Epilepsy and intellectual disability linked protein Shrm4 interaction with GABABRs shapes inhibitory neurotransmission

Shrm4, a protein expressed only in polarized tissues, is encoded by the KIAA1202 gene, whose mutations have been linked to epilepsy and intellectual disability. However, a physiological role for Shrm4 in the brain is yet to be established. Here, we report that Shrm4 is localized to synapses where it regulates dendritic spine morphology and interacts with the C terminus of GABAB receptors (GABABRs) to control their cell surface expression and intracellular trafficking via a dynein-dependent mechanism. Knockdown of Shrm4 in rat severely impairs GABABR activity causing increased anxiety-like behaviour and susceptibility to seizures. Moreover, Shrm4 influences hippocampal excitability by modulating tonic inhibition in dentate gyrus granule cells, in a process involving crosstalk between GABABRs and extrasynaptic δ-subunit-containing GABAARs. Our data highlights a role for Shrm4 in synaptogenesis and in maintaining GABABR-mediated inhibition, perturbation of which may be responsible for the involvement of Shrm4 in cognitive disorders and epilepsy.

T he actin-binding proteins Shroom (Shrm) play an important role in cytoskeletal organization and consist of an N-terminal PDZ domain, a central Apx/Shrm Domain 1 (ASD1) and a C-terminal ASD2 domain 1 . There are four evolutionarily conserved Shrm proteins (Shrm1-4) 2 that are localized to polarized tissues, including neurons 1,3 . Of these, only Shrm4 lacks the actin-targeting ASD1 motif, and the role of its PDZ domain is unknown. Murine Shrm4 possesses putative binding sites for EVH1 (poly-proline rich domain), a PDZ (SNF) binding motif 4 and a stretch of glutamine and glutamate residues preceding the C-terminal ASD2 motif that is unique to Shrm4, but not other family members 4 . Shrm4 is ubiquitously expressed throughout embryonic and adult murine brains 2 and also binds to F-actin in non-neuronal cells 4 .
The importance of Shrm4 is illustrated by two de novo balanced X-chromosomal translocations, which disrupts the KIAA1202 gene (Xp11.2) that encodes for Shrm4. In addition, a pathogenic missense mutation was identified in an unrelated large family, with carriers exhibiting mild-to-severe intellectual disability (ID) and increased susceptibility to seizures 2,5 . Recent studies reinforce the role of Shrm4 in ID [6][7][8] , however, how disruption of KIAA1202 causes these neuropathological conditions is unknown. Indeed, the role of Shrm4 in the brain is also unknown, but given the pathological profile, it may regulate GABA-mediated inhibition 9,10 . GABA activates ionotropic GABA A (ref. 11) (GABA A R) and metabotropic GABA B receptors (GABA B Rs) 12 to control inhibition which is important for synaptic plasticity 13,14 . The importance of these receptors is emphasized during dysfunction, which occurs in different neurological disease 14,15 .
Here, we report that Shrm4 interacts with GABA B Rs to facilitate trafficking to dendrites using a dynein-dependent mechanism. For cell surface expression, GABA B Rs are obligate heterodimers comprising GABA B R1 (GABA B1 ) and R2 (GABA B2 ) subunits that mediate long-lasting synaptic inhibition 16 . However, the motor-protein-dependent trafficking of these receptors is not fully understood.
We found that loss of Shrm4 compromises GABA B R delivery to postsynaptic compartments, impairs GABA B R-mediated K þ currents and GABA A R-mediated tonic inhibition in the hippocampus, and reduces dendritic spine density altering the composition of synaptic proteins resulting in increased anxietylike behaviour and susceptibility to seizures in rats. Our study suggests a possible underlying mechanism by which Shrm4 may be involved in epilepsy and ID.

Results
Shrm4 is an interacting partner of GABA B receptors. We first investigated the subcellular localization of Shrm4 in cultured rat hippocampal neurons at 18 days in vitro (DIV) by immunostaining and found colocalization with presynaptic (Bassoon) and postsynaptic markers of excitatory (PSD-95) and inhibitory (GABA A b3) synapses ( Supplementary Fig. 1a). By using electron microscopy and post-embedding immunogold methods with an anti-Shrm4 antibody 2 , gold nanoparticles were identified in preand post-synaptic areas and along dendrites (Supplementary Fig. 1b; Supplementary Table 1). Biochemical fractionation of adult rat brain hippocampi and cortices revealed enrichment of Shrm4 in the postsynaptic density (PSD) fraction further confirming its presence at synapses (Supplementary Fig. 1c).
To explore the role(s) of Shrm4 in neurons, we searched for binding partners using yeast two-hybrid (Y2H) screening. The PDZ domain of Shrm4 (residues 1-91; Fig. 1a) was selected as the bait to screen against an adult human brain cDNA library, as this domain participates in protein-protein interactions. Twenty positive cDNA clones were isolated; six of these encoded a 100 amino acids stretch of the C-terminal tail of GABA B1 present in both splice variants B1a and B1b, which differ by the inclusion of two Sushi domains in the N terminus only in GABA B1a (refs 12,14) (Fig. 1a). The interaction between Shrm4 and the GABA B1 C-tail was crucially reliant on the Shrm4-PDZ domain (PDZ domain 1-91: þ þ þ ; DPDZ 91-1,492: negative; DASD2 1-1,213: þ þ þ ; Fig. 1a). We confirmed this interaction using rat brain lysates and found that endogenous Shrm4 co-precipitates with GABA B1a/b (Fig. 1b). GABA B R antibody specificity was confirmed by western blots from GABA B1a and GABA B1b knockout mice 17 (Supplementary Fig. 1d). Moreover, direct stochastic optical reconstruction microscopy (dSTORM) of 14DIV hippocampal neurons revealed that Shrm4 and GABA B Rs co-cluster in neurons ( Fig. 1c; Supplementary Fig. 1e) 18 .
We used immunoprecipitation to determine the minimal region of the GABA B1 C-tail that interacts with Shrm4 in HEK293 cells expressing Shrm4-GFP, GABA B2 and GABA B1a cDNAs with differing C termini (Fig. 1f). This revealed that R 859 LITRGEWQSEA 870 in the C-tail of GABA B1 interacted with the Shrm4-PDZ domain (Fig. 1f). To confirm this interaction, we produced a cell-permeable peptide fragment (Tat-peptide) encompassing the Shrm4-GABA B1 binding site on GABAB1 (859-870). We verified the efficiency of this peptide in pulldown experiments with GST-PDZ incubating different concentration of Tat control or Tat-859-870 with lysates of HEK293 cells overexpressing GABA B1a -GFP and GABA B2 ( Supplementary  Fig. 1f) and a dose-dependence was observed. We then incubated lysates from HEK293 cells overexpressing GABA B1a and GABA B2 with this peptide before GST pull-down using Shrm4-PDZ or its mutant form (K26G27/AA). Consistent with previous results, co-precipitation of GABA B1 was reduced after preincubation with the Tat-peptide compared with its scrambled control (Fig. 1g). These results reveal a new association between the PDZ domain of Shrm4 and residues 859-870 of the GABA B1 C terminus.
Shrm4 regulates the surface expression of GABA B Rs. Given that the C-tail of GABA B1 is important for trafficking to the cell surface 19,20 , we next examined whether Shrm4 regulates GABA B R trafficking. We designed two shRNAs that specifically targeted rat (shRNA#1 and #2) and human (shRNA#2) Shrm4 transcripts. To test their effectiveness as Shrm4 silencers, HEK293 cells were transfected with human HA-Shrm4 cDNA (Rescue) with or without each shRNA. Both shRNAs reduced Shrm4 expression while co-transfection of a rescue construct with shRNA#1 restored expression ( Supplementary Fig. 2a,b). We extended this approach to neurons, transfecting Scrambled, shRNA#1, shRNA#2 or Rescue constructs at 8DIV for later analysis of surface levels of GABA B1 at 18DIV. Interestingly, both shRNAs reduced surface levels of GABA B1 that can be rescued by the re-expression of Shrm4 (Fig. 1h, full neurons in Supplementary  Fig. 2c). As an additional control, we tested the effect of knocking down Shrm3 and found no effect on GABA B1 surface expression indicating a unique role played by Shrm4 in GABA B1 trafficking ( Supplementary Fig. 2d). We next used lentiviral delivery to corroborate our immunostaining results. For this we used cell surface biotinylation assays for GABA B Rs (Fig. 1i). Shrm4 knockdown reduced surface expression of GABA B Rs without affecting the total expression of GABA B Rs or the total and surface expression of GABA A R a1 subunits used as negative controls (Fig. 1i). Thus, Shrm4 is important for GABA B1 trafficking in neurons.
Shrm4 regulates synaptic structure and protein composition. Since Shrm4 colocalizes with PSD-95 and Bassoon, we investi-gated if its knockdown affected excitatory synapses in rat pyramidal hippocampal neurons. In these excitatory neurons, expressing scrambled or shRNA#1 before synaptogenesis (8DIV) did not alter the branching or dendritic diameters at 18DIV (Fig. 2a). We therefore refined our analysis and studied dendritic spines, where defects are commonly linked to ID 21 Supplementary Fig. 11).
using immunofluorescence (Fig. 2b, full neurons shown in Supplementary Fig. 3). The numbers and intensities of fluorescent puncta for all four markers were reduced in Shrm4-silenced neurons and rescued by HA-Shrm4 (Fig. 2c). The decrease occurred in parallel with a reduction in spine density and this was also rescued by HA-Shrm4 (Fig. 2d). These results were replicated by silencing via shRNA#2 ( Supplementary  Fig. 4a). Finally, to attribute these changes to Shrm4 specific knockdown, we silenced Shrm3 and found no changes to dendritic spine density as shown in Supplementary Fig. 2d.
Shrm4 silencing also had profound consequences on spine morphology reducing spine length, without affecting spine head width (Fig. 2e,f) while increasing the number of stubby spines, compared with scrambled controls (Fig. 2g). These effects were reversed using the rescue construct discounting non-specific effects of the shRNA (Fig. 2e-g). Interestingly, no changes in spine density were observed with Shrm4 knockdown after synaptogenesis at 12DIV (Supplementary Fig. 4b) suggesting a role for Shrm4 in synaptogenesis or in synapse maintenance.   To assess whether Shrm4 silencing reduced the density of spines via down-regulation of dendritic GABA B Rs, we transfected hippocampal neurons with GABA B1 shRNA, or scrambled shRNA 22 or GABA B1 shRNA together with shRNA-insensitive GABA B1b cDNA before synaptogenesis (8DIV), for processing at 18DIV for immunofluorescence. GABA B1 knockdown reduced GABA B1 immunofluorescence ( Supplementary Fig. 5a), and similar to Shrm4 silencing, also reduced the density of spines, which was rescued by shRNA-insensitive GABA B1b cDNA (Fig. 2h).
Then, we evaluated the specific involvement of Shrm4-GABA B Rs interaction on spine density by applying the Tat-859-870 peptide to hippocampal cultured neurons from 8DIV to 13DIV during the synaptogenesis peak. This treatment was sufficient to induce a reduction in spine density similar to that observed in Shrm4 or GABA B1 knockdown condition demonstrating that the interaction between the two proteins is necessary for normal spine development (Fig. 2i). Interestingly, silencing of Shrm4 also resulted in changes in spontaneous dendritic Ca 2 þ -signals (Fig. 2j,k), assessed using GCaMP6f, without a change in mean amplitudes of spikes (Fig. 2l) in hippocampal neurons. In particular, we observed slower rise (Fig. 2m,n) and decay times (Fig. 2m,o) of Ca 2 þ transients in Shrm4-knockdown neurons, which could have important consequences for spine morphology. Moreover, these defects were not rescued by the concomitant overexpression of GABA B1 DC mutant demonstrating the dependency of these effects on GABA B Rs surface expression ( Supplementary Fig 5b).
Thus, Shrm4 is localized at excitatory synapses and plays a crucial role in determining the morphology and molecular structure of dendritic spines most likely by modulating the surface expression of GABA B1 receptors.
Shrm4 and GABA B1 are associated in a complex with DIC. As Shrm4 knockdown reduced surface GABA B1 receptors without affecting the total level of protein expression, we wondered whether this occurred because of a trafficking impairment. Immunostaining for GABA B1 in permeabilized hippocampal neurons after Shrm4 knockdown revealed an increased intensity of intracellular GABA B R expression in the soma compared with scrambled controls, indicating an accumulation in intracellular compartments which could be reversed by the rescue construct (Fig. 3a).
Accumulation of GABA B Rs in the soma in Shrm4-silenced neurons could reflect increased internalization, or defective transport from the soma to the cell surface, or both. To distinguish, GABA B R internalization was analysed in live hippocampal neurons 23 . These were transfected at 7DIV with cDNAs encoding for GABA B1a tagged with a bungarotoxin (BTX) binding site, and GABA B2 , together with either knockdown or scrambled shRNAs. No difference in GABA B R internalization rates between Shrm4-silenced neurons and control neurons were found ( Supplementary Fig. 6a-e).
As a consequence, we investigated the transport of GABA B Rs to distal compartments, which should be driven by microtubule-based motor proteins: kinesins and dyneins 24 . Kinesin-1 is responsible for GABA B1a transport through the axonal endoplasmic reticulum (ER) and ER-Golgi intermediate compartment (ERGIC 25 ); while dynein targets GABA B1 to the dendrites 26 .
Using anti-Shrm4 antibody in adult rat brain lysates, we coimmunoprecipitated the dynein intermediate chain (DIC), GABA B1 and GABA B2 , but not any of KIF5 isoforms ( Fig. 3b; Supplementary Fig. 6f). We then confirmed the existence of this complex in transfected HEK293 cells by immunoprecipitating Shrm4 with Shrm4-V5, and either GABA B1a -myc or GABA B1b -myc (Fig. 3c). Interestingly, DIC co-immunoprecipitated with HA-Shrm4 even in the absence of GABA B1 suggesting that these two proteins are associated in a molecular complex (Fig. 3d).
We found histidine-tagged DIC associated with Shrm4-PDZ domain also in pull-down experiment in vitro demonstrating that the two proteins can interact (Fig. 3e). Of course, these results cannot exclude other factors, such as dynein light chains, that can participate in a Shrm4-DIC complex in vivo. Nevertheless, to identify which domain of Shrm4 was responsible for the interaction, we used a GST-tagged PDZ domain mutated in the classical PDZ binding site (K26G27/AA). This mutant was still able to pull-down histidine-tagged DIC in vitro (Fig. 3e) suggesting that another region of the Shrm4-PDZ domain was involved. The analysis of the Shrm4-PDZ solution structure (Protein Data Bank (PDB) code: 2EDP) highlighted the N-terminal segment (first 14 amino acids) as one of the portions more likely involved in interaction with DIC. Indeed, this segment of Shrm4-PDZ is solvent accessible and is opposite to the PDZ region that participates in classic protein's C-termini recognition (b2-a2 interface 27 ). We truncated 14 amino acids at the N terminus of Shrm4-PDZ (GST-tagged PDZD14) and found that eliminating these residues completely abolished DIC binding in adult rat brain lysates (Fig. 3f). These results suggest that Shrm4 has the ability to bind GABA B1 and DIC simultaneously, forming a ternary complex. Using in silico modelling, we propose a structure for this complex (Supplementary Fig. 6g). Collectively, the results of in vitro experiments suggest that the b1 and b2 strands of Shrm4 bind the R 859 LITRGEWQSEA 870 segment of GABA B1 and DIC, respectively. By combining the information from our in vitro experiments with the structure availability of the Shrm4-PDZ, DIC 110-138 fragment in complex with LCs, and the coil-coil C-terminal heterodimer of GABA B1b and GABA B2 (PDB: 4PAS 28 ), we could predict a low resolution model of the Shrm4-PDZ in complex with the GABA B1b -GABA B2 heterodimer, DIC and LC8 ( Supplementary Fig. 6g).
By using dSTORM on 18DIV hippocampal neurons, we detected Shrm4 puncta on tubulin-positive filaments ( Fig. 3g) along dendrites where Shrm4 could associate with DIC and GABA B1 .
Finally, to ascertain if Shrm4, DIC and GABA B Rs associate in a physiological context, we generated adeno-associated virus serotype 5 (AAV5) expressing either scrambled (AAV5-scrambled#1) or Shrm4 knockdown (AAV5-shRNA#1) shRNAs and injected these into rat CA1 hippocampi (Fig. 3h). Using western blots, the AAV5-knockdown-injected hemisphere exhibited markedly lower Shrm4 protein levels than the AAV5-scrambled-injected hemisphere ( Supplementary Fig. 6h). Subsequent anti-DIC immunoprecipitation from hippocampal lysates, previously injected with either AAV5-scrambled#1 or AAV5-shRNA#1, revealed co-immunoprecipitation of GABA B R with DIC in scrambled controls; however, this was significantly reduced in lysates from the Shrm4-knockdown hemisphere suggesting that dynein-GABA B R co-association is dependent on endogenous levels of Shrm4 (Fig. 3i). Taken together, these results highlight the central role played by Shrm4 in the association of dynein and GABA B Rs in the hippocampus.
Shrm4 mediates GABA B R dynein-dependent dendritic transport. As dynein transports postsynaptic proteins and Golgi outposts 29  distribution of endogenous GABA B Rs in the dendrites and axons of 8DIV hippocampal neurons expressing: GFP (control), GFPcoexpressing knockdown shRNA#1 either with or without rescue-shRNA, and a GFP-tagged dominant-negative dynactin construct (GFP-p150-cc1) that inhibits dynein activity 30 .
GABA B R fluorescence intensity in dendrites of GFP-shRNA#1 and GFP-p150-cc1-expressing neurons was markedly lower compared with GFP controls (Fig. 4a), while fluorescence in axons was unaffected (P40.05). Expressing the rescue construct in Shrm4-silenced neurons recovered the GABA B R fluorescence in dendrites (Fig. 4a). We also quantified the polarity index (PI), which has a higher value the greater the abundance of receptors in dendrites (Fig. 4a, full neurons are shown in Supplementary Fig. 7a). Interestingly, by expressing shRNA#1 in neurons at 12DIV, the PI was significantly lower compared with controls, suggesting that Shrm4 is important for the normal abundance of dendritic GABA B Rs not only before, but also after synaptogenesis ( Supplementary  Fig. 7b). Finally, a second GFP-Shrm4 knockdown construct (shRNA#2) also reduced the GABA B R PI as shown in Supplementary Fig. 4a.
Since both GABA B1a and GABA B1b are present in dendrites 31 , we considered if Shrm4 knockdown or dynein inhibition differentially modified their intracellular transport. Then we expressed shRNA#1 or GFP-p150-cc1 that significantly reduced the PI of expressed GABA B1a and GABA B1b subunits, consistent with the reduced intensity noted for endogenous GABA B Rs (Fig. 4b, full neurons are shown in Supplementary Fig. 8a). Importantly, these results demonstrate that Shrm4 acts through dynein to drive GABA B Rs to dendrites in hippocampal neurons and overexpression of GABA B1a or GABA B1b does not rescue the reduction in PI.  Supplementary Fig. 11). (g) Conventional (top left) and super-resolution direct stochastic optical reconstruction microscopy (dSTORM) (bottom left) imaging of Tubulin-ATTO 488 (shown in red) and Shrm4-Alexa 647 (shown in green). Shrm4 positive puncta are localized along microtubule-positive filaments, as evidenced in the details of the super-resolution image (top right). The super-resolution data has been quantified by cross-correlation analysis. The positive, higher than 1 cross-correlation, indicates co-clustering of the two fluorescent signals (Errorbars are s.e.m.s, number of regions for the cross-correlation measurement: 30, number of fields: 3. Scale bar, 0.4 mm). (h) Schematic illustrating unilateral local injection of AAV5-scrambled#1 (left hemisphere) and AAV5-shRNA#1 (right hemisphere) into rat brain CA1 hippocampus, with time line for recovery and experiments. (i) Western blots and histograms showing monoclonal anti-DIC immunoprecipitation from scrambled and shRNA lysates of infected hippocampus extracts (n ¼ 6 rats). Co-precipitated GABA B1 levels were normalized to DIC immunoprecipitation levels and the normalized percentage of GABA B1 co-precipitation was calculated. Histograms show mean ± s.e.m.; *Po0.05, t-test.
Furthermore, we verified that Shrm4 knockdown was not altering GABA B1 -GABA B2 dimerization 32 . We therefore cotransfected GABA B1 -myc and GABA B2 -flag with either shRNA#1 or its scrambled control. This was followed by immunostaining for the receptors. The analysis revealed high correlation in both scrambled and Shrm4-knockdown conditions excluding dysfunctional receptor dimerization ( Supplementary Fig. 8b).
To address whether the reduction in dendritic GABA B Rs was a consequence of decreased transport, fluorescence recovery after photobleaching (FRAP) was applied to cultured hippocampal neurons coexpressing GABA B1 -RFP with GFP-shRNA#1 or GFP-p150-cc1. For both, fluorescence recovery of GABA B1 -RFP was significantly delayed compared with neurons expressing the GFP control. Only an increased half-time for fluorescence recovery, without affecting the mobile fraction (plateau reached after 350 s), was observed in Shrm4-silenced and GFP-p150-cc1-expressing neurons, suggesting that transport of dendritic receptors, occurring either intracellularly or at the surface level, was severely impaired (Fig. 4c).
These results suggest that Shrm4 mediates dendritic transport of the GABA B Rs heterodimer via its interaction with GABA B1 and DIC. Merge Brevican Physiological role of Shrm4 in vitro and in vivo. GABA B Rs activate G protein-coupled inwardly-rectifying K þ channels, generating slow inhibitory postsynaptic currents (IPSCs) 33 . As Shrm4 regulates GABA B R dendritic cell surface number, we examined K þ currents induced by the GABA B R agonist baclofen (10-100 mM) on 14DIV hippocampal neurons transfected at 7DIV with either GFP-coexpressing knockdown shRNA#1 or scrambled#1 or p150-cc1. Peak K þ current densities (pA/pF) were lower in Shrm4-silenced and dynein-inhibited neurons compared with scrambled controls (Fig. 4d), consistent with a reduced number of dendritic GABA B Rs. This reduced current was reversed by expressing shRNA-insensitive Shrm4 (Supplementary Fig. 9a). By contrast, K þ currents activated by metabotropic glutamate receptors were unaffected by Shrm4 silencing and dynein inhibition ( Supplementary Fig. 9b). GABA A Rmediated miniature IPSCs (mIPSCs) were also unaffected by Shrm4 silencing (Supplementary Fig. 9c), indicating that Shrm4 and dynein/dynactin selectively regulate GABA B R-mediated responses.
To assess if Shrm4 silencing affected the neurophysiology of GABA B Rs, we injected AAV5-scrambled (left hemisphere) and AAV5-shRNA#1 (right hemisphere) into the hippocampal CA1 region of 3 months-old rats 33 (Fig. 4e). Input/output currents were measured to evaluate the basal excitatory transmission and we did not observe any statistically significant difference between the two conditions ( Supplementary Fig. 10a). We also recorded field excitatory postsynaptic potentials (fEPSPs) in the apical dendritic layer of CA1 3 weeks later by inducing long-term potentiation (LTP) or long-term depression (LTD) via Schaffer collateral stimulation. LTP and LTD induction and maintenance were unaffected ( Supplementary Fig. 10b,c) consistent with the unchanged PSD-95 intensity in brain slices from injected animals ( Supplementary Fig. 10d). The apparent contradiction with the defect observed in dendritic spine number can be explained by noting that these animals were injected at three months of age, when the peak of synaptogenesis has passed. We know that Shrm4 knockdown in mature neurons did not affect spine number ( Supplementary Fig. 5b), whereas it is still able to impair GABA B1 trafficking ( Supplementary Fig. 4b) as shown by the decreased polarity index.
Whole-cell K þ currents evoked by baclofen in the CA1 of acute slices from injected animals were significantly reduced by Shrm4 knockdown compared with scrambled controls (Fig. 4f). The injection of AAV5-shRNA#2 induced similar reductions in K þ current confirming the specificity of our results. Thus, Shrm4 silencing reduces functional GABA B R responses in vitro and in vivo, in accord with a reduced number of surface dendritic GABA B Rs.
Shrm4 silencing in vivo affects hippocampal tonic inhibition. Recent evidence suggests that GABA B R activation enhances the conductance of extrasynaptic d subunit-containing GABA A Rs of dentate gyrus granule cells (DGGCs) [34][35][36] . As Shrm4 silencing reduced GABA B R activity, we explored if this affected tonic inhibition in DGGCs.
Animals were injected with AAV5-scrambled (left hemisphere) and AAV5-knockdown (right hemisphere) shRNAs into the DG (Fig. 5a). Whole-cell recordings 3 weeks later from DGGCs in acute slices 35,36 subject to scrambled shRNA revealed that GABA (5 mM) increased the bicuculline-sensitive baseline current and current noise variance (Fig. 5b). For cells expressing Shrm4-shRNA#1, both current and noise variance were reduced compared with controls (Fig. 5b). As tonic inhibition in DGGCs relies on d subunit-containing GABA A Rs 34 and can be regulated by GABA B Rs, we considered whether the reduced tonic current was a direct consequence of a reduction in surface GABA B Rs. Repeating these experiments in the presence of the GABA B R antagonist CGP54628, showed similar reduction in tonic current and noise variance evident in the knockdown condition without CGP pre-treatment (Fig. 5c). These data suggested that reducing cell surface GABA B Rs has no direct effect on tonic inhibition, which is more likely due to an indirect impairment of d subunitcontaining GABA A Rs. This was confirmed using the agonist THIP at d subunit-selective concentrations (3 mM) and as expected, the mean current and noise variance were reduced in DGGCs in the hemisphere carrying the Shrm4 knockdown compared with the control hemisphere (Fig. 5d). To exclude off-target effects of shRNA#1, we injected a second knockdown construct, AAV5-shRNA#2, which produced identical effects on THIP-induced currents (Fig. 5e). By contrast, GABA A R-mediated mIPSCs in DGGCs were unaffected ( Supplementary Fig. 10e), indicating that Shrm4 specifically regulates d subunit-containing GABA A R-mediated responses.
Interestingly, co-immunoprecipitation using brain extracts and monoclonal anti-GABA A R d subunit revealed that GABA B Rs and d-subunit-containing GABA A Rs co-associate. By contrast, the synaptic GABA A R g2 subunit 37 was absent (Fig. 5f). Thus, by controlling postsynaptic GABA B Rs, Shrm4 is also able to regulate tonic inhibition mediated by d subunit-containing GABA A Rs.
In vivo Shrm4 silencing causes behavioural deficits. A role for GABA B Rs in anxiety 38 and epilepsy 14,39 is well-known. GABA transmission has been also linked to neurodevelopmental disorders such as autism spectrum disorders (ASD) 40,41 . Furthermore, GABA B R agonist application has been proposed as therapeutic strategy for social deficits, repetitive behaviours and other aspects of ASD in different mouse models 42 . To understand if reducing GABA B R numbers and tonic inhibition by Shrm4 knockdown has behavioural implications, we injected rats bilaterally with either Shrm4 knockdown (AAV5-shRNA#1 or AAV5-shRNA#2), or AAV5-scrambled shRNA (AAV5-scrambled#1 or AAV5-scrambled#2) and extensively analysed the behaviour of these animals (Fig. 6a). Locomotor activity of the injected animals was unaffected allowing us to perform subsequent behavioural analyses without locomotor bias (Fig. 6b).
The elevated plus maze (EPM) and the marble-burying test measured anxiety levels 43 whereas social behaviours were evaluated in a three-chamber apparatus 44 and in the tube test for aggressivity 45 . AAV5-knockdown-shRNA-injected rats (with either shRNA#1 or shRNA#2) exhibited increased anxiety and impaired social behaviour (Fig. 6c-f). In the EPM, Shrm4knockdown rats made fewer open-arm entries (Fig. 6c) and spent less time in open arms compared with AAV5-scrambled shRNAinjected controls (Fig. 6c). The total number of arm entries was unaffected by AAV5-knockdown-shRNA, confirming that locomotion was unaffected.
In the marble-burying test, animals injected with AAV5-shRNA#1 buried a higher number of marbles and spent less time before burying compared with AAV5-scrambled controls confirming an increased anxiety level (Fig. 6d). Social behaviours of animals injected with AAV5-shRNA#1 were also found impaired with reduced time spent close to a stranger naive animal (Fig. 6e, sociability) and to a second new stranger animal (Fig. 6e, social novelty) compared with AAV5-scrambled controls. These animals also showed greater aggression in the tube test in terms of percentage of wins versus the controlinjected animals (Fig. 6f).
We then assessed involvement of Shrm4's deficiency in epilepsy by pentylenetetrazole (PTZ) administration to evaluate seizure sensitivity 46 and recorded electro-encephalograms (EEG) to measure spontaneous electrical activity (Fig. 7a).
Electrical activity evaluated for 24 h in freely moving awake animals showed a significant spontaneous spike activity in all the AAV5-shRNA#1 injected rats compared with the AAV5-scrambled#1 (Fig. 7b,c) suggesting an increased general excitability.
In fact, following PTZ injections (i/p 45 mg kg À 1 ) 46 , the latency to the first seizure was reduced and the seizure duration was longer for knockdown shRNA-injected animals (Fig. 7d). The severity of the response to PTZ was also increased with greater number of tonic-clonic seizures (Fig. 7d). This indicates that in vivo Shrm4 silencing in CA1 produces defects in anxiety, social behaviour and susceptibility to seizures. Similar effects on seizure susceptibility were obtained when we injected rats intraperitoneally with Tat-859-870 peptide administering PTZ 12 h after (Fig. 7e,f) demonstrating that the disruption of Shrm4-GABA B Rs interaction was responsible of these effects.
These phenotypes parallel the defects in GABA B R trafficking and transmission observed in vitro and in vivo in Shrm4-silenced hippocampal neurons.

Discussion
Several neurodevelopmental IDs including autism and Fragile X syndrome are characterized by a reduction of GABA B R expression levels 47,48 , and treatment with GABA B R agonists have been reported to improve susceptibility to seizures 49 and social and cognitive behaviour [50][51][52] . In this study, we have characterized an interaction between the ID-linked protein Shrm4, GABA B Rs and dynein motor protein. The disruption of this complex leads to dysfunction of GABA B Rs cell surface targeting and subsequent reduction of signalling efficacy, and this has interesting parallels with these neurodevelopmental disorders.
We have characterized physiological roles for Shrm4 by discovering a new interaction between its PDZ domain and the (c) Tonic currents activated by 5 mM GABA in the presence of CGP54626 (5 mM) of DGGCs in acute brain slices taken from hemispheres injected with either AAV5-shRNA#1 or AAV5-scrambled#1. A lower GABA-mediated current shift and current variance in slices pre-treated with CGP54268 from AAV5-shRNA#1 compared with AAV5-scrambled#1 injected animals were observed (current shift: n ¼ 6, 7; *P ¼ 0.0352; t-test; noise variance: n ¼ 5, 7; *P ¼ 0.0215; t-test). (d) THIP (3 mM)-activated currents of DGGCs in the presence of kynurenic acid (3 mM). THIP current shift and current variance were lower in slices injected with AAV5-shRNA#1 compared with AAV5-scrambled#1 (current shift: n ¼ 9, 11; *P ¼ 0.0225; t-test; noise variance: n ¼ 10, 13; *P ¼ 0.0419; t-test). (e) THIP (3 mM)-activated currents from DGGCs in the presence of kynurenic acid (3 mM). A lower THIP current shift and current variance was also observed in slices injected with a second knockdown construct AAV5-shRNA#2 compared with AAV5-scrambled#2. Application of bicuculline (20 mM) demonstrates the GABA A R-mediated specificity of the THIP current (current shift: n ¼ 9, 8; *P ¼ 0.0341; t-test; noise variance: n ¼ 10, 8; *P ¼ 0.0465; t-test). (f) Co-immunoprecipitation of GABA B Rs and d but not g2 subunits of GABA A Rs using brain extracts (full blot is shown in Supplementary Fig. 11) All histograms were presented as mean±s.e.m.  19,20 . Only one other PDZ domain-containing protein, Mupp1, is known to interact with GABA B Rs, but its physiological role remains unknown 53 . The localization of Shrm4 to microtubule-positive filaments in dendrites suggests its interaction with GABA B Rs is potentially critical for delivering these receptors to membrane-delimited signalling domains. Consistent with this, using shRNA silencing to reduce Shrm4 levels we observed decreased GABA B R in dendrites, causing their accumulation in the soma. This most likely involved dysfunctional GABA B R transport to the dendrites, a process driven by microtubule-based motor proteins: kinesins and dyneins. In this regard, we found that Shrm4 directly binds the dynein intermediate chain with a distinct portion of the PDZ domain (that is, the b1 and b2 strands) allowing for the simultaneous binding of Shrm4 with both GABA B1 and DIC. In fact we found that GABA B Rs and the dynein intermediate chain co-associate, and significantly, dynein inhibition reduced GABA B R transport to the dendrites, but had no effect on trafficking into axons. Therefore, both dynein and Shrm4 are required to target GABA B Rs to dendrites.
We know that Shrm4 can associate with both GABA B1a and GABA B1b , and that Shrm4 silencing reduced the levels of both isoforms in the dendrites. However, GABA B1a is preferentially sorted to axons under physiological conditions 31 being targeted via its Sushi domains in pre-Golgi ER or ERGIC 25 . We surmised that only GABA B1a subunits that escape axonal targeting in pre-Golgi compartments can then associate with Shrm4 in the Golgi to be re-directed (like GABA B1b ) to dendrites 54 .
On the basis of this premise, we propose that Shrm4, similar to the ID protein PQBP1, functions as an adaptor protein 55 for intracellular trafficking of cargo, by binding the C termini of GABA B1 and also dynein to target these receptors to dendrites. This is further supported by our finding that in vivo Shrm4 silencing reduced the association between GABA B Rs and dynein and differs from other post-translational modifications that influence the kinetics and pharmacological properties of GABA B Rs 12 .
trafficking and depletion of surface dendritic GABA B Rs without affecting GABA A R-mediated mIPSCs and glutamate-mediated K þ currents. Depleting postsynaptic GABA B Rs in vivo increased seizure susceptibility, as noted for GABA B1 À / À knockout mice 56 ; the weaver mouse (with mutated GIRK2 channel), and the Girk2 À / À null mouse, all of which are characterized by a significant loss of GABA B R-mediated inhibition 53,57 . Thus it is not surprising that Shrm4 silencing, or the disruption of Shrm4-GABA B Rs interaction, are associated with augmented neuronal excitability, as showed by the increase in spontaneous spikes recorded by EEG and in seizures susceptibility after PTZ injection.
GABA B R involvement in anxiety-related disorders remains unclear. While GABA B1 À / À knockouts show increased anxiety 58 , mice lacking either GABA B1a or GABA B1b isoforms are not anxious 40 , possibly due to isoform compensation. However, hippocampal Shrm4 silencing was associated with anxiogenic behaviour, and as this impairs trafficking of both GABA B R isoforms to dendrites, the increased anxiety is likely due to a reduction in postsynaptic GABA B R numbers. Social behaviours also appeared to be impaired on Shrm4 knockdown; this is not surprising since GABA transmission has been linked to ASD symptoms [40][41][42] . In addition, down-regulating GABA B Rs by Shrm4 silencing surprisingly affected tonic inhibition. We found GABA B Rs and d subunit-containing GABA A Rs co-associated, which is consistent with their extrasynaptic localization in the molecular layer of the DG 59,60 . By silencing Shrm4, tonic inhibition mediated by d subunit-containing GABA A Rs is impaired, which could also contribute towards increased seizures, a feature noted previously for Gabrd À / À mice 61 .
Defects in dendritic spine density and shape are established pathological correlates of X-linked IDs 21,62 and Shrm4 silencing, before synaptogenesis, reduced spine density, length and affected their pre-and postsynaptic molecular composition. Shrm4 binds ARTICLE F-actin and can influence actin remodelling in non-neuronal cells 4 and F-actin dynamic is crucial for synaptogenesis and spine plasticity 63,64 . Although Shrm4 could regulate the spine cytoskeleton, this is considered unlikely because Shrm4 silencing after maturation has no effect on spine morphology, even though dendritic GABA B R number were still reduced. This could explain why LTP induction and stability was unaffected by Shrm4 knockdown and is consistent with the maintenance of LTP in GABA B1b À / À knockout mice, indicating that postsynaptic GABA B Rs are not affecting LTP induction 31 . The dendritic spine defects might derive indirectly from impaired GABA B R trafficking caused by Shrm4 silencing. Indeed, GABA B R silencing before synaptogenesis, or the inhibition of Shrm4-GABAB interaction with Tat-859-870, induced a similar reduction in spine density. In addition, given GABA B Rs modulate dendritic Ca 2 þ signals 65 , by inhibiting voltage-gated Ca 2 þ channels, perturbation in Ca 2 þ signalling due to altered trafficking of GABA B Rs could also underlie spine defects.
In conclusion, our data identify Shrm4 as an important protein for synaptogenesis and for maintaining the inhibitory equilibrium mediated by the GABA B Rs and extrasynaptic d subunit-containing GABA A Rs. The consequences of disrupting Shrm4 expression are severe and manifest by increased anxiety, social behaviours impairments and a predisposition towards epilepsy.

Methods
Animals. Experimental procedures were performed in accordance with the European Communities Council Directive (86/809/EEC) on the care and use of animals and the UK Animals (Scientific Procedures) Act 1986, and were approved by the CNR Institute of Neuroscience.
Cell-permeable peptides (CPPs) were obtained from Primm (Italy) and China Peptide (China) (Tat control: YGRKKRRQRRR-TETWGQRSIE; Tat-859-870: YGRKKRRQRRR-ITRGEWQSET). Lyophilized CPPs were resuspended in sterile deionized water and used at the final concentration of 10 mM for pull-down treatments, 20 mM for hippocampal culture treatments whereas were administered by intraperitoneal injection at 3 nmol g À 1 in vivo 66 .
Cell cultures, transfection and lentiviral infection. HEK293 (Thermo Fisher Scientific) cells at 50-70% confluence (24 h after plating) were transiently transfected with cDNA expression constructs (0.5-1 mg DNA per well in optiMEM, Invitrogen) using lipofectamine 2000 (Invitrogen) for 1 h in 5% CO 2 at 37°C. Transfected cells were washed twice with PBS and grown for 48 h in Dulbecco's modified Eagle medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin, before fixation for immunocytochemistry, lysis for co-immunoprecipitation, or GST pull-down. The 293FT cell line was used to generate lentivirus and was grown in 1% G418 antibiotic. Primary hippocampal neurons were prepared from Wistar E18 rat brains 68,69 and plated onto coverslips coated with poly-D-lysine (0.25 mg ml À 1 ) at 75,000 per well for immunochemistry, or 300,000 per well for GST pull-down, co-immunoprecipitation and lentivirus infection. Neurons were transfected or infected at 7-8DIV or 12-13DIV and processed for experiments at 16-20DIV. For transfection, lipofectamine 2000 was used. Infection with Shrm4-shRNA or scrambled shRNA was performed as described previously 70 .
GST pull-down and immunoprecipitation. GST fusion proteins were prepared in BL21 E. coli and purified using standard procedures. For co-immunoprecipitation, HEK293 cells, cultured neurons or rat brain homogenates (homogenization buffer: 50 mM TRIS-HCl, 200 mM NaCl, 1 mM EDTA, 1% NP40, 1% Triton X-100, pH 7.4, protease inhibitor cocktail) were centrifuged at 10,000g for 30 min at 4°C, and supernatants incubated with antibodies (see below) at 4°C overnight. Protein A-agarose beads (GE Healthcare, USA) were then incubated with the supernatants at 4°C for 2 h. Beads were washed three times with lysis buffer, resuspended in 3 Â sample buffer and after boiling for 5 min, the resulting solution was analysed by SDS-PAGE, followed by western blotting with antibodies (see below).
For in vitro interaction assays, His-tagged DIC was expressed in E. coli BL21 strain and purified with a Protino Ni-Ted Kit (Macherey-Nagel). Eluted DIC using imidazole gradient was incubated with GST-PDZ and mutant GST-PDZ (AA) for 3 h at room temperature, washed four times with PBS and analysed.
Colocalization analysis-Pearson correlation coefficient (r). Pearson correlation coefficient (r) statistic was used to analyse the linear colocalization between fluorophores using the JACoP plugin of ImageJ. For two directly interacting proteins, the colocalization value of r tends towards 1. For high correlation, r is between 0.5 and 1; medium correlation: 0.3 and 0.5; low correlation: 0.1 and 0.3.
Image acquisition. Confocal images were obtained as described previously 72 . Briefly, fluorescence images were acquired with an LSM510 Meta confocal microscope (Carl Zeiss; gift from F. Monzino) and a Â 63 objective (numerical aperture 1.4) with sequential acquisition setting, at 1,280 Â 1,024 pixels resolution.
Image data were Z series projections of about 6-10 images, each averaged four times and taken at depth intervals of 0.75 mm.
Quantification of GABA B R fluorescence intensity. Images were quantified as previously described in ref. 72. Images were acquired using a Â 63 objective and average intensity of signals in proximal axon and primary dendrites were measured in ImageJ. Brevican, a proteoglycan present in the growth cone was used to identify the beginning of the axon and confirmed axonal identification. To prevent selection bias during quantification, the axon and dendritic segments were selected in one channel (GFP to visualization neuronal morphology) and quantified in the other channel (GABA B Rs). A third channel was used to identified the axon (Brevican). The axon and dendritic signals were measured in segments of the same size. To control for background signals, we measured the intensity near the axon or dendrite (same segment size) and subtracted the random fluorescence intensity in these images. The average dendrite intensity I d and average axonal intensity I a was used to calculate the polarity index (PI) using PI ¼ (I d À I a )/(I d þ I a ). For uniformly distributed proteins I d ¼ I a and PI ¼ 0, whereas PI40 or PIoindicates polarization towards dendrites and axons, respectively 30 .
Calcium imaging and analysis. Ca 2 þ transients were imaged at 20 Hz in Krebs using GCaMP6 in hippocampal neurons transfected at 7DIV with scrambled, Shrm4 knockdown and GABAB1DC constructs along with cDNAs for GCaMP6f and dsRed. Fluorescence intensities (F) from ROIs, drawn around dendrites were normalized to baseline fluorescence (F 0 ) to obtain values of DF/F 0 . DF/F 0 peaks were detected in Matlab using the Peakfinder plugin (Nathanael Yoder, Mathworks). Ca 2 þ transients less than Â 3 the signal-to-noise ratio were excluded from the analysis.
dSTORM and analyse of cross-correlation. Super-resolution localization imaging was carried out by direct stochastic optical reconstruction microscopy (dSTORM) 73 . Briefly, the fluorescent molecules of a sample prepared according to a standard immunofluorescence protocol (see above) were induced to blink on and off by modifying the chemical composition of the medium. If few molecules fluoresce at a given time, such that the diffraction limited spots corresponding to individual fluorophores are well separated, the position of each molecule can be quantified with higher accuracy than the resolution limit 74 . By acquiring multiple images of the same field (tens to hundreds of thousands), and storing the position of the localized molecules at each frame it is therefore possible to obtain an image of the sample with resolution higher than the diffraction limit. dSTORM was performed on an optical setup based on a Leica SR GSD-3D (Leica Microsystems Srl, Milan, Italy) super-resolution microscope equipped with a Â 160 1.43 NA objective, an Andor iXon Ultra-897 EM-CCD sensor and three (405 nm 30 mW, 488 nm 300 mW and 642 nm 500 mW) solid state lasers. The sample was mounted on the stage and the medium was substituted before acquisition with a mix of glucose oxidase (560 mg ml À 1 ), caltalase (400 mg ml À 1 ) and Cysteamine HCl (100 mM) in TN buffer with 10% glucose w/v at pH 8 to induce blinking of the fluorophores 75 . Alexa 647 and Atto 488 were imaged sequentially, starting from the red channel. 30,000 images were collected for each channel with 5-20 ms exposure times and an increasing ramp of 405 nm laser intensity, with powers between 0 and 0.8 mW, to reactivate the molecules in long-lived dark states. The super-resolution images were reconstructed using the proprietary analysis software, discarding all the detected events with less than 20 photons/pixel for Alexa 647 and 40 photons for Atto 488. To quantify the co-clustering between the two labelled proteins, the list of particles detected for each channel were used to compute the spatial cross-correlation between the two signals 76 , using previously published routines based on fast Fourier transformation written in Matlab 77 . Briefly flat crosscorrelation curves are representative of no co-clustering between the two labelled proteins, while curves that significantly differ from 1 indicate that the two proteins cluster together. For each condition, the cross-correlation was evaluated on B30 randomly chosen square regions of 2 mm sides. For visualization purposes, superresolution images were rendered as a 2D histograms with pixel sizes equal to 20 nm.
Y2H screening. For Y2H experiments, a fragment corresponding to the Shrm4 N-terminal PDZ domain (aa 1-91) was cloned in-frame with the GAL4 DNAbinding domain of the pGBKT7 vector and used as bait to screen an adult human brain cDNA library (Clontech, Mate and Plate Library). Positive yeast clones grew on plates containing X-GAL and Aureobasidin A (QDO/X/A plates) and expressed all four reporter genes: HIS3, ADE2, AUR1C and MEL1 under the control of three distinct Gal4-responsive promoters. cDNA plasmids from positive clones were recovered with the Easy Yeast plasmid isolation kit and transformed into DH5 E. coli grown on ampicillin plates, followed by sequencing.
Cell surface biotinylation assay. Hippocampal neuron membrane proteins were biotinylated using membrane-impermeable sulfo-NHS-SS-biotin (0.3 mg ml À 1 , Pierce) for 5 min at 37°C. The neurons were then washed with tris-buffered saline (TBS) supplemented with 0.1 mM CaCl 2 , 1 mM MgCl 2 and 50 mM glycine at 37°C, and rinsed with TBS supplemented with 0.1 mM CaCl 2 and 1 mM MgCl 2 (without glycine) on ice, followed by lysis in extraction buffer (50 mM Tris-HCl, pH 7.4, 1 mM EDTA, 150 mM NaCl, 1% SDS, and protease inhibitors). Lysates were boiled for 5 min and biotinylated membrane proteins were precipitated with streptavidinconjugated beads (Dynabeads, Invitrogen). The beads were washed with lysis buffer, boiled for 5 min in sample buffer, filtered, and the proteins in solution separated by SDS-PAGE for western blotting.
Live-cell internalization assay. GABA B R internalization in live hippocampal neurons was investigated as described previously 78,79 . Neurons at 14-21DIV expressing GABA B R heterodimers (R1a with bungarotoxin (BTX) binding site along with R2) and either GFP-shRNA or GFP-scrambled-shRNA were incubated in 1 mM d-tubocurarine for 2 min at room temperature followed by 3 mg ml À 1 BTX-AF555 (which attaches to the R1a binding site) for 10 min at room temperature. Receptor internalization was then followed as decay in surface fluorescence at 30-32°C under the confocal microscope.
Live-cell imaging and FRAP. For live-cell imaging, neurons co-transfected with GABA B1 -RFP together with either GFP, Shrm4-shRNA or GFP-p150-cc1, were placed in an incubator with imaging medium at 37°C and 5% CO 2 mounted on a Zeiss LSM510 Meta confocal microscope. Neurons expressing Shrm4-shRNA or scrambled shRNA were imaged with the 458 nm laser; neurons expressing GABA B1 -mRFP were imaged with the 543 nm laser.
FRAP experiments were performed following Fossati et al. 80 on GABA B1 -RFPpositive dendrites; regions of interest (ROI) on dendrites were defined and a pre-bleaching image acquired at the start. The ROI was next bleached by scanning 30 times with 405 and 458 nm lasers until no fluorescence signal was detectable. Fluorescence signal recovery was imaged over 10 min, and normalized to the total fluorescence of the pre-bleached ROI, which was verified as constant over time. The images were analysed with ImageJ and the results analysed with Prism (GraphPad).
Computational modelling. The structural models of GABA B1b and GABA B2 were developed by comparative modelling using MODELLER 81 . We first modelled the 591-918 sequence GABA B1b , comprising the seven-helix bundle, the intracellular and extracellular loops, and the C terminus up to the end of the coil-coil region (879-918 sequence) by using the crystal structure of the metabotropic glutamate receptor 5 (mGluR5; PDB 5CGC) as a template. One hundred models were generated by randomizing all the Cartesian coordinates of standard residues in the initial model. Extra-helical restraints were imposed to the following portions of the receptor: 766-772 and 879-918, corresponding, respectively, to the N terminus of helix 5 (H5) and the C terminal coil-coiled helix. In the model selected according to a corroborated procedure 82 , the 887-918 portion of the C-terminal helix was replaced with the corresponding helix from the crystallographic coil-coil heterodimer (PDB: 4PAS). The best selected model of the GABA B1b served as a template to build the seven-helix bundle and the loops of the GABA B2 R. The best model among the 100 produced was docked onto the structural model of GABA B1b R following an already described procedure 82 . The receptor orientation corresponding to an H1-H1, H7-H7 dimer, the most compatible with the C terminal coil-coil heterodimer, was employed for model completion by adding the C terminus. In this respect, the C-terminal helix was extracted from the crystallographic heterodimer (PDB: 4PAS), whereas the segment connecting the C-terminal end of H7 and the coil-coiled helix was built by means of the loop routine in MODELLER.
The information from our in vitro experiments were employed to drive the docking between the 859-870 GABA B1b R and the b2 strand of Shrm4-PDZ and between the b1 strand of Shrm4-PDZ (PDB: 2EDP) and the 128-135 strand of IC2 in complex with dimeric LC8 of dynein (PDB: 2PG1). The two strands were docked so as to form an extension of the antiparallel sheet between IC2 and LC8.
Injection of adeno-associated virus constructs. Five to seven-weeks-old male Sprague Dawley rats, kept under 12 h light/dark conditions, with ad libitum food and water, were anaesthetized with zoletil 100 (0.1 ml per 100 g) and xylazine (0.1 ml per 100 g) and secured in a Kopf stereotaxic frame. A cannula was inserted into each CA1 (ref. 22). AAV5-shRNA#1 (or #2) or AAV5-scrambled shRNA#1 (or #2) (Penn Vector Core, University of Pennsylvania, USA) were injected directly into the CA1 (coordinates AP: À 3.8 mm relative to Bregma; L: þ / À 2 mm; DV: À 2 mm from skull; Paxinos and Watson, 1985) or into the Dentate Gyrus (coordinates AP: À 3.6 mm relative to Bregma; L: 3.6 þ / À mm; DV: À 3.5 mm from skull; Paxinos and Watson, 1985). Zoanthus sp. green fluorescence protein was also incorporated into the AAV construct to report injection coordinates and volumes. The animals were allowed to recover for 3 weeks before experimentation.
Whole-cell patch-clamp recording of GIRK currents. Current densities of G protein-coupled inwardly rectifying K þ (GIRK) channel conductance in response to GABA B activation by baclofen, or mGluR activation by glutamate, were recorded from hippocampal neurons in culture at 12-16DIV, using whole-cell patch-clamp electrophysiology as described previously. Neurons were transfected at 7DIV with shRNA, scrambled shRNA or the coiled-coil domain of dynactin (GFP-p150 glued -cc1), and the cells were identified by their expression of GFP. Rescue experiments were performed from neurons co-transfected at 7DIV with shRNA#1 and HA-Shrm4 and the cells were identified by their expression of GFP. Patch pipettes made of thin-walled borosilicate glass (outer diameter, 1.5 mm; inner diameter, 1.17 mm; GC-150TF-10; Harvard Apparatus, Kent, UK), (4)(5) were filled with internal solution (mM: 120 KCl, 2 MgCl 2 , 11 EGTA, 30 KOH, 10 HEPES, 1 CaCl 2 , 1 GTP, 2 ATP, pH 7.4). Neurons were continuously perfused with Krebs solution (mM): 140 NaCl, 2.5 CaCl 2, 4.7 KCl, 1.2 MgCl 2 , 11 glucose, and 5 HEPES, pH 7.4 and voltage-clamped at À 70 mV in whole-cell configuration, and series-resistance compensation applied. Currents for 20 sweeps of À 10 mV hyperpolarising pulses were recorded with an Axopatch 200B amplifier (Molecular Devices, CA, USA) at 10 kHz filtering. Similar sweeps were recorded at regular intervals after applying series-resistance compensation of B40%. Cells were discarded in the event of a change in series resistance 410%. To increase the size of GIRK currents and to convert them to inward currents, before baclofen or glutamate application, the KCl concentration was increased to 25 mM and the NaCl concentration reduced to 120 mM in the Krebs solution. This changed the E K from approximately À 90 to À 47 mV. In addition, 2 mM kynurenic acid and 20 mM bicuculline were added to the Krebs to block ionotropic glutamate and GABA receptors. Peak K þ current amplitudes were filtered at 5 kHz before storage for analysis with Clampex 10 software. Whole-cell capacitance was calculated using WinWCP V4.7 software from the area under the membrane discharge curve from a clean averaged set of À 10 mV hyperpolarizing pulses before applying compensation. Current density was calculated by dividing the peak amplitude by the whole-cell capacitance.
Patch-clamp recordings of whole-cell mIPSCs. Whole-cell mIPSCs were recorded in hippocampal neuron cultures at 14DIV after transfection with either shRNA#1 or scrambled shRNA#1 at 8DIV. mIPSCs were recorded at a holding potential of À 70 mV over 2-5 min, filtered at 2 kHz and digitized at 20 kHz using Clampex 10.1 software. Neurons were perfused with external solution (mM: 130 NaCl, 2.5 KCl, 2.2 CaCl 2 , 1.5 MgCl 2 , 10 D-glucose, 10 HEPES-NaOH, pH 7.4). Sodium channel blocker (500 mM lidocaine) and NMDA, AMPA/kainate receptor blocker (3 mM kynurenic acid) were added to the external solution before recording. The pipette solution was caesium chloride based (140 CsCl, 2 MgCl 2 , 1 CaCl 2 , 10 EGTA, 10 HEPES, 2 ATP (disodium salt) mM, adjusted to pH 7.3 with CsOH). Recordings were performed with a Multiclamp 700B amplifier (Axon CNS Molecular Devices, USA). Pipette resistance was 2-3 MO and series resistance was always below 20 MO. Analyses were performed offline with pCLAMP 10.1 software using a threshold crossing principle; the detection level was set at 5 pA. Raw data were inspected visually to eliminate false events; data from cells with noisy or unstable baselines were also discarded. mIPSC population averages were obtained by aligning events at the mid-point of the rising phase.
Whole-cell GABA B R-mediated currents where recorded from CA1 pyramidal neurons at a holding potential À 80 mV following perfusion of baclofen 100 mM to evoke GABA B R currents and kynurenic acid (3 mM) to block glutamatergic transmission. At the end of each recording, the slices were perfused with GABA B R antagonist CGP55845 (5 mM) to check current specificity. The pipettes were filled with (126 K gluconate, 4 NaCl, 1 EGTA, 1 MgSO 4 , 0.5 CaCl 2 , 3 ATP (Mg salt), 0.1 GTP (Na salt), 10 glucose, 10 HEPES-KOHmM, pH 7.3). fEPSPs were evoked by stimulation (frequency 0.05 Hz) of the Schaffer collateral pathway of the CA1 region using aCSF-filled monopolar glass pipettes. fEPSPs were recorded (acquired at 20 kHz, filtered at 5 kHz) from the dendritic field of CA1 pyramidal neurons again using aCSF-filled electrodes. Input-output (I-O) curves were constructed by measuring the slope of fEPSPs evoked in response to stimulation with increasing intensity (0-1.0 mA). Stimulus strength was adjusted to give 50% maximal response. Long-term potentiation was elicited by high frequency stimulation (100 stimuli at 250 Hz) while long-term depression was elicited by low frequency stimulation (900 stimuli at 1 Hz). Recordings were acquired using Clampex 10.1 software and analysed offline with Clampfit 10.1 software.
For whole-cell patch-clamp electrophysiological recordings on dentate gyrus granule cells, coronal hippocampal slices (thickness, 250-300 mm) were prepared and incubated first for 40 min at 36°C and then for 30 min at room temperature in oxygenated (95% O 2 /5% CO 2 ). aCSF. Slices were transferred to a recording chamber perfused with aCSF at 33°C temperature at a rate of about 2 ml min À 1 .
Tonic GABAergic currents and mIPSCs were recorded at a holding potential of À 65 mV in the presence of kynurenic acid (3 mM) with the caesium chloride internal solution (above) supplemented with 5 mM QX-314 (lidocaine N-ethyl bromide, only for tonic currents).
Access resistance was between 10 and 20 MO; if it changed by 420% during the recording, the recording was discarded. For tonic currents recordings, after a baseline period of 2-5 min, GABA (5 mM) or THIP (3 mM) was added to the aCSF to increase the tonic component of the GABAergic transmission and to measure modification in extrasynaptic GABA A R subunits composition. At the end of the experiments bicuculline (20 mM) was added to block all GABAergic currents. For the recording of mIPSCs, lidocaine (500 mM) was added in the external solution and recordings were performed as previously described for culture experiments. Analysis was performed offline with Clampfit 10.1 software. CPG 54628, the selective GABA B R antagonist, was used at the final concentration of 5 mM.
Behavioural tests. Behavioural tests were carried out 3 weeks after rats were injected bilaterally with either Shrm4 AAV5-shRNA#1 (or #2) or AAV5-scrambled shRNA#1 (or #2). The rats were maintained in 2 per cage post injections before experiments were performed. Behavioural tests were carried out in the following order with a gap of a week between tests: spontaneous motor activity, elevated plus maze, marble-burying, EEG and PTZ-induced seizures. Ten animals per condition were submitted to the behavioural tests for 3 consecutive weeks and during the fourth and fifth weeks, animals of each condition were divided in two subgroups: one submitted to PTZ-induced seizures and the other to EEG. Behavioural experiments were carried out during the light phase of the light/dark cycle between 1000 hours and 1400 hours, and performed by trained observers blind to treatment.
Spontaneous motor activity test. Motor measurements were taken in AAV5-shRNA#1 (or #2) or AAV5-scrambled shRNA#1 (or #2) bilaterally injected rats. Spontaneous motor activity was evaluated as previously described 83 in an activity cage with the following dimensions: 43 cm long Â 43 cm wide Â 32 cm high (Ugo Basile, Varese, Italy), placed in a sound attenuating room. The cage was fitted with two parallel horizontal infrared beams 2 cm off the floor. Cumulative horizontal movements were counted every 10 min for 30.
Elevated plus maze test. The elevated plus maze procedure was used as described previously 84  Marble-burying test. This test of anxiety was used according to ref. 85. The test was conducted in a 43 cm Â 26 cm Â 22 cm box cage with 5 cm of fresh hardwood chip bedding. Each animal was habituated for 15 min to the cage. Then, an array of 24 standard marbles was arranged uniformly over the surface. Individual subjects were placed again in the test cage for 15 min. The number of marbles buried and the latency to the first burying was recorded. A marble was scored as buried if more than two-thirds of it was covered with sawdust. New bedding was used for each animal and marbles were cleaned with 10% acetic acid solution between animals. The same subjects were used for this test and in the elevated plus maze and were tested in a counterbalanced order.
Sociability and social novelty test. Social behaviour was carried out in a threechamber apparatus according to Leite et al. 85 The apparatus was an acrylic rectangular box divided into three compartments of equal size (35 cm height, 50 cm width, 50 cm deep) provided with doors. The sociability test was divided in three sequential phases of 10 min each. During the habituation period, the test rat was placed in the middle chamber for 10 min where the rat was free to explore the three compartments. Each of the two sides contained an identical empty wire cage. In the sociability phase, an unfamiliar rat (stranger 1) of the same strain, sex and weight was enclosed in one of the wire cages and the time spent in each compartment with the object or social stimulus was measured. In the social novelty phase, a new unfamiliar rat (stranger 2) was enclosed into the wire cage in the opposite compartment and the time spent in each compartment was measured. Before the introduction of a social stimulus the test rat was trapped in the central chamber. The test was videotaped and the time spent in each compartment was measured offline.
Tube test. The tube test is a well-known testing paradigm designed to measure social hierarchies, and thus is relevant when investigating social dominance in mice 86 but it can be adapted to rats 87 . This test measures dominant/submissive behaviour without allowing them to fight and injure each other. Rats were initially habituated to the testing apparatus, which consisted of a 10 cm (diameter) by 50 cm (length) transparent plastic tube, of sufficient size to allow one but not two rats to move through the tube. Over two consecutive days, rats were allowed to run through the tube on eight occasions, with alternate trials in which the entry and exit ends were switched. Competition trials involved simultaneously releasing two competing rats into opposite ends of the tube. The individual rat that was able to travel forwards through the tube to exit the other side 'won' and was deemed dominant; the rat that retreated was considered subordinate. The number of wins (%) on the total number of competitions was measured.
Electroencephalogram recording. AAV5-scrambled#1 and AAV5-shRNA#1 rats were anaesthetized with an i.p. injection of chloral hydrate dissolved in saline and given at a volume of 10 ml kg À 1 . Under anaesthesia (450 mg kg À 1 , body weight of chloral hydrate, i.p.; Sigma-Aldrich), all the rats were placed in a stereotaxic instrument and four silver-silver chloride ball electrodes were fixed epidurally with dental acrylic cement, as described in detail elsewhere 88 , on the right and left of the parieto-occipital cortex according to the Paxinos and Watson brain atlas, 2 mm anterior, 2 mm lateral from the midline and 3 mm posterior from the bregma. The four electrodes, and a fifth inserted into the nasal bone and used as ground, were connected to a microconnector attached to the rat's head with dental cement (New Galetti e Rossi, Milan, Italy). Animals were treated with ceftriaxone (50 mg kg À 1 i.p.) for three days. One week after electrode placement the rats were allowed to acclimatize themselves to a sound-attenuated Faraday chamber for 1 h a day for 3 days. Then, each freely moving awake rat was connected to the microconnector. Signals were amplified by an Animal Bio-Amplifier (AD Instruments), band-pass filtered (0.2-30 Hz) and then connected to a PC for signal acquisition (PowerLab system, AD Instruments, Castle Hill, Australia) at a sampling rate of 100 Hz and a resolution of 0.2 Hz. Each animal was continuously recorded for 24 h, under basal conditions. All collected data were analysed for abnormalities by an experienced observer blinded to rat condition. All EEG traces were scored for the presence of isolated spikes or repetitive spiking using an additional software (LabChart v8 Pro Windows). Spikes were defined as having a duration o200 ms with baseline amplitude set to 4.5 times the standard deviation of the EEG signal (determined during inter-spike activity periods, whereas repetitive spiking activity was defined as three or more spikes lasting o5 s).
Immunohistochemistry. AAV-injected rats have been anesthetized with zoletil100 (0.1 ml per 100 g) and xylazine (0.1 ml per 100 g) and transcardially perfused with PBS and PFA 4%/ Sucrose 4%. Dissected rat brains have been fixed ON in PFA 4%, rinsed in PBS and cryoprotected with OCT. The blocks have been sectioned (25 mm-thick) with a cryostat. Sections have been collected on glass slides, blocked with 3% BSA,10% goat serum, 0.4% Triton in PBS for 1 h at RT and then incubated with anti-PSD95 antibody (1:300 in blocking solution, Neuromab) at 4°C ON followed by three washes with PBS. Sections have been incubated for 1 hr with the secondary antibody (1:400, Alexa 555, Life Technlogies) at RT in blocking solution and washed with PBS. Sections have been mounted on coated super-adhesive glass slides and covered with Mowiol prior to confocal imaging. Images have been deconvoluted with ImageJ plugin Iterative Deconvolve 3D.
Immuno-Electron Microscopy. Paraformaldehyde fixed neurons were incubated with a polyclonal primary anti-Shrm4 diluted (1:100) in PBS containing 5% normal goat serum, 0.1% saponin for 1 h; they were then incubated with a secondary antibody conjugated with 1.4 nm gold particles (Life Technologies, CA, US) and fixed with 1% glutaraldehyde in PBS. After washing, gold enhancement was performed using the GoldEnhance EM 2113 kit (Nanoprobes, NY, US). Cells were postfixed with 0.2% osmium tetroxide in 0.1 M phosphate buffer, stained with 0.25% uranyl acetate, dehydrated and embedded in epoxy resin. The distribution of gold particles on ultrathin sections of cortical neurons was assessed adapting a randomness test (Mayhew et al., 89 ). To evaluate the distribution of Shrm4 on our sections, the following compartments were defined: synaptic boutons, post-synaptic terminals, dendrites, un-determined structures and areas where cell structure is absent.
Data availability. The authors declare that the data supporting the findings of this study are available within the paper and its Supplementary Information files.