Deficiency of Neuronal p38α MAPK Attenuates Amyloid Pathology in Alzheimer Disease Mouse and Cell Models through Facilitating Lysosomal Degradation of BACE1*

Amyloid β (Aβ) damages neurons and triggers microglial inflammatory activation in the Alzheimer disease (AD) brain. BACE1 is the primary enzyme in Aβ generation. Neuroinflammation potentially up-regulates BACE1 expression and increases Aβ production. In Alzheimer amyloid precursor protein-transgenic mice and SH-SY5Y cell models, we specifically knocked out or knocked down gene expression of mapk14, which encodes p38α MAPK, a kinase sensitive to inflammatory and oxidative stimuli. Using immunological and biochemical methods, we observed that reduction of p38α MAPK expression facilitated the lysosomal degradation of BACE1, decreased BACE1 protein and activity, and subsequently attenuated Aβ generation in the AD mouse brain. Inhibition of p38α MAPK also enhanced autophagy. Blocking autophagy by treating cells with 3-methyladenine or overexpressing dominant-negative ATG5 abolished the deficiency of the p38α MAPK-induced BACE1 protein reduction in cultured cells. Thus, our study demonstrates that p38α MAPK plays a critical role in the regulation of BACE1 degradation and Aβ generation in AD pathogenesis.

Alzheimer disease (AD) 2 is pathologically characterized by the extracellular deposits of amyloid ␤ peptide (A␤). A␤ injures neurons in the neocortex and limbic system directly (1) and indirectly by triggering microglial release of various neurotoxic inflammatory mediators, including cytokines (tumor necrosis factor-␣ and interleukin-1␤ (IL-1␤)) and reactive oxygen species (2). A␤ is generated after serial digestion of Alzheimer amyloid precursor protein (APP) by the membrane-anchored ␤-site APP-cleaving enzyme (BACE1, ␤-secretase) and ␥-secretase (3). It has been observed that knock-out of BACE1 or administration of the BACE1 inhibitor dramatically decreases A␤ levels in the brain and attenuates behavioral and electrophysiological deficits in APP-transgenic mice (4 -6). Thus, extensive investigations have focused on the direct inhibition of BACE1 to reduce A␤ load in the AD brain; however, these studies have unfortunately not yet led to any efficacious therapy for AD patients due to the various physiological roles of BACE1 (7). Using alternative methods to inhibit BACE1 might be a preferable investigative approach.
Inflammatory activation might lead to up-regulation of neuronal BACE1 expression in the AD brain, as NF-B signaling enhances (8), and PPAR␥ activation suppresses (9), the activity of bace1 gene promoter. Accumulating evidence has shown that posttranslational modification of BACE1 is extremely important for the activity, intracellular trafficking, and lysosomal degradation of BACE1. For example, phosphorylation of BACE1 at Thr-252 by p25/Cdk5 increases the secretase activity (10), and phosphorylation at Ser-498 facilitates retrograde transport of BACE1 from endosomes to the trans-Golgi network (11). Ubiquitination at Lys-501 targets BACE1 to late endosomes/lysosomes for degradation (12). Finally, bisecting N-acetylglucosamine modification blocks delivery of BACE1 to lysosomes (13).
p38 mitogen-activated protein kinases (p38 MAPKs) are a class of mitogen-activated protein kinases that are responsive to stress stimuli, such as inflammatory cytokines and reactive oxygen species. Phosphorylation of p38 MAPK has been observed in the postmortem brain in the early stages of AD (Braak stages IV-V) (14,15). A␤ and glutamate have each been shown to activate p38 MAPK in cultured neurons by increasing reactive oxygen species (16,17), and the A␤-triggered microglial release of inflammatory mediators, especially IL-1␤, is hypothesized to activate neuronal p38 MAPK (18 -20). p38␣ MAPK mediates A␤-initiated microglial inflammatory activation (21,22), phosphorylates Tau protein in neurons (20,23,24), and mediates A␤-induced synaptic impairment in the cultured hippocampal slice (25). However, no experiments have yet elucidated whether p38 MAPK phosphorylates BACE1 and regulates A␤ generation.
Previous studies demonstrating the effects of p38 MAPK in hippocampal slices or APP-transgenic mice used p38 MAPK inhibitors, which cannot distinguish neuronal p38 MAPK effects and microglial p38 MAPK effects (22,25). In attempting to dissect the role of p38 MAPK in AD, it is essential to distin-guish between p38␣ and p38␤ MAPK enzymes, as they have different functions (26). To investigate the pathogenic function of neuronal p38␣ MAPK, we used Cre-Lox or knockdown techniques to ablate p38␣ MAPK specifically in neurons and examined the effect on BACE1 degradation in this study.

Experimental Procedures
Animal Models and Cross-breeding-APP/PS1 double transgenic mice over-expressing human mutated APP (KM670/ 671NL) and PS1 (L166P) under Thy-1 promoters (27) were kindly provided by M. Jucker, Hertie Institute for Clinical Brain Research, Tübingen, Germany; p38 fl/fl mice with loxP siteflanked mapk14 gene were imported from BioResource Center, RIKEN Tsukuba Institute, Japan (28); and Nex-Cre mice expressing Cre recombinase from the endogenous locus of the nex gene that encodes a neuronal basic helix-loop-helix (bHLH) protein were kindly provided by K. Nave, Max-Planck-Institute for Medicine, Göttingen, Germany (29). All three mouse strains were on a C57BL6 genetic background. APPtransgenic mouse models with deletion of p38␣ MAPK specifically in neurons of the neocortex and hippocampus had been established by mating APP/PS1, p38 fl/fl , and Nex-Cre mice. Animal experiments were performed in accordance with all relevant national rules and were authorized by the local research ethical committee.
Tissue Collection for Histological and Biochemical Analysis-Animals were euthanized at 4 months of age by isofluorane inhalation. Mice were perfused with ice-cold PBS, and the brain was removed and divided. The left hemisphere was for immunohistochemistry. A 0.5-m thick piece of tissue was sagittally cut from the right hemisphere. The cortex and hippocampus were carefully separated and homogenized in TRIzol (Life Technologies) for RNA isolation. The remainder of the right hemisphere was snap frozen in liquid nitrogen and stored at Ϫ80°C until biochemical analysis.
Immunohistological Analysis-For A␤ analysis, the left hemibrains derived from 4-month-old APP-transgenic mice with and without deficiency of p38␣ MAPK were immediately fixed in 4% paraformaldehyde (Sigma) in PBS for 48 h. The tissue was embedded in paraffin. Serial 50-m thick sagittal sections were cut and mounted on glass slides. Immunofluorescent staining with the primary antibody, mouse monoclonal anti-human A␤ antibody (clone 6F/3D; Dako Deutschland GmbH, Hamburg, Germany), was performed on these sections. The staining was visualized by incubating sections with Cy3-conjugated goat anti-mouse IgG (Jackson ImmunoResearch Europe Ltd., Suffolk, United Kingdom). All images were acquired by Zeiss Axio-Imager.Z2 microscope equipped with a Stereo Investigator system (MBF Bioscience, Williston). In the whole hippocampus and cortex, volumes of A␤ and brain tissues were estimated with the Cavalieri probe as we described in the previous study (30) with a 15-m grid size, which provided coefficient of error estimates of Ͻ0.05. The A␤ load was demonstrated as the ratio of A␤ volume to relevant brain tissue volume. To demonstrate Nex-Cre-mediated deletion of p38 MAPK in neurons, the isolated brain was directly embedded in Tissue-Tek O.C.T. Compound (Sakura Finetek Europe B.V., AJ Alphen aan den Rijn, the Netherlands) and frozen in 2-methylbutane on liquid nitrogen. The brain was cut on a freezing-sliding microtome in 5-m coronal sections. For immunofluorescent staining, the rabbit polyclonal antibody against p38 MAPK (catalog number 9212, Cell Signaling Technology, Danvers) and Alexa 488-conjugated goat anti-rabbit IgG (Life Technologies) were used. Neurons were identified by staining the tissue with mouse monoclonal antibody against NeuN (clone A60; Merck Chemicals GmbH, Darmstadt, Germany) and Cy3-conjugated goat anti-mouse IgG (Jackson ImmunoResearch). The experimenter was blinded to the genotypes of mice.
Construction of Knockdown and Transgenic Vectors-Two pcDNA6.2-GW/EmGFP-miR vectors (Life Technologies) (kd509 and kd709) were engineered to contain different select hairpins targeting human p38␣ MAPK-encoding gene, mapk14, using the protocol we established in the previous study (31). Sequences of the DNA oligomers are listed in Table 1. After transfection, the vector will transcribe artificial miRNA, which have 100% homology to the gene sequence of interest, resulting in target RNA cleavage. pcDNA6.2-GW/EmGFP-miR-neg control plasmid (Life Technologies) (kd-ct) containing scrambled sequence was used as a control knockdown vector. pEGFP-hAPP695 transgenic vector was constructed by replacing the eGFP-encoded sequence in pEGFP-N1 vector (Takara Bio Europe/Clontech, Saint-Germain-en-Laye, France) with the PCR products between HindIII and NotI. The PCR was performed using genomic DNA derived from TgCRND8 APPtransgenic mice (32) as the templates and the following primers (Life Technologies) containing HindIII or NotI restriction sites (underlined): 5Ј-GGGAAGCTTCCCACCATGCTGCCCGG-TTTGG-3Ј and 5Ј-ATAAGAATGCGGCCGCTACTAGT-TCTGCATCTGC-3Ј.
Cell Culture and Establishment of Cell Lines-SH-SY5Y neuroblastoma cells were obtained from LGC Standards GmbH (Wesel, Germany) and maintained in DMEM supplemented with 10% fetal calf serum (FCS; PAN Biotech, Aidenbach, Germany). Different cell lines were established: 1) SH-SY5Y cell lines with knockdown of p38␣ MAPK expression were established by transfecting cells with kd-ct, kd509, and kd709 vectors; 2) SH-SY5Y cell lines overexpressing wild-type and dominant-negative (DN) human ATG5 were created by transfecting cells with expression gift vectors from N. Mizushima (Addgene plasmids numbers 22948 and 22949) (33). To evaluate the different autophagic activity, these two different cells were treated  (34). All transfected cells were selected with blasticidin (Life Technologies) and/or G418 (Sigma) until the stable transfection was established. All genetic modifications were confirmed by Western blot detection of their encoding proteins using specific antibodies. Western Blot Analysis of A␤ Pathology-For detection of A␤, C99, and APP in the mouse brain, frozen brain tissues derived from p38␣ MAPK-deficient and wild-type mice were homogenized in 5 ml/g of tissue in radioimmunoprecipitation assay buffer (RIPA; 50 mM Tris, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Nonidet P-40, and 5 mM EDTA (pH 8.0)) supplemented with protease inhibitor mixture (Roche Applied Science, Mannheim, Germany) on ice and centrifuged at 16,100 ϫ g for 30 min at 4°C to collect the supernatants. The protein concentration in the supernatant was measured using a Bio-Rad Protein Assay (Bio-Rad). To evaluate the effects of p38␣ MAPK on A␤ secretion, the FCS-contained culture medium of APP695-transgenic SH-SY5Y cells was replaced with serum-free medium for 16 h. The proteins in the brain homogenate or in the cell culture medium were separated by 10 -20% pre-casted Tris-Tricine gels (Anamed Elektrophorese GmbH, Groß-Bieberau/Rodau, Germany). For Western blot, anti-human amyloid ␤ mouse monoclonal antibody (clone W0 -2; Merck Chemicals GmbH) and anti-␤-actin rabbit monoclonal antibody (clone 13E5; Cell Signaling Technology) were used. Western blots were visualized via the Plus-ECL method (PerkinElmer Life Technologies). Densitometric analysis of band densities was performed with Image-Pro PLUS software version 6.0.0.260 (Media Cybernetics, Inc., Rockville, MD). For each sample, the level of protein in the brain was calculated as a ratio of target protein/␤-actin from that sample. In cultured cells, the A␤ level was adjusted by the secreted APP protein level.
Measurement of the Half-life of BACE1-To measure the lifetime of BACE1, SH-SY5Y cells were treated with 100 g/ml of cycloheximide (Sigma) to inhibit protein translation. The cell lysate was harvested at 0, 4, 8, and 12 h after drug treatment. The remaining BACE1 in the cell lysate was analyzed by SDS-PAGE and Western blotting using BACE1 antibody (clone D10E5) and ␣-tubulin antibody (clone DM1A). Kinetics of the disappearance of BACE1 in p38␣ MAPK knocked down and control cells were compared.
Preparation and Characterization of Lysosome-The procedure was processed as described (35,36) with minor modifications. Cultured SH-SY5Y cells were harvested by trypsinization and washed with PBS. All subsequent manipulations were carried out at 4°C using pre-cooled reagents. Washed cells were homogenized in the HB buffer (0.25 M sucrose, 10 mM Hepes, 1 mM EDTA, (pH 7.4)) supplemented with protease inhibitor mixture (Roche Applied Science) at 1 ϫ 10 8 /ml. Homogenates were spun at 800 ϫ g for 10 min to pellet nuclei and unbroken cells, which were then re-homogenized in a half-volume of HB. The two supernatants were pooled, incubated for 10 min at 37°C in the presence of 2 mM CaCl 2 , and then centrifuged at 3,000 ϫ g for 10 min to remove large heavy mitochondria. The resultant supernatant was centrifuged for 10 min at 18,000 ϫ g, obtaining a pellet that was re-suspended in 0.5 ml of HB buffer and layered on 4 ml of iso-osmotic Percoll (GE Healthcare, München, Germany) at a concentration of 30% (pH 7.4). Under Percoll, 0.5 ml of 2.5 M sucrose was laid. Centrifugation was performed at 4°C for 40 min at 44,000 ϫ g. The subsequent gradients were carefully collected with 0.9 ml/fraction into tubes from the top. Protein concentrations were determined and Western blot analysis was performed using antibodies against the lysosomal marker, LAMP-1 (clone C54H11), and non-lysosomal markers, calnexin (rabbit polyclonal to Calnexin; catalog number ab22595, Abcam) and ␤-actin (clone 13E5).
Preparation of Membrane Components-To measure ␤and ␥-secretase activity, membrane components were purified according to the published protocol (37). Briefly, brain tissues or SH-SY5Y cell pellets were transferred into sucrose buffer (10 mM Tris-HCl, pH 7.4, including 1 mm EDTA, and 200 mM sucrose) and homogenized on ice. The homogenate was centrifuged at 1,000 ϫ g for 10 min at 4°C to delete nuclei. The resulting postnuclear supernatant was transferred to a new tube and centrifuged again at 10,000 ϫ g at 4°C for 10 min. Finally, the resulting supernatant was centrifuged at 187,000 ϫ g in an Optima MAX Ultracentrifuge (Beckman Coulter GmbH, Krefeld, Germany) for 75 min at 4°C. The resulting supernatant was discarded, and pellets were re-suspended using cannulas (10 strokes per cannula) of decreasing diameter (0.6, 0.4, and 0.33 mm) in sucrose buffer.
␤and ␥-Secretase Activity Assays-␤-and ␥-secretase activities were measured by incubating the crude membrane fraction with secretase-specific FRET substrates according to our established methods (37). For measurement of ␤-secretase activity, the crude membrane fraction was resuspended in 500 l of ␤-secretase assay buffer (0.1 M sodium acetate, pH 4.5). The final concentrations for the ␤-secretase assay were: 0.1 mg/ml of membrane protein (125 g of protein/well in 96-well plates), 10% dimethyl sulfoxide, and 8 M ␤-secretase substrate IV (Calbiochem, Darmstadt, Germany). For measurement of ␥-secretase activity, the crude membrane fraction was resuspended in 500 l of ␥-secretase assay buffer (50 mM Tris-HCl, pH 6.8, 2 mM EDTA). Final concentrations for the ␥-secretase assay were: 1 mg/ml of membrane protein (1250 g of protein/ well in 96-well plates) and 8 M ␥-secretase substrate (Calbiochem). For both secretase assays, kinetics were performed at 37°C and fluorescence intensity in each well was measured for 73 cycles with intervals of 5 min with Synergy Mx Monochromator-based Multi-mode Microplate Reader (BioTek, Winooski, VT). Fluorescence intensity of the first cycle was considered as background and subtracted for each well.
BACE Phosphorylation Assay-Human BACE1 sequence with UniProt accession number P56817 was used as the query sequence to predict potential p38 MAPK-mediated phosphorylation sites with the NetPhosK 1.0 Server. Thereafter, p38␣ MAPK knocked down (kd509 or kd709) and wild-type control (kd-ct) cells were lysed in RIPA buffer supplemented with sufficient proteinase and phosphatase inhibitors (50 mM Tris, pH 8.0, 2 mM EDTA, 2 mM EGTA, 50 nM okadaic acid, 5 mM sodium pyrophosphate, 2 mM sodium vanadate, 1 mM DTT, 50 mM NaF, 1% Triton X-100 and protease inhibitor mixture; Roche Applied Science). All operations were on ice and the cell lysate was frozen in liquid nitrogen and stored at Ϫ80°C immediately if not followed by further experiments. The endogenous BACE1 in 200 l of cell lysate (adjusted to 1 mg/ml) was pulled down with 10 g of BACE1 antibody (clone D10E5, Cell Signaling Technology) and 50 l of Dynabeads-conjugated protein G (Life Technologies) for 2 h at 4°C. Samples were detected with Western blot serially with anti-phosphoserine rabbit polyclonal antibody (catalog number SAB5200086; Sigma) and anti-BACE1 antibody (clone D10E5). The amount of phosphorylated BACE1 was shown as the ratio of total BACE1 protein.
Reverse Transcription and Quantitative PCR for Analysis of Gene Transcripts-Total RNA was isolated from the brain homogenate in TRIzol or from cultured SH-SY5Y cells with RNEasy Plus Mini Kit (Qiagen, Hilden, Germany). First-strand cDNA was synthesized by priming total RNA with hexamer random primers (Life Technologies) and using Superscript III reverse transcriptase (Life Technologies). For quantification of gene transcripts, real-time PCR was performed with SYBR Green (Roche Applied Science) with the 7500 Fast Real-time PCR System (Life Technologies). The primer sequences used for detecting transcripts of mouse and human bace1, ps1, nicastrin, and gapdh are listed in Table 2. The amount of doublestranded PCR product synthesized in each cycle was measured by detecting SYBR Green, which binds to double-stranded DNA. Threshold cycle (C t ) values for each test gene from the replicate PCRs were normalized to the C t values for gapdh control from the same cDNA preparations. The ratio of each gene transcript was calculated as 2 (⌬Ct) Confocal Analysis of Autophagic Activity and Colocalization between BACE1 and Autophagic Vacuoles-To investigate the effect of p38␣ MAPK on autophagy, LC3-GFP-RFP-transgenic SH-SY5Y cells cultured on coverslips in a 24-well plate (BD Bioscience, Heidelberg, Germany) at a density of 1 ϫ 10 5 cells per well were treated with p38 MAPK inhibitor, SB203580, at 0, 10, and 30 M for 2 h. The cells were then directly fixed with 4% paraformaldehyde and examined under a Zeiss LSM 510 Meta Confocal Microscope. More than 40 areas under a ϫ40 objective were randomly chosen and Ͼ300 cells were counted. The density of puncta with pure red or green (with week red) fluorescence in each cell was calculated as the total number of puncta divided by the total number of cells. The experiments were repeated independently 3 times. To investigate the relationship between BACE1 and autophagic vacuoles, SH-SY5Y cells were cultured in a 8-well chamber slide (BD Biosciences) at a density of 0.5 ϫ 10 5 cells per well. The cells were treated with or without 200 nM bafilomycin (Sigma) for 6 h. After fixation in 4% paraformaldehyde at room temperature or in methanol at Ϫ20°C, cells were incubated with 0.3% Triton to increase the permeability and blocked with 5% BSA. The cells were first stained with mouse monoclonal antibody (clone number 137612; R&D Systems) or rabbit polyclonal antibody (catalog number ab10716, Abcam) against BACE1 and Alexa 488-conjugated second antibodies (Life Technologies). After sufficient washing, cells were further incubated with rabbit monoclonal antibody against LC3A/B (Clone D3U4C, Cell Signaling Technology) or mouse monoclonal antibody against SQSTM1/p62 (catalog number ab56416, Abcam) and then Cy3-conjugated second antibodies. Whether BACE1 co-localized with autophagic vacuoles was analyzed under confocal microscopy.
Statistics-Data were presented as mean Ϯ S.E. For multiple comparisons, we used one-way or two-way ANOVA followed by Bonferroni, Tukey, or Dunnett T3 post hoc tests (dependent on the result of Levene's test to determine the equality of variances). Two independent samples of t test were used to compare means for two groups of cases. All statistical analyses were performed with SPSS version 19.0 for Windows (IBM, New York, NY). Statistical significance was set at p Ͻ 0.05.

Results
Deletion of p38␣ MAPK in Neurons Reduces A␤ Load in the Alzheimer Mouse Model-To confirm that p38 MAPK is activated in the AD mouse brain, we compared the phosphorylation levels of p38␣ MAPK in 4-month-old APP/PS1-transgenic mice and their wild-type littermate controls and observed that phosphorylation levels were significantly higher in the APP/ PS1-transgenic mice (Fig. 1, A and B; t test, p Ͻ 0.05). We then deleted p38␣ MAPK in neurons of the neocortex and hippocampus of AD mice (mating APP/PS1 mice, with p38 fl/fl , and Nex-Cre mice; see "Experimental Procedures"). Western blot analysis showed that the protein levels of p38␣ MAPK in the homogenate of whole cortex and hippocampus were significantly decreased in 1-month-old APP tg p38 fl/wt Nex-Cre ϩ/Ϫ (p38␣ heterozygote) mice compared with APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ (p38␣ wild-type) littermates (Fig. 1, C and D; t test, p Ͻ 0.05). In a further experiment, we co-stained p38 MAPK and NeuN, a neuronal marker, in the brain of 4-month-old p38 fl/fl -Nex-Cre ϩ/Ϫ (p38␣ homozygote knock-out) and p38 fl/wt or p38 fl/fl Nex-Cre Ϫ/Ϫ (p38␣ wild-type) littermate mice. In the cortex of p38␣ MAPK wild-type mice, p38 MAPK protein was stained and shown in puncta around the NeuN antibodystained nucleus. By contrast, no p38 MAPK was observed in neurons of p38␣ homozygote knock-out mice (Fig. 1E). In the hippocampal neurons, the fluorescent staining of p38 MAPK was weak in both p38␣ MAPK-deficient and wild-type mice, which made the comparison of these two groups of mice difficult (data not shown).
Interestingly, the histological analysis using the stereological Cavalieri method (30) revealed that heterozygous deletion of p38␣ MAPK in neurons resulted in strongly reduced A␤ levels in both the hippocampus and cortex of APP-transgenic mice, even as early as 4 months of age (APP tg p38 fl/wt Nex-Cre ϩ/Ϫ mice: 0.295 Ϯ 0.036 and 0.565 Ϯ 0.095 (adjusted by the volume of analyzed tissues) in hippocampus and cortex, respectively, and APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ littermates: 0.504 Ϯ 0.067 and 0.989 Ϯ 0.188 in hippocampus and cortex, respectively; see Fig.  2, A and B; t test, p Ͻ 0.05). Similarly, Western blot analysis with human A␤-specific antibody showed a 40% reduction with ␤-actin as an internal control, or a 25% reduction with APP as an internal control in the monomers, dimers, or trimers of A␤ in the homogenate of cortex and hippocampus derived from 4-month-old APP tg p38 fl/wt Nex-Cre ϩ/Ϫ mice compared with APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ littermate controls (Fig. 2 Deletion of p38␣ MAPK in Neurons Reduces Both Activity and Protein Levels of BACE1 in the Mouse Brain-To investigate the underlying mechanism by which ablation of neuronal p38␣ MAPK reduces cerebral A␤ levels, we measured ␤and ␥-secretase activity in 4-month-old littermate mice with the genotypes of p38 fl/fl Nex-Cre ϩ/Ϫ , p38 fl/wt Nex-Cre ϩ/Ϫ , and p38 fl/wt or p38 fl/fl Nex-Cre Ϫ/Ϫ , but without the expression of human APP. We observed that neuronal p38␣ MAPK deletion significantly reduced the activity of ␤-secretase, but not of ␥-secretase, in a copy-dependent manner (Fig. 3, A and B; twoway ANOVA, p Ͻ 0.05). However, Bonferroni post hoc tests only showed significant differences between p38 fl/fl Nex-Cre ϩ/Ϫ (p38␣ homozygote knock-out) and p38 fl/wt or p38 fl/fl -Nex-Cre Ϫ/Ϫ mice (p38␣ wild-type), not between p38 fl/wt Nex-Cre ϩ/Ϫ (p38␣ heterozygote knock-out) and p38␣ MAPK wild-type mice. Interestingly, the ␤-secretase activity in the brain tissue of 4-month-old APP tg p38 fl/wt Nex-Cre ϩ/Ϫ mice (p38␣ heterozygote knock-out) that express human APP was already lower than the ␤-secretase activity in the brain tissue of APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ control mice (p38␣ wild-type) (Fig.  3C, two-way ANOVA, p Ͻ 0.001). Similarly, p38␣ MAPK deletion in neurons did not affect the ␥-secretase activity in APPtransgenic mice (Fig. 3D). Moreover, we observed that brain homogenates derived from 4-month-old APP tg p38 fl/wt Nex-Cre ϩ/Ϫ mice had lower amounts of APP-C99 fragment, a ␤-secretase product, compared with brain homogenates derived from APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ littermates (Fig. 3E, t test, p Ͻ 0.05).
We used Western blot analysis to examine ␤-secretase (BACE1) protein levels in the 4-month-old APP-transgenic mice. As shown in Fig. 3G, p38␣ MAPK deletion in neurons significantly decreased the amount of BACE1 protein in the brain of APP-transgenic mice compared with wild-type control mice (t test, p Ͻ 0.05). Surprisingly, real-time PCR showed that FIGURE 1. The cerebral p38␣ MAPK is efficiently ablated by cross-breeding p38 fl/fl and Nex-Cre mice. The brain homogenates derived from 4-month-old APP-transgenic (tg) and wild-type (wt) littermates were analyzed for the phosphorylation of p38␣ MAPK with Western blot (A and B, t test, n Ն 4 per group). Similarly, the cerebral p38␣ MAPK protein in 1-month-old APP tg p38 fl/wt Nex-Cre ϩ/Ϫ (p38␣ MAPK heterozygote) and APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ (p38␣ MAPK wild-type, wt) littermates was detected with Western blot (C and D, t test, n ϭ 7 per group). A or C, grouping images from different parts of the same gel. E, 4-month-old p38 fl/fl Nex-Cre ϩ/Ϫ (p38␣ MAPK ko) and p38 fl/wt or p38 fl/fl Nex-Cre Ϫ/Ϫ (p38␣ MAPK wt) littermate mice were co-stained with fluorophore-conjugated antibodies against p38 MAPK and NeuN in the brain cortex. p38 MAPK protein was shown in green puncta around the NeuN staining in red, whereas, no p38 MAPK was stained in NeuN staining-positive neurons in p38␣ MAPK ko mice. Nuclei were stained with DAPI in blue.
neuronal deletion of p38␣ MAPK expression up-regulated transcription of the bace1 gene, perhaps indicating compensation following BACE1 protein reduction (Fig. 3H, t test, p Ͻ 0.05). Transcription of ␥-secretase-encoding genes (like ps1 and nicastrin) was not significantly altered (Fig. 3H). Thus, we hypothesized that deficiency of p38␣ MAPK in neurons decreases A␤ generation by suppressing ␤-secretase activity and reducing the levels of BACE1 protein.
In additional experiments, we quantified the protein levels of APP in APP tg p38 fl/wt Nex-Cre ϩ/Ϫ and APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ mice with Western blot. We observed no differences (see Fig. 3F, p Ͼ 0.05), thereby excluding the possibility that A␤ reduction in the brain observed in our study was due to the decreased protein levels of APP.
Knockdown of p38␣ MAPK Expression Reduces Both Activity and Protein Amount of ␤-Secretase in the Neuronal Cells-To confirm the finding that deletion of p38␣ MAPK in neurons specifically reduces ␤-secretase activity in APP-transgenic mice, we established both p38␣ MAPK knockdown and control neuronal cell lines by stably transfecting SH-SY5Y cells with artificial human mapk14 miRNA-transcribing vectors. Vectors kd509 and kd709 were targeted to the mRNA of mapk14 at different sites, and the kd-ct vector transcribed miRNA that was not complementary to any mammalian mRNA (see "Experimental Procedures"). Western blot detection of p38␣ MAPK protein demonstrated the efficiency of both vectors 509 and 709-induced knockdown of mapk14 gene expression with ϳ50% reduction of p38␣ MAPK protein compared with vector kd-ct-transfected cells (Fig. 4A, one-way ANOVA, p Ͻ 0.001). Interestingly, reduction of p38␣ MAPK resulted in ϳ40% reduction of BACE1 levels in the neuronal cell lines (Fig. 4B, one-way ANOVA, p Ͻ 0.05) and almost 50% suppression of ␤-secretase activity (Fig. 4C, two-way ANOVA, followed by Bonferroni post hoc test between cells with different geno-types). We also detected ␥-secretase activity in p38␣ MAPK knocked-down and control cells and observed that deficiency of p38␣ MAPK did not change ␥-secretase activity (Fig. 4D, twoway ANOVA, p Ͼ 0.05). With these cell models, we further used real-time PCR to confirm that the knockdown of mapk14induced reduction of BACE1 proteins was not due to the decreased genetic transcription of BACE1 (Fig. 4E, one-way ANOVA, p Ͼ 0.05). The transcription of ps1 was not changed, whereas the transcription of nicastrin was slightly up-regulated (one-way ANOVA for ps1 and nicastrin, p Ͼ 0.05 and ϭ 0.024, respectively).
In additional experiments, we also knocked down p38␣ MAPK expression in APP-transgenic SH-SY5Y cells. We confirmed the decreased protein level of p38␣ MAPK (data not shown) and detected secreted A␤ in the conditional culture medium with quantitative Western blot. As shown in Fig. 4F, significantly less A␤ was released from p38␣ MAPK-deficient cells than from p38␣ MAPK-wild-type control cells (one-way ANOVA, p Ͻ 0.05).
Knockdown of p38␣ MAPK Expression Facilitates Lysosomal Degradation of BACE1 in the Neuronal Cells-As reduction of p38␣ MAPK expression decreased BACE1 proteins in both the mouse brain and cultured cell without down-regulating bace1 gene transcription, we hypothesized that BACE1 degradation was enhanced by p38␣ MAPK inhibition. As predicted, knockdown of p38␣ MAPK with either vector (kd509 or kd709) facilitated the degradation of BACE1 and significantly shortened its lifetime in neuronal cells in which protein translation was blocked by cycloheximide treatment (Fig. 5, A and B; two-way ANOVA, p Ͻ 0.05).
As BACE1 is degraded by lysosomes (38), we investigated BACE1 protein levels in lysosomes isolated from cultured cells (Fig. 5, C and D). We collected 4 times fractions using Percoll density gradient centrifugation. Western blot analysis showed FIGURE 2. Deletion of neuronal p38␣ MAPK reduces cerebral A␤ load in APP-transgenic mice. Four-month-old APP tg p38 fl/wt Nex-Cre ϩ/Ϫ (p38␣ heterozygote, ko) and APP tg p38 fl/wt Nex-Cre Ϫ/Ϫ (p38␣ wild-type, wt) littermate mice were analyzed for cerebral A␤ load after immunofluorescent staining with human A␤-specific antibody (A). The A␤ volume adjusted by relevant brain volume as estimated with Cavalieri method is reduced in p38␣ MAPK-deficient APP mice (B; t test, n ϭ 10 and 8 for p38 ko and wt groups, respectively). The cerebral A␤ in these APP-transgenic mice was also evaluated by detecting A␤ in the brain homogenate with quantitative Western blot (C-E, t test, n ϭ 7 and 8 for p38 ko and wt groups, respectively). C, the figure is grouping images from different parts of the same gel.
that LAMP-1, a lysosome-specific protein, was enriched in Fraction 4, whereas non-lysosomal proteins, such as calnexin and ␤-actin, were nearly absent (Fig. 5C). Thus, we used LAMP-1 as an internal protein control in measuring BACE1 protein in Fraction 4.
Significantly more BACE1 protein were detected in lysosomes derived from p38␣ MAPK-knocked down cells compared with p38␣ MAPK-wild-type cells (Fig. 5, D and E, p Ͻ 0.05). Interestingly, the lower efficiency of BACE1 degradation in kd509-mediated p38␣ MAPK knocked down cells in the lifetime assay (Fig.  5B) was correlated with a lower protein level of BACE1 in lysosomes (Fig. 5E) compared with kd709-transfected knockdown cells, which strongly suggested that p38␣ MAPK knockdown facilitates the lysosomal degradation of BACE1.  A and B), p38 fl/fl Nex-Cre ϩ/Ϫ (ko), p38 fl/wt Nex-Cre ϩ/Ϫ (heterozygote), and p38 fl/wt or p38 fl/fl Nex-Cre Ϫ/Ϫ (wt) littermate mice were used to prepare membrane components, brain homogenates, and total RNA. ␤and ␥-secretase activity was measured by incubating membrane components with fluorogenic ␤and ␥-secretase substrates, respectively (A-D, two-way ANOVA followed by Bonferroni post hoc test between mouse groups with different genotypes; A and B, n Ն 4 per group; C and D, n Ն 10 per group). The protein levels of APP, APP-C99, and BACE1 in the APP tg mouse brain homogenate were determined by quantitative Western blots (E-G, n Ն 4 per group). Moreover, the transcripts of bace1, nicastrin, and ps1 genes in the APP tg mouse brain were measured with real-time PCR, which showed that transcription of bace1 was up-regulated after deletion of p38␣ MAPK specifically in neurons (H, t test; n Ն 4 per group). E or G, the figure is grouping images from different parts of the same gel.

Knockdown of p38␣ MAPK Expression Does Not Change the Phosphorylation of BACE1 in the Neuronal Cells-BACE1
is potentially phosphorylated by p38␣ MAPK at serine and threonine, which might modulate the intracellular trafficking and degradation of BACE1 (39,40). Thus, we asked whether the decreased cerebral A␤ load after the inhibition of neuronal p38␣ MAPK was due to the altered phosphorylation of BACE1. We first utilized NetPhosK 1.0 Server and human BACE1 sequence with UniProt accession number P56817 to predict the phosphorylation sites. We observed that only Ser-83 without any threonine residues in BACE1 was potentially phosphorylated by p38 MAPK. In the following experiments, we only investigated phosphorylation of BACE1 at serine. Phosphorylated serine-specific antibody detected no changes in serine phosphorylation in BACE1 taken from cell lysate derived from p38␣ MAPK-knocked down and wild-type control cells (Fig. 6, A and B).
Deletion of p38␣ MAPK Enhances Autophagy That Is Involved in the Decrease of BACE1 Protein-Phosphorylation of BACE1 was not altered by p38␣ MAPK deficiency, suggesting that another mechanism must facilitate the lysosomal degradation of BACE1. p38 MAPK has been observed to block autophagy by phosphorylating ATG5 (41); we asked whether enhanced autophagy mediated p38␣ MAPK deficiency induced BACE1 degradation.
Autophagic activity was enhanced in APP-transgenic mice with p38␣ MAPK deletion in neurons, compared with control mice with normal p38␣ MAPK. Specifically, both the ratio of LC3-II/I protein and the amount of beclin1 protein detected in RIPA buffer-soluble brain homogenate were significantly higher in APP tg p38 fl/wt Nex-Cre ϩ/Ϫ mice than in APP tg p38 fl/wt -  (kd509 and kd709). The control cell line (kd-ct) was transfected with miRNA-transcribing vectors containing scramble encoding sequence. The protein levels of p38␣ MAPK and BACE1 were determined by quantitative Western blots (A and B, one-way ANOVA followed by Bonferroni post hoc test; n Ն 4 per group). For detection of ␤and ␥-secretase activity, membrane components prepared from different cell lines (kd-ct, kd509, and kd709) were incubated with fluorogenic substrates (C and D, two-way ANOVA followed by Bonferroni post hoc test). The transcripts of bace1, nicastrin, and ps1 genes in the SH-SY5Y cell lines with and without knockdown of the mapk14 gene were also measured with real-time PCR, which showed that transcription of nicastrin was slightly up-regulated after cells were transfected with the vector kd709 (E, one-way ANOVA followed by Tukey post hoc test; n ϭ 4 per group). To investigate the effects of p38␣ MAPK on A␤ generation, SH-SY5Y cells were transfected first with APPtransgenic vectors and then with p38␣ MAPK knockdown vectors (kd-ct, kd509, and kd709). The A␤ in the culture medium was detected with quantitative Western blot and adjusted by APP secretion from the same cell line (E, one-way ANOVA followed by Bonferroni post hoc test, n Ն 6 per group). A, B, or E, grouping images from different parts of the same gel. Nex-Cre Ϫ/Ϫ mice (Fig. 7, A-C, t test, p Ͻ 0.05). Similarly, we observed that deletion of p38␣ MAPK in SH-SY5Y cell lines increased both the ratio of LC3-II/I protein and the amount of beclin1 protein (Fig. 7, D-F, one-way ANOVA, p Ͻ 0.05). To exclude the possibility that impairment in autophagy or lysosomal digestion caused the increase of LC3, we treated LC3-GFP-RFP-transgenic autophagy reporter SH-SY5Y cells (34) with the p38␣ MAPK inhibitor SB203580, and observed that inhibition of p38 MAPK significantly increased the numbers of autophagosomes and autolysosmes in a dose-dependent manner (Fig. 7, G-I, one-way ANOVA, p Ͻ 0.05). We also used Western blot to compare the protein levels of LAMP-1 and Gm2A, which are degraded by lysosomes (42), in p38␣ MAPKdeficient and wild-type SH-SY5Y cells. As shown in Fig. 7, J-L, deletion of p38␣ MAPK does not affect the protein levels of LAMP-1 and Gm2A in SH-SY5Y cells (one-way ANOVA, p Ͼ 0.05). We concluded that deficiency of p38␣ MAPK did enhance neuronal autophagy.
We next investigated the effects of inhibition of autophagy on the decrease in BACE1 protein caused by p38␣ MAPK deficiency. We established cell lines overexpressing wild-type and dominant-negative (DN) human ATG5 by stably transfecting SH-SY5Y cells with the established expression vectors (33). After adding 0.2 g/ml of rapamycin in the culture medium for 24 h, LC3-II/I levels were significantly lower in ATG5DNtransgenic cells than in wild-type ATG5-overexpressing cells, demonstrating ATG5DN-induced inhibition of autophagy (Fig.  8, A and B, t test, p Ͻ 0.05). We then further reduced p38␣ MAPK expression in both ATG5DN and ATG5 (wild-type)transgenic cells by transfecting cells with the miRNA-transcribing vector kd709. We also inhibited autophagy with treatment of 1 mM 3-methyladenine (3-MA) in human ATG5transgenic SH-SY5Y cells with and without knockdown of p38␣ MAPK. Interestingly, both approaches to autophagic inhibition (overexpression of ATG5DN and 3-MA treatment) increased BACE1 protein levels and abolished the p38␣ MAPK knockdown-induced decrease of BACE1 protein in SH-SY5Y cells (Fig. 8C, t test, p Ͻ 0.05).
In further experiments, we investigated the co-localization of BACE1 and autophagic vacuoles under confocal microscopy. BACE1 rarely co-localized with autophagosomes or autolysosomes as stained with antibodies against LC3A/B or p62. Treatment of bafilomycin, which was supposed to block the fusion of autophagosomes with lysosomes (43), did not change the results of our observation. Here we showed the co-localization of BACE1 and LC3A/B within one cell that had not been treated with bafilomycin (Fig. 8D).

Discussion
The pathogenic molecule, A␤, which is produced from the ␤and ␥-cleavages of APP (7), damages neurons and induces inflammatory activation in the brain (1, 2). Our study demonstrates for the first time that partial deletion (via heterozygous knock-out or knockdown) of the inflammation-sensitive kinase, p38␣ MAPK, in neurons reduces A␤ generation and decreases cerebral A␤ load by promoting macroautophagy (hereafter referred to as autophagy)-associated BACE1 degradation.
Microglial inflammatory activation is correlated with disease progression in AD patients (44,45). p38 MAPK, which responds to inflammatory and oxidative stress, is activated in the AD brain (14,15), and mediates A␤-induced inflammatory activation in cultured microglia (21). Studies have suggested that inhibition of p38␣ MAPK retards AD pathogenesis; for example, oral administration of p38␣ MAPK inhibitor successfully reduced neuroinflammation and synaptic dysfunction in APP-transgenic mice (22,46). However, the effect of p38␣ MAPK on A␤ pathology was not examined.
Multiple binding sites for the inflammatory and oxidative stimuli-responsive transcriptional activators, including AP1, AP2, NF-B, STAT1, and SP1 have been identified within the promoter region of the bace1 gene, suggesting that neuroinflammation may increase BACE1 expression (8,9,47,48). Elevated BACE1 protein expression and activity have been observed in sporadic AD cases (49) and during aging (50). The up-regulated expression of BACE1 tends to further increase A␤ generation, creating a BACE1-based vicious cycle in AD pathogenesis (51). Our study demonstrates that reducing p38␣ MAPK expression in neurons facilitates BACE1 degradation and decreases A␤ generation. Moreover, this effect is more apparent in APP-transgenic mice than in their non-APP-transgenic littermates.
Our study further demonstrates that reduction of p38␣ MAPK expression enhances autophagy in neurons. Autophagy is a basic cellular catabolic mechanism, through which the unnecessary or dysfunctional cellular proteins and organelle components are degraded through the lysosomal machinery (52). Growing evidence has shown that autophagy is impaired in AD (53)(54)(55). Histological analysis shows accumulation of autophagosomes and autolysosomes in neurons of AD brains (56). These accumulated autophagic vacuoles potentially provide membrane structures to produce A␤, as BACE1-cleaved APP and ␥-secretase components are enriched in autophagic vacuoles (57). Moreover, impaired autophagy was reported to promote the intracellular accumulation of A␤ and subsequent neuronal degeneration (58). Thus, inhibition of p38␣ MAPK FIGURE 6. p38␣ MAPK deficiency does not change the phosphorylation of BACE1 in SH-SY5Y cells. BACE1 proteins were pulled down from the cell lysate derived from p38␣ MAPK knocked down (kd509 or kd709) or control (kd-ct) cells with anti-BACE1 (C-terminal specific) and protein G; thereafter, they were serially detected with quantitative Western blot using phosphoserine and BACE1-specific antibodies (A and B, one-way ANOVA, p Ͼ 0.05, n ϭ 6 per group). A, grouping images from different parts of the same gel. may act to repair dysfunctional autophagy in the AD brain. Furthermore, increased autophagy is helpful in the degradation of APP and its C-terminal fragment C99, as well as phosphorylated Tau proteins (59 -61).
How p38␣ MAPK regulates autophagy is unclear. p38 MAPK inhibits starvation-induced autophagy by interfering with the binding between ATG9 and p38-interacting protein in HEK293 cells (62), and by phosphorylating ATG5 in fibroblasts and macrophages (41); however, p38 MAPK also mediates interferon-␥ or oxidative stress-triggered autophagy in macrophages or HeLa cells (63,64). Thus, the effect of p38␣ MAPK on autophagy is likely cell type-specific and may even depend on the type of stimuli that activate autophagy. In our neuronal cell models, inhibition of p38␣ MAPK appears to speed delivery of BACE1 to lysosomes in an ATG5-dependent pattern, but the mechanism by which enhanced autophagy facilitates the lysosomal degradation of BACE1 remains unclear. We have observed that BACE1 rarely co-localizes with autophagic vacuoles. This evidence is not sufficient to support the hypothesis that BACE1 is directly degraded through autophagy. However, FIGURE 7. Deletion of p38␣ MAPK enhances autophagy in both the brain and cell. Autophagy activity was evaluated with quantitative Western blot detection of LC3 and beclin1 in neuronal p38␣ MAPK-deleted (ko) and wild-type (wt) APP-transgenic mouse brains (A-C), as well as in p38␣ MAPK knocked down (kd509 or kd709) and control (kd-ct) SH-SY5Y cells (D-F). Deletion of p38␣ MAPK increased both the ratio of LC3-II/I and beclin1 protein levels (B and C: t test, n ϭ 11 per group; E and F, one-way ANOVA followed by Tukey post-hoc test, n Ն 5 per group). Additionally, autophagy-reporting cells expressing the fusion protein of LC3-GFP-RFP were treated with the p38 MAPK inhibitor, SB203580, at different concentrations. Autophagosomes are shown as green puncta (overlap with weak red fluorescence, marked with arrowheads) and autolysosomes are shown as red puncta (marked with arrows) (G). The numbers of autophagosomes and autolysosomes are significantly increased by the treatment with SB203580 in a dose-dependent manner (H and I, one-way ANOVA followed by Dunnett T3 post hoc test; n ϭ 4). The lysosomal digestion in p38␣ MAPK-knocked down (kd509 or kd709) and control (kd-ct) SH-SY5Y cells was evaluated by Western blot detection of LAMP-1 and Gm2A in the cell lysate (J-L, one-way ANOVA, p Ͼ 0.05; n Ն 5 per group). A, D, or J, grouping images from different parts of the same gel. this result may be misleading, as it could not be excluded that the BACE1 epitope for the antibodies used in this study was potentially damaged or lost in the autophagic vacuoles, rendering the intravacuolar BACE1 unrecognizable by the antibodies. Consequently, questions concerning the p38␣ MAPK deficiency promoted autophagic degradation of BACE1 must await clarification of the mechanisms through which BACE1 is delivered to the autophagic pathway.
For transport to lysosomal degradation, BACE1 needs to be carried by its sorting adaptor proteins, especially GGA1 and GGA3 (65,66). BACE1 contains an acidic di-leucine (DXXLL) motif at the C terminus, which binds the VHS domain in the sorting adaptor proteins. This binding is strongly enhanced by phosphorylation of Ser-498 in the DXX(S)LL motif (40,67). We have observed that knockdown of p38␣ MAPK does not affect BACE1 phosphorylation. Thus, we should further investigate whether p38␣ MAPK deficiency reduces the phosphorylation of GGA1 or GGA3 and subsequently alters GGA1 or GGA3mediated BACE1 degradation, in the following projects.
In summary, we have shown that inhibition of p38␣ MAPK in neurons enhances autophagy and promotes BACE1 degradation, which thereby reduces neuronal A␤ generation and decreases cerebral A␤ load. p38␣ MAPK might be a promising target for AD therapy. However, as a signal transduction mediator, p38 MAPK signaling has been implicated in many cellular responses including inflammation, cell cycle, cell death, development, cell differentiation, senescence, and tumor genesis (68). Possible toxic side effects should be carefully evaluated during any translational study of p38 MAPK inhibitors in AD therapy (69).
Author Contributions-Y. L. and K. F. designed research; L. S., W. H., Y. Q., S. L., I. T., X. L., and Y. L. performed research; K. F. contributed analytic tools; L. S. and Y. L. analyzed data; Y. L. wrote the paper.