Oligomeric β-Amyloid Suppresses Hippocampal γ-Oscillations through Activation of the mTOR/S6K1 Pathway

Neuronal synchronization at gamma frequency (30-100 Hz: γ) is impaired in early-stage Alzheimer's disease (AD) patients and AD models. Oligomeric Aβ1-42 caused a concentration-dependent reduction of γ-oscillation strength and regularity while increasing its frequency. The mTOR1 inhibitor rapamycin prevented the Aβ1-42-induced suppression of γ-oscillations, whereas the mTOR activator leucine mimicked the Aβ1-42-induced suppression. Activation of the downstream kinase S6K1, but not inhibition of eIF4E, was required for the Aβ1-42-induced suppression. The involvement of the mTOR/S6K1 signaling in the Aβ1-42-induced suppression was confirmed in Aβ-overexpressing APP/PS1 mice, where inhibiting mTOR or S6K1 restored degraded γ-oscillations. To assess the network changes that may underlie the mTOR/S6K1 mediated γ-oscillation impairment in AD, we tested the effect of Aβ1-42 on IPSCs and EPSCs recorded in pyramidal neurons. Aβ1-42 reduced EPSC amplitude and frequency and IPSC frequency, which could be prevented by inhibiting mTOR or S6K1. These experiments indicate that in early AD, oligomer Aβ1-42 impairs γ-oscillations by reducing inhibitory interneuron activity by activating the mTOR/S6K1 signaling pathway, which may contribute to early cognitive decline and provides new therapeutic targets.

of aberrant Aβ has been attributed to suppressed autophagy [3]. However, in mild cognitive impairment (MCI) and early stages of AD, there is a pathological heterogeneity, and a significant proportion of nondemented aged subjects show typical AD pathologies [4,5]. Therefore, the quest for the immediate cause of early cognitive decline in AD is imperative.
Synchronization of neuronal activity at frequencies in the gamma band (30-100 Hz: γ) provides a millisecondprecision timing matrix that facilitates inter-neuronal communication and determines the dynamic coupling of activity between brain areas [6,7]. Hippocampal γoscillations of the local field potential measure the synchronization of neuronal activity, which correlates with indexes of working memory [8][9][10] and spatial memory [11]. γ-oscillations in the hippocampus have been associated with navigation [12] and hippocampusbased memory encoding/retrieval [13,14].
Decreased γ-oscillations have been reported in AD animal models' prefrontal, lateral entorhinal, and olfactory cortex areas [9,15,16] and in AD patients [17]. A reduction in the global synchronization index in the γ band, a measure used to quantify synchronization between brain areas, correlates with cognitive decline in AD [18] and is already observed in MCI [19]. In transgenic AD models that produce high levels of Aβ, γ-oscillations are reduced at very early ages [16,[20][21][22][23]. Oligomeric Aβ causes a significant reduction in spontaneous network activity in the hippocampal area CA1 [24]. Fibrillar Aβ acutely degrades mouse hippocampal γ-oscillations in a concentration-and time-dependent manner [25,26]. Interestingly, AD has been associated with neuronal hyper-excitability and epileptiform activity, primarily during reduced γ-oscillatory activity, which has been implicated in the progression of AD [27].
These studies suggest that impairments of synchronization at γ frequencies contribute to cognitive deficits in the early stages of AD [28,29], and treatments that restore γ-oscillations could be a promising therapeutic target [23], but the molecular mechanisms underlying the Aβ-induced change of hippocampal γ-generating circuits remain elusive.
Gamma oscillations emerge from rhythmic inhibitory postsynaptic currents (IPSCs) that synchronize neuronal firing [30]. Because γ-oscillations are very energydemanding [31], Aβ may reduce hippocampal γoscillations as the result of impairment of key metabolic processes involved in energy homeostasis. One option we pursued here is that Aβ affects energy homeostasis through changes in the mechanistic target of rapamycin (mTOR), a conserved Ser/Thr kinase that forms two multi-protein complexes known as mTOR complex 1 (mTOR1) and 2 (mTOR2) [32,33], which weakens synaptic transmission [34] and mitochondrial function [35]. Evidence from AD in both patients and mouse models shows that hyper-activation of the mTOR pathway occurs at the early stages of AD [36,37], and an increase in two mTOR downstream targets, p70 S6 kinase polypeptide 1 (S6K1) and eukaryotic translation initiation factor 4E (eIF4E)-binding proteins (4E-BPs), occurs already in MCI [38]. Reducing S6K1 expression improves spatial memory and synaptic plasticity in a mouse model of AD [39].
In the present study, we demonstrate that oligomeric Aβ causes impairment of hippocampal γ-oscillations through activating mTOR/S6K1 signaling.

Animals
All animal experiments followed the "Principles of laboratory animal care" (NIH publication No. 86-23, revised 1985), as well as guidelines and regulations of the Ethics Committee of Xinxiang Medical College. C57BL/6J mice of either sex (3-4-week-old, unless specifically mentioned) were obtained from Beijing HFK Bioscience Co. The APP/PS1, double transgenic mice, were obtained from cross-breeding single transgenic mice expressing human APPK670N/M671L with single transgenic mice expressing human PS1M146L [40]. The mouse colonies were kept on a 12-hour light/dark cycle in temperature-and humidity-controlled rooms. Food and water were available ad libitum.
DNA samples were isolated from the tail tip of APP/PS1 transgenic mice at 3 weeks old for genotyping, using PCR with human APP primers and human PS1 oligo primers. For APP, the forward primer was 5'-GAC TGA CCA CTC GAC CAG GTT CTG -3'and the reverse primer was 5'-CTT GTA AGT TGG ATT CTC ATA TCC G -3'. For PS1, the forward primer was 5'-GAC AAC CAC CTG AGC AAT AC-3', and the reverse primer was 5'-CAT CTT GCT CCA CCA CCT GCC-3'. APP/PS1 double transgenic mice displayed two target bands, while wild-type mice displayed no bands. APP/PS1 mice and wild-type mice were kept under standard conditions for up to 4-6 months in age.

Extracellular field potential recordings
Extracellular field potentials were recorded from the stratum pyramidal of area CA3, with aCSF-filled glass pipette recording electrodes (3-5 MΩ) [42]. Field potentials were amplified with Neurolog NL106 ACcoupled amplifiers (Digitimer, Welwyn Garden City, UK) and band-pass filtered at 2-200 Hz with Neurolog NL125 filters (Digitimer). After the mains line noise was removed with Humbug noise eliminators (Digitimer), the signal was digitized and sampled at 2 kHz using a CED-1401 Plus (Cambridge Electronic Design, Cambridge, UK) and Spike-2 software (Cambridge Electronic Design).
Ten minutes after placing the slices in the recording chamber, kainate (100 nM) was added to the aCSF to induce γ-oscillations. Gamma oscillations were recorded from area CA3c, where oscillation amplitude is normally the largest and least affected by the faster intrinsic γoscillations in CA1 [42].
Oscillation power was calculated from the power spectrum, generated by fast Fourier transforms over 60-s epochs (1Hz bin size, Hanning window, FFT size 2048). The summated power in the γ frequency range (set at 20-60 Hz for 32 °C): γ power, was used for quantification of the γ-oscillation strength. The peak frequency was determined as the local maximum of the power spectrum in the γ frequency range. Waveform auto-correlograms were calculated over 60-s band-pass (10-200 Hz) filtered epochs.
To ensure γ power was stable before drugs and/or Aβ peptide were added, the kainate-induced oscillatory activity was left to develop for at least 60 minutes. The average over the 5 minutes before the application, was taken as a baseline control. To quantify the effect of the drug or Aβ peptide on γ-oscillations, the average measure during the last 5 minutes of the application was normalized to the baseline control. In a separate set of experiments, the slices were pretreated with drugs for 10 minutes after 60 minutes in kainate, followed by 60 minutes of Aβ peptide addition. The average measure during the last 5 minutes of the Aβ peptide addition was normalized to the baseline control, to quantify the effect of the drug plus Aβ peptide on γ-oscillations. The power of γ-oscillations varies between slices and continues to grow gradually in kainate. To control for the effect of time, the changes induced by drugs or Aβ peptides were compared to the changes with time in a control group, where slices were treated with kainate only for 120 minutes, and the average measure over the last 5 minutes was normalized to the baseline measure.

Patch-clamp recordings
Whole-cell patch-clamp recordings were conducted in a submerged recording chamber perfused with aCSF (3-4 ml/minute at 30 °C). Pyramidal CA3 neurons were visually identified with a 40x water-immersion objective in an upright microscope (FN1, Nikon, Tokyo, Japan) and recorded using the whole-cell patch voltage-clamp technique, using glass pipettes (3-5 MΩ) as previously described [43]. Membrane currents were recorded with a Multiclamp 700B patch-clamp amplifier (Molecular Devices, Sunnyvale, USA), filtered at 2 kHz with a lowpass Bessel filter, and then digitized and sampled at 10 kHz with a Digidata 1440 (Molecular Devices).
All neurons included in this study had a resting membrane potential below -55 mV. The series resistance was tested before and after completion of each recording by measuring the current transient elicited by a 10-mV hyperpolarizing voltage step, and recordings were not analyzed for neurons with access resistance > 25 MΩ or if series resistance deviated >20% from the initial value.
Aβ peptides or drugs were added to the incubation solution containing 100 nM kainate 30 minutes before the start of patch-clamp recordings. While incubating, slices were continuously supplied with aCSF saturated with carbogen.
The biophysical and biological properties of Aβ 1-42 were characterized by transmission electron microscopy (TEM) analysis [47][48][49]. Samples (5 μl) were diluted and deposited onto carbon-coated copper mesh grids for 5 minutes, and the liquid was absorbed with paper and negatively stained with 2% (w/v) uranyl acetate. Then, the sample grids were allowed to air dry. The samples were viewed with a JEOL JEM-1400 microscope (JEOL Ltd, Japan), and digital images were acquired with an Advanced Microscopy Techniques camera. Aβ 1-42 oligomers appeared as fibril-free small globular structures that were <10 nm in diameter, and fibrillar Aβ 1-42 showed long threads measuring >1 µm in length with some aggregated Aβs (Fig. 1). The morphological characteristics of these fibrils were identical to those of Aβ fibrillar structures, as has been reported [49,50].

Western blotting
Western blotting was performed by the methods established in our laboratory [43]. Briefly, for the preparation of total cell extracts, the dissected hippocampal area CA3 tissue was homogenized in lysis RIPA buffer containing 1% sodium dodecyl sulfate buffer in Tris-EDTA (pH7.4), 1× protease inhibitor cocktail (P8340, Sigma-Aldrich), 5mM NaF, and 1× phosphatase inhibitor cocktail (P2580, Sigma-Aldrich). The homogenate was centrifuged at 5,000 g for 10 minutes, the supernatant was collected, and total protein concentrations were measured using Bradford Assays. Proteins were electrophoretically separated in 12% SDS-PAGE gels and were transferred to a polyvinylidene fluoride membrane. The proteins were analyzed by Western blotting using antibodies against mTOR (1:500; Cell Signaling, 2972), phospho-mTOR (Ser 2448) (1:500; Cell Signaling, 2971), β-actin (1: 1,000; Abcam, ab6272). Membranes were then incubated with a horseradish peroxidenzyme-conjugatedated secondary antibody (1:2,000; Abcam, ab288151) to reveal the location of the protein bands. Immunoreactive bands were visualized by chemiluminescence using enhanced ECL reagents (BeyoECL Plus, Beyotime). Subsequently, X-ray films were exposed to the membranes and then quantitatively analyzed by Image J software, which measures the intensity of protein bands after background subtraction. Relative protein level was calculated by normalizing phosphorylated form levels and total protein levels to the β-actin levels, respectively.

Statistics
Data are expressed as mean ± standard error of the mean or medians ± min-max for non-normally distributed data. The normality of the distribution of data was assessed by the Shapiro-Wilk test. N indicates the number of hippocampal slices tested or cells recorded. Statistical comparisons between experimental conditions were made using the unpaired Student's t-test or one-way ANOVA with Tukey's post hoc tests for normally distributed data or the Mann-Whitney U test for non-normally distributed data. Significance was assumed when P < 0.05.
To study whether the Aβ 1-42 -induced reduction of γoscillations is specific to kainate-induced γ-oscillations, we tested the effect of Aβ 1-42 on a different in vitro model of γ-oscillations. Using the same approach as for kainateinduced γ-oscillations, Aβ 1-42 reduced γ power of activity induced by carbachol (10 µM) to 47% of control power (the control power was 105% of baseline, t 8 = 6.80, P< 0.001, data not shown). Aβ 1-42 increased the peak frequency by 7% (t 8 = 3.80, P= 0.005, data not shown). These data suggest that oligomeric Aβ 1-42 impairs γoscillations in hippocampal CA3 irrespective of the in vitro model.

Activation of mTOR/S6K1 signaling mediates Aβ 1-42induced impairment of γ-oscillations
Involvement of the mechanistic target of rapamycin (mTOR) activation in AD has been shown previously [36,54]. To explore the potential role of mTOR in modulating γ-oscillations, we tested the effect of inhibition and activation of mTOR on γ-oscillations.
Slices were pretreated with mTORC1 inhibitor rapamycin (100 nM for 10 min) and then perfused with Aβ 1-42 and rapamycin for 60 minutes. Compared with the reduction of γ power induced by Aβ 1-42 only (by 47% of the control, n= 10 slices from 10 mice Aβ 1-42 only), pretreatment with rapamycin prevented the effect of Aβ 1-42 on γ power (increased by 12% compared with baseline control, F 3,42 = 37.01, P= 0.369, one-way ANOVA with Tukey's post hoc test, n= 10 slices from 10 mice, Fig. 3A-C). Rapamycin also prevented the increase in peak frequency of γ-oscillations (increased by 4% compared with the baseline control, F 3,42 = 13.56, P= 0.217, Fig.  3D). To control the effects of rapamycin itself, we tested the effect of rapamycin only for 60 minutes. Rapamycin-only treatment had no significant impact on the γ power (increased by 4% compared with the control, F 3,42 = 37.01, P= 0.158, one-way ANOVA with Tukey's post hoc test, Fig. 3A-C) or on the peak frequency (decreased by 3% compared with the baseline control, F 3,42 = 13.56, P= 0.742, Fig. 3D).

Oligomeric Aβ 1-42 affects EPSC and IPSC properties through the mTOR/S6K1 pathway
γ oscillations emerge from the rhythmic, mutual interactions of pyramidal neurons and interneurons in the local network [30], and extracellularly recorded γoscillations reflect mainly the amplitude and synchronicity of rhythmic IPSCs. To assess whether the Aβ 1-42 -induced impairment of γ-oscillations can result from the changes in synaptic properties and/or synchronicity, we tested the effect of Aβ 1-42 on EPSCs and IPSCs, recorded in CA3 pyramidal neurons in a kainateactivated neuronal network (Fig. 5).
Since PF4708671 could prevent the effect of Aβ 1-42 on γ-oscillations, we tested whether S6K1 activation was involved in the effect of Aβ 1-42 on IPSC and EPSC properties. As described for rapamycin, hippocampal slices were pretreated with PF4708671 (20 μM). PF4708671 alone had no significant effect on the amplitude or frequency of IPSCs and EPSCs (Fig. 5, Tables 1 and 2) but slightly reduced the EPSC decay time (by 20%, t 24 = 2.46, P= 0.021, n=12 neurons from 5 mice, Fig. 5, Table 2). In the presence of PF4708671, Aβ 1-42 did not affect IPSC amplitude or the frequency of IPSCs or EPSCs (Tables 1 and 2), but slightly reduced the EPSC decay time (by 22%, t 23 = 2.24, P= 0.035, n=14 neurons from 6 mice, Table 2). These data suggest that the Aβ 1-42induced changes to IPSC and EPSC properties are mediated through mTOR/S6K1 activation.

Degeneration of hippocampal γ-oscillations in APP/PS1 mice and rescue by inhibiting mTOR/S6K1 signaling
Gamma oscillations are markedly reduced in patients diagnosed with AD [59]. To test whether γ-oscillations are affected in an AD animal model that over-expresses Aβ, we compared γ-oscillations induced by 100 nM kainate in slices from APP/PS1 mice (4-6-month-old) with those from age-matched wild-type mice (Fig. 6A, B). The γ power of the oscillations recorded in slices from APP/PS1 mice was 64% lower than that recorded in slices from wildtype mice (Mann-Whitney U test P= 0.008, n= 18 slices from 18 mice for WT, n= 15 slices from 15 mice for APP/PS1 mice, Fig. 6A-C) and the peak frequency was higher (by 3%, Mann-Whitney U test P< 0.05, Fig. 6D). Whereas the γ-oscillation in slices of wildtype mice was very regular, reflected by peaks of 0.42±0.05 at 32±1 ms in the auto-correlogram, the autocorrelation peak of APP/PS1 mice was smaller (0.29±0.03, t 31 = 2.141, P= 0.040) at 29±1 ms (example in Fig. 6E), indicating that the oscillation was less regular. Representative Western blot quantification of mTOR and the serine 2448 phosphorylated mTOR (p-mTOR). Relative protein level was calculated by normalizing phosphorylated form levels to their corresponding total protein levels and total protein levels to the β-actin levels, respectively. Data are presented as mean ± SEM. *, P< 0.05, compared with WT, t-test, n=12 mice for each group.
Because inhibition of the mTOR/S6K1 pathway could prevent the Aβ 1-42 -induced reduction of γoscillations, we predicted that inhibition of the mTOR/S6K1 pathway would restore γ-oscillations in APP/PS1 mice. To test this, 200 nM rapamycin or 20 μM PF4708671 were applied after the kainate-induced γoscillations were established for 60 minutes (Fig. 7A). Compared with the APP/PS1 control, the application of rapamycin increased γ power (example in Fig. 7A and B) by 25% (the control γ power was 105% of baseline, P< 0.05, Mann-Whitney U test, n= 15 slices from 15 mice for control, n= 14 slices from 14 mice for rapamycin, Fig. 7C) and reduced the peak frequency by 3% (P< 0.05, Mann-Whitney U test, Fig. 7D). Compared with the APP/PS1 control, the application of PF4708671 increased the γ power (example in Fig. 7A and B) by 19% (P< 0.05, Mann-Whitney U test, n=15 slices from 15 mice for PF4708671, Fig. 7C) and reduced the peak frequency by 3% (P< 0.05, Mann-Whitney U test, Fig. 7D). To determine the role of eIF4E-eIF4G interactions in the suppression of γ-oscillations in APP/PS1 mice, we also tested the effect of 4EGI-1 on γ-oscillations. 4EGI-1 (50 μM, application for 60 min) had no significant effect on γ power (increased by 1% compared with the control, P< 0.05, Mann-Whitney U test, n= 14 slices from 14 mice, Fig. 7A, B and C) or the peak frequency (increased by 6% compared with the control, P< 0.05, Mann-Whitney U test, Fig. 7D). This suggests that the degradation of γ oscillations in APP/PS1 mice is partly caused by an overactivity of the mTOR/S6K1 pathway.
To examine whether mTOR was over-activated in APP/PS1 mice, hippocampal CA3 tissue was dissected for Western blotting. Total mTOR levels were not different between APP/PS1 and wild-type mice (t 4 = 0.551, P= 0.601, each sample collected from 3 mini slices of hippocampal CA3 region of 3 mice, n= 4 independent experiments from 12 mice, Fig. 7E). However, compared to that in wildtype, mTOR serine 2448 phosphorylation in CA3 of APP/PS1 mice was increased by 38% (t 4 = 3.216, P= 0.018, Fig. 7E). The increased mTOR serine 2448 phosphorylation was strikingly similar to the increase induced by Aβ 1-42 in wildtype slices (Fig. 3E) and suggests that the mTOR/S6K1 pathway mediates the disruption of γ-oscillations in AD.

DISCUSSION
In this study, we investigated the role of oligomeric Aβ 1-42 in modulating hippocampal γ-oscillations in vitro. Aβ 1-42 strongly suppressed γ-oscillations while increasing the peak frequency. Aβ 1-42 -induced hyper-activation of mTOR was both necessary and sufficient to suppress γoscillations through activation of S6K1 rather than through inhibition of eIF4E. Activation of the mTOR/S6K1 pathway decreased spontaneous EPSC amplitude and frequency and IPSC frequency. Inhibition of the mTOR/S6K1 pathway rescued the impairment of hippocampal γ-oscillations in APP/PS1 mice.
Oligomeric Aβ 1-42 is increased in the brain tissue of AD patients and animal models [60]. Because increased levels of oligomer Aβ 1-42 are observed early in the development of AD, it has been the target of AD therapies [61,62].
The oligomeric Aβ 1-42 concentrations that were effective in suppressing γ-oscillations are within the concentration range of oligomeric Aβ, found in wet brain tissue of AD patients [53]. Still, no information is available about the oligomeric Aβ 1-42 concentration in the extracellular space. Aβ 1-42 was more potent in its oligomeric or fibrillar form than as monomers, which confirms a study by Kurudenkandy et al. [25] and is in line with the increased toxicity of the oligomeric form [63,64]. Although the fibrillar Aβ 1-42 was equally potent as the oligomeric form, it is unlikely that large fibrils can affect γ-oscillation-generating networks directly [25] because the average extracellular space is only ~50 nm [51]. Most likely, fibrils are turned into oligomers in the tissue through a fibril-catalyzed secondary nucleation reaction [52]. The < 10 nm diameter oligomers will probably diffuse into the tissue and affect the γ-generating networks.
A similar impairment of γ-oscillations was observed in the 4-month-old APP/PS1 mouse [16] and has been described at very early stages in other AD animal models [20,21,23]. Early-stage AD is associated with the hyperactivation of mTOR in patients and animal models [36,37]. Oligomeric Aβ 1-42 caused a hyper-activation of mTOR that was necessary for the effect of Aβ 1-42 on γoscillations, as shown by the effect of the mTORC1 inhibitor rapamycin. Hyperactivation of mTOR was also demonstrated in the hippocampus of 4-month-old APP/PS1 mice. Rescuing impaired γ-oscillations in slices from APP/PS1 mice by rapamycin confirms that mTOR1 hyper-activation is necessary for the Aβ 1-42 effect on γoscillations. Hyperactivation of mTORC1 was also sufficient for the Aβ 1-42 -induced impairment of γoscillations, since activating the mTOR1 pathway in neurons by leucine [65], could mimic the Aβ 1-42 effect. The rapamycin-sensitive mTORC1 suppresses autophagy and promotes protein synthesis by phosphorylating S6K1 and 4E-BPs [66]. Whether the acute impact of Aβ 1-42 on γ-oscillations is caused by suppression of autophagy remains to be determined. Our results identify the mTOR/S6K1 pathway as responsible for the suppression of γ-oscillations. Because eIF4E phosphorylation levels are elevated significantly in the later stages of AD, where they are associated with tau hyper-phosphorylation [56], the mTOR/S6K1-induced suppression of γ-oscillations by oligomeric Aβ 1-42 may be instrumental in the cognitive deficits, especially in the early stages of AD. In addition, a lack of γ-oscillations was proposed to contribute to the development of AD-associated pathology [23].
The suppression of γ-oscillations by mTOR hyperactivation, whether by Aβ 1-42 , leucine or in the APP/PS1 AD model, was invariably accompanied by an increase in peak frequency, which was reduced by rapamycin that restored γ power as well. This inverse relation between γ power and peak frequency is expected since the IPSC amplitude determines the cycle length [67]. It suggests that the Aβ 1-42 -induced reduction in power and increase in the frequency of γ-oscillations, results from a reduced amplitude of the combined IPSCs of interneurons contributing to the synchronization. However, mTOR hyper-activation did not affect the amplitude of spontaneous (not rhythmic) IPSCs, indicating normal GABAergic synaptic transmission. The mTOR/S6K1-induced reduction of IPSC frequency points to a reduced presynaptic GABA release probability. This can be due to a reduction of AMPA receptor-mediated fast EPSCs on interneurons. Fuchs et al. [10] showed that a reduced AMPA receptor-mediated recruitment of parvalbumin-expressing interneurons caused a reduction of γ power, an increase in peak frequency, and a reduction in regularity, a pattern identical to the Aβ 1-42 effect on γoscillations. The Aβ 1-42 -induced reduced amplitude and frequency of spontaneous EPSCs in CA3 pyramidal neurons can explain a reduced interneuron activation. Our results are in line with the observations of Ramirez et al. [68], who showed that rapamycin increases the frequency of miniature EPSCs of rat hippocampal primary neurons by modulating neurotransmitter release [68]. Interestingly, a decreased glutamate release probability and a reduced spontaneous EPSC frequency were also observed in the CA3 region of 4-month-old APP/PS1 mice [69,70]. Our observations differ from those of Kurudenkandy et al. [25], who report increased