Primary cultures of cortical neurons were prepared from post-natal mice (P1–P3) as previously described (Gullo et al., 2009). All the procedures concerning animal handling and sacrifice followed the Principles of Laboratory Animal Care (2010/63/UE Directive), and were approved by the University of Milan-Bicocca Ethics Committee and the Italian Ministry of Health (D.Lgs 26/2014). Briefly, the cerebral cortex (excluding the hippocampus) was removed from decapitated mice, cut into 1 mm³ pieces, and digested by trypsin (0.15%) and DNAse (10 μg/ml) at 37°C for 20 min. After enzyme digestion, cells were mechanically dissociated by means of trituration, and plated at densities of 600–900 × 10³ cells/ml on glass coverslips (for immunocytochemistry) or MEA Petri dishes pre-coated with polyethyleneimine 0.1% (wt/vol) and laminin 20 μg/ml (30 μm diameter ITO electrodes spaced 200 μm apart, Multichannels System, Germany). After 3 h incubation, the plating medium was replaced by neurobasal medium (NB) with B27 (Invitrogen, Italy), glutamine 1 mM and basic fibroblast growth factor (bFGF) 10 ng/ml, and the culture was maintained at 37°C in 5% CO₂. One-half of the medium volume was replaced every 3 days. The cultures in MEA dishes were covered with gas-permeable covers (MEA-MEM, Ala Scientific Instruments, Inc., USA) throughout the culture period (12–22 days-in-vitro, DIV).

As previously described (Gullo et al., 2009), we report results obtained within a few hours after positioning the MEA dish into the incubator, which can be considered at the steady-state. The recording area in our MEA dishes was ~2 mm², and in this area, the average number of neurons was in the order of ~5,000, plus about the same number of astrocytes (see Figure 1 of Supplementary Material in Gullo et al., 2010, at http://www/frontiersin.org/neuralcircuits/paper/10.3389/fncir.2010.00011/). The average space between cells was therefore relatively large. Neuroinflammation was induced by incubating cultures with LPS, isolated and purified from E. coli R515 by Enzo Life Sciences (Alexis Biochemicals, code 581-007). This product does not contain detectable protein or DNA contaminants with agonistic Toll-like receptor (TLR) activity. Since it is a strong activator of TLR4 but it does not activate TLR2 or the other TLRs, it was considered suitable for our experiments. The drugs to be tested were kept as frozen stock solutions in distilled water (or DMSO <0.1%) at −20°C, until diluted as appropriate with MEA culture medium before each experiment. The drugs used were: crocin (CR), the polyphenols resveratrol (RE) and curcumin (CU), purchased from Sigma, Italy; exendin-4 (EX-4) and the GLP-1 receptor antagonist, exendin (9–39), purchased from Tocris, UK. All experiments were performed by adding the drugs in volumes that were always <1% of the total conditioned media volume bathing the neurons. When indicated, a washout was carried out with a solution pre-conditioned by the same network under control conditions.

We used the same procedures previously described in Gullo et al. (2009, 2010). Briefly, analog signals sampled at 40 kHz were recorded at 36°C in CO₂-controlled incubators using MEA-1060BC or 1060INV pre-amplifiers (bandwidth 1–8,000 Hz, Multichannel Systems, Germany) connected to a MEA Workstation (bandwidth 100–8,000 Hz, Plexon Inc., USA). Data were sorted into timestamp files by the MEAWorkstation Sorter software (MEAWS, see details below) and cleaned of artifacts using the OFFLine Sorter program (Plexon Inc., USA). Next, during the PCA-based waveform sorting and for multi-unit electrodes, we applied one of the following procedures: (i) spike removal with a Mahalanobis threshold in the range 1.8 to 1.4; we concurrently checked that the P-value of multivariate ANOVA sorting quality statistics was <0.01 among the identified units; (ii) when the previous procedure led to excessive spike invalidation, we manually removed the spikes invading the adjacent unit ellipsoids (the latter method was very effective in decreasing the P-values, with only a limited number of erased spikes).

The method of classification into excitatory or inhibitory neurons is described in detail in Gullo et al. (2009, 2010) and Becchetti et al. (2012). Briefly, for each identified unit and each burst, the following characteristics were computed in defined time segments: the autocorrelation function (ACF), the burst duration (BD), the spike number (SN), the spike rate (SR), the intra-burst spike rate (IBSR), the inter-burst intervals (IBIs), and the Fano factor (FF; Teich, 1989; Baddeley et al., 1997). We classified the neurons on the basis of an unsupervised learning approach consisting of data reducing principal component analysis (PCA) based on FF as a feature (Becchetti et al., 2012), followed by the K-means clustering procedure. The large differences in these burst metrics was the basis for adopting FF as the best feature to clusterize neurons. As previously described (Becchetti et al., 2012), these procedures normally separated two statistically different clusters composed of numbers of excitatory (~60–90) and inhibitory (~15–25) neurons, whose ratio always fitted the ratio present in the neocortex i.e., from 4 to 5 (Gullo et al., 2010; Sahara et al., 2012). To give an approximate idea of the computing time involved in the analysis of a 60-electrode MEA dish with ~100 identified neurons, the following are the typical parameters: 20 h-long experiment, 0.5 GB memory size, 10 min to obtain results.

The global network burst structure was analyzed with standard techniques as well as procedures recently developed by us (Gullo et al., 2012). Briefly, we applied a running window of variable duration (from 10 ms to 1 s) in order to search for the start and end of each burst and simultaneously collecting all of the spikes which were precisely tagged to the engaged excitatory and inhibitory neurons, already designated as previously explained above. In conclusion, for each spike in a generic burst, we knew exactly which neuron fired it and how many other spikes were fired by all of the other neurons. This had the consequence that we could also compute for the two identified neuronal clusters (defined above) the average propensity for firing, namely, the average number of spikes per burst, here called “excitability” (EXC) in each time segment of interest (shown in Figure 1A). Normally, it is much more interesting to compare EXC data normalized to control for both clusters, namely, normEXC. To investigate the heterogeneity of the burst length, BD, we studied its distribution in the form of cumulative probabilities in each time segment of interest, cumBD, as previously done in Gullo et al. (2014). To compare experiments, two types of plots are shown as follows: (i) time plots where normEXC describes how clusters change firing in the presence of drugs with respect to control (open, □, and closed, ■, symbols for excitatory and inhibitory clusters, respectively). Aligned characters, attached to the time axis, are control (c), washout (w), or drug names (short names were: CU, CR, RE, EX-4) indicating the presence or absence of drugs (note axis breaks necessary to zoom to particular time segments); (ii) cumBD plots (bin 0.2 s) vs. burst duration describing how drugs modify the distribution of the burst lifetimes (thick continuous lines for control; open stars for drugs; open rightward triangles, ⊳, for LPS and open upward triangles, Δ, for washout. In our previous paper (Gullo et al., 2014), we found that, at 6–7 h after LPS application, its action was either present or blocked if the values of two variables, namely normEXC and cumBD₉₅ (the value of BD at cumBD of 95%), resulted in an increase with respect to control beside fixed thresholds. The normEXC threshold was +45 ± 10% (with respect to control value, in this case 1) and cumBD threshold was +40 ± 5% (with respect to the BD value in control computed at cumBD₉₅). Therefore, to highlight if drugs were able or unable to block LPS action in the normEXC and cumBD₉₅ plots, in the figures these thresholds were represented by a thin straight line and an arrow respectively.

In each MEA dish, small (150 μl) aliquots of the incubation medium were collected in control and during different times (6 and 12 h, representing the peak-time of LPS-induced TNF-α release, Gullo et al., 2014) after the addition of LPS (3 μg/ml) or polyphenols/crocin/exendin-4 + LPS. Samples (50 μl) were analyzed in triplicate for murine TNF-α with enzyme-linked immunosorbent assay (ELISA) kits (KCM3012, Invitrogen, Italy) according to manufacturer's instructions. The data were expressed as pg/ml following interpolation on the basis of a standard curve. As suggested by the manufacturer, the inter-assay and intra-assay variability were 3.5–4.3 and 2.6–8.2%, with a lower limit of detection <3 pg/ml. During the analysis we did not correct the data for the small change in the total incubation medium volume resulting from the aliquots removal. Data were analyzed using GraphPad 7.0 software employing ANOVA followed by Dunnet's test for group comparison. P < 0.05 was considered statistically significant.

All of the data are expressed as mean values ± SEM, with n indicating the number of experiments. We used OriginPro 9.1 to test normality of data such as cumBD and FSH. Moreover, for statistical significance among cumBD and FSH distributions from different experiments, we used the non-parametric tests Wilcoxon signed rank test or Kruskal–Wallis test both available from OriginPro 9.1. The data were analyzed and the figures prepared using OriginPro 7.0 or OriginPro 9.1 software.

For each tested drug, we performed a dose-response curve using concentrations based on published data, and then we verified if the substances per-se introduced relevant excitability changes with respect to control and washout.

In Figure 1, we show some specific exemplary experiments for CU; similar experiments were done also for the other drugs. As shown in Figure 1A (an exemplary experiment out of 4 similar results), the CU dose-response relationships consistently confirmed that in the low nanomolar region, both excitatory and inhibitory neurons decreased their activity immediately (over a few minute time course). On the contrary, in the region from 30 nM up to 1 μM the CU effect was opposite. Finally, the effects were almost completely reversible on washout. This plot shows the time course of the average spikes/burst elicited by neurons engaged in the activity (“excitability,” EXC, see details in Section Methods) by the two inhibitory (in control ~20) and excitatory (~4.5) neuronal clusters. EXC is characterized by values that are very different between the clusters and can be changed by pharmacological manipulations (see Becchetti et al., 2012). The neuronal clusters responded immediately and differently after each drug application (10 min) starting from 1 nM up to 1 μM (see legend).

To compare different experiments, it was necessary to normalize data to the control data, and in Figures 1B,C these plot-types (now of “normEXC”) are shown superimposed. It can be noticed that from the response in B, the IC₅₀ for inhibitory neurons (closed square) was ~3 nM with a saturation after 10 nM and a recovery during washout. In Figure 1C we show the average of 4 similar experiments and the normEXC dose-response curves were found to have two different shapes in the two concentration regions from 1 to 30 nM and from 0.1 to 1 μM, respectively. In the first region only inhibitory cells were inhibited, and the IC₅₀ was 4.5 ± 0.6 nM; on the contrary, in the second region, inhibitory cells recovered their activity and excitatory cells increased their activity. On average, at 1 μM, the global effect was an increase of activity that recovered on washout. It has previously been shown that an early and fast effect of CU, characterized by an IC₅₀ in the micromolar range, consists of inhibiting glutamate (glu) release from synaptosomes from rat cortex nerve endings (Lin et al., 2011, 2012).

On the whole, these results are only partially in line with those obtained by Lin et al. (2011) by using a different technique. Indeed, since in the cortex microcircuits, inhibitory neurons receive glutamatergic synaptic inputs (in both feedback and feedforward loops, see Isaacson and Scanziani, 2011), we expected to observe a consistent decrease in their activity only in the lower CU concentration region. Compared to those of Lin et al. (2011) (glutamate release from synaptosomes), our methods were characterized by the functional synapse integrity and the astrocyte-mediated [glu]_o uptake (Wanke et al., 2016), suggesting a much higher sensitivity. Moreover, our dose-response curves cannot be considered at steady-state because each drug concentration was changed every 10 min.

Since our experiments with LPS-induced inflammation imply very long time-segments of the order of about 18 h, we investigated if the inhibitory CU effects seen in Figure 1 were indeed transient and therefore eventually negligible over this time-frame. Moreover, we wanted to use CU concentrations lower than those that have been recently used in animals (~100 μg/Kg/day, see Ojha et al., 2012; Hoppe et al., 2013; Zhu et al., 2014) approximatively equivalent to ~100 μM. Indeed, since we did not have any blockade induced by the blood brain barrier (see Wang et al., 2016) and the drug action should be directly on neurons, we decided to fix 1 μM as a reasonable concentration, which is in line with the cited experiments performed by Lin et al. (2011).

Experiments were done by sequentially recording activity in control (3 h), during a preconditioning in CU (about 1 h at 1 μM) and successively in LPS (at least 7 h at 3 μg/ml); a final washout of 5 h was also recorded. The fast effects of a single drug application were studied in detail by averaging data every 15 min. After 3 days, in the same dish we performed a standard experiment by applying only LPS to be sure that the effects of the induced inflammation were still working. The results of such a dual experiment are shown in Figures 2A–D, respectively. In the experiment of Figure 2A that plots normEXC, 1 μM CU produced a fast and transient activity increase which actually decayed in ~45 min. In the inset are shown the original data before normalization: we specifically analyzed the average number of spikes elicited in each burst and in each cluster of neurons; this analysis was performed counting for each neuron the spikes engaged in each burst. In the first 15 min, the drug caused an immediate firing increase of the inhibitory and excitatory clusters from 25.4 ± 1.5 to 38.9 ± 2.2 and from 5.5 ± 0.2 to 8.73 ± 0.4 spikes/burst (the bursts were 82, number of inhibitory and excitatory cells, 21 and 74, respectively) (see other details in legend). The subsequent LPS application, in fact, caused a further decline of normEXC for both clusters of neurons and the trend of normEXC for excitatory and inhibitory clusters was +14 ± 3 and +9 ± 1% respectively, suggesting that LPS was unable to exert its expected excitatory inflammatory function as previously observed (Gullo et al., 2014). Indeed, as shown in Figure 2B, the effects of a standard LPS application (3 μg/ml), done in the same dish 3 DIV later, illustrate that, after 4 and up to 8 h, a strong increase of the activity of excitatory neurons (+86 ± 6%) took place, although we did not observe any effect on inhibitory cells.

The cumBD data associated with the Figures 2A,B experiments are shown in Figures 2C,D, respectively. In C four curves are superimposed as follows: (i) control (thick line), (ii) CU at 15 min (line + asterisk), (iii) CU at 60 min (thin line), (iv) LPS at 7 h (line + right-triangles). The corresponding 90% BD values were: 2.6, 3.8, 2.6, 3.7 s, respectively. These results suggest that, as compared to control, CU only transiently increased cumBD_90%, and LPS over 7 h produced only a 37% increase, confirming that the expected LPS action did not take place. Statistical significance by Kruskal–Wallis test applied to control data and CU at 60 min resulted in a P < 0.12, and at 7 h in LPS data resulted in P < 0.05. In contrast, as shown in Figure 2D, a typical effect of LPS on cumBD (in the same dish) was to further right-shift, with respect to control (line, 95% at 7.2 s), the curves at 5 h (right-triangles) and at 9 h (upward-triangles) to 12 s (+66%) and 17 s (+136%) values, respectively, with “seizure-like” burst durations reaching 22 s. The Kruskal–Wallis test of significance applied to control data and LPS data resulted in P-values which were always smaller than 10⁻⁴.

On the whole, these results demonstrate that 1 μM CU by itself suddenly produced changes of activity in bursting, by promoting long BD values and significant increases in the excitability of both neuronal clusters. These transient modifications rapidly returned to quasi-control values as if the effects were partially compensated, only producing scarce or null effects in the early hours after adding LPS. On all of the occasions (n = 10) in which we performed these types of experiments, the results were similar. In other experiments (n = 5) in which we used smaller CU concentrations (i.e., 0.1 μM), the results were not clear and these data were discarded. Few (n = 4) experiments performed in dishes whose LPS responses were not significant were also discarded.

Crocin and crocetin have been found to block the release of NO and various chemokines under the action of LPS (Nam et al., 2010). As for curcumin, we did fast dose-response experiments initially to test putative effects of crocin (CR) on the activity of our networks (Supplementary Figure 1). In the range 1–100 μM, some inhibitory effects were seen only beyond 20 μM (n = 3). We here show that during the first hour of incubation, experiments using CR at 3 (n = 3) and 20 μM (n = 3) had either negligible or transient inhibitory effects, respectively. These CR effects were accompanied by other different actions caused by the further LPS presence.

In Figure 3 are shown the results from two exemplary experiments at 3 (Figures 3A,C) and 20 (Figures 3B,D) μM CR, respectively. At 3 μM it can be observed in Figure 3A that the early CR-induced activity was barely modified and LPS induced a very slow, but sizeable increase of activity up to ~52% at eighth hours. An analysis to test the cumBD plots, shown in Figure 3C, suggests that a significant increase by 46% of BD was present at 7 h LPS as compared to control.

On the contrary, at 20 μM CR, the effects were found to be largely depressive. After the CR application (see Figure 3B), we detected an early negative peak of 45 and 16%, respectively in inhibitory and excitatory cells, but this effect decayed in the next hour. During the next 7 h in LPS, the activity remained stable but depressed. Furthermore, by analyzing in Figure 3D the cumBD plots, it can be concluded that the curves shifted to the left, suggesting that the burst durations decreased, in line with a weaker activity of the network.

On the whole, it can be concluded that CR is able to block the LPS-induced inflammation only at concentrations higher than 15–20 μM. Interestingly, at these concentrations, although the fast earlier transient inhibitory effects were present, a strong general inhibition of activity remained and was washed out only during recovery.

We studied the action of resveratrol (RE) at concentrations in the lower micromolar range, much lower than those reported in many papers present in the literature (Gambini et al., 2015). As reported above for CU, also RE has been described to inhibit the release of glutamate from nerve terminals (Chang and Wang, 2009). These authors showed that the threshold to detect the start of the inhibition was between 10 and 100 nM, but at 1 μM the effect reached the IC₅₀. When tested in our dishes, 200 nM RE (Figure 4B) exerted immediately a decrease in the neuronal firing, but the network slowly recovered after ~30 min.

Due to this reason, we decided to search for a concentration range in which effects were either unable or able to cause changes in the plots of normEXC and the cumBD in opposite directions. In Figures 4A,B are shown two exemplary experiments in which RE applied at 50 and 200 nM induced a negligible or a fast inhibitory effect, respectively, in the normEXC plots. The average duration of these intrinsic responses of RE were never longer than ~45 min. The subsequent application of LPS was tested every hour over 8 h.

In Figures 4A,B it can be observed that the development of the response was divergent, i.e.: normEXC data either increased in Figure 4A (at 50 nM) or decreased in Figure 4B (at 200 nM), respectively, suggesting that the drug preconditioning produced opposite results during the LPS action. Similar results were also seen in other 4 dishes at 50 nM and 3 dishes at 200 nM. The fluctuations of the activity shown in Figure 4A, during control and RE, were not significant up to the first hour of LPS application, whereas from 2 to 8 h in LPS, the activity increased as expected by 71% for excitatory cells and 127% for inhibitory cells (see in Gullo et al., 2014). On the other hand, in Figure 4B, the normEXC slowly decayed during the following 8 h.

To complete the analysis, we computed the distribution of the BD durations of these experiments and the results are shown in Figures 4C,D, respectively. Although the profile properties of the control curves (thick line) were different in the two experiments, the data acquired during the presence of RE alone show, in both cases, a cumBD curve (see the asterisk symbols) very similar to control. On the contrary, comparing the last hours of LPS (right-pointing triangles) data in Figure 4C suggest that the 90% cumBD showed an 87% increase, and in 4 D a dramatic decrease, thus confirming that the RE doses used in these experiments either did not block or completely blocked the LPS action, respectively.

In hippocampal pyramidal neurons in vivo, it has been shown that a brief (1 s) application of GLP-1 (7–36) amide, a naturally produced active fragment of GLP-1, first rapidly increased (i.e., 1 s) and then slowly decreased (>10 s) single unit firing activity (Oka et al., 1999). Interestingly, these effects were inhibited either by the specific GLP-1 receptor antagonist, exendin (9–39) or by the specific non-NMDA glutamate receptor antagonist 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), suggesting that the initial GLP-1 action involved a fast and transient release of glutamate.

Our in vitro networks of cortical neurons from neonatal mice contain fractions of excitatory and inhibitory cortical neurons, which always fit the standard ratio of 4:1 known to be present in adult cortex (Sahara et al., 2012). As for curcumin and crocin, we did fast dose-response experiments to test putative effects of EX-4 on the activity of our networks (Supplementary Figure 1). Surprisingly, we found that in about ~80% (n = 24) of the experiments done using EX-4, the fast effects consisted, as expected, of a transient, brief pro-excitability response, but there were also some networks (~20%, n = 6) whose responses consisted of a transient decreased excitability. This was probably caused by unknown heterogeneity factors intrinsic to our cultured preparation of dissociated cells, and therefore we excluded these experiments from the results (although they were characterized by late LPS responses where atypical activity was present). EX-4, from 10 up to 500 nM, elicited an early fast excitatory response and prevented the late LPS-induced neuroinflammation. Amongst the EX-4 experiments in which the early effects of the drug were pro-excitable, in Figure 5 are shown the results of two experiments, performed in the same dish (at two different DIVs) with 10 and 500 nM EX-4 and one experiment done at 50 nM. In Figures 5A–C are compared the normEXC data and it is evident that concentrations differing by 5 and 50 times indeed produced fast transient excitatory effects, suggesting a dose-dependent mechanism of action. On the contrary, the long-term normEXC curves at the end of the LPS application did not show any significant change, thus suggesting that activity of the network was not different with respect to control. We analyzed in Figures 5D–F the related cumBD histograms in control, during the early EX-4 action and at the end of the long-term LPS action. In the three cases, the early EX-4 histograms showed a conspicuous rightward shift of the curves. On the contrary, the final effect of LPS was either negligible or without atypical bursts showing a BD increase at 90% in the cumBD curves.

In order to confirm that the EX-4 action shown in Figure 5 is really due to the activation of GLP-1Rs, we performed experiments (n = 6) in which we preconditioned the EX-4 application (50 nM) with the specific GLP-1R antagonist exendin (9–39) at 100 nM. The results of such an exemplary experiment are shown in Figure 6. In Figure 6A, the normEXC plot shows that the early application of exendin (9–39) (e) induced a very short (10 min) and small negative deflection, followed by the application of EX-4 (E), and of LPS, which increased the slow normEXC up to 57 ± 3%. After 7 h, the activity stopped increasing. In Figure 6B the cumBD plots show that the 90% value was 8.1 and 12.2 s, in control and LPS, respectively, thus suggesting that a normal excitatory action of LPS took place as if EX-4 was not present. On the whole, these experiments confirm that the results shown in Figure 5 robustly indicate EX-4 as a potent agent to prevent LPS-induced inflammation.

As reported in our previous paper (Gullo et al., 2014) where we showed that minocycline blocked the microglial-release of TNF-α induced by LPS treatment, we performed, for all the polyphenols and EX-4, appropriate experiments to test if the drug concentrations used in the experiments (Figures 2–5) were able to block also TNF-α production. As shown in Figure 7, the amount of this cytokine was, both at 6 and 12 h after LPS treatment, significantly smaller than that found when the preconditioning drugs were absent. On the whole, these results confirm that the drugs used here were not only able to block the atypical seizures induced by LPS, but also the release of TNF-α in the network dishes.

Microglia are highly ramified cells which rapidly scan the local environment and react to its modification (Davalos et al., 2005). Under physiological conditions, microglia react rapidly to neuronal activity by modulating the contacts that their processes continuously establish with synaptic elements (Wake et al., 2009). Amongst the microglial-released molecules with a putative role in neurotransmission, TNF-α was shown to control basal synaptic functions (Santello et al., 2011) as well as plasticity (Stellwagen and Malenka, 2006; Costello et al., 2011), even if this role was attributed to TNF-α produced by astrocytes. However, the astrocytes have often been thought to release TNF-α because cultures of astrocytes are consistently contaminated by microglia (Saura, 2007). In fact, other data reveal no TNFα-encoding transcript in astrocytes (Foo et al., 2011; Zamanian et al., 2012). Thus, the TNF-α that controls several aspects of synaptic transmission is produced by microglial cells. Similarly, TLR4, the specific LPS receptor, was detected in mixed astrocyte/neuronal cultures, but when microglia were removed from astrocyte cultures, TLR4 expression was no longer present (Lehnardt et al., 2002).

Apart from the obvious concentration-dependence, the effects of polyphenols can be subdivided into various types of action according to: (i) time-dependence (fast or long-term), (ii) cell-dependence (nervous or non-nervous system), and (iii) disease-dependence (i.e., Parkinson's, Alzheimer's diseases, or cancer). CU was found to be neuroprotective in the MPTP model of Parkinson's disease (Ojha et al., 2012) by inhibiting the generation of pro-inflammatory cytokines through the prevention of NF-κB traslocation into the nucleus (Karunaweera et al., 2015). In addition, CU attenuated inflammation in experimental traumatic brain injury (Zhu et al., 2014) and decreased human TNF-α levels (Sahebkara et al., 2016). The nanoencapsulated CU version was also found to reduce β-amyloid-induced cognitive impairment in rats (Hoppe et al., 2013). In the same way as CU, RE was also able to protect dopamine neurons (Zhang et al., 2010) from LPS-induced inflammation. Moreover, RE was found to protect cortical neurons from oxygen-glucose deprivation-induced apoptosis (Gao et al., 2014). In general, RE studies in in vitro and in vivo showed interesting data of metabolism and bioavailability in various animal models and in humans (Smoliga and Blanchard, 2014; Gambini et al., 2015). Interestingly, RE was shown to ameliorate the clinical severity observed in an animal model of multiple sclerosis, by maintaining the integrity of the blood-brain barrier (Wang et al., 2016). In summary, we believe that the principal mechanisms by which polyphenols were able to protect against LPS neuroinflammation in our system are those related to the TLR4 pathway and microglia activation.

There is evidence that the gut hormone GLP-1 is also produced in the brain (Korol et al., 2015; Thiebaud et al., 2016). In particular, it was shown that the GLP-1 receptor agonist exendin-4 (EX-4) protected dopaminergic neurons by inhibiting TNF-α release from microglia (Kim et al., 2009). A reduction of brain TNF-α levels was also described in an animal model of Alzheimer disease (AD) exposed to EX-4 treatment (Solmaz et al., 2015). In LPS-treated rats, GLP-1 protected hippocampal neurons from synaptic impairments (Iwai et al., 2014) and inhibited IL-1β production from cultured astrocytes (Iwai et al., 2006). Furthermore, in vitro administration of EX-4 in rats increased the expression of the astrocytic glutamate transporter GLT-1 in the hippocampus, lowering extracellular glutamate concentration (Kobayashi et al., 2013). This is in line with the lower long-term excitability data shown in Figures 5B,C during LPS inflammation. The results shown in Figures 5, 6 split the presumed fast AMPAR-induced transient firing increase following GLP-1R activation (by EX-4), as characterized by Oka et al. (1999), from the long-term effects related to the switch-off of the LPS-induced inflammation.

Our data show, quite importantly, and for the first time, that these effects are not peculiar to the hippocampus in vivo, but can also be reproduced in vitro, in networks from neocortical dissociated cells containing neurons, astrocytes, and microglia. Moreover, the data shown in Figure 6 demonstrate that the GLP-1R antagonist exendin (9–39) functionally blocked the ability of EX-4 to counteract the LPS sterile inflammation. Thus in summary, we think that the intriguing connection between the fast electrophysiological responses and the slow microglia pathway merits a specialized investigation, difficult to be carried out by means of our procedures.

Only recently, ion channels have also been described to be functionally expressed in microglia (for a minireview, see Madry and Attwell, 2015). Among the various types of identified channels, the voltage-gated proton channel Hv1 (not present in neurons) was thought to have a dramatic importance during damage induced by ischemic stroke when reactive oxygen species (ROS) are generated (Wu et al., 2012). Accordingly, knock-out mice lacking the Hv1 channel were protected from ROS-mediated neuronal death and brain damage. Furthermore, since acute neuroinflammation provokes intracellular acidification (Tyrtyshnaia et al., 2016), it cannot be excluded that voltage-gated proton channels or acid-sensing channels (Yu et al., 2015) could also be involved in this type of process. Indeed, a polyphenol such as epigallocatechin-3-gallate (EGCG) is known to inhibit the Hv1 currents in microglial BV2 cells (Jin et al., 2012). Interestingly, the reduction in TNF-α release produced by some antidepressants was supposed to be related to the same effect (Song et al., 2012). The putative role of this ionic mechanism in the LPS-induced sterile inflammation is also currently under investigation in our laboratory (EW, FG, and ML).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.