons[4, 5]. Beyond transcription, post-transcriptional mechanisms—mediated by RNA-binding proteins (RBPs) and microRNAs—further fine-tune gene expression by influencing splicing, mRNA stability, localization, and translation. In Drosophila and C. elegans, by contrast to mammals, ChAT and VAChT are co-expressed through differential splicing of a transcript initiated from a common promoter[6, 7].

While protein-coding genes and small RNAs have been extensively studied in this context, far less is known about the roles of long non-coding RNAs (lncRNAs) in coordinating neuronal gene expression. lncRNAs can modulate transcription, splicing, and RNA metabolism through diverse mechanisms in both the nucleus and cytoplasm[8]. Although many lncRNAs are expressed in the brain, only a few have well-defined functions in neurons, particularly in the cytoplasm[9, 10]. One of the best-characterized examples is rodent brain cytoplasmic RNA 1 (BC1; BC200 in primates), which localizes to synapses and regulates local protein synthesis by interacting with RBPs and the translational machinery[11, 12].

In this work, we describe a critical role for a putative lncRNA in signaling of the primary inhibitory neurotransmitter, GABA, within the Drosophila nervous system that appears to be conserved in Diptera. Like ChAT and VAChT in cholinergic neurons, the GABA synthetase and transporter, Gad1 and VGAT, are specifically co-expressed in GABAergic neurons of the fly. For Gad1, this expression is controlled transcriptionally, as Gad1 mRNA is only found within GABAergic neurons. However, VGAT mRNA is expressed in seemingly all neurons, and we found that the specificity of its expression is the result of repression of its translation outside of GABAergic neurons. We identified the link that coordinates Gad1 transcription with VGAT translation as CG14989, a putative lncRNA located next to the Gad1 gene locus that is also expressed specifically in GABAergic neurons. The 3’ end of the CG14989 RNA contains three stretches of complementary regions to the 3’UTR of VGAT, each of which completely overlap predicted target sites for the conserved microRNA miR-7. Using reporter constructs in-vivo, we show that these regions are essential for translational repression of the VGAT 3’UTR in non-GABAergic neurons, whereas the presence of CG14989 RNA is sufficient to de-repress VGAT translation in GABAergic neurons. Finally, we identify several other mRNAs, with known functions in nearly all aspects of GABA metabolism and signaling, which also contain complementary regions to CG14989. Together, we suggest a novel function for intermolecular RNA-RNA interaction in specifying the spatial (cell-type) translation of functionally related genes in the nervous system by antagonizing microRNA targeting.

Results

Specificity of VGAT expression in GABAergic neurons requires post-transcriptional regulation

VGAT is specifically required in GABAergic neurons for trafficking GABA into vesicles[13], which only account for ~10% of Drosophila neurons[14]. Surprisingly, our analysis of existing single cell RNA-seq data [15, 16] revealed that all Drosophila neurons of the visual system (optic lobe) contain mRNA encoding VGAT (Fig. 1A). Furthermore, data collected from the FlyCell Atlas[17], which covers each region of the Drosophila nervous, suggests all neurons of the fly transcribe VGAT. Using mRNA in-situ hybridization, we confirmed that, although VGAT mRNA levels are higher in the GABAergic lineage, it is present in all neurons of the optic lobe (Fig. 1B).

Figure 1. VGAT expression is determined by repression of its translation outside of GABAergic neurons.

A) Normalized expression per cluster of VGAT and Gad1 from scRNA-seq of the adult optic lobe[15]. Gad1 expressing clusters (red) contain higher levels of VGAT mRNA than Gad1-negative clusters (black). Annotated clusters corresponding to GABAergic neurons in the optic lobe are labeled.

B) RNA HCR-FISH in the adult optic lobe shows VGAT mRNA (yellow) is not limited to GABAergic (Gad1+, magenta) neurons.

C) Antibody labeling shows VGAT protein (yellow) is present in cell bodies of GABAergic neurons, marked by Gad1-LexA expression (magenta).

D) Expression of a reporter for VGAT translation is restricted to GABAergic neurons, marked by Gad1 mRNA. VGAT-T2A-Gal4 contains an insertion of T2A-Gal4 in the endogenous VGAT coding sequence, resulting in Gal4 only being expressed in neurons that translate VGAT.

E) Expression of a reporter for VGAT transcription (yellow) is not restricted to GABAergic neurons (GABA+ magenta). VGAT-lexA contains a replacement of the endogenous VGAT coding sequence with a LexA coding sequence followed by the Sv40 and Hsp70 transcriptional terminators.

F) Expression of a transgenic GFP for sensor the VGAT 3’UTR (green) is restricted to GABAergic neurons (GABA+ magenta)

To determine if the VGAT mRNA in non-GABA neurons is translated into protein, we used Antibody labeling that detects mature VGAT protein, and an allele containing a T2A-Gal4 insertion in frame with the C-terminus of VGAT, which provides a sensitive read out of VGAT translation (through Gal4 driving UAS-GFP) [18]. Antibody staining in the presence of a genetically labeled GABA marker, Gad1-LexA, revealed that VGAT protein is exclusively present in GABA neurons (Fig. 1C). Imaging of Gal4 activity via UAS-GFP in the VGAT-T2A-Gal4 line confirmed a specific expression pattern within the GABAergic lineage (Fig. 1D). As the Gal4 protein is separated from VGAT via the 2A peptide, it is not subjected to any post-translational regulation that may affect VGAT. We can therefore conclude that VGAT mRNA is not translated outside of GABA neurons.

To determine whether the difference in VGAT mRNA levels between GABA vs non-GABA neurons is mediated by differences in transcription or post-transcriptional regulation, we made use of an existing allele in which the VGAT coding sequence was replaced by homologous recombination with the LexA transcriptional activator[19]. Importantly, the LexA coding sequence is immediately followed by two exogenous transcriptional terminators, and therefore is not subjected to post-transcriptional regulation through the VGAT 3’UTR, although any potential inputs of the 5’UTR are preserved. Expression of LexA activity in the optic lobe of flies containing this allele revealed a uniform expression pattern in neurons, providing strong evidence that VGAT is transcribed pan-neuronally and requires post-transcriptional inputs for restricted expression in GABAergic neurons (Fig. 1E).

Finally, to confirm that post-transcriptional regulation through the 3’UTR is sufficient for restricting VGAT expression, we constructed a VGAT-3’UTR sensor in which the Tubulin promoter drives ubiquitous transcription of GFP with the 3’UTR of VGAT. Expression of GFP protein from this reporter was restricted to the GABAergic lineage, confirming that VGAT 3’UTR restricts expression spatially in the Drosophila nervous system (Fig. 1F).

The VGAT 3’UTR contains complementary sequences to a GABAergic-specific non-coding RNA

We next sought to determine the mechanism by which VGAT translation is active specifically in GABAergic neurons. The 3’UTR of VGAT contains three predicted binding sites for the microRNA miR-7, the most of any gene in the Drosophila genome. Each of these is a perfect complement to the miR-7 seed sequence, has an additional Adenosine following the seed which promotes stability of binding by the microRNA-AGO complex[20], and all are conserved within other Drosophila species (Fig. S1). miR-7 is expressed in the Drosophila nervous system based on existing small RNA sequencing[21, 22], and we confirmed its activity in neurons using a miR-7 genetic sensor [23] (Fig. S2). However, we also found that miR-7 is active within GABA positive neurons, suggesting an additional factor is required that potentially antagonizes the interaction between miR-7 and VGAT mRNA.

One potential mechanism in which a microRNA targeting can be influenced locally is by RBPs which can either enhance binding cooperatively or inhibit binding through competition[24]. However, our analysis of a comprehensive scRNA-seq dataset[15, 16] of the optic lobe failed to identify any RBP candidates that were exclusively expressed, or specifically absent, in the GABAergic neuron population.

A second possibility we considered is based on the observation of antisense transcripts that promote stabilization of their sense mRNA counterpart through complementary binding of their 3’UTR[25]. While VGAT does not contain antisense transcripts, we performed a BLAST search using the full VGAT 3’UTR to identify other transcripts that could potentially interact with VGAT mRNA through complementary base pairing. The top scoring hit was within the 3’UTR of an uncharacterized gene named CG14989, which shares a 22 nucleotide stretch of perfect complementarity (Fig. 2A) as well as an additional two smaller regions of complementarity (see below). The CG14989 gene is positioned in the genome directly next to and transcribed divergently from the Gad1 gene. Gad1 is transcribed exclusively in GABA neurons, where it is required for synthesizing GABA from Glutamate. Analysis of scRNA-seq data revealed that CG14989 RNA shares this pattern (Fig. 2B), suggesting the two genes may share regulatory elements. Then, using RNA in-situ hybridization we confirmed that CG14989 marks all GABA neurons in the optic lobe with a perfectly matching cell expression pattern to Gad1 (Fig. 2C).

Figure 2. Complementary regions between VGAT and CG14989.

A) Complementary regions between the VGAT 3’UTR and CG14989 RNA. Predicted miR-7 target sites indicated (orange bar). Optic lobe neuronal (ELAV population [35]) expression of Gad1 and CG14989 (bottom) shows that the extended isoform of CG14989 which contains all three complementary regions is expressed in neurons.

B) Normalized expression per cluster of CG14989 and Gad1 from scRNA-seq of the adult optic lobe[15]. Gad1 expressing clusters (red) contain higher levels of CG14989 mRNA than Gad1-negative clusters (black). Annotated clusters corresponding to GABAergic neurons in the optic lobe are labeled.

C) RNA HCR-FISH of CG14989 (green) and Gad1 (magenta) in the Adult optic lobe shows an overlapping expression pattern.

D) Ribosome profiling of neurons (ELAV+, whole brain) showing Translation efficiency vs RNA levels[26].

Strikingly, the position of each complementary region directly overlaps a predicted miR-7 target site (encompassing the 7 bp seed target plus trailing Adenosine) in the VGAT 3’UTR (Fig. 2A), i.e. CG14989 contains three sequences identical to the miR-7 seed perfectly coinciding with the three target sites within the VGAT 3’UTR. One of these is embedded in the longer 22 nucleotide complementary region, while the other two encompass the entirety of the predicted miR-7 target site without further complementarity beyond this.

Although CG14989 is annotated as a coding mRNA, several pieces of evidence support that it may in fact encode a non-coding RNA, at least in neurons. First, the isoform expressed in neurons is annotated to encode a protein utilizing the non-canonical start codon of Leucine instead of Methionine, which is extremely rare in eukaryotes. The first Methionine-encoding AUG codons within this isoform contain stop codons shortly following them. Second, there are no experimentally identified peptides in Drosophila proteomic databases corresponding to CG14989 despite its RNA being present at high levels in the nervous system. This contrasts with Gad1, which is well represented in protein databases despite being expressed in the same cell type and at comparable mRNA levels compared to CG14989. Finally, the strongest evidence for the lack of coding potential of CG14989 comes from our analysis of existing Ribosome profiling performed in Drosophila adult neurons[26]. CG14989 is amongst the lowest scoring genes in the transcriptome for translation efficiency, which is the ratio of transcripts to Ribosome protected fragments (Fig. 2D). In summary, CG14989 is a likely non-coding RNA exclusively expressed in GABAergic neurons and contains multiple regions of complementarity to the VGAT 3’UTR that directly overlap miR-7 target sites.

CG14989 regulates translation of VGAT mRNA

Given the presence of multiple complementary regions to VGAT mRNA in CG14989 that overlap predicted miR-7 sites, and the expression of CG14989 exclusively in neurons where VGAT is translated into protein, we next tested the hypothesis that CG14989 can de-repress VGAT mRNA translation. To ectopically activate CG14989 in non-GABAergic neurons, we utilized the CRISPR activation (CRISPRa) system, in which transcriptional activators are recruited to the CG14989 promoter using sequence-specific guide RNAs[27, 28]. We expressed CG14989 in clones throughout the Drosophila visual system and investigated VGAT protein levels in regions that do not contain GABA neurons including the Lamina cortex[29] and a region of the Lobula Plate that produces the cholinergic T4/T5 neurons[30, 31]. We found widespread induction of VGAT protein within CRISPRa clones in both lamina and T4/T5 neurons, suggesting that CG14989 is sufficient for promoting the translation of VGAT mRNA in non-GABAergic neurons (Fig. 3A). Next, we generated a modified VGAT 3’UTR sensor in which all three complementary regions (8bp + 8bp + 22bp), and therefore all predicted miR-7 binding sites, were deleted from the VGAT mRNA 3’UTR. Loss of these sites led to complete de-repression of sensor expression in the visual system, confirming that these regions are required for repression in the non-GABAergic population (Fig. 3B).

Figure 3. Expression of CG14989 is sufficient for derepression of VGAT.

A-A’) CRISPRa expressing clones (Red) in the Lobula Plate (A) and Lamina cortex (A’) generated randomly by heatshock-dependent Flippase with antibody labeling of VGAT (Green). Dotted line in (A) marks the lower limit of the C2/C3 domain, two types of GABAergic projection neurons.

B-B’) VGAT 3’UTR Sensor with wildtype (B, as in Fig. 1F) and deleted complementary sequences (B’)

Complementarity between CG14989 and other mRNAs involved in GABA metabolism and signaling

We next asked whether other mRNAs outside of VGAT contain complementary sequences to CG14989 within their respective 3’UTRs, which may suggest a similar type of GABAergic-specific translational regulation. Strikingly, a blast search of CG14989 revealed that mRNAs that encode conserved proteins involved in, both directly and indirectly, nearly every aspect of GABA metabolism and signaling contain complementary regions (Fig. 4A–B). These include 1) P5CDh1 (ALDH4A1) and Glts (glutamate synthetase) that are both involved in the synthesis of Glutamate, the precursor of GABA from proline or glutamine, respectively 2) sgll (PNPO) involved in the synthesis of PLP, an essential cofactor of Gad1, 3) stol (CACNA2D3/4), a component of calcium voltage-gated channel, 4) Gfrl: A predicted GDNF receptor homolog and 5) Gabat (ABAT) and ScsβG (SUCLG2) that are components of the GABA shunt pathway that breaks down GABA. Taken together with our findings that CG14989 regulates the translation of VGAT, this suggests that CG14989 functions as a novel ‘operon-like’ coordinator of GABA levels and signaling in Drosophila.

Figure 4. Other GABA-related genes contain complementary bases to CG14989.

A) Schematic showing GABA metabolism and signaling. Genes with complementary sequences in their 3’UTRs to CG14989 shown in black.

B) Sequence alignments of CG14989 and GABA-related genes. Predicted miR-7 target sites in orange.

Discussion

Our findings suggest a novel mechanism for coordinating expression in the nervous system by linking restricted transcription with cell-type specific translation of functionally related genes through the activity of lncRNAs. Specifically, we demonstrate that a lncRNA transcribed divergently in GABAergic neurons, likely under control of the same regulatory sequence, from the gene encoding the enzyme required for synthesis of GABA, Gad1, is sufficient for translation of the GABA vesicular VGAT, which is transcribed pan-neuronally in the Drosophila nervous system. As a result, two essential components of GABA signaling are produced together in the proper neuronal context.

Two major types of ncRNA-miRNA interactions have been characterized in mRNA regulation. One involves competing endogenous RNAs (ceRNAs), which contain microRNA binding sites and act as “decoys” to reduce the availability of microRNAs for their mRNA targets, as demonstrated for the circular RNA CDR1as[32]. The other involves antisense transcription of a microRNA target gene, producing a lncRNA with sequence complementarity that competes for microRNA binding, thereby modulating mRNA stability[25].

Our data suggest that CG14989 may function in a manner analogous to the latter mechanism but with two important distinctions that have broader implications for the coordination of functionally related genes. First, because CG14989 ultimately determines cell-type specificity of VGAT translation, the regulatory constraints on VGAT transcription are loosened, as evident by its pan-neuronal transcription pattern. Second, because this mechanism does not require antisense transcription, it provides a more flexible framework for coupling transcriptional and translational regulation of gene pairs, potentially enabling broad coordination of gene expression programs within GABAergic neurons. The implications of this are not only apparent for VGAT, but also for the various other GABA-related mRNAs with complementary sites to CG14989.

It should be noted that miR-7 target sites within the Drosophila bearded gene family, a group of genes that are regulated by, and in turn antagonize Notch signaling, also contain complementary sequences within several proneural mRNAs, which encode transcription factors that specify neural identity from epidermal precursors[33, 34]. Bearded and proneural genes are co-expressed in many neurogenic contexts, and while the role of these complementary sequences in the context of proneural fate remains to be investigated, it is possible that they perform a similar function to the case we describe for VGAT and CG14989. Taken together, these observations have two broader implications: (1) the regulatory mechanism uncovered for VGAT by CG14989 may be a more general strategy employed across systems, and (2) such interactions could involve both lncRNA-mRNA and mRNA-mRNA base-pairing.

CG14989 is present in all insects with conserved synteny and orientation in relation to Gad1, suggesting its role in regulating gene expression in GABAergic neurons may be evolutionarily conserved. Furthermore, mammalian genomes also contain a lncRNA that is transcribed divergently from the beginning of the GAD1 gene, like the case of insects. Critically, a recent survey of developmental gene expression in several mammalian species found that this lncRNA was significantly correlated with GAD1 in multiple species including humans[10].

In summary, our findings reveal a previously unrecognized mechanism for coordinating genes involved in the essential process of GABA signaling.

Materials and Methods

Fly strains and fly rearing

Flies were reared on molasses-cornmeal-agar media at 25°C. The fly lines used are described in Supp. Table 1. VGAT 3’UTR sensor plasmids were injected by BestGene Inc. in the 3rd chromosome (attp2 landing sites). For generating ectopic CRISPRa clones, pupa (~48h APF) were heat shocked at 38°C for 15min and then reared at 25°C until one day after adult eclosion, when they were dissected. In all the experiments at least 3 different brains were analyzed.

Immunohistochemistry

Flies were anesthetized on ice and brains dissected in ice-cold Schneider’s media and then transferred to Schneider’s media on ice. Brains were fixed using 4% paraformaldehyde diluted in PBS 1X for 20 min at room temperature. Then, they were rinsed 3 times in 0.5% Triton-X diluted in PBS 1X (PBTx), washed for 30 min in PBTx, and incubated for 30 min in PBTx with 5% goat serum (PBTx-block) at room temperature. They were then incubated at 4°C for 2 overnights in primary antibodies diluted in PBTx-block. They were then rinsed 3 times in PBTx, incubated two times for 30 min in PBTx, and incubated at 4°C overnight in secondary antibodies diluted in PBTx-block with 5ug/mL DAPI. Finally, they were rinsed 3 times in PBTx, incubated two times for 30 min in PBTx, and mounted in Slowfade before imaging with either a Leica SP8 or Zeiss NLO 980 confocal microscope using a 63x glycerol objective. Image stacks were acquired at 0.8–1 μm optical sections. Images were processed in Fiji and Adobe Illustrator. The antibodies used are described in Supp. Table 1.

Hybridization Chain Reaction Fluorescence In Situ Hybridization (HCR-FISH)

Flies were anesthetized on ice and brains dissected in ice-cold PBS and then transferred to PBS media on ice. Brains were fixed using 4% paraformaldehyde diluted in PBTx 1X for 20 min at room temperature. Then, they were washed 1 time in PBTx for 15 min, and then washed 2 times in 0.5% Tween-20 diluted in PBS 1X (PBTw). They were then incubated at 37°C for 30 min in Hybridization solution (Molecular Instruments) with shaking. Following pre-hybridization, the brains were incubated in hybridization solution with RNA-specific probes (diluted 1/100) at 37°C overnight with shaking. The following day, brains were washed 4 times in Probe Wash Buffer (Molecular Instruments) at 37°C for 15 min on a rocker. They were then incubated for 30 min at room temperature in Amplification Buffer (Molecular Instruments) with rocking, followed by Amplification Buffer with fluorescently coupled amplifiers synthesized by Molecular Instruments (Diluted 1/50) and incubated at 25°C for 4 hours in the dark. Next, they were washed 5 times for 15 min in 0.1% Tween-20 diluted in 5x sodium chloride sodium citrate (SSCt). Brains were then incubated in PBSTw containing 5ug/mL DAPI and GFP booster (only for VGAT-T2A-Gal4>GFP) before two final washes in PBSTw for 30 min and mounting in SlowFade medium. The VGAT HCR-probe was synthesized by Molecular Instruments. Gad1 and CG14989 probe sets were designed using HCR Probe Maker[36] and ordered as oligo pools from IDT. Sequences of Gad1 and CG14989 probes are listed in Supp. Table 2.

Molecular Biology

To generate the VGAT 3’UTR sensors, we first modified pJFRC177-10XUAS-FRT>-dSTOP-FRT>-myr::GFP (addgene #32149) to make Tubulin-2xMsGFP2-PEST-NLS, with the coding sequence followed by an Sv40 terminator flanked by SapI restriction sites [Plasmid RLV72]. Wild-type and mutant versions of the VGAT 3’UTR were synthesized by Genewiz (Azenta) and inserted via Golden-Gate SapI assembly. The plasmid sequences can be found in Supp. Table 2.