Knockdown of astrocytic Grin2a aggravates β‐amyloid‐induced memory and cognitive deficits through regulating nerve growth factor

Abstract Synapse degeneration correlates strongly with cognitive impairments in Alzheimer's disease (AD) patients. Soluble Amyloid‐beta (Aβ) oligomers are thought as the major trigger of synaptic malfunctions. Our earlier studies have demonstrated that Aβ oligomers interfere with synaptic function through N‐methyl‐D‐aspartate receptors (NMDARs). Our recent in vitro study found the neuroprotective role of astrocytic GluN2A in the promotion of synapse survival and identified nerve growth factor (NGF) derived from astrocytes, as a likely mediator of astrocytic GluN2A buffering against Aβ synaptotoxicity. Our present in vivo study focused on exploring the precise mechanism of astrocytic GluN2A influencing Aβ synaptotoxicity through regulating NGF. We generated an adeno‐associated virus (AAV) expressing an astrocytic promoter (GfaABC1D) shRNA targeted to Grin2a (the gene encoding GluN2A) to perform astrocyte‐specific Grin2a knockdown in the hippocampal dentate gyrus, after 3 weeks of virus vector expression, Aβ were bilaterally injected into the intracerebral ventricle. Our results showed that astrocyte‐specific knockdown of Grin2a and Aβ application both significantly impaired spatial memory and cognition, which associated with the reduced synaptic proteins PSD95, synaptophysin and compensatory increased NGF. The reduced astrocytic GluN2A can counteract Aβ‐induced compensatory protective increase of NGF through regulating pNF‐κB, Furin and VAMP3, which modulating the synthesis, mature and secretion of NGF respectively. Our present data reveal, for the first time, a novel mechanism of astrocytic GluN2A in exerting protective effects on synapses at the early stage of Aβ exposure, which may contribute to establish new targets for AD prevention and early therapy.


| INTRODUC TI ON
Alzheimer's disease (AD) is characterized by progressive deterioration of memory and other cognitive functions. Synapse degeneration, as an early event in AD pathogenesis (Selkoe, 2002), correlates strongly with cognitive impairments in AD patients (Müller et al., 2017;Wang & Mandelkow, 2016). Emerging evidence implicates soluble Amyloid-beta (Aβ) oligomers as the major trigger of synaptic malfunction and loss in AD patients brain.
Our own earlier studies and others evidence suggest that soluble Aβ oligomers interfere with synaptic function by acting as gain-of-function ligands that bind near, or to, ionotropic glutamate N-methyl-D-aspartate receptors (NMDARs), which play a central role in mediating both physiological synaptic plasticity and glutamate-induced neurotoxicity Ferreira et al., 2015;Liu et al., 2010;Wang et al., 2013). In contrast to earlier views that NMDARs are exclusively expressed in neurons, recent two decades work demonstrates that fully functional NMDARs are present on astrocytes (Cahoy et al., 2008;Conti et al., 1996Conti et al., , 1999Lee et al., 2010), although the precise function of astrocytic NMDARs remains enigmatic.
NMDARs, which are encoded by the Grin gene family, exist as hetero-tetrameric complexes consisting of two obligatory GluN1 along with two GluN2(A-D) and/or GluN3 (A-B) subunits (Bai et al., 2013;Dumas, 2005;Myers et al., 2019). Our previous in vitro studies have found the neuroprotective role of astrocytic GluN2A subunit in the promotion of synapse survival, and identified nerve growth factor (NGF), a neurotrophin derived from astrocytes, as a likely mediator of astrocytic GluN2A buffering against Aβ synaptotoxicity . NGF, as the first discovered and best characterized neurotrophin, its beneficial effects during early stage of AD have been verified (Guo & Mattson, 2000;Mattson, 2008;Schulte-Herbruggen et al., 2007). NGF composes of α, β and γ three subunits, among them β subunit (β-NGF) is responsible for its biological activity (Shooter, 2001;Young et al., 1988). Several researches have demonstrated that β-NGF can protect synapses against Aβ-induced neurotoxicity (Canu et al., 2017;Cuello et al., 2019). However, the exact nature of astrocytic GluN2A-NGF interaction and the mechanisms involved in Aβ-induced synaptotoxicity remain unclear. To clarify the related mechanisms of astrocytic GluN2A which protect neurons against the early Aβ synaptotoxicity through regulating NGF could provide a pharmacological strategies aiming to increase an endogenous neuroprotective NGF level for early treatment of AD.
The in vivo study reported here focused on whether and how astrocytic GluN2A subunit can influence the synaptotoxic effects of Aβ through regulating astrocytic NGF. Our results show that astrocyte-specific knockdown of Grin2a (the gene encoding GluN2A) in the rat hippocampus can counteract Aβ-induced compensatory protective increase of β-NGF through modulating pNF-κB, Furin and VAMP3, which affects the synthesis, mature and secretion of NGF respectively. Our in vivo findings reveal, for the first time, a novel mechanism of astrocytic GluN2A exerting protective effects on synapses at the early stage of exposure to Aβ oligomers.

Aβ-induced memory and cognitive impairments
Our previous initial study has demonstrated that there were higher normal level and significant Aβ-induced increase of astrocytic GluN2A in the neuron-astrocyte mixed cultures compared to that in pure astrocytic cultures , which implies that astrocytic GluN2A may play an enigmatic role in the complex cross talk of neuron-astrocyte, including the neuronal response to Aβ. We also found that astrocytic GluN2A promotes synapse survival likely through NGF derived from astrocytes buffering against Aβ 1-40 synaptotoxicity . In our recent in vitro experiments, with antiβ-NGF antibody to deplete β-NGF in medium of neuronastrocyte co-cultures, we have further confirmed astrocytic NMDA receptor protects neurons against Aβ 1-42 -induced synaptotoxicity through NGF secreted by astrocytes (data not shown). But the precise mechanism of astrocytic GluN2A in the execution of the synaptoprotective effects, especially in AD rat model, remains undefined.
Our previous in vitro findings prompt us to further investigate the potential mechanism of astrocytic GluN2A in rat AD model.
Initially, in order to reduce GluN2A expression in astrocytes of rat hippocampus, we established characterized Grin2a shRNA adeno-associated viral (AAV) vector equipped with the promoter GfaABC1D, which allows astrocyte-specific knockdown of Grin2a. F I G U R E 1 Knockdown of rat astrocytic Grin2a aggravates Aβ-induced memory and cognitive impairments. (a) Schematic diagram of the experimental design. (b, c) Knockdown efficacy of astrocyte-specific Grin2a was verified with confocal imaging. The fluorescence distribution along the linescan profile depicting the spatial distribution of GluN2A and GFAP. The subregion images (a and b, indicated by white rectangle on the left images) are magnified for clarity. White arrows indicating the GluN2A positive astrocytes, and white arrowhead indicating GluN2A positive neurons. Bar, 100 μm, n = 3 rats/group. (d) Representative western blot band and quantification showing the relative expression of GluN2A after AAVs injection. (e) The relative expression of GluN2A at mRNA level after AAVs injection. (f-i) Barnes maze test showing astrocytic Grin2a knockdown aggravates Aβ-induced memory and cognitive impairments. The representative locomotor traces of the four groups rats in the probe trial (f), latency to the target holes (g), average locomotor speed (h) during the training days and the latency to the target hole in the probe trial (i). (j, k) Novel objective recognition (NOR) test showing astrocytic Grin2a knockdown aggravates Aβ-induced memory and cognitive impairments. The representative locomotor traces (j) and preference index (k) of the four groups rats. preference index is calculated as the time exploring the novel object divided by total time spent in exploring the familiar and the novel objects in the probe session. FO, familiar object; NO, novel object. Numerical data are shown as means ± SEM. n = 5-8 rats/group, * vs. vehicle, *p < 0.05, **p < 0.01, ***p < 0.01, # vs. Aβ, # p < 0.05, ## p < 0.01 One control shRNA virus pAAV2/9-GfaABC1D-EGFP-3FLAG, and three (1#-3#) Grin2a-shRNA viruses (pAAV2/9-GfaABC1D-EGFP-3FLAG-micro30 shRNA(Grin2a)) with EGFP after expression were produced. AAVs were bilaterally injected into the rat hippocampal dentate gyrus. After 21 days of virus expression and the behavior tests, the rats were sacrificed and the knockdown efficacy of Grin2a was confirmed with immunofluorescent staining (nearly 100%) ( Figure 1), the reduced GluN2A at the protein and mRNA levels were also demonstrated ( Figure 1).
Barnes maze test was used to evaluate the spatial learning and memory. After the first day's habituation, the rats were trained from day 2-5. At the third day of the training phase, both Aβ and shGrin2a + Aβ rats showed significantly increased latencies to locate the target box (Figure 1f,g,i), which could not be explained by reduced locomotor activity, as there were no differences among the rats in the average locomotor speed during the training days (Figure 1h). During the probe test of the seventh day, not only Aβ and shGrin2a+Aβ rats, shGrin2a rats also showed significantly increased latencies to enter the target hole, moreover shGrin2a + Aβ rats showed a stronger increase of latency (Figure 1i) (Aβ: 22.33 ± 1.60 s; shGrin2a + Aβ: 77.51 ± 16.27 s versus vehicle: 5.81 ± 1.40 s, p < 0.001; shGrin2a: 21.18 ± 4.5 s versus vehicle, p < 0.01). These results indicated Aβ exposure or astrocyte-specific Grin2a knockdown can impair spatial learning and memory, and astrocytic Grin2a knockdown further aggravates Aβ-induced spatial learning and memory decline.
In addition, Fluoro-Jade C staining (Schmued et al., 2005) results showed there was no positive staining neurons (data not shown), which indicating no significant neurodegenerative neurons in the hippocampus. The above data suggest that astrocytic Grin2a knockdown exacerbate Aβ-induced synaptotoxicity.

| Knockdown of astrocytic Grin2a inhibits Aβinduced increase of pNF-κB
To further investigate the mechanism of GluN2A regulating Aβinduced increase of astrocytic β-NGF, we next examined several transcription factors that may affect the synthesis of NGF. As shown in Figure 4a

| Knockdown of astrocytic Grin2a inhibits Aβinduced increase of Furin
Furin is an important proprotein convertase which can cleave proNGF and convert it into mature NGF, which is the main form of neuroprotective action. In order to explore whether knockdown of astrocytic Grin2a inhibits Aβ-induced β-NGF elevation through regulating NGF maturation, we observed the expression of Furin with immunofluorescence and western blot methods. As shown in Figure

| DISCUSS ION
The data presented here demonstrate that reduction of astrocytic GluN2A in the hippocampus of AD model rat aggravates Aβ oligomer-induced spatial memory and cognition impairments through regulating astrocytic NGF synthesis, mature and secretion.
To the best of our knowledge, our present data are the only in vivo findings published so far through the use of AD rats with a conditional knockdown of astrocytic Grin2a to demonstrate the involvement of the astrocytic NMDA receptor subunit in Aβ-induced early synaptotoxicity (Skowrońska et al., 2019). Our research shed light on a novel mechanism of astrocytic GluN2A in exerting protective effects on synapses at the early stage of Aβ synaptotoxicity, which could contribute to establish new targets for AD prevention and early therapy.
NGF is an essential mediator of synaptic plasticity. The accumulation of proNGF and an enhanced degradation of the mature NGF has been found in brains of AD patients, which indicated the dysregulation of NGF metabolism (Cuello et al., 2019). NGF addition has been demonstrated to improve cognitive decline in AD (Cuello et al., 2019;Siegel & Chauhan, 2000). Under physiological conditions, NGF is thought to be mainly produced by neurons (Balkowiec & Katz, 2000;Lewin & Barde, 1996;Miklic et al., 2004), while under pathological conditions, astrocytes become the major source of NGF production (Bakhit et al., 1991;Goss et al., 1998). An increased β-NGF release in astrocytes with Aβ exposure has been described previously by our team and others (Bakhit et al., 1991;Goss et al., 1998;Li et al., 2016;Schulte-Herbruggen et al., 2007). Our current findings of increased β-NGF production in AD model rat might reflect an early endogenous autoprotective attempt of astrocytes to minimize Aβ-induced neuron loss, since there was significant synaptic dysfunction but no obvious neuronal degeneration under our experiment conditions. This view is supported by in vitro AD models and some clinical data available on NGF treatment in early stages of AD (Enciu et al., 2011;Iulita & Cuello, 2014;Schulte-Herbruggen et al., 2007).
Our results showed that astrocyte-specific knockdown of Grin2a significantly counteracted Aβ-induced compensatory elevation of β-NGF, with concomitant aggravated memory and cognitive decline. Together with our previous study , the present finding further confirmed astrocytic GluN2A could mediate neuroprotection through regulating NGF production in astrocytes.
One possible mechanism regulating NGF synthesis by astrocytic GluN2A may be mediated by NF-κB, since astrocytic Grin2a knockdown inhibited Aβ-induced activation of NF-κB. This result is supported by the evidence that sublethal concentrations of Aβ activate NF-κB through a mechanism involving NMDA receptors (Kawamoto et al., 2008). In addition, a feedback loop between NGF and NF-κB triggered by Aβ might occur. Through astocytic GluN2A, Aβ activated NF-κB, in turn increasing NGF that, when interacting with its receptor TrkA, led to further NF-κB activation and NGF increase (Chacón et al., 2010).
At the same time, under our experimental condition, the involvement of BDNF, another important neurotrophic factor related with synaptic efficacy, could not be excluded. As demonstrated by Miklic et al., (2004), astrocytes have ability to synthesize NGF and BDNF, and Aβ could increase BDNF by activating NF-κB through NMDA receptor, in which BDNF contributed to mediate a neuroprotective response to Aβ (Kawamoto et al., 2008). Our results showed that either Aβ treatment or knockdown of astrocytic Grin2a caused declined pCREB, which may indicate decreased BDNF production (Tong et al., 2001). Several reports suggest there is a cross talk between NF-κB and CREB, while BDNF is proposed as a common gene regulated by the two transcription factors (Caviedes et al., 2017;Kaltschmidt & Kaltschmidt, 2015). Whether astrocytic GluN2A can directly or indirectly regulate BDNF under Aβ neurotoxic effect and the precise mechanism need to be clarified in future.
Furin, as a calcium-dependent serine endoprotease (Yamada et al., 2018), can cleave proNGF and convert it into mature NGF to exert neuroprotective effects. Here our results show that the im- After cleaved by Furin, mature NGF can be delivered to the extracellular space through vesicle transport, fusion and exocytosis.
During this serial processes, VAMP3 and SNAP23 play important roles (Ravichandran et al., 1996;Schubert et al., 2011). Our results showed that astrocytic Grin2a knockdown strongly inhibited Aβ-induced increase of VAMP3 in astrocytes and aggaravated the decrease of total VAMP3 level, but did not show any effects on Aβinduced alterations of SNAP23. Our finding suggests that astrocytic GluN2A reduction has more critical effects on VAMP3, the sharply decreased VAMP3 disrupted NGF transport and exocytosis, ultimately impaired synaptic efficacy. It has been demonstrated that VAMP3 contributed to Ca 2+ -dependent trafficking of astrocytic vesicles (Schubert et al., 2011), and the mechanism of astrocytic Grin2a knockdown elicited VMAP3 decrease may be related to astrocytic NMDA receptor-coupled Ca 2+ release from the intracellular Ca 2+ stores (Jimenez-Blasco et al., 2015;Verkhratsky & Chvátal, 2020).
In the meantime, it should be noticed that Aβ and astrocytic Grin2a knockdown led to the decreased total levels of both VAMP3 and SNAP23, so the precise mechanism that how astrocytic GluN2A regulates VAMP3 and SNAP23 in different cells under Aβ exposure remains to be investigated.
There are some limitations to our research. We have examined some key factors associated with NGF synthesis, mature and secretion, but no further assessment for the related regulators necessary for proNGF cleavage and NGF degradation in the extracellular space upon Aβ stimulation. In addition, to sort astrocytes and neurons using fluorescence-activated cell sorting (FACS) is necessary for neuron-astrocyte communication research in the future. As a beneficial factor in the early stage of AD, if supplementary NGF to be performed in Grin2a knockdown AD rats to rescue the memory and cognition deficits will make this research more perfect.
In summary, our present work illustrates the complex nature of neuron-astrocyte communication. Specifically, our work shows that at the early stage of Aβ oligomers synaptotoxicity, the reduction of astrocytic GluN2A can exacerbate synaptotoxicity through regulating NGF synthesis, mature and secretion (Figure 7). This novel information contributes to establish early targeted interventions to delay development of AD.

| Animals
All Wistar rats used in this study were obtained from Beijing HFK Bioscience Co., Ltd (Beijing, China). All experimental procedures were approved by the Animal Ethics Committee of Capital Medical University. All possible efforts were made to minimize rats suffering and the number of rats used.

F I G U R E 7 A proposed model
for knockdown of astrocytic Grin2a aggravated Aβ-induced synaptotoxicity. The reduced astrocytic GluN2A can counteract Aβ-induced compensatory protective increase of β-NGF through modulating pNF-κB, Furin and VAMP3, which modulating the synthesis, mature and secretion of NGF respectively

| Virus Generation
The adeno-associated virus (AAV) shRNA knockdown vector system has become a powerful tool of modulating gene expression.

| Barnes maze test
The Barnes maze test was used for assessing spatial learning and memory. The modified protocol used here was previously described (Sadeghian et al., 2019). A high (90 cm to the floor) black circular platform (120 cm in diameter), containing 20 uniform holes in its periphery (with 19 false boxes and one escape box), was used in the present study. Prior to the first trial, rats were habituated to the escape box and the platform. An acquisition trial consisted of placing a rat in the starting box for 30 s. Then, the box was raised, and an aversive stimulus (bright light) was switched on, and the rat was allowed to explore the maze freely for 240 s. Starting from the second training, the maze was randomly rotated one to several holes, but the target box was always fixed in the same position. The rats were trained 2 times per day for 5 consecutive days. After the sixth day's rest, on the seventh day, the rats were submitted to a probe trial for 240 s on the maze without an escape box. The platform and holes were carefully cleaned with 70% ethanol after every trial. All trials were recorded with an overhead camera and tracked using Smart 3.0 video tracking system (Harvard Apparatus, Holliston, MA, USA).
The following behavioral parameters were recorded: the latency of reaching the target box/hole, the average speed and the number of errors, which is defined as the rat's head reaching or exploring any non-target hole.

| Novel object recognition (NOR)
The novel object recognition (NOR) test is based on the nature tendency of rats to show preference for novel objects. The present protocol was previously described (Zappa et al., 2018) with some little modification, three objects A, B and C were prepared, among which A and B are identical, and C are completely different from A and B in shape and color, but with similar dimension. After a 5 min habituation session in an empty arena, the rats were submitted to a training session, in which rats explored two identical objects (A and B) during 5min and the time spent exploring each object was recorded.
Objects A and B were placed equidistant from the center and walls of the arena. One hour after the training session, the rats were again placed in the arena for the probe session, one of the two familiar objects (A or B) was replaced with novel object C. Again, the time spent in exploring the familiar object (FO) and novel object (NO) was acquired and analyzed. The location of the novel object was counterbalanced among animals. In each trial, the rat was allowed to explore the objects freely. Exploration was considered effective when the orientation of the rat's nose was toward the object, within a range of ~1 cm or closer to the object. Climbing was not considered exploration. The objects and the floor were cleaned thoroughly with 70% ethanol between trials. All trials were recorded with Smart 3.0 video tracking system. The preference index (PI) was calculated as the time exploring the novel object divided by total time spent in exploring the familiar and the novel objects in the probe session.

| Immunofluorescence and imaging analysis
In brief, brain sections were fixed (ice cold 4% paraformaldehyde), washed, permeabilized (0.3% Triton X-100 in PBS) and blocked with appropriate non-immune sera. After that, specimens were in-

| Western blotting
In brief, total cellular proteins of the rats hippocampi were extracted, homogenized and subsequent brief sonication. Homogenates were centrifuged at 4℃ (30 min, 14000 rpm), the pellet was discarded and protein contents of lysates were determined using the BCA protein assay (Thermo). Total protein from each sample was boiled for AF-556-NA; R&D Systems), followed by incubation with goat antimouse or anti-rabbit IgG (H+L)-HRP, or rabbit anti-goat IgG (H+L)-HRP (1:5000, Absin) in TBST. Protein bands were acquired using ECL Plus, and analyzed in a luminescent image analyzer system (LAS-500, GE Healthcare). Densitometric quantification was carried out using Image J software. The results were expressed as the band densities of the target protein/β-actin or GAPDH ratio and then normalized to the values from the corresponding controls.

| Transmission electron microscopy and synapse morphometric analysis
The rats were deeply anesthetized and perfused transcardially with PBS for 1 min, then with a fixative composed of 0.25% glutaraldehyde and 4% paraformaldehyde in 0.1 M PB (pH 7.4). The hippocampus was isolated and cut into ~1 mm 3 cubes, post-fixed in 1% osmium tetroxide, dehydrated, embedded in Epon resin 812, before ultrathin sections (70 nm) were prepared and collected on copper grids.
Sections were stained with 2% uranyl acetate/lead citrate, and grids were examined with a HITACHI H-7700 electron microscope (acceleration voltage: 100 kV). Thickness of postsynaptic density (PSD) was evaluated with Image J software, and the number of the vesicles and the vesicle-active zone distance (within 200 nm distance of active zones) were measured with LoClust software (Nikonenko & Skibo, 2004).

| Statistical analysis
Data are depicted as mean ± standard error (SE). Statistical analysis (SPSS 13.0 or Prism 8.0 software) included one-way ANOVA or Ttest for comparison between two groups, two-way ANOVA for interaction of two factors, p < 0.05 was considered significant.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.