Protein expression profiling of inflammatory mediators in human temporal lobe epilepsy reveals co-activation of multiple chemokines and cytokines

Neocortical (medial temporal gyrus) (CX) and hippocampal (HC) tissue samples of pharmaco-resistant mTLE patients were obtained after surgery at the University Medical Centre Utrecht. Patients were selected for surgery according to the criteria of the Dutch Epilepsy Surgery Program [31]. The excision was based on clinical evaluations, interictal and ictal electroencephalography (EEG) studies, magnetic resonance imaging (MRI) and intraoperative electrocorticography (iEcOG). Informed consent was obtained from all patients and all procedures were approved by the Institutional Review Board. Immediately after resection, the tissue was cooled in physiological saline (4°C) and cut on a precooled plate. The neocortex sample was cut in half perpendicular to the surface of the gyrus, the hippocampus into three slices perpendicular to its longitudinal axis. One half of the neocortex sample and the two outer parts of the hippocampus were used for clinical pathological analysis. The remaining neocortical and hippocampal tissue samples were once more divided into a part that was immediately frozen on powdered dry ice and a part that was immersion-fixed in 4% paraformaldehyde in 0.1 M phosphate buffer for 24 h at 4°C. Following fixation, tissue was embedded in paraffin and stored at 4°C. Frozen samples were stored at −80°C. Frozen and paraffin-embedded control brain tissue samples were obtained from non-epileptic post mortem cases without hippocampal aberrations from the Netherlands Brain Bank (NBB no. 618, http://​www.​brainbank.​nl). All control material was collected from donors with written informed consent for brain autopsy and the use of the material and clinical information for research purposes. Only patient samples obtained after a relatively short post mortem delay (range 4 to 20.5 h; mean 6.8 h) and with a pH value close to 6.5 (range 6.36 to 6.88; mean 6.68) were used. Detailed histological examination of the brain material from all patients used in this study showed that all samples were devoid of tumor tissue. Neocortical samples that displayed cortical dysplasia or any other abnormalities that might trigger an immunological response were excluded from the study [32, 33]. The quality of all tissue samples was further analyzed in a parallel study [30], showing that after RNA isolation all samples had RNA integrity values (RIN) >6 (range 6.4 to 8.4; mean 7.2) and RIN values did not significantly differ between autopsy control and mTLE patient samples. Table 1 provides a summary of the clinical data of all patients included in the study. Cortical and hippocampal specimens were divided into three groups: a non-epileptic autopsy control group (control, n = 10), a group of mTLE patients without signs of hippocampal sclerosis (mTLE-HS, n = 10) and an mTLE group with hippocampal sclerosis (mTLE + HS, n = 10). The severity of HS was determined by a neuropathologist using the Wyler classification method [34] defining W0 as hippocampal tissue without HS and W4 as tissue with the most severe type of HS. Wyler classification was independently verified on sections of paraffin-embedded tissue.

Nissl-stained cryosections were generated to ensure that neocortical samples contained 50% gray and 50% white matter, and that in hippocampal samples all anatomical subregions were equally represented. Subsequently 25 μm cryosections were cut until approximately 20 mg of tissue was collected. This material was stored at −80°C until all samples were collected. The frozen slices were homogenized at 4°C in 400 μl lysis buffer (Lysis M, Roche, Basel, Switzerland), sonicated, centrifuged and passed through a filtering column (0.22 μm, Spin-X, Costar, Sigma-Aldrich, St Louis, MO, USA), frozen and stored (final concentration 0.5ug/ul) at −80°C [36].

Concentrations of 39 immune modulators (listed in Table 2) were measured using a multiplex bead-based immunoassay (MIA) as previously described. [37]. Capture and detection antibody pairs and recombinant proteins used for the standard curves were purchased from commercial sources as described previously [38] (and Table 3). Carboxylated polystyrene microspheres were purchased from Bio-Rad Laboratories (Hercules, CA, USA). Covalent coupling of the capture antibodies to the microspheres was performed as previously described [37, 38]. Optimal working conditions and sample dilutions were determined for each antibody and calibration curves from recombinant protein standards were prepared using two-fold dilution steps in Lysis M buffer used for homogenization [38]. All assays were carried out directly in a 96 well 1.2 μm filter plate (Millipore, Billerica, MA, USA) at room temperature and protected from light. A mixture containing 1000 microspheres per antigen (total volume 10 μl/well) was incubated together with a standard, homogenate or lysis buffer for 1 h at room temperature. Next, 10 μl of a cocktail of biotinylated antibodies (16.5 μg/ml each) was added to each well and incubated for an additional 60 min Beads were then washed with phosphate buffered saline (PBS) supplemented with 1% bovine serum albumin (BSA) and 0.5% Tween 20 at pH of 7.4. After incubation for 10 min with 50 ng/well streptavidin R-phycoerythrin (BD Biosciences, San Diego CA, USA) and washing twice with PBS-1% BSA-0.5%-Tween 20 pH 7.4. Fluorescence intensity of the beads was measured in a final volume of 100 μl HPE buffer (Sanquin Reagents, Amsterdam, The Netherlands) and buffer values were subtracted from all readings. Measurements and data analysis were performed using the Bio-Plex system in combination with the Bio-Plex Manager software version 4.1 (Bio-Rad Laboratories). CCL4 levels were measured separately using a fluorokine kit (R&D Systems, Abingdon, UK) according to the manufacturer’s instructions.

Hippocampal immunohistochemistry (IHC) was performed on 7 μm paraffin sections of the same patients (n = 6 per group) as used in the MIA (CCL4 polyclonal goat, 1:200, Santa Cruz Biotechnology, Santa Cruz, CA, USA; IL-25 monoclonal mouse, 1:800, R&D systems, Abingdon, UK). We tested several antibodies against immune modulators with pattern 1 and 2 (Figure 1). Based on their superior quality for immunocytochemistry on paraffin section we choose IL-25 and CCL4. Serial antibody dilution curves were prepared in phosphate buffered saline (PBS) containing 0.2%Triton-X100 to determine optimal working concentrations. After rehydration, all sections were subjected to antigen retrieval using microwave treatment (in 0.01 M, Na⁺ Citrate pH 6.0), non-specific binding was blocked by incubating sections with 0.3% H₂0₂ for 30 min at room temperature and with fetal calf serum (IL-25) or normal rabbit serum (CCL4) for 30 min at 37°C. Sections were incubated with primary antibodies overnight at 4°C. The secondary antibodies used were biotinylated rabbit anti-goat and biotinylated horse anti-mouse (Dako Cytomation, Glostrup, Denmark). Signals were visualized using the avidin-biotin method (Vectastain ABC Elite kit; Vector Laboratories, Burlingame, CA, USA) with 3,3′-diaminobenzidinetetrachloride (DAB) as the chromogen (Sigma Chemical Co., St. Louis, MO, USA). By visual inspection, sections were ranked according to overall IR levels by two independent observers blinded to the experimental design. No immunostaining was observed when the protocol was completed with primary antibodies preabsorbed with their respective antigens, or without the primary antibodies.

Double immunofluorescence labeling was performed as described for IHC, but without the 0.3% H₂0₂ blocking step. Anti-GFAP (polyclonal rabbit, 1:6000 Dako Cytomation, Glostrup, Denmark), Iba1 (polyclonal rabbit, 1:200, WAKO Pure Chemical Industries, Neuss, Germany), Vimentin (monoclonal mouse,1:2400, Dako Cytomation, Glostrup, Denmark) and NeuN (polyclonal rabbit, 1:800, Millipore, Billerica, MA, USA) were used as CNS cell-type markers. Secondary antibodies used were donkey anti-goat Alexa 488 and goat anti-mouse Alexa 488 (Invitrogen, Molecular Probes, Eugene, OR, USA) for CCL4 and IL-25, donkey anti-mouse Alexa 555 and donkey anti-rabbit Alexa 555 (Invitrogen, Molecular Probes, Eugene, OR, USA) for the cell-type markers. Finally, Sudan Black dye was used to reduce autofluorescence [39]. Controls without primary antibody were devoid of fluorescent staining.

To determine overall statistical significance of differences in protein expression between patient groups, data from all detectable inflammatory mediators were analyzed using a one-way ANOVA (if normally distributed) or Kruskal-Wallis test (if not normally distributed). When significant group differences were present (Bonferroni corrected for multiple testing: P = <0.0015), additional post hoc tests (Table 4) were performed (Students t-test or Mann–Whitney U with Bonferroni correction: P = <0.008).

In the autopsy control data set Pearson’s correlation tests did not reveal significant correlations between protein expression and brain pH or post mortem delay (PMD), thus ruling out pH and PMD as confounders (CC < 0.7 P = > 0.0015). Also, using RNA integrity numbers (RIN) measured for a different study of these control samples, no correlation was found between inflammatory mediator levels and RIN values [30]. To assess correlations between expression levels of inflammatory mediators (all proteins with P = <0.05) and known clinical parameters (Engel score, seizure frequency, iEcOG spikes), we performed correlation analyses using a Pearson’s correlation (if normally distributed) or a Spearman’s rank test (if not normally distributed).The cut-off for a correlation was set at 0.8 with a Bonferroni corrected P value P = <0.0017. All positive correlations between inflammatory mediators were plotted to form networks using Cytoscape.

Finally, all data from the detectable inflammatory mediators were clustered into components (factors) using an unbiased principle component analysis (PCA) (see Marengo et al. 2010 [40]) (in SPSS also known as factor analysis). As bivariate analyses showed that expression data for multiple proteins were correlated, which is in line with functional relations between cytokines, we chose the oblimin rotation, which allows correlation between factors, rather than the often used varimax rotation, which does not allow correlation between factors. The number of extracted components was based on examination of the scree plot and eigenvalues (>1.0). Interpretation of the components was based on mediators with factor loadings above 0.4 (16% explained variance). When there were missing values (in one patient, for two cytokines) they were replaced with the group mean. Rotations for all data converged within 50 iterations. All inflammatory mediators were used to calculate individual components scores for each participant (using regression to determine the component or factor scores). These were saved and subsequently used to analyze group differences and construct PCA plots.

Programs used for the statistical analyses and graphical representations were SPSS (SPSS 15 for Windows), Graphpad Prism 4 and Cytoscape 2.8.

To generate protein expression profiles of inflammatory mediators in individual mTLE patients, we analyzed 40 proteins in each brain sample using a multiplex bead-based immunoassay. To assess mTLE-associated differences in expression profiles in hippocampus (HC) and neocortex (CX), we compared three patient groups, mTLE patients without HS (mTLE-HS), mTLE patients with HS (mTLE + HS), and autopsy controls (for patient details see Table 1). This experimental setup has been previously used by us and others [41‐44]. In samples of mTLE patients, we could reliably detect 35 proteins, whereas due to the lower expression levels in autopsy control tissue we could reliably detect 32 proteins. Table 2 summarizes the data set (median and concentration range) of all inflammatory mediators measured in all HC and CX samples in the three patient groups. Statistical analysis revealed that 21 inflammatory mediators showed differential expression across patient groups (Bonferroni-corrected P value <0.0015); mediators are ranked according to level of significance. A univariate general linear approach and Pearson’s correlation tests revealed no significant covariate influence or correlations for gender, age, post mortem delay, anti-epileptic drugs (AED) or pH. Post hoc testing showed that all 21 differentially expressed mediators were upregulated in mTLE + HS and/or mTLE-HS tissue compared to autopsy controls. Two inflammatory mediators (MIF, P = 0.0024 and Cathepsin S, P = 0.0025) showed a tendency to upregulation in autopsy controls compared to both mTLE groups (Table 2 and Figure 2D), but this upregulation did not pass the strict Bonferroni correction.

To further analyze the upregulation of individual inflammatory mediators between patient groups, we generated scatter plots per mediator showing the protein concentration in the HC and CX of each individual patient (Figure 1 and Figure 2). These plots clearly showed that upregulation of mediators in HC and CX tissue can be divided into two patterns (classified as pattern I and II). For instance, IL-10 and IL-25 showed upregulation in mTLE + HS tissue compared to mTLE-HS and autopsy controls (pattern I, Figure 1A and 1B), whereas CCL4 and IL-7 showed upregulation in both mTLE groups compared to autopsy controls (pattern II, Figure 1C and 1D). Interestingly, IL-1RA was only upregulated in the CX of mTLE + HS patients (Figure 1E).

Upregulation of some mediators, which we classified type A, was only found in the hippocampus (Figure 1A and 1C), whereas others (classified as type B) showed upregulation in both hippocampal and neocortical tissue (Figure 1B and 1D). A summary of the response types of all significantly upregulated mediators is provided in Figure 1G.

These results show that in both mTLE-HS and mTLE + HS patients multiple components of the immune response are upregulated. Upregulation can be found in the HC and the CX, and inflammatory mediators show distinct patterns of upregulation.

Our next step was to independently verify upregulation and to analyze the distribution and cell types expressing one inflammatory mediator in each of the two major expression patterns. We performed immunohistochemistry (IHC) for IL-25 (pattern I) and CCL4 (pattern II) on hippocampal sections of a subset (n = 6) of patients from each of the three groups also used in the MIA. Visual inspection of overall immunoreactivity (IR) for IL-25 revealed increased expression in all sections from mTLE + HS patients compared to mTLE-HS and autopsy control patients, thus confirming our MIA data (Figure 1B).

IL-25 IR was found in all three patient groups in the principle neuronal layers of the hippocampus, the pyramidal layer of the cornu ammonis (CA) (subfields 1 to 4) and the granule cell layer in the dentate gyrus (DG). In mTLE patients the increased IL-25 IR was most apparent in the CA1 (Figure 3A) and the DG-CA4 region (Figure 3B) of the hippocampus. This increase in IL-25 IR was due to labeling of cells with a glial morphology (see also Figure 4A and 4B). As expected, the number of IL-25 IR neurons was decreased in mTLE patients compared to mTLE-HS and controls as a result of neuron loss in the sclerotic hippocampus [5].

Hippocampal CCL4 IR was higher in all sections of both mTLE groups compared to autopsy controls (Figure 3),thus confirming our MIA data (Figure 1C). CCL4 expression in both mTLE groups was most pronounced in the CA1 and the hilar region of the hippocampus (Figure 3).

To identify the cell types expressing IL-25 and CCL4 IR, we performed double-label immunofluorescence on mTLE + HS patients (Figure 4 and Figure 5). IL-25 IR co-localized with Iba1-positive microglial cells, GFAP-positive astrocytes and NeuN-positive neuronal cells (Figure 4). CCL4 was predominantly detected in Vimentin-positive reactive (Figure 5A) and GFAP-positive astrocytes (Figure 5B). Interestingly, both IL-25 and CCL4 IR were also detected in GFAP-positive astrocytic endfeet surrounding blood vessels (Figure 4B and 5A, small insets). To exclude T cells as prominent source of cytokines we performed IHC for these cells (data not shown). In line with results published by others [12, 45], we only detected a small number of T cells in the hippocampus of mTLE patients. Together, our immunocytochemistry data show distinct hippocampal staining patterns for IL-25 and CCL4, particularly in the dentate gyrus, where IL-25 but not CCL4 expression is prominent in the principle neurons in the granule cell layer. In mTLE + HS patients the increased IL-25 and CCL4 expression (compared to controls) appears to involve glia cells, whereas increased IL-25 expression is also found in principle neurons.

To investigate relationships between various upregulated mediators, we conducted unbiased correlation network and principle component analyses on data from all patients, irrespective of their pathology. First, correlation analyses using Pearson’s correlation showed that many cytokines were co-regulated in patterns that were remarkably similar in the three patient groups and segregated in two main networks (Figure 6). In most cases, patients with high expression of proteins in the first network had lower levels of proteins in the second network, and vice versa (data not shown).

Principle component analyses (PCA) on all inflammatory mediators above the detection limit (see Table 2), confirmed the correlation analyses and revealed clustering of the proteins in eight principle components. These eight components explained 88.5% (HC) and 84% (CX) of the variance detected in all samples (Table 5A and B). The major components, which explained 63% (HC) and 50.1% (CX) of the variance, are depicted in Figure 7 with their corresponding plots. The first and the third HC components contained all three types of inflammatory mediators we investigated (chemokines, interleukins, growth factors and adhesion molecules), while the second HC component predominantly consisted of chemokines (Figure 7A). Follow-up analysis on the individual factor scores obtained from the PCA, revealed that HC components 1 and 2 differed as whole clusters between both mTLE groups and controls, whereas the HC component 3 differed between mTLE + HS versus mTLE-HS and controls (Table 5A). The PCA plots (Figure 7A, bottom panel) clearly showed that patient groups can be discriminated based on the expression patterns of inflammatory mediators in components 1 and 3. Interestingly, patient 32 of the mTLE + HS group did not cluster with the rest of the mTLE + HS group (Figure 7A). Neuropathological reassessment revealed only mild sclerosis (W2 score) for this patient (Table 1). Component 1 of the CX tissue contained sixteen inflammatory mediators, of which nine overlapped with the HC first component. Follow-up analyses of the components scores obtained from the PCA of the neocortical expression data (Figure 7B) also showed that the three patient groups could be discriminated, yet to a lesser degree, as only CX component 1 differed significantly between patient groups (Table 5B). Interestingly, two mTLE-HS patients clustered more with mTLE + HS patients (patients 20 and 28). We could not find any clinical parameters to explain this aberrant clustering (Table 1).

Subsequently, we investigated possible correlations between the top three principle components of the hippocampal PCA analysis and several clinical parameters. Pearson’s correlation analyses revealed a significant correlation between HC component 3 and seizure frequency (including and excluding clustered seizures, see Table 1) in mTLE-HS patients. No correlation was found with the occurrence of spikes on the iEcOG. Interestingly, the expression levels of the two inflammatory mediators with the strongest loadings in HC component 3; CCL3 and IL-10, showed a correlation with seizure frequency (correlation coefficient of >0.7 with a P value below 0.05), however this correlation did not pass the strict Bonferroni correction for multiple testing.

In all, the data obtained from the PCA analyses showed that patient groups can be discriminated based on their expression profiles. The clustering in major components indicates that upregulation of inflammatory mediators in mTLE may involve multiple immunological pathways.

We measured 40 inflammatory proteins in the hippocampus and neocortex of autopsy control and mTLE patients. In mTLE tissue, 35 of these proteins were expressed above the detection limit. Sixty percent of these detectable mediators were significantly upregulated in mTLE tissue compared to autopsy controls, the remaining 40% showed no significant difference between the three patient groups (Table 2). We could distinguish two main patterns of upregulation in mTLE patients. Inflammatory mediators showing the first pattern (for example, IL-10 and IL-25) were upregulated in mTLE + HS patients (Figure 1A, 1B and 1G) compared to mTLE-HS and autopsy controls. Mediators showing the second pattern (for example, CCL4 and IL-7) were upregulated in both mTLE patient groups (+ and –HS) compared to controls (Figure 1C, 1D and 1G). Significantly more proteins (71.4%) showed the second pattern (upregulation in hippocampus and cortex) than the first (23.8%). Only in 19% of the patients upregulation of inflammatory mediators was confined to the hippocampus (Figure 1G, type A pattern). Thus, our results show that activation of inflammatory mediators is more widespread, and not restricted to the hippocampus, which often is the primary source of epileptic activity. Interestingly, IL1ra was upregulated only in the neocortex, but not in the hippocampus. The significance of the upregulation of this endogenous anti-inflammatory protein [46] remains to be determined. Upregulation may be a response of the cortex to seizure activity. Alternatively, the lack of IL1ra upregulation in the hippocampus may contribute to the epileptic properties of the hippocampus.

The upregulation of a specific subset of inflammatory mediators in the hippocampus of mTLE + HS patients is most likely associated with hippocampal sclerosis. These data indicate that the increased expression in mTLE patients is not just caused by post mortem changes in the autopsy control tissue. This latter point is further substantiated by the fact that covariate analysis in the controls revealed no effects of post mortem delay on any of the parameters measured. The human tissue samples used in our study are from patients suffering from repetitive seizures over a prolonged period. Therefore, part of the effects observed may be seizure-induced rather than specific for the mTLE pathogenesis. Moreover, we cannot rule out effects of antiepileptic drugs when comparing autopsy control patients with mTLE patients, although no correlations with any AED were found in our study. Further studies, particularly in animal models for TLE will be required to analyze the role of individual or groups of inflammatory mediators in epileptogenesis. Animal studies will also be essential to establish whether upregulation of specific inflammatory mediators have epileptogenic, or rather anti-epileptogenic, properties.

Our data show upregulation of protein levels of several inflammatory mediators, which previously have been found upregulated in human mTLE tissue. These include IL1β, CCL2, CCL3, CCL4, VCAM1, ICAM1, and VEGF [10, 11, 18, 20, 47, 48], which were found to have pro-epileptogenic properties in animal models for TLE (for references: see Table 6). Particularly, IL-1β has frequently been associated with pro-epileptogenic properties, a protein which affects neuronal Ca²⁺ influx through NMDA-dependent signaling. As mentioned above, the endogenous antagonist of IL-1β, IL-1ra, has been attributed with seizure-inhibiting properties (Table 6).

Two proteins identified in this study have been described in human genetic studies before: single nucleotide polymorphisms (SNPs) present in IL-1α and IL-10 genes are associated with TLE and febrile seizures respectively (Table 6).

Our data showing upregulation of these pro-epileptogenic proteins in mTLE confirm that these genes are regulated, and that upregulation of transcripts translates into increased production of protein in mTLE tissue.

In this study, we identified a series of new inflammatory mediators that show increased protein levels in human mTLE tissue. These include CCL5, CCL19, CCL22, CXCL9, IL-5, IL-7, IL-13, IL-22, IL-25, IL-27, IFNα and HGF. As the majority of these new mTLE-associated inflammatory mediators have not been studied in animal models of TLE, information on pro- or anti-epileptogenic properties is limited. However, most of these proteins have previously been studied in relationship to other CNS pathologies. Processes such as neuroprotection, excitotoxicity through Ca²⁺ influx, glutamate/GABA signaling and disruption of the BBB have been studied. These processes potentially can influence epileptogenesis [94‐97]. Based on these studies, we categorized the newly identified inflammatory mediators in mTLE as potentially pro-epileptogenic or anti-epileptogenic (Table 6). Potentially pro-epileptogenic are all chemokines, the interleukins IL-5, 7 and 22, and the adhesion molecules VCAM1 and ICAM1. Potentially anti-epileptogenic are IL-1ra, IL-10, 13 and 25, IFNα and HGF. Studies on the potential role of IL-5, IL-7 and VEGF in processes related to epileptogenesis provide inconsistent information. [47, 98].

IL-6 or TNFα protein levels did not differ between the three patient groups. Even though both cytokines have been implicated in epilepsy before [8, 9, 15, 99], as we did not repeat the experiment with multiple antibodies, we cannot rule out technical issues. Our results however, are in line with data from mRNA profiling studies on human TLE tissue [10, 11, 14, 15, 100] as in these studies no significant differences in mRNA or protein were detected in brain tissue. Increased levels of IL-6, but not of TNFα, were found in plasma or cerebrospinal fluid (CSF) of TLE patients [101, 102]. Therefore, increased IL-6 levels might not arise from brain tissue [103‐105] but more likely from peripheral blood mononuclear cells (PMBCs). Indeed, upon stimulation, PMBCs from epileptic patients released increased amounts of IL-6 compared to controls [106].

Thus, our total set of upregulated inflammatory mediators comprises (potentially) pro- as well as anti-epileptogenic proteins in both mTLE patient groups.

Interestingly, we detected both CCL4 and IL-25 in glial cells surrounding blood vessels (Figures 4 and 5). Here, these proteins probably exert opposing influences on epileptogenesis, as CCL4 is thought to disrupt the blood–brain barrier (BBB) and attract immune cells across the BBB while IL-25 has been proposed to have a protective effect on the BBB (Table 6).

To study co-regulation of inflammatory mediators, we first performed bivariate correlation network analysis on the datasets within each patient group. This analysis revealed co-regulation of inflammatory mediators in one large and one small network in each patient group (Figure 6). We observed that in most cases, patients with high levels of the smaller network components (for example, CCL2, CCL4, IL-8) had lower expression levels of larger network components (for example, IL-13, IL-5 CCL9) and vice versa. Subsequent PCAs on all investigated brain samples confirmed these networks and revealed eight components of clustered inflammatory proteins in both the HC and the CX separately (Figure 7, Table 5A and B). Each component typically contained both putative pro- and anti-epileptogenic proteins, suggesting that these proteins are co-regulated in all patient groups. We detected the majority of mTLE-specific differences in the HC of mTLE patients. Further analysis of the HC components shows that individual patients segregate into the three patient groups (Figure 7A), indicating that each patient group has a unique immune profile. The first two components, which explained the greatest variance of protein expression in the HC, contained only proteins of a type II pattern (that is, upregulated in both mTLE patient groups), whereas the majority of component 3 consists of proteins with a type I expression pattern, and this is clearly shown in the PCA plots (Figure 7A). As expected, the majority of HC component 1 consisted of proteins also detected in the large network detected in the bivariate correlation network analysis. HC component 2 comprises chemokines represented in the smaller correlation network (Figure 6).

The PCA analysis on the data set from the CX tissue also showed segregation of patients into patient groups (Figure 7B), although less pronounced, possibly due to the smaller number of regulated proteins. As expected, there was substantial overlap between the major clusters found in the HC and CX, as nine of the ten inflammatory mediators clustered in the HC component 1 were also present in the CX component 1 (Figure 7, Table 5).

The data obtained from the PCAs suggest that the activated immune system in mTLE is complex and may comprise different networks and pathways. The biological function and pathological outcome of activation of the immunological networks needs to be further investigated. Surprisingly, pro- and anti-epileptogenic inflammatory mediators cluster together, suggesting co-regulation of these two types of responses. Correlation analysis of upregulated whole networks with clinical parameters revealed a significant correlation between the HC component 3 and seizure frequency in the hippocampus of mTLE-HS patients. Indeed, the top two proteins in this component (CCL3 and IL-10) showed a weak correlation with seizure frequency in the hippocampi of mTLE-HS patients only. No correlations were found with iEcOG spikes. Larger patient populations will be required to further elucidate the relevance of upregulation of inflammatory networks or specific inflammatory mediators for patient pathology.

In summary, we observed a widespread upregulation of inflammatory mediators in the mTLE brain. Co-regulation of upregulated mediators indicates that they cluster in complex immunological networks. Within these networks we identified both pro- and anti-inflammatory mediators. Detailed analysis of these complex pathways in animal models will be required to assess their putative role in epileptogenesis. Our data suggests that the success of any future therapeutic strategies to treat mTLE will require a multifactorial approach aimed at blocking detrimental effects of proinflammatory pathways while promoting endogenous anti-inflammatory pathways. Promising targets for treatment would be, for instance, CCR5 receptors, which can be activated by a number of chemokines [54]. CCR5 inhibitors are currently being tested in clinical trials for HIV/AIDS treatment [107, 108]. Treatment with CCR5 antagonists will inhibit the action of pro-epileptogenic chemokines CCL3, 4 and 5, the major chemokines in the second immunological network we identified in the HC. Inhibition of CCL3, 4 and 5 signaling, and in general proinflammatory pathways, may contribute to tipping the balance from a perturbed pro-epileptogenic to a more protective and anti-epileptogenic immune system in mTLE patients.