Microglia TREM2R47H Alzheimer-linked variant enhances excitatory transmission and reduces LTP via increased TNF-α levels

To study the mechanisms by which the p.R47H variant of the microglia gene and Alzheimer’s disease (AD) risk factor TREM2 increases dementia risk, we created Trem2R47H KI rats. Trem2R47H rats were engineered to produce human Aβ to define human-Aβ-dependent and -independent pathogenic mechanisms triggered by this variant. Interestingly, pre- and peri-adolescent Trem2R47H rats present increased brain concentrations of TNF-α, augmented glutamatergic transmission, suppression of Long-term-Potentiation (LTP), an electrophysiological surrogate of learning and memory, but normal Aβ levels. Acute reduction of TNF-α activity with a neutralizing anti-TNF-α antibody occludes the boost in amplitude of glutamatergic transmission and LTP suppression observed in young Trem2R47H/R47H rats. Thus, the microglia-specific pathogenic Trem2 variant boosts glutamatergic neuronal transmission and suppresses LTP by increasing brain TNF-α concentrations, directly linking microglia to neuronal dysfunction. Future studies will determine whether this phenomenon represents an early, Aβ-independent pathway that facilitates dementia pathogenesis in humans.

Introduction AD, the most common progressive neurodegenerative disorder in the elderly, is diagnosed when clinical dementia occurs concomitantly with Ab plaques, neurofibrillary tangles and neuronal loss (James et al., 2014). Genome-wide association studies have discovered Triggering Receptor Expressed on Myeloid Cells 2 (TREM2) human variants that increase the risk of developing sporadic AD (Guerreiro et al., 2013). In the central nervous system (CNS), TREM2 is exclusively expressed in microglia (Schmid et al., 2002). This genetic evidence directly implicates microglia function in AD pathogenesis.
Microglia surround amyloid plaques both in AD patients (McGeer et al., 1987) and Ab plaquesbearing mice (Frautschy et al., 1998). Recent data suggest that these disease-associated microglia (DAM) possess enhanced activities, including the Ab plaque-clearing activity (Keren-Shaul et al., 2017;Mazaheri et al., 2017). Through diverse mechanisms Mazaheri et al., 2017;Schlepckow et al., 2017;Song et al., 2018;Ulland et al., 2015), disease--associated TREM2 variants cause a loss of function of TREM2 that inhibits microglia transition to DAMs and impairs Ab plaque-clearing activities . As for the AD-associated p. R47H TREM2 variant, in vitro studies suggest that it destroys an essential lipid and Ab-binding site within the TREM2 ectodomain, reducing the Ab-phagocytosis capabilities of microglia (Yeh et al., 2016).
Model organisms are useful tools to study how human pathogenic mutation/variants alter protein's functions and promote disease in humans; thus, to dissect the pathogenic mechanisms of the p.R47H TREM2 variant, we generated Trem2 R47H knock-in (KI) rats, which carry the p.R47H variant in the rat endogenous Trem2 gene . Rat and human APP differ by 3 amino acids in the Ab region. These differences may be crucial since human Ab may have higher propensity to form toxic species as compared to rodent Ab and the pathogenic role of the p.R47H TREM2 variant may be linked to toxic Ab clearance deficits. To eliminate this potential issue, together with the Trem2 mutation we introduced mutations to 'humanize' the rat Ab sequence (App h allele) (Tambini et al., 2019). In Trem2 R47H KI rats, transcription, splicing and translation of this pathogenic variant is controlled by endogenous regulatory elements, allowing to study pathogenic mechanisms triggered by the p.R47H TREM2 variant in a model organism mimicking the genetics of the human disease and expressing physiological levels of human Ab.
Pre-adolescent (Sengupta, 2013) Trem2 R47H rats showed no significant alterations in brain levels of human Ab40 and Ab42, the latter is considered the pathogenic Ab species . In addition, the Ab42/Ab40 ratio, another indicator of Ab-mediated pathogenesis, is not altered. Thus, it is likely that the consequences of the Ab-clearance deficits caused by the p.R47H TREM2 variant fully manifest in vivo in an aging-dependent manner. Yet, the p.R47H TREM2 may trigger both human Ab-dependent and Ab-independent pathogenic mechanism. Ab-independent mechanism may precede and, perhaps, participate in mechanisms leading to dementia.
In macrophages, TREM2 functions to inhibit pro-inflammatory cytokines production, especially TNF-a (Turnbull et al., 2006). Thus, it is possible that loss of TREM2 function caused by the p.R47H variant may favor pro-inflammatory cytokine production by microglia. In this study, we tested this hypothesis in young Trem2 R47H rats with the purpose of determining potential early pathogenic mechanisms caused by the p.R47H variant.

Results
Increased concentration of TNF-a and other pro-inflammatory cytokines in the CNS and CSF of young animals carrying the Trem2 R47H variant Pre-adolescent (4 weeks old) Trem2 R47H rats showed no significant alterations in CNS levels of human Ab40, Ab42 and the Ab42/Ab40 ratio , even though the Trem2 R47H variant reduces binding and clearance of human Ab in vitro (Zhao et al., 2018). This discrepancy prompted us to assess further Ab metabolism in Trem2 R47H rats. Reduction of Ab42 concentration in the cerebrospinal fluid (CSF) is a biomarker for AD. Aggregation of Ab42 in brain parenchyma appears the most likely cause for the decreased CSF Ab42 concentration because the aggregated state inhibits Ab42 from being transported from the interstitial fluid to the CSF. In absence of aggregation, Ab40 and Ab42 CSF levels reflect the concentrations in the CNS. Thus, we measured concentrations of human Ab38, Ab40, and Ab42 in the CSF of peri-adolescent (6-8 weeks old) rats using a human Ab specific-ELISA, as described previously . Whereas previously, a decrease in Ab38 was present in the whole brain lysate of Trem2 R47H/R47H rats, no significant decrease in Ab38 was seen in the CSF of Trem2 R47H rats ( Figure 1). No differences were seen in Ab40 and Ab42 levels and the Ab42/Ab40 ratio between Trem2 w/w , Trem2 R47H/w , and Trem2 R47H/R47H rats ( Figure 1). Thus, the reduced Ab clearance caused by the Trem2 R47H variant in vitro does not result in significant alterations of Ab steady-state in vivo, at least in young rats.
Next, we looked for other potential early changes in CNS physiology prompted by the Trem2 R47H variant. Macrophages lacking Trem2 produce more TNF-a in response to LPS, zymosan and CpG, suggesting that Trem2 functions to inhibit cytokine production by macrophages (Turnbull et al., 2006). Thus, it is possible that the p.R47H variant may alter the anti-inflammatory TREM2 function in microglia. To test for this, we used the Proinflammatory Panel 2 (rat) ELISA multiplex Kit from MSD. This multiplex kit allows for quantitative determination of 9 cytokines (IFN-g, IL-1b, IL-4, IL-5, IL-6, CXCL1, IL-10, IL-13, and TNF-a) that are important in inflammation response and immune system regulation. First, we measure cytokine levels in the CNS of 4 weeks old rats, the same rats tested previously for Ab levels . IL-1b, CXCL1 A and TNF-a were significantly increased in Trem2 R47H/R47H rats whereas IFN-g, IL-4, IL-5, IL-6, IL-10 and IL-13 were not ( Figure 2).
Augmented excitatory synaptic transmission at hippocampal SC-CA3 >CA1 synapses of peri-adolescent rats carrying the Trem2 R47H variant Among the cytokines tested, TNF-a is the only one that is significantly increased in both pre-adolescent CNS tissue and peri-adolescent CSF. Physiological levels of TNF-a produced by glia cells are required for normal surface expression of AMPA (a-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptors at synapses, and increased TNF-a concentrations cause a rapid exocytosis of AMPA receptors in hippocampal pyramidal neurons increasing excitatory synaptic strength (Beattie et al., 2002;Ogoshi et al., 2005;Stellwagen et al., 2005;Stellwagen and Malenka, 2006). Interestingly, Amyloid Precursor Protein (APP, which codes for the precursor of Ab) (Del Prete et al., 2014;Fanutza et al., 2015;Groemer et al., 2011;Kohli et al., 2012;Lundgren et al., 2015;Norstrom et al., 2010;Tambini et al., 2019;Yao et al., 2019a), Presenilin 1/2 (PSEN1 and PSEN2, which code for the catalytic components of the g-secretase complex, an enzyme essential for Ab production) (Wu et al., 2013;Xia et al., 2015) and Integral Membrane Protein 2B (ITM2b, whose protein product binds APP and regulates APP processing) (Fotinopoulou et al., 2005;Matsuda et al., 2005;Matsuda et al., 2008;Matsuda et al., 2011;Yao et al., 2019b) play a physiological role in glutamatergic transmission. Moreover, some APP, PSEN1 and ITM2b mutations linked to familial dementia alter this physiological function (Tamayev et al., 2010a;Tamayev et al., 2010b;Tambini et al., 2019;Xia et al., 2015;Yao et al., 2019b). Based on these data, we examined the effects of the Trem2 R47H variant on glutamatergic synaptic transmission in the hippocampal Schaffercollateral pathway. First, we analyzed miniature excitatory postsynaptic currents (mEPSC). The amplitude of mEPSC, which is dependent on postsynaptic AMPA receptors, was significantly increased in Trem2 R47H/R47H rats ( Figure 5B,E and F) while the decay time was unchanged ( Figure 5C and E). In addition, the frequency of mEPSC was significantly increased in Trem2 R47H/R47H rats ( Figure 5D and G). To test further whether postsynaptic AMPAR-mediated responses are increased in Trem2 R47H Figure 4. Treml1 mRNA expression is normal in Trem2 R47H rats. (A) Levels of Treml1 mRNA were measured and normalized to Gapdh mRNA expression. We used microglia purified from 2 male and 2 female 5/6 weeks old rats for each genotype (Trem2 w/w and Trem2 R47H/R47H ). Data were analyzed by unpaired student's t-test (p=0.5198), and presented as average (Treml1/Gapdh)± SEM. (B) Treml1 mRNA expression was measured in total brains. We used 5 female and 5 male (6-8 weeks of age) for each genotype (Trem2 w/w , Trem2 R47H/w and Trem2 R47H/R47H ). Data are represented as (Treml1/Gapdh) mean ± SEM and were analyzed by ordinary one-way ANOVA (F (2, 27) =1.940, p=0.1632). The online version of this article includes the following source data for figure 4: Source data 1. Related to Figure 4A,B. rats, we measured AMPAR and NMDAR-dependent synaptic responses. Consistent with the hypothesis that the Trem2 R47H variant increases AMPAR-mediated responses, the AMPA/NMDA ratio was increased in both Trem2 R47H/R47H and Trem2 R47H/w rats and the significance of the increase is gene dosage-dependent ( Figure 5H).
Treatment of primary neurons with recombinant TNF-a (Beattie et al., 2002;Grassi et al., 1994) and Trem2 deletion in mice (Filipello et al., 2018) enhance frequency of mEPSC. Several mechanisms, including an increase in release Probability (Pr) of glutamatergic synaptic vesicles and/or synaptic density, can enhance frequency of mEPSC. Paired-pulse facilitation (PPF), a form of short-term synaptic plasticity, is determined, at least in part, by changes in Pr; an increase in Pr leads to a decrease in facilitation (Zucker and Regehr, 2002). PPF was not significantly changed in Trem2 R47H/ w and Trem2 R47H/R47H rats as compared to Trem2 w/w animals ( Figure 5I). This evidence argue against changes in Pr in Trem2 R47H/R47H rats and suggests that enhanced mEPSC frequency may be due to increased synaptic density, a phenomenon that in Trem2 -/mice is due to impaired synapse elimination by microglia (Filipello et al., 2018).
Supraphysiological TNF-a concentrations boost glutamatergic transmission in hippocampal SC-CA3 >CA1 synapses of young Trem2 R47H/R47H rats The similarities between glutamatergic transmission alterations observed in young Trem2 R47H rats and those induced by TNF-a are striking. To test whether the supraphysiological TNF-a concentrations trigger glutamatergic deficits in Trem2 R47H/R47H rats, we treated hippocampal slices with a neutralizing antibody to rat TNF-a (anti-TNF-a), which functions as a TNF-a antagonist. To control for off-target effects of the antibody, we used a Goat IgG isotype control. The 50% neutralization dose (ND 50 ) of this anti-TNF-a antibody against the cytotoxic effect of recombinant rat TNF-a (0.25 ng/ mL) is about 500 ng/ml. Since physiological levels of TNF-a are necessary for normal glutamatergic transmission and most of the activities of TNF-a can be rapidly reversed (Beattie et al., 2002;Ogoshi et al., 2005;Stellwagen et al., 2005;Stellwagen and Malenka, 2006), we tested the acute effects 10 ng/ml of anti-TNF-a, a concentration~50 times lower than the ND 50 . At this concentration, anti-TNF-a occluded the increase in mEPSC amplitude ( Figure 6A,B,E and F) and AMPA/ NMDA ratio ( Figure 6H). Anti-TNF-a did not alter decay time of mEPSC ( Figure 6C) and PPF ( Figure 6I), which were not significantly changed in Trem2 R47H/R47H rats (see Figure 5C and I). Surprisingly, anti-TNF-a did not reduce mEPSC frequency of Trem2 R47H/R47H SC-CA3 >CA1 synapses to WT-like levels ( Figure 6D and F).
The evidence that the goat IgG isotype control did not occlude glutamate transmission alterations observed in the mutant rats ( Figure 6) indicates that the effects of anti-TNF-a are specific. The finding that these low doses of anti-TNF-a do not alter glutamatergic transmission in Trem2 w/w rats ( Figure 6) demonstrates that, at least at this dosage, anti-TNF-a only targets synaptic transmission alterations triggered by excess TNF-a set off by the Trem2 R47H variant. Overall, these data indicate   F (2, 30) =7.092, p=0.0030**; post-hoc Tukey's multiple comparisons test: w/w vs. RH/w, p=0.4881 (ns); w/w vs. RH/RH, p=0.0025**; RH/w vs. RH/RH, p=0.0345*]. (E) Average mEPSC of the three groups depicts differences in amplitude. As can also be noted in B, Trem2 R47H/w rats show mEPSC with increased amplitude, albeit this increase did not reach statistical significance. Cumulative probability of AMPAR-mediated mEPSC amplitudes (F) and inter event intervals (G). (H) AMPA/NMDA ratio is significantly increased in both Trem2 R47H/R47H and Trem2 R47H/w rats in a gene dosage dependent manner [F (2, 27) =0.1397, p=0.8700]. Representative traces are shown on top of the panels. Data are represented as mean ± SEM and were analyzed by ordinary one-way ANOVA followed by post-hoc Tukey's multiple comparisons test when ANOVA showed significant differences. For each type of recordings, we indicate the number of animals by genotype and sex, plus the number of recording by genotype and sex as follow: 1) genotypes: w/w = Trem2 w/w, RH/w = Trem2 R47H/w, RH/RH = Trem2 R47H/ R47H ;2) sex: R47H/R47H = female, M = males; 3) number of animals and number of recordings from animals: n/n', were n = number of animals, n'=number of recordings from the n animals. For example, the w/w: F = 4/6; M = 4/5 in A indicates that data for mEPSC for the Trem2 w/w rats were obtained from 4 females and 4 males, and that 6 recordings were obtained from the 4 females and 5 recordings from the 4 males. The online version of this article includes the following source data for figure 5: Source data 1. Related to Figure 5B,C,D,H,I. The increase in mEPSC amplitude caused by the Trem2 R47H variant is occluded by anti-TNF-a application but not by the IgG isotype control [F (3, 54) =18.79, p<0.0001****; post-hoc Tukey's multiple comparisons test: w/w + anti-TNF-a vs. RH/RH + anti-TNF-a, p=0.7884 (ns); w/w + anti-TNF-a vs. RH/RH + Isotype, p<0.0001****; w/w + anti-TNF-a vs. w/w + Isotype, p=0.9252 (ns); RH/RH + anti-TNF-a vs. RH/RH + Figure 6 continued on next page that the increase in amplitude of AMPAR-mediated responses at SC-CA3 >CA1 synapses of Trem2 R47H/R47H rats is due to the acute action of supraphysiological TNF-a concentrations prompted by theTrem2 R47H variant and is rapidly reversable.
LTP is suppressed at hippocampal SC-CA3 >CA1 synapses of young Trem2 R47H/R47H rats: this deficit is rescued by reducing TNF-a activity LTP, a long-lasting form of synaptic plasticity, has been described at glutamatergic synapses throughout the brain and remains one of the most attractive cellular models for learning and memory. Given the alterations in glutamatergic transmission observed in Trem2 R47H rats, we determined whether the Trem2 R47H variant could also impact this electrophysiological surrogate of memory in young rats (6-8 weeks old peri-adolescent rats). Before recording LTP, baseline was recorded every minute at an intensity that elicited a response 40% of the maximum evoked response. Maximum evoked responses were assessed by measuring the slope of the field excitatory postsynaptic potential (fEPSP) elicited by stimuli of increasing intensity (basal synaptic transmission or BST). Consistent with the evidence that Trem2 R47H variant enhances AMPAR-mediated responses, BST was increased in Trem2 R47H/R47H rats ( Figure 7A). Surprisingly perhaps given the young age of the animals, LTP was reduced in Trem2 R47H/R47H rats ( Figure 7B and C), indicating that the Trem2 R47H variant compromises LTP early in life. Interestingly, anti-TNF-a occluded the increase in BST ( Figure 7D) and suppression of LTP ( Figure 7E and F). This effect is specific and indicates that also LTP suppression in young Trem2 R47H/R47H rats is caused by the sustained and acute action of supraphysiological TNFa concentrations prompted by theTrem2 R47H variant and is rapidly reversable.

Discussion
Here, we studied young rats carrying the Trem2 R47H variant, which increases the risk for AD in humans, with the purpose of probing early dysfunctions that may underlie initial pathogenic mechanisms leading to dementia. We found no alteration in Ab metabolism ( Figure 1); however, proinflammatory cytokines, especially TNF-a, were significantly increased in the CNS and CSF of rats carrying Trem2 R47H variant. Overall, these increases were more significant in Trem2 R47H/R47H as opposed to Trem2 R47H/w rats (Figures 2 and 3), suggesting that early neuroinflammation caused by the Trem2 R47H variant is gene dosage-dependent. In the CNS, TREM2 expression is restricted to microglia (Schmid et al., 2002) and activated microglia secrete pro-inflammatory cytokines: thus, increased secretion of pro-inflammatory cytokines by Trem2 R47H -microglia may cause early neuroinflammation. Also, microglia may promote cytokine production by other CNS cell types -such as astrocytes, for example-and/or that excess cytokines present in the brain of Trem2 R47H rats may be produced by non-CNS resident cells. It is also possible that cytokine clearance is altered in Trem2 R47H rats. These possibilities do not need to be mutually exclusive.
The frequency of mEPSC is significantly increased in Trem2 R47H/R47H rats as well ( Figure 5D and G) independently of changes in Pr of glutamatergic synaptic vesicles ( Figure 5I). However shortterm treatment with anti-TNF-a did not completely occlude this increase ( Figure 6D and G). Thus, either the increased mEPSC frequency in Trem2 R47H/R47H rats is TNF-a independent or TNF-a underlies this alteration via long-lasting mechanisms that are not fully occluded by short-term treatments with anti-TNF-a. The evidence that recombinant TNF-a enhances mEPSC frequency in primary neurons (Beattie et al., 2002;Grassi et al., 1994) and that anti-TNF-a reduces the statistical significance of the differences between Trem2 R47H/R47H and Trem2 w/w recordings (w/w + anti-TNF-a vs. RH/RH + anti-TNF-a, p=0.0046**; w/w + anti-TNF-a vs. RH/RH + Isotype, p<0.0001****; w/w + Isotype vs RH/RH + anti-TNF-a. p=0.0109*; w/w + Isotype vs RH/RH + Isotype, p<0.0001****) would support the latter possibility. Increase in synaptic density, a phenomenon that in Trem2 -/mice is due to impaired synapse elimination by microglia (Filipello et al., 2018), can also enhance frequency of mEPSC. Further studies will be needed to determine whether the Trem2 R47H variant operates with a similar mechanism and to determine the role(s) of other pro-inflammatory cytokines whose brain concentration is increased by the Trem2 R47H variant.
Supraphysiological TNF-a concentrations also cause LTP suppression at hippocampal SC-CA3 >CA1 synapses of young Trem2 R47H/R47H rats. Given that LTP is the most attractive cellular model for learning and memory and is considered an electrophysiological surrogate of memory (Nicoll, 2017), it would not be surprising if subsequent studies will find that rats carrying the Trem2 R47H variant develop learning and memory deficits. If this were the case, this model organism may be used to test therapeutic approaches, including reduction of TNF-a activity and glutamatergic transmission, targeting early pathogenic mechanisms leading to dementia.
In conclusion, we present evidence that the Trem2 R47H KI rat represents a useful model to dissect early events leading to sporadic forms of dementia. Our study indicates that the microglia-specific pathogenic Trem2 R47H variant boosts glutamatergic neuronal transmission and suppresses LTP by enhancing brain TNF-a concentrations, directly linking microglial to neuronal dysfunctions. More studies will be needed to determine whether this microglia-neuronal axis represents an early, Abindependent pathway that facilitates dementia pathogenesis in humans. Rat brain preparation for RNA extraction 6-8 week old rats were perfused with PBS (N = 5 for each genotype (RH/RH, RH/w and w/w) and each gender (N = 30 total) and froze immediately after dissection and saved in À80 for further preparation.

RNA extraction and quantitative RT-PCR
Total brain RNA or microglia RNA was extracted with RNeasy RNA Isolation kit (Qiagen 74106) and used to generate cDNA with a High-Capacity cDNA Reverse Transcription Kit (Thermo 4368814) with random hexamer priming. For total brain RNA, frozen hemispheres were homogenized in buffer (250 mM sucrose, 20 mM Tris-base pH 7.4, 1 mM EDTA, 1 mM EGTA) and an aliquot (~40 mg) was used as input for RNA extraction. Real time polymerase chain reaction was carried out with TaqMan Fast Advanced Master Mix (Thermo 4444556), and the appropriate TaqMan (Thermo) probes.
Brain slice preparation and CSF collection 6-8 week old rats were deeply anesthetized with isoflurane, and CSF was immediately collected from the cisterna magna using customized glass pipettes connected with syringes. The rats were then intracardially perfused with an ice-cold cutting solution containing (in mM) 120 Choline Chloride, 2.6 KCl, 26 NaH CO3, 1.25 NaH2PO4, 0.5 CaCl2, 7 MgCl2, 1.3 Ascorbic Acid, 15 Glucose, prebubbled with 95% O2/5% CO2 for 15 min. The brains were rapidly removed from the skull. Coronal brain slices containing the hippocampal formation (400 mm thick) were prepared in the ice-cold cutting solution bubbled with 95% O2/5% CO2 using Vibratome VT1200S (Leica Microsystems, Germany) and then incubated in an interface chamber in ACSF containing (in mM): 126 NaCl, 3 KCl, 1.2 NaH2PO4; 1.3 MgCl2, 2.4 CaCl2, 26 NaHCO3, and 10 glucose (at pH 7.3), bubbled with 95% O2% and 5% CO2 at 30˚C for 1 hr and then kept at room temperature. The hemi-slices were transferred to a recording chamber perfused with ACSF at a flow rate of~2 ml/min using a peristaltic pump.

Electrophysiological recording
Whole-cell recordings in the voltage-clamp mode (À70 mv) were made with patch pipettes containing (in mM): 132.5 Cs-gluconate, 17.5 CsCl, 2 MgCl2, 0.5 EGTA, 10 HEPES, 4 ATP and 5 QX-314, with pH adjusted to 7.3 by CsOH. Patch pipettes (resistance, 8-10 MW) were pulled from 1.5 mm thin-walled borosilicate glass (Sutter Instruments, Novato, CA) on a horizontal puller (model P-97; Sutter Instruments, Novato, CA). Basal synaptic responses were evoked at 0.05 Hz by electrical stimulation of the hippocampal SC-CA3 >CA1 afferents using concentric bipolar electrodes. CA1 neurons were viewed under upright microscopy (FN-1, Nikon Instruments, Melville, NY) and recorded with Axopatch-700B amplifier (Molecular Devices, San Jose, CA). Data were low-pass filtered at 2 kHz and acquired at 5-10 kHz. The series resistance (Rs) was consistently monitored during recording in case of reseal of ruptured membrane. Cells with Rs >20 MW or with Rs deviated by >20% from initial values were excluded from analysis. EPSCs were recorded in ACSF containing 15 mM bicuculline methiodide to block GABA-A receptors. The stimulation intensity was adjusted to evoke EPSCs that were 40% of the maximal evoked amplitudes ('test intensity'). 5-10 min after membrane rupture, EPSCs were recorded for 7 min at a test stimulation intensity that produced currents of~40% maximum. For recording of paired-pulse ratio (PPR), paired-pulse stimuli with 50 ms or 200 ms inter-pulse interval were given. The PPR was calculated as the ratio of the second EPSC amplitude to the first. For recording of AMPA/NMDA ratio, the membrane potential was firstly held at-70 mV to record only AMPAR current for 20 sweeps with 20 s intervals. Then the membrane potential was turned to +40 mV to record NMDAR current for 20 sweeps with perfusion of 5 mM NBQX to block AMPAR. Mini EPSCs were recorded by maintaining neurons at À70 mV with ACSF containing 1 mM TTX and 15 mM bicuculline methiodide to block action potentials and GABA-A receptors respectively. mEPSCs were recorded for 5-10 mins for analysis. Data were collected and analyzed using the Axopatch 700B amplifiers and pCLAMP10 software (Molecular Devices) and mEPSCs are analyzed using mini Analysis Program.
Field potential recordings in the current-clamp mode were made with pipettes (resistance, 3-4 MW) containing 3 M NaCl. Basal synaptic responses were evoked at 0.05 Hz by electrical stimulation of the Schaffer collateral afferents using concentric bipolar electrodes. fEPSPs were recorded from CA3-CA1 synapse area near Schaffer collateral afferents. For the input-output curve, the stimulus intensity was raised from 0.4 mA in steps of 0.2 mA. LTP was induced after 10-20 mins of baseline recording with 5 trains of 1 s 100 Hz stimulation with 5 s intervals. Following assessment of basal synaptic transmission (BST) by plotting the stimulus voltages against slopes of field Excitatory Post-Synaptic Potentials (fEPSP), baseline was recorded every minute at an intensity that evoked a response 35% of the maximum evoked response. Responses were measured as fEPSP slopes expressed as percentage of baseline.

Antibodies treatment
10 ng/ml Goat anti-TNF-a (AF-510-NA, R and D Systems) or Goat IgG control (AB-108-C, R and D Systems) was incubated right after slice cutting and perfused throughout recording. Experiments were performed at 28.0 ± 0.1˚C.

Statistical analysis
Data were analyzed using GraphPad Prism software and expressed as mean ± s.e.m. Statistical tests used to evaluate significance are shown in Figure legends. Significant differences were accepted at p<0.05.