PKN1 promotes synapse maturation by inhibiting mGluR-dependent silencing through neuronal glutamate transporter activation

Abnormal metabotropic glutamate receptor (mGluR) activity could cause brain disorders; however, its regulation has not yet been fully understood. Here, we report that protein kinase N1 (PKN1), a protein kinase expressed predominantly in neurons in the brain, normalizes group 1 mGluR function by upregulating a neuronal glutamate transporter, excitatory amino acid transporter 3 (EAAT3), and supports silent synapse activation. Knocking out PKN1a, the dominant PKN1 subtype in the brain, unmasked abnormal input-nonspecific mGluR-dependent long-term depression (mGluR-LTD) and AMPA receptor (AMPAR) silencing in the developing hippocampus. mGluR-LTD was mimicked by inhibiting glutamate transporters in wild-type mice. Knocking out PKN1a decreased hippocampal EAAT3 expression and PKN1 inhibition reduced glutamate uptake through EAAT3. Also, synaptic transmission was immature; there were more silent synapses and fewer spines with shorter postsynaptic densities in PKN1a knockout mice than in wild-type mice. Thus, PKN1 plays a critical role in regulation of synaptic maturation by upregulating EAAT3 expression. Generating mice lacking protein kinase N1 (PKN1), Yasuda et al. find that PKN1 loss leads to abnormal input-nonspecific mGluR-dependent long-term depression. The authors also observe reduced glutamate uptake and immature synaptic transmission, suggesting an important role for PKN1 in synapse maturation.

N eurons in the central nervous system are generated in the embryo, and synaptogenesis starts when pre-and postsynaptic structures meet together. Electrophysiological and anatomical studies proved that there are many silent synapses, which have NMDA receptors (NMDARs) without functional AMPA receptors (AMPARs), in the neonatal hippocampus and the barrel cortex, and the number of synapses with AMPARs increases during development [1][2][3][4][5][6] . The mechanisms of maturation of synapses remain unclarified. Since both "wakingup" of silent synapses and NMDAR-dependent long-term potentiation (LTP) involve the incorporation of AMPARs at postsynaptic sites, an LTP-like mechanism has been suggested to underlie the maturation of newborn synapses 1,3,6,7 . On the other hand, inhibition of NMDAR function using pharmacological tools or genetic manipulation in the immature brain causes a decrease 8 or an increase in expression of AMPARs 5,9 at synapses, suggesting that developmental expression of synaptic AMPARs is also regulated by an NMDAR-dependent long-term depression (NMDAR-LTD)-like mechanism, which is an activity-dependent decrease in synaptic efficacy and predominant in the immature hippocampus 10,11 .
Group 1 metabotropic glutamate receptors (mGluRs) also induce LTD 7,12,13 and have been implicated in brain disorders, including autism, anxiety disorders, and depression 14 . For example, fragile X syndrome is an autism spectrum disorder, caused by abnormality of neural development, and its mouse model shows constitutive activity of mGluR5, a group 1 mGluR 12,14,15 . LTD enhanced by an enhanced mGluR5 function may underlie mental retardation in fragile X syndrome 12,14,15 . On the other hand, mGluR5 activators also improved autistic behaviors in Shank2 knockout (KO) mice, another model for autism 16 , suggesting that precise control of mGluR function is essential for normal brain development, although regulation mechanisms for mGluR function are still poorly understood.
Protein kinase N (PKN), which we identified in 1994, is a serine/threonine protein kinase and three subtypes compose the PKN family, which is closely related to the protein kinase C family; however, PKN has a unique regulatory region and is activated by unsaturated fatty acids or Rac and Rho GTPases regulating cytoskeletal organization [17][18][19][20] . PKN1 is widely distributed in the entire body of mammals including the brain 21,22 . PKN1 is predominantly expressed in neurons 22 and accounts for 0.01% of total protein in the normal brain 17 . PKN has been implicated in cell proliferation or metastasis of many types of tumor including prostate and bladder cancers, and PKN inhibitors may treat them 20,23 . Very recently, PKN1 has been reported to be involved in axonal outgrowth and presynaptic differentiation of parallel fibers of cerebellar granule cells 24 ; however, the physiological roles of PKN in the brain have not been fully clarified yet. Therefore, we investigated the function of PKN1 in the hippocampus by generating PKN1a knockout (KO) mice ( Supplementary Fig. 1). PKN1a is a major isoform of PKN1 in the brain ( Supplementary Fig. 2). We found that PKN1 normalizes group 1 mGluR function through upregulation of neuronal glutamate transporters. PKN1 masks abnormal mGluR-dependent LTD (mGluR-LTD), prevents silencing of AMPAR synapses, and supports maturation of synapses. We propose that PKN1 is critical for normalizing mGluR activity and essential for normal brain development.

Discussion
Group 1 mGluRs mediate LTD and are implicated in many brain disorders 12,14,15,41 ; however, regulation of their action has remained unclarified. We developed PKN1a KO mice and found that deleting PKN1a unmasks group 1 mGluR-dependent input-nonspecific LTD in the developing CA1 region. PKN1a elevates glutamate uptake activity of EAAT3, a neuronal glutamate transporter. Inputnonspecific mGluR-LTD is mimicked by a glutamate transporter inhibitor, TBOA, in wild-type mice, and the effects of TBOA are occluded in KO mice. These data indicate that PKN1 upregulates neuronal glutamate transporters and normalizes group 1 mGluR function (Fig. 9c). Furthermore, LFS induced silencing of AMPAR synapses only in PKN1a KO mice, and the hippocampus in PKN1a KO mice possesses more silent synapses, which is a characteristic feature of the immature brain 1-5 . Thus, PKN1 supports maturation of synapses through inhibition of LTD-like mechanisms by upregulating neuronal glutamate transporters.
PKN1 upregulates glutamate transporters and normalizes mGluR activity. In a mouse model of fragile X syndrome, mGluR5s are constitutively active and account for phenotypes including excessive mGluR-LTD 12,15 . LFS induced abnormal mGluR5-and translation-dependent LTD in PKN1a KO mice (Fig. 3). However, DHPG-induced LTD in KO mice was comparable to that in wild-type mice ( Supplementary Fig. 7), indicating that PKN1 does not generally downregulate group 1 mGluR function. Rather, PKN1 upregulates activity of neuronal glutamate transporters, reduces glutamate concentration and prevents abnormal mGluR-LTD (Fig. 9c). EAAT3 is modestly expressed mainly at perisynaptic sites in hippocampal neurons 42 and has less capacity of glutamate clearance than glial EAAT2 (ref. 43 ). However, knocking out PKN1a reduced EAAT3 expression and PKN1 inhibition downregulated EAAT3 activity (Fig. 5e, f). mGluR-LTD was also induced in wild-type mice in the presence of 10 μM DL-TBOA. Ten micromolar DL-TBOA inhibits EAAT2 and EAAT3 but not EAAT1 because the IC 50 values of DL-TBOA for EAAT1, EAAT2, and EAAT3 are 70, 6, and 6 μM, respectively 32 . mGluR-LTD was not induced in the presence of an EAAT2-selective inhibitor, DHK (Fig. 5a, d). These data indicate that reduced EAAT3 activity is responsible for mGluR-LTD in wild-type mice in the presence of TBOA. Many synapses are not covered by glial processes in the hippocampus 44,45 , so EAAT3 is likely to uptake glutamate at synapses more effectively than EAAT2 and downregulates synaptic group 1 mGluR activity at least in the developing hippocampus where astrocytes are immature 45 . mGluR1 and mGluR5 are located at perisynaptic sites 28,46 ; therefore, presumably a higher glutamate concentration is needed to activate group 1 mGluRs than it is to activate Fig. 7 Inhibition of PKN1 and endocytosis in postsynaptic neurons unmasks homo-and heterosynaptic LTD. a Sample experiments illustrating homo-(left) and heterosynaptic LTD of EPSPs (right) in P11 wild type (upper) and P14 PKN1a KO mice (lower) in whole-cell recordings. b Summary of homo-(left; wild type, n = 16 from 14 mice; KO, n = 20 from 10 mice; t (34) = 2.05, p = 0.048; Welch's t-test) and heterosynaptic LTD of EPSPs (right; t (30) = 4.33, p = 0.00015; Welch's t-test) in P10-16 mice. No significant difference was observed between PPRs at baseline and 60 min after LFS in homo-and heterosynaptic pathways in KO mice (lower graph; n = 9 from 6 mice). c Summary of the effects of postsynaptic injection of a PKN inhibitor peptide, PRL, and a control peptide, DNER, on homo-(left; PRL, n = 17 from 11 mice; DNER, n = 15 from 11 mice; t (30) = −2.53, p = 0.017; Welch's t-test) and heterosynaptic LTD (right; t (27) = −2.85, p = 0.0083; Welch's t-test) in P11-16 wild-type mice. d Summary of the effects of postsynaptic injection of dynamin inhibitor peptide (dyn) and its scrambled peptide (cont) on homo-(left; dyn, n = 9 from 7 mice, cont, n = 12 from 7 mice; t (14) = −3.91, p = 0.0015; Welch's t-test) and heterosynaptic LTD (right; t (19) = −5.08, p = 0.000066; Welch's t-test) in P12-18 PKN1a KO mice. All data are shown as mean ± s.e.m.
NMDARs, which are located in PSDs just below presynaptic active zones, and mGluR-LTD is efficiently induced in PKN1a KO mice in which EAAT3 activity is reduced.
Previous anatomical studies suggest that CA1 pyramidal cells mainly have mGluR5 and minimally express mGluR1 (ref. 28,46 ). However, more recent electrophysiological studies showed that mGluR1 activation is involved in group 1 mGluR-dependent LTD 33,47 and regulation of LTP induction 48 in the CA1 region of the hippocampus. We presume that mGluR5 is more abundantly expressed than mGluR1, but mGluR1 is still functional at perisynaptic sites on the CA1 pyramidal neurons. Therefore, single-pulse LFS did not activate mGluR1 (Fig. 3b, c and Table 1) and a more intensive PP-LFS elevated glutamate concentration at synaptic clefts and activated perisynaptic mGluR1 as well as mGluR5. However, mGluR1 was found to be involved in homoand heterosynaptic LTD induced by PP-LFS in PKN1a KO mice s t-test) and heterosynaptic pathways (lower; wild type, 128.5 ± 24.6%, n = 8 from 7 mice; KO, 64.8 ± 11.9%, n = 15 from 14 mice; t (10) = 2.33, p = 0.042; Welch's t-test). Wild-type mice were at P8-11 and KO mice were at P8-12. d Summary of failure rate in homo-(upper) and heterosynaptic (lower) pathways in LTD experiments. The pairing protocol did not induce significant changes in failure rate between baseline, 30 min after pairing at −70 mV and at +40 mV in homo-and heterosynaptic pathways in wild-type mice. However, failure rate 30 min after pairing was significantly higher than that at baseline in homosynaptic (F (2, 28) = 38.41, p = 0.0001; one-way repeated-measures ANOVA with Holm's post hoc test) and heterosynaptic (F (2, 28) = 23.76, p = 0.00033; one-way repeated-measures ANOVA with Holm's post hoc test) pathways in PKN1a KO mice. Also, failure rate at +40 mV was significantly lower than that at baseline (homosynaptic pathways, p = 0.019; heterosynaptic pathways, p = 0.0067; one-way repeated-measures ANOVA with Holm's post hoc test) and that 30 min after pairing at −70 mV (homosynaptic pathways, p = 0.00000058; heterosynaptic pathways, p = 0.00025; one-way repeated-measures ANOVA with Holm's post hoc test). *p < 0.05; **p < 0.01, ***p < 0.001. All data are shown as mean ± s.e.m.
( Fig. 4 and Table 1). We also assume that glutamate concentration during single-pulse LFS in the presence of TBOA in wildtype mice might be higher than that during single-pulse LFS in KO mice and lower than that during PP-LFS in KO mice, because mGluR1 is involved in heterosynaptic LTD but not in homosynaptic LTD in the presence of TBOA in wild-type mice (Fig. 5b, d and Table 1). Very interestingly, D-APV plus antagonists of mGluR1 (YM298198) and mGluR5 (MPEP or MTEP) did not completely inhibit homosynaptic LTD induced by PP-LFS in KO mice (Fig. 4a, b). LTD induced by PP-LFS in the normal rat hippocampus is not inhibited by APV + 20 μM LY341495, LY367385, and MPEP, but it is suppressed by APV + 100 μM LY341495 (ref. 33 ). LY341495 inhibits group 2 and/or group 3 mGluRs with K d or K i values of less than 0.1 μM except for mGluR4, the K i value for which is 22.0 μM 49 . Twenty micromolar LY341495 may not completely inhibit mGluR4 and it may induce LTD in the presence of APV + 20 μM LY341495, LY367385, and MPEP. Therefore, mGluR4 might also be involved in LTD induced by PP-LFS in PKN1a KO mice.
So far a few reports have argued that neuronal glutamate transporter dysfunction facilitates induction of NMDAR-LTD 31 or internalization of AMPARs 44 . mGluR-dependent LTD was induced only when fewer pulses of LFS were applied in the presence of amyloid-β proteins, which inhibit neuronal glutamate transporters 31 . We are not sure why NMDAR-LTD was not enhanced in PKN1a KO mice in our hands. Weak stimulation prefers induction of mGluR-LTD 38 , and we always record smaller fEPSPs for baseline (approximately 0.2-0.4 mV, see Fig. 3). Our experimental conditions might be appropriate for mGluR-LTD rather than NMDAR-LTD.
PKN1 deletion unmasked mGluR5-dependent heterosynaptic LTD (Fig. 3). There are few reports on heterosynaptic LTD 50,51 and induction mechanisms are poorly understood. Intracellular calcium mediates homosynaptic LTD 7 , and spread of calcium is a candidate signal for induction of heterosynaptic LTD 51 . In contrast, heterosynaptic LTD in PKN1a KO mice is independent of calcium ( Supplementary Fig. 6b). mGluR-LTD in the normal immature hippocampus does not require protein translation 29 ; however, abnormal mGluR-LTD in immature PKN1 KO mice is translationdependent. Therefore, PKN1 might restrict translation and diffusion of produced proteins to inhibit spread of LTD.
PKN1 supports maturation of synapses through EAAT3 activity during development. Single-pulse LFS elicits only modest NMDAR-LTD in the normal hippocampus 52,53 . However, LFS induced NMDAR-LTD and additional mGluR5-dependent LTD that spread also to inactive synapses (Fig. 3) and generated silent synapses in the developing hippocampus lacking PKN1a (Fig. 8). Paired-pulse LFS induced mGluR1-and mGluR5-dependent homo-and heterosynaptic LTD in KO mice (Fig. 4). On the other hand, three different types of induction protocol induced almost the same amount of LTP in both wild-type and PKN1a KO mice ( Supplementary Fig. 5). Therefore, we consider that a group 1 mGluR-dependent LTD-like mechanism induces much more silent synapses in in vivo PKN1a KO mice (Fig. 9), which resulted in the attenuated input-output relationship (Fig. 1a), a lower AMPAR/NMDAR ratio (Fig. 1b), a lower mEPSC frequency (Fig. 1c), and a smaller number of spines (Fig. 2). These are all the features of young synapses [1][2][3][4][5]25 . Previously, not only NMDAR-LTD 54 but also mGluR-LTD 55,56 was suggested to induce shrinkage and retraction of spines. Furthermore, shrinkage of large spines requires group 1 mGluR activation 56 . Therefore, mature mushroom spines might be preferentially shrunk and reduced by mGluR-LTD in PKN1a KO mice (Fig. 2c). Interestingly, filopodia were also decreased; however, immature stubby and thin spines were not changed in PKN1a KO mice (Fig. 2c). LTP is intact in PKN1a KO mice ( Supplementary Fig. 5), so LTPlike mechanisms might convert filopodia into immature stubby and thin spines, but enhanced mGluR-LTD might reduce further maturation of spines. Many mGluRs are expressed at perisynaptic sites 28 . Because PKN1a KO mice have smaller spines (Supplementary Fig. 4) with shorter PSDs (Fig. 2d) than wild-type mice, mGluRs could be physically closer to presynaptic active zones. The possible closeness of group 1 mGluRs to active zones might also help induction of mGluR-LTD in PKN1a KO mice. EAAT3 is modestly expressed in hippocampal neurons and has less capacity for glutamate clearance than EAAT2 (ref. 43 ), and its physiological roles have not been well elucidated. In the cerebellum, neuronal glutamate transporters (EAAT3 and EAAT 4) distribute closely to mGluRs at postsynaptic sites and negatively regulate mGluR-LTD 57 . mGluR5 colocalizes with EAAT3 also in the CA1 region of the hippocampus 58 . Furthermore, EAAT3 expression precedes that of glial glutamate transporters in the first postnatal week, peaks in the second postnatal week, and decreases thereafter in the brain of rodents 45,59 . Considering that many synapses are not covered by glial processes in the hippocampus 44,45 , EAAT3 may be more important in synapses specifically in the developing hippocampus. A recent study suggested a role of PKN1 in development of cerebellar circuits 24 . In this study we identify PKN1 as an important promoter of synapse maturation through upregulation of neuronal glutamate transporters and inhibition of mGluRdependent synapse silencing in the hippocampus.

Methods
Animal use and all experimental procedures were approved by the Ethical Committee for Animal Experiments of Gunma University Graduate School of Medicine and Institutional Animal Care, the Animal Care and Use Committee of Saga University (#30-007-0, 30-008-0), and Use Committee of Kobe University. All experiments were performed in accordance with the guidelines of these committees.
Generation of PKN1a KO mice. A genomic fragment of the mouse PKN1 gene was isolated from a 129Sv/J phage library and probed with a mouse PKN1 cDNA. For PKN1a disruption, a replacement-type targeting vector was prepared (Supplementary Fig. 1b). It contained an~1.6 kbp HindIII-NaeI DNA fragment including the 5′ part of exon 1a, neomycin selection cassette, and an~4 kbp XbaI/ SmaI DNA fragment including exon 1b and the 5′ part of exon 2 followed by the diphtheria toxin (DT) gene for negative selection. 129 Sv/J embryonic stem (ES) cells were transfected with the linearized targeting vector by electroporation. One of the 192 G418-resistant ES cell clones carried a correctly targeted PKN1a allele, as assessed by PCR analysis and Southern blot analysis of genomic DNA.
The ES clone was injected into C57BL/6J blastocysts to generate chimeric mice. High-percentage male chimeric mice, as determined from the agouti coat color, were mated with C57BL/6J mice to determine germline transmission. F1 mice were backcrossed at least 12 times with mice with the C57BL/6J background before phenotypic analysis. PKN1a knockout mice carrying the homozygous deletion (PKN1a−/−) were viable, born at a frequency expected for Mendelian inheritance.
PKN1 expression in various tissues from wild type (PKN1a+/+) and knockout (PKN1a−/−) mice was analyzed by immunoblot analysis using the αC6 antibody against the C-terminal region of PKN1 ( Supplementary Fig. 2a). The content of PKN1 in mutant mice was less than that in wild-type mice, but PKN1 remained to some extent depending on the tissue type. Biochemical measurement of PKN1 showed that the content in the brain of PKN1a−/− mice was reduced tõ 1/10 of that of wild-type mice and the PKN1 content in the spleen was reduced tõ 60% of that of wild-type mice ( Supplementary Fig. 2b). To confirm that PKN1b remains in PKN1a−/− mice, an anti-PKN1b specific antibody, αN1b2, was prepared by immunizing rabbits with N-terminal 12 aa encoded by exon 1b conjugated with KLH ( Supplementary Fig. 2c). The sensitivity of the αN1b2 antibody was not sufficient to detect endogenous PKN1b in crude tissue lysates. However, PKN1b was detected by αN1b2 in the immunoprecipitates obtained using αN2, a variant nonselective anti-PKN1 antibody (Supplementary Fig. 2d). The αN2 immunoprecipitates from PKN1a−/− mice contained more PKN1b detected by αN1b2 than those from wild-type mice, which was evident in the spleen. This finding suggests that PKN1b remains in PKN1a−/− mouse tissues, and is particularly abundantly in the spleen, consistent with the contents of PKN1 shown in Supplementary Fig. 2b.
As shown in Supplementary Fig. 2e, the amount of PKN2 did not significantly change even in the absence of PKN1a expression in various brain regions. PKN2 seems not to compensate, at least at the protein level, for the loss of PKN1a.
Genotyping. Genomic DNA was isolated from ES cells and mouse tail snips by standard techniques and subjected to Southern blot analysis and PCR analysis for identification. Southern blot analysis was performed using genomic DNA digested with SacI/BamHI and XhoI/BamHI and probed with probe A (SacI/HindIII fragment) and probe B (SmaI/SacI fragment), respectively. Wild type (+) and mutant alleles (−) were indicated by the presence of a 2.8 kbp (+) versus 5.8 kbp (−) SacI/ BamHI for probing with probe A, and a 8.7 kbp (+) versus 11.7 kbp (−) XhoI/ BamHI DNA fragment for probing with probe B (Supplementary Fig. 1c).
Antibody. The polyclonal antibodies αN2 and αC6 were prepared by immunizing rabbits with the fragments of N-terminal 390 aa of rat PKN1a and C-terminal 84 aa of rat PKN1 synthesized in bacteria, respectively 60 . The polyclonal antibody αN1b2 was raised by immunizing rabbits with the N-terminal 12 aa of PKN1b conjugated with KLH. The αParN2 antibody was raised by immunizing rabbits with the Nterminal 506 aa of human PKN2 synthesized in bacteria. The monoclonal anit-PKN1 and polyclonal anti-EAAT3 antibodies were purchased from BD Transduction Laboratories and Cell Signaling Technology (#12179), respectively.
Immunoblot analysis. Samples were subjected to SDS-PAGE, and separated products were subsequently transferred to a polyvinylidene difluoride membrane. The membrane was then blocked with phosphate-buffered saline (PBS) (20 mM sodium phosphate, pH 7.5, 137 mM NaCl) containing 0.05% Triton X-100 (PBST) and 5% normal goat serum for 1 h at room temperature. The membrane was then incubated in PBST and the primary antibody for 1 h at room temperature. The membrane was washed three times (5 min each time) in PBST before incubating the blot in PBST containing the secondary antibody conjugated to horseradish peroxidase at 1:2000 dilution for 45 min. After this incubation, the membrane was subjected to three 10 min washes in PBST. Blots were developed by the enhanced chemiluminescence method.
Quantification of PKN1. The contents of PKN1 in various tissues were determined by immunochemical analysis using the αC6 antibody as described in ref. 61 . Various amounts (1-10 ng) of the purified GST-fused C-terminal 310 aa region of mouse PKN1 were used as standards. Mouse tissues were removed quickly after decapitation and added to 10 vol of 50 mM Tris/HCl, pH 7.5, containing 5 mM EDTA, 5 mM EGTA, 0.5 mM dithiothreitol (DTT), 10 μg/ml leupeptin, and 1 mM phenylmethyl sulfonylfluoride (PMSF). The tissues were homogenized with 10 strokes of a Teflon/glass homogenizer and the crude lysates were used for quantification of PKN1.
Immunoprecipitation experiment. The brain and spleen from wild type and PKN1a−/− mice were homogenized with a polytron in 19 vol of ice-cold buffer A containing 50 mM Tris-HCl, pH 7.5, 1 mM EDTA, 1 mM EGTA, 1 μg/ml leupeptin, 1 mM DTT, 1 mM PMSF, 0.5% Triton X-100, and 150 mM NaCl. After centrifugation at 100,000 × g for 15 min, the supernatant was transferred to new tubes and incubated with 3 μl of αN2 serum at 4°C for 2 h with rotation. Eighty microliters of 50% slurry of protein A Sepharose preequilibrated with buffer A was added to the mixture and incubated for 1 h. After centrifugation at 5000 × g for 1 min, the resulting pellet was washed three times with buffer A, and immunoprecipitated proteins were eluted with SDS sample buffer. Eluates were subjected to immunoblotting with the αPKN1 monoclonal antibody or αN1b2 antibody.
Immunohistochemistry. Wild-type mice were deeply anesthetized with sodium pentobarbital (25 mg/kg, i.p.). They were perfused transcardially with 0.1 M PBS (pH 7.4) followed by a fixative consisting of 2% paraformaldehyde and 0. Electron microscopy. PKN1a KO and wild-type mice were intravascularly perfused with a fixing solution containing 2.5% glutaraldehyde and 2.0% paraformaldehyde in 0.1 M PB (pH 7.4). After perfusion, tissue blocks of the hippocampi were extracted and immersed in the same fixative, and washed with 7.5% sucrose for 5 min. After postfixation with 1.0% OsO 4 for 2 h, the small blocks of specimens were dehydrated and embedded in an Epon-812 mixture. Ultrathin sections were cut using an ultramicrotome (LKB Ultratome® NOVA). The sections were stained with uranyl acetate and lead citrate, and then the stratum radiatum of the CA1 region approximately 50-100 μm apart from cell layer was observed under a transmission electron microscope (JEOL JEM-1010) at an accelerating voltage of 80 kV. The length of PSD was measured using Object J (https://sils.fnwi.uva.nl/bcb/ objectj/index.html), a plug-in for Image J (https://imagej.nih.gov/ij/index.html).
Golgi staining. Rapid Golgi staining was performed using the FD Rapid GolgiStain Kit (FD NeuroTechnologies) as previously described 62 . Mice were killed by decapitation under deep isoflurane anesthesia and the removed brains were washed in ice-cold PB. Whole brains were treated for silver impregnation for 2 weeks, cryoprotected for 1 week, and sectioned at 100 μm on a cryostat. After sectioning, sections were developed, clarified, then coverslipped in resinous medium. During staining, image acquisition, and analysis, observers were blind to the genotype of each animal. Using an upright microscope (Axioplan, Carl Zeiss) with a ×40 0.75 numerical aperture objective lens and a cooled CCD camera (CoolSnap fx, Photometrics), we obtained z-stack images (0.5 μm interval) of a primary apical dendritic segment (approximately 50-100 μm from the soma) of a CA1 pyramidal neuron in the hippocampus. We analyzed 30 dendritic segments from three KO mice or three wild-type littermates blind to genotype, and measured spine head width, spine length, and spine density per 1 μm dendrite using Object J. Dendritic protrusions were classified into filopodia (processes longer than 1.0 μm without a head), stubby (shorter than 1.0 μm without a head or with a head whose diameter was smaller than half of its length), thin (longer than 1.0 μm with a head whose diameter was smaller than half of the length), and mushroom (with a head whose diameter was larger than half of its length) spines 26,27 .
Slice electrophysiology. Synaptic transmission was recorded from mouse hippocampal slices as described previously 6,11,13 . Slices were prepared from postnatal 7-to 18-day-old (P7-18) PKN1a KO or their wild-type littermates in ice-cold oxygenated (95% O 2 /5% CO 2 ) artificial cerebrospinal fluid (ACSF), consisting of (in mM) 119 NaCl, 2.5 KCl, 26.2 NaHCO 3 , 1 NaH 2 PO 4 , 4 CaCl 2 , 4 MgSO 4 , and 11 glucose (pH 7.4). Slices were incubated in a submersion-type incubation chamber for at least 2 h at room temperature, and then transferred to a recording chamber mounted on an upright microscope (BX51WI, Olympus) equipped with IR-DIC optics. Slices were perfused with the oxygenated (95% O 2 /5% CO 2 ) ACSF that contained 100 μM picrotoxin at~32°C. We recorded fEPSPs using a patch pipette that had a broken tip and was filled with normal ACSF from the CA1 hippocampal region using a Multiclamp 700A (Molecular Devices). Acquisition and analysis were performed using custom Igor Pro (WaveMetrics) software routines. Two separate Schaffer collateral/commissural pathways were stimulated using two glass electrodes placed in the stratum radiatum on each side of the recording electrodes (two-pathway experiments). When we performed one-pathway experiments (e.g., for LTP and PPF), we put two hippocampal slices in a recording chamber, placed one recording electrode and one stimulating electrode in a slice, and recorded fEPSPs from two slices simultaneously.
EPSPs were recorded from CA1 pyramidal cells in the current-clamp mode with a Multiclamp 700A. Recording electrodes were filled with an internal solution containing (in mM) 135 K gluconate, 10 HEPES, 0.2 EGTA, 10 KCl, 4 Mg-ATP, and 0.5 Na 3 GTP (pH 7.2 with KOH, osmolarity adjusted to 275-285 mOsm). Basal synaptic transmissions were obtained at 0.05 Hz. We applied LFS (1 Hz for 15 min) only to one pathway, induced homosynaptic LTD of fEPSPs or EPSPs, and observed whether heterosynaptic LTD was induced in another pathway. The independence of two pathways was confirmed as follows: single stimuli to one and the other pathways were applied 100 ms apart, and we observed the lack of crossfacilitation between the two pathways.
When AMPAR/NMDAR, mEPSCs, NMDAR-EPSCs, or silent synapses were assessed, EPSCs were recorded in the voltage-clamp mode through a glass recording pipette with an internal solution containing (in mM) 135 Cs-MeSO 4 , 10 HEPES, 0.2 EGTA, 8 NaCl, 4 Mg-ATP, and 0.3 Na 3 GTP (pH 7.2 with CsOH, osmolarity adjusted to 275-285 mOsm). AMPAR-EPSCs were recorded at −70 mV and NMDAR-EPSCs were recorded at +40 mV in the presence of 10 μM CNQX. In silent synapse experiments, small AMPAR-EPSCs were evoked at −70 mV at 0.1 Hz through minimal stimulation at a reduced intensity at which some failures were surely identified visually, and then NMDAR-EPSCs were recorded using the same stimuli at +40 mV. Initially, failure rates at −70 mV (F −70 ) and +40 mV (F −70 ) were estimated by doubling that of events with an amplitude of more than and less than zero, respectively 63,64 . The percentage of silent synapses was calculated using the formula 1 -Ln (F −70 )/Ln (F +40 ) 63,64 . In these minimal stimulation experiments, LTD was induced by 1 Hz 5 min afferent stimulation paired with −50 mV 150 ms pulses. mEPSCs were recorded at 2 kHz in the presence of TTX (1 μM) and analyzed with Mini Analysis software (Synaptosoft) blind to genotype. The threshold mEPSC amplitude was set at 5 pA and each mEPSC was initially detected automatically and verified visually. NMDAR-EPSCs were recorded at −70 mV in low-Mg 2+ ACSF containing (in mM) 119 NaCl, 2.5 KCl, 26.2 NaHCO 3 , 1 NaH 2 PO 4 , 4 CaCl 2 , 0.1 MgSO 4 , 11 glucose, 0.1 picrotoxin, and 0.01 NBQX 31 . TBOA was loaded through a patch pipette, whose tip was filled with a TBOA-free internal solution with 0.2% DMSO and backfilled with an internal solution with 400 μM DL-TBOA in 0.2% DMSO. In all, 10-90% rise time and decay time constant of averaged NMDAR-EPSCs were measured using Stimfit ver. 0.13 (https://github.com/neurodroid/stimfit/wiki/Stimfit) 65 . Decays were fitted to a double exponential: I (t) = Aexp (−t/τf) + Bexp (−t/τs), where I is the amplitude of NMDAR-EPSCs, A and B are the peak amplitudes of fast and slow components, and τf and τs are the decay time constants, respectively, and the weighted time constant (τw) was calculated as τw = (A × τf + B × τs)/(A + B) 11 .
Glutamate uptake assay. SH-SY5Y cells derived from human neuroblastoma were transfected with the expression plasmid for FLAG-tagged mouse EAAT3 (pTB701/EAAT3) using NEPA21, and were seeded within a 24-well plate at approximately 40,000 cells/well. After 24 h the cells were transfected with the expression plasmids for FLAG-tagged full-length PKN1a (pRc/CMV/PKN1a-FL), the kinase-negative mutant (pRc/CMV/PKN1a-T774A-FL), or mock using FuGENE HD transfection reagent (Roche). [ 3 H]-Glu uptake assay was carried out 72 h after the EAAT3 transfection 66 . In brief, the culture medium was removed and replaced with Krebs-Ringer-HEPES (KRH) buffer containing (in mM) 120 NaCl, 4.7 KCl, 2.2 CaCl 2 , 25 HEPES, 1.2 MgSO 4 , 1.2 KH 2 PO 4 , and 10 glucose (pH 7.4). After a 15-min pre-incubation at 37°C, the cells were incubated for an additional 15 min in the presence of 100 nM [ 3 H]-Glu. The uptake of [ 3 H]-Glu was stopped by three washes with cold KRH buffer containing 1 mM DL-TBOA, and the cells were then treated with RIPA buffer containing (in mM) 25 HEPES, 0.5% Triton X-100, 100 NaCl, and 2 EDTA (pH 8.0). The radioactivity of cell extracts diluted in scintillation cocktail (Clear-Sol II, Nakalai Tesque) was measured with a liquid scintillation counter (Beckman). The obtained results were used as the total uptake of the cells. The [ 3 H]-Glu uptake level in the presence of 1 mM DL-TBOA was also measured and used as the nonspecific glutamate uptake level of the cells. The specific glutamate uptake level was obtained by subtracting the nonspecific glutamate uptake level from the total uptake level. To calculate the uptake level per mg protein of the cells, after counting the [ 3 H]-Glu uptake, the protein concentration was measured from the rest of the cell extracts using the BCA Protein Assay Reagent (Thermo Scientific).
Statistics and reproducibility. Results are reported as mean ± s.e.m. For multiple statistical comparisons, one-way ANOVA with Tukey's post hoc test in EZR (easy R) 67 was used, except when Holm's test was adopted for multiple comparisons with repeated measures (Fig. 8d). Comparisons between two groups were performed using unpaired, two-tailed t-test with Welch's correction (Welch's t-test) or paired t-test in Excel. Two groups that do not follow a normal distribution were compared using Mann-Whitney's U test in Excel (Supplementary Fig. 4). We did not statistically calculate sample sizes; however, the numbers of data we got in experiments presented here were similar to those in papers previously published in our field. We usually used at least three mice for one group in each experiment.