Stochastic RNA Editing of the Complexin C-Terminus Within Single Neurons Regulates Neurotransmitter Release

Characterizing the functional significance of alternative splicing of Cpx exon 7

Before characterizing the impact of RNA editing on Cpx7A function, we first examined the role of Cpx alternative splicing to gain greater insight into the endogenous roles of Cpx7A and Cpx7B. In cpx null mutants that lack both isoforms (cpx^SH1), spontaneous mini frequency is dramatically elevated (>50-fold), evoked release is decreased, and larval NMJ synaptic growth is enhanced⁷. Overexpression of either Cpx7A or Cpx7B in cpx^SH1 partially rescues these phenotypes, with Cpx7A having more robust clamping properties and Cpx7B over-rescuing evoked release¹⁶. Although both variants support aspects of Cpx function when overexpressed, endogenous Cpx7A mRNA is >100-fold more abundant than Cpx7B in Drosophila adults and larvae based on quantitative RT-PCR¹⁶. As such, Cpx7A is hypothesized to play a more critical role in synaptic transmission. To directly test the endogenous function of the two splicing isoforms, CRISPR mutants disrupting 7A (cpx^Δ7A) or 7B (cpx^Δ7B) were generated by introducing an early stop codon at the beginning of exon 7A or 7B, respectively. Western analysis of adult head lysates with a Cpx antibody that recognizes both variants showed an 87% reduction in overall Cpx expression in cpx^Δ7A (p < 0.0001) and a milder 10% reduction in cpx^Δ7B (p = 0.9398), consistent with Cpx7A being the predominant isoform (Figure 2A). Immunostaining for Cpx at 3^rd instar larval NMJs showed a similar effect (Figure 2B, ​,C),C), with a >85% reduction in Cpx at synapses in cpx^Δ7A mutants (p < 0.0001) and no detectable decrease in cpx^Δ7B mutants (p = 0.9567).

To assay functional requirements for the splice variants in vivo, adult and larval motor behavior were examined in cpx^Δ7A and cpx^Δ7B and compared to control and cpx^SH1 nulls. Complete loss of Cpx in cpx^SH1 severely disrupts behavior and reduces viability, with the few escaper adults that emerge from the pupal case displaying a profound loss of motor control and an inability to walk in a coordinated manner. Loss of the predominant Cpx7A isoform also strongly disrupted motor behavior, though not as severely as cpx^SH1. In contrast to cpx^SH1 where homozygous adults are rarely observed, cpx^Δ7A could be maintained as a homozygous stock, indicating the remaining endogenous Cpx7B can improve viability and fertility compared to animals lacking both isoforms. However, cpx^Δ7A adults showed a complete inability to climb in a negative geotaxis assay (Figure 2D). In contrast, cpx^Δ7B adults were fully viable and did not display obvious motor defects. In addition, cpx^Δ7B adults were moderately hyperactive in climbing compared to controls (control: 50.4 ± 8.3% pass rate, n = 8 cohorts of ten flies; cpx^Δ7A: 0 ± 0% pass rate, n = 8 cohorts of ten flies, p < 0.0001 to control; cpx^Δ7B: 79 ± 5.5% pass rate, n = 8 cohorts of ten flies, p = 0.0055 to control, p < 0.0001 to cpx^Δ7A). To examine larval locomotion, crawling velocity of 3^rd instar larvae was assayed in cpx^Δ7A and cpx^Δ7B mutants (Figure 2E). Similar to cpx^SH1, cpx^Δ7A larvae displayed a large reduction in crawling velocity (p = 0.0011). In contrast, cpx^Δ7B mutants showed only a mild decrease in velocity (p = 0.1155). Together, these data indicate Cpx7A has a more prominent role in supporting larval and adult motor behavior.

To examine synaptic morphology and neurotransmitter release in cpx^Δ7A and cpx^Δ7B mutants, the well-characterized 3^rd instar larval glutamatergic NMJ preparation was used²⁹. Synaptic bouton and active zone (AZ) number were quantified by immunostaining for neuronal membranes (anti-HRP) and the AZ protein Bruchpilot (Brp) (Figure 2B, ​,F,F, ​,G).G). In contrast to the large increase in bouton (72%, p = 0.0012) and AZ (65%, p < 0.0001) number in cpx^SH1, synaptic growth was largely unaffected in cpx^Δ7A (AZ#: p = 0.2809, bouton#: p = 0.4576) and cpx^Δ7B (AZ#: p = 0.3121, bouton#: p = 0.7781) mutants. Despite the differences in overall levels of the two splice isoforms at NMJs (Figure 2B), cpx^Δ7A mutants had similar NMJ morphology to cpx^Δ7B (p = 0.9999). We conclude that Cpx7A and Cpx7B are functionally redundant in their ability to regulate synaptic growth, with Cpx7B having a more important role than expected based on its lower expression level.

To assay synaptic function, two-electrode voltage-clamp (TEVC) was used to measure evoked and spontaneous neurotransmitter release at 3^rd instar larval muscle 6 in abdominal segment A3. In contrast to the dramatic increase in spontaneous release rate observed in cpx^SH1 (>55-fold), endogenously expressed Cpx7A or Cpx7B was able to substantially lower mini frequency (Figure 3A, ​,B).B). The presence of only Cpx7A in the cpx^Δ7B mutant returned spontaneous release rates to control levels (p = 0.2641). The residual Cpx7B in cpx^Δ7A mutants was not able to fully clamp spontaneous fusion, with a 2-fold increase compared to controls (p < 0.0001). In contrast to spontaneous release, the presence of only one of the two splice isoforms was not as effective in driving normal levels of evoked fusion (Figure 3C, ​,D).D). A 71% reduction in the peak amplitude of the evoked excitatory junctional current (eEJC) was observed in cpx^SH1 compared to control (p < 0.0001). Although less severe, cpx^Δ7A displayed a 46% reduction (p < 0.0001) and cpx^Δ7B a 25% reduction (p = 0.0269) compared to controls. These data indicate Cpx7A and Cpx7B are both required for wildtype evoked responses, with the higher levels of Cpx7A in cpx^Δ7B mutants supporting more evoked fusion than the lower levels of Cpx7B in cpx^Δ7A.

In addition to the total number of SVs that fuse during an action potential, Cpx also modulates SV release kinetics by promoting fast synchronous fusion and reducing the slower asynchronous pathway⁴. While peak eEJC amplitude primarily captures synchronous release, eEJC charge, half-width, and the time course of cumulative release provide kinetic insights into synchronous and asynchronous fusion. These evoked variables were compared across control, cpx^Δ7A, cpx^Δ7B and cpx^SH1 NMJs (Figure 3E–I). Like eEJC amplitude, release kinetics in cpx^Δ7B mutants were more similar to controls. Loss of Cpx7A in the cpx^Δ7A line resulted in reduced charge transfer (Figure 3E) and a mild increase in asynchronous release (Figure 3F, ​,G),G), though less than the large amount of asynchronous fusion in cpx nulls. The onset of synchronous release was slightly enhanced in cpx^Δ7A mutants (Figure 3G, ​,H),H), consistent with prior data showing overexpression of Cpx7B enhanced the speed of onset of evoked fusion⁴. In summary, we conclude endogenous levels of either Cpx isoform are sufficient to clamp spontaneous SV fusion. The requirement for both isoforms to fully recapitulate evoked responses indicate they have both shared and independent roles in Ca²⁺-dependent SV release. Given the more severe behavioral defects observed in animals lacking Cpx7A, some neuronal subtypes are likely to be more highly reliant on this splice variant for supporting synaptic transmission, in contrast to motoneurons.

Single-cell RNAseq reveals stochastic RNA editing of Cpx7A in larval motoneurons

Having analyzed functions for the two C-terminal splice isoforms of Cpx, we next examined RNA editing profiles for Cpx7A in larval motoneurons. Larval abdominal muscles are innervated by several populations of glutamatergic motoneurons in Drosophila, including tonic Type Ib and phasic Type Is subclasses^30,31. Type Ib and Is motoneurons are the primary drivers of muscle contraction and display unique morphological and functional properties^32–37. Given Ib and Is neurons have distinct presynaptic release output, we hypothesized RNA editing of Cpx7A could contribute to these differences if: (1) Cpx7A edit variants were differentially expressed between the two neuronal populations; and (2) editing of Cpx7A altered its function. To determine the abundance and diversity of Cpx7A edit variants (I125M, N130G, N130S, N130D), single-cell RNAseq datasets we previously generated from ~200 larval Ib and Is motoneurons were analyzed³³. For these experiments, whole cell electrodes were used to collect cytosolic and nuclear content from individual Ib or Is neurons that expressed GFP using Gal4 drivers specific to each cell type³². High-resolution paired-end deep single-cell RNA sequencing was performed on RNA extracted from each individual motoneuron, generating ~4 million reads per cell and allowing identification of RNA editing diversity at single neuron resolution. Prior studies using mRNA obtained from pooled neuronal populations proposed RNA editing occurs in an “all-or-none” fashion at individual edit sites within a cell. For example, more than 99% of mammalian AMPA GluA2 receptor subunit transcripts undergo RNA editing at specific times during brain development^38,39. Given our method provides individual neuron resolution of RNA editing, we could directly test the “all-or-none” model. Strikingly, individual larval motoneurons showed highly stochastic RNA editing of the three adenosine bases (position 375, 388 and 389) that are subject to possible editing in Cpx exon 7A. Editing rates for these adenosines ranged from 0 to 100% across the ~200 neurons, with 98% of Ib cells and 94% of Is cells showing some level of exon 7A editing (Figure 4A, ​,B).B). For the adenosine 375 edit site, the average I125M (adenosine 375 to inosine 375) edited transcript level per neuron was 31.7 ± 2.4% in Ib (n = 95 cells) and 30.3 ± 2.4% in Is (n = 86 cells). Only 3% of Ib and 1% of Is neurons fully edited all Cpx mRNA to I125M. For all Cpx7A mRNA edit variants, the average Ib motoneuron expressed 53% unedited Cpx, 32% Cpx^I125M, 0.0% Cpx^N130D, 1.5% Cpx^N130S, 0.3% Cpx^N130G, 0.2% Cpx^I125M,N130D, 9.5% Cpx^I125M,N130S and 3.7% Cpx^I125M,N130G (Figure 4A, ​,B).B). The average Is motoneuron expressed a similar ratio of edited Cpx7A transcripts (Figure 4A, ​,B).B). As such, differential RNA editing of Cpx7A is unlikely to drive the distinct release properties of Ib and Is motoneurons, though stochastic editing of Cpx7A could contribute to individual neuronal heterogeneity in presynaptic output. We conclude that RNA editing of Cpx exon 7A is stochastic across individual edit sites and generates similar diversity of edited isoforms in Is and Ib motoneurons.

Prior studies proposed editing at adenosine 375 serves to enhance exon-intron base pairing within the 7A pre-spliced mRNA to generate a more favorable double-stranded RNA structure for ADAR to edit the downstream adenosines 388 and 389 that form the AAT codon (N130)¹⁶. Consistent with this model, RNAseq data from single motoneurons showed editing for both I125 and N130 was far more common than single edits to N130 alone (Figure 4A). Ib neurons expressing Cpx^I125M,N130S mRNA were six-fold more abundant that those expressing Cpx^N130S alone (I125M, N130S: 9.5 ± 1.5 edit % per cell; N130S: 1.47 ± 0.51 edit % per cell, p = 0.0035). A similar ratio was observed in Is neurons (Figure 4A, ​,B).B). In the 56% of Ib cells expressing Cpx^I125M,N130S edited transcripts, an average of 17% of total cpx mRNA were of this variant, similar to the 16% of total cpx mRNA in the 59% of Is cells that expressed Cpx7A^I125M,N130S. Rarely, Cpx^I125M,N130S represented the only Cpx mRNA detected within a neuron (Figure 4A). In contrast to the more abundant Cpx^I125M,N130S, only 6% of Ib neurons expressed Cpx^I125M,N130D and it represented just 3% of the total cpx mRNA in these cells. The Cpx^I125M,N130G variant, which requires A-to-I editing at all three adenosines, was observed in 40% of Ib neurons, representing 9% of total cpx mRNA in cells in which it was expressed. Similar patterns were observed in Is neurons (Figure 4A, ​,B).B). For the most abundant edit variant (Cpx^I125M), no correlation of editing percentage at this site and expression levels of adar mRNA in that cell was observed for Ib or Is motoneurons (Figure 4C, ​,D).D). We conclude that Cpx^I125M,N130S is the highest expressed variant in larval motoneurons that alters the N130 residue on the amphipathic helix in Cpx7A.

To determine if RNA editing of other synaptic target genes showed similar stochastic single neuron editing, several additional mRNAs known to undergo RNA editing were examined in the single neuron RNAseq dataset. RNA editing percentages at three sites (N15D, R19G, R20G) within Synapsin (Syn) and one site (M244V) in Syntaxin 1A (Syx1A) were compared in the same motoneurons that edited Cpx7A to I125M (Figure 4E). Both Syn and Syx1A displayed stochastic editing rates that ranged from 0–100% across motoneurons. For example, Syn R20G editing was observed at an average of 55 ± 11.6% (n = 11 Is cells), while Syx1A M244V was edited in the same neurons at a rate of only 5.4 ± 3.8% per cell (n = 11 Is cells). Similar to Cpx, no significant difference in editing percent of Syn or Syx1A was observed between Ib and Is neurons (Figure 4E). We conclude ADAR-mediated RNA editing is not all-or-none in individual Drosophila motoneurons, with stochastic editing at single adenosine base sites and across multiple mRNAs having the potential to generate unique synaptic proteomes within the same population of neurons.

Transgenic constructs

QuikChange Lightning (Agilent) was used for site-directed mutagenesis on unedited Cpx7A to generate specified Cpx edit variants that were subcloned into modified pValum construct, as previously described¹¹. The resulting constructs were injected into a yv;;attP 3^rd chromosome docking strain by BestGene Inc. (Chino Hills, CA, USA). UAS lines were recombined into the cpx^SH1 null mutant background and elav^C155-GAL4 (BDSC #8765) was used for pan-neuronal expression of transgenes.

Generation of CRISPR-modified Cpx strains

Two endogenous Cpx truncation lines were generated (cpx^Δ7A and cpx^Δ7B) using a CRISPR genome engineering approach. Four guide RNAs (gRNAs) flanking the splice acceptor site of exon 7A or 7B were selected using the CRISPR Optimal Target Finder⁶⁵. gRNAs were cloned into the pCFD5 expression vector (Addgene #73914)⁶⁶ and donor constructs were generated to encode a floxed P3>DsRed reporter cassette (Addgene #51434) in the reverse orientation flanked with one kb homology arms upstream and downstream of the splice acceptor site of either exon 7A or 7B by Gibson assembly protocol using NEBuilder HighFidelity DNA Assembly Cloning Kit (E5520). An early stop codon was inserted between homology arms for each respective exon construct, with several amino acid coding sequences maintained to preserve proper exon splicing. gRNA binding sites of donor template were mutated using silent mutations that did not alter amino acid sequence. Template and gRNA plasmids were co-injected into vasa-Cas9 embryos (BDSC #56552) by BestGene Inc (Chino Hills, CA, USA) and Ds>Red positive transformants were selected by BestGene Inc (Chino Hills, CA, USA). The modified locus with stop codons inserted into exon 7A or 7B were confirmed by sequencing.

Locomotion analysis

Adult geotaxis was measured in adult male flies aged 2–3 days as previously described⁶⁷. Briefly, eight cohorts of ten adult males (80 total flies) per genotype were separated after eclosion and allowed to recover from CO₂ for 24 hours on standard fly medium. After 24 hours, each cohort was moved to a chamber made from two clear plastic vials taped together, with a line drawn around the lower vial 8 cm from the bottom. Each cohort was allowed to acclimate in the chamber for five minutes before assays began. Negative geotaxis was measured as percent of the cohort that crossed the 8 cm line within ten seconds after being tapped to the bottom of the chamber. Each cohort was subjected to ten rounds of negative geotaxis assay, with a one-minute rest period between each. The percent of flies that crossed the 8 cm line after each round was averaged to produce a pass rate per cohort. Larval crawling was assayed in 3^rd instar larvae of both sexes, as previously described^68,69. Larvae were briefly washed in room-temperature water before placing onto the center of a 5 cm petri dish containing 2% agarose, with five animals from a single genotype placed together (n = 10 larvae per genotype). The petri dish was placed over a grid and velocity was measured as average distance (in mm) traveled during the first 30 seconds following placement.

Immunohistochemistry

Larvae were dissected in hemolymph-like HL3.1 solution (in mM: 70 NaCl, 5 KCl, 4 MgCl₂, 10 NaHCO3, 5 trehalose, 115 sucrose, 5 HEPES, pH 7.2) and fixed in 4% paraformaldehyde for 18 minutes. Larvae were washed three times for five minutes with PBST (PBS containing 0.1% Triton X-100), followed by a thirty-minute incubation in block solution (5% NGS in PBST). Fresh block solution and primary antibodies were then added. Samples were incubated overnight at 4°C and washed with two short washes and three extended 20 minutes washed in PBST. PBST was replaced with block solution and fluorophore-conjugated secondary antibodies were added. Samples were incubated at room temperature for two hours. Finally, larvae were rewashed with PBST and mounted in Vectashield (Vector Laboratories, Burlingame, CA). Antibodies used for this study include: mouse anti-Brp, 1:500 (NC82; Developmental Studies Hybridoma Bank (DSHB), Iowa City, IA)); rabbit anti-Cpx, 1:5000⁷; goat anti-rabbit Alexa Fluor 488, 1:500 (A-11008; ThermoFisher Scientific, Waltham, MA, USA); goat anti-mouse Alexa Fluor 546, 1:500 (A-11030; ThermoFisher); DyLight 649 conjugated anti-HRP, 1:500 (#123–605-021; Jackson Immuno Research, West Grove, PA, USA).

Confocal imaging and imaging data analysis

Imaging was performed on a Zeiss Pascal confocal microscope (Carl Zeiss Microscopy, Jena, Germany) using a 63X 1.3 NA oil-immersion objective (Carl Zeiss Microscopy). Images were processed with the Zen (Zeiss) software. A 3D image stack was acquired for each NMJ imaged (muscle 4 Ib NMJ of abdominal segment A3) and merged into a single plane for 2D analysis using FIJI image analysis software⁷⁰. No more than two NMJs were analyzed per larva. Anti-HRP labeling was used to identify neuronal anatomy (axons and NMJs) and quantify synaptic bouton number and NMJ area. Brp puncta quantification was used to measure AZ number. Muscle 4 area was used to normalize quantifications for muscle surface area. For Cpx fluorescence quantification, the HRP-positive area was used to outline NMJs and axons. Total Cpx fluorescent intensity was measured in the outlined area, with background fluorescence of mean pixel intensity of non-HRP areas subtracted. For NMJ/axon ratios, background subtracted mean NMJ Cpx fluorescence was compared to background subtracted mean axon Cpx fluorescence within the same image.

Two-electrode voltage-clamp electrophysiology

Postsynaptic currents were recorded from 3^rd instar muscle 6 at segment A3 using two-electrode voltage clamp with a −80 mV holding potential. Experiments were performed in room temperature HL3.1 saline solution (in mM, 70 NaCl, 5 KCl, 10 NaHCO3, 4 MgCl2, 5 trehalose, 115 sucrose, 5 HEPES, pH 7.2). Final [Ca²⁺] was adjusted to 2 mM unless otherwise noted. Motor axon bundles were cut and suctioned into a glass electrode and action potentials were stimulated at 0.5 Hz (unless indicated) using a programmable stimulator (Master8, AMPI; Jerusalem, Israel). Data acquisition and analysis was performed using Axoscope 10.0 and Clampfit 10.0 software (Molecular Devices, Sunnyvale, CA, USA) and inward currents were labeled on a reverse axis for clarity.

Western blot analysis

Western blotting of adult head lysates (three heads per sample with one head loaded per lane) was performed using standard laboratory procedures with mouse anti-Tubulin (T5168; Sigma) at 1:10000 (UAS rescue experiments) or 1:1000000 (CRISPR experiments) and rabbit anti-Cpx at 1:5000. IRDye 680LT-conjugated goat anti-mouse, 1:5000 (926–68020; LICOR) and IRDye 800CW conjugated goat anti-rabbit, 1:5000 (926–32211; LICOR) were used as secondary antibodies. Blocking was performed in a solution containing four parts TBS (10 mM Tris Base pH 7.5, 150 mM NaCl) to one part Blocking Buffer (Rockland) for one hour. Antibody incubations were performed in a solution containing four parts TBST (1X TBS with 1% Tween-20) to one part Blocking Buffer. A LI-COR Odyssey Imaging System (LI-COR Biosciences, Lincoln, MA, USA) was used for visualization and analysis was performed using FIJI image analysis software. Relative Cpx expression was calculated by normalizing to Tubulin intensity.

Purification of Complexin for in vitro phosphorylation assays

QuikChange Lightning (Agilent) was used for site-directed mutagenesis of unedited Cpx7A to generate Cpx7A^I125M,N130S (termed N130S). Recombinant Cpx fused with GST was expressed in E. coli (BL21) and purified using glutathione sepharose 4B (Fisher Scientific). Peak fractions were concentrated and further purified by gel filtration as previously described¹¹. In vitro kinase assays were performed using purified recombinant Cpx proteins and the catalytic subunit of CK2 (C70–10G, SignalChem). Briefly, 10 mg of purified GST-fusion protein (unedited Cpx7A^I125,N130 or edited Cpx7A^I125M,N130S) was used per reaction and incubated with 2,500 units of recombinant kinase and [³²P]ATP (Perkin Elmer). Reaction products were separated by SDS-PAGE and gels were stained with Bio-Safe Coomassie Blue (Bio-Rad), dried, and exposed to autoradiography film at room temperature. Mean integrated density of each band was quantified using FIJI and relative density of phospho-Cpx (pCpx) was calculated by normalizing to input band intensity determined by Coomassie staining.

RNAseq analysis of RNA editing

RNAseq data from 105 single MN1-Ib and 101 single MNISN-Is 3^rd instar larval motoneurons obtained using isoform Patchseq protocols³³ were analyzed using the Integrative Genomics Viewer (IGV)⁷¹. To create single-cell Cpx RNA editing expression profiles, single RNA reads were analyzed for Cpx and included in the analysis if all three C-terminal Cpx7A edited bases were represented on a continuous single read. The percent of each Cpx7A edit variant was determined by the number of edited variant reads divided by total RNA reads for each cell, creating an RNA editing profile for each neuron. To compare single base editing across different genes, the edit percent at each base of interest was analyzed and compared to known edits in other genes within the same neurons. Neurons were excluded if each base of interest did not contain ten or more reads for all edits of interest.

We thank the Bloomington Drosophila Stock Center (Indiana University, Bloomington, IN; NIH P40OD018537), the Developmental Studies Hybridoma Bank (University of Iowa, Iowa City, IA), Dina Volfson and Richard Cho for assistance with experimental methods and Drosophila strains, and members of the Littleton lab for helpful discussions. This work was supported by The JPB Foundation and a National Institutes of Health grant (NS40296) to J.T.L., and a NSF grant (1122374) to E.A.B.