Control of neuronal synapse specification by a highly dedicated alternative splicing program

See allHide authors and affiliations

Science  20 May 2016:
Vol. 352, Issue 6288, pp. 982-986
DOI: 10.1126/science.aaf2397

Synapse identity through specific splicing

What are the mechanisms that create different, highly specific types of synapses? Traunmüller et al. found that the RNA-binding protein SLM2 regulated a very small number of alternative splicing decisions in mouse hippocampal neurons. Disrupting this splicing program resulted in specific defects in transsynaptic protein complexes. Glutamatergic transmission and synaptic plasticity were impaired. Genetically restoring one single alternative splicing event was enough to recover synaptic plasticity. SLM2 thus activates a highly specific alternative splicing program that controls the properties of glutamatergic synapses.

Science, this issue p. 982


Alternative RNA splicing represents a central mechanism for expanding the coding power of genomes. Individual RNA-binding proteins can control alternative splicing choices in hundreds of RNA transcripts, thereby tuning amounts and functions of large numbers of cellular proteins. We found that the RNA-binding protein SLM2 is essential for functional specification of glutamatergic synapses in the mouse hippocampus. Genome-wide mapping revealed a markedly selective SLM2-dependent splicing program primarily consisting of only a few target messenger RNAs that encode synaptic proteins. Genetic correction of a single SLM2-dependent target exon in the synaptic recognition molecule neurexin-1 was sufficient to rescue synaptic plasticity and behavioral defects in Slm2 knockout mice. These findings uncover a highly selective alternative splicing program that specifies synaptic properties in the central nervous system.

Alternative splicing provides a key mechanism for neuron-specific gene expression (13). An array of RNA-binding proteins broadly expressed in neuronal cells has been implicated in controlling developmentally regulated and neuron-specific alternative splicing programs, with single proteins regulating hundreds of target transcripts (46). However, some RNA-binding proteins are selectively expressed in neuronal populations, raising the possibility that they may control cell type– and synapse-specific functions (7, 8).

The KH-domain RNA-binding protein SLM2 is highly expressed in glutamatergic pyramidal cells of the mouse hippocampus and in a specific subset of γ-aminobutyric acid (GABA)–releasing interneurons (9, 10). In Slm2KO knockout (KO) mice, glutamatergic spine synapses formed at normal numbers on the primary apical dendrites of hippocampal CA1 neurons (Fig. 1, A and B). Western blot analysis of synaptosome fractions from wild-type (WT) and Slm2KO hippocampi revealed overall normal concentrations of glutamatergic synapse proteins. However, there was an increase in the AMPA-type glutamate receptor (AMPAR) subunit GluA1, in particular in detergent-soluble fractions from adult Slm2KO mouse synaptosomes (Fig. 1, Cand D). In acute slices from adolescent mice (postnatal day 25), GluA1 amounts were elevated in total cell lysates and cell surface fractions, whereas N-methyl-d-aspartate (NMDA)–receptor GluN1 subunit expression was unaltered (Fig. 1, E and F). Whole-cell voltage-clamp recordings from CA1 neurons in Slm2KO hippocampal slices showed no difference in the miniature excitatory postsynaptic current (mEPSC) amplitude and frequency between WT and Slm2KO mice (Fig. 2, A and B). However, mEPSC events showed a modest increase in the speed of rise and decay times in Slm2KO CA1 neurons (Fig. 2, C and D). AMPAR/NMDAR ratios were significantly increased in Slm2KO mice (Fig. 2, E and F). In Slm2KO CA1 neurons, stimulation of Schaffer collaterals elicited larger postsynaptic responses as compared to WT (Fig. 2G; see also fig. S1A for data from field EPSP recordings). Paired-pulse facilitation was normal in Slm2KO (Fig. 2H). In sum, these experiments demonstrate an elevation in postsynaptic AMPAR surface expression and function in CA1 neurons of Slm2KO mice. Finally, long-term potentiation (LTP) induced by theta-burst stimulation of Schaffer collaterals was significantly reduced in acute slices from Slm2KO mice (Fig. 2I).

Fig. 1 Synaptic structure and synaptic composition in Slm2KO mice.

(A) Thy1::YFP-H and Thy1::YFP-H:Slm2KO hippocampi [yellow fluorescent protein (YFP) in green, anti-SLM2 in red; scale bar, 200 µm (upper panels); 25 µm (lower panels). (B) Dendritic spine densities along the primary apical dendrite of CA1 cells in the stratum radiatum (100 µm from pyramidal cell layer). WT: n = 5 mice (38 dendrites); Slm2KO: n = 6 mice (44 dendrites); unpaired t test. (C) Western blot analysis of glutamatergic proteins in total lysate (LYS), synaptosomes (SYN), Triton X100–soluble fraction at pH = 6 (SOL), and Triton X100–insoluble postsynaptic density (PSD) in hippocampal tissue from adult WT and Slm2KO mice. (D) Mean intensities of expression levels in WT and Slm2KO hippocampi (n ≥ 3 mice per genotype, One-way analysis of variance (ANOVA) with Dunnett’s multiple comparison test, **P < 0.01). (E) Total and surface glutamate receptor pools were assessed in acute slices from P25 mice by surface biotinylation and streptavidin pull-down assay; nonbiotinylated samples (“no biotin”) served as negative control. (F) Quantification for Fig. 1E (one-way ANOVA with Tukey’s post-hoc test, ***P < 0.001, n ≥ 4 for WT and Slm2KO).

Fig. 2 SLM2 is required for normal glutamatergic transmission and plasticity.

(A and B) Amplitude and frequency of spontaneous mEPSCs recorded from CA1 pyramidal neurons in acute hippocampal slices (n ≥ 20 per genotype, n ≥ 3 animals per genotype, mean and SEM, unpaired t test). (C and D) Average traces of mEPSCs from (A) and (B) scaled to peak amplitude. (E and F) Average traces and summary data of evoked EPSCs at –70 mV (lower trace) and +40 mV (upper trace), respectively. Peak EPSC amplitudes were measured at –70 mV (peak marked by red circle). NMDAR currents were measured at +40 mV (red circle, n ≥ 22 per genotype, n ≥ 4 animals; mean and SEM, unpaired t test). (G) Representative traces of EPSCs evoked with varied stimulation intensities in WT or Slm2KO CA1 neurons (n ≥ 8 per genotype, n ≥ 3 animals; mean and SEM, two-way ANOVA). (H) Paired-pulse ratio in CA1 neurons from acute hippocampal slices (n ≥ 11 per genotype, n ≥ 4 animals, mean and SEM, two-way ANOVA). (I) Averaged responses from Slm2KO CA1 neurons following a TBS (theta-burst stimulus) delivered at Vc= –70 mV. The gray bar indicates the interval quantified in the histogram (n ≥ 7 per genotype, n ≥ 6 animals, mean and SEM, unpaired t test).

Candidate gene approaches have identified some transcripts that are altered in Slm2KO brains (912). However, a comprehensive global analysis of SLM2 targets is lacking. Using Illumina paired-end sequencing we mapped SLM2-dependent alternative splicing events at a genome-wide level. This analysis revealed highly correlated expression of transcripts between genotypes, indicating that SLM2 does not play a major role in tuning overall transcript levels (Fig. 3A). Moreover, transcripts encoding ionotropic and metabotropic glutamate receptors were not significantly changed (fig. S1, B to E). Most significantly altered was the transcript encoding the SLM2 paralogue SLM1, which has previously been shown to be up-regulated in Slm2KO hippocampus (11).

Fig. 3 Genome-wide mapping of SLM2-dependent alternative splicing program.

(A) Correlation analysis of log10-transformed, normalized total transcript counts (for all genes with count ≥ 100 in Slm2KO). Transcripts with significant alteration between genotypes (P < 0.01, ΔFC > 50%) are highlighted in blue (n = 4 mice per genotype). (B) Frequency distribution of exon incorporation rates for 107,482 exons expressed as fold change in splicing index (FC SI) plotted on a log2 scale. (C) Fold change SI and P-values for all exons with an FC SI > 10% (total of 9110 exons). Exons exhibiting changes of FC SI > 30% and P < 0.01 are marked in red (Pecam1, Epha5, Dgkb, Gm1673, Cpne5, Hnrnpul1, and Cdk16; threshold indicated by dashed lines). Highly deregulated exons (FC SI > 50% and P < 0.01) are marked in blue (Nrxn1, Nrxn2, Nrxn3, Stxbp5l). Slm2 exon 2, deleted in Slm2KO, is not displayed. (D) Experimental assessment of candidate SLM2-dependent exons by reverse transcription–polymerase chain reaction (RT-PCR) in WT and Slm2KO hippocampi; alternative exons (red), primer sites (arrows). (E) Table for fold change SI and P-values for validated exons shown in (D).

Genome-wide splicing patterns were extracted on the basis of annotations from FAST DB, and splicing indices calculated. The vast majority of exons remained essentially unchanged between WT and Slm2KO hippocampi (Fig. 3B and fig. S2A; analysis includes 4965 microexons). Notably, alternative exons in four genes showed disproportionately strong deregulation in the Slm2KO hippocampus. There was a >2-fold increase in the incorporation of exons at the alternatively spliced segment four (AS4) of neurexins (Nrxn1, 2, and 3), three genes that encode synaptic cell surface receptors. Moreover, exon 24 incorporation was 1.58-fold elevated in tomosyn-2 (Stxbp5l), a component of the vesicle fusion machinery. Another seven exons were identified that exhibited significant alterations (P < 0.01) although with modest fold-change (Fig. 3C and table S1). Independent experimental validation confirmed that deregulation of additional candidate target exons was small or in some cases not detectable (Fig. 3, D and E, and fig. S2B). The paralogue SLM1 may compensate for the loss of SLM2 (11). However, comparison of candidate exons in Slm1KO, Slm2KO, and Slm1:Slm2DKO (double knockout) mice revealed that deregulation was not significantly more severe in the DKO mice (fig. S2C).

Alternative splicing at Nrxn AS4 regulates selective trans-synaptic interactions of neurexins in the presynaptic terminal with several synaptic receptors (1316). For an unbiased identification of interaction partners regulated by this alternative splicing event, we performed affinity purifications on recombinant NRX1β4(+) and NRX1β4(-) isoforms, followed by shot-gun mass spectrometry (Fig. 4A and fig. S3, A and B). We identified 21 candidate binding partners, including known neurexin-binding proteins (Cbln2, the leucine-rich repeat proteins LRRTM1,2,4, neuroligin-1,2, 3, and the sortilin-related VPS10-domain containing receptor SORCS) and a number of previously unknown candidate ligands (complement C3, chondroadherin-like protein, neuronal pentraxin and neuronal pentraxin receptor 1, and astrotactin). For six of these proteins, we observed significant differences in the interaction with AS4(+) and AS4(-) NRX1β isoforms, including neuroligins, Chadl, LRRTMs, and complement C3 (Fig. 4B and table S2; P < 0.05). Thus, there is an array of synaptic interactions that can be modified by deregulation of neurexin alternative splicing in Slm2KO mice.

Fig. 4 Heterozygous deletion of Nrxn1 exon 21 restores synaptic phenotypes in Slm2KO hippocampus.

(A) Affinity purification with recombinant NRX1β AS4(+) and NRX1β AS4(-). Correlation of log10-transformed median spectra counts for all extracellular proteins identified with ≥2 peptides recovered on at least one of the two neurexin isoforms. Paralogues highlighted: NLs (red), Cblns (blue), LRRTMs (orange), SORCs (purple). (B) Fold change against P-value for all bound extracellular proteins when comparing NRX1β AS4(+) to NRX1β AS4(-). (C) Anti-NRX1 immunoprecipitation from forebrain lysates of WT and Slm2KO animals probed with anti-NRX1, anti-NL1, and anti-NL3. (D) Representative RT-PCR for analysis of Nrxn1,2,3 AS4 in the hippocampus of WT, Slm2KO, NrxnΔex21/,+ and Slm2KO;Nrxn1Δex21/+ animals. (E) NRX1 immunoprecipitation from forebrain lysates of WT, Slm2KO, and Slm2KO;Nrxn1Δex21/+ animals probed with anti-NRX1, anti-NL1 and anti-NL3. (F) Glutamate receptor expression in acute slices from P25 WT and Slm2KO and Slm2KO;Nrxn1Δex21/+ hippocampi. Protein concentrations were probed by Western blotting of total lysates (average FC Slm2KO/WT = 1.94, n = 5, average FC Slm2KO;Nrxn1Δex21/+/WT = 1.13, n = 3). (G) Averaged responses to a theta-burst stimulation (TBS) in Slm2KO;Nrxn1Δex21/+ CA1 neurons (green) is overlaid on WT (gray) and Slm2KO responses (blue) from Fig. 2I (n = 8 cells, n ≥ 5 animals per genotype). Comparison of the responses at 40 to 60 mins after induction between Slm2KO and Slm2KO;Nrxn1Δex21+/- by ANOVA with Tukey’s multiple comparison test (**P < 0.01). (H and I) Behavioral alterations of WT, Slm2KO, and Slm2KO;Nrxn1Δex21 mice in object recognition task. Time spent investigating a novel and a familiar object during a 5-min trial 1 hour after acquisition and discrimination ratio are plotted [(H), paired t test; (I), unpaired t test; WT, n = 15; Slm2KO, n = 20; Slm2KO;Nrxn1Δex21, n = 9, ns P > 0.05; ***P < 0.001].

We hypothesized that the modification of endogenous neurexin alternative splicing in Slm2KO hippocampus may disrupt interactions with these splice insertion-sensitive ligands. In coimmunoprecipitation experiments with an antibody against neurexin-1 (anti-neurexin-1), we observed that neuroligin-1 (NL1) and neuroligin-3 (NL3) are abundant in immunoprecipitates from WT mice. However, both NLs were reduced in precipitates prepared from Slm2KO mice (Fig. 4C). Conversely, coprecipitation of C3, a component of the complement system that has been implicated in synapse elimination (17), was slightly elevated in the same precipitates from Slm2KO mice (fig. S3C). Thus, the loss of SLM2 indeed switches synaptic receptor-ligand interactions. NL1 recruits synaptic NMDA receptors (18, 19), whereas NL3 and LRRTMs mediate the synaptic recruitment and stabilization of AMPA receptors (20). Loss of the trans-synaptic interactions with these neurexin ligands might result in the postsynaptic glutamate receptor deficits in Slm2KO mice. To test this hypothesis, we generated Nrxn1Δex21 mice in which Nrxn1 exon 21 (the alternative cassette exon at AS4) is specifically deleted but total Nrxn transcript levels are unchanged (fig. S4, A to C). Heterozygous removal of one Nrxn1Δex21 allele restored normal Nrxn1 AS4(-) transcript levels in the Slm2KO background (Fig. 4D). Restoration of Nrxn1 AS4(-) transcripts rescued trans-synaptic cell surface interactions of endogenous neurexin with NL1 and NL3 in the coimmunoprecipitation assays (Fig. 4E), normalized GluA1 amounts in acute hippocampal slices (Fig. 4F), and partially recovered theta burst–induced Schaffer collateral LTP (Fig. 4G). In an object recognition test, a behavioral task that involves the hippocampus (21), Slm2KO mice differed significantly from WT mice in that they did not preferentially explore novel objects (Fig. 4, H and I, and fig. S4D). Also, this phenotype and other behavioral alterations were rescued in Slm2KO;Nrxn1Δex21 /+ mice (Fig. 4, H and I, and fig. S4, E to G). Thus, the control of a single alternative exon by SLM2 has a major contribution to the specification of glutamatergic synapse function, plasticity, and mouse behavior.

Forced expression of Nrxn3 AS4(+) isoforms was previously shown to reduce postsynaptic AMPAR localization and to impair LTP in the subiculum (22). By contrast, in Slm2KO CA1 neurons, total surface AMPAR amounts and AMPAR function are elevated. These differences are most likely due to the simultaneous disruption of alternative splicing in all three neurexin genes in the Slm2KO, as well as the different cell type–specific context. Compared to other RNA-binding proteins, SLM2 exhibits highly selective neuronal cell type–specific expression, and only a small set of alternative exons is particularly reliant on SLM2 function. Our genetic rescue experiments demonstrate that restoration of a single alternative exon has a major impact on the SLM2KO phenotype. We hypothesize that targeted, cell type–specific splicing regulation of surface receptor recognition systems as reported here for the SLM2-neurexin system represents a general mechanism for the control of synapse specification in neuronal circuits.


Materials and Methods

Figs. S1 to S4

Table S1

References (2332)


  1. Acknowledgments: We thank O. Mauger, D. Schreiner, and L. Xiao for insightful comments on the manuscript and Y. Hayashi and J. Bischofberger for expert advice. We thank C. Bornmann and P. Demougin, F. Lemoine, P. de la Grange, A. Schmidt, and M. Rüegg for suggestions and experimental support. L.T. was financially supported by the Boehringer Ingelheim Fonds. A.M.G. was supported by a European Molecular Biology Organization long-term fellowship. T.-M.N. was supported by the International PhD Program University of Basel. This work was supported by funds to P.S. from the Swiss National Science Foundation; European Autism Interventions EU-AIMS, which receives support from the Innovative Medicines Initiative; and the Kanton Basel-Stadt. Sequencing data were deposited at the functional genomics data repository Gene Expression Omnibus (accession number GSE79902).
View Abstract

Stay Connected to Science

Navigate This Article