|Home | About | Journals | Submit | Contact Us | Français|
Deposition of amyloid β protein (Aβ) to form neuritic plaques in the brain is the pathological hallmark of Alzheimer’s disease (AD). Aβ is generated from sequential cleavages of the β-amyloid precursor protein (APP) by the β- and γ-secretases, and β-site APP-cleaving enzyme 1 (BACE1) is the β-secretase essential for Aβ generation. Previous studies have indicated that glycogen synthase kinase 3 (GSK3) may play a role in APP processing by modulating γ-secretase activity, thereby facilitating Aβ production. There are two highly conserved isoforms of GSK3: GSK3α and GSK3β. We now report that specific inhibition of GSK3β, but not GSK3α, reduced BACE1-mediated cleavage of APP and Aβ production by decreasing BACE1 gene transcription and expression. The regulation of BACE1 gene expression by GSK3β was dependent on NF-κB signaling. Inhibition of GSK3 signaling markedly reduced Aβ deposition and neuritic plaque formation, and rescued memory deficits in the double transgenic AD model mice. These data provide evidence for regulation of BACE1 expression and AD pathogenesis by GSK3β and that inhibition of GSK3 signaling can reduce Aβ neuropathology and alleviate memory deficits in AD model mice. Our study suggests that interventions that specifically target the β-isoform of GSK3 may be a safe and effective approach for treating AD.
Alzheimer’s disease (AD) is the most common neurodegenerative disorder leading to dementia. The characteristic neuropathological features of AD include neuritic plaques, neurofibrillary tangles, and neuronal loss. Amyloid β protein (Aβ), the central component of neuritic plaques, is produced from sequential endoproteolytic cleavages of the type 1 transmembrane glycoprotein β-amyloid precursor protein (APP) by β-secretase and γ-secretase. Proteolytic processing of APP at the β site is essential for generating Aβ, and β-site APP-cleaving enzyme 1 (BACE1) is the β-secretase in vivo (1–4). BACE1 cleaves APP at two β-sites, Asp+1 and Glu+11 of the Aβ domain, to generate C99 and C89 fragment, respectively (5). Subsequently, γ-secretase cleaves C99 within its transmembrane domain to release Aβ and APP C-terminal fragment γ (CTFγ). In addition to APP, BACE1 substrates also include other proteins: LRP (6), APLP1 (7), APLP2 (8), ST6Gal I (9), and PSGL-1 (10).
BACE1 expression is tightly regulated at the level of transcription (5, 11, 12) and translation (13–16). It was reported that a G/C polymorphism in exon 5 of the BACE1 gene might be associated with some sporadic cases of AD (17–19). Although genetic analyses from our and other laboratories have failed to uncover any mutation in the BACE1 coding sequence or any disease-associated SNP in its promoter region in AD patients (20–22), increased β-secretase levels and activity have been reported in AD (23–27). BACE1 levels were elevated in neurons around plaques (28). BACE1 mRNA levels tended to increase as miR-107 levels decreased in the progression of AD (29). We reported that hypoxia, a common vascular component among AD risk factors, increased BACE1 expression, resulting in both increased Aβ deposition and memory deficits in AD transgenic mice (30). Recently we found that both NF-κB and BACE1 levels were increased in sporadic AD patients, and NF-κB facilitated BACE1 gene expression and APP processing (27). Thus, increased BACE1 expression by NF-κB signaling in the brain could be one of the mechanisms underlying AD development (27). Together, these studies indicate that BACE1 dysregulation plays an important role in AD pathogenesis.
BACE1 has been considered as one of the major targets for AD drug development. Bace1-knockout mice have abolished Aβ generation (31–33). Suppression of BACE1 by RNA interference reduced APP processing and Aβ production in primary cortical neurons derived from both wild-type and Swedish APP mutant transgenic mice (34), and disruption of the Bace1 gene rescued memory deficits and cholinergic dysfunction in Swedish APP mice (35). Oral administration of a potent and selective BACE1 inhibitor decreased β-cleavage and Aβ production in APP transgenic mice in vivo (36). Bace1-KO mice were reported to display hypomyelination of peripheral nerves and aberrant axonal segregation (37, 38), suggesting that inhibiting β-secretase may have unwanted collateral effects. However, even a partial reduction in BACE1 can have dramatically beneficial effects on AD pathology (39), suggesting that therapeutic inhibition of BACE1 is a valid therapeutic target for AD treatment.
Glycogen synthase kinase 3 (GSK3) is a proline-directed serine/threonine protein kinase originally identified as playing an important role in glycogen metabolism. Since its discovery, many studies have shown that GSK3 has pleiotropic functions, including embryonic development, gene transcription, and neuronal cell function (40). In mammals, two GSK3 isoforms are encoded by distinct genes: GSK3A produces a 51-kDa GSK3α protein and GSK3B a 47-kDa GSK3β protein (41). These two isoforms are highly homologous, sharing greater than 95% amino acid identity in the catalytic domains. Although both isoforms are ubiquitously expressed, the β isoform is expressed at higher levels in neuronal tissues (42).
GSK3 activity is regulated at several levels. Phosphorylation of Tyr279/Tyr216 on GSK3α/β is important for enzymatic activity (43). Inactivation of GSK3 can be achieved through phosphorylation of Ser21/Ser9 residues within the N-terminal domain on GSK3α/β, respectively. Stimulation of tissues by insulin and growth factors activates the PI3K/PKB/Akt signal transduction cascade, leading to phosphorylation of these inhibitory serine residues (44, 45). GSK3 is also regulated upon interaction of the Wnt ligand and its receptor Frizzled and co-receptor LRP5/6. This interaction releases GSK3 from a multi-protein complex formed by β-catenin, axin, and adenomatous polyposis coli (APC) (46, 47), which prevents GSK3-mediated β-catenin degradation and induces β-catenin–dependent gene transcription.
Dysregulation of GSK3 activity has been implicated in AD. Increased GSK3β activity was found in postmortem AD brains (48). GSK3β has been found to phosphorylate the tau protein on various conserved sites and contribute to tau hyperphosphorylation and neurofibrillary tangle formation (49, 50). GSK3α was reported to regulate Aβ production by positively modulating the γ-secretase complex (51), although this finding has recently been challenged (52). Inhibition of GSK3 activity with the commonly known GSK3 inhibitors LiCl and valproic acid in cell culture (53) and animal models of AD decreased Aβ production (51, 54). Although LiCl and valproic acid are known to inhibit GSK3 activity, these compounds also activate a plethora of signaling cascades that could differentially regulate APP processing independent of GSK3 (55–58).
In this study, we examined the effects of GSK3β-specific inhibition on AD neuropathology and behavioral deficits and identified its underlying mechanism. We found that specific inhibition of GSK3β, but not GSK3α, reduced BACE1-mediated cleavage of APP. Treatment with the GSK3 inhibitor AR-A014418 (ARA) reduced neuritic plaque formation and alleviated memory deficits in AD transgenic model mice. Furthermore, we found that GSK3β regulated BACE1 transcription via NF-κB signaling. Our work provides evidence that specific inhibition of GSK3β may be an effective therapeutic approach for treating AD.
Previous studies showed that LiCl and valproic acid modulated GSK3 signaling and reduced Aβ production (51, 54). However, the underlying mechanism is not well defined, and these compounds also have many confounding GSK3-independent effects. To examine the specific effect of GSK3 signaling on APP processing, we applied ARA, a highly selective and potent inhibitor of GSK3, to 20E2 cells, a stable cell line expressing human Swedish mutant APP (5). ARA at 1, 2.5, and 5 μM significantly decreased the levels of the β-secretase cleavage product APP C99 to 95.7% ± 1.6%, 66.4% ± 0.7%, and 31.3% ± 0.4%, respectively (P < 0.001 by 1-way ANOVA) (Figure (Figure1A1A and B). The treatment had no significant effect on APP expression (Figure (Figure1A).1A). The Aβ ELISA was performed to assess the levels of Aβ40 and Aβ42 in the conditioned media of 20E2 cells. ARA markedly reduced Aβ generation in a dose-dependent manner. Aβ40 was decreased to 99.2% ± 3.8%, 73.2% ± 4.6%, and 48.7% ± 1.3% with 1, 2.5, and 5 μM of ARA treatment, respectively (P < 0.05) (Figure (Figure1C);1C); and Aβ42 was reduced to 99.2% ± 3.1%, 73.2% ± 3.5%, and 48.7% ± 4.5% with 1, 2.5 and 5 μM of ARA treatment, respectively (P < 0.01 by 1-way ANOVA) (Figure (Figure1D).1D). As expected, inhibition of GSK3 stabilized β-catenin (Figure (Figure1A),1A), and ARA treatment resulted in a significant increase in β-catenin levels to 152.7% ± 11.1%, 221.3% ± 17.0%, and 233.7% ± 25.0% with 1, 2.5, and 5 μM, respectively (P < 0.05). These data indicate that specifically inhibiting GSK3 reduced BACE1-mediated APP processing and C99 and Aβ production.
To further examine the effect of GSK3 on β-secretase cleavage of APP and Aβ production, we pharmacologically blocked γ-secretase activity with the γ-secretase–specific inhibitor L658,458 in 20E2 cells while co-treating with the GSK3 inhibitor ARA and G2. G2 is another structurally distinct GSK3 inhibitor, with an IC50 of 304 nM (59). As expected, γ-secretase inhibition resulted in markedly increased generation of the APP CTFs C83 and C99 (Figure (Figure1E).1E). Addition of ARA or G2 reduced C99 levels to 78.2% ± 1.1% and 63.9% ± 6.6% (Figure (Figure1,1, E and F). ARA treatment had no significant effect on Notch cleavage (Supplemental Figure 1B; supplemental material available online with this article; doi: 10.1172/JCI64516DS1). These data demonstrated that specific inhibition of GSK3 reduced β-secretase cleavage of APP to generate C99 and Aβ production in cells.
Our study has shown that GSK3 regulated β-secretase processing of APP, an essential step for Aβ generation. Since BACE1 is the β-secretase in vivo, we first examined whether GSK3 affects BACE1 gene expression. GSK3 has two highly homologous isoforms, GSK3α and GSK3β. We used RNA interference to specifically knock down the expression of GSK3α or GSK3β isoforms in human neuroblastoma SH-SY5Y cells to determine whether both isoforms or one of the isoforms play a major role in regulating BACE1 gene expression. Specific knockdown of GSK3β expression by the siRNA significantly reduced BACE1 mRNA levels to 76.0% ± 6.6% as compared with control (P < 0.05) (Figure (Figure2,2, A and B), whereas knockdown of GSK3α did not affect BACE1 mRNA expression (P > 0.05) (Figure (Figure2,2, A and B). Additional exogenous expression of GSK3β rescued the reduction of BACE1 expression resulting from knockdown of GSK3β by siRNA (Supplemental Figure 2). To examine whether specific knockdown of GSK3β expression also affected β-secretase processing of APP, we transfected siRNA specific to GSK3α or GSK3β into 20E2 cells while inhibiting γ-secretase activity with L658,458. Western blotting showed that GSK3α or GSK3β siRNA specifically reduced GSK3α or GSK3β expression, respectively (Figure (Figure2C).2C). Knockdown of GSK3α expression did not significantly affect BACE1-mediated APP processing, whereas GSK3β knockdown reduced the production of the BACE1 cleavage product C99 to 51.9% ± 6.4% (P < 0.05) (Figure (Figure2,2, C and D).
To further confirm that GSK3β regulates BACE1 gene expression at the transcription level, we assayed BACE1 mRNA in the S9A-GSK3β inducible SHSY5Y stable cell line. This stable cell line carries the constitutively active mutant GSK3β, and expression of the mutant gene is under control of a tetracycline-inducible promoter (60). Addition of tetracycline induced the expression of active GSK3β (Figure (Figure2E)2E) and resulted in significantly increased expression of BACE1 to 251.1% ± 70.0% relative to the control (P < 0.05) (Figure (Figure2F).2F). These data demonstrated that GSK3β, but not GSK3α, specifically regulates BACE1 gene expression and contribute to APP processing.
To investigate the molecular mechanism underlying the effect of GSK3β on BACE1 gene expression at the transcription level, we performed a BACE1 gene promoter assay. Two human BACE1 gene promoter deletion plasmids were constructed. Regions of the BACE1 promoter from –2890 to +292 bp (pB1-A) and from –9 to +292 bp (pB1-B) were inserted into promoterless vector pGL3-basic upstream of the firefly luciferase reporter gene (Figure (Figure3A).3A). To examine the effect of GSK3 on BACE1 gene promoter activity, we transfected N2a cells with pB1-A and then treated them with ARA. Inhibition of GSK3 signaling by ARA significantly decreased the promoter activity of pB1-A in N2a cells to 54.9% ± 6.2% (P < 0.005) (Figure (Figure3B).3B). To further investigate the underlying mechanism and determine the BACE1 promoter region that mediates the transcriptional activation by GSK3 signaling, we co-transfected N2a cells with either pB1-A or pB1-B plasmid together with the S9A-GSK3β plasmid, which carries a constitutively active form of GSK3β. Expression of active GSK3β markedly increased the luciferase activity of pB1-A in the S9A-GSK3β transfected cells to 144.0% ± 2.0% as compared with the control (P < 0.05), but had no significant effect on the luciferase activity of pB1-B (P > 0.05) (Figure (Figure3C).3C). This result showed that enhancing GSK3β signaling upregulated BACE1 gene promoter activity and the 2.8-kb promoter region is responsible for GSK3β-mediated upregulation of BACE1 transcription. To further confirm this finding, we transfected these two deletion promoter plasmids into Gsk3b-KO or wild-type cells. Ablation of Gsk3b expression in the Gsk3b-KO cells resulted in significant reduction in luciferase activity of pB1-A to 47.0% ± 7.5% as compared with the wild-type control cells (P < 0.01) (Figure (Figure3D).3D). However, knockout of the Gsk3b gene had no significant effect on the luciferase activity of pB1-B (Figure (Figure3D).3D). Taken together, these results demonstrated that GSK3β regulates BACE1 gene expression via its effect on the BACE1 promoter.
Recently we reported that levels of both BACE1 and NF-κB are increased in AD brains, and NF-κB signaling upregulates human BACE1 gene expression by acting on the cis-acting p65-binding element in its promoter (27). Previous studies in Gsk3b-KO mice demonstrated that GSK3β is required for proper activation of NF-κB (61). As our data indicate that GSK3β regulates BACE1 gene transcription, we next examined whether regulation of BACE1 transcription by GSK3β signaling is dependent on NF-κB. TNF-α is a strong activator of NF-κB p65 expression. In order to examine the specific role of NF-κB in GSK3β-regulated BACE1 transcription, we transfected N2a cells with a pBACE1-4κB promoter plasmid that contained only the NF-κB–binding elements in the human BACE1 promoter (27). After transfection, the cells were cotreated with TNF-α and the GSK3 inhibitor ARA. ARA treatment alone consistently reduced BACE1 promoter activity (Figure (Figure4A).4A). TNF-α stimulation increased pB1A promoter activity to 132.6% ± 7.5% of the control (P < 0.01). However, ARA treatment reduced the TNF-α–induced BACE1 promoter activation to 115.6 ± 6.9% as compared with TNF-α treatment alone (P < 0.05) (Figure (Figure4A).4A). Overexpression of p65 NF-κB significantly increased the BACE1 promoter activity to 487.1% ± 102.2% as compared with the control (P < 0.001) (Figure (Figure3D).3D). However, ARA did not affect NF-κB p65’s upregulating effect on BACE1 promoter activity (P > 0.05), indicating that GSK3 signaling may have its effect upstream of NF-κB in the modulation of BACE1 transcription (Figure (Figure4B).4B). To further confirm ARA’s effect on NF-κB activity, we co-transfected N2a cells with pNF-κB–Luc and NF-κB p65 or an empty vector, followed by treatment with ARA to inhibit GSK3 signaling. ARA reduced pNF-κB-Luc promoter activity to 51.9% ± 9.1% as compared with control (P < 0.05) (Figure (Figure4C).4C). Overexpression of NF-κB p65 significantly increased pNF-κB–Luc promoter activity to 203.36 ± 87.38–fold (P < 0.001) (Figure (Figure4C).4C). Additional ARA treatment did not have any significant effect on the transcriptional activation of pNF-κB–Luc by NF-κB p65 overexpression (P > 0.05) (Figure (Figure4C).4C). These data indicate that the p65-binding elements in the BACE1 promoter mediated the effect of GSK3β on BACE1 transcription.
To confirm the role of the NF-κB signaling pathway in GSK3β-mediated BACE1 transcription, we transfected NF-κB p65 knockout RelA-KO cells and wild-type control cells with the BACE1 promoter pB1-A and the constitutively active GSK3β expression plasmid S9A-GSK3β. In wild-type cells, activation of GSK3β signaling by expression of S9A-GSK3β markedly increased BACE1 promoter activity to 189.6% ± 20.9% as compared with the vector control (P < 0.05), whereas activation of GSK3β signaling by expression of S9A-GSK3β had no effect on BACE1 promoter activity in RelA-KO cells (P > 0.05) (Figure (Figure4D),4D), indicating that disruption of NF-κB p65 expression in RelA-KO cells abolished GSK3β’s effect on the transcriptional activation of the human BACE1 gene promoter. To further examine the effect of GSK3 inhibition on NF-κB p65 expression, we treated N2a cells with ARA and subjected them to subcellular fractionation (Figure (Figure4E).4E). ARA treatment significantly reduced nuclear NF-κB p65 levels to 46.9% ± 9.1% as compared with control (P < 0.01) (Figure (Figure4F).4F). Moreover, ARA treatment also reduced cytosolic NF-κB p65 levels to 56.1% ± 7.1% as compared with control (P < 0.001) (Figure (Figure4G).4G). These data show that GSK3 inhibition with ARA reduced NF-κB activity by decreasing NF-κB p65 levels. Taken together, our results demonstrate that NF-κB signaling mediates the regulatory effect of GSK3β on BACE1 gene expression.
Our data provide strong evidence that GSK3β activates BACE1 gene expression, resulting in enhanced β-secretase processing of APP and Aβ production in vitro. To examine the effect of GSK3 signaling on BACE1 gene expression and APP processing in vivo, we first assayed APP CTFs and Aβ production in the brain of APP23/PS45 mice by Western blot analysis (Figure (Figure5,5, A and B). APP23/PS45 double transgenic mice, an AD mouse model, were generated by crossing APP23 mice, carrying the human Swedish mutant APP751 transgene driven by the neuron-specific Thy1.2 promoter, and PS45 mice, carrying the human familial AD–associated G384A mutant presenilin-1 (PS1) (54). The mice were treated with 5 mg/kg of the GSK3 inhibitor ARA at 6 weeks of age daily for 4 weeks, while age-matched control mice received vehicle solution. ARA treatment significantly decreased the brain levels of the β-secretase–generated CTFβ fragments C99 and C89, to 38.4% ± 4.8% relative to controls (P < 0.05) (Figure (Figure5B).5B). The levels of Aβ40 and Aβ42 were reduced to 71.2% ± 8.3% and 65.6% ± 11.0% in ARA-treated mice relative to controls, (P < 0.05) (Figure (Figure5,5, C and D). These data demonstrate that inhibition of GSK3 activity by ARA treatment reduces β-secretase cleavage of APP and Aβ production in vivo.
We then examined whether the level of BACE1 was altered by GSK3 signaling in vivo. Western blot analysis showed that inhibition of GSK3 signaling significantly reduced the protein level of BACE1 to 64.7% ± 7.3% in ARA-treated mice, as compared with the control mice (P < 0.01), and the treatment had no significant effect on APP and PS1 protein levels (P > 0.05) (Figure (Figure5,5, A and B). We further showed that GSK3 inhibition with ARA reduced tau phosphorylation, which is consistent with previous findings (ref. 62 and Supplemental Figure 1A). Our in vitro study has shown that GSK3β regulates the transcription of the BACE1 gene. To confirm that the decrease in BACE1 protein level seen in the brains of the ARA-treated mice was due to reduced BACE1 gene transcription, the endogenous Bace1 mRNA levels were measured (Figure (Figure5,5, E, G, and I). ARA treatment markedly reduced Bace1 mRNA level to 36.7% ± 11.8% (P < 0.01) (Figure (Figure5F),5F), but did not significantly change the mRNA levels of App and PS1 genes (P > 0.05) (Figure (Figure5,5, H and J). These data demonstrate that, consistent with the in vitro results, inhibition of GSK3 specifically inhibited Bace1 gene expression and its β-secretase activity in vivo.
Our study has demonstrated that NF-κB signaling is required for GSK3β’s regulatory effect on BACE1 gene expression in vitro. To confirm this effect in vivo, we examined whether NF-κB activity was affected in APP23/PS45 double transgenic mice administered ARA. EMSA was used to assess NF-κB consensus DNA binding in whole brain lysates. ARA treatment inhibited the binding of NF-κB p65 protein to the cis-acting consensus oligonucleotide probe, resulting in a reduction in the intensity of the p65 NF-κB shifted bands (Figure (Figure6A).6A). The specificity of the bands was confirmed by competition assay with addition of ×10 and ×100 unlabeled wild-type NF-κB oligonucleotides (oligos), while the mutant NF-κB oligos did not have any significant effect on the shifted bands (Figure (Figure6B).6B). Consistent with the in vitro experiment, APP23/PS45 mice receiving ARA treatment showed a significant reduction in NF-κB levels to 62.9% ± 6.9% of those in the sham-injected controls (P < 0.005) (Figure (Figure6,6, C and D). Thus, inhibition of GSK3 signaling attenuates NF-κB binding to cis-acting p65 binding elements by reducing NF-κB levels in the AD transgenic model mice in vivo.
To examine the specific effect of GSK3 signaling on AD pathogenesis, we treated APP23/PS45 double transgenic mice with ARA. The double transgenic mice develop detectable neuritic plaques in the neocortex and hippocampus as early as 1 month of age. The mice were treated with 5 mg/kg ARA at 6 weeks of age daily for 4 weeks, while age-matched control mice received vehicle solution. 4G8 immunostaining and thioflavin-S staining were used to detect Aβ-containing neuritic plaques in the brains (ref. 63 and Figure Figure7).7). ARA treatment significantly decreased the number of neuritic plaques in the transgenic mice relative to the vehicle-injected group (Figure (Figure7,7, A and B). Quantification showed that overall ARA treatment reduced plaque number by approximately 50% (23.8 ± 4.7 vs. 10.4 ± 1.3 per slice, P < 0.01) (Figure (Figure7C).7C). Thioflavin-S staining also confirmed that ARA treatment dramatically reduced the number of Aβ-containing neuritic plaques in the brains of APP23/PS45 double transgenic mice (Figure (Figure7,7, D and E). The inhibitory effect of ARA was reversible, and the treated mice (Figure (Figure7G)7G) had plaque numbers similar to those in the control mice (Figure (Figure7F)7F) when examined 3 months after the end of the drug treatment (36.5 ± 5.1 vs. 32.1 ± 4.2, P > 0.05) (Figure (Figure7H).7H). During the 4-week injection period, ARA treatment of the mice did not affect food consumption, and no significant weight changes were observed between the treatment and control groups (data not shown).
To investigate whether GSK3 inhibition by ARA treatment affects the memory deficit in AD pathogenesis, we used the Morris water maze to test spatial memory after APP23/PS45 mice received 1 month of ARA treatment (64). In the visible platform tests, ARA-treated and control APP23/PS45 mice had similar escape latency (42.4 ± 4.3 and 44.8 ± 2.7 seconds, P > 0.05) (Figure (Figure8A)8A) and path length (6.6 ± 0.7 m and 6.1 ± 0.5 m, P > 0.05) (Figure (Figure8B),8B), indicating that ARA treatment did not affect mouse mobility or vision. In the hidden platform test, ARA-treated mice showed significant improvements as compared with the vehicle-treated controls. The escape latency on the third and fourth days of the hidden platform test was shorter (12.0 ± 1.6 and 17.4 ± 3.1 seconds) than that of sham-treated mice (23.8 ± 4.0 s and 24.9 ± 2.7 s) (P < 0.05, Figure Figure8C).8C). The ARA-treated mice swam significantly shorter distances to reach the platform (2.9 ± 0.7 m and 3.2 ± 0.3 m) as compared with control mice (3.7 ± 0.6 and 4.2 ± 0.5 m) on the third and fourth days (P < 0.05, Figure Figure8D).8D). In the probe trial on the last day of testing, the platform was removed. ARA treatment significantly improved the spatial memory in the APP23/PS45 mice, as these mice spent more time searching for the platform in the appropriate quadrant (Supplemental Figure 3A). The number of times the mice traveled into the third quadrant, where the hidden platform was previously placed, was significantly greater with ARA treatment as compared with control (6.0 ± 1.0 and 2.0 ± 1.2 times, P < 0.05) (Figure (Figure8E).8E). Moreover, ARA-treated mice spent nearly twice as long in the southwest (SW) quadrant, where the platform was originally placed during the hidden platform test (NE 16.3.9% ± 1.3%, SE 19.2% ± 1.6%, SW 44.0% ± 3.1%, NW 20.4% ± 2.0%). The sham-treated mice, however, spent approximately equal amounts of time in each quadrant (NE 24.9% ± 1.9%, SE 20.4% ± 2.0%, SW 28.2% ± 1.7%, NW 26.6% ± 2.1%) (Supplemental Figure 3B). These data demonstrate that inhibition of GSK3 signaling significantly improved the memory deficits seen in the AD model mice.
GSK3 signaling has been previously shown to be strongly associated with several neuropathological changes of AD. GSK3 is involved in tau hyperphosphorylation (65), neuronal apoptosis (66), and synaptic dysfunction (67). Moreover, GSK3 signaling has also been implicated in Aβ production and neuritic plaque formation (51, 54, 68). However, the mechanism by which GSK3 affects APP processing and Aβ production has not been clearly elucidated, and its pharmaceutical potential as an effective drug target remains to be validated. The current study clearly indicates that GSK3β, but not the related GSK3α isoform, facilitates Aβ production by upregulating BACE1 gene expression via NF-κB p65 cis-acting elements on the BACE1 gene promoter. The C-terminal end of BACE1 undergoes phosphorylation, which regulates BACE1 trafficking and cellular levels (69, 70). In our study, inhibition of GSK3 signaling did not affect the phosphorylation status of BACE1 (data not shown). Thus, specifically inhibiting GSK3 decreases BACE1 expression and markedly reduces β-secretase processing of APP and Aβ production, resulting in inhibition of neuritic plaque formation and amelioration of memory deficits in AD model mice.
GSK3 appears to be a common molecular link between amyloidogenesis and tau abnormalities in AD pathology. Therefore, GSK3 inhibition has been proposed to be a valid therapeutic target for treating AD. Lithium chloride and valproic acid are known to have some inhibitory effects on GSK3 (71, 72) and have been used in the clinic for many decades for the treatment of bipolar disorders and epilepsy. Recent work has demonstrated that lithium and valproic acid could reduce Aβ levels and improve cognitive performance in mouse models of AD (54, 68). While the inhibitory effects of lithium and valproic acid on GSK3 are known, both compounds have also been found to affect signaling cascades independent of GSK3. To further investigate the role of GSK3 on APP processing, we administered ARA, a highly selective and potent inhibitor of GSK3 (IC50 = 104 ± 27 nM) (73) to APP Swedish stable cell lines and APP23/PS45 mice. We found that with specific GSK3 inhibition, the BACE1 major product C99 is reduced, accompanied by a significant reduction in Aβ levels. Previous reports indicated that ARA treatment reduced tau hyperphosphorylation and tangle formation in transgenic mice overexpressing human tau, a model of AD (62). Our study shows that ARA also reduces plaque pathology and rescues cognitive deficits in an AD mouse model. However, continuous ARA treatment is required in order to reduce plaque pathology, as APP23/PS45 mice sacrificed 3 months after the last injection had plaque levels similar to those in the control group.
Previous work by Phiel et al. (51) demonstrated that inhibition of GSK3α with isoform-specific siRNA decreases γ-secretase activity but found no effect of GSK3β. In our study, we found that in a system in which γ-secretase activity is pharmacologically inhibited, GSK3α knockdown did not have any significant effect on APP processing. On the other hand, GSK3β knockdown with inhibition of γ-secretase activity reduced the level of C99 fragment, indicating that BACE1 activity is affected. We further provided evidence that knocking down GSK3β reduced BACE1 mRNA levels, but knocking down GSK3α did not affect BACE1 expression. Previous genetic studies have found that during development GSK3α could not compensate for the loss of GSK3β, as Gsk3b-KO mice are embryonic lethal while Gsk3a-KO mice are viable (61). This argues that GSK3α and GSK3β isoforms may have distinct cellular functions with respect to BACE1 expression. Moreover, GSK3β is the predominant isoform in the brain and therefore has been implicated in many CNS disorders (74). However, Jaworski et al. recently showed the GSK3α/β knockout mice did not show any changes to APP processing and Aβ production (52). This could be due to a physiological compensation of BACE1 expression in the KO mice compared with the acute effect of the GSK3 inhibitor. Additionally, our data showed that GSK3β regulated NF-κB–mediated BACE1 expression. Gsk3a knockout had no effect on NF-κB signaling. Taken together, these studies suggest that GSK3 isoforms have distinct roles in regulating APP processing. We do not believe that gene silencing of GSK3 isoforms would be a valid approach for treating AD. Complete ablation of each isoform may lead to compensatory elevated activity of the other isoform or even other functionally protein kinases. Moreover, it has been reported that complete ablation of GSK3α/β expression in mice results in various abnormalities (75), which may interfere with any therapeutic values.
Epidemiological and experimental studies have suggested a significant inflammatory component in AD (76–79). However, it remains controversial whether neuroinflammation is a driving force for AD or simply a byproduct of the disease (80). Moreover, the detrimental role of astrocyte and microglia during neuroinflammation remains elusive. The release of various proinflammatory and anti-inflammatory cytokines has been shown to affect APP processing, Aβ production, and neurodegenerative changes in AD brains. There have been preclinical studies to show that nonsteroidal anti-inflammatory drugs (NSAIDs) or anti-inflammatory cytokines have beneficial effects in treating AD (80, 81). However, there are reports demonstrating that anti-inflammatory stimuli could potentiate Aβ production, possibly through suppression of glia-mediated Aβ clearance. On the other hand, proinflammatory stimuli could also have extreme effects on AD pathogenesis. There is a large amount of evidence to show that increased proinflammatory cytokines correlates with Aβ production and neurodegeneration. In contrast, there are also a number of studies indicating that proinflammatory cytokines precondition the system for Aβ challenge and protect against neurodegeneration (82–85). Clearly, inflammation is a major component of the AD brain, and more work will be required in order to elucidate its role in AD pathogenesis. Since several proinflammatory cytokines including the interleukins and TNF have been found to activate GSK3β, and GSK3 inhibitors showed anti-inflammatory effects (86–88), we speculate that GSK3 plays an important role in the inflammatory response driving Aβ production.
The chronic inflammatory response induced by Aβ may depend on transcription factors to exert neurodegenerative effects. Regulation of the transcription factor NF-κB and its transcriptional activity may in part play a role in Aβ-mediated neurodegeneration (27, 89). It has been previously reported that the BACE1 promoter contains NF-κB–binding elements (27, 90) and exacerbated Aβ levels modulate the BACE1 promoter activity via NF-κB–dependent pathways (91). More recently, Chen et al. (27) found increased NF-κB p65 and BACE1 expression in postmortem AD brains. Furthermore, overexpression of NF-κB p65 was found to increase human BACE1 promoter activity, whereas inhibiting NF-κB signaling reduced BACE1 expression (27).
Exogenous application of Aβ has been found to increase GSK3β activity and NF-κB levels. There are many reports suggesting that GSK3β regulates gene transcription in an NF-κB–dependent manner (61, 92, 93). Therefore, we argue that the GSK3β/NF-κB signaling pathway is involved in regulating BACE1 gene transcription. In this study, we found that interfering with GSK3β activity using Gsk3b-KO cells or ARA treatment reduced BACE1 promoter activity and gene expression. Moreover, disrupting NF-κB expression also blocked GSK3β-induced BACE1 transcription.
Long-term treatment with GSK3 inhibitors always raises the concern about unwanted side effects. Many of the Wnt signal transduction components such as the APC gene or β-catenin gene carry mutations that render the protein nondegradable (94). Prolonged activation of the Wnt pathway has been linked to development of various cancers (95). GSK3 inhibitors including SB216763, SB415286, Kenpaullone, and lithium chloride have all been shown to elevate the level of β-catenin (51, 96, 97). Consequently, it may be possible that GSK3 inhibitors may mimic Wnt signaling and could potentially be oncogenic. However, long-term use of lithium chloride to treat bipolar disorder has not been linked to increased risk of cancer. Interestingly, administration of the GSK3 inhibitors CHIR99021 and ARA did not change β-catenin levels (65). Moreover, β-catenin levels are largely unaffected in Gsk3b-KO embryos (61). These findings indicate that the inhibition of GSK3 by itself might not be sufficient to elevate the level of β-catenin. It is possible that there are specific pools of GSK3 that are insensitive to certain GSK3 inhibitors, thereby sparing the effect of increasing β-catenin levels. In order to validate the therapeutic efficacy of GSK3 inhibition, while minimizing oncogenic side effects, long-term treatment studies in cells and animals will be required.
In conclusion, we have found that the GSK3β/NF-κB signaling pathway regulates BACE1 transcription and thereby facilitates Aβ production in AD. Since GSK3β and NF-κB are involved in AD pathogenesis, our results suggest that direct interference of this pathway may be a promising drug target for AD therapy.
All cells were maintained at 37°C in an incubator containing 5% CO2. HEK293 (human embryonic kidney), N2a (mouse neuroblastoma), SHSY5Y (human neuroblastoma), and wild-type MEFs (mouse embryonic fibroblasts) were maintained in complete DMEM supplemented with 10% fetal bovine serum, 1% l-glutamine, 1% penicillin/streptomycin, and 1% sodium pyruvate (Invitrogen). Gsk3b-KO fibroblast cells were derived from E12.5 Gsk3b-KO mouse embryos and were maintained in complete DMEM (61). The 20E2 cell line is a Swedish mutant APP695 stable HEK293 cell line cultured in complete DMEM with 50 μg/ml geneticin (98). 293B2 cells stably expressing human BACE1 in HEK293 cell line were cultured in complete DMEM with 100 μg/ml Zeocin (99). The RelA-KO fibroblast cell line, derived from E12.5–E14.5 mouse embryo fibroblasts, was maintained in DMEM supplemented with 15% FBS, β-mercaptoethanol, and ESGRO (LIF) (100). S9A-GSK3β stably transfected SHSY5Y cells, under the control of the tetracycline-regulated mammalian expression T-Rex system (Invitrogen), were maintained in complete DMEM (60). S9A-GSK3β gene induction was achieved by stimulation with 1 μg/ml tetracycline (60). Cells were transfected with plasmid DNA using either calcium phosphate or Lipofectamine 2000 (Invitrogen). L685,458, G2, and ARA (EMD Biosciences) were dissolved in DMSO and diluted with complete cell culture medium. The final DMSO concentrations in each experiment were less than 0.5%.
BACE1 promoter constructs were transfected into N2a cells and wild-type and Gsk3b-KO MEFs. The Renilla (sea pansy) luciferase vector pCMV-Rluc was co-transfected to normalize transfection efficiency. A luciferase assay was performed 48 hours after transfection with the Dual-Luciferase Reporter Assay system (Promega) as previously described (27).
RNA was isolated from cells using TRI Reagent (Sigma-Aldrich). A Thermoscript Reverse Transcription kit (Invitrogen) was used to synthesize the first-strand cDNA from an equal amount of RNA following the manufacturer’s instructions. The newly synthesized cDNA templates were further amplified via Platinum Taq DNA polymerase in a 20-μl reaction. The following primers were used to specifically amplify BACE1, GSK3A, GSK3B, APP, and PS1 genes: BACE1 forward 5′-cccgcagacgctcaacatcc and reverse 5′-GCCACTGTCCACAATGCTCTT; GSK3A forward 5′-TGAAGCTGGGCCGTGACAGCGG and reverse 5′-ACATGTACACCTTGACATAG; GSK3B forward 5′-TCAGGAGTGCGGGTCTTCCGAC and reverse 5′-CTCCAGTATTAGCATCTGACGCT; APP forward 5′-GCTGGCCTGCTGGCTGAACC and reverse 5′-GGCGACGGTGTGCCAGTGAA; PS1 forward 5′-GAGACACAGGACAGTGGTTCTGG and reverse 5′-GGCCGATCAGTATGGCTACAAA. β-Actin was used as an internal control. The samples were resolved and analyzed on a 1.2% agarose gel.
Brain tissues or cells were lysed in RIPA lysis buffer (1% Triton X-100, 1% sodium deoxycholate, 4% SDS, 0.15 M NaCl, 0.05 M Tris-HCl, pH 7.2) supplemented with 200 mM sodium orthovanadate, 25 mM β-glycerophosphate, 20 mM sodium pyrophosphate, 30 mM sodium fluoride, 1 mM PMSF, and a complete Mini Protease Inhibitor Cocktail Tablet (Roche Diagnostics). The samples were diluted in 4× SDS-sample buffer, boiled, and resolved on 12% Tris-glycine SDS-PAGE or 16% Tris-tricine SDS-PAGE, followed by transfer to polyvindylidine fluoride (PVDF-FL) membranes. For immunoblot analysis, membranes were blocked for 1 hour in PBS containing 5% nonfat dried milk, followed by overnight incubation at 4°C in primary antibodies diluted in the blocking medium. Rabbit anti-APP C-terminal polyclonal antibody C20 was used to detect APP and its CTF products. PS1 was detected by anti-PS1 N-terminal antibody 231. Total GSK3β was determined using a pan-specific mouse anti-GSK3α/β antibody (BioSource International Inc.). Total β-catenin was determined using a pan-specific rabbit anti–β-catenin antibody (Cell Signaling Technology). The NF-κB p65 subunit was determined using mouse anti-p65 (Sigma-Aldrich). Internal control β-actin was analyzed using monoclonal antibody AC-15 (Sigma-Aldrich). BACE1 was detected with the anti-BACE1 antibody 208 recognizing the C-terminal end (101).
HEK293 cells stably expressing the human APP gene containing the Swedish mutation were maintained in cell culture medium supplemented with 1% FBS. Following ARA treatment for 24 hours, conditioned medium was harvested, and protease inhibitors and AEBSF (Roche Diagnostics) were added to prevent degradation of Aβ peptides. APP23/PS45 double transgenic mouse cortical tissues were prepared according to the manufacturer’s instructions prior to carrying out the ELISA protocol (Invitrogen). The concentration of Aβ40 and Aβ42 were detected using an Aβ1–40 or Aβ1–42 Colorimetric ELISA kit (Invitrogen) according to the manufacturer’s instructions.
Whole brain extracts were prepared by homogenizing the tissue in Buffer C (20 mM HEPES pH 7.5, 400 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 1 mM PMSF, 10% glycerol) supplemented with protease inhibitor. EMSA was performed as previously described with a few changes (102). Briefly, 20–40 μg protein was incubated with an IRDye 700–labeled NF-κB oligo (5′-AGTTGAGGGGACTTTCCCAGGC), and the gels were scanned using the Odyssey system (LI-COR Biosciences). In the competition assay, unlabeled wild-type and mutant (5′-AGTTGAGGCCACTTTCCCAGGC) NF-κB oligos at ×10 and ×100 molar excess were used to compete for binding.
APP23 mice overexpressing the Swedish APP751 (KM→NL) mutant transgene driven by the mouse Thy1.2 promoter were originally generated at Novartis Pharma as previously described (30, 103). APP23/PS45 double transgenic mice were generated by crossing APP23 mice with mice overexpressing the human G384A-mutated PS1 under the control of the murine Thy1 promoter (B6,D2-TgN(Thy1-PS1G384A)45) (54, 103–105). APP23 and PS45 mice had been backcrossed to C57BL/6 mice for more than 7 generations prior to breeding for the double transgenic mice. The genotype of the mice was confirmed by PCR using DNA from tail tissues. The treatment group contained 14 animals (8 females, 6 males), and the sham-injected group contained 12 animals (7 females, 5 males). The treatment group received 5 mg/kg ARA diluted in 0.9% saline daily via intraperitoneal injection at the same time each day. Mice in the control group were injected with the vehicle solution containing DMSO diluted in 0.9% saline only. We tabulated the daily food consumption and weight for each mouse.
Mice were sacrificed after behavioral testing, and half brains were immediately homogenized for protein, RNA, or DNA extraction. The other halves of the brains were fixed in 4% paraformaldehyde and sectioned with a Leica Cryostat to 30-μm thickness. Every twelfth slice with the same reference position was mounted onto slides for staining. Immunocytochemical staining was performed on floating sections. The plaques in the sections were detected with biotinylated monoclonal 4G8 antibody (Signetlabs) in 1:500 dilution, visualized by the ABC and DAB method, and counted under a ×40 objective as previously described (54, 63). Plaques were quantified, and the average plaque count per slice was recorded for each mouse. Thioflavin-S staining of plaques was performed with 1% thioflavin-S visualized using fluorescence microscopy (54).
The Morris water maze test was performed as previously described (30, 64). The APP23/PS45 mice treated with or without ARA were subjected to the Morris water maze test 1 day after the last injection regimen. The test was performed in a 1.5-meter-diameter pool with a 10-cm-diameter platform placed in the SE quadrant of the pool. The procedure consisted of 1 day of visible platform tests and 4 days of hidden platform tests, plus a probe trial 24 hours after the last hidden platform test. In the visible platform test performed on the first day, the mice were tested for 5 continuous trials, with an inter-trial interval of 60 minutes. In the hidden platform tests, mice were trained for 5 trials, with an inter-trial interval of 1 hour. Each mouse was allowed 60 seconds to search for the platform. In the scenario where the mouse could not locate the platform, the experiment guided it to the platform and allowed the mouse to rest there for 15 seconds. On the last day of the test, each mouse was subjected to the probe trial in which the platform was removed. Each mouse was given 60 seconds to locate where the platform was originally placed. Mouse behavior including distance traveled, escape latency and the number of passes through the platform was recorded by automated video tracking (ANY-maze, Stoelting). The data were analyzed by 2-way ANOVA, followed by Tukey’s post hoc test.
All results were presented as mean ± SEM and analyzed by ANOVA or 2-tailed Student’s t-test. Statistical significance is accepted when P < 0.05.
Animal experiments were conducted in accordance with the University of British Columbia Animal Care and Use Committee and Canadian Institutes of Health Research (CIHR) guidelines.
We thank Steven R. Vincent and Jing Zhang for helpful comments. This work is supported by the Canadian Institutes for Health Research (CIHR) (MOP-97825 and TAD-117948), the National Natural Science Foundation of China (NSFC) (NSFC81161120498), and the Jack Brown and Family Alzheimer’s Research Foundation. W. Song is the holder of the Tier 1 Canada Research Chair in Alzheimer’s Disease. P.T.T. Ly is supported by the Natural Sciences and Engineering Research Council (NSERC) Alexander Graham Bell Canada Graduate Scholarship and Michael Smith Foundation for Health Research (MSFHR) Senior Graduate Studentship. Y. Wu is the recipient of the Arthur and June Willms Fellowship. R. Wang and Y. Yang are supported by the Chinese Scholarship Council award.
Conflict of interest: The authors have declared that no conflict of interest exists.
Citation for this article: J Clin Invest. 2013;123(1):224–235. doi:10.1172/JCI64516.