[1] 13w Loss of TDP-43 function underlies hippocampal and cortical synaptic deficits in TDP-43 proteinopathies

[1] 152w TDP-43 proteinopathy is linked to neurodegenerative diseases that feature synaptic loss in the cortex and hippocampus, although it remains unclear how TDP-43 regulates mature synapses. We report that, in adult mouse hippocampus, TDP-43 knockdown, but not overexpression, induces robust structural and functional damage to excitatory synapses, supporting a role for TDP-43 in maintaining mature synapses. Dendritic spine loss induced by TDP-43 knockdown is rescued by wild-type TDP-43, but not ALS/ FTLD-associated mutants, suggesting a common TDP-43 functional deficiency in neurodegenerative diseases. Interestingly, M337V and A90V mutants also display dominant negative activities against WT TDP-43, partially explaining why M337V transgenic mice develop hippocampal degeneration similar to that in excitatory neuronal TDP-43 knockout mice, and why A90V mutation is associated with Alzheimer's disease. Further analyses reveal that a TDP-43 knockdown-induced reduction in GluN2A contributes to synaptic loss. Our results show that loss of TDP-43 function underlies hippocampal and cortical synaptic degeneration in TDP-43 proteinopathies.

[1] 104w Nuclear protein TAR DNA/RNA binding protein 43 (TDP-43) has been shown to mediate physiological functions by regulating DNA transcription, as well as RNA splicing, stability and translation [1][2][3][4][5]. TDP-43 pathology was first identified in postmortem spinal cord, cortex, and hippocampus pathological samples from amyotrophic lateral sclerosis (ALS) and frontotemporal lobar degeneration (FTLD) patients [6], and then later found in samples from patients with other neurodegenerative diseases, including Alzheimer's disease (AD), Lewy body diseases (LBD), and Huntington's disease (HD) [7][8][9]. More than 50 TDP-43 mutations associated with ALS/ FTLD have been identified, further supporting a causal role of TDP-43 pathology in the these diseases [10,11].

[2] 207w Typical TDP-43 pathology is characterized by abnormal formation of TDP-43 cytosolic aggregations concurrent with loss of nuclear TDP-43, which has led to conflicting hypotheses that both gain-of-function and loss-of-function TDP-43 mutations can lead to pathogenic outcomes [2,[12][13][14]. Transgenic expression of wild-type or disease-associated TDP-43 mutants in mice recapitulated certain features of related diseases in some studies, supporting the gain-of-function theory. However, conditional TDP-43 knockout in excitatory neurons also caused neurodegeneration, arguing for a loss-of-function hypothesis [15]. Thus, to better understand and treat these associated neurodegenerative diseases, research is urgently needed to resolve whether gain or loss of TDP-43 function is the salient mechanism underlying neurodegeneration [2]. TDP-43 pathology has been observed in 80-90% of hippocampal sclerosis (HS) cases. HS is a severe condition of hippocampal degeneration associated with cognitive decline that commonly accompanies some neurodegenerative diseases, including FTLD, LBD, and AD. Approximately 30-70% of all the AD patients have TDP-43 pathology [16][17][18]. Notably, AD patients with TDP-43 pathology have greater hippocampal atrophy and more severe cognitive decline compared to those without TDP-43 pathology [19,20], suggesting that TDP-43 abnormalities contribute to greater hippocampal degeneration. In addition, the TDP-43 A90V mutation is associated with increased AD risk [21,22], further supporting an essential role for TDP-43 in regulating hippocampal functions.

[3] 163w Mice with conditional TDP-43 knockout (cKO) in the forebrain principal neurons develop robust hippocampal degeneration by 12 months of age [15], suggesting that loss of TDP-43 is sufficient to cause hippocampal degeneration. Consistent with this possibility, substantial hippocampal degeneration has only observed with M337V transgenic mice, not with the other TDP-43 transgenic mice. Transgenic mice expressing the human M337V TDP-43 mutant (but not wild-type TDP-43) driven by the Thy-1 promoter develop severe neuronal loss, astrogliosis, and microgliosis in the hippocampal CA region [23], similar to effects observed in TDP-43 cKO mice. It thus remains unclear why M337V expression, but not other reported TDP-43 mutations, leads to severe hippocampal degeneration in transgenic mice while its overexpression results in degenerative effects resembling TDP-43 cKO in the hippocampus [15,24,25]. Moreover, all of these above-mentioned manipulations of TDP-43 begin early during development, so it remains uncertain whether TDP-43-associated neurodegeneration is a lateonset result of these developmental defects, or if it truly reflects TDP-43 dysfunction in adult brains.

[4] 222w Synaptic loss/degeneration is the early pathological event of neurodegeneration in many neurodegenerative diseases [26][27][28][29], including those associated with TDP-43 pathology [30]. TDP-43 has also been established as a regulator of early dendrite and synapse development in hippocampal neurons, although its precise role is still controversial [31][32][33][34]. Germane to understanding the etiology of neurodegenerative diseases is the question of whether TDP-43 also drives commonly observed morphological and functional changes in mature synapses. To explore these questions, we used a highly tractable organotypic rat hippocampal slice culture system in conjunction with in vivo molecular manipulations to systematically investigate whether and how TDP-43 dysfunction affects mature hippocampal and cortical synapses. Surprisingly, TDP-43 knockdown (KD), but not its overexpression, reduced the density and function of mature synapses both in vitro and in vivo. This effect was rescued by wildtype TDP-43, but not by truncation mutants deleted for the RRM or GRD domains, nor by ALS/FTLD/AD-associated TDP-43 mutants. Among these ALS/FTLD/AD-associated mutations, we found that M337V and A90V functioned in a dominant negative manner, which explains why M377V transgenic expression, and not other mutations, resulted in hippocampal degeneration similar to TDP-43 cKO. Our data strongly support an essential role for TDP-43 in maintaining mature hippocampal and cortical excitatory synapses, and further indicate that loss of TDP-43 function plays a major role in hippocampal degeneration in TDP-43 proteinopathies.

[1] 109w In DIV10 organotypic hippocampal slice cultures, CA1 pyramidal neurons at seven days of TDP-43-HA (TDP-43 with a C-terminal tag) co-expression showed similar dendritic spine density as the control neurons expressing GFP alone (Fig. 1A, B; GFP alone, 0.598 ± 0.029/μm; TDP-43: 0.546 ± 0.032/μm, P = 0.2348). In dissociated mature rat hippocampal neurons, TDP-43-HA was transiently expressed in the nucleus without notable cytosolic inclusions (Supplementary Fig. S1A). Expression of TDP-43-HA in rat primary neurons and the specificity of TDP-43 antibody were also verified (Supplementary Fig. S1B). This indicated that transient overexpression of TDP-43 in rat hippocampal neurons did not result in formation of inclusions or damage to excitatory synapses.

[2] 119w In order to study the impact of TDP-43 deficiency, two shorthairpin RNAs (shRNA; in pSuper) targeting rodent TDP-43 (shTDP_1 and shTDP_2) were selected based on their efficacy for knockdown of rat TDP-43 overexpression in HEK293T cells. The two shRNAs were then expressed with a lentiviral vector in dissociated rat hippocampal neurons. Immunostaining showed that approximately 90% of the neurons (positive for neuronspecific marker MAP2) were infected (labeled by GFP expression from the same viral vector) (Supplementary Fig. S1C). Both mRNA (Fig. 1C) and protein (Fig. 1D) levels of endogenous TDP-43 were much lower in neurons expressing shTDP_1 or 2 than those of the neurons expressing control shRNA (shLuc: a shRNA targeting firefly luciferase; Fig. 1E and quantified in F.

[3] 113w shTDP_1 targets the 3' non-coding region of rodent TDP-43 mRNA. shTDP_2 targeting a specific rodent coding region that is different from that in human TDP-43. Therefore, the human TDP-43-HA (hTDP-43) construct, which contains only human coding sequence, was resistant to both shTDP_1 and 2 (Supplementary Fig. S1D). Co-expression of hTDP-43 robustly rescued shTDP_1-or 2-induced loss of spines (Fig. 1G, quantified in H). In addition, unlike shTDP_2, shTDP_1 also failed to knockdown wildtype rat TDP-43 containing only the coding region (Supplementary Fig. S1E). We generated a rat TDP-43 mutant with silent mutations in the shTDP_2 targeting sites (rTDP-43 RES -HA) and found that it was resistant to both shTDP_1 and 2 (Supplementary Fig. S1F).

[4] 48w Similar to hTDP-43, co-expression of rTDP-43 RES also significantly rescued shTDP_1-or 2-induced spine loss (Supplementary Fig. S1G, quantified in S1H). These experiments confirmed that the reduction in dendritic spines from neurons expressing shTDP_1 or 2 was specifically caused by knockdown of endogenous TDP-43, not from their off-target effects.

[5] 16w Furthermore, in vivo, mature rat hippocampal pyramidal neurons infected for ten days with lentivirus expression vector

[6] 127w A B C D + TDP-43 + shLuc + shTDP_1 + shTDP_2 E F Merge G 0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 GFP-PSD-95 mKate2-F-actin 0.0 0.2 0.4 0.6 Tuj1 TDP-43 0.0 0.5 1.0 1.5 **** **** H I ** shLuc shTDP_1 J + filler + TDP-43 + TDP-43 + filler K GFP-PSD-95 mKate2-F-actin Merge shTDP_1 + GFP shTDP_2 + GFP shTDP_2 + GFP 0.0 0.1 0.2 0.3 0.4 0.5 shTDP_1 + GFP **** **** L GFP GFP **** **** 0.0 0.2 0.4 0.6 M 7 6 7 16 14 15 12 14 1 2 2 2 8 1 0 2 10 8 11 8 5 5 5 p=0.0003 p=0.0007 p=0.0054 p=0.0004 p＜0.0001 p＜0.0001 p＜0.0001 p＜0.0001 GFP only shLuc shTDP_1 shTDP_2

[7] 108w co-encoding shTDP_1 or _2 and GFP had significantly lower spine density (about 40-50%) compared with the neurons expressing shLuc (Fig. 1I, J). This result suggests that TDP-43 is required for the maintenance of mature dendritic spines in adult rat hippocampus. However, loss of dendritic spines might not always entail loss of synapses [35]. To address this question, GFP-PSD-95 and mKate2β-actin were co-expressed with shTDP_1 to simultaneously label postsynaptic density of excitatory synapses and dendritic spines in rat organotypic hippocampal cultures. The densities of both PSD-95 and F-actin puncta were significantly downregulated in CA1 pyramidal neurons expressing shTDP_1 (Fig. 1K-M), thus supporting that TDP-43 knockdown also reduced synapse density.

[8] 95w To test whether TDP-43 was also important for maintaining excitatory synapses in mice, we analyzed the density of dendritic spines from CA1 pyramidal neurons infected with lentivirus encoding shTDP_1, 2 or shLuc with GFP in adult mice (Fig. 2A). Consistently, the spine density of CA1 pyramidal neurons expressing shTDP_1 or 2 for 2 weeks was 29-41% significantly lower than that of the neurons expressing shLuc (Fig. 2B, quantified in 2 C). Furthermore, the spine density of adult mouse cortical neurons was also decreased by approximately 25% under shTDP_1-mediated suppression (Supplementary Fig. S2A, quantified in S2B).

[9] 94w Excitatory synaptic transmission was then measured by wholecell patch-clamp recording of cortical neurons infected with lentivirus encoding shLuc or shTDP_1, as well as neighboring uninfected neurons. The miniature excitatory postsynaptic current (m) frequency was about 60% lower in neurons expressing shTDP_1 (compared with the neighboring uninfected neurons), but not in neurons expressing shLuc (Fig. 2D, E). In addition, mEPSC amplitude was not altered in neurons expressing either shTDP_1 or shLuc (Fig. 2D, F). These results together suggested that TDP-43 knockdown led to reduced excitatory synapse density without affecting the strength of the remaining synapses.

[10] 54w To test the behavioral impact of TDP-43 knockdown in hippocampus, we bilaterally injected Adeno-associated viruses (AAV) encoding shTDP_1 or shLuc with GFP into adult mouse hippocampus (Fig. 2G). GFP fluorescence imaging showed that the AAV construct was expressed throughout the entire hippocampal region (Fig. 2G). Four to five weeks later, animal behavior was examined.

[11] 44w Using Open Field Tests (OFT), which revealed that mice expressing shTDP_1 traveled slightly longer distances with a higher speed than the mice expressing shLuc (Supplementary Fig. S2C, D). The rearing frequency and duration were not different between the two groups (Supplementary Fig. S2E, F).

[12] 193w In Fear Conditioning Tests (FCT), mice expressing shTDP_1 had similar baseline levels of freezing time as mice expressing shLuc before training (Fig. 2H), thus suggesting that the changes in locomotion did not affect the baseline freezing behavior. Interestingly, mice expressing shTDP_1 froze for significantly shorter durations (~50%) than those expressing shLuc in the hippocampus-dependent contextual FCTs after training (Fig. 2H, left). Whereas these two groups performed similarly in hippocampus-independent Cued FCTs (Fig. 2H, right). These findings supported that the hippocampal loss of TDP-43 in adult mice was sufficient to disturb hippocampal function and cause hippocampus-dependent cognitive dysfunction. We also found that after each of the three training sessions in Day1, mice expressing shTDP or shLuc froze gradually more than their baseline before training, indicating that both groups had learned from the training (Supplementary Fig. S2G). The mice expressing shTDP_1 appeared to freeze slightly less than the shLuc control group after each of the training sessions, but none reached statistical significance (Supplementary Fig. S2G). It suggested that both groups of mice had similar learning phase in this test. Therefore, loss of TDP-43 in hippocampus mainly impaired the retrieval, not formation of contextual fear memory.

[13] 129w These two groups of mice performed similarly in Y-maze Tests, suggesting that they had similar spatial working memory (Supplementary Fig. S2H). Then, in Elevated Plus Maze Tests, mice expressing shTDP_1 stayed slightly longer in the open arms and spent less time in the closed arms (Supplementary Fig. S2I, J), suggesting they were anxiolytic compared with the control group. In Tail Suspension Tests and Forced Swimming Tests, these two groups exhibited no difference in performance (Supplementary Fig. S2K, L), indicating that the loss of hippocampal TDP-43 did not alter depression-like behaviors. We also verified the reduction of TDP-43 expression in the CA1 region of these mice by immunostaining (Fig. 2I). Taken together, these results show that TDP-43 is necessary for maintaining excitatory synapses and related cognitive function in adult mice.

[14] 106w The RNA/DNA recognition motif and Glycine-rich domains were both necessary for TDP-43 to maintain hippocampal excitatory synapses TDP-43 protein has 414 amino acid residues, including an N-terminal domain (NTD: aa 1-77), a nuclear localization sequence (NLS: aa 82-98), two RNA/DNA recognition motifs (RRM1: aa 105-181; RRM2: aa 191-261), a nuclear export sequence (NES: aa 239-250) and a Glycine-rich domain (GRD: aa 274-414) (Fig. 3A) [14,[37][38][39][40]. To further study the molecular mechanisms of which domain is necessary for maintaining synapse, we generated a series of whole-domain deletion mutants, then investigated changes in dendritic spine density associated with expression of the mutant variants and shTDP_1 or 2 constructs.

[15] 402w Most truncation mutants were similarly expressed in a diffused subcellular distribution pattern in both nuclei and cytosol of dissociated rat hippocampal neurons except for the ΔNLS, ΔRRM1, Fig. 1 Transient TDP-43 knockdown, not overexpression, significantly reduced mature excitatory synapses in rat hippocampus. Representative images (A) and spine density (B) of basal dendrites from CA1 pyramidal neurons transiently expressing GFP alone or together with human TDP-43 for 7 days in rat organotypic hippocampal slice cultures. RT-qPCR measurement of TDP-43 mRNA levels (C) and western blots showing TDP-43 protein levels (D) in primary hippocampal neuronal cultures infected with lentivirus expressing either shLuc or TDP-43 shRNAs. Neuronal cultures were infected at 14 days in vitro (DIV14) and cultured for 5-6 days before analysis at DIV20. GAPDH was used as the internal control for RT-qPCR and mRNA expression was normalized to the shLuc control. Mock control shows neuronal cultures without viral infection. E, F Representative dendrites and quantitation of their spine density in CA1 pyramidal neurons transfected with shRNAs and GFP. Representative images of basal dendrites (G) and quantitation of their spine density (H) in CA1 pyramidal neurons from rat organotypic slice cultures co-expressing GFP and two different TDP-43 shRNAs with shRNA-resistant TDP-43 or a filler plasmid (see Methods) for 5-6 days. Representative images of basal dendrites (I) and spine density quantitation (J) from 8-week-old rat CA1 pyramidal neurons infected with lentivirus expressing GFP and TDP-43 shRNAs or shLuc for 10 days. Representative images (K) and quantitation of PSD-95 (L) and F-actin (M) puncta in secondary basal dendrites of CA1 pyramidal neurons transiently co-expressing shLuc (K, left) or shTDP_1 (K, right) with GFP-PSD-95 (K, top) and mKate2-β-actin (K, middle) for 5 days in rat organotypic hippocampal cultures. Scale bar, 10 μm. Numbers of neurons measured for each condition are indicated in the individual bars. Student's t test for B, C, H, L and M; One-way ANOVA with multiple comparison test for F and J; Quantifications are presented as mean ± SEM, **p < 0.01, ***p < 0.001, ****p < 0.0001. and ΔRRM (the mutant without both RMM domains) mutants (Supplementary Fig. S3A). The majority of ΔNLS variants were localized in the cytosol, with a small fraction in the nucleus. ΔRRM1 and ΔRRM variants formed puncta-like structures in the nucleus (Supplementary Fig. S3A). All of these truncation variants were expressed at similar levels as the WT TDP-43 in HEK293T cells and dissociated neurons (Supplementary Fig. S3B, C).

[16] 28w Transient expression of ΔRRM significantly reduced spine density of CA1 pyramidal neurons by about 30% in organotypic rat cultures, whereas expression of the other mutant variants did not

[17] 100w A B C D shLuc shTDP_1 uninfected E ** shTDP_1 shTDP_2 shTDP_1 shTDP_2 0.0 0.2 0.4 0.6 0.8 1.0 uninfected infected F G I * 100 80 60 40 20 0 shLuc shTDP_1 shTDP_1 0 50 100 150 shLuc **** **** Fear conditioning: cued shLuc shTDP_1 H infected 1s 10 pA contextual **** ** n=14 n=13 n=14 n=13 injection Cue test Context test Training 4-5 weeks Day1 2 3 8 weeks age immunostaining 0 5 10 15 uninfected infected 0.0 0.2 0.4 0.6 0.8 **** **** 35 48 39 n=14 n=13 n=14 n=13 p=0.0036 p=0.01 p=0.0433 p＜0.0001 p＜0.0001 p＜0.0001 p＜0.0001

[18] 151w alter the density of spines (Fig. 3B, quantified in C). Co-expression of ΔNTD, ΔNLS, ΔRRM1, ΔRRM2, or ΔNTDΔRRM1 significantly reversed the spine loss caused by shTDP_1 (Fig. 3D, quantified in E). Whereas ΔRRM and ΔGRD failed to rescue the spine loss (Fig. 3D, quantified in E). These results showed that at least one of the RRM domains and GRD were required for TDP-43 to maintain hippocampal synapses, while NTD was dispensable. Co-expression of ΔRRM with shTDP_1 did not further reduce spine density, suggesting that ΔRRM functioned in a dominant negative manner. Both ΔRRM1 and ΔRRM variants formed puncta-like structures in the nuclei, however only ΔRRM displayed a dominant negative activity. This finding indicated that formation of the puncta-like structures was likely unrelated to ΔRRM's dominant negative activity. Taken together, these data indicate that DNA/RNA binding and protein-protein interaction function of TDP-43 are necessary for its ability to maintain hippocampal excitatory synapses.

[19] 266w ALS/FTLD/AD-associated TDP-43 mutants lost their capability to maintain hippocampal synapses More than 50 TDP-43 mutations have been associated with ALS/ FTLD/AD [10,11]. We selected eight representative mutants based their regional distribution within the TDP-43 protein and their disease association for further investigation into their ability to maintain dendritic spines [7,41]. We first confirmed that all mutants were expressed at comparable levels in HEK 293T cells (Supplementary Fig. S3D). Then, in primary hippocampal neurons, we observed that all of these mutants were localized diffusely in the neuronal nuclei and cytosol, similar to that of WT TDP-43 (Supplementary Fig. S3E). Only Q331K localized predominantly in the nucleus. Similar to that of WT TDP-43, transient expression of six of the disease-associated mutants did not alter spine density in CA1 pyramidal neurons from rat organotypic hippocampal cultures. However, expression of A90V and M337V did significantly reduce dendritic spine density (Fig. 4A, quantified in B). Interestingly, in the rescue experiment, six of the mutants completely failed to reverse the spine loss caused by shTDP_1. By contrast, P112H significantly reversed the spine loss caused by TDP suppression (Fig. 4C, quantified in D), to a similar, but lesser, extent than that observed with WT TDP-43, thus suggesting that P112H was almost fully functional in maintaining hippocampal dendritic spines. Additionally, Q331K also reversed shTDP_1-induced spine loss (Fig. 4C, quantified in D), but to a significantly lower level than that of rescue by WT-TDP-43. These results indicated that Q331K was a partial loss of function mutant. Notably, A90V and M337V did not further reduce shTDP_1-mediated spine loss, suggesting they both worked in dominant negative manners.

[20] 162w To test the impact of TDP-43 overexpression in vivo, we generated lentivirus constructs for expression of A90V or M337V at similar levels to that of WT TDP-43 in cultured rat hippocampal neurons (Fig. 4E). CA1 pyramidal neurons were then co-infected with two viruses encoding either GFP or TDP-43 (WT or mutants) at a 1:4 ratio in adult mice. Two weeks after infection, almost all the GFP positive neurons were also positive for TDP-43 (Fig. 4F-I), thereby confirming successful co-expression of TDP-43 in GFP positive cells. Similar to that of WT TDP-43, A90V and M337V were generally expressed at comparable levels in the nuclei without abnormal inclusions in vivo (Fig. 4F-I, Supplementary Fig. S4A, B). Consistent with the in vitro results, WT TDP-43 co-expression did not alter spine density of CA1 pyramidal neurons in adult mice (Fig. 4J, quantified in K). In contrast, A90V or M337V co-expression significantly reduced dendritic spine density by about 20% or 35% respectively (Fig. 4J, quantified in K).

[21] 85w We next used HEK 293T cells to examine the effects of mouse M337V mutant (mM337V) expression at similar levels to that of WT hTDP43 (Supplementary Fig. S4C). Different from hTDP-43, mM337V tended to mainly localize in the neuronal nucleus (Supplementary Fig. S4D). Transient expression of mM337V failed to decrease spine density of CA1 pyramidal neurons in rat organotypic hippocampal cultures (Supplementary Fig. S4E, F). Together, these results showed that mM337V was a pure loss of function mutant without dominant negative activity in regulating hippocampal synapses.

[22] 123w To further explore the mechanisms underlying the TDP-43 mutant dominant negative activity, we generated three M337V truncation mutant variants. To this end, the first 10 residues from the N-terminus (NT10), which were required for TDP-43 homodimerization, were deleted in the ΔNT10 M337V variant [42,43]. ΔNT10 M337V was expressed in both nuclei and cytosol of neurons, whereas ΔNLS M337V was localized only in the cytosol (Supplementary Fig. S4G). Interestingly, both ΔNT10 M337V and ΔNLS M337V failed to reduce the density of dendritic spines when expressed in rat CA1 pyramidal neurons (Fig. 5A, quantified in B) [42,43]. In contrast, ΔRRM M337V was primarily localized in the nucleus, and the co-expression of which successfully reduced the density of dendritic spines (Fig. 5A, B, Supplementary Fig. S4G).

[23] 117w Notably, we found that ΔNT10 M337V co-expression failed to rescue shTDP_1-induced spine loss (Fig. 5C, quantified in D), suggesting that it was a loss of function mutant without dominant negative activity. We then examined how the loss of NT10 affected interactions between ΔNLS and M337V. Co-expression of M337V with the ΔNLS mutant increased the cytosolic subcellular localization of M337V (Fig. 5E, top), whereas ΔNLS missing the first ten residues (ΔNT10ΔNLS) failed to alter M337Vsubcellular localization (Fig. 5E, bottom). This was consistent with previous reports that the first 10 residues were critical for TDP-43 dimerization. Together these data indicate that the dominant negative effects of M337V require both homodimerization (likely with endogenous TDP-43) as well as nuclear localization.

[24] 157w However, expression of the three A90V truncation mutants all significantly reduced the density of dendritic spines (Fig. 5F, quantified in G), and co-expression of ΔNT10 A90V with shTDP_1 also failed to rescue TDP-43 silencing-induced spine loss (Fig. 5H, quantified in I). These results supported that homodimerization with endogenous TDP-43 was not required for the dominant negative activity of A90V, unlike M337V. While ΔNT10 A90V was expressed in both the nuclei and cytosol of neurons (Supplementary Fig. S4H), ΔRRM1 A90V and ΔRRM A90V were barely detectable in the neuronal cytosol (Supplementary Fig. S4H), implying that the dominant negative activity of A90V likely occurred in the nuclei, in agreement with results of the M337V mutant. We confirmed that all of these M337V and A90V truncation mutants were expressed at similar levels as WT TDP-43 in HEK 293T cells (Supplementary Fig. S4I), suggesting their differences in regulatory activity toward dendritic spines were not caused by changes in their expression levels.

[25] 243w Based the mutagenesis study, we postulated that the DNA/RNA binding and protein-protein interaction activity of TDP-43 were both necessary for its dendritic spine maintenance activity (Fig. 3B-E). TDP-43 is known to regulate mRNA by binding to mRNA and interacting with other proteins. To further investigate how loss of TDP-43 regulates mRNA in neurons, we performed a quantitative genome-wide RNA-sequencing (RNAseq) for gene expression profiling of mature dissociated rat hippocampal A B GFP only + M337V + T115A + P112H + C173S + A90V + Y374X D + filler + P112H + T115A + C173S + Q331K + A315T + M337V + Y374X + A90V C + A315T + Q331K shTDP_1 + GFP * 0.0 0.2 0.4 0.6 ** E F GFP only + WT + M337V + A90V J H TDP-43 Tuj1 GFP HA(TDP-43) Merge G I K GFP HA(A90V) Merge GFP HA(M337V) Merge shTDP_1 + GFP 0.0 0.1 0.2 0.3 0.4 0.5 * ** **** 12 8 9 11 14 12 10 9 7 11 8 9 14 10 12 15 12 13 13 * * **** 0.0 0.4 0.8 1.2 12 14 8 14 p=0.0349 p=0.0075 p=0.0055 p=0.0194 p=0.0328 p=0.0292 GFP Analysis of RNAseq data showed that TDP-43 could modulate the alternative splicing of many genes, including genes encoding neurodegenerative disease-associated proteins [4,[44][45][46]. In neurons expressing shTDP_1, 479 alternative splicing events were found to be significantly changed compared with the shLuc control group (Supplementary Fig. S5A, B). We also found 405

[26] 426w A B G GFP only + ΔNLS M337V + ΔNT10 M337V + ΔRRM M337V 0.0 + filler + ∆NT10 M337V H C GFP only + ΔRRM + ΔRRM1 D E F + filler + ΔNT10 shTDP_1 + GFP ** 0.0 0.2 0.4 0.6 0.1 0.2 0.3 0.0 0.1 0.2 0.3 I HA Flag DAPI Merge GFP M337V-HA + ∆NLS-Flag M337V-HA + ∆NT10∆NLS-Flag shTDP_1 + GFP 13 13 9 8 13 15 10 10 0.0 0.2 0.4 0.6 11 13 12 8 p=0.0071 p=0.0021 p=0.0299 p=0.0011 exon skipping (ES) events that were significantly altered in neurons expressing shTDP_1. Among these, 221 mRNA isoforms included more exons, and 184 mRNA isoforms excluded more exons. Gene Ontology analysis found that synapse-or dendriterelated pathways were enriched for these genes with altered ES in the shTDP_1 group (Supplementary Fig. S5C). Knockdown of TDP-43 in rat neurons led to changes in the ES events of 16 genes, resulting in altered ES patterns that resembled those observed in the Q331K knock-in mice (Supplementary Fig. S6A, B), suggesting a similar effect of this mutation as that incurred by loss of TDP-43 function. Critically, in neurons expressing shTDP_1, splicing of MAPT mRNA, the gene encoding the key neurodegenerative molecule Tau, was also significantly changed to increase the abundance of isoforms with exon 3 (Supplementary Fig. S6C, D; Supplementary Table S1). This altered splicing of Tau was noteworthy because it could potentially affect the aggregation property of Tau [47,48]. This analysis revealed that 180 genes were significantly upregulated and 164 genes were downregulated in neurons expressing shTDP_1 (vs. shLuc control; Fig. 6A; Supplementary Table S2). Gene Set Enrichment Analysis (GSEA) revealed specific enrichment in downregulated genes in the Ribosome pathway (Fig. 6B, top), but not in upregulated genes (Fig. 6B, bottom). Interestingly, similar enrichment was reported in the genes altered in cortical samples of ALS patients with TDP-43 pathologies, suggesting loss of TDP-43 function contributes to the gene expression changes associated with translation machineries in these ALS patients [49]. Enrichment plots also showed that the genes involved in the Ribosome pathway had a strong tendency toward downregulation in neurons expressing shTDP_1 (Fig. 6C; 86% downregulated vs 14% upregulated), although the majority of these genes were not significantly downregulated at the individual level. These genes included major components of ribosomal subunits, such as RPL31, RPS9, RPL15, RPL3, RPS2, and RPS6, etc. These results suggested that TDP-43 knockdown could potentially impair protein translation, consistent with reports by others [50][51][52][53], and implying that loss of TDP-43 in neurons might alter translation machineries by regulating expression of key ribosomal proteins.

[27] 158w To investigate how proteins necessary for synaptic function are regulated by TDP-43, we subsequently assessed the expression levels of a group of glutamate receptors. Interestingly, we found that none of these receptors were transcriptionally affected (Fig. 6D), although three NMDA receptor subunits (GluN2A, GluN2B, and GluN1) were found to have reduced protein levels in neurons expressing shTDP_1 or shTDP_2 (Fig. 6E). In addition to biogenesis and metabolism of mRNA, TDP-43 also modulates protein translation [48,[50][51][52]. GluN1 and GluN2A mRNA were both previously shown to bind to TDP-43 protein [4], supporting direct regulation of these transcripts by TDP-43. While we found that neither the mRNA levels nor the splicing patterns of GluN2A and GluN2B were affected by TDP-43 knockdown in our RNAseq analysis (Fig. 6A, Supplementary Fig. S5A, B), also supporting that their expression was not impacted at the mRNA level. Taken together, these analyses support a potentially critical role for TDP-43 in maintaining the translation of these NMDARs.

[28] 177w Importantly, co-expression of GluN2A, but not GluN2B, significantly reversed the dendritic spine loss caused by shTDP_1-or shTDP_2-mediated TDP-43 suppression (Fig. 6F, G; Supplementary Fig. S7A, B), consistent with the established role of GluN2A as the main synaptic NMDAR in mature neurons [54][55][56] (Successful expression of GluN2A-HA or GluN2B-Flag in neurons was confirmed by immunostaining; Supplementary Fig. S7C). In addition, expression of GluN2A or 2B alone did not alter the density of dendritic spines (Fig. 6H, quantified in I). Deletion of the C-terminal domain of GluN2A (GluN2AΔCTD) or replacing it with the CTD of GluN2B (GluN2A2B) impaired the ability of GluN2A to rescue spine loss induced by shTDP; whereas GluN2B with a replaced CTD from GluN2A significantly rescued the spine loss induced by shTDP (Supplementary Fig. S7D, quantified in E). These data supported that the CTD of GluN2A, not its channel property, played an essential role in maintaining dendritic spines. Combined with our RNAseq data, these results showed that silencing of TDP-43 induces a reduction in GluN2A expression, which at least partially contributes to the loss of spines.

[1] 350w TDP-43 inclusions were originally discovered as hallmarks of ALS/ FTLD in the spinal cord as well as in other brain regions, such as the hippocampus and cortex [6]. Interestingly, TDP-43 pathology is also associated with HS and increased severity of hippocampal degeneration in AD [19,20,57,58]. Many mutations of TDP-43 (A90V, P112H, T115A, C173S, A315T, Q331K, M337V, and Y374X, etc.) were associated with high risks of ALS/FTLD. Among them, A90V is also associated with an increased risk of AD [21,22]. These previous findings suggest a critical role for TDP-43 in hippocampal degeneration. However, it remains largely unclear how TDP-43 regulates hippocampal function. Currently, three predominant hypotheses have been proposed to interpret how TDP-43 pathology can change its function and consequently induce neurodegenerative effects: gain of normal function, gain of toxic function, or loss of function [2,12,13]. Transgenic expression of M337V leads to severe degeneration in the hippocampus and cortex, suggesting a gain of toxic function by the M337V mutant [23]. However, conditional TDP-43 KO in the forebrain excitatory neurons can also lead to hippocampal degeneration, supporting the likelihood that the loss of TDP-43 function can also introduce severe hippocampal damage [15]. These paradoxically similar effects induced by either gain or loss of function could be explained by our discovery that M337V is a dominant negative mutant in regulating hippocampal synapses. It is therefore plausible that M337V-induced loss of TDP-43 function also underlies hippocampal degeneration in the M337V transgenic mice. Different to M337V which located in the GRD domain, mutation A90V is located between the bipartite nuclear localization signal sequence of TDP-43. A90V has been reported to disturb the nuclear localization of TDP-43 in varies types of cells [22,59,60], and expression of TDP-43 containing A90V mutation led to sequestration of endogenous TDP-43 in the insoluble cytoplasmic aggregates [59]. These results, together with our discoveris, could provide a potential explanation of how A90V disturbs the function of endogenous TDP-43 in a dominant negative manner. Besides M337V and A90V, most of the tested disease-associated TDP-43 mutants are only loss of function (partial or full) mutations without dominant negative activity in maintaining hippocampal dendritic spines.

[2] 140w In our work, transient overexpression of WT TDP-43 did not lead to inclusion formation nor does damage hippocampal synapses, indicating that gain of TDP-43 normal function does not lead to hippocampal synaptic loss. This finding is consistent with a previous report that WT TDP-43 transgenic mice do not exhibit as severe a hippocampal degeneration, in contrast to that observed in M337V transgenic mice, regardless of their comparable TDP-43 expression levels [23]. Transgenic mice expressing other diseaseassociated TDP-43 mutants (such as A315T or Q331K) also do not show obvious hippocampal degeneration phenotypes [24,25], in line with our results which strongly suggested that these TDP-43 mutants are only loss of function mutations without dominant negative activity in maintaining hippocampal dendritic spines. It warrants mention, here, that abnormal TDP-43 inclusions were not formed in our studies (Figure S1A), so it therefore remains unclear

[3] 91w A shTDP_1 vs shLuc C D GluN2A GluN2B GluN1 Tuj1 TDP-43 **** 0.0 0.5 1.0 1.5 GFP only + GluN2A shTDP_1 + GFP + GluN2B + GluN2A + GluN2B + filler 0.2 0.3 0.4 0.5 0.0 0.1 F I G E B shTDP_1 vs shLuc Enrichment plot: KEGG_RIBOSOME Enrichment profile Hits -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 NES=-2.3597806 p<0.0001 FDR<0.0001 0.0 0.1 0.2 0.3 0.4 shTDP_1 + GFP ** Tardbp 0 10 20 30 -6 -3 0 3 log2FoldChange Up Down Not Sig Downregulated pathways Pathway NES NOM p-val

[4] 64w KEGG_RIBOSOME -2.3597806 <0.0001 KEGG_MISMATCH_REPAIR -1.8769225 <0.0023 KEGG_TYROSINE_METABOLISM -1.7282209 <0.0048 Upregulated pathways Pathway NES KEGG_RIG_I_LIKE_RECEPTOR_ SIGNALING_PATHWAY 2.0039675 <0.0001 KEGG_CYTOSOLIC_DNA_ SENSING_ PATHWAY 1.8297424 <0.0001 KEGG_NEUROACTIVE_LIGAND_ RECEPTOR _INTERACTION 1.7479858 <0.0001 KEGG_NOD_LIKE_RECEPTOR_ SIGNALING_PATHWAY 1.5843012 0.0236 KEGG_TOLL_LIKE_RECEPTOR_ SIGNALING _PATHWAY 1.5388591 0.0068 NOM p-val N=180 N=164 17 15 17 * * 19 19 19 p=0.0044 how a gain of toxic function due to abnormally aggregated TDP-43 impacts hippocampal synaptic function.

[5] 194w M337V is an aggressive TDP-43 mutant that co-segregates with early onset familial ALS [61,62]. Based on our results, we propose that M337V functions in a dominant negative manner resulting in hippocampal synaptic damage. This hypothesis predicts that M337V carriers should also develop hippocampal dysfunctions, although these associated dysfunctions have not yet been reported. One possible explanation is that human spinal cord motor neurons are more sensitive to M337V, and thus M337V carriers die of motor deficits before development of hippocampal dysfunctions. TDP-43 germline KO mice are embryonic lethal, while M337V KI mice (in the mouse TARDBP gene) develop normally [11,63,64]. This suggests that TDP-43 function during embryonic development is different from its function in maintaining mature hippocampal synapses. Surprisingly, mouse M337V TDP-43 loses its dominant negative activity in reducing hippocampal dendritic spines (Supplementary Fig. S4F), although human and mouse TARDBP genes are highly conserved around the M337V mutation site. This loss of dominant negative activity may explain why no neurodegeneration was observed in heterozygous M337V KI mice and further reminds us that homogenous mutations in the rodent TARDBP gene might result in considerably different effects from their human counterparts, even in highly conserved domains.

[6] 170w Interestingly, Q331K KI mice were shown to develop cognitive deficits without obvious motor defects [48], which is consistent with our finding that Q331K loses its activity partially in maintaining hippocampal synapses. Moreover, these findings together indicate that Q331K impacts motor neurons via a different mechanism than in cortical and hippocampal neurons. It may be germane to this point that our RNASeq of silenced TDP-43 neurons, as well as previous reports in Q331K mutant knock-in mice hippocampus both found altered splicing isoforms of Mapt, another pivotal dementia-associated gene encoding Tau [48]. This was interpreted as a gain of TDP-43 function by the Q331K mutant in the original publication. Surprisingly, we found a significant enhancement specifically in the inclusion of Mapt exon 3 in hippocampal neurons suppressed for TDP-43 (Supplementary Fig. S6C, D). The fact that the loss of TDP-43 is sufficient to cause the same alteration in Mapt mRNA splicing as Q331K further supports that loss of TDP-43 function underlies the pathological changes incurred in the hippocampus of Q331K KI mice.

[7] 138w Given the discovery that shTDP-43 downregulated mRNA of key ribosomal proteins, it is also possible that TDP-43 controls NMDARs translation by regulating ribosome. Loss of TDP-43 leads to reduction of their translation. Among them, loss of GluN2A as the main synaptic NMDAR contributes to the loss of dendritic spines at least partially. In contrast, loss of GluN2B does not contribute the spine loss. This is likely due to the extrasynaptic distribution or the relative abundance of GluN2B in mature neurons. Analysis of chimeric receptors supports that the C-terminal intercellular domain of GluN2A is critical for its role in maintaining dendritic spines, whereas its channel property is not. However, it remains to be further examined which function of CTD is essential. The impacts of CTD in the receptor's spatial distribution, downstream signaling or interacting proteins could all contribute [65][66][67][68][69].

[8] 108w In summary, we show that physiological TDP-43 function is necessary for maintaining mature hippocampal synapses, while the loss of TDP-43 is sufficient to cause hippocampal dysfunction at both physiological and behavioral levels. In addition, we found that the vast majority of ALS/FTLD/AD-associated TDP-43 mutants are loss of function mutants in the maintenance of hippocampal synapses. Together, these results strongly argue that loss of TDP-43 function contributes to the neurodegeneration associated with TDP-43 proteinopathies. Importantly, NMDARs partially mediated the TDP-43-silencing-induced synaptic damage. Further investigation of NMDARs [70,71] and their translation machineries as potential drug targets to reverse TDP-43 pathology-associated synaptic degeneration may provide valuable and desperatelyneeded clinical treatment strategies.

[1] 109w The following antibodies were used in this study: rabbit TDP-43 antibody (Proteintech, 10782-2-AP), chicken MAP2 antibody (Synaptic Systems, 188006), rabbit β-Tubulin III antibody (Sigma, T 2200), rabbit HA-Tag antibody (CST, 3724S), mouse Flag-Tag antibody (Sigma, F1804), chicken GFP antibody (Abcam, ab13970), rabbit GluN2A antibody (NOVUS, NB300-105), mouse GluN2B antibody (Millipore, 05-920), mouse GluN1 antibody (Millipore, 05-432), mouse GAPDH antibody (Proteintech, 60004-1). Peroxidase-Conjugated secondary antibodies: Goat Anti-Rabbit IgG (YESEN, 33101ES60), Goat Anti-Mouse IgG (YESEN, 33201ES60). Fluorescent secondary antibodies: Alexa 488 Goat anti-chicken IgG (Thermo Fisher, A11039), Alexa 568 Goat anti-chicken IgY (Thermo Fisher, A11041), Alexa 568 Goat anti-rabbit IgG (Thermo Fisher, A11036), Alexa 647 Goat anti-Mouse IgG (Thermo Fisher, A21236).

[2] 221w The expression cassettes of HA or Flag tagged wild-type human/rat/mouse TDP-43, domain deletion mutants, disease-associated point mutations, and GFP were sub-cloned into pCAGGS mammalian expression vector [35]. The virus constructs of TDP-43-HA, A90V-HA, M337V-HA and GFP were sub-cloned into pFHTrePW vector backbone under the control of Tet Response Element (TRE 3G). pFHTrePW is modified from pFHSynPW (Provided by Dr. Carlos Lois, MIT) by switching the human Synapsin I promoter to TRE 3G promoter [35]. rtTA for tet-off system was expressed by pFHSynPW under the control of human synapsin I promoter. Plasmids were packaged into lentivirus to acquire high infection efficiency in primary neurons. For shRNA constructs used in hippocampal slice cultures and HEK293T cells, the oligonucleotides were inserted between HindIII and BglII sites of the pSuper vector (target sequences, Rat TDP-43 #shTDP_1: 5'-GCTGATGGGCTGCGAACAT-3'; Rat TDP-43 #shTDP_2: 5'-GTAGATGTCTTCATTCCCAAA-3'; Firefly luciferase shRNA #shLuc: 5'-CGTACGCGGAATACTTCGA-3'). For shRNA lentivirus used in neuronal cultures and in vivo, shRNAs were expressed by a modified dual-expression plasmid system which contains the pFHSynPW backbone with H1 promoter driving shRNA and human synapsin I promoter driving GFP [35,36]. For the AAV constructs, shRNA expression was driven by a H1 promoter and EGFP expression was driven by a CAG promoter. shLuc (sequence: CGCTGAGTACTTCGAAATGTC) served as the negative control [35]. AAV packaging service was provided by a third party company (Taitool Bioscience).

[3] 219w Organotypic slice cultures were prepared from P7-8 Sprague-Dawley (SD) rats from both genders, which were biolistically transfected with a gene gun (Biorad) at DIV3-DIV4. At DIV10, Live-cell images of transfected neurons were captured by fluorescence confocal microscope [35]. In overexpress experiments, 5 mg gold particles (1.6 μm in diameter; Bio-Rad) were coated with 15 μg of cDNA plus 5 μg of GFP. In knockdown experiments, 5 mg gold particles were coated with 25 μg of shRNA plasmids plus 5 μg of GFP. In rescue experiments, 5 mg gold particles were coated with 30 ug of shRNA plasmids plus15 μg of cDNA or empty vector plasmid (filler) and 5 μg of GFP. In the PSD-95 and F-actin imaging experiments, 5 mg gold particles were coated with 2.5 μg of PSD-95-GFP and 2.5 μg of mKate2-β-actin plus 25 μg of shRNAs. Morphologically healthy neurons expressing GFP/mKate2 from the transfected slices were imaged by a Nikon confocal microscopy system using a water-immersion 60X objective (NA = 1.0) with Z-axial stacks (step size = 0.8 μm). Protrusions longer than 0.4 μm of the secondary basal dendrites were counted as spines [35]. For spine density analysis, 3-5 different dendrites were imaged from each pyramidal neuron; and spine densities were then analyzed by ImageJ. Data acquisitions and analyses were all performed in a blinded manner.

[4] 85w Hippocampal neurons were prepared from embryonic day 17 (E17) SD rats or day 16 (E16) wild-type mice (C57BL/6) [35]. The dissociated cells were cultivated in Neurobasal media (Thermo Fisher Scientific) containing 2% B27 (Life technology), 0.25% glutamax (Thermo Fisher Scientific) and 100 units/mL Penicillin/Streptomycin (Thermo Fisher Scientific). Neurons were cultured at 37 °C with 5% CO 2 , and fresh media was added once a week for 7-15 days before infected by lentivirus or transfected with Lipofectamine 2000 (Thermo Fisher Scientific) following the manufacturer's protocol.

[5] 120w Culture medium in 24-well plates was removed and the cells were rinsed once with ice-cold PBS before lysed with (100 μl per well for neurons, 400 μl per well for HEK cells) western blot sample buffer (100 mM Tris-Hcl, 4% SDS, 20% Glycerol). Lysates were shaken for 10 min at room temperature, followed by 10 min of boiling at 100 °C and 5 min of centrifugation at 12,000 rpm. Proteins were separated with 8-12% Trisglycine SDS-PAGE and transferred to Nitrocellulose Membranes (Millipore, 0.22 μm). Primary antibodies were incubated in blocking buffer overnight at 4 °C, and secondary antibodies in RT for 2 h. The signal was detected using an Amersham Imager 600 (GE Healthcare) and densitometry was measured using ImageJ.

[6] 188w For immunofluorescent staining, neurons were washed with PBS (plus 1 mM CaCl 2 and 5 mM MgCl 2 ) once, HEK293T cells were washed with PBS, followed by 10 min of fixation with 4% PFA/4% sucrose in 1×PBS at room temperature. Next, cells were washed 3X with 1×PBS and permeabilized with 0.15% Triton-X in 1×PBS for 15 min, then blocked with 5% albumin bovine V in PBS for 30 min at room temperature. Cells were incubated with primary antibodies in blocking buffer at 4 °C overnight. On the next day, samples were washed 1X with PBS, 1X with 0.5 M NaCl in PBS and 2X with PBS, followed by secondary antibodies incubation in blocking buffer at room temperature for 2 h; then samples were washed 3X with PBS and mounted on glass slides using Mounting Medium with DAPI (Solarbio, S2110). Fluorescent images were captured by Leica TCS SP8 (60× water lens, NA = 1.2) and Nikon confocal microscopy system (63× oil lens, NA = 1.4) with properly adjusted settings and internal controls to ensure the homogeneity of imaging. All images were processed and analyzed by ImageJ and Imaris.

[7] 100w After infected by lentivirus for 5-6 days, total RNA sample was isolated with TRIZOL (Thermo Fisher) from neurons according to the user guide. Following the DNase (Promega) treatment to remove the genomic DNA, real-time PCR was performed on the QuantStudio™ 6 Flex Real-Time PCR system (Life Technologies) using a one-step protocol with the SYBR Green qPCR Master Mix (QIAGEN). GAPDH mRNA level was used as an internal control to normalize the mRNA level. The primers used for amplification of gene encoding rat TDP-43 was Tardbp-F: 5'-CGACTGGTGGAAG-GAATTCTGC-3'; Tardbp-R: 5'-TCACTTTCACTGCAGAGGAAGC-3'. All RT-qPCR primers for glutamate receptors and GAPDH were purchased from Qiagen.

[8] 59w mRNA profiling with RNA-sequencing After 15 days in vitro, dissociated rat hippocampal neurons were infected with lentivirus expressing either shLuc or shTDP and allowed to grow for another 6 days for shRNA expression. Total RNA was then isolated using TRIZOL (Thermo Fisher). Samples passing an mRNA quality check proceeded to quantitative Genome-wide gene expression profiling analysis using RNA-sequencing (RNAseq).

[9] 50w For rat hippocampal injection, 7-8 weeks old rats were injected with shRNA lentivirus (ML: ± 4.3 mm; AP: -4 mm; DV: -5 mm). 10 days later, the animals were sacrificed, perfused with PBS, then fixed with 4% PFA/4% sucrose in PBS and analyzed by immunofluorescence staining with an anti-GFP antibody.

[10] 67w Mice at 7-8 weeks age were injected with shRNA encoded by lentivirus or AAV in hippocampus (ML: ± 1.68 mm; AP: -2 mm; DV: -1.67 mm), or cortex (ML: ± 1 mm; AP: -1.67 mm; DV: -1.2 mm). After 10-20 days, animals were subjected to spine density analysis by immunostaining, electrophysiology studies, or behavior tests before sacrificed and analyzed to verify the injection accuracy and infection efficiency.

[11] 155w Patch-clamp recordings were performed from cortical pyramidal cells in acute brain slice. Recording external solution used artificial cerebrospinal fluid (ACSF) consisting of (in mM): 127 NaCl, 2.5 KCl, 12.5 NaH 2 PO 4 , 25 NaHCO 3 , 25 D-glucose, 2.5 CaCl 2 , 1.3 MgCl 2 , aerated with 95% O 2 /5% CO 2 to maintain pH around 7.25. Pipettes (TW150F-4, World Precision Instruments) with resistance of 4-6 mΩ were filled with an internal solution consisting of (in mM): 115 cesium methanesulfonate, 20 CsCl, 10 HEPES, 2.5 MgCl 2 , 4 Na 2 ATP, 0.4 Na 3 GTP, 10 Na phosphocreatine and 0.6 EGTA (pH 7.25). mEPSCs were recorded at -70 mV holding potential in the presence of 1 uM tetrodotoxin (TTX, MedChem express, HY-12526A) and 100 uM picrotoxin (PTX, Sigma, P1675) in the external solution. The events were then picked and analyzed by MiniAnalysis software (Synaptosoft) with amplitude threshold of 5 pA.

[12] 14w Animals were habituated for 1 h in the room before all the behavioral tests.

[13] 19w The tests were performed in the sequence as they were described below and were analyzed in a blind manner.

[14] 75w Open field test: The open field test used 40 cm × 40 cm boxes without roofs. Mouse activities were recorded using the Ethovision videotracking system (Noldus Information Technology Inc., Leesburg, VA, USA). The field was digitally divided into center area (24 × 24 cm) and whole arena, data were collected continually for 1 h. The travel distance, velocity, the rearing time, and the time spent in the inner area were all recorded and measured automatically.

[15] 63w Y maze: To investigate their spatial working memory, mice were placed in a Y maze (with arms 30 cm in length) and allowed to explore the maze freely for 8 min. Each mouse was placed at the same end of one arm and then video recorded throughout the whole process. The number of alternations and entries were analyzed using the Ethovision videotracking system.

[16] 78w Elevated plus maze: The test used an elevated (0.5 m), plus-shaped (+) apparatus consists of two open arms and two closed arms with 15 cm walls and open roofs. For testing, the subjects were individually placed in the center of the maze and facing to the open arm, and then allowed to explore the apparatus for 5 min freely. Distance and time spent in the open arms, center and closed arms were measured by the Ethovision videotracking system.

[17] 151w Fear conditioning: For the contextual fear conditioning test, mice were trained on the first day following this procedure: the mice were placed in the conditioning boxes and allowed to freely explore the arena for 3 min before they were presented with an 80 dB, 2 kHz tone for 30 s (conditioned stimulus, CS). During the last 2 s of the tone, an inescapable 0.3 mA foot shock (unconditioned stimulus, US) was delivered. This procedure was repeated three times with a 30-s inter-stimulus-interval. Then mice were taken back to their home cages. To evaluate their learned aversion for the shock associated environment, on the second day, the mice were placed in the same conditioning box in which they were trained on the previous day for 5 min, and their freezing behavior was recorded using a visual camera in the absence of tone or foot shock, and analyzed with the Ethovision videotracking system.

[18] 77w For the cued fear conditioning test, the same trained animals after contextual fear conditioning test were placed in a new context different from their original training chamber (color, lighting, and odor [lemon essence]) on the third day. Their freezing behavior was measured during the first 3 min free exploration of the new context and another 3 min after the tone stimulus (2 kHz, 80 dB). The freezing behavior was measured and analyzed with the Ethovision videotracking system.

[19] 99w T Tail suspension test: Each mouse was individually suspended by its tail with tape in a three-walled rectangular compartment (30 cm height × 15 cm width × 15 cm depth). The mice dangled in the air, facing downward. Mice intended to struggle to face upward or climb to a solid subject initially, but would slowly stop struggling and hung still later on. The animals were recorded with a video camera from the side and their immobile time (cessation of limb movements more than 2 s) during the last 4 min of a 6 min session was quantified for analysis.

[20] 81w Forced swim test: Each mouse was individually placed in a transparent cylindrical glass container (14 cm in diameter, 30 cm tall) filled halfway with water (23-25 °C) and swam for 6 min. The animals were recorded with a video camera in front of the container. The immobile time during the last 4-min recording was quantified for further analysis. When the mice remained floating or only had limb movements to keep balance for more than 2 s, they were counted as immobile.

[21] 59w Unless otherwise noted in the figure legends, the statistical significance in this study is determined by unpaired, two-tailed Student's t test. Error bars represent the standard error of the mean (SEM). All animal procedures were reviewed and approved by the Institutional Animal Care and Use Committee of Interdisciplinary Research Center on Biology and Chemistry, Shanghai Institute of Organic Chemistry.