Mossy cell synaptic dysfunction causes memory imprecision via miR-128 inhibition of STIM2 in Alzheimer's disease mouse model.

Abstract Recently, we have reported that dentate mossy cells (MCs) control memory precision via directly and functionally innervating local somatostatin (SST) inhibitory interneurons. Here, we report a discovery that dysfunction of synaptic transmission between MCs and SST cells causes memory imprecision in a mouse model of early Alzheimer's disease (AD). Single‐cell RNA sequencing reveals that miR‐128 that binds to a 3′UTR of STIM2 and inhibits STIM2 translation is increasingly expressed in MCs from AD mice. Silencing miR‐128 or disrupting miR‐128 binding to STIM2 evokes STIM2 expression, restores synaptic function, and rescues memory imprecision in AD mice. Comparable findings are achieved by directly engineering MCs with the expression of STIM2. This study unveils a key synaptic and molecular mechanism that dictates how memory maintains or losses its details and warrants a promising target for therapeutic intervention of memory decays in the early stage of AD.


| INTRODUC TI ON
Alzheimer's disease (AD) is the most common cause of brain degeneration characterized by a progressive memory decline and a subsequent loss of broader cognitive functions (Selkoe, 2001;Sisodia & St George-Hyslop, 2002). There are two major pathological hallmarks, including extracellular deposition of amyloid-β (Aβ) plaques and intracellular aggregates of abnormal tau protein tangles that are seen in the brain of AD patients and have long been considered to be the causes of neuronal degeneration and memory loss (Duffy et al., 2015;Lazarov et al., 2005;Ohno et al., 2004;Vassar, 2007). However, recent studies indicate that memory loss is caused by synaptic dysfunction rather than neuronal death (Hsia et al., 1999;Ittner et al., 2010;Jacobsen et al., 2006;Selkoe, 2002;Shu et al., 2016;X. Yang et al., 2018;Zhu et al., 2017).
In AD patients, memory impairments correlate with a reduction of excitatory synaptic terminals in the cerebral cortex (Hsia et al., 1999;Ittner et al., 2010). In APPswe/PSEN1dE9 mice (AD mice) that carry a transgene encoding the 695-amino-acid isoform of the human Aβ precursor protein with the Swedish mutation and a mutant human presenilin 1 (PS1-dE9), which exhibit plaque pathologies similar to those in AD patients, synaptic dysfunctions in the hippocampus, a brain region responsible for learning and memory, reduce the capability of spatial information acquisition (Reiserer, Harrison, Syverud, & McDonald, 2007;Scheff, Price, Schmitt, DeKosky, & Mufson, 2007;Yang et al., 2018). However, there were three key In our present study, we have shown that memory is imprecise in a mouse model of early AD and an imprecision of memory is associated with a selective degeneration of MC synapses. Single-cell RNA sequencing reveals that miR-128, which binds to a 3′UTR region of STIM2 and inhibits STIM2 translation, is increasingly expressed in MCs from AD mice. Silencing miR-128 or uncoupling miR-128 binding to STIM2, and/or expressing exogenous STIM2, restores synaptic function of MCs and significantly improves the quality of memories in AD mice. Thus, dysfunction of mossy cell synaptic transmission causes memory imprecision via miR-128 inhibition of STIM2 translation.

F I G U R E 1 Memory is imprecise in a mouse model of early AD.
(a) Illustration shows a touch screen spatial discrimination model of high separation tasks (HST). (b) A plot shows the accuracy, as defined the percentage of correct performance versus a total number of trials in training sessions and the trials that are required to reach the criterion in testing sessions of the individual AD mice (red circles) at 3, 6, 8, or 10 months old of age and the agematched individual controls (blue circles) and their averages per group (triangles, 76 ± 3.8 accuracy in AD mice at 3 months versus 75 ± 4.4 in controls, p = .6471, t = 0.4647; 81 ± 3.8 accuracy in AD mice at 6 months versus 78 ± 3.6 in controls, p = .6218, t = 0.5010; 79 ± 3.9 accuracy in AD mice at 8 months versus 77 ± 3.8 in controls, p = .7529, t = 0.3192; 75 ± 4.4 accuracy in AD mice at 10 months versus 80 ± 3.4 in controls, p = .4049, t = 0.8510; triangles, 12 ± 2.2 trials in AD mice at 3 months versus 12 ± 2.2 in controls, p > .9999, t = 0; 12 ± 2.3 trials in AD mice at 6 months versus 12 ± 2.3 in controls, p = .8456, t = 0.1972; 11 ± 2.1 trials in AD mice at 8 months versus 11 ± 2.1 in controls, p = .9037, t = 0.1225; 13 ± 1.6 trials in AD mice at 10 months versus 20 ± 2.8 in controls, *p = .0465, t = 2.123, mean ± SEM, n = 11 mice per group, two-tailed t test). (c) Illustration shows a touch screen spatial discrimination model of low separation tasks (LST). (d) A plot shows the accuracy, as defined the percentage of correct performance versus a total number of trials in training sessions and the trials that are required to reach the criterion in testing sessions of the individual AD mice (red circles) at 3, 6, 8, or 10 months old of age and the age-matched individual controls (blue circles) and their averages per group (triangles, 72 ± 5.1 accuracy in AD mice at 3 months versus 74 ± 4.3 in controls, p = .7889, t = 0.2714; 70 ± 4.6 accuracy in AD mice at 6 months versus 73 ± 3.8 in controls, p = .6221, t = 0.5007; 68 ± 4.6 accuracy in AD mice at 8 months versus 69 ± 3.2 in controls, p = .9104, t = 0.1140; 66 ± 3.5 accuracy in AD mice at 10 months versus 72 ± 3.4 in controls, p = .2002, t = 1.325; triangles, 18 ± 2.7 trials in AD mice at 3 months versus 19 ± 3.9 in controls, p = .7632, t = 0.3054; 33 ± 1.9 trials in AD mice at 6 months versus 18 ± 2.6 in controls, **p < .0001, t = 4.917; 36 ± 0.9 trials in AD mice at 8 months versus 18 ± 2.7 in controls, **p < .0001, t = 6.448; 38 ± 0.6 trials in AD mice at 10 months versus 19 ± 2.5 in controls, **p < .0001, t = 7.174, mean ± SEM, n = 11 mice per group, two-tailed t test.)

| Memory is imprecise in a mouse model of early AD
We assessed the neurological functions of male AD mice during the course of aging and found that AD mice before 6 months old of age were normal in motor activity throughout the light-dark phases (Figure S1a-d) and performed normally in elevated plus maze ( Figure S2a,b), novel object recognition ( Figure S2c,d), the rotarod test ( Figure S3a,b), the Morris water maze ( Figure S4a-c), and a high separation task (HST) of two-choice discrimination (Figure 1a,b), as compared to the age-matched C57BL/6 controls. Abnormal performance in the Morris water maze was observed in AD mice only when they were at 8 months old of age or older ( Figure S4a-c). In low separation task (LST) of a two-choice spatial discrimination test that detects the quality and/or the precision of memories, AD mice were severely impaired even when they were at 6 months old of age (Figure 1c,d). These findings reveal that memory is imprecise in a mouse model of early AD.

| Mossy cell synaptic transmission is dysfunctional in a mouse model of early AD
We next determined the underlying cellular and molecular mecha-  and other skewed peaks ( Figure 2c). Notably, however, and in stark contrast to controls, the distribution of EPSCs in AD mice was dramatically shifted toward the smaller numbers of the events, with a significant increase of the failures ( Figure 2d). The mean amplitude (M) was unchanged, whereas the CV that reflects the probability of presynaptic transmitter release obtained by the method of M 2 /ς 2 was dramatically reduced in synaptic terminals of MCs.
We next recorded the paired-pulse facilitation of the evoked EPSCs. To measure glutamate release from synaptic terminals of MCs accurately and to avoid the polysynaptic responses, we recorded N-methyl-D-aspartic acid (NMDA) receptor-mediated EPSCs in the presence of 20 μm AMPA receptor antagonist NBQX and 20 μM bicuculline ( Figure 2e). We found that AD mice displayed greater paired-pulse facilitation of synaptic transmission from MCs to SST cells than controls, with the greatest effects at the shortest inter-stimulus interval (Figure 2f). This facilitation increased with an elevation of Ca 2+ (Figure 2g), showing a reduction of glutamate transmitter release probability from synaptic terminals of MCs in the early stage of AD. To further validate a reduction of presynaptic transmitter release from MCs, we next recorded excitatory synaptic transmission between MCs and the dentate granule cells (GCs, Figure 2h). NMDA receptor-mediated EPSCs in GCs were evoked by delivery of BLL onto axon fibers of MCs ( Figure 2i). Notably, AD mice at 6 months old of age displayed greater paired-pulse facilitation of synaptic transmission from MCs to GCs than age-matched control mice, with the greatest effects at the shortest inter-stimulus interval ( Figure 2j). Together, these data demonstrate that neurotransmitter release from synaptic terminals of MCs is reduced in the early stage of AD.

| Activation of MCs rescues memory imprecision in a mouse model of early AD
To establish a causal link between synaptic degeneration and memory imprecision in early stage of AD, we examined whether functional

restoration of synaptic transmission from MCs to SST cells is sufficient
to improve the quality of learned memories. To do so, we engineered reliably and repeatedly evoked action potential firings ( Figure S5).
Subsequently, we bilaterally implanted optic fibers into the hilar regions and applied BLL, consisting of 60 pulses at 20 Hz, repeated 15 times, every 2 min through the fibers once per day for 30 con-

| Activation of MCs rescues LTP in a mouse model of early AD
Long-term potentiation (LTP) of synaptic transmission is widely considered as a cellular substrate of learning and memory. Accordingly, we wanted to determine whether LTP of synaptic transmission between MCs and SST cells is impaired in early AD and whether this impair-

| MiR-128 inhibits STIM2 translation in a mouse model of early AD
Next, we searched for the signaling molecules responsible for synaptic decay of MCs in AD mice by focusing on the altered expression of synaptic genes in MCs of AD mice at 6 months old of age. STIM2, as a member of the stromal interaction molecule family, was interested as it detects Ca 2+ homeostasis in the endoplasmic reticulum and involves Ca 2+ -dependent cellular events (Novello, Zhu, Feng, Ikura, & Stathopulos, 2018;Pchitskaya et al., 2017). A dramatic reduction of STIM2 was verified by single-cell PCR and Western blots in MCs from AD mice at 6 months old of age (Figure 5a-c).
In addition to a reduction of STIM2, we also identified a small, noncoding RNA, miR-128 that was increased by ~13-folds in MCs of AD mice at 6 months old of age, as compared to controls ( Figure 5c). MiR-128 binds to a complementary sequence (CACUGUGA) in the mRNA 3′-untranslated region (3′UTR) of STIM2 ( Figure 5d). We subsequently created a wild-type 3′UTR segment and its mutant of STIM2 and placed these segments into the luciferase reporter system. When expressed with miR-128, a wild-type reporter showed significant inhibition, as compared to its mutant control ( Figure 5e).
To determine further the inhibition of STIM2 translation, we expressed miR-128 in MCs of adult control mice (Figure 5f). Western blots of MC proteins revealed that miR-128 significantly inhibited STIM2 protein expression, without affecting the expression of the other synaptic proteins including syntaxin-1a, Cav2.1, and synaptotagmin-1 ( Figure 5g). Thus, miR-128 directly targets to and inhibits STIM2 translation in the MC synaptic terminals.

| Inhibition of miR-128 restores STIM2 translation in a mouse model of early AD
Next, we examined whether inhibition of miR-128 promotes STIM2 protein expression and hence restores synaptic transmission from MCs to SST cells in AD mice. We created a small interfering RNA that specifically targets miR-128 (miR-128I). As a control, we used a MCs with the expression of exogenous STIM2 in AD mice ( Figure S7).

| Therapeutic intervention of memory decays by interfering miR-128 binding to STIM2 in AD
Having established that miR-128 inhibition of STIM2 translation in MCs play a key role in synaptic degeneration and memory imprecision in early stage of AD, we sought to test an alternate, simpler way to relieve an inhibitory effect of STIM2 translation by miR-128. We designed a locked nucleic acid (LNA)-modified oligonucleotide (LNA-STIM2) tiling the entire region of overlap between miR-128 and STIM2 (Figure 7a). In this study, a scrambled F I G U R E 2 Mossy cell synapses are dysfunctional in a mouse model of early AD. (a-d) Double whole-cell patch clamp recordings (a) from MCs (green) paired with nearby SST cells (red) and representatives (b) of action potential firings from MCs and the evoked EPSCs at a holding potential of −65 mV in SST cells, and distributions (c) of the evoked EPSCs amplitude with bin sizes of 3 pA (inset, CV 2 1.07 ± 0.1 in control versus 0.6 ± 0.09 in AD, *p = .00418, t = 3.337, mean ± SEM, n = 9 mice per group, two-tailed t test) and the failure rates (d) from the individual (circles) 14.27 ± 2.19 in control versus 23.82 ± 2.94 in AD mice and their averages per group (triangles) are plotted, *p = .0168, t = 2.608, mean ± SEM, n = 9 mice per group, two-tailed t test. (e, f) Representatives of EPSCs (e) are recorded from SST cells at holding potential of +60 mV. The EPSCs were evoked by paired-pulse stimulations of MCs in the presence of 20 μm NBQX and 20 μm bicuculline. A plot shows the paired-pulse ratios (P2/P1) versus intervals of paired stimulation pulses (f) from the individual AD mice (green circles) at 6 months old of age and the age-matched individual control mice (blue circles) and their averages per group (triangles, 2.1 ± 0.17 control mice versus 2.9 ± 0.19 AD mice at 50-ms interval, **p = .00358, t = 3.41; 1.6 ± 0.12 control mice versus 2.1 ± 0.16 AD mice at 100-ms interval, *p = .0123, t = 2.819; 1.3 ± 0.09 control mice versus 1.9 ± 0.17 AD mice fat 150-ms interval, **p = .00998, t = 3.41; mean ± SEM, n = 9 mice per group, two-tailed t test). (g) The paired-pulse ratios (P2/P1) from the individual AD mice (green circles) and age-matched individual controls (blue circles) and their averages per group (triangles) with elevation of extracellular Ca 2+ in the recording solutions (1.6 ± 0.13 control mice versus 2.2 ± 0.17 AD mice at 1 mM, *p = .0452, t = 2.235; 2.3 ± 0.23 control mice versus 2.9 ± 0.27 AD mice at 3 mM, p = .167, t = 1.471; 2.6 ± 0.27 control mice versus 3.5 ± 0.31 AD mice at 5 mM, *p = .0405, t = 2.295; mean ± SEM, n = 7 mice per group, two-tailed t test). (h) Double whole-cell patch clamp recordings from MCs (green) paired with nearby granule cells (GC). (i) Representatives of EPSCs are recorded from granule cells (GC) at holding potential of + 60 mV. The EPSCs were evoked by paired-pulse stimulations of MCs in the presence of 20 μm NBQX and 20 μm bicuculline. (j) The paired-pulse ratios (P2/P1) from the individual (circles) AD mice at 6 months old of age and age-matched controls and their averages per group (triangles, 2.4 ± 0.23 control mice versus 3.1 ± 0.22 AD mice at 50-ms interval, *p = .0392, t = 2.246; 1.6 ± 0.13 control mice versus 2.2 ± 0.15 AD mice at 100-ms interval, **p = .0040, t = 3.358; 1.4 ± 0.11 control mice versus 1.9 ± 0.15 AD mice at 150-ms interval, *p = .0249, t = 2.475; mean ± SEM, n = 9 mice per group, two-tailed t test) LNA-STIM2 (sLNA-STIM2) was used as a control. Intravenous and Figure S8). F I G U R E 3 Activation of MCs improves the quality of memory in AD mice. (a-d) Illustrations show the injection of the rAAV1/2-DIO-GFP (green) and rAAV1/2-DIO-ChR2-tdT (red) virus into the dentate gyrus of the AD/Calb2-CRE (a) and the AD/C57BL/6 control (b) mice, respectively. Representative images show the expression of GFP (green) and ChR2 (red) specifically in MCs 12 days after the virus injection in the dentate gyrus of the AD/Calb2-CRE mice (c), but not the AD/C57BL/6 mice (d). (e-g) Experimental schedule (e) shows the applications (top) of blue laser lights onto ChR2-expressing MCs of AD mice beginning at 5 months old of age and a touch screen spatial discrimination model (bottom) of HST and LST to examine memory precision. The trials that are required to reach the criterion in HST (f) (triangles, 12 ± 2.1, 11 ± 1.6, 13 ± 2.1, 14 ± 2.0 trails in Control/ChR2, AD/ChR2 Control/tdT, AD/tdT; F(3, 40) = 0.4633, p = .7095, mean ± SEM, n = 11 mice per group, one-way ANOVA) and LST (g) (triangles, 19 ± 2.8, 21 ± 3.3, 20 ± 2.7, 36 ± 1.2 trails in Control/ChR2, AD/ChR2 Control/tdT, AD/ tdT; F(3, 40) = 9.661, p < .0001, mean ± SEM, n = 11 mice per group, one-way ANOVA) from the individual mice (circles) and their averages per group. (h-j) Illustration (h) shows blue light applications onto synaptic terminals of ChR2-expressing MCs and whole-cell patch clamp recordings from SST cells. Representative recordings (i) of EPSCs recorded from SST cells at holding potential of +60 mV in the presence of 20 μm NBQX and 10 μm bicuculline. The paired-pulse ratios at an interval of 50 ms from the individual mice (circles) and their averages per group (triangles) are plotted (j). (2.2 ± 0.2 in control/ChR2 mice, 2.1 ± 0.16 in AD/ChR2 mice, 2.2 ± 0.18 in Control/tdT mice, 3.4 ± 0.31 in AD/tdT mice, F(3, 32) = 6.314, p = .0017, mean ± SEM, n = 9 mice per group, one-way ANOVA)  and STIM2 mRNA (c) in MCs from the individual (circles) AD mice (A) at 3 or 6 months old of ages versus age-matched control mice (C) and their averages per group (triangles, 0.52 ± 0.1 in 6 months control mice versus 0.19 ± 0.05 in AD mice, n = 5 mice per group,*p = .013, t = 3.179, two-tailed t test) (d, e) MiR-128 binds to sequence (d) in the 3′UTR region of STIM2 and inhibits the wild-type (W), but not the mutant (M), STIM2 3′UTR luciferase activity (e) from the individual assays (circles) and their averages per group (triangles, 0.97 ± 0.08 in mutant versus 0.28 ± 0.09 in wild-type, n = 5 mice per group, **p = .00033, t = 5.983, two-tailed t test). (f, g) Representative images (f) show the expression of GFP and miR-128 in MCs by injecting the rAAV1/2-DIO-GFP (green) and the rAAV1/2-DIO-miR-128-IRES-tdT (red) virus into the dentate gyrus of Calb2-CRE (top) and control (bottom) mice. Western blots (g) show that the expression of miR-128 in MCs dramatically reduces STIM2 protein expression without affecting the expression of other synaptic proteins. The normalized protein levels of STIM2, syntaxin-1a (synt), Cav2.1, and synaptotagmin-1 (synap) versus α-tubulin from the individual mice, in which MCs are expressed with miR-128 (red circles) or miR-124 (blue circles) and their averages per group (triangles), (0.61 ± 0.12 STIM2 expression in miR-124 versus 0.13 ± 0.03 in miR-128, n = 5 mice per group, *p = .00471, t = 3.875, two-tailed t test) (4.8 ± 0.8 frequencies of miR-128C in control mice (C) versus 2.2 ± 0.5 in AD mice (A), *p = .0103, t = 2.905, mean ± SEM, n = 9, two-tailed t test). (g, h) Plots show the accuracy of the performance during the training sessions and the trials that are required to reach the criterion during the testing sessions in HST (g) and LST (h) from the individual (circles) and the averages per group (triangles) of mice, in which MCs are expressed with miR-128C or miR-128I (21 ± 2.5 trails in control/miR-128C mice in LST testing versus 37 ± 1.1 in AD/miR-128C mice, **p < .001, t = 5.718, mean ± SEM, n = 11 mice per group, two-tailed t test) general memory deficit is detectable in the same strain of AD mice 2 months late. This finding reveals that memory imprecision is an earliest neurological sign of AD.

| D ISCUSS I ON
In the present study, we have investigated the pathological events in a mouse model of early AD. We have uncovered that dysfunctional synaptic transmission from MCs to SST cells rather than neuronal loss is associated with memory imprecision in AD mice. Using single-cell RNA sequencing, we have identified that the levels of miR-128 that binds to and inhibits STIM2 translation in MCs from AD mice are dramatically increased, as compared to controls (Figure 7g). Experimentally reducing levels of miR-128 and/or expressing exogenous STIM2 in MCs restore synaptic functions of MCs and significantly improve the quality of learned memories in AD mice. This finding demonstrates that miR-

inhibition of STIM2 translation impairs synaptic transmission from
MCs to SST cells and hence causes memory imprecision.
STIM2 is a member of the stromal interaction molecule family and detects Ca 2+ homeostasis in the endoplasmic reticulum and involves Ca 2+ -dependent cellular events (Novello et al., 2018;Pchitskaya et al., 2017). In our present study, we have shown that STIM2 expression is dramatically reduced in MCs from AD mice at 6 months old of age. We have further demonstrated that a reduction of STIM2 impairs glutamate transmitter release from presynaptic terminals of MCs at Ca 2+ -dependent manner. Thus, it is probable that STIM2 acts as Ca 2+ sensor at the presynaptic terminals for glutamate transmitter release.
But, how the expression of miR-128 is regulated at a transcriptional level in the brain is yet to be studied. Thus, it would be important to determine the mechanisms underlying transcriptional regulation of miR-128 expression in AD patients and hence provide the promising targets for therapeutic intervention of AD via inhibition of miR-128 transcription.
MCs are a small group of excitatory neuronal types scattered under the granule cell layer and are distinct from the other excitatory cell types including granule cells in the dentate gyrus in a number of ways including morphology, intrinsic physiological properties, afferent inputs, and axonal projections (Leranth & Hajszan, 2007;Leutgeb, Leutgeb, Moser, & Moser, 2007;Stevenson et al., 2018).

Because of the lack of reliable methods to identify and differentiae
MCs

| Generation and breeding of mutant mice
Male APP/PS1 mice (AD mice, Stock No: 34829, the Jackson Laboratory) are double transgenic mice expressing a chimeric mouse/human amyloid precursor protein (Mo/HuAPP695swe) and a mutant human presenilin 1 (PS1) mice, were used, unless otherwise specified. Mice were bred and reared under the same conditions in accordance with institutional guidelines and the Animal Care and Use Committee of the animal core facility at Huazhong University of Science and Technology, Wuhan, China, and housed in groups of F I G U R E 7 Disruption of miR-128 binding to STIM2 is therapeutic effective against memory imprecision in AD. (a) LNA-STIM2 disrupts miR-128 binding to 1 3′-UTR region of STIM2. (b, c) Administration of LNA-STIM2 increases STIM2 expression in MCs of AD mice without affecting miR-128. Plots show the expression levels of miR-128 and STIM2 mRNA (b) and STIM2 protein (c) in MCs of the individual (circles) and the averages per group (triangles) of mice, in which LAN-STIM2 or a scrambled control (sLNA-STIM2) is applied (−5.70 ± 0.89 STIM2 mRNA in AD sLNA-STIM2 mice versus 0.40 ± 0.39 in AD LNA-STIM2 mice, **p = .0002, t = 6.304; 0.29 ± 0.08 STIM2 protein in AD sLNA-STIM2 mice versus 0.77 ± 0.12 in AD LNA-STIM2 mice, **p = .0113, t = 3.272, mean ± SEM, n = 5 mice per group, two-tailed t test). (d, e) Representative recordings (d) of the evoked EPSCs in SST cells at holding potential of +60 mV in the presence of 20 μm NBQX and 10 μm bicuculline. The paired-pulse ratios (e) at an interval of 50 ms from the individual control and AD mice (circles) and the averages per group are plotted (3.2 ± 0.36 in AD sLNA-STIM2 mice versus 2.1 ± 0.22 in AD LNA-STIM2 mice, *p = .024, t = 2.581, mean ± SEM, n = 7 mice per group, two-tailed t test). (f) Plots show the trials that are required to reach the criterion from the individual (circles) control and AD mice and the averages per group (triangles) (36 ± 1.2 in AD sLNA-STIM2 mice in LST versus 24 ± 3.1 in AD LNA-STIM2 mice, **p = .0021, t = 3.655, mean ± SEM, n = 9 mice per group, two-tailed t test). (g) Working model: Under physiological conditions, STIM2-dependent transmitter release from synaptic terminals of MCs mediates synaptic transmission from MCs to SST cells and control memory precision. In an early stage of AD, miR-128 is increasingly expressed in MCs. MiR-128 binds to and inhibits STIM2 translation and impairs STIM2-dependent transmitter release from MC synaptic terminals, thereby causing memory imprecision three to five mice/cage under a 12-hr light-dark cycle, with lights on at 8:00 a.m., at a consistent ambient temperature (21 ± 1°C) and humidity (50 ± 5%). All behavioral tests were conducted during the light phase of the cycle. For a two-choice spatial discrimination test, mice were maintained on a restricted diet and kept at 85% of free-feeding body weight during behavioral testing. Calb2-CRE (stock No: 010774) was purchased from the Jackson Laboratory, Bar Harbor, Maine, USA. C57BL/6 mice were used as controls. Because male AD mice had the memory imprecision, as the same as the agematched female AD mice did when they were at 6 months old of age ( Figure S9), for the clarity, we used male mice in this study throughout the experiments.

DIC) Axioskop 2FS upright microscopy equipped with a Hamamatsu
C2400-07E infrared camera, as described before (Li et al., 2019;Scheff et al., 2007;Tu et al., 2010;Yang et al., 2018;Yang et al., 2012). 20 μM NBQX (TOCRIS, 0190) was added before the end of each recording. The row electrophysiological data were collected at 10 kHz and filtered with a low-pass filter at 2 kHz and was analyzed using ClampFit 10.2 software (Molecular Devices) with template matching at a threshold of 5 pA.

| Optogenetics in vivo
In Figure S6, we anesthetized mice with 6% chloral hydrate (0.06 ml/10 g; intraperitoneally) and coated four tetrodes of twisted 17-μm HM-L with platinum-iridium (10% or 20% platinum, Cat#: 100-167, California Fine Wire Company) and connected tetrodes to a microdrive for dorsal-ventral adjustment after the implantation, as described before Yang et al., 2018). We positioned the tetrodes directly above the recording site and secured the microdrive to the skull using jeweler's screws and dental cement. A jeweler's screw was used as a ground electrode. We screened cells and behaviors daily for each experimental procedure. During the screening procedures, we lowered the tetrodes slowly over sev-

| Open-field, object recognition, and rotarod tests
We measured motor activity within clear boxes that measured 100 cm × 100 cm and outfitted with photo-beam detectors for monitoring horizontal and vertical activity (Li et al., 2019;Scheff et al., 2007;Tu et al., 2010;Yang et al., 2018;Yang et al., 2012). The The mice were not exposed to the chamber prior to testing.
The data were recorded for each individual animal during 30-min intervals.
To test the performance in the object recognition task, we subjected mice for two sessions of one trial each: acquisition and retrieval trials. During the acquisition trial, we placed mice in an arena that contained two identical objects for 5 min. The mice that did not explore the objects for 20 s within the 5-min period were excluded from further experiments. We defined exploration as a mouse approaching its nose within 1 cm of the object. This approaching was associated with looking, sniffing, or touching. The retrieval session was done 2 hr after the acquisition trial. In this trial, we replaced one of the objects presented in the first trial with a novel object. We then placed mice back in the arena for 5 min and recorded the total time spent in the exploration of each object. New objects were different in shape and color, but made of the same materials and with similar general dimensions. The objects and arenas were thoroughly cleaned with 70% ethanol between trials. New objects and positioning of new objects were counterbalanced through all experiments to avoid bias. Motor activity and time spent in active exploration of the familiar or novel objects during the retrieval trial were calculated. Recognition index was expressed as the time exploring the novel object divided by the total time exploring both objects and multiple by 100. For rotarod tests, mice were trained to perform the task for 6 consecutive days. After training, mice were subjected to accelerating rotarod tests twice per week. Task performance that mice were unable to remain on the rotating bar for more than 10 s on three consecutive attempts was defined as failure.

| Two-choice spatial discrimination test
A mouse touch screen task was conducted in an automated touch screen platform, which was consisted of the Bussey-Saksida mouse touch screen chamber (Lafayette Instrument Company), and equipped with a house light, a reward port, holding a reward magazine with an infrared sensor for detection of a mouse entering into the port, and a touch-sensitive monitor on the front side.
All trials in the chamber were mouse-initiated and independent of the experimenters. A Perspex plate containing 6 windows, which is approximately 1.6 cm above the floor of the chamber, was located in front of the touch screen to allow spatially localization of stimuli on the touch screen and to prevent a mouse from incident triggering the touch screen (e.g., with its tail), as described before (Li et al., 2019;Scheff et al., 2007;Tu et al., 2010;Yang et al., 2018;Yang et al., 2012).
In pretraining, mice were initially habituated to the apparatus and learned to nose poke to the stimuli presented in one of the 6 windows, and then through several stages to associate the touching of stimuli on the screen with the delivery of reward (15 μl sweet milk) in the reward magazine as described before. Once a mouse returned to the magazine to retrieve the reward, the magazine light was turned off and an ITI (intertrial interval) of 20 s initiated, followed by the initiation of the next trial.
In training, mice were then trained to perform a two-choice spatial discrimination task. In this task, two illuminated squares from closed (LST) to far apart (HST) were presented on the touch screen, one the correct (e.g., the leftmost square) and the other the incorrect. A nose poke to a correct square resulted in a tone, magazine light and a reward pellet. Incorrect responses were followed by a 5-s time-out in which the house light was extinguished, followed by a correction procedure in which the stimulus display was repeated until the mouse made a correct response. Performance was measured by calculating the percent correct choices per session of 30 (noncorrection) trials. Once reaching more than 80% correct performance, the other location (e.g., the leftmost square) was designed as correct, which was defined as a reversal. At the initial training, one reversal with a maximum of 41 trials per day and a criterion of 9 out of 10 consecutive touches were required.
HST (separation with 4 empty/dark locations between the two illuminated locations) and LST (separation with 0 empty/dark locations between two illuminated locations) were presented on the touch screen by separated illuminated boxes. Mice were examined for 6-day blocks on each separation and counterbalanced across the separations.
In testing, after a completion of a 12-day training (each separation of 6-day blocks), the protocol was modified for testing as follows: Mice were allowed to reverse unlimitedly within 60 min (up to 81 trials/day) for a further testing of 6 days (each separation of 3-day blocks), and a criterion was set at 7 of 8 correct consecutive touches. All mice completed 81 trials per day, and only reversals, in which the mouse reached a criterion, were included in the analysis.
If a mouse failed to reach the first reversal within 60 min (up to 81 trials), a maximum score of 41 trials was used for this trial. This maximum score represented half of the total trials per session, which was a conservative estimate.
We designed a locked nucleic acid (LNA)-modified oligonucleotide (LNA-STIM2) tiling the entire region of overlap between miR-128 and STIM2. In this study, a scrambled LNA-STIM2 (sLNA-STIM2) was used as a control. Intravenous administration of LNA-STIM2 or sLNA-STIM2 with multiple doses (0.5 mg/kg per dose, single dose per day for 5 consecutive days), as described before (Yang et al., 2012). We tested three different doses of LNA-STIM2 (0.1 mg/kg per dose, single dose per day for 5 consecutive days; or 1.0 mg/kg per dose, single dose per day for 3 consecutive days) and found that a dose (0.5 mg/kg per dose, single dose per day for 5 consecutive days) was effective in restoration of STIM2 expression in AD mice to the control level.

| Statistical analysis
All values in the text and Figure legends are represented as the mean ± SEM. Unpaired two-tailed Student's t tests (t test) and repeated one-way ANOVA and post hoc Bonferreni's following twoway analyses of variance (BF ANOVA) were used when assumptions of normality and equal variance (F test) were met. Significance was accepted for p < .05. Power calculations were performed using G*power software v3.1.9.2 (IDRE Research Technology Group, Los Angeles, USA). Group sizes were estimated based on recent studies and were designed to provide at least 80% power with the following parameters: probability of type I error (α) = 0.05, a conservative effect size of 0.25, and 3-8 treatment groups with multiple measurements obtained per replicate.

ACK N OWLED G M ENTS
This work was supported by the National Natural Science Foundation of China (Grants: 31721002 to YL; 91632306 to YL; 51627807 to YL).

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

AUTH O R CO NTR I B UTI O N S
YL conceived and designed the studies and wrote the paper. MD, QZ, ZW, and TM carried out the experiments including electrophysiology, behavioral tests, and genetic and optogenetic studies. AH, TZ, XK, and QY performed the experiments including genotyping, PCR, and cell counting. All authors contributed to the data analysis and presentation in the paper.

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.