Transmembrane protein 108 is required for ...

Transmembrane protein 108 is required for glutamatergic transmission in dentate gyrus Hui-Feng Jiaoa,b,c,1, Xiang-Dong Sunc,1, Ryan Batesc,1, Lei Xiongc, Lei Zhangc, Fang Liuc, Lei Lic, Hong-Sheng Zhangc, Shun-Qi Wanga, Ming-Tao Xionga,b, Mihir Patelc, Alexis M. Stranahanc, Wen-Cheng Xiongc,d, Bao-Ming Lia,e,2, and Lin Meia,c,d,e,2 a

Institute of Life Science, Nanchang University, Nanchang 330031, China; bSchool of Life Sciences, Nanchang University, Nanchang 330031, China; Department of Neuroscience and Regenerative Medicine, Medical College of Georgia, Augusta University, Augusta, GA30912; dCharlie Norwood Veterans Administration Medical Center, Augusta University, Augusta, GA30912; and eJiangxi Medical School, Nanchang University, Nanchang 330031, China c

Edited by Solomon H. Snyder, The Johns Hopkins University School of Medicine, Baltimore, MD, and approved December 16, 2016 (received for review November 7, 2016)

dentate gyrus schizophrenia

| spine | glutamatergic transmission | AMPA receptors |

S

chizophrenia is a disabling psychiatric disorder that affects 1% of the general population. It is thought to be a neurodevelopment disorder, as many symptoms appear or worsen during adolescence, a time of great transition and refinements in brain structure and function (1, 2). Consequently, patients display characteristic positive symptoms including delusions and hallucinations, negative symptoms including abnormal emotional reactivity and anhedonia and cognitive deficits. Underlying pathophysiological mechanisms have been explored extensively. The medial temporal lobe, including hippocampal dentate gyrus (DG), is thought to be involved in mediating aspects of psychosis and memory deficits in schizophrenia (3, 4). Impaired glutamatergic transmission in DG causes deficits in spatial coding, learning, and memory and emotion processing (5–7). However, detailed molecular mechanisms of DG dysfunction in schizophrenia remain unclear. Identification of risk genes in recent genetic studies has contributed to a better understanding of pathophysiological mechanisms of schizophrenia. Transmembrane protein 108 (TMEM108) has recently been linked with schizophrenia and alcoholism in genome-wide association studies (8, 9). In human, TMEM108 is located on chromosome 3q21-q22, a risk locus for bipolar disorder, schizophrenia and other psychosis (10, 11). In particular, an intronic single nucleotide polymorphism (SNP) (rs7624858) is associated with schizophrenia (8). These findings raise an important question regarding the physiological function of TMEM108 and whether abnormal expression levels of TMEM108 impair neural development or function.

www.pnas.org/cgi/doi/10.1073/pnas.1618213114

Tmem108 is a transmembrane protein, initially identified as a protein (retrolinkin) that interacts with a neuronal isoform of bullous pemphigoid antigen 1 (BPAG1n4) and promotes retrograde axonal transport in dorsal root ganglia neurons (12). Tmem108 is also present in dendrites of hippocampal neurons and has been implicated in BDNF-induced TrkB endocytosis and dendrite outgrowth in cultured neurons (13, 14). However, genetic evidence is lacking regarding the in vivo function of Tmem108 and whether its mutation impairs neural development and causes schizophrenia-relevant behavioral deficits. Here we show that Tmem108 was highly enriched in DG granule neurons and that its expression is regulated by neural development. Knocking down Tmem108 impaired spine development in cultured DG granule cells; in agreement, Tmem108 mutant (MT) mice displayed fewer and smaller spines. Both the frequency and amplitude of excitatory postsynaptic currents (EPSCs) of DG granule neurons were reduced in Tmem108 MT mice. Further molecular studies suggest that Tmem108 is required for maintaining synaptic AMPA receptors on DG granule neurons. Consequently, deletion of Tmem108 impaired spatial recognition memory, contextual fear memory, as well as sensorimotor function. Together, these observations indicate that Tmem108 is necessary for proper development of DG neuron circuitry and its deletion leads to hypofunction of the glutamatergic activity in the brain and behavioral deficits. Considering that Tmem108 is a susceptibility gene of schizophrenia, our study sheds light on potential pathophysiological mechanisms of this disorder. Significance Dentate gyrus (DG) dysfunction has been implicated in schizophrenia, a disabling psychiatric disorder. However, underlying pathophysiological mechanisms are not clear. We provide evidence that Tmem108, a novel schizophrenia-associated gene, is highly enriched in DG granule neurons. Tmem108 is required for spine development and glutamatergic transmission. Further investigations indicate a critical role of Tmem108 for AMPA receptor expression in postsynaptic compartments. Mutation of Tmem108 leads to schizophrenia-related behavioral deficits. These results provide insight into a potential pathophysiological mechanism for DG dysfunction in schizophrenia. Author contributions: H.-F.J., X.-D.S., R.B., W.-C.X., B.-M.L., and L.M. designed research; H.-F.J., X.-D.S., R.B., L.Z., and M.-T.X. performed research; L.X., L.Z., F.L., L.L., H.-S.Z., S.-Q.W., and A.M.S. contributed new reagents/analytic tools; H.-F.J., X.-D.S., R.B., M.-T.X., and M.P. analyzed data; and H.-F.J., X.-D.S., and L.M. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. 1

H.-F.J., X.-D.S., and R.B. contributed equally to this work.

2

To whom correspondence may be addressed. Email: [email protected] or bmli@ncu. edu.cn.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1618213114/-/DCSupplemental.

PNAS | January 31, 2017 | vol. 114 | no. 5 | 1177–1182

NEUROSCIENCE

Neurotransmission in dentate gyrus (DG) is critical for spatial coding, learning memory, and emotion processing. Although DG dysfunction is implicated in psychiatric disorders, including schizophrenia, underlying pathological mechanisms remain unclear. Here we report that transmembrane protein 108 (Tmem108), a novel schizophrenia susceptibility gene, is highly enriched in DG granule neurons and its expression increased at the postnatal period critical for DG development. Tmem108 is specifically expressed in the nervous system and enriched in the postsynaptic density fraction. Tmem108-deficient neurons form fewer and smaller spines, suggesting that Tmem108 is required for spine formation and maturation. In agreement, excitatory postsynaptic currents of DG granule neurons were decreased in Tmem108 mutant mice, indicating a hypofunction of glutamatergic activity. Further cell biological studies indicate that Tmem108 is necessary for surface expression of AMPA receptors. Tmem108-deficient mice display compromised sensorimotor gating and cognitive function. Together, these observations indicate that Tmem108 plays a critical role in regulating spine development and excitatory transmission in DG granule neurons. When Tmem108 is mutated, mice displayed excitatory/inhibitory imbalance and behavioral deficits relevant to schizophrenia, revealing potential pathophysiological mechanisms of schizophrenia.

Results Enriched Expression of Tmem108 in the DG. Tmem108 is expressed

in the nervous system and barely detectable in peripheral tissues (12) (Fig. S1A). In the brain, Tmem108 was enriched in DG of the hippocampus, compared with other brain regions (Fig. 1A). In agreement, quantitative real-time PCR (qRT-PCR) indicated that Tmem108 mRNA was highly expressed in the DG, relative to other hippocampal subfields and extrahippocampal regions (Fig. 1B). To further study the regional expression, we generated Tmem108 mutant reporter mice because the available antibodies function poorly for immunohistochemical staining (15, 16). In this strain, the lacZ-containing cassette was inserted in the first coding exon (exon 3) (Fig. S1B). Under the control of the endogenous promoter, betagalactosidase (β-gal) activity was expected to faithfully indicate where Tmem108 is expressed. To avoid possible effect of Tmem108 mutation on brain structures, β-gal assay was performed using samples from heterozygous mice. As shown in Fig. 1C, β-gal activity was highly enriched in both suprapyramidal and infrapyramidal blades of the DG, areas where granule neurons locate. Little β-gal activity was detected in layers of CA1 or CA3 regions where pyramidal neurons are enriched, suggesting that Tmem108 was rather specifically expressed in DG granule neurons. β-Gal activity was detectable at a much lower level in the cortical region, mostly in layers 2/3. Together, these results suggest that Tmem108 is highly enriched in the DG of the hippocampus. Tmem108 expression in the hippocampus was developmentally regulated. As shown in Fig. 1 D and E, β-gal activity as well as

Fig. 1. Enriched expression of Tmem108 in the DG. (A) Tmem108 was highly expressed in DG regions in the brain. Tissues of indicated brain regions were collected from 2-mo-old WT mice and homogenized for Western blotting. β-Actin served as loading control. OB, olfactory bulb; Ctx, cortex; AH, ammon’s horn; DG, dentate gyrus; Hypo, hypothalamus; Str, striatum; and CB, cerebellum. (B) DG-enriched expression of Tmem108 mRNA in the brain. Total RNA of indicated brain regions was subjected to qRT-PCR. (C) X-Gal staining of coronal (Left) and sagittal (Right) brain sections of Tmem108 heterozygous mice. Arrow, DG. (Scale bar: 1 mm.) (D) Temporal regulation of β-gal activity in DG. Coronal sections of Tmem108 heterozygous mice at indicated ages were subjected to X-gal staining. (E) Tmem108 expression in the hippocampus at different stages. β-Actin served as loading control. (F) Colocalization of β-gal with DG granule neuron marker Prox1. Sections were stained with antibodies against β-gal, PSA-NCAM, and Prox1. Images in the dotted areas were enlarged and shown on the Right. (Scale bars: Left, 50 μm; Right, 30 μm.) (G) Tmem108 was enriched in the postsynaptic dense (PSD) fraction. Subcellular fractions of hippocampal tissues were probed for postsynaptic marker PSD95, presynaptic marker synaptotagmin, and Tmem108. SYT, synaptotagmin; S1, supernatant 1; S2, supernatant 2; P2, synaptosome-enriched pellet 2; Syn, synaptosome; Pre, presynaptic fraction; PSD, postsynaptic density fraction.

1178 | www.pnas.org/cgi/doi/10.1073/pnas.1618213114

Tmem108 protein was undetectable in the DG at postnatal day 1 (P1) and became detectable at P7, although at a low level. The levels seemed to peak between P15 and P21 and remained at a high level at adult age. The enrichment of β-gal activity in suprapyramidal and infrapyramidal blades suggests that Tmem108 is expressed in granule cells. To test this hypothesis, we costained β-gal with different cell markers. As shown in Fig. 1F, β-gal colocalized with prospero homeobox protein 1 (Prox1), a marker of granule neurons (17), but not with polysialylated neuronal cell adhesion molecule (PSA-NCAM), a marker of neuronal precursors (18). Finally, we determined subcellular localization of Tmem108 and found that Tmem108 is enriched in the postsynaptic dense (PSD) fraction, but not presynaptic fraction (Fig. 1G). These results demonstrate that Tmem108 is specifically expressed in DG neurons and enriched in the PSD and that its expression temporally correlates with a period critical for spine development. Spine Abnormality of Tmem108-Deficient DG Granule Neurons. To determine whether Tmem108 regulates spine formation, we investigated the effects of changing Tmem108 levels. Neurons were isolated from P0 pups, transfected at DIV13, and fixed at DIV20 for spine analysis. Neurons were stained with anti-Prox1 antibody, which helped to identify DG granule neurons (17) (Fig. S2A). As shown in Fig. 2 A and B, spine number was increased in DG neurons that were transfected with Flag-Tmem108, suggesting that higher levels of Tmem108 promotes spine formation. On the other hand, neurons transfected with Tmem108 shRNA, which was able to reduce Tmem108 expression (Fig. S2B), formed fewer spines, which was associated with reduced spine width and increased spine length (Fig. 2 A and B). These effects were specific as they were not observed in neurons transfected with scrambled shRNA, and it could be diminished by cotransfecting a shRNA-resistant Tmem108 (Fig. S2B and Fig. 2A). These results indicate an important role of Tmem108 in spine development. To determine whether Tmem108 deficiency alters spine development in vivo, we characterized Tmem108 MT mice. As shown in Fig. S1 B and C, the insertion of the lacZ-Neo cassette introduces a stop codon with a polyadenylation termination signal (15, 16), which would terminate or severely reduce transcription of the Tmem108 gene. In agreement, mRNA and protein of Tmem108 were dramatically reduced in homozygous mice (Fig. S1 D and E). Homozygous Tmem108-lacZ mice were viable and showed no difference in body weight, compared with wild-type littermates. Unless otherwise specified, WT and MT indicate, respectively, wild-type and homozygous mutant littermates in the study (Fig. S1F). Tmem108 mutation seemed to have no detectable effect on global anatomic structures of the brain (Fig. S3A). The hippocampal organization and number of granule neurons in DG were comparable between WT and MT mice (Fig. S3 B and C). Previous studies suggest that Tmem108 regulates dendritic outgrowth of hippocampal neurons (13, 14). As shown in Fig. S4, dendrite length and complexity of DG granule cells, revealed by Golgi staining, were similar between WT and MT mice (Fig. S4). Next, we quantified dendritic spines of DG granule neurons (Fig. 2C). The spine density in the molecular layer of MT mice was significantly reduced, which was associated with decreased spine width and increased spine length (Fig. 2D), consistent with the results of in vitro knockdown experiments (Fig. 2 A and B). To determine that the spine abnormality was due to loss of Tmem108, we used in vivo electroporation to specifically express Tmem108 in DG neurons in MT mice (Fig. 2E). As shown in Fig. 2 F–H, Tmem108 reintroduction was able to diminish spine morphological deficits observed in Tmem108 MT mice. Abnormal Excitatory Transmission in Tmem108 MT DG Granule Neurons.

Next, we examined synaptic transmission of DG granule neurons by characterizing spontaneous excitatory postsynaptic and inhibitory Jiao et al.

(Fig. 3 A–C), these results indicate compromised glutamatergic transmission. To investigate whether glutamate release was impaired, we examined paired-pulse ratios of evoked EPSCs (Fig. 3K). As shown in Fig. 3L, paired-pulse ratios in DG granule neurons were similar between WT and MT mice, indicating that Tmem108 deficiency has little effect on glutamate release, suggesting a postsynaptic deficit in MT mice. Together with morphological findings (Fig. 2), these electrophysiological results suggest that Tmem108 is necessary for synapse development and that glutamatergic hypofunction is due to a postsynaptic mechanism. Decreased AMPA Receptor Surface Level of Tmem108-Deficient DG Granule Neurons. Reduced spine number and EPSC could suggest

reduced AMPA receptor levels in DG neurons. Unexpectedly, similar levels of AMPA receptors (GluA1 and GluA2) and NMDA receptors (GluN1, GluN2A, and GluN2B) were detected between control WT and Tmem108 MT tissue homogenates, suggesting Tmem108 mutation did not alter total amounts of these proteins in the hippocampus. Considering that Tmem108 is enriched in the PSD (Fig. 1G), we next tested whether AMPA receptors in the PSD were reduced. In fractions that are labeled by PSD95, but not synaptotagmin (Fig. 1G), GluA1 and GluA2 were reduced by 13%

postsynaptic currents (sEPSCs and sIPSCs, respectively). sEPSCs and sIPSCs were recorded in same cells by alternately clamping at reversal potentials of GABAA receptor-mediated (−70 mV) and glutamate receptor-mediated (0 mV) currents, respectively (19, 20) (Fig. 3A). The total excitatory charge transfer was decreased in MT mice compared with WT mice (Fig. 3B). However, the inhibitory synaptic charge was comparable (Fig. 3C). Consequently, the sEPSC/ sIPSC ratio was decreased by almost 50% (Fig. 3D). To determine whether Tmem108 mutation altered evoked postsynaptic currents, we stimulated medial perforant pathway (MPP) with stimuli at gradually increasing intensity. As shown in Fig. 3 E–G, amplitudes of eEPSCs, but not eIPSCs, were reduced. These results suggest that the excitatory/inhibitory (E/I) balance was disrupted by Tmem108 mutation, mostly due to impaired excitatory synaptic activity. To further dissect how excitatory strength was suppressed in Tmem108 MT mice, we recorded miniature EPSCs (mEPSCs) in DG granule neurons (Fig. 3H). Both the frequency and amplitude were significantly decreased in MT hippocampal slices (Fig. 3 I and J). However, no difference was found in mIPSC frequency or amplitude (Fig. S5). Together with sEPSC and sIPSC data Jiao et al.

Fig. 3. Tmem108 is required for excitatory synapse transmission of DG granule neurons. (A) Representative sEPSC (Top) and sIPSC (Bottom) traces. sEPSCs and sIPSCs were recorded in the same granule neuron at −70 mV and 0 mV, respectively. (Scale bars: 2 s and 10 pA.) (B) Redued sEPSC charge transfer. CsEPSC, charge transfer of sEPSC. pC, picocoulomb. (C) Similar sIPSC charge transfer between WT and MT mice. CsIPSC, charge transfer of sIPSC. (D) Decreased charge transfer ratio of sEPSC/sIPSC. (E) Representative eIPSC (Top) and eEPSC (Bottom) traces. Medial perforant pathway was stimulated at gradual increasing intensity (5–40 μA). eEPSC and eIPSC were recorded in the same granule neuron at −70 mV and 0 mV, respectively. (Scale bars: 20 ms and 200 pA.) (F and G) Quantatitive analysis of eEPSC and eIPSC. amp., amplitude; Sti., stimulus. (H) Representative mEPSC traces in DG granular neurons. (Scale bars: 2 s and 10 pA.) (I and J) Reduced mEPSC frequency (freq.) and mEPSC amplitude (amp.). (K) Representative sweeps with interstimulus interval of pair-pulse stimulations at 25 ms. (Scale bars: 10 ms and 20 pA.) (L) Similar paired-pulse ratio of the two genotypes. Three to five mice were used for each genotype. n = 19 neurons for WT or 18 neurons for MT in B–D; n = 20, 25, and 25 neurons for both genotypes in F, I, and J, respectively; n = 13 neurons for WT, or 15 neurons for MT in L. *P < 0.05; **P < 0.01; ***P < 0.0001; Student’s t test for B–D, I, J, and L; two-way ANOVA, F(1,271) = 36.61 and F(1,263) = 1.84 for F and G, respectively.

PNAS | January 31, 2017 | vol. 114 | no. 5 | 1179

NEUROSCIENCE

Fig. 2. Abnormal spine development of Tmem108-deficient DG granule neurons. (A) Representative images of dendritic spines of cultured DG neurons. Neurons were isolated from P0 pups, transfected at DIV13 with indicated constructs, and fixed and stained at DIV20. Veh, GFP vector; Tmem108, FlagTmem108; shRNA, Tmem108 short hairpin RNA; Scr shRNA, Scrambled short hairpin RNA; rTmem108, shRNA-resistant Flag-Tmem108. (Scale bar: 5 μm.) (B) Quantitative analysis of data in A. Spine number/10 μm (Left), width (micrometers, Middle), and length (micrometers, Right) were analyzed. (C) Representative spine images from Golgi staining. Dendrite segments were chosen from mature granule neurons, which located in the superficial granule cell layer. (Scale bar: 5 μm.) (D) Reduced spine number/10 μm (Left), width (micrometers, Middle), and length (micrometers, Right) of DG granule neurons in Tmem108 MT mice. (E) Diagram illustrating in vivo electroporation. Mouse pups at P0 were injected with indicated constructs into bilateral ventricles and electroporated. Twenty-eight days later, mice were subjected to spine analysis. (F) Representative images of electroporated granule neurons. Sections were stained with anti-GFP antibody. Image in the dotted areas was enlarged and are shown on the Right. (Scale bars: Left, 200 μm; Right, 50 μm.) (G) Representative spine images of granule neurons electroporated with indicated constructs. Veh, GFP vector; Tmem108, Flag-Tmem108. (Scale bar: 5 μm.) (H) Quantitative analysis of data in G. Spine number/10 μm (Left), width (micrometers, Middle), and length (micrometers, Right) were analyzed. Data were collected from three to four dendrite segments of each neuron; n = 15 and 20 neurons in B and H, respectively; n = 20 neurons for WT or 19 for MT in D. n.s., not significant; *P < 0.05; **P < 0.01; Student’s t test.

and 25%, respectively (Fig. 4A). Similar reductions were obtained by concentration-dependent Western blot analysis (Fig. S6). The reduction of AMAP receptors was specific because levels of NMDA receptors were similar between WT and MT PSD fractions (Fig. 4 A and B). These results suggest that Tmem108 is necessary for proper expression of AMPA receptors at excitatory synapses. To test this hypothesis, we characterized GluA2 surface expression in cultured granule neurons. First, we determined whether Tmem108 was present in excitatory synapses of hippocampal granule cells. Due to lack of anti-Tmem108 antibody for staining, we cotransfected hippocampal neurons with Flag-Tmem108 and GFPGluA2. As shown in Fig. 4C, Flag-Tmem108 staining appeared as puncta in neurons. Tmem108 puncta colabeled with GFP-GluA2 in spines (arrow, Fig. 4C) as well as dendrites (triangle, Fig. 4C). These results indicate that Tmem108 is present at excitatory synapses, in agreement with subcellular fraction data. Next, we stained neurons for endogenous GluA2 under permeabilizing and nonpermeabilizing conditions to assess total and surface AMPA receptors, respectively (21) (Fig. 4 D and E). Granule neurons were identified by Prox1 antibody. GluA2 staining was similar between permeabilized WT and MT granule neurons (Fig. 4D), indicating little change in total GluA2 level, in agreement with Western blot data (Fig. 4A). However, GluA2 staining was reduced in nonpermeabilized MT granule neurons, compared with that of WT (Fig. 4E). Quantitatively, reduction was

observed in the number of GluA2 puncta, the puncta area, and soma GluA2 intensity (Fig. 4F), suggesting that Tmem108 may regulate GluA2 trafficking. To test this hypothesis in the same neurons, we transfected GFP-GluA2 in granule cells. Surface GluA2 in live neurons was first labeled with chicken anti-GFP antibody (visualized by donkey anti-chicken antibody, red). Neurons were then fixed and stained with mouse anti-GFP antibody (visualized by goat anti-mouse antibody, green). As shown in Fig. S7 A and B, the GluA2 surface/total ratio was reduced in Tmem108 MT granule neurons, compared with WT neurons. These observations are in agreement with reduced eEPSC and mEPSC amplitudes in MT granule neurons. Together, these results suggest that Tmem108 promotes GluA2 surface expression, without changing total levels, and thus maintains spine morphology. This notion is supported by the observations that spine morphological deficits in Tmem108 MT DG neurons could be rescued by overexpressing GluA2 (Fig. 4 G and H). Notice that the effect of Tmem108 mutation was specific to Prox1-positive neurons (i.e., granule cells) and not to Prox1-negative neurons (presumably hippocampal pyramidal neurons) (Fig. S7 C–E). Behavioral Deficits in Tmem108 MT Mice. Abnormal locomotor activity is thought to correspond to psychomotor agitation of schizophrenic patients (22, 23). We examined MT mice in the open field test (Fig. 5A). Tmem108 MT mice traveled similar distances,

Fig. 4. Decreased surface AMPA receptor of Tmem108 MT DG granule cells. (A) Reduced GluA1 and GluA2 in PSD fractions of MT mice. Subcellular fractions of hippocampal tissues were subjected to Western blot for different glutamatergic receptors. S1, supernatant 1; PSD, postsynaptic density fraction. (B) Quantitative analysis of data in A. Band densities of interested proteins were normalized by loading control β-actin; values of WT mice were taken as 1. n = 3. (C) Colocalization of Tmem108 with GluA2 in dendrites and spines. DIV9 hippocampal neurons were cotransfected with Flag-Tmem108 and GFP-GluA2 and stained with anti-Flag and anti-GFP antibodies at DIV13. Image in dotted area was enlarged as shown in the Bottom. Arrow, double-positive puncta in spines; triangle, double-positive puncta in dendrites. (Scale bars: 30 μm and 5 μm.) (D and E) Reduced surface GluA2 levels (E), but no change in total GluA2 levels (D) in MT granule neurons. DIV15–17 hippocampal neurons were stained under permeabilized and nonpermeabilized conditions to assess total and surface GluA2 levels, respectively. Anti-Prox1 antibody was used to idenitify granule neurons. Image in dotted area was enlarged as shown in the Bottom. Side bar, glow scale of GluA2 staining intensity in arbitrary unit. (Scale bars: Top, 30 μm; Bottom, 5 μm.) (F) Quantitative analysis of data in D and E. Shown are GluA2 puncta number/10 μm (Top Left), GluA2 puncta area (square micrometers, Top Right), and GluA2 soma intensity (normalized to WT, Bottom). (G) Representative images of DG granule neurons transfected with indicated constructs. Veh, mCherry vector; Tmem108, Flag-Tmem108; GluA2, GFP-GluA2. (Scale bar: 5 μm.) (H) Quantitative analysis of data in G. Spine number/10 μm (Left), width (micrometers, Middle), and length (micrometers, Right) were analyzed. Data were collected from four dendrite segments of each neuron. n = 17 neurons for both genotypes in F; n = 15 neurons for each group in H. n.s., not significant; *P < 0.05; **P < 0.01; ***P < 0.001; paired Student’s t test for B; Student’s t test for F and H.


Jiao et al.

Fig. 5. Impaired behaviors of Tmem108 MT mice. (A) Representative traces of first 5 min in the open field test. Mice were placed in a chamber and movements were monitored for 30 min. (B and C) Similar distance traveled during 30 min between WT and MT mice. Activity was summated at 5-min intervals over a 30-min period (B). (D) Diagram of PPI test. Response to auditory-evoked startle stimulus (120 dB) was measured. (E) Similar baseline startle responses of the two genotypes. (F) Reduced PPI in MT mice. (G) Diagram of Y-maze test. Mice were put in Y-shape maze for 8 min, and total arm entry number and spontaneous alternation were recorded. (H) Similar total arm entries of the two genotypes. (I) Fewer spontaneous alternations in MT mice. (J) Diagram of contextual fear conditioning. Footshocks were delivered four times (FS, 0.7 mA, 2 s) during training. Twenty-four hours later, mice were reintroduced to the same box and freezing time was recorded for 5 min. (K) Similar fear acquisition, but reduced freezing time 24 h after training. BS, baseline; FS, footshock. n = 12–14 mice of both genotypes for each behavior test. *P < 0.05; *P < 0.05; Student’s t test for C, E, H, I, and K (consolidation); two-way ANOVA, F(1,75) = 5.45 for F; repeated two-way ANOVA, F(1,100) = 0.22 and F(1,100) = 3.52 for B and K (acquisition), respectively.

Jiao et al.

test associative memory formation and consolidation. MT mice displayed similar freezing response to footshocks during training (Fig. 5K), compared with WT mice, suggesting comparable ability in fear acquisition. However, the freezing time of MT mice in the absence of footshocks when reintroduced to the same cage 24 h later was significantly less than that of WT mice. This result suggests that deletion of Tmem108 suppressed fear memory consolidation (Fig. 5K). These results indicate that Tmem108 is required for proper cognitive function. Together, these observations indicate that Tmem108 mutation specifically impairs PPI and cognitive function without altering locomotor activity. Discussion In this paper, we provide evidence that Tmem108 was enriched in DG granule neurons and its expression increased at postnatal days, a period critical for neural development. Tmem108 knockdown in cultured neurons and mutation in mice reduced spine number of DG granule neurons, and this effect could be rescued by reintroduction of Tmem108. Concomitantly, mEPSC frequency and amplitude as well as evoked EPSCs in DG granule neurons were reduced. These results indicate hypofunction of the glutamatergic entorhino–hippocampal pathway when Tmem108 is deficient. Cell biological studies indicate that Tmem108 is necessary for surface expression of AMPA receptors. Behaviorally, MT mice exhibit impaired sensorimotor gating and cognitive function. Together, these observations indicate that Tmem108 plays a critical role in regulating spine development and excitatory transmission in DG granule neurons. When Tmem108 is mutated, mice displayed E/I imbalance and behavioral deficits relevant to schizophrenia, revealing potential pathophysiological mechanisms of schizophrenia. DG is a critical region for higher brain functions, including spatial coding, learning memory, and emotion processing (5, 6). Hypofunctional glutamatergic signaling in the DG has been observed in patients with schizophrenia (3, 4). However, underlying pathophysiological mechanisms are less clear. During the first 2 wk after birth in mice, precursor granule cells migrate from the hilus to the granule cell layer of the DG, where they form synapses with other neurons (28, 29). Tmem108 expression in DG begins to increase at P7 and plateaus between P21 and P30. This unique temporal expression correlates with active synaptic pruning in the hippocampus (30, 31), suggesting a role of Tmem108 after the migration of granule precursor cells. In support of this hypothesis, in Tmem108 MT mice, the number of NeuN+ cells in the DG and dendritic arborization of granule cells were not changed. In contrast, dendritic spine density and size of DG granule neurons were reduced, and spine length was increased in Tmem108 MT mice, indicating that Tmem108 is necessary for spine formation and maturation. Ensuing hypofunction of the glutamatergic transmission leads to behavioral deficits associated with schizophrenia. AMPA receptors within the postsynaptic domain are critical for maintaining and strengthening spine structure and function (32– 34). Knockout of GluA2, a subunit of AMPA receptor, causes spine deficits in DG granule cells (33). Tmem108 deficiency reduced AMPA receptor surface level of DG neurons without changing the total level. There was a concomitant reduction of AMPA receptors in the PSD fraction. This effect is specific because Tmem108 mutation had no effect on total or surface levels of NMDA receptors of DG neurons. Importantly, spine deficits in Tmem108 MT DG neurons could be rescued by overexpressing GluA2. A parsimonious interpretation of these results is that Tmem108 promotes surface expression of AMPA receptors that is necessary for spine development. AMPA receptor dynamics in spines are regulated by proteins that control the cytoskeleton. For example, the stabilization of postsynaptic AMPA receptors as well as spine morphology are regulated by small G proteins of the Rho family (35, 36). Rac and Cdc42 regulate spine stabilization by activating the Arp2/3 complex to promote actin nucleation and inhibit PNAS | January 31, 2017 | vol. 114 | no. 5 | 1181

NEUROSCIENCE

compared with control WT mice within 30 min of test (Fig. 5 B and C), indicating no change in locomotor activity. Prepulse inhibition (PPI) is a test of sensory-motor gating that is often decreased in schizophrenic patients (24, 25). We used a combination of an auditory-evoked startle stimulus (120 dB) and three levels of prepulse stimuli (70, 75, and 80 dB) to measure PPI of MT mice (Fig. 5D). The baseline startle responses of WT and MT mice were similar (Fig. 5E), suggesting normal hearing and acoustic startle reflex. However, the level of PPI was substantially lower in MT mice than in WT mice (Fig. 5F). These results implicate that the Tmem108 MT mice were impaired in sensorimotor gating. Patients with schizophrenia have a wide range of cognitive function deficits, including impairment in learning and memory, executive function, and intelligence (26, 27). We tested spatial recognition memory of MT mice by using the Y maze (Fig. 5G). Tmem108 MT mice exhibited comparable number of arm entries (Fig. 5H), in agreement with no change in locomotor activity. However, the numbers of spontaneous alterations were significantly decreased in MT mice, compared with WT mice (Fig. 5I). These results suggest that Tmem108 deletion impaired spatial recognition memory. To further characterize the effects of Tmem108 mutation on cognitive function, MT mice were subjected to contextual fear conditioning (Fig. 5J), a classical behavioral paradigm to

of spines in DG granule neurons in culture, whereas reducing its level diminishes the spines in cultured neurons as well as in Tmem108 MT mice. These observations suggest that a proper level of Tmem108 needs to be maintained for homeostasis of spines. Altered level, either high or low, could serve as a pathophysiological mechanism (43).

actin depolymerization (37, 38). A recent study identified Tmem108 as a binding partner of cytoplasmic FMRP-interacting protein 1/2 (CYFIP1/2) to regulate Arp2/3 by promoting the formation of the wave regulator complex (14). The CYFIP1 gene is located in the 15q11.2 region of the human genome, which is implicated in the development of neurological and neuropsychiatric conditions such as autism spectrum disorder, epilepsy, intellectual disability, and schizophrenia (39–41). Its copy number variation is linked to both schizophrenia and autism spectrum disorder (40, 42). Down-regulating CYFIP1 levels increases the ratio of immature-to-mature spines and the mobility of surface AMPA receptors (41). Taken together, these observations could suggest that Tmem108 may regulate spines of DG granule neurons via interacting with CYFIP1/2. Exact mechanisms by which Tmem108 regulates spines and synaptic expression of AMPA receptors warrant further investigation. The SNP (rs7624858) that associates with schizophrenia is located between the first coding exon (exon 3) and exon 4. Being intronic, this SNP may interfere with the expression of the Tmem108 gene, although there are no data at the present that this SPN predicts a higher or lower level of mRNA or protein. We found that elevating the level of Tmem108 increased the number

Reagents, generation of Tmem108 MT mice, qRT-PCR, X-gal assay, subcellular fractionation, cell culture and transfection, Golgi staining, in vivo electroporation of neonatal mice, electrophysiological analysis, immunostaining, behavior tests, and statistic analysis are described in SI Materials and Methods. Experimental procedures were approved by the institutional animal care and use committee (IACUC) of Augusta University.

1. Lewis DA, Lieberman JA (2000) Catching up on schizophrenia: Natural history and neurobiology. Neuron 28(2):325–334. 2. Owen MJ, Sawa A, Mortensen PB (2016) Schizophrenia. Lancet 388(10039):86–97. 3. Tamminga CA, Southcott S, Sacco C, Wagner AD, Ghose S (2012) Glutamate dysfunction in hippocampus: Relevance of dentate gyrus and CA3 signaling. Schizophr Bull 38(5):927–935. 4. Stan AD, et al. (2015) Magnetic resonance spectroscopy and tissue protein concentrations together suggest lower glutamate signaling in dentate gyrus in schizophrenia. Mol Psychiatry 20(4):433–439. 5. Kesner RP (2007) A behavioral analysis of dentate gyrus function. Prog Brain Res 163: 567–576. 6. Redondo RL, et al. (2014) Bidirectional switch of the valence associated with a hippocampal contextual memory engram. Nature 513(7518):426–430. 7. Taylor AM, et al. (2013) Hippocampal NMDA receptors are important for behavioural inhibition but not for encoding associative spatial memories. Philos Trans R Soc Lond B Biol Sci 369(1633):20130149. 8. O’Donovan MC, et al.; Molecular Genetics of Schizophrenia Collaboration (2008) Identification of loci associated with schizophrenia by genome-wide association and follow-up. Nat Genet 40(9):1053–1055. 9. Heath AC, et al. (2011) A quantitative-trait genome-wide association study of alcoholism risk in the community: Findings and implications. Biol Psychiatry 70(6):513–518. 10. Beveridge NJ, Cairns MJ (2012) MicroRNA dysregulation in schizophrenia. Neurobiol Dis 46(2):263–271. 11. Kondo K, et al. (2013) Genetic variants on 3q21 and in the Sp8 transcription factor gene (SP8) as susceptibility loci for psychotic disorders: A genetic association study. PLoS One 8(8):e70964. 12. Liu JJ, et al. (2007) Retrolinkin, a membrane protein, plays an important role in retrograde axonal transport. Proc Natl Acad Sci USA 104(7):2223–2228. 13. Fu X, et al. (2011) Retrolinkin cooperates with endophilin A1 to mediate BDNF-TrkB early endocytic trafficking and signaling from early endosomes. Mol Biol Cell 22(19):3684–3698. 14. Xu CC, Fu XP, Zhu SX, Liu JJ (2016) Retrolinkin recruits the WAVE1 protein complex to facilitate BDNF-induced TrkB endocytosis and dendrite outgrowth. Mol Biol Cell 27(21):3342–3356. 15. Tao Y, et al. (2009) Erbin regulates NRG1 signaling and myelination. Proc Natl Acad Sci USA 106(23):9477–9482. 16. Tang T, et al. (2010) A mouse knockout library for secreted and transmembrane proteins. Nat Biotechnol 28(7):749–755. 17. Lavado A, Oliver G (2007) Prox1 expression patterns in the developing and adult murine brain. Dev Dyn 236(2):518–524. 18. Seki T, Arai Y (1993) Distribution and possible roles of the highly polysialylated neural cell adhesion molecule (NCAM-H) in the developing and adult central nervous system. Neurosci Res 17(4):265–290. 19. Dani VS, et al. (2005) Reduced cortical activity due to a shift in the balance between excitation and inhibition in a mouse model of Rett syndrome. Proc Natl Acad Sci USA 102(35):12560–12565. 20. Zhou YD, et al. (2009) Arrested maturation of excitatory synapses in autosomal dominant lateral temporal lobe epilepsy. Nat Med 15(10):1208–1214. 21. Tao Y, et al. (2013) Erbin interacts with TARP γ-2 for surface expression of AMPA receptors in cortical interneurons. Nat Neurosci 16(3):290–299. 22. Jones CA, Watson DJ, Fone KC (2011) Animal models of schizophrenia. Br J Pharmacol 164(4):1162–1194. 23. Stanford SC (2007) The Open Field Test: Reinventing the wheel. J Psychopharmacol 21(2):134–135. 24. Gainetdinov RR, Mohn AR, Caron MG (2001) Genetic animal models: Focus on schizophrenia. Trends Neurosci 24(9):527–533. 25. Yin DM, et al. (2013) Reversal of behavioral deficits and synaptic dysfunction in mice overexpressing neuregulin 1. Neuron 78(4):644–657.

26. Keefe RS, Harvey PD (2012) Cognitive impairment in schizophrenia. Handb Exp Pharmacol (213):11–37. 27. Johnstone EC, Crow TJ, Frith CD, Husband J, Kreel L (1976) Cerebral ventricular size and cognitive impairment in chronic schizophrenia. Lancet 2(7992):924–926. 28. Altman J, Bayer SA (1990) Migration and distribution of two populations of hippocampal granule cell precursors during the perinatal and postnatal periods. J Comp Neurol 301(3):365–381. 29. Li G, Kataoka H, Coughlin SR, Pleasure SJ (2009) Identification of a transient subpial neurogenic zone in the developing dentate gyrus and its regulation by Cxcl12 and reelin signaling. Development 136(2):327–335. 30. Zafirov S, Heimrich B, Frotscher M (1994) Dendritic development of dentate granule cells in the absence of their specific extrinsic afferents. J Comp Neurol 345(3):472–480. 31. Holtmaat AJ, et al. (2005) Transient and persistent dendritic spines in the neocortex in vivo. Neuron 45(2):279–291. 32. McKinney RA, Capogna M, Dürr R, Gähwiler BH, Thompson SM (1999) Miniature synaptic events maintain dendritic spines via AMPA receptor activation. Nat Neurosci 2(1):44–49. 33. Medvedev NI, et al. (2008) The glutamate receptor 2 subunit controls post-synaptic density complexity and spine shape in the dentate gyrus. Eur J Neurosci 27(2):315–325. 34. Passafaro M, Nakagawa T, Sala C, Sheng M (2003) Induction of dendritic spines by an extracellular domain of AMPA receptor subunit GluR2. Nature 424(6949):677–681. 35. Tashiro A, Yuste R (2004) Regulation of dendritic spine motility and stability by Rac1 and Rho kinase: Evidence for two forms of spine motility. Mol Cell Neurosci 26(3):429–440. 36. Hotulainen P, Hoogenraad CC (2010) Actin in dendritic spines: Connecting dynamics to function. J Cell Biol 189(4):619–629. 37. Penzes P, Rafalovich I (2012) Regulation of the actin cytoskeleton in dendritic spines. Synaptic Plasticity (Springer, Vienna), pp 81–95. 38. Spence EF, Kanak DJ, Carlson BR, Soderling SH (2016) The Arp2/3 complex is essential for distinct stages of spine synapse maturation, including synapse unsilencing. J Neurosci 36(37):9696–9709. 39. van der Zwaag B, et al. (2010) A co-segregating microduplication of chromosome 15q11.2 pinpoints two risk genes for autism spectrum disorder. Am J Med Genet B Neuropsychiatr Genet 153B(4):960–966. 40. Stefansson H, et al.; GROUP (2008) Large recurrent microdeletions associated with schizophrenia. Nature 455(7210):232–236. 41. Pathania M, et al. (2014) The autism and schizophrenia associated gene CYFIP1 is critical for the maintenance of dendritic complexity and the stabilization of mature spines. Transl Psychiatry 4(3):e374. 42. Levy D, et al. (2011) Rare de novo and transmitted copy-number variation in autistic spectrum disorders. Neuron 70(5):886–897. 43. Mei L, Nave KA (2014) Neuregulin-ERBB signaling in the nervous system and neuropsychiatric diseases. Neuron 83(1):27–49. 44. Wu H, et al. (2012) β-Catenin gain of function in muscles impairs neuromuscular junction formation. Development 139(13):2392–2404. 45. Ting AK, et al. (2011) Neuregulin 1 promotes excitatory synapse development and function in GABAergic interneurons. J Neurosci 31(1):15–25. 46. Tang FL, et al. (2015) VPS35 deficiency or mutation causes dopaminergic neuronal loss by impairing mitochondrial fusion and function. Cell Reports 12(10):1631–1643. 47. Stranahan AM, et al. (2009) Voluntary exercise and caloric restriction enhance hippocampal dendritic spine density and BDNF levels in diabetic mice. Hippocampus 19(10):951–961. 48. Yin DM, et al. (2013) Regulation of spine formation by ErbB4 in PV-positive interneurons. J Neurosci 33(49):19295–19303. 49. Ito H, Morishita R, Iwamoto I, Nagata K (2014) Establishment of an in vivo electroporation method into postnatal newborn neurons in the dentate gyrus. Hippocampus 24(12):1449–1457. 50. Wen L, et al. (2010) Neuregulin 1 regulates pyramidal neuron activity via ErbB4 in parvalbumin-positive interneurons. Proc Natl Acad Sci USA 107(3):1211–1216.


Materials and Methods

ACKNOWLEDGMENTS. We thank members of the L.M. and W.-C.X. laboratories for helpful discussions, Dr. Jia-jia Liu for antibody, and Dr. Richard Huganir for constructs. This work was supported in part by grants from the US National Institutes of Health (L.M. and W.-C.X.) and Veterans Affairs (L.M. and W.-C.X.), “Thousand Talents” Innovation Project from Jiangxi Province (L.M.), and National Natural Science Foundation of China (NSFC) Grants 31271171 and 81471116 (to B.-M.L). L.M. is a Georgia Research Alliance Eminent Scholar in Neuroscience.

Jiao et al.

Supporting Information Jiao et al. 10.1073/pnas.1618213114 SI Materials and Methods Reagents, Antibodies, and Plasmids. Chemicals were purchased from

Sigma-Aldrich unless otherwise indicated. DL-AP5 (0105), CNQX (0190), and BMI (0130) were purchased from Tocris Bioscience. Information of primary antibodies is as follows: chicken anti– β-galactosidase (Abcam) (ab9361; 1:500 for staining); chicken antiGFP (Aves Labs) (GFP-1020; 1:2,000 for staining); rabbit anti-Prox1 (Millipore) (ab5475; 1:1,500 for staining); rabbit anti-GluN2A (R&D Systems) (PPS012; 1:1,000 for blotting); rabbit anti-Flag (Sigma) (F7425; 1:500 for staining); mouse anti-GFP (Abcam) (ab1218, 1:500 for staining); mouse anti-PSANCAM (Millipore) (MAB5324; 1:300 for staining); mouse anti-Flag (Sigma) (F3165; 1:300 for staining); mouse anti-NeuN (Neuromics) (MO22122; 1:1,000 for staining); mouse anti-PSD95 (Millipore) (MAB1598; 1:1,000 for blotting); mouse anti-synaptotagmin (DSHB) (mAB 30; 1:20 for blotting); mouse anti-GluA1 (Millipore) (MAB2263; 1:300 for staining, 1:1,000 for blotting); mouse anti-GluA2 (Millipore) (MAB397; 1:300 for staining, 1:1,000 for blotting); mouse antiGluN1 (Synaptic Systems) (114011; 1:1,000 for blotting); mouse antiGluN2B (DSHB) (75-097; 1:1,000 for blotting); mouse anti–β-actin (Upstate) (MAB1501; 1:8,000 for blotting); and mouse antiGAPDH (Santa Cruz) (SC-32233; 1:4,000 for blotting). Tmem108 antibody was generated against a GST-fusion protein containing the C-terminal amino acids 490–574 and affinity purified as described previously (22). pCAG-GFP and mCherry-C1 were obtained from Addgene. To generate Flag-Tmem108, the Tmem108 cDNA was generated by PCR and subcloned in pFlag-CMV1 downstream of an artificial signal peptide sequence and a Flag epitope (Sigma, E7273). GFP- tagged GluA2 was a gift given by Richard Huganir (Johns Hopkins University, Baltimore). The authenticity of all constructs was verified by DNA sequencing and Western blot analysis. Animals. Mice were derived from mutant embryonic stem (ES) cells

obtained from University of California, Davis Mutant Mouse Regional Resource Center (MMRRC: 032633-UCD), where pKOS.29 containing a lacZ/neomycin cassette was targeted at Tmem108 exon 3 (the first coding exon) by homologous recombination (15, 16). In Tmem108 mutant (MT) mice, exon 3 was partially deleted, which generates a premature stop codon. In addition, the polyadenylation termination signal contained in the cassette severely reduces the transcription of downstream DNA. Genotyping primers for wildtype allele were: 5′ AACCC CCAAC CATGA ACTTA TTTT 3′ and 5′ AAATG CTGCG TGGAC TTACT TA 3′ (547 bp) and for mutant allele: 5′ GAATC CCGCA TAACT ACGCA GAAT 3′ and 5′ GCAGC GCATC GCCTT CTATC 3′ (496 bp). Experimental procedures were approved by the institutional animal care and use committee of Augusta University. Quantitative RT-PCR Analysis. Quantitative real-time PCR (qRTPCR) was performed as described previously (44). Briefly, total RNA was isolated with TRIzol (Invitrogen). Equal amounts of total RNA were reverse transcribed with GoScript reverse transcriptase and Oligo(dT) primers (Promega). The SYBR Green/ROX (Fermentas) was used for qRT-PCR analysis. Samples were assayed in triplicates, with each plate having loading standards in duplicate. mRNA levels of Tmem108 were normalized to those of GAPDH. Primer sequences were: Tmem108-N, 5′ AAGTT TACAG GCCCT CTATT GC 3′ and 5′ GGAGA TGGTT CGTGG ACAGC 3′ and Tmem108-C, 5′ ACTGG AACAA TGCCA TCACA AT 3′ and 5′ AGTGT CTCGA TAGTC GCCAT TG 3′ Jiao et al. www.pnas.org/cgi/content/short/1618213114

X-Gal Assay. Mice were anesthetized and decapitated. Brain samples were isolated and rapidly frozen in OCT and cut into 40-μm sections and mounted on SuperFrost Plus slides (Fisher). Sections were fixed for 2 min in a buffer containing (in millimoles): 2 MgCl2, 5 EGTA with 0.2% glutaraldehyde, and 2% (wt/vol) paraformaldehyde. Sections were washed in ice-cold PBS and stained in X-gal solution [1 mg/mL X-gal, 5 mM K3Fe(CN)6, 5 mM K4Fe(CN)6, 0.02% Nonidet P-40, 0.01% deoxycholate, and 2 mM MgCl2 in PBS] at 37 °C overnight. After washing with PBS, slices were counterstained with nuclear Fast Red (Vector Laboratories), mounted in CC/Mount (Sigma), and sealed with coverslips. Subcellular Fractionation. Hippocampal postsynaptic density (PSD) fraction was prepared as described previously (21). Briefly, mouse hippocampi were homogenized in Solution A containing (in millimoles) 320 sucrose, 1 NaHCO3, 1 MgCl2, 0.5 CaCl2, 1 PMSF, and protease inhibitors. Homogenates were centrifuged at 470 × g for 2 min. Resultant supernatants (S1) were centrifuged at 10,000 × g for 10 min to obtain mitochondria- and synaptosome-enriched pellets (P2) and supernatants (S2) containing soluble proteins. P2 fractions were resuspended in 0.32 M sucrose, which was then layered onto 0.8 M sucrose in a centrifuge tube. After being centrifuged at 9,100 × g for 15 min in a swinging bucket rotor, synaptosomes (most of loose pellets) were collected from the 0.8 M sucrose layer and resuspended with equal volume of 20 mM Hepes (pH 7.0), 2% (vol/vol) Triton X-100, and 150 mM KCl. Samples were centrifuged at 20,800 × g for 45 min using a fixedangle rotor, and resulting supernatants were collected as the presynaptic fraction. Pellets were resuspended in a solution of 1% Triton X-100 and 75 mM KCl using a Dounce minihomogenizer and centrifuged again at 20,800 × g for 30 min. The pellets were washed with 20 mM Hepes, dissolved in 1× SDS/PAGE sample buffer, and collected as the PSD. shRNA Interference. Target sequence of Tmem108 shRNA was designed using the web-based Block-iT program (Invitrogen) and subcloned in XhoI and HpaI sites in pLentiLox 3.7 (Addgene). Tmem108 shRNA sequence was 5′ GCTTG AGCAA GATGG ATATT G 3′ (1,627–1,647 bp). Scramble shRNA sequence was 5′ AGAGT GGACG TCGAT GATTA T 3′. shRNA-resistant Tmem108 was constructed by mutating Tmem108 at 1,627–1,647 bp to the sequence of 5′ GTCTC AGTAA AATGG ACATC G 3′, with altering the amino acid sequence. All constructs were confirmed by sequencing. Cell Culture and Transfection. HEK293 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Corning) supplemented with 10% fetal bovine serum (FBS) (Gemini). Transient transfection was performed using polyethylenimine (Sigma), as described (21). Briefly, cells were cultured in six-well plates and at ∼70% confluence were incubated with precipitates formed by 3 μg of plasmid DNA and 140 μL of polyethylenimine 0.05% (wt/vol). Cells were harvested 36 h posttransfection. Hippocampal neurons were cultured as described previously (45). Briefly, brains were collected from P0 pups in ice-cold HBSS (Gibco). Hippocampi were isolated and kept separate from one another in HBSS on ice. Following digestion in 0.25% trypsin (one hippocampus in 1 mL) at 37 °C for 20 min. Dissociated cells were resuspended in plating media (DMEM supplemented with 10% FBS) and plated at a density of 3 × 104 or 6 × 104 per well onto poly-L-lysine–coated 8-mm coverslips (Fisher) in 12-well plates (Thermo Fisher). Cells were incubated for 4 h before 1 of 6

replacing with maintenance medium [neurobasal medium (Gibco) supplemented with 2% B-27 supplement (Gibco), 1% GlutaMax (Gibco), and 1% penicillin/streptomycin (Gemini)]. Neurons were maintained at 37 °C in 5% CO2, with half of the medium changed every 2–3 d. Neurons were transfected using calcium phosphate as described previously (46). Briefly, 4 μg DNA in 1–2 μL was mixed with 3 μL 2 M CaCl2 in ddH2O (total volume 30 μL), and further mixed with 30 μL of Hepes-buffered saline containing (in millimoles): 274 NaCl, 10 KCl, 1.4 Na2HPO4, 15 glucose, and 42 Hepes, pH 7.05. Resulting DNA-calcium phosphate precipitates were added into each well of neurons. Morphology was studied 3–6 d later. Prox1-positive neurons were subjected to morphology analysis. Golgi Staining. Golgi staining was performed by using the FD Rapid GolgiStain Kit following the manufacturer’s protocol (FD NeuroTechnologies). Brain tissues were incubated in mixed solutions A and B for 2 wk in the dark at room temperature and put into solution C for 3 d. Tissues were cut into slices with 100-μm thickness, stained with solutions D and E, dehydrated in gradient ethanol, cleared with xylene, and mounted on slides for imaging. Images of neurons in the superficial granule cell layer (GCL) were taken and imported into Image J for analysis. Dendrites were reconstructed and analyzed using the ImageJ Sholl analysis plugin, with 20-μm incremental increases in concentric circular diameter from soma. Spines of secondary and tertiary dendritic branches of randomly selected segments (20 μm each) of DG granule neurons were quantified (47, 48). In Vivo Electroporation of Neonatal Mice. In vivo electroporation was performed according to a previously published method (49). In brief, P0 pups were anesthetized, a sharp glass electrode with a beveled tip containing a solution of plasmid (1.5 mg/mL in TE with 0.2% Fast Green) was pierced through skin and skull. For rescue experiments, 3× of Tmem108 plasmid was mixed with GFP plasmid. After injection with 1 μL of DNA solution into both lateral ventricles, pups were immediately electroporated with tweezer-type electrodes using ECM830 (Harvard Apparatus) (five pulses of 110 V, 50 ms, with 950-ms interval). Electroporated animals were placed on a hot pad at 37 °C for recovery before being returned to dams. Electrophysiological Analysis. Mice (5- to 7-wk-old males) were anesthetized with ketamine/xylazine (Sigma, 100/20 mg/kg, respectively, i.p.). Brains were quickly removed and chilled in ice-cold modified artificial cerebrospinal fluid (ACSF) containing (in millimoles): 250 glycerol, 2 KCl, 10 MgSO4, 0.2 CaCl2, 1.3 NaH2PO4, 26 NaHCO3, and 10 glucose. Coronal hippocampal slices (300 μm) were cut in ice-cold modified ACSF using VT-1200S vibratome (Leica) and transferred to and incubated in a storage chamber containing regular ACSF (in millimoles, 126 NaCl, 3 KCl, 1 MgSO4, 2 CaCl2, 1.25 NaH2PO4, 26 NaHCO3, and 10 glucose) at 32 °C for 30 min and at room temperature (25 ± 1 °C) for an additional 1 h before recording. All solutions were saturated with 95% O2/5% CO2 (vol/vol). Slices were placed in the recording chamber, which was superfused (2 mL/min) with ACSF at 32–34 °C. Granule neurons from the outer layer of DG upper blade were visualized with infrared optics using an upright microscope equipped with a 40× water-immersion lens (Axioskop 2 Plus, Zeiss) and infraredsensitive CCD camera (C2400-75, Hamamatsu). Pipettes were pulled by a micropipette puller (P-97, Sutter) with a resistance of 3–5 MΩ and filled with the solution contained (in millimoles): 125 Cs-methanesulfonate, 5 CsCl, 10 Hepes, 0.2 EGTA, 1 MgCl2, 4 Mg-ATP, 0.3 Na-GTP, 10 phosphocreatine, and 5 QX314 (pH 7.4, 285 mOsm). Recordings were made in wholecell patch-clamp configuration with MultiClamp 700B amplifier and 1440A digitizer (Molecular Devices). Series resistance was controlled below 20 MΩ and not compensated. Jiao et al. www.pnas.org/cgi/content/short/1618213114

Spontaneous EPSC and IPSC were recorded in the presence of ongoing spontaneous activity at reversal potential of GABAA receptor-mediated (−70 mV) and glutamate receptor-mediated (0 mV) currents, respectively. Total excitatory and inhibitory charge was calculated by integrating baseline-subtracted spontaneous synaptic current. Evoked EPSCs and IPSCs were recorded by stimulating the medial perforated pathway with a concentric bipolar electrode (FHC) with various intensities at reversal potential of IPSC (−70 mV) and EPSC (0 mV), respectively. Miniature EPSCs were recorded at holding potential of −70 mV in the presence of 20 μM BMI and 1 μM TTX. To record miniature IPSCs, recording pipettes were filled with the solution contained (in millimoles): 140 CsCl, 10 Hepes, 0.2 EGTA, 1 MgCl2, 4 Mg-ATP, 0.3 Na-GTP, 10 phosphocreatine, and 5 QX314 (pH 7.4, 285 mOsm). Glutamatergic activity was blocked by the treatment of 20 μM CNQX and 100 μM DL-AP5. For paired-pulse ratio recording, EPSCs were evoked by stimulating the medial perforated pathway at holding potential of −70 mV in the presence of 20 μM BMI. Interval of paired stimulations was set at 25 ms. Ratio was defined as [(p2/p1] × 100, where p1 and p2 are the amplitude of the EPSCs evoked by the first and second pulse, respectively. Cells were rejected if membrane potentials were more positive than −60 mV, or if eEPSPs were unstable 10 min after establishing whole-cell configuration, or if series resistance fluctuated more than 20% of initial values. Data were filtered at 1 kHz and sampled at 10 kHz. Immunostaining. Animals were deeply anesthetized with isoflurane and perfused with PBS (incubated at 37 °C before using) followed by around 20 mL ice-cold 4% (wt/vol) paraformaldehyde until bodies became stiff. Brains were postfixed overnight at 4 °C in 4% paraformaldehyde and dehydrated using 15% (wt/vol) sucrose followed by 30% sucrose at 4 °C. Brain tissues were embedded in OCT, rapidly frozen, and cut into 40-μm sections. Slices were blocked and permeabilized using a solution of 0.3% Triton X-100 plus 5% goat serum PBS for 2 h at room temperature. They were incubated at 4 °C overnight with primary antibodies in PBS containing 5% goat serum and 2% BSA. Samples were washed with PBS and incubated at room temperature for 1–2 h with Alexa488–, Alexa-594–, or Alexa-647–conjugated secondary antibodies. For staining of surface GluA2, DIV15–17 neurons were fixed at 4 °C for 30 min, blocked with 5% goat serum in PBS at 4 °C, and incubated, without being permeabilized, at 4 °C overnight with monoclonal anti-extracellular GluA2 antibody. Staining was visualized by Alexa-594 goat anti-mouse antibody. Neurons were permeabilized with 0.1% Triton for 5 min at room temperature and incubated with anti-Prox1 antibody to mark DG neurons. For staining of total GluA2, neurons were permeabilized and incubated with both anti-GluA2 and anti-ProX1 antibodies. To analyze surface and total GluA2 levels in same neurons, GFP-GluA2–transfected neurons were incubated with anti-GFP antibody for 15 min at room temperature to label the surfaceexpressed protein. Neurons were fixed in 4% paraformaldehyde for 15 min and incubated with Alexa-594–conjugated (red) antichicken secondary antibody for 30 min. Neurons were then permeabilized with 0.1% Triton X-100, incubated with 10% normal goat serum, and labeled with anti-GFP and anti-Prox1 antibodies for 30 min at room temperature. After washing, samples were incubated with Alexa-488–conjugated (green) anti-mouse and 647-conjugated (blue) anti-rabbit secondary antibodies for 30 min. Slices and neurons on coverslips were mounted on SuperFrost Plus in Vectashield mounting medium (Vector). Images were taken by a Zeiss LSM510 confocal microscope and analyzed by ImageJ (NIH). Behavioral Analysis. Behavioral analysis was carried out with 8- to 12-wk-old mice by investigators unaware of genotypes. Male mice were handled for habituation for 3 d before behavioral tests. Locomotor activity was measured as described previously (50). 2 of 6

Mice were placed in a chamber (50 × 50 × 10 cm) and movement was monitored for 30 min using an overhead camera and tracking software (EthoVision, Noldus). Prepulse inhibition (PPI) was tested in sound-attenuated chambers (Med Associates). Before the test, mice were allowed to habituate to the chamber with a 65-dB background white noise for 5 min. During the test, mice were placed in a Plexiglass tube mounted on a plastic frame to monitor motion by a piezoelectric accelerometer and subjected to 12 startle trials (120 dB, 20 ms) and 12 prepulse/startle trials (20-ms white noise at 70, 75, or 80 dB at 100-ms intervals and 20-ms 120-dB startle stimulus). Different trial types were presented pseudorandomly with each trial type presented 12 times, and no two consecutive trials were identical. Mouse movement was measured during 100 ms after startle stimulus onset (sampling frequency 1 kHz). PPI (%) was calculated according to the formula: [100 − (startle amplitude on prepulsepulse trials/startle amplitude on pulse alone trials) × 100]. For working memory, mice were placed at the center of a Y-shaped maze with three arms (35 cm) and allowed to move

freely through the maze for 8 min. The total number and series of arm entries were recorded. Nonoverlapping entrance sequences (e.g., ABC, BCA) were defined as spontaneous alternations. For contextual fear conditioning, mice were placed in soundattenuating chambers (7 × 7.5 × 15 in., Coulbourn Instruments) on metal grid for 120 s before delivery of four footshocks (intensity 0.7 mA; duration 2 s, interval 90 s between shocks) with stimulator during the acquisition phase. Twenty-four hours later, fear memory was tested for 5 min in the same context. Time spent freezing during fear acquisition and recall was measured by Freezeframe 4 software (Actimetrics). Statistical Analyses. Statistical analysis was done by the GraphPad Prism version 5.0 (GraphPad Software). Two-way ANOVA was used in behavioral, morphological, and electrophysiological studies that analyze more than two parameters. Student’s t test or paired Student’s t test was used to compare data from two groups. All tests were two sided. Data were presented as mean ± SEM. P < 0.05 was considered to be statistically significant.

Fig. S1. Generation of Tmem108 MT mice. (A) Tmem108 was specifically expressed in the nervous system. Indicated organs or tissues were collected from 2-mo-old WT mice and homogenized for Western blotting. β-Actin and GAPDH served as loading controls. Hippo, hippocampus. (B) Diagrams of wild-type and mutant Tmem108 genes. (C) PCR genotyping. Values of wild-type and Tmem108 mutant bands are 547 bp and 496 bp, respectively. (D) Dramatically reduced Tmem108 mRNAs in DG tissues of MT mice. DG tissues, collected from WT and MT mice, were subjected to qRT-PCR with the primer Tmem108-N targeting exons 3 and 4 and the primer Tmem108-C targeting exons 5 and 6. n = 3 pairs of mice. **P < 0.01; ***P < 0.001; paired Student’s t test. (E) Tmem108 level was dramatically reduced in MT mice. Hippocampi of each genotype were homogenized for Western analysis. Shown were representative blots of more than three independent experiments with similar results. (F) Similar body weight between the two genotypes. n = 8 WT mice or 11 MT mice. P > 0.05; Student’s t test.

Fig. S2. Identification of Prox1-positive DG granule neurons and characterization of shRNA-Tmem108 constructs. (A) Representative image of a Prox1-positive neuron. DIV13 hippocampal neurons were cotransfected with Flag-Tmem108 and pCAG-GFP constructs, and quadruple-stained with DAPI, anti-Flag, anti-GFP, and anti-Prox1 antibodies at DIV20. (Scale bars: 30 μm and 5 μm.) (B) Reduction of Tmem108 level by shRNA-Tmem108 and resistance of rTmem108. HEK293 cells were cotransfected with indicated constructs and were lysed 36 h later for Western blot analysis. β-Actin served as loading control.

Jiao et al. www.pnas.org/cgi/content/short/1618213114

3 of 6

Fig. S3. Characterization of gross morphology and neuron numbers of Tmem108 MT mice. (A) No detectable effect of Tmem108 mutation on global morphology. Brain sections from 1-mo-old mice of each genotype were subjected to Nissl staining. (B) Comparable numbers of granule neurons between WT and MT mice. Brain sections were stained with anti-NeuN antibody. (Scale bars: Top, 300 μm; Bottom, 100 μm.) (C) Quantitative analysis of data in B. n = 3 mice for each genotype. P > 0.05; Student’s t test.

Fig. S4. Characterization of dendritic morphology of DG granule neurons in Tmem108 MT mice. (A) Representative reconstructed dendrite morphology of DG granule neurons. (Scale bar: 20 μm.) (B) Similar total dendrite length between the two genotypes. n = 20 neurons, three mice for each genotype. P > 0.05, Student’s t test. (C) Normal dendrite complexity by Sholl analysis. n = 20 neurons, three mice for each genotype. F(1,396) = 0.06, P > 0.05; repeated two-way ANOVA.

Fig. S5. Similar sIPSC between WT and MT mice. (A) Representative mIPSC traces in DG granule neurons. (Scale bars: 2 s and 10 pA.) (B and C) Quantitative analysis of mIPSCs frequency (freq.) (B) and amplitude (amp.) (C). n = 14 neurons, 3 WT mice; n = 15 neurons, 3 MT mice. P > 0.05; Student’s t test.


4 of 6

Fig. S6. Quantitative analysis of protein reduction. (A) Different amounts of PSD fractions (in micrograms of protein) were resolved by SDS/PAGE and blotted with indicated antibodies. Densities of bands were quantified by NIH ImageJ. (B) Band densities were plotted against protein concentrations and analyzed by linear regression. (C) Histograms of protein level reductions in Tmem108 MT mice. Protein level of control WT mice was used as 1. *P < 0.05; paired Student’s t test.


5 of 6

Fig. S7. Reduced surface/total ratio of GluA2 level in Tmem108-deficient DG granule neurons. (A) Surface/total ratio of GluA2 intensity was decreased. Cultured hippocampal neurons were transfected with GluA2-GFP at DIV10. At DIV15–17, neurons were first incubated with chicken anti-GFP antibody for 15 min at room temperature to label surface-expressed proteins. They were then permeabilized and costained with mouse anti-GFP and rabbit anti-Prox1 antibodies. GluA2 staining intensity in dendrites was quantified. Image in dotted area was enlarged as shown at Right. Arrow, Prox1-positive neurons. (Bars: Left, 30 μm; Right, 5 μm.) (B) Quantitative analysis of data in A. Four dendrites were used in each neuron. n = 20 neurons of each genotype from three independent experiments. *P < 0.05; Student’s t test. (C and D) No change in total or surface GluA2 levels in Prox1-negative neurons in MT mice. DIV15–17 neurons were stained with anti-extracellular GluA2 antibody under permeabilizing and nonpermeabilizing conditions to assess total and surface GluA2. GluA2 staining intensity in dendrites and soma was quantified. Image in dotted area was enlarged as shown at Bottom. Side bar, glow scale of GluA2 staining intensity in arbitrary units. (Scale bars: Top, 30 μm; Bottom, 5 μm.) (E) Quantitative analysis of data in C and D. Shown were GluA2 puncta number/10 μm (Left), GluA2 puncta area (square micrometers, Middle), and GluA2 soma intensity (normalized to WT, Right). Four dendrites were used from each neuron. n = 15 neurons for each genotype from three independent experiments. P > 0.05; Student’s t test.


6 of 6