Neuronal Cx3cr1 Deficiency Protects against Amyloid β-Induced Neurotoxicity

Cx3cr1, the receptor for the chemokine Cx3cl1 (fractalkine), has been implicated in the progression and severity of Alzheimer’s disease-like pathology in mice, but the underlying mechanisms remain unclear. A complicating factor is that Cx3cr1 has been demonstrated in both neurons and microglia. Here, we have dissected the differences between neuronal and microglial Cx3cr1, specifically by comparing direct amyloid-β-induced toxicity in cultured, mature, microglia-depleted hippocampal neurons from wild-type and Cx3cr1-/- mice. Wild-type neurons expressed both Cx3cl1 and Cx3cr1 and released Cx3cl1 in response to amyloid-β. Knockout of neuronal Cx3cr1 abated amyloid-β-induced lactate dehydrogenase release. Furthermore, amyloid-β differentially induced depression of pre- and postsynaptic components of miniature excitatory postsynaptic currents, in a peptide conformation-dependent manner. Knockout of neuronal Cx3cr1 abated effects of both amyloid-β conformational states, which were differentiable by aggregation kinetics and peptide morphology. We obtained similar results after both acute and chronic treatment of cultured neurons with the Cx3cr1 antagonist F1. Thus, neuronal Cx3cr1 may impact Alzheimer’s disease-like pathology by modulating conformational state-dependent amyloid-β-induced synaptotoxicity.


Introduction
Alzheimer's disease (AD) is a neurodegenerative disorder characterized clinically by progressive cognitive dysfunction. Pathologic hallmarks include senile plaques of aggregated amyloidβ (Aβ) and neurofibrillary tangles of hyperphosphorylated tau, coupled with dysfunctional and degenerating neurons [1]. In mouse models, Aβ evokes behavioral deficits, synaptic dysfunction, and neurotoxicity, either directly by interacting with neurons [2] or indirectly by activating microglia [3].
(Worthington Biochemical, Lakewood, New Jersey) for 45 minutes at 37°C. For immunohistochemistry, cells were plated at a density of 1.0 × 10 5 /cells per cm 2 on 18 mm glass coverslips pre-coated with 20 μg/ml poly-L-lysine at least 24 hours prior to plating. For immunoblotting, RT-PCR, and ELISA analyses of supernatants, cells were plated at 5 × 10 6 per 9.6 cm 2 well in 6-well Nunc multi-well plates pre-coated with 20 μg/ml poly-L-lysine at least 24 hours prior to plating. On DIV3, cells were treated with 2 μM cytosine arabinoside, an inhibitor of DNA replication, (Sigma-Aldrich, St. Louis, Missouri) for 3 days to prevent mitotic, non-neuronal cell growth. Under these conditions, Iba1 + microglial contamination was <<0.1% (Fig 1), confirming previous reports [2]. Cells were maintained in B27/neurobasal (NB) medium containing 2 mM glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin in a humidified atmosphere of 95% air and 5% CO 2 at 37°C for 21 days with partial media changes every 3 days. Each experiment (n) consisted of the collective mixed cortical or hippocampal tissue extracted from a single litter.

Microglial cultures
Primary microglia were cultured from P1-P10 C57Bl/6 and Cx3cr1 -/pups. After removal of cerebellum, midbrain, and hindbrain structures, whole brain extracts were mechanically dissociated and plated in 10% FBS-containing DMEM. Cells were grown at 37°C in 5% CO 2 , with partial media changes once weekly. Monocyte colony stimulating factor (MCSF)was added on DIV10. After 2 weeks of culture, flasks were gently shaken and the supernatant containing the microglia was collected. After centrifugation, microglia were plated and maintained in 10% FBS-containing DMEM until harvested for RT-PCR.

Immunocytochemistry
Neurons were fixed with 4% formaldehyde for 30 min at 25°C; washed with PBS; permeabilized in 0.1% Triton X-100-containing PBS for 10 minutes; and blocked in 10% donkey serum-containing PBS for 30 min. Cells were incubated overnight at 4°C with primary antibodies: NFH, polyclonal, 1:1000 dilution, Covance, Princeton, New Jersey; GFP, 1:1000 dilution, Life Technologies, Grand Island, New York; CD11b, monoclonal 1:1000 dilution, Abcam, Cambridge, Massachusetts; IBA1, polyclonal, 1:500 dilution, Wako Chemicals, Richmond, Virginia) in 0.05% Triton X-100 and 1% donkey serum-containing PBS. After washing with PBS, cells were incubated with Alexa Fluor 488-and Alexa Fluor 568-conjugated secondary antibodies (1:500 dilution; Life Technologies Molecular Probes, Grand Island, NY), in 0.05% Triton X-100 and 1% donkey serum-containing PBS for 30 min at 25°C. Coverslips were then mounted using Gel/Mount (Biomeda, Foster City, California). Fluorescent images were acquired with a Zeiss Axiovert 100M laser scanning confocal microscope. Pearson's colocalization coefficient and Mander's coefficients for each respective channel were obtained with Image J software. AT-8 intensity was quantified by determining integrated density of the 488 channel in all images with integrated density of the 568 channel within one standard deviation of the mean integrated density value of all images in ImageJ (NIH).

Quantitative PCR
Total RNA was purified from cultured wild-type and Cx3cr1 -/neurons and wild-type microglia at DIV21. Cells were first washed with ice-cold PBS, then placed in TRIzol (Life Technologies) reagent after washing cells with ice-cold PBS. RNA was first extracted by addition of chloroform and precipitation with methanol and then converted to cDNA using a Superscript III Supermix kit (Life Technologies Invitrogen). Real-time PCR was performed in a total reaction volume of 22 μl using 2 μl cDNA, 11.5 μl 2X TaqMan PCR Master mix, and 1.25 μl primer Preparation and maintenance of mature hippocampal cultures results in glial depletion at DIV21. Hippocampal primary neurons prepared from C57Bl/6 mice on P0 using a densitometric gradient centrifugation protocol were treated with 2 μM cytosine arabinoside (AraC) on DIV3 to enrich for neurons and deplete microglia. Prior to treatment at DIV3, GFAP+ cell contamination (A) was 21% ± 5 and IBA1+ cell contamination (B) was < 0.003%. At DIV 21, GFAP+ cell contamination was 6% ± 0.5 and IBA1+ cell contamination was < 0.003%. Microglial depletion was confirmed by staining for Cd11b (data not shown). n = 3 experiments in each sample category, with 300 cells/experiment. Scale bars, 10 μm. Cx3cr1 in Alzheimer's Disease and Aβ Neurotoxicity -probe mix. Mouse Cx3cr1 primers used for quantitative PCR were purchased from Life Technologies Applied Biosystems (Mm00438354_m1). Real-time PCR reactions were run on an Applied Biosystems 7900HT system using the standard protocol provided by Invitrogen. Calculated copies were normalized against copies of housekeeping gene GAPDH (Applied Biosystems, cat no. Mm99999915_g1).

LDH Cytotoxicity
Aβ (1-42) cytotoxicity was quantified by analyzing LDH content in culture medium from Aβ (1-42)-and vehicle-treated neurons using a Cytotoxicity Detection Kit (Roche Life Science, Indianapolis, Indiana), according to the manufacturer's instructions. Results were expressed as a percentage of maximal LDH release in culture medium from neurons treated with 0.2% Triton X-100 for 5 minutes immediately prior to assay.

ELISA assay
Conditioned media collected from neurons treated with Aβ (1-42) for 24 hours on DIV 21 was assayed for soluble murine Cx3cl1 using an ELISA kit (R&D Systems, Minneapolis, Minnesota), according to the manufacturer's instructions. Optical density was determined at 490 nm. The detection limit was 25 ng/mL.
For electrophysiology experiments, DIV14-21 microglia-depleted hippocampal cultures were incubated at room temperature with 200 nM or 2 μM Aβ (1-42) for 30 minutes prior to recording and were superperfused with identical concentrations of Aβ (1-42)-containing artificial CSF at a rate of 1.5 mL/min.
For electrophysiology experiments, non-HFIP-solubilized synthetic Aβ (1-42) was equilibrated to room temperature for 30 minutes, reconstituted to 4 μM in artificial CSF (composition described under electrophysiology methods) and sonicated for 10 minutes, before dilution to final treatment concentrations in either NB medium or artificial CSF.

Recombinant met-Aβ (1-42) and met-Aβ (1-40)
Recombinant met-Aβ (1-42) (MDAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVG GVV IA) and met-Aβ (1-40) (MDAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVG GVV) were expressed in the E. coli BL21 Gold DE3 strain (Stratagene, La Jolla, CA) and purified as described previously, with slight modifications [43]. Briefly, the purification procedure involved sonication of E. coli cells, dissolution of inclusion bodies in 8 M urea, and ion exchange in batch mode on diethylaminoethyl cellulose resin. Eluates were analyzed using SDS-PAGE for the presence of the desired protein product. The fractions containing the recombinant protein were combined, frozen using liquid nitrogen, and lyophilized. Solutions of monomeric peptides were prepared by dissolving lyophilized peptide in 6 M GnHCl. Monomeric forms were purified from potential oligomeric species and salt using a Superdex 75 10 ⁄ 300 GL column (GE Healthcare, Buckinghamshire, UK) at a flow rate of 0.7 mL/min, and were eluted in either a 50 mM ammonium acetate buffer of sodium phosphate buffer, pH 8.5. The monomeric fractions were then lyophilized and stored as identical 0.1 mg aliquots at −80°C for further experiments. Each aliquot was only thawed and reconstituted once. Post monomer purification, oligomeric species of met-Aβ (1-42) and met-Aβ  were prepared for culture treatment as described above with and without solubilization in HFIP.

Amino terminus-modified Cx3cr1 ligand F1
The Cx3cr1 antagonist F1 produced and characterized as described previously [44] was reconstituted in PBS at a stock concentration of 100 μM. Cultures were treated at a working concentration of 1 μM on DIV3 of culture or concurrently to treatment with met-Aβ (1-42).

Aβ aggregation analysis by Thioflavin T (ThT) fluorescence
For aggregation kinetic analysis, 200 nM and 2 μM synthetic and recombinant Aβ, respectively, with and without HFIP solubilisation, were prepared as described above. Samples were supplemented with 20 μM ThT from a 1 mM stock in half-area, black polystyrene, clear-bottomed, PEG-coated 96-well plates (Corning 3881). Experiments were performed in artificial CSF under quiescent conditions at room temperature (22-25°C). Fluorescence spectra of 20 μM ThT (obtained from 1 mM stock) in separate solutions (80 μL/sample) containing different Aβ preparations were measured with an Optima Fluostar plate reader (BMG Labtech, Ortenburg, Germany) (constant excitation wavelength 440 nm, emission 480 nm) for a period of 22 hours. Background fluorescence spectra obtained from 20 μM ThT in plain artificial CSF was subtracted, and resultant fluorescent measurements were plotted in Graphpad Prism (Version 6). All reported fluorescence values are an average of triplicate repeats per sample. Half-times of sigmoidal aggregation curves to maximal ThT fluorescence were determined by obtaining polymer mass, accounting for monomer-dependent secondary nucleation, with the following equation, as described previously [39].

Aβ compositional analysis by atomic force microscopy (AFM)
Samples were deposited onto freshly cleaved mica surfaces and allowed to dry for 30 minutes. At least two replicates of each condition were prepared. Samples were then washed with Milli-Q water, dried with nitrogen, and stored at 4°C. Between 3 and 7 AFM images were collected from each sample condition.
AFM images were acquired using a VEECO Dimension 3100 atomic force microscope (Brucker) and JPK Nanowizard software (Cambridge, UK). The instrument was operated in tapping mode in air using n-type silicon cantilevers with resonant frequencies between 65 and 130 kHz and force constants between 0.6 and 2 N/m. Tip (MikroMasch, Wetzlar, Germany) radii, heights, and cone angles were 8 nm, between 12 and 18 μm, and 40 degrees, respectively. Bulk probe sensitivity was between 0.01 and 0.025 O/cm. Probe mat Images were collected at a scan rate of 0.5 Hz, and with resolution varying between approximately 0.5 and 24 nm/pixel. Image preprocessing was conducted using Gwyddion software and included background leveling with a fourth order polynomial correction, and correcting horizontal lines by matching height median, and horizontal scars, if necessary. Image features were identified in Gwyddion. A thresholding algorithm was used for the recombinant samples with and without HFIP with height thresholds of between 8.5 and 9.5% of maximal values. The synthetic HFIP and non-HFIP samples had complex and feature-rich backgrounds, so in these cases a watershed algorithm was used for enhanced segmentation. Image specific parameters were applied for optimal extraction of features over background. In each case, several thousand image features were considered. The maximum value of each image feature was measured One or several Gaussian distributions were fit to these data in Matlab and the mean and standard deviation of the peak(s) were extracted.

mEPSC recording and analysis parameters
Whole-cell mEPSC recordings were obtained using an Axon Multiclamp 700B amplifier (Molecular Devices, Sunnyvale, CA). Recordings were low pass-filtered at 2 kHz and acquired at 5 kHz using an Instrutech ITC-16 acquisition board (Instrutech, Port Washington, NY) and custom-made acquisition protocols programmed in Igor Pro software (WaveMetrics, Lake Oswego, OR). All experiments were performed in voltage clamp mode at a holding potential of −80 mV. GABAergic currents were assessed at 0 mV and were completely blocked under the described recording conditions. Recordings were discarded if leak current was greater than 200 pA, input resistance was greater than 100 MO, series resistance was greater than 30 MO, or if series resistance fluctuated more than 15%. mEPSC detection and analysis were conducted by a semi-automated method using custom protocols programmed in Igor Pro (Wavemetrics). Analysis was limited to recordings with a minimum of 50 events in each one-minute interval analyzed and averaged. We first calculated the first derivative of the raw data trace with respect to time; events were defined as negative amplitude fluctuations of 5-100 pA with monotonic rise slopes of the first derivative of the tracing at least twice that of baseline noise and exponential decay time constants less than 25 ms. Frequency, peak amplitude, and decay times were averaged over at least 3 recordings per coverslip. At least 3 coverslips were averaged per experimental n.

Statistical Analysis
Unless otherwise noted, all data plots are mean ± SEM. We used Kolmogorov-Smirnov testing to determine normal distribution of data; for non-normal distributed data or for data whose sample size was insufficient to confirm normal distribution, we applied the non-parametric Mann-Whitney test. For multi-factor interaction testing, one-and two-way ANOVAs were used, as applicable. Values are considered statistically significant if P < 0.05 or, in the cases of multiple-factor testing, if P values meet post-hoc Bonferroni-corrected thresholds for significance. All two-tailed Student t-tests, one-way and two-way ANOVAs, non-parametric Mann-Whitney tests were performed using Prism, version 6 (GraphPad Software).

Cultured mature microglial-deficient hippocampal neurons express Cx3cr1
Because neuronal Cx3cr1 expression in mice is controversial [11][12][13][14][15][16] and has not yet been investigated in our system, we first assayed mature microglia-depleted hippocampal cultures from postnatal Cx3cr1 +/gfp heterozygous mice for Cx3cr1 expression. Microglial depletion was attained using a rigorous densitometric segregating centrifugation protocol (see Materials and Methods) and ensured for the lifespan of the culture by treatment with 2 μM cytosine arabinoside (AraC) on DIV3 (Fig 1).
We found that EGFP expressed under the endogenous Cx3cr1 promoter significantly colocalized with Neurofilament H (NFH)-positive cell bodies, axons, and dendrites in hippocampal cultures (Fig 2A). Manual cell counts of NFH-and EGFP-positive cell bodies indicated that the majority of NFH-positive neurons in hippocampal culture (91.9% ± 2.4%) were also positive above background for EGFP, a result corroborated by Cx3cr1 RNA transcript detection ( Fig  2B). While mixed cortical cultures also expressed Cx3cr1 RNA, less than half of NFH-positive neurons (36% ± 7.4%) in these cultures were also positive for EGFP above background. These results indicate that NFH-positive neurons within our system of mature, microglia-deficient neuronal culture expressed Cx3cr1 RNA and protein and that the degree of protein expression was regionally selective (P = .01, two-tailed Student's t-test).
To test whether or not knockout of neuronal Cx3cr1 abates Aβ (1-42) neurotoxicity, we treated wild-type and Cx3cr1 -/neurons from microglia-depleted mixed cortical and hippocampal cultures with 2 μM human Aβ (1-42) on DIV21, conditions previously identified to induce neurotoxicity directly in the absence of glia [2,46]. We used LDH release and βIII-tubulin immunoreactivity ( Fig 3A) to assess Aβ(1-42)-induced cytotoxicity, and found significantly reduced levels of LDH release and cell loss in cultured Cx3cr1 -/neurons compared to wild type neurons (Fig 3B).
Consistent with the neuronal Cx3cr1 expression data, the reduction in LDH release conferred by Cx3cr1 deficiency was more pronounced in hippocampal cultures than in mixed cortical cultures. In contrast, we found that Aβ(1-42) induced significantly increased levels of LDH release and cell loss as assessed by tubulin immunoreactivity in purified microglia cultures from Cx3cr1 -/mice compared to those from wild type mice, indicating that Cx3cr1 differentially affects cellular responses to Aβ  in neurons and in microglia (Fig 3C and 3D).

Aβ(1-42) potentiates Cx3cl1 cleavage from neurons
Cx3cl1 is a plasma membrane protein that can be cleaved to release the N-terminal chemokine domain, which signals either in a paracrine or autocrine manner through Cx3cr1. Since inflammatory cytokines and Aβ itself enhance Cx3cl1 signalling preceding cell death in various disease Since cognitive dysfunction correlates more closely with synapse loss than with cell loss in humans and because synaptic dysfunction precedes degeneration in AD animal models [23], we investigated the effects of Aβ (1-42) treatment on AMPAR-dependent miniature excitatory postsynaptic currents (mEPSCs) in mature microglia-depleted hippocampal cultures. AMPAR-dependent mEPSCs, representative of network stabilization, are perturbed in AD mouse models [24-26, 28-30, 47]; mEPSC frequency primarily reflects presynaptic release probability, while mEPSC amplitude primarily reflects postsynaptic mechanisms of receptor trafficking and sensitization.
Because synaptotoxicity appears to be critically dependent on precise conformational states of Aβ and because oligomeric intermediates of Aβ  are considered the most potent synaptotoxic species [41], we characterized the physico-chemical properties of several Aβ preparations, known to induce either cytotoxicity or acute synaptic plasticity deficits, by assessing the aggregation kinetics of fibril formation for each preparation under experimental recording conditions.
Because aggregation is critically dependent on properties of seeding species [38], we performed AFM to identify the initial species corresponding to these distinct aggregation profiles ( Fig 4B) and found that 2 μM HFIP Aβ (1-42) (the only 2 μM sample that did not aggregate during the experimental window) demonstrated two distinct populations of spherical oligomers with z-height values of 0.5-1.0 nm or 14-20 nm, either of which may have precluded sigmoidal aggregation during the experimental window. All other 2 μM preparations were comprised primarily of polydisperse spherical oligomers without evidence of the two distinct populations observed in the 2 μM HFIP synthetic Aβ (1-42) preparation described above. Specifically, relative to its comparable synthetic preparation, HFIP 2 μM recombinant met-Aβ (1-42) demonstrated a more homogenous distribution of spherical oligomers between 3 and 6 nm and no significant spherical oligomer populations with z-heights of 0.5-1.0 nm or 14-20 nm (Fig 4C). When synthetic Aβ (1-42) preparations were added to microglia-depleted hippocampal culture, we found that 2 μM non-HFIP Aβ (1-42) (aggregating Aβ) reduced both mean mEPSC frequency and amplitude, while 2 μM HFIP Aβ (1-42) and both 200 nM preparations (non-aggregating) decreased mean mEPSC frequency only (Fig 5A and 5B). See Table 2 for all parameters. These results suggest that Aβ(1-42) species present in both non-aggregating and aggregating preparations of Aβ selectively depress presynaptic parameters of mEPSCs, while species present during the rapid sigmoidal conformational transition of Aβ to aggregated fibrils affect both pre-and postsynaptic parameters of mEPSCs. Similarly, only aggregating preparations of recombinant met-Aβ also depressed mEPSC amplitude (Fig 6A and 6B).
2 μM recombinant met-Aβ (1-40) preparations, which demonstrated no aggregation, did not affect mEPSCs, indicating that Aβ (1-42) effects were not due to residual monomeric species (Fig 7). These results indicate that presynaptic mEPSC effects are attributable to species present in all Aβ(1-42) preparations but that postsynaptic effects are associated with rapid conformational conversion of Aβ(1-42) during the experimental interval.
Because neuronal Cx3cr1 deficiency inhibits neurotoxicity, we asked if it might also be protective of Aβ (1-42)-induced synaptic dysfunction. Interestingly, we found that relative to wild-type neurons, baseline mEPSC mean frequency and amplitude were significantly increased and decreased in Cx3cr1 -/neurons, respectively (Fig 8). The quantile coefficient of variation of quantal amplitude in Cx3cr1 -/hippocampal neurons was significantly greater (.383) than that in wild-type hippocampal neurons (.136; P = 0.0001, Mann-Whitney). We also found that no preparation of synthetic (Fig 5C and 5D) or recombinant Aβ(1-42) (Fig 6C and 6D) 3 IPP refers to peptide addition immediately post-purification without any pre-treatment preparation 4 All samples were aggregated in aCSF per materials and methods except for those denoted as prepared in sodium-phosphate (NaP) buffer 5 Table 3 for all parameters.
To further investigate whether functional neuronal Cx3cr1 facilitates Aβ(1-42)-induced suppression of AMPA-mediated mEPSCs, we treated wild-type hippocampal culture with the modified amino terminus Cx3Cl1 ligand, F1, which is a Cx3cr1 antagonist, either concurrently with (acute) or on DIV3 preceding treatment with (chronic) 200 nM non-HFIP Aβ . We found that both acute and chronic treatment with 1 μM F1 blocked Aβ (1-42)-induced reduction in AMPA-mediated mEPSC frequency and rightward shifting of the cumulative frequency distribution curve of inter event intervals (Fig 9). Thus, genetic inactivation, as well as acute and chronic pharmacologic blockade of Cx3cr1 protect against Aβ-induced synaptotoxicity.
Our results suggest that neuronal Cx3cr1 deficiency may play a neuroprotective role in the context of Aβ toxicity, but previous studies have been inconsistent with regards to neuronal expression of the receptor [4,11]. Regional specificity of neuronal Cx3cr1 expression might be one reason for these conflicting reports. In our experiments, we found that both cultured mixed cortical and hippocampal neurons expressed Cx3cr1 mRNA but that receptor protein was expressed more robustly in hippocampal neurons, in accordance with previous reports of selective hippocampal CA1 neuron marker colocalization with Cx3cr1 [14].
Our results are consistent with previous reports that hippocampal neurons are more susceptible than cortical neurons to Aβ toxicity [48]. Because Cx3cr1 expression is required for Aβ toxicity in neuronal culture, selective expression of Cx3cr1 in hippocampal neurons might contribute to their heightened susceptibility to Aβ. Discrepancies in reports of neuronal Cx3cr1 expression [11][12][13][14][15][16] might also be a result of developmental and inducible receptor regulation. All reports of cultured neuronal Cx3cr1 expression occur at DIV7 or later, and several excitotoxic stimuli, including Aβ, elevate Cx3cl1-Cx3cr1 signaling [12,16]. Interestingly, studies conducted within the human central nervous system report neuronal Cx3cr1 expression [13,16].
Dissociated, Cx3cr1 -/microglia and neurons exhibited opposite cytotoxic responses to Aβ, indicating different receptor effector functions. Whether Cx3cl1 signals through microglial or neuronal Cx3cr1 may depend on relative receptor affinities for the ligand, a suggestion supported by reports of microglial activation by low concentrations of Cx3cl1 and direct neuronal activation by higher concentrations [12,19]. Our study suggests that cell-specificity of receptor expression may be an important consideration when developing therapeutics targeting the Cx3cl1-Cx3cr1 signaling axis and other chemokine receptors expressed by both neurons and glial cells in the CNS.
In our study, Aβ increased soluble Cx3cl1 from cultured hippocampal neurons. Previous studies have implicated Cx3cl1 in concentration-dependent reduction of neurotransmitter  Tables 2 and 3 for all parameters and P values. See Fig 6E and 6F for summary of comparison between parameters of wild-type and Cx3cr1-/hippocampal neurons. doi:10.1371/journal.pone.0127730.g005 Cx3cr1 in Alzheimer's Disease and Aβ Neurotoxicity  Cx3cr1 in Alzheimer's Disease and Aβ Neurotoxicity release probability [32], reduction of amplitude of evoked AMPA currents, and direct AMPAR internalization [12,19,31,32], which are downstream effects of Aβ. Potentiation of Cx3cl1 signalling through neuronal Cx3cr1, therefore, might be one mechanism of Aβ toxicity. These in vitro results warrant validation in vivo. We found that Aβ(1-42) selectively depressed pre-or both pre-and postsynaptic parameters of AMPAR-dependent mEPSCs depending on peptide aggregation-state and that Cx3cr1 was required for both effects of Aβ. It should be noted that while mEPSC frequency is considered to be primarily a presynaptic measure of release probability, we cannot rule out potential postsynaptic mechanisms that might contribute to frequency alterations. Further, due to decreased baseline mEPSC amplitude in Cx3cr1 -/neurons, we cannot rule out inability to detect Aβ-induced amplitude reduction in these cells. While concentration-dependent selective Aβ reduction of mEPSC frequency has been reported previously [30], the effect we observed was not simply concentration-dependent, since only 2 μM Aβ preparations that also demonstrated rapid sigmoidal kinetics of aggregation during the recording interval induced postsynaptic  Table 2 for all parameters. Data were inspected for normality of distribution of frequency values by the Kolmogorov-Smirnov test, analyzed for statistical significance by two-way ANOVA by genotype and treatment condition, with post-hoc Bonferroni correction for multiple comparisons.
doi:10.1371/journal.pone.0127730.g008 effects. Collectively, the kinetic and AFM data suggest that metastable conformations associated with fibril formation were required to affect both pre-and postsynaptic parameters of mEPSCs in hippocampal culture. This finding is supported by previous studies indicating that severity and mechanisms of Aβ toxicity are known to vary, dependent on peptide aggregation state [49,[52][53][54]. It should be noted that neurons treated with HFIP Aβ demonstrated slightly decreased baseline frequency and amplitude, with more pronounced effects in Cx3cr1 -/neurons, corroborating previous reports that residual solvent may be toxic [55]. Residual solvent may also contribute to slowed aggregation kinetics of HFIP Aβ(1-42).
2 μM non-HFIP synthetic and recombinant Aβ(1-42) demonstrated comparatively high baseline ThT fluorescence and impeded aggregation profiles relative to model sigmoidal kinetics curves; as assessed by AFM, both preparations contained pre-existent fibrils, which likely contributed to these aggregation profiles. However, the ability of Aβ to affect postsynaptic mEPSC parameters did not require these pre-existent fibrils as non-fibril containing 2 μM HFIP met-Aβ (1-42) elicited similar mEPSC amplitude reduction. There are several limitations to the application of this study's results to pathophysiologic mechanisms of the aging human CNS, owing to its in vitro nature, its dependence on neonatal murine tissue, and the requirement for doses of Aβ that exceed physiological concentrations in humans to elicit neurotoxicity in this system.
In conclusion, our study shows that Aβ  intermediates present in all preparations decrease presynaptic release probability, while metastable conformations populated during the rapid phase of sigmoidal aggregation involve additional postsynaptic mechanisms of receptor internalization and desensitization. Because Cx3cr1 deficiency conferred resistance to both presynaptic and postsynaptic mEPSC parameter depression and because baseline mEPSC frequency and amplitude, were increased and decreased, respectively, in Cx3cr1 knockout neurons when compared to wild-type neurons, the neuroprotective effects of receptor knockout could be related to global network alterations or to specific receptor-Aβ interactions.
Long-term modulation of synaptic scaling by AMPAR stabilization has been suggested as a viable disease-modifying strategy in AD [56]. Our study implicates neuronal Cx3cr1 in network modifications that alter susceptibility to Aβ-induced mEPSC depression, indicating Cx3cr1 as a potential therapeutic target.

Author Contributions
Conceived and designed the experiments: JD TPK CMD CB OP PMM. Performed the experiments: JD BR JH EVY. Analyzed the data: JD BR JH EVY TPK CMD CB OP PMM. Contributed reagents/materials/analysis tools: CC. Wrote the paper: JD OP PMM.