Rheb activation disrupts spine synapse formation through accumulation of syntenin in tuberous sclerosis complex

Rheb is a small GTP-binding protein and its GTPase activity is activated by the complex of Tsc1 and Tsc2 whose mutations cause tuberous sclerosis complex (TSC). We previously reported that cultured TSC neurons showed impaired spine synapse morphogenesis in an mTORC1-independent manner. Here we show that the PDZ protein syntenin preferentially binds to the GDP-bound form of Rheb. The levels of syntenin are significantly higher in TSC neurons than in wild-type neurons because the Rheb-GDP-syntenin complex is prone to proteasomal degradation. Accumulated syntenin in TSC neurons disrupts spine synapse formation through inhibition of the association between syndecan-2 and calcium/calmodulin-dependent serine protein kinase. Instead, syntenin enhances excitatory shaft synapse formation on dendrites by interacting with ephrinB3. Downregulation of syntenin in TSC neurons restores both spine and shaft synapse densities. These findings suggest that Rheb-syntenin signalling may be a novel therapeutic target for abnormalities in spine and shaft synapses in TSC neurons. Tuberous sclerosis complex (TSC) arises from mutations in an activator of the small GTPase Rheb. Here the authors show that syntenin binds to GDP-bound Rheb, and loss of this interaction in TSC leads to increased syntenin expression and downstream signalling defects leading to aberrant spine synapse morphogenesis.

D endritic spine formation and synaptogenesis are often preceded by a period in which numerous elongated filopodia extend from developing dendrites. The impairment in the conversion of filopodia to dendritic spines may lead to childhood cognitive disorders 1 . Synaptic abnormalities in neurodevelopmental disorders may be caused by mutations that affect synaptic proteins. For example, mutations in fmr1, identified in fragile X syndrome, alter the properties of the FMRP protein, resulting in spine dysmorphogenesis 2 . The E3 ubiquitin ligase Ube3A, linked to Angelman syndrome, also controls spine formation 3 . Moreover, another autism-related protein neurobeachin regulates spine density and shaft synapse formation 4 . Thus, childhood mental disorders may be accompanied by morphological abnormalities of excitatory spine and/or shaft synapses.
Tuberous sclerosis complex (TSC) is a systemic genetic disease that causes the growth of hamartomas in the brain and other organs 5 . The neuropsychiatric symptoms of TSC patients include intellectual disability, epilepsy and/or autism 5 . TSC results from inactivating mutations in either Tsc1 (ref. 6) or Tsc2 (ref. 7), which encode hamartin and tuberin, respectively. Hamartin and tuberin form a complex that exhibits guanosine triphosphatase (GTPase)-activating protein activity towards the small GTPbinding protein Rheb [8][9][10] . Inactivating mutations in Tsc1 or Tsc2 result in the accumulation of the GTP-bound form of Rheb, which activates mammalian target of rapamycin complex 1 (mTORC1; ref. 11). A Golgi study of surgical specimens from a TSC patient reported that cortical neurons showed aberrant spine development 12 . Several lines of evidence also suggest that Tsc1 þ / À or Tsc2 þ / À rodents show impairments in dendritic spine formation 13,14 and cognitive behaviours 15 , suggesting that the activation of mTORC1 in TSC neurons may cause memory disturbances, likely through dendritic spine abnormalities. However, the mTORC1 signalling pathway has been shown to be activated during learning and is required for spatial memory formation. This discrepancy may indicate that an mTORC1-independent mechanism may be involved in memory impairments in TSC.
Syndecan-2 is a transmembrane heparan sulphate proteoglycan that promotes spine formation [16][17][18] . The C-terminal EFYA motif of syndecan-2 is necessary for dendritic spinogenesis, and this motif binds to the PSD-95, Dlg and ZO1 (PDZ) domains of calcium/calmodulin-dependent serine protein kinase (CASK). CASK is known to associate with F-actin via its binding to protein 4.1 (ref. 18), and it may mediate spine morphogenesis through syndecan-2. However, the C terminus of syndecan-2 also binds to syntenin, a scaffold protein with two PDZ domains 19 . In contrast to CASK, overexpression of syntenin in hippocampal neurons increases filopodia-like dendritic protrusions 20 , suggesting that syntenin competes with CASK for syndecan-2.
In the present study, we propose a novel mechanism for dendritic spine morphogenesis. We previously showed that activation of Rheb, but not mTORC1, impaired spine synapse formation in TSC neurons 14 . To further elucidate this mechanism, we sought to detect a new Rheb-mediated signalling pathway in the brain and to identify syntenin as a novel Rheb-binding protein. Syntenin preferentially binds to the GDP form of Rheb, and this complex is prone to proteasomal degradation in wild-type (WT) neurons. The consequent reduction in syntenin levels allows CASK to associate with syndecan-2, leading to stabilization of spine structure. In Tsc2 þ / À neurons, the syntenin released from Rheb-GTP associates with syndecan-2, resulting in decreased spine synapse density. Syntenin also interacts with ephrinB3 and increases the density of excitatory shaft synapses. In fact, downregulation of syntenin restores spine and shaft synapse densities in Tsc2 þ / À neurons. These results suggest that syntenin is a regulator of spine and shaft synapse density, and link syntenin to a novel pathophysiological mechanism implicated in neurodevelopmental disorders.

Results
Rheb associates with syntenin. We previously reported that dendritic spine synapse density was decreased in TSC neurons through an mTOR-independent mechanism 14 . We found that activation of Rheb, but not mTORC1, reduced spine synapse density, and the expression of a dominant-negative form of Rheb or treatment with farnesyl transferase inhibitors largely restored spine synapse density in Tsc2 þ / À neurons. These results suggest that the mTOR-independent mechanism controlling spine synapse density involves Rheb, and that this mechanism may be dysregulated in TSC neurons.
To elucidate this mechanism, in this study, we searched for a Rheb-associating protein using the yeast two-hybrid system. We identified a PDZ domain-containing scaffold protein, syntenin 19 , as a novel Rheb-binding partner. The potential interaction between Rheb and syntenin was confirmed by coimmunoprecipitation of Rheb with syntenin from rat brain lysates (Fig. 1a). Furthermore, Rheb D60I protein was pulled down with glutathione S-transferase (GST)-fused syntenin (Fig. 1b). The association of syntenin with Rheb appeared to depend on the nucleotide-binding state of Rheb, as the GDP-bound form (Rheb D60I ) exhibited a significantly stronger binding to syntenin than did the GTP-bound form (Rheb Q64V ; Fig. 1c,d).
We localized the Rheb-binding region of syntenin to its C-terminal segment but not to either PDZ domain because the affinity of Rheb to syntenin was significantly reduced by the C-terminal deletions (D278-300 and D289-300) but not by the mPDZ1 or mPDZ2 (ref. 19; Fig. 1e,f). Furthermore, syntenin  showed no interaction with Rheb ( Fig. 1g), indicating that the Rheb-binding region of syntenin is localized between residues 289 and 294. Conversely, syntenin interacted with the switch I region (also called the effector region) of Rheb in pull-down experiments. Specifically, a T38M, but not I39K, mutation in Rheb D60I reduced the binding to syntenin (Fig. 1h), suggesting that residues exposed only in Rheb-GDP form may mediate the binding of the two proteins 21 .
Syntenin regulates dendritic spine synapse density. To elucidate the role of the Rheb-syntenin interaction in spine synapse morphogenesis, we first compared syntenin levels in WT and TSC neurons. We used spontaneous Tsc2-mutated rats (Eker) as a model of TSC 22 . We cultured hippocampal neurons of WT and heterozygous Eker (Tsc2 þ /-) rat embryos, fixed the cultures at 21 days in vitro (DIV) and immunostained the neurons with antisyntenin antibody. As shown in Fig. 2a,b, syntenin levels were significantly higher in Tsc2 þ / À neurons than in WT neurons. This result was consistent with an increase in syntenin levels in Tsc2 þ / À brains compared with those in WT brains (Fig. 2c,d), suggesting that the Tsc2 mutation may affect neuronal syntenin levels, possibly via Rheb.
To test whether the nucleotide-binding state of Rheb regulates syntenin levels, we overexpressed the GTP-bound form of Rheb (Rheb Q64V ) in cultured hippocampal neurons 14 . The expression of active Rheb significantly increased both the somatic and dendritic levels of syntenin in hippocampal neurons compared with adjacent untransfected neurons ( Supplementary Fig. 1a,b). By contrast, the expression of Rheb D60I decreased syntenin levels in cultured neurons ( Supplementary Fig. 1c,d). These results suggest that the nucleotide-binding state of Rheb controls the syntenin levels in hippocampal neurons.
To confirm the effects of neuronal syntenin levels on spine and synapse formation, we overexpressed or knocked down syntenin in primary cultures of hippocampal neurons. Overexpression of syntenin in WT neurons decreased and increased dendritic protrusion width and length, respectively (Fig. 2e,f). Furthermore, syntenin expression drastically reduced spine synapse density (Fig. 2e,g), as observed previously in Rheb Q64V -expressing WT neurons or Tsc2 þ / À neurons 14 . We then knocked down the endogenous syntenin levels in Tsc2 þ / À neurons, whose syntenin levels were higher than those in WT neurons. Syntenin short interfering RNA (siRNA) reduced the expression of exogenous syntenin by 92.4% in HEK293T cells ( Supplementary Fig. 1e). The transfection of syntenin siRNA into Tsc2 þ / À neurons reduced the endogenous syntenin levels ( Supplementary Fig. 1f) and induced wider and shorter dendritic protrusions (Fig. 2h,i). Syntenin knockdown also drastically enhanced spine synapse density in Tsc2 þ / À neurons (Fig. 2j). Accordingly, a reduction in syntenin levels may restore spine synapse density in Tsc2 þ / À neurons. By contrast, unexpectedly, knockdown of syntenin in WT neurons also disrupted spine synapse morphogenesis (Fig. 2h,j) and produced narrower and longer filopodia-like protrusions (Fig. 2i). Thus, both increasing and decreasing syntenin levels can affect spine synapse formation, suggesting that neuronal syntenin levels must be maintained within a narrow range for normal spine development.
The Rheb-syntenin complex is degraded by proteasomes. What mechanism underlies the increase in syntenin levels in TSC neurons? One possibility is the upregulation of syntenin production, and another is the inhibition of syntenin degradation. The former is unlikely because the mTORC1 inhibitor rapamycin failed to reduce syntenin levels in Tsc2 þ / À neurons (unpublished observations). Recent work has demonstrated that the proteasome system regulates activity-dependent spine outgrowth 23 . Therefore, we assumed that syntenin levels may be regulated by the ubiquitin-proteasome system and that normal steady-state levels of syntenin degradation are required for spine synapse development. We treated Rheb D60I -expressing Tsc2 þ / À neurons with the proteasomal inhibitors lactacystin (0.5 mM) or MG132 (0.1 mM) for 2 days, and observed that the Rheb D60Iinduced decreases and increases in syntenin levels and spine synapse density, respectively, were prevented by either inhibitor (Fig. 3a-d; unpublished data). Proteasomal inhibitor alone did not change spine synapse density in WT neurons (Fig. 3d), consistent with a previous study 23 . Furthermore, knockdown of   syntenin rescued spine synapse density in lactacystin-treated neurons (Fig. 3c,d). Thus, Rheb D60I -induced spine synapse morphogenesis in Tsc2 þ / À neurons may be closely related to the reduction in syntenin levels by proteasomal degradation.
To further elucidate the mechanism of syntenin downregulation, we investigated whether proteasomal degradation of syntenin depends on the nucleotide-binding state of Rheb. Syntenin was co-expressed with Rheb D60I or Rheb Q64V in COS-7 cells, which were treated with MG132 (ref. 24) or untreated. We compared the syntenin levels between MG132-treated and -untreated cells, and observed that the proteasomal degradation of syntenin was enhanced in the Rheb-GDP state compared with that in the Rheb-GTP state (Fig. 3e,f). Moreover, the GDP-bound form of Rheb was more sensitive to proteasomal degradation than Rheb-GTP (Fig. 3e,g). Next, we tested whether the association between syntenin and Rheb is required for the proteasomal degradation of both proteins. We co-expressed Rheb D60I with either intact syntenin or syntenin DC(289-294) , a mutant with a deletion of the Rheb-binding site, in COS-7 cells. Syntenin DC(289-294) drastically reduced the degradation of Rheb D60I (Fig. 3h,i), indicating that the proteasomal degradation of Rheb is enhanced by its interaction with syntenin.
Syntenin competes with CASK for association with syndecan-2.
The PDZ2 domain of syntenin has been associated with various proteins 25 . Of the more than 20 syntenin-binding proteins, only syndecan-2 has been demonstrated to regulate dendritic spine formation 17,18,26 . To clarify whether syndecan-2 is involved in aberrant spinogenesis in TSC neurons, we downregulated both syntenin and syndecan-2 in Tsc2 þ / À neurons. Syndecan-2 siRNA reduced endogenous and exogenous syndecan-2 levels in cultured neurons and HEK293T cells, respectively ( Supplementary Fig. 2d,e). Knockdown of syndecan-2 alone abolished normal spine synapse formation in WT neurons, confirming that syndecan-2 plays a crucial role in normal spine development. Similarly, syndecan-2 knockdown in Tsc2 þ / À neurons did not restore spine formation, but concomitant reduction in syndecan-2 abolished the recovery of spine synapse density produced by syntenin siRNA (Fig. 4a,b). These results suggest that increased syntenin may disrupt spine synapse formation through association with syndecan-2.
Similar to syntenin, CASK interacts with the C-terminal EFYA motif of syndecan-2 and promotes dendritic spine maturation 16,27 . We examined whether the binding of syndecan-2 to CASK and syntenin is mutually exclusive. We first tested the ability of CASK to induce spine synapse formation. The overexpression of CASK in Tsc2 þ / À neurons also increased spine synapse density as well as syntenin knockdown (Fig. 4c,d); however, syntenin levels were not changed by CASK expression ( Supplementary Fig. 2f,g), suggesting that an increase in CASK relative to syntenin restores spine synapse formation in Tsc2 þ / À neurons. Further increases in syntenin impaired the ability of CASK to enhance spine synapse formation (Fig. 4e,f). We then examined whether the inhibitory effects of syntenin reflect competition between CASK and syntenin for the binding to the C terminus of syndecan-2. We fused GST to the intracellular domain (IC) of syndecan-2 to pull-down CASK protein. In the absence of FLAG-syntenin, CASK was pulled down with GST-syndecan-2 IC. However, increasing the amount of FLAGsyntenin in cell lysates suppressed the pull-down of CASK with GST-syndecan-2 in a dose-dependent manner (Fig. 4g,h), indicating that syntenin competes with CASK for the binding to syndecan-2.
The protein 4.1-binding motif of CASK is involved in spine morphogenesis 27 . Protein 4.1 binds to spectrin 28 and promotes the interaction between spectrin and filamentous actin 28,29 ; thus, CASK may act as a linker between transmembrane proteins and the actin cytoskeleton, thereby stabilizing dendritic spines. We therefore sought to determine whether the protein 4.1-binding site of CASK plays a pivotal role in spine synapse morphogenesis. The CASK mutant lacking the protein 4.1-binding motif (CASK Dp4.1 ; Fig. 4c, bottom) lost the ability to increase spine synapse density in Tsc2 þ / À neurons (Fig. 4d). These results indicate that the protein 4.1-binding site of CASK is required for spine synapse formation in Tsc2 þ / À neurons. Taken together, these findings indicate that syntenin may reduce spine synapse density by competing with CASK and preventing it from stabilizing spine morphology via protein 4.1.
Binding of syntenin to ephrinB3 increases shaft synapses. We previously showed that in Tsc2 þ / À neurons, excitatory synapses were rarely located on dendritic protrusions but preferentially localized to dendritic shafts 14 . Syntenin has been demonstrated to be a downstream effector of ephrinB3 (ref. 30), which modulates glutamatergic synaptic density 31 , particularly on shaft synapses 32 . To test whether syntenin is involved in excitatory shaft synapse formation, we knocked down syntenin in WT and Tsc2 þ / À neurons. Downregulation of syntenin drastically reduced excitatory shaft synapse density in Tsc2 þ / À neurons but not in WT neurons (Fig. 5a). Next, we examined the involvement of ephrinB3 in the syntenin-mediated shaft synapse formation. We knocked down the endogenous ephrinB3 in Tsc2 þ / À hippocampal neurons using ephrinB3 siRNA (Supplementary Fig. 3a,b). Eleven days after the ephrinB3 siRNA transfection, we observed a significant reduction in vGlut1 puncta on the dendritic shafts in Tsc2 þ / À neurons but not in WT neurons (Fig. 5b). EphrinB3 knockdown did not affect spine synapse density in either WT or Tsc2 þ / À neurons (Fig. 5b). Conversely, overexpression of ephrinB3 increased excitatory shaft synapse density in WT neurons (Fig. 5c), and simultaneous expression of syntenin siRNA effectively inhibited this increase (Fig. 5c), suggesting that ephrinB3 may regulate shaft synapse density via syntenin. In our experiment, the overexpression of ephrinB3 significantly reduced spine synapse density, which was inconsistent with a previous study 32 . The reason for this difference is unclear, but our overexpression of ephrinB3 plasmid was longer and stronger (10 days under the control of the CAG promoter) than that in the other study 32 (4 days under the control of the cytomegalovirus promoter). Because ephrinB3 regulates synapse density by inhibiting the postsynaptic mitogenactivated protein kinase signalling 31 that is required for increases in spine density 33 , long and strong expression of ephrinB3 might reduce spine synapse density. Future studies will be needed to characterize this mechanism.
To examine the direct interaction between ephrinB3 and syntenin, we co-immunoprecipitated ephrinB3 with syntenin from HEK293T cells expressing green fluorescent protein (GFP)syntenin and ephrinB3 (Fig. 5d). This binding likely occurred through a PDZ domain because the ephrinB3 mutant lacking the PDZ recognition motif was not associated with syntenin (Fig. 5e). Together, these results indicate that syntenin may enhance shaft synapse formation by binding to ephrinB3 through the PDZ domain.
Excitatory shaft synapses in Tsc2 þ / À neurons are functional. In a previous work, to investigate excitatory synaptic transmission in TSC neurons, we systematically investigated the electrophysiological properties of WT and Tsc2 þ / À neurons 14 . Unexpectedly, Tsc2 þ / À neurons showed sufficient synaptic transmission, with no significant differences in presynaptic or postsynaptic features between WT and Tsc2 þ / À neurons. Therefore, we proposed that increased shaft synapses in Tsc2 þ / À neurons may functionally compensate for the excitatory synaptic transmission of decreased spine synapses. To test this hypothesis, we reduced shaft synapse density in autaptic WT and Tsc2 þ / À neurons by expressing ephrinB3 siRNA, and we electrophysiologically recorded synaptic transmission in these neurons ( Fig. 6a-g). Although the control siRNA did not affect synaptic transmission in either WT or Tsc2 þ / À neurons ( Supplementary Fig. 4a-g), ephrinB3 siRNA markedly reduced the excitatory postsynaptic current (EPSC) amplitude and miniature EPSC (mEPSC) frequency without changes in the probability of vesicular release in Tsc2 þ / À neurons ( Fig. 6a-f). These data suggest that excitatory synaptic transmission between Tsc2 þ / À neurons is functionally compensated for by increases in the ephrinB3-mediated shaft synapses, leading to the apparent stabilization of synaptic transmission in Tsc2 þ / À neurons. Furthermore, we immunostained Tsc2 þ / À neurons with anti-GluR1 (GluA1) and vGlut1 antibodies to confirm that postsynaptic receptors are localized next to presynaptic puncta. Indeed, vGlut1 clusters were co-localized with GluA1 puncta on the dendritic shaft of Tsc2 þ / À neurons ( Supplementary Fig. 4h), consistent with the dendritic shaft synapses in Tsc2 þ / À neurons being functional.
Furthermore, we noticed that the number of synaptic vesicles in the readily releasable pool (RRP) was significantly reduced by ephrinB3 siRNA in Tsc2 þ / À neurons but not in WT neurons (Fig. 6g). In this experiment, we estimated the number of synaptic vesicles in the RRP by simply dividing the hypertonic sucroseevoked RRP charge by an individual average mEPSC charge; thereby, the RRP charge should involve all synaptic releases from presynaptic terminals projecting to both spine and shaft synapses. Thus, the finding that number of releasable vesicles was reduced by ephrinB3 siRNA in Tsc2 þ / À neurons also implies that dendritic shaft synapses are functional. Because vesicular release probability was unchanged in ephrinB3-knockdown neurons, we believe that presynaptic effects can be excluded. However, because presynaptic development has been shown to be regulated by syntenin 34 , exploring whether presynaptic function is changed in TSC neurons will be the focus of our next study.
We show that increased syntenin diminished spine synapse density in TSC neurons, whereas a previous study demonstrated that syntenin overexpression in WT neurons did not reduce spine synapse density 30 . This discrepancy may be caused by the use of different antibodies to detect presynaptic terminals; we immunolabelled only excitatory synapses with anti-vGlut1, whereas Xu et al. 30 used an anti-synapsin antibody that reacts with both excitatory and inhibitory terminals. Because inhibitory synapses were also increased in TSC neurons (unpublished data), the use of an anti-synapsin antibody might produce different results from this study. The other difference between this study and that by Xu et al. 30 is in the mEPSC frequency in ephrinB3-reduced neurons. We found no significant change in the mEPSC frequency in ephrinB3-knockdown WT neurons ( Fig. 6d and Supplementary Fig. 4d), whereas Xu et al. 30 showed a clear decrease in mEPSC frequency in ephrinB3-deficient neurons 30 . This discrepancy may be because of the differences in neuronal maturation. We used cultured hippocampal neurons at DIV 14-23, whereas Xu et al. 30   shafts (shaft synapses) to spines (spine synapses) during neuronal development 43 . We used mature neurons with fewer shaft synapses, which are the main targets of regulation by ephrinB3 (ref. 32). This may explain why ephrinB3 knockdown failed to reduce synaptic transmission in this study.
Reducing the syntenin levels in WT neurons also decreased the spine synapse density (Fig. 2h,i). Thus, the deviation of syntenin levels from a narrow physiological range may disrupt spine synapse formation. In contrast to Tsc2 þ / À neurons, however, syntenin þ / À neurons did not show increased shaft synapse density (Fig. 7f), suggesting that syntenin reduction may affect only spine synapses. How the reduction in syntenin levels decreases spine synapse density remains unclear; however, it is noteworthy that syntenin associates with multiple proteins, such as adhesion molecules (SynCAM 44 , neurexin and neurofascin 45 ) or cytoplasmic proteins (PICK1; ref. 46). The absence of these associations may contribute to impaired spine formation. It is possible that such distinct syntenin function regulates dendritic spinogenesis in different ways. Clarification of this mechanism will be the next step in understanding the multiple functions of syntenin.
Here, we focused on the contributions of postsynaptic syntenin and CASK to the control of spine formation; however, both proteins are also localized to presynaptic terminals and bind to b-neurexin 49,50 . To examine the presynaptic effects of Tsc2 mutations on excitatory synapse formation, we compared the size of vGlut1 puncta in WT-WT synapses and Tsc2 þ / À -WT synapses; however, we found no significant differences (0.11 ± 0.013 mm 2 for WT and 0.10 ± 0.011 for Tsc2 þ / À boutons). Because syntenin knockdown in postsynaptic TSC neurons results in spine synapse formation with presynaptic TSC neurons, spine formation may be regulated by the postsynaptic balance between syntenin and CASK rather than by presynaptic effects. Future studies are needed to characterize the effects of presynaptic syntenin and CASK on spine and shaft synapse formation.
In this study, we found that Rheb interacted with syntenin and formed a novel part of the TSC-Rheb pathway, which may help us to better understand the pathophysiology of neurological symptoms in TSC. Several preclinical studies and early human case reports have demonstrated that mTORC1 inhibitors such as sirolimus or everolimus are useful for treating seizures 39,51-54 , whereas more recent prospective trials have shown that benefits could not be demonstrated when seizure frequency was analysed as a secondary outcome measure 55,56 . Moreover, 10-20% of TSC patients showed exacerbation of seizures after everolimus treatment, even in the clinical trial that demonstrated the efficacy of everolimus 57 . Thus, in addition to mTORC1 inhibitors, new approaches for the treatment of neurological symptoms in TSC patients are needed to improve not only seizure control but also cognitive function and patient quality of life. Because Rheb expression is regulated by neuronal activity 58 , Rheb-syntenin signalling may be involved in activity-dependent learning and memory processes. Appropriate modulation of this signalling pathway can be required for normal cognitive function. Thus, the Rheb-syntenin pathway may also be a therapeutic target for relieving TSC and other cognitive disorders involving abnormalities of spine and shaft synapses.

Methods
Antibodies and reagents. Antibodies used for western blotting are summarized in Supplementary Table 1. Rapamycin, MG132 and lactacystin were purchased from LKT Laboratories, Millipore and the Peptide Institute, respectively. The siRNAs used in this study are summarized in Supplementary Table 2. Oligonucleotides were synthesized and duplexed by Invitrogen (Stealth, syntenin and syndecan-2) and Dharmacon (Accell, ephrinB3). Their suppressive activities were confirmed in rat neurons and HEK293T cells using immunostaining and western blotting, respectively.
Animals. The Eker (Tsc2 þ /-) and WT genotypes were determined using PCR 59 . Syntenin-floxed mice were obtained from the Knock-Out Mouse Project (www.komp.org; project no. CSD26463). CaMKII Cre mice 35 were obtained from European Mouse Mutant Archive. Tsc2 þ /mice were generated previously 36 Mice were maintained on a C57BL/6 background. All of the animal experiments were approved by the Animal Care and Use Committees of the Tokyo Metropolitan Institute of Medical Science and Fukuoka University.
Immunoprecipitation. Rat brains, COS-7 cells and HEK293T cells were homogenized with lysis buffer (20 mM Tris-HCl (pH 8.0), 0.15 M NaCl, 1 mM EDTA, 1% Triton X-100, protease inhibitor cocktail tablet (Roche) and 1 mM phenylmethylsulfonyl fluoride). The homogenates were centrifuged to obtain clear protein extracts. Anti-syntenin (goat, Santa Cruz Biotechnology), anti-GFP (rabbit, Invitrogen) or anti-FLAG (mouse, Sigma-Aldrich) antibodies were added to the protein extracts on ice for 1 h, and the mixtures were incubated overnight at 4°C with packed protein G Sepharose (GE Healthcare). The precipitated immune complex was boiled in SDS sample buffer and analysed using western blotting with anti-Rheb (rabbit, CST), anti-FLAG (mouse, Sigma-Aldrich), anti-GFP (rabbit, Invitrogen) or anti-ephrinB3 (rabbit, Invitrogen) antibodies 63 . All representative western blots included in the manuscript are shown as full scans in Supplementary  Fig. 6.
Recombinant protein expression and pull-down assay. For the bacterial expression of recombinant GST-tagged syntenin and the syntenin mutants, the corresponding sequences were cloned into the pGEX-4T2 vector (GE Healthcare) and expressed in BL21(DE3) cells. Each GST-fused syntenin protein was purified with glutathione beads (GE Healthcare). To examine the binding between Rheb and syntenin, the same amount of purified GST-syntenin (WT or mutant) was first attached to glutathione beads and then incubated with cell extracts expressing various versions of enhanced green fluorescent protein (EGFP)-tagged Rheb overnight at 4°C. To determine whether the nucleotide-binding state of Rheb affects the binding of syntenin to the IC of syndecan-2, the purified GST-syndecan-2 IC protein was first attached to glutathione beads and then incubated with cell extracts expressing FLAG-tagged syntenin and either EGFP-Rheb D60I or EGFP-Rheb Q64V overnight at 4°C. To examine the binding between syntenin and ephrinB3, the purified GST-syntenin was first incubated with cell extracts expressing ephrinB3 or DPDZ overnight at 4°C. In all pull-down experiments, the  However, Tsc2 disruption in tuberous sclerosis increases Rheb-GTP, which rescues syntenin from proteasomal destruction. Accumulated syntenin may inhibit the binding of CASK to sdc2 and promote its binding to ephrinB3, resulting in impaired spine morphogenesis and enhanced shaft synapse formation (right, Tsc2 þ / À ).
beads were washed with lysis buffer and boiled in SDS sample buffer. The precipitates were analysed using western blotting with anti-GFP, anti-FLAG or anti-ephrinB3 antibodies. All representative western blots included in the manuscript are shown as full scans in Supplementary Fig. 6.
For spine and shaft synapse density analysis, dendrites were chosen randomly and their lengths were measured in two-dimensional projection of threedimensional image stacks. Spines were defined as any protrusions from the dendritic shaft o8 mm in length 32 . The number of vGlut1 puncta on spine and shaft synapses of the dendritic segment is determined by counting the number of vGlut1 puncta that partially overlap with the GFP-positive shaft or spines. Any vGlut1 puncta that are not touching the GFP-positive dendrite are not included.
Image acquisition and quantification. Confocal images of neurons were obtained using a Zeiss 63 Â lens (numerical aperture 1.4) with sequential acquisition settings at the resolution of the confocal microscope (512 Â 512 pixels). Each image was a composite constructed from a series of images captured throughout the z axis of each cell. The parameters of each composite image were optimized for the particular lens and pinhole setting. Identical confocal microscope settings were maintained for all scans in each experiment. All morphometric analyses and quantifications were performed using MetaMorph image analysis software (Universal Imaging). For the spine morphology studies, individual dendritic protrusions were manually traced, and the maximum length and head width of each dendritic protrusion were measured and recorded in Microsoft Excel. To measure the average pixel fluorescence intensity of syntenin, the neurons were carefully traced and the background intensity was subtracted from the intensity in the traced regions. Neurons were selected blindly based on the GFP or MAP2 fluorescence.
Autaptic culture. Astrocytes were obtained from cerebral cortices of newborn (P0) ICR mice (Kyudo). Cerebral cortices were removed from the brains, placed in cold Hank's Balanced Saline Solution (Invitrogen) and dissociated using 0.05% trypsin-EDTA (Wako). The cells were plated in plating medium composed of Dulbecco's modified Eagle medium with GlutaMax and pyruvate (DMEM, Invitrogen) supplemented with 10% fetal bovine serum (Invitrogen) and 0.1% MITO þ Serum Extender (BD Biosciences) in 75 cm 2 culture flasks (250 ml, BD Falcon). The next day, the culture flask was gently rinsed once with plating medium to remove nonadherent cells. After 2 weeks, when the glial cells had almost reached confluence, microglia and other small cells were isolated and removed by tapping the culture flask several times. The remaining astrocytes adhered to the bottom of the culture bottle were subsequently trypsinized and plated onto microdot-coated coverslips (see above) in the plating medium at a density of 6,000 cells per cm 2 . Once microislands were uniformly established on the coverslip, within a week, an antimitotic agent (8 mM 5-fluoro-2 0 -deoxyuridine (Sigma-Aldrich) and 20 mM uridine (Sigma-Aldrich)) was added to curtail glial proliferation and maintain the purity of the microisland cultures.
The microisland culture plates were established a few days before the neuronal cultures. The autaptic cultures of rat E18-19 and mouse P0-1 hippocampal neurons were generated [65][66][67] . Briefly, CA1-CA3 hippocampi were isolated and enzymatically dissociated in papain (2 units per ml, Worthington) in DMEM for 60 min at 37°C. The dissociated hippocampal neurons were plated at a density of 1,500 cells per cm À 2 onto the above astrocyte island plates. Before plating the dissociated hippocampal neurons, the astrocyte-conditioned media was replaced with Neurobasal-A medium (Invitrogen) supplemented with 2% B27 supplement (Invitrogen) and GlutaMax. The medium was not changed during the neuronal culture periods.
Electrophysiology. All synaptic recordings were undertaken on autaptic cultures after genotyping. In all cases, the synaptic responses from autaptic neurons were recorded in the whole-cell configuration under the voltage-clamp mode at a holding potential (V h) of À 70 mV (MultiClamp 700B, Molecular Devices) and room temperature. The patch pipette resistance was 3-6 MO, and the series resistance was compensated by 70-90%. The a-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor-mediated evoked glutamatergic synaptic transmission was recorded in response to an action potential elicited by a brief (2 ms) somatic depolarization pulse (to þ 0 mV) from the patch pipette. Spontaneous glutamatergic mEPSCs were recorded in the presence of 0.5 mM tetrodotoxin. The size of the RRP of synaptic vesicles was estimated using a synaptic transient charge elicited by a 10-s application of 0.5 M sucrose. The number of synaptic vesicles in the RRP was subsequently calculated by dividing the RRP charge by the individual average mEPSC charge. The vesicular release probability was calculated as the ratio of the charge of the evoked EPSC to the RRP charge successfully obtained from the same neuron. The synaptic responses were recorded at a sampling rate of 20 kHz and filtered at 10 kHz. The data were excluded from the analysis if a leak current 4300 pA was observed. The template for the mEPSC data analysis was set as a rise time constant of 0.5 ms, a decay time constant of 4 ms, a baseline of 5 ms, and a template length of 10 ms. The data were analysed offline using Axograph X 1.2 software (AxoGraph Scientific). Recordings of synaptic responses were performed at DIV 14-23 and DIV 13-16 from rat and mouse neurons, respectively.
The standard extracellular solution contained (in mM) NaCl 140, KCl 2.4, HEPES 10, glucose 10, CaCl 2 2 and MgCl 2 1, with pH 7.4 and an adjusted osmotic pressure of 315-320 mOsm. Patch pipettes were filled with an intracellular solution containing (in mM) K-gluconate 146.3, MgCl 2 0.6, ATP-Na 2 4, GTP-Na 2 0.3, phosphocreatine 12, EGTA 1 and HEPES 17.8, as well as creatine phosphokinase 50 U ml À 1 (pH 7.4). Hypertonic solutions for determining the RRP size were prepared by adding 0.5 M sucrose to the standard extracellular solution. The extracellular solutions were applied using a fast-flow application system (Warner Instruments). Each flow pipe had a diameter of 430 mm, ensuring that the solution was applied to all parts of an autaptic neuron on the astrocytic microisland (300 Â 300 mm). This configuration was necessary to facilitate the application of sucrose to induce the release of RRP vesicles from all nerve terminals of the recorded neuron. All chemicals were purchased from Sigma-Aldrich unless otherwise specified.
DiOlistic labelling. Six-to eight-week-old mice were deeply anaesthetized with 0.1 mg kg À 1 phenobarbital and perfused transcardially with 0.1 M phosphate buffer, followed by 50 ml of 2% paraformaldehyde. The brains were dissected and postfixed in the same fixative for 4 h. Subsequently, sagittal brain sections were obtained using a vibratome and collected in PBS. Twenty-five milligrams of gold particles (Bio-Rad) were coated with 7.0 mg of the lipophilic dye DiI (Invitrogen) and injected into Tefzel tubing, which was subsequently cut into bullets. These particles were delivered DiOlistically at 80 p.s.i. using a Helios gene-gun system (Bio-Rad). The slices were fixed again with 4% PFA at 4°C for 1 h and mounted on slides in VectaShield Mounting Medium (Vector Laboratories).
Statistical analysis. Statistical analyses were performed using Student's t-test, two-way analysis of variance with proper post hoc tests or the Kolmogorov-Smirnov test, as specified in each figure legend. Differences were considered significant when Po0.05.