AMPA-ergic regulation of amyloid-β levels in an Alzheimer’s disease mouse model

The mice used for these studies were hemizygous APPswe/PS1ΔE9 (APP/PS1) and bred on a wild-type C3H/B6 background, C57BL/6j-IL-6 ^tm1Kopf mice (hereafter referred to as IL-6^−/− mice), or littermate controls (WT) [23, 24].

Original APP/PS1 transgenic breeders as well as IL-6^−/− mice were purchased from Jackson Laboratory (Bar Harbor, Maine), and colonies were maintained at Washington University. Equal numbers of male and female mice were used in each study at 2–4 months of age. All studies were performed in accordance with the guidelines of AAALAC and the IACUC at Washington University.

In vivo microdialysis was performed in awake and behaving APP/PS1 mice as previously described [12, 25]. Briefly, guide cannulas (BR-style, Bioanalytical Systems, West Lafayette, IN) were stereotaxically implanted above the left hippocampus, coordinates bregma − 3.1 mm, 2.5 mm lateral to midline, and 1.2 mm below dura at a 12° angle. The cannulas were securely affixed to the head with dental cement, and microdialysis probes (BR-2, 2 mm, 38 kDa MWCO, Bioanalytical Systems) were inserted into the hippocampus through the guide cannula. In APP/PS1 mice, probes were perfused with artificial cerebrospinal fluid (aCSF; 1.3 mM CaCl₂, 1.2 mM MgSO₄, 3 mM KCl, 04 mM KH₂PO₄, 25 mM NaHCO₃, and 122 mM NaCl, pH 7.35) with 0.15% bovine serum albumin (BSA; Sigma-Aldrich, St. Louis, MO) at a rate of 1.0 μL/min with samples of hippocampal ISF collected every 90 min during basal collection or every 60 min during treatment. Because WT murine Aβ concentrations are lower than in amyloidogenic transgenic mice, microdialysis was run at 0.5 μL/min and samples collected every 3 h to increase concentration of each sample. Murine Aβ was also analyzed in the experiment using IL-6^−/− mice. For this experiment, microdialysis was run at 1.0 μL/min and samples were collected every 2.5 h. Basal sampling began at least 16 h following surgery. These experiments took place under constant light conditions to diminish circadian-related fluctuation in Aβ levels. At the conclusion of the experiment, all ISF samples were analyzed for either human or murine Aβ_x-40 or Aβ_x-42 levels by sandwich ELISA.

Reverse microdialysis was used to administer compounds directly into the hippocampus. Drugs were diluted into the perfusion buffer of artificial CSF and 0.15% BSA, allowing the drugs to diffuse into the brain continuously for the duration of the experiment at the same time that Aβ is collected. Due to the complexity of determining the final concentration of compound delivered to the brain, only the starting concentrations of drugs in the perfusion buffer are given. We estimate approximately 10% of the drug is delivered across the probe membrane where it is further diluted in the brain CSF. AMPA (0.5, 2, 5, 7.5, and 10 μM), MK801 (100 μM), NMDA (40 μM), and thiorphan (10 μM) were purchased from Sigma. Cyclothiazide (CTZ; 300 μM), tetrodotoxin (TTX; 5 μM), NBQX (100 μM), and GM6001 (25 μM) were purchased from Tocris Bioscience (Ellisville, MO). LY411575 (Sigma-Aldrich) was diluted in corn oil and administered subcutaneously at 5 mg/kg.

ISF samples were analyzed for Aβ_x-40 or Aβ_x-42 concentration using methods previously described (Fisher et al., 2016). A mouse monoclonal anti-Aβ₄₀ capture antibody (mHJ2) or anti-Aβ₄₂ capture antibody (mHJ7.4) made in-house was used in conjunction with a biotinylated central domain detection antibody (mHJ5.1) and streptavidin-poly-HRP-40 (Fitzgerald Industries, Acton, MA). Super Slow ELISA TMB (Sigma-Aldrich) was then used to develop, and absorbance was read by a BioTek Epoch plate reader at 650 nm. The same assay can be used for both human and murine Aβ_x-40. Standard curves for ELISAs were generated using synthetic human Aβ₄₀ or Aβ₄₂ (American Peptide, Sunnyvale, CA). Basal levels of ISF Aβ levels were calculated by averaging the Aβ concentrations taken every 90 min for 9 h prior to drug treatment. All Aβ levels for each mouse were then normalized by calculating percent of basal for each point. Mean ± SEM per group are shown.

Guide cannula implantation and microdialysis were performed as described above using 2–4 month old APP/PS1 mice. 5 μM AMPA or vehicle was administered to APP/PS1 mice via reverse microdialysis for 8 or 14 h. Immediately following treatment, perfusion buffer was changed to aCSF containing 0.1% Evans Blue dye for 30 min. During this period, the area of the hippocampus directly surrounding the microdialysis probe was dyed blue, approximating the area of tissue affected by reverse microdialysis drug delivery. Following the 30-min of Evans Blue administration, the mice were sacrificed and the dyed tissue surrounding the probe was microdissected and snap frozen on dry ice, generating approximately 5-7 mg of tissue per mouse. The collected hippocampal tissue was homogenized by sonication at a 10:1 volume:wet weight in 150 mM NaCl, 50 mM Tris, pH 7.4, 0.5% deoxycholic acid, 0.1% SDS, 1% Triton X-100, 2.5 mM EDTA, and protease inhibitors. Gel electrophoresis of 20 μg protein samples was performed under reducing conditions using 4–12% Bis-Tris NuPAGE gels (ThermoFisher Scientific, Waltham, MA) and then transferred to nitrocellulose membrane. Blots were probed for glial fibrillary acidic protein (GFAP; 1:500; ThermoFisher), low density lipoprotein receptor-related protein 1 (LRP1; 1:5000; Abcam, Cambridge, MA), insulin-degrading enzyme (IDE; 1μg/mL; Abcam), neprilysin (1:1000; Millipore, Billerica, MA), matrix metalloproteinase-9 (MMP-9; 1:1000; Millipore), C-terminal fragments of APP (1:1000; Sigma-Aldrich), total soluble APP (22C11; 1:5000; Millipore), soluble APP-α (poly18268; BioLegend, San Diego, CA), soluble APP-β (poly8134; 1:1000; BioLegend), β-amyloid 1–16 (6E10; 1:500; BioLegend), glutamate receptor 2 (GluR2; 1:1000; Millipore), tubulin (1:2500; Sigma), and glyceraldehyde 3-phosphate dehydrogenase (GAPDH; 1:10,000; Sigma). HRP-conjugated goat anti-rabbit IgG (1:1000; Cell Signaling Technology, Danvers, MA) and HRP-conjugated Amersham ECL sheep anti-mouse IgG (1:1000; GE Healthcare, Chicago, IL) were used as secondary antibodies. Membranes were developed using SuperSignal West Pico Substrate (ThermoFisher) or Lumigen-TMA6 (GE Healthcare) and imaged using the Kodak ImageStation 440CF (Rochester, NY). Band intensity was quantified using the Kodak 1D Image Analysis software, and normalized using tubulin or GAPDH signals as loading controls. Values shown are these normalized band intensities relative to the experimental control group. Mean ± SEM per group are shown.

Using the same tissue preparation as used for Western blotting (described above), APP/PS1 mice were treated with 5 μM AMPA for 8 or 14 h, followed by 30 min of 0.1% Evans Blue solution via reverse microdialysis. Dyed tissue around the probe was microdissected and frozen. Quantitative PCR was performed as described previously (Fisher et al., 2016). The RNeasy Mini Kit (Qiagen, Valencia, CA) was used to extract RNA, which was then reverse transcribed with a High Capacity cDNA Reverse Transcription kit (ThermoFisher). The Harvard Medical School Primer Bank was used to design primers [26‐28]. Real-time detection of PCR product was performed using the Fast SYBR Green Master Mix (Applied Biosystems, Foster City, CA) in ABI 7900HT (Applied Biosystems) with the default thermal cycling program. cFos was used as a positive control due to its established role as a mark of neuronal activity [29]. Gapdh was used as a reference gene for relative expression calculations. Relative mRNA levels were calculated using the comparative Ct method using the formula 2^-ΔΔCt. Mean ± SEM per group are shown.

2–4 month-old wild-type mice (n = 6 per group) or APP/PS1 mice (n = 3 per group) were treated with 8 h or 14 h, respectively, of AMPA or artificial CSF via reverse microdialysis then immediately transcardially perfused with ice-cold phosphate buffer saline (PBS) with 0.3% heparin. Brains were removed, fixed in 4% paraformaldehyde for 24 h at 4 °C, then placed in 30% sucrose prior to freezing and sectioning. Coronal brain sections 50 μm wide were sliced in 300 μm intervals using a freezing sliding microtome. Sections were then immunostained to visualize astrocytes or microglia using antibodies against glial fibrillary acidic protein (GFAP; 1:500, ThermoFisher) as an astrocytic marker or against ionized calcium-binding adaptor molecule 1 (Iba1; 1:500; Wako Laboratory Chemicals, Richmond, VA) as a microglial marker. Biotinylated secondary antibody, horseradish peroxidase-conjugated streptavidin, and DAB reaction (Sigma) were used to develop. Brain sections were imaged with a Nanozoomer slide scanner (Hamamatsu Photonics, Bridgewater, NJ). Staining density was qualitatively evaluated by blinded observers and vehicle- and AMPA-treated groups were compared. Images shown are representative.

Half-life of ISF Aβ was measured using methods described previously [25]. Microdialysis was performed as detailed above and basal ISF Aβ levels were collected. Reverse microdialysis was then used to treat APP/PS1 mice with either 5 μM AMPA or vehicle for 14 h, followed by co-administration with LY411575, a potent and selective γ-secretase inhibitor (Sigma-Aldrich; 5 mg/kg in corn oil, subcutaneous injection) to block Aβ production. ISF Aβ levels were measured using sandwich ELISA, and the half-life was calculated using the slope of the semi-log plot of percent change in Aβ levels versus time. The slope was calculated based only on Aβ values that were continually decreasing, excluding points at which levels plateaued. Mean ± SEM per group are shown.

Hippocampal tissue was collected from APP/PS1 mice treated with either vehicle (n = 7) or AMPA (n = 9) for 14 h via reverse microdialysis. Only tissue directly surrounding the probe was used. Tissue was homogenized following the manufacturer protocol in 500 mM NaCl, 50 mM Tris, pH 7.4, 0.5% deoxycholic acid, 0.1% SDS, 1% Triton X-100, 2 mM EDTA, and protease inhibitors (MesoScale Discovery, Rockville, MD, USA). Samples were assayed for interleukin(IL)-1β, IL-6, and tumor necrosis factor (TNF)-α using a custom MSD Proinflammatory Panel multiplex assay using the manufacturer’s protocol. Samples were assayed duplicate. Data analysis was performed using MSD Workbench software.

Littermate mice were randomly assigned into treatment groups, with equal numbers of male and females. Based on power analyses for detecting changes in ISF Aβ in microdialysis experiments, we used n = 4–8 mice per treatment group. A full description of statistical tests and the number of mice used can be found in the figure legends. Two-tailed unpaired t-tests were used to compare between two groups. One-way or two-way ANOVA was used when comparing one or two independent variables, respectively, between multiple groups. The appropriate correction for multiple comparisons was used (Sidak, Tukey, or Bonferroni; refer to figure legends). Analysis of microdialysis experiments was performed by averaging the final three data points of a specific treatment period and using one-way or two-way ANOVA with an appropriate correction for multiple comparisons. Values were accepted as significant is p ≤ 0.05. Data in figures are presented as mean ± SEM. Prism 6.0b for Mac OS X (GraphPad, San Diego, CA) was used for all statistical analyses.

Both synaptic activity and NMDA-Rs have distinct, established roles in regulating Aβ, but the involvement of AMPA-R signaling in Aβ regulation has been largely unexplored. To address this, we used in vivo microdialysis to measure the concentration of ISF Aβ in the hippocampus of mice [9, 25]. Crucially, this technique allows us to monitor changes in ISF Aβ levels over time in freely moving mice with functional glutamatergic synapses and intact neuronal networks. Through reverse microdialysis, we are also able to locally and continuously deliver small-molecule compounds, such as AMPA, into the hippocampus without needing to cross the blood-brain barrier.

Using microdialysis in the hippocampus of young, plaque-free (2–4 month old) APPswe/PS1Δe9 hemizygous (APP/PS1) mice [23, 24], we collected hourly samples of ISF while infusing AMPA in increasing concentrations from 0.5 μM to 10 μM for 8 h each (Fig. 1a). AMPA delivered at 0.5 μM or 2 μM had no effect on ISF Aβ. However, beginning with the 5 μM AMPA concentration, ISF Aβ levels gradually decreased over time before stabilizing at a 32% decrease from baseline levels. An even greater decrease is seen following 10 μM AMPA treatment, with levels of Aβ stabilizing at a 75% decrease from baseline levels (Fig. 1a). In the following experiments, we used 5 μM AMPA in order to observe further increases and decreases in ISF Aβ levels after they are already lowered by AMPA treatment. In this study, we focus primarily on ISF Aβ₄₀ because it is produced in much higher quantities than Aβ₄₂ in our mouse model and therefore simpler to detect using microdialysis. To determine if AMPA treatment acts on both species of Aβ similarly, ISF samples from 5 μM AMPA-treated mice were measured for Aβ_42. We found that AMPA decreases ISF Aβ₄₂ similarly to Aβ₄₀, indicating that it acts on both species of Aβ in the same manner (Fig. 1b). Next, wild-type (WT) mice were treated with 5 μM AMPA to eliminate potential confounds due to the transgenes in APP/PS1 mice. Murine ISF Aβ levels in WT animals reacted to 5 μM AMPA treatment similarly to APP/PS1 mice with a 45% decrease from baseline levels (Fig. 1c).

AMPA-Rs rapidly desensitize following AMPA or glutamate exposure [30]. One possible explanation for the observed effect on ISF Aβ, therefore, could be reduced activity due to decreased AMPA-R signaling. To test this possibility, we treated the APP/PS1 mice with cyclothiazide (CTZ), a thiazide diuretic, which inhibits desensitization and potentiates AMPA-mediated glutamate currents [31]. The mice were pre-treated with CTZ for 4 h before and then during treatment with increasing doses of AMPA (0.5 μM–5 μM) lasting 4 h each (Fig. 1d). Potentiated AMPA-R signaling enhanced the suppression in ISF Aβ levels with AMPA treatment starting at just 0.5 μM, a dose that has no effect on ISF Aβ without CTZ. This decrease is dose-dependent, with a maximal decrease in ISF Aβ of 83% from basal levels (Fig. 1d). These data indicate that the observed decrease in ISF Aβ is due to AMPA-R activity and not desensitization.

The exogenous application of AMPA through reverse microdialysis allows us to directly and selectively target AMPA-Rs. However, infusion of AMPA does not necessarily reproduce endogenous AMPA-R signaling. To address this, we treated mice with NBQX, a competitive AMPA-R antagonist (Fig. 2a). When baseline levels of AMPA-R signaling were blocked, ISF Aβ levels decreased by 32%, suggesting that AMPA-R activation increases Aβ during normal activity.

Next, we treated mice with tetrodotoxin (TTX) for 16 h to prevent the production of action potentials and therefore block evoked presynaptic release of glutamate (Fig. 2b). Following 16 h of TTX treatment, we co-infused TTX with NBQX. As previously reported, treatment with TTX alone decreases ISF Aβ levels by about 40% from basal levels [8]. Blocking AMPA-Rs in addition to TTX treatment leads to a further decrease in Aβ levels of 33% despite the cessation of presynaptic activity (Fig. 2b). Thus, AMPA-Rs activated during steady-state, tonic levels of activity appear to drive higher ISF Aβ levels independently of evoked glutamatergic signaling. Interestingly, antagonizing basally active AMPA-Rs induces a full effect on ISF Aβ levels regardless if action potentials are intact or blocked with TTX, suggesting that basal AMPA-ergic regulation of Aβ is driven by spontaneous glutamate release via miniature EPSCs (“minis”) as opposed to evoked activity.

We next determined the extent to which AMPA-mediated Aβ regulation relies on presynaptic activity. As before, mice were pre-treated with TTX followed by co-treatment with TTX and AMPA. During the initial 8 h of TTX and AMPA treatment, the decrease in Aβ levels caused by AMPA treatment (Fig. 2c) is abolished. However, a longer AMPA treatment of 14 h significantly decreased ISF Aβ levels by 30% of post-TTX levels (Fig. 2b). These results imply that, initially, evoked glutamatergic transmission is necessary for AMPA treatment to decrease ISF Aβ. With longer treatment, however, ISF Aβ levels are reduced through postsynaptic AMPA-R signaling alone, without the need of action potentials or further glutamatergic activity stimulation.

Given that high levels of NMDA-R activation result in decreased Aβ levels through calcium–dependent ERK signaling [32, 33], we hypothesized that AMPA treatment might reduce ISF Aβ levels through the indirect activation of NMDA-Rs expressed on downstream postsynaptic neurons. To determine the contribution of NMDA-Rs to the changes in Aβ levels following AMPA treatment, mice were pre-treated with MK801, a NMDA-R open channel blocker, via reverse microdialysis for 6 h before co-treatment with MK801 and 5 μM AMPA (Fig. 2d). Within the first 8 h of treatment, co-application of MK801 and AMPA does not effect an AMPA-related change in Aβ levels. The ability of AMPA to alter ISF Aβ is therefore dependent on NMDA-R activation at this time point. By hour 14 of AMPA treatment, however, Aβ levels began to decline regardless of the presence of MK801 (Fig. 2d). These data imply that AMPA’s effects on ISF Aβ levels are dependent on NMDA-R signaling for only a limited period. After prolonged treatment with AMPA, Aβ levels decrease through an NMDA-R-independent mechanism.

In consideration of these results, we questioned if AMPA-R signaling might be responsible for any part of NMDA-Rs’ effect on Aβ levels. To test this, we first treated the mice with 100 μM NBQX, a competitive AMPA-R antagonist, through reverse microdialysis then co-treated with NMDA (Fig. 2a). As observed in previous experiments [32], 40 μM NMDA reduced ISF Aβ levels to approximately 50% of basal levels within 6 h of treatment, even in the presence of an AMPA-R antagonist (Fig. 2a). Though the effect of AMPA treatment on ISF Aβ in part relies on NMDA-R involvement, the opposite does not appear true; NMDA treatment decreases Aβ levels independently from AMPA-R activation. To ensure the specificity of AMPA treatment, animals were treated with NBQX to block AMPA-Rs prior to the addition of AMPA. As was expected, NBQX completely blocked the effect of AMPA-Rs on Aβ (Fig. 2a).

Previous data show that activation of NMDA-R signaling rapidly decreases ISF Aβ levels by approximately 50% [32]. Once NMDA is no longer administered, ISF Aβ gradually returns to baseline levels within 30 h. AMPA treatment, however, results in a longer-lasting change in Aβ levels. APP/PS1 mice were perfused with 5 μM AMPA for 8 h. After this period, AMPA treatment ended and Aβ levels were monitored every 1–2 h for an additional 44 h (Fig. 3a). Levels of ISF Aβ decreased steadily during the AMPA treatment and continued to decrease for 3 h into the washout period to reach a maximal decrease of 60% from basal levels. From this lowest point, Aβ levels significantly increased from the trough to reach a level only 35% decreased from basal levels after 44 h of recovery (Fig. 3a). The washout study was terminated after a total of 60 h of ISF collection due to limitations in the reliable duration of microdialysis experiments, so it is possible that Aβ levels may completely recover from AMPA treatment with a longer washout period. A recovery in ISF Aβ suggests that AMPA treatment does not cause major cell death and that the area surrounding the microdialysis probe continues to function normally following treatment.

APP/PS1 mice were treated with AMPA for 8 h followed by co-administration with NBQX (Fig. 3b). The decrease in Aβ levels following AMPA application did not recover to baseline levels with the addition of NBQX despite the cessation of AMPA-R activation. Because the Aβ decrease was preserved without AMPA-ergic transmission, the effect on Aβ is likely due to a long-lasting intracellular event and not a feed-forward increase in continued glutamatergic transmission. This observed long-lived change in Aβ levels was initiated by an AMPA treatment period of only 30 min, which resulted in a 30% decrease in ISF Aβ (Fig. 3c).

We demonstrated above that extended treatment with AMPA influences ISF Aβ levels without the need for NMDA-R activation. NMDA-Rs receptors are often associated with intracellular signaling and transcriptional regulation, while AMPA-Rs are generally thought of in terms of neuronal depolarization. However, there is growing evidence to suggest that AMPA-Rs may also play an active role in cellular signaling. For example, Plant et al. (2006) found that transient calcium signaling through calcium-permeable AMPA-Rs promotes the maintenance of long-term potentiation (LTP) [34]. Additionally, AMPA-R signaling, independent of depolarization, is sufficient to activate the transcription factor CREB as well as to initiate ERK signaling [17, 18, 35]. Given these results, the AMPA-R-dependent decrease in ISF Aβ that we observe could be due to the initiation of a signaling cascade by AMPA-Rs. First, we tested if AMPA-Rs affect the transcription of genes related to APP processing or Aβ clearance (Fig. 4a, b). APP/PS1 mice were administered 5 μM AMPA for 8 or 14 h by reverse microdialysis. At the end of treatment, probes were infused with Evans Blue for 30 min to mark the surrounding tissue reached by reverse microdialysis. The dyed hippocampal tissue was lysed and used for quantitative real-time PCR (qPCR) for a selection of genes involved in Aβ metabolism. Expression of the immediate early gene, cFos, was used as a control due to its increased expression following glutamatergic transmission [29]. As expected, AMPA treatment increased the expression of cFos in both the 8- and 14-h groups. However, we found no significant changes in the expression of APP, in genes related to α-secretase (ADAM10 and ADAM17), in genes related to β-secretase (BACE1), nor in genes related to ϒ-secretase (PS1, PS2, PSEN2, APH1, BSG, and NIC) following 8 or 14 h of AMPA treatment (Fig. 4a, b). Further, AMPA treatment did not change expression in ERK1 or ERK2 or in genes associated with Aβ clearance (LRP1, LRPR, AQP4, NEP, MMP2, and MMP9). Finally, none of the AMPA-R subunits genes (GRIA1–4) were altered by AMPA treatment (Fig. 4a, b).

To the best of our knowledge, all previous studies investigating the relationship between synaptic signaling and alterations in Aβ levels, including several from our laboratory, have found that synaptic signaling primarily affects Aβ production [4, 6‐8, 25, 36]. However, after 14 h of AMPA administration, we found no change in full-length APP levels or in the cleavage products β-C-terminal fragment (β-CTF), soluble APP-α (sAPP-α), and sAPP-β as determined by Western blot (Fig. 5a). In combination with the lack of transcriptional changes in production-related genes (Fig. 4a, b), these data suggest that extended treatment with AMPA does not have a pronounced effect on Aβ production. It is important to note, however, that small changes in gene or protein levels, such that occur when only a subpopulation of cells is affected, can be masked when total brain lysates are analyzed. Considering the large effect that AMPA has on ISF Aβ levels, though, we hypothesized that AMPA-Rs act on ISF Aβ through a different mechanism, namely by altering its clearance.

Aβ is eliminated from the ISF through five main pathways: receptor-mediated transport across the blood brain barrier (BBB), enzymatic degradation, cellular uptake, glymphatic-mediated clearance, or passive bulk-flow clearance for (reviews see [37‐39]). If any of these pathways is targeted by AMPA treatment, the rate of ISF Aβ clearance could increase. To test this possibility, we measured half-life of ISF Aβ in mice treated with either 5 μM AMPA or vehicle using reverse microdialysis (Fig. 5b). AMPA treatment leads to a rapid decrease in Aβ that stabilizes by 6–8 h of treatment. After 14 h, mice were subcutaneously injected with LY411575, a potent γ-secretase inhibitor that rapidly inhibits Aβ production. LY411575 enters the brain and within 15 min reaches a concentration approximately 200-fold in excess of its IC₅₀ for γ-secretase inhibition [25]. Once γ-secretase is inhibited, all new production of Aβ is precluded and microdialysis is used to monitor the levels of remaining ISF Aβ over time. The rate at which Aβ in the ISF is eliminated can be measured by calculating the slope of the semi-log plot of percentage baseline Aβ levels versus time. This elimination rate was determined for both groups, and the Aβ half-life calculated. Interestingly, the half-life of ISF Aβ was significantly shorter by over 30% in mice receiving AMPA treatment (t_1/2 = 0.93 h) than those in the control group (t_1/2 = 1.38 h), indicating that AMPA treatment increases the clearance of ISF Aβ (Fig. 5c). It is important to note that 6 of 12 AMPA-treated mice had ISF Aβ levels decrease so much that a reliable half-life could not be calculated. If this greater decrease following AMPA treatment was also due to enhanced clearance, then the observed effect of AMPA on Aβ would be even greater so, we could be underestimating the effect of AMPA on Aβ clearance. Next, we measured the levels of key proteins involved in Aβ clearance in the hippocampal tissue surrounding the microdialysis probe for mice treated with 14 h of AMPA or vehicle (Fig. 6a). Similar to the qPCR experiments (Fig. 4b), only the positive control cFos showed a significant change in protein levels with AMPA treatment (Fig. 6a). Though these data suggest that none of the Aβ clearance-related proteins selected is involved in AMPA-mediated regulation of Aβ, Western blots do not detect cell type-specific changes in protein levels, alterations in protein function, or changes in protein localization. To test if AMPA treatment increases protease activity and thus Aβ degradation, we pre-treated APP/PS1 mice with the neprilysin inhibitor, thiorphan, or with the broad-spectrum metalloproteinase (MMP) inhibitor, GM6001, before co-treating with AMPA. Inhibition of neprilysin and all MMP family members both blocks Aβ clearance pathways and potentially inhibits α-secretase, which increases ISF Aβ levels when those agents are administered singly (Fig. 6b). Importantly, the addition of AMPA still decreased Aβ by a comparable amount as observed without the protease inhibitors, indicating that AMPA does not affect degradation of Aβ through these proteases.

A potential concern is that AMPA treatment decreases ISF Aβ by causing cellular toxicity and/or creating a lesion through increased glutamatergic activity [40]. If AMPA does cause cellular damage, an inflammatory response would involve the recruitment and activation of microglia and astrocytes [41‐43]. To monitor inflammatory responses, mice were treated with 5 μM AMPA or with vehicle for 8 or 14 h before brains were collected and fixed in 4% formaldehyde. The brains were stained for Iba1, a marker of microglia [44], and GFAP, a marker for astrocytes [42, 45]. As expected, we found increased Iba1 and GFAP staining around the microdialysis probe tract, but no change in staining density between the AMPA- and vehicle-treated tissue at either time point (Fig. 7a). For confirmation, we measured protein levels of GFAP and CD45, another microglial marker [46], using hippocampal lysates from APP/PS1 mice treated with either 5 μM AMPA or vehicle for 14 h (Fig. 7b). In agreement with the immunostaining results, AMPA treatment did not increase GFAP or CD45 protein levels, indicating a lack of glial recruitment (Fig. 7b). In addition to monitoring the glial response, we measured pro-inflammatory cytokines levels in the hippocampal lysates of mice following AMPA treatment. Though IL-1β and TNF-α levels were unchanged, the levels of IL-6 showed a dramatic increase of over 500% (Fig. 7c). IL-6 is a neuropoietic cytokine with both neuromodulatory and neuroprotective roles, known to be induced by neuronal activity [47‐49]. Without a visible increase in gliosis and with no significant increase in IL-1β or TNF-α, there does not appear to be a broad inflammatory response. These data, along with the partial recovery of ISF Aβ in the 44 h sampled following AMPA treatment (Fig. 3a), strongly suggest AMPA is not causing widespread toxicity accounting for the effects on Aβ observed in this study.

IL-6 is a neuropoietic cytokine with both neuromodulatory and neuroprotective roles, known to be induced by neuronal activity [47‐49]. Intriguingly, IL-6 has previously been linked to enhanced Aβ clearance [50, 51]. Because levels of IL-6 increased greatly following AMPA treatment, we tested the possibility that enhanced IL-6 signaling is involved in the decrease in ISF Aβ levels following AMPA-R stimulation. To do this, we utilized 3-month-old IL-6-deficient mice (IL-6^−/− mice). These mice develop normally and produce normal levels of murine Aβ in the ISF. We treated IL-6^−/− and WT mice with MK801 to block NMDA-R signaling for 6 h, then added 7.5 μM AMPA into the perfusion buffer for an extended period (Fig. 8). Because murine Aβ levels are much lower in these mice than in our amyloidosis models, samples were collected every 2.5 h. Similar to our observations in APP/PS1 mice, AMPA treatment led to a decrease in ISF Aβ by approximately 67% in WT mice. Conversely, AMPA failed to produce a significant change in ISF Aβ levels in IL-6^−/− mice, suggesting that IL-6 signaling is necessary for AMPA-R regulation of Aβ.

We found that infusion of AMPA directly into the hippocampus of APP/PS1 mice through reverse microdialysis decreases ISF Aβ levels by up to 75% following the maximal dose of 10 μM. Treatment with AMPA induces a potent, long-lasting effect on Aβ levels, with even a brief application initiating a full response. AMPA-Rs, therefore, appear to be significant regulators of Aβ levels in the extracellular space. Factors that influence extracellular levels of Aβ have the potential to directly influence AD pathogenesis by altering the likelihood of Aβ to aggregate [52]. That AMPA increases activity but suppresses Aβ levels is somewhat surprising considering previous reports that synaptic activity drives production of Aβ. Treatment with the GABA_A receptor antagonist picrotoxin, high levels of potassium chloride, or electrical stimulation promotes Aβ secretion into the extracellular space [7‐9]. In a more physiological setting, increasing activity within the barrel cortex through vibrissal stimulation results in higher levels of ISF Aβ in APP/PS1 mice [4, 53]. In humans, the highest levels of amyloid deposition are found in brain regions with the highest baseline metabolic activity [10].

Considering these findings, it would be reasonable to hypothesize that AMPA-Rs, as excitatory channels, should increase Aβ levels. Paradoxically, however, we found increasing AMPA-R activation through exogenous AMPA treatment significantly decreases ISF Aβ. Because AMPA-Rs are susceptible to rapid desensitization, we considered the possibility that AMPA-Rs act on Aβ levels through induced synaptic depression [30, 31]. However, when receptor desensitization was blocked with cyclothiazide, the decrease in Aβ in response to AMPA was potentiated. Receptor desensitization only limited Aβ suppression, and receptor activation is directly responsible for the reduction of Aβ levels.

Though general increases in synaptic activity upregulate Aβ production, the activation of certain postsynaptic signaling systems can alter APP processing to yield varied effects on Aβ levels, particularly when α-secretase is targeted. As mentioned above, serotonin receptor activation decreases Aβ levels through PKA and ERK activation [12, 13]. The serotonin receptor illustrates the specificity involved in Aβ regulation; only the G_s-linked receptors decrease Aβ whereas the other G-protein coupled serotonin receptors have no effect or increase Aβ [13]. Additionally, M1 muscarinic acetylcholine (mACh) receptor agonists decrease Aβ production, and deleting this receptor leads to increased Aβ and amyloid pathology [54‐56]. Within the glutamate receptor family, muscarinic glutamate receptor 5 has been shown to trigger Aβ production [57, 58], and NMDA-Rs can modulate Aβ levels bidirectionally [32, 33, 59]. Clearly, postsynaptic effects on Aβ are diverse and markedly context-specific.

In these studies we have shown that AMPA-R regulation of Aβ is multifarious (see model Fig. 9). When basal AMPA-R activity is antagonized, ISF Aβ decreases by 20%. The same decrease occurs even after action potentials are blocked and evoked synaptic transmission is inhibited, indicating that the basal AMPA-R signaling that increases Aβ levels is likely due to spontaneous transmission (Fig. 9a). Conversely, application of AMPA via reverse microdialysis causes direct AMPA-R activation as well as stimulates evoked glutamatergic transmission. In this scenario, AMPA-R activation decreases Aβ levels. This dual effect of AMPA-Rs, depending on the mode of transmission, has been seen in various contexts. Sara and colleagues (2011) utilized a use-dependent AMPA-R antagonist to show that spontaneous and evoked transmission activate discrete populations of AMPA-Rs [60]. Additionally, several studies found that receptors that respond differentially to spontaneous and evoked transmission are physically and functionally distinct [61‐66]. Intriguingly, spontaneous activity appears to suppress protein synthesis while evoked activity stimulates translation. Another possible explanation is that the effects of AMPA-Rs on Aβ are dependent on relative levels of AMPA-R activation. During basal transmission, a smaller set of AMPA-Rs is active compared to the AMPA-Rs targeted by action potentials or exogenous AMPA treatment. How endogenous AMPA-Rs promote increased levels of ISF Aβ remains unknown, though we speculate that basal AMPA-ergic signaling induces amyloidogenic APP processing through increased endocytosis within or near the presynaptic terminal, as described in previous studies [7‐9, 67].

Adding an additional layer of complexity, exogenous AMPA treatment appears to act on Aβ levels through two distinct pathways. Within the first 8 h of treatment, AMPA’s ability to modulate Aβ levels depends on NMDA-R signaling (Fig. 9b). This pathway relies on presynaptic activity to increase glutamatergic transmission, thus stimulating NMDA-R activation on downstream neurons to decrease Aβ production in these cells [32, 33]. The reverse is not true, however; AMPA-Rs do not appear to play a role in NMDA-R-mediated decreases in Aβ. Following longer AMPA treatment, a novel pathway by which AMPA-Rs influence Aβ independently of both presynaptic activity and NMDA-Rs emerges.

As detailed above, studies regarding synaptic and postsynaptic regulation of Aβ have primarily addressed the effects of activity on Aβ production. However, we were unable to detect changes in APP processing-related gene expression or in APP fragment levels in hippocampal lysates following either 8 or 14 h of AMPA treatment. Instead, using microdialysis along with a potent γ-secretase inhibitor, we found that treatment with AMPA for 14 h decreased the half-life of ISF Aβ, implying that AMPA-Rs modulate Aβ clearance (Fig. 9c). This does not appear to involve glial recruitment, a broad inflammatory response, or changes in key clearance-related proteins. We found that one proinflammatory cytokine, IL-6, increased dramatically following AMPA treatment. IL-6 has been shown to have both normal physiological as well as inflammatory, pathological roles in the CNS [49, 51, 68, 69] and has been shown to be secreted in response to neuronal depolarization [47, 48]. Furthermore, IL-6 signaling has been linked to increased Aβ clearance through microglial phagocytosis [50, 51]. Given the substantial increase in IL-6 following AMPA treatment, we propose that AMPA could be causing IL-6 release and enhanced phagocytosis of Aβ. In support of this hypothesis, mice deficient in IL-6 fail to show decreased ISF Aβ levels in response to AMPA treatment, suggesting that IL-6 signaling is involved in AMPA-R regulation of ISF Aβ levels. The IL-6 receptor is expressed on neurons, microglia, and astrocytes [70, 71], so this synaptic activity-dependent clearance pathway could be mediated by multiple cell types. Though we propose a connection between neuronal IL-6 release and microglial clearance, our data do not indicate which cell types are involved. Furthermore, we have only tested a handful of cytokines in response to AMPA treatment thus far, leaving open the possibility that multiple cytokines are involved in this pathway. Future experiments will address the mechanism through which AMPA-Rs affect Aβ clearance.

The finding that AMPA treatment decreases Aβ levels is supported by a previous study by Hoey and colleagues conducted in primary cortical neurons [22]. Unlike our study, however, the authors conclude that Aβ production is decreased when AMPA directly acts to increase non-amyloidogenic APP processing. In contrast, our in vivo studies suggest that AMPA treatment requires an intermediary step of NMDA-R activation in order to increase non-amyloidogenic processing of APP. Additionally, our studies model a second pathway in which AMPA directly acts on Aβ through enhanced clearance. Because this pathway likely involves multiple cell types interacting, experiments using neuronal cultures would not recapitulate the effects we observed. Furthermore, the discrepancies in findings could also be explained by developmental differences between our two systems. Hoey et al. (2013) found that AMPA-mediated alterations in APP processing are at least partially due to calcium-permeable AMPA-Rs. There is evidence that GluA2, the receptor subunit responsible for determining the receptor’s calcium permeability, is developmentally regulated [72‐75]. Finally, we have found that even slight changes in AMPA concentration can change Aβ’s response, and our two studies used very different doses. We administered 5 μM AMPA through the microdialysis probe of which only an estimated 10% diffuses into the extracellular space. In contrast, Hoey et al. (2013) administered 50 μM AMPA, potentially activating a different pathway than we observed.

Though both production and clearance determine the steady state levels of Aβ in the extracellular space, late-onset AD (LOAD) is primarily characterized by dysfunctions in Aβ clearance [38, 76]. In 2003, we found that ISF Aβ half-life as measured by microdialysis is doubled in an aged APP transgenic model compared to young animals [25]. In human studies, metabolic labeling and CNS analysis revealed impaired clearance rates in participants with LOAD, though Aβ production was unaltered [76]. Furthermore, many of the genetic factors associated with LOAD are related to clearance, including APOE, CLU, CR1, and CD33. Given the evident prominence of Aβ clearance in AD, our results highlight the importance of understanding the ways in which synaptic activity impinges on previous clearance-related studies.