High frequency oscillations associate with neuroinflammation in low-grade epilepsy associated tumors

OBJECTIVE
High frequency oscillations (HFOs) in intraoperative electrocorticography (ioECoG) are thought to be generated by hyperexcitable neurons. Inflammation may promote neuronal hyperexcitability. We investigated the relation between HFOs and inflammation in tumor-related epilepsy.


METHODS
We identified HFOs (ripples 80-250 Hz, fast ripples 250-500 Hz) in the preresection ioECoG of 32 patients with low-grade tumors. Localization of recorded HFOs was classified based on magnetic resonance imaging reconstructions: in tumor, in resected non-tumorous area and outside the resected area. We tested if the following inflammatory markers in the tumor or peritumoral tissue were related to HFOs: activated microglia, cluster of differentiation 3 (CD3)-positive T-cells, interleukin 1-beta (IL1β), toll-like receptor 4 (TLR4) and high mobility group box 1 protein (HMGB1).


RESULTS
Tumors that generated ripples were infiltrated by more CD3-positive cells than tumors without ripples. Ripple rate outside the resected area was positively correlated with IL1β/TLR4/HMGB1 pathway activity in peritumoral area. These two areas did not directly overlap.


CONCLUSIONS
Ripple rates may be associated with inflammatory processes.


SIGNIFICANCE
Our findings support that ripple generation and spread might be associated with synchronized fast firing of hyperexcitable neurons due to certain inflammatory processes. This pilot study provides arguments for further investigations in HFOs and inflammation.


Introduction
Epilepsy is thought to arise from uncontrollable neural excitation and epilepsy surgery can be an effective treatment for focal drug-resistant epilepsy (Ryvlin et al. 2014). One of the commonly encountered pathologies in epilepsy surgery practice are lowgrade epilepsy-associated tumors (LEATs), mostly represented by https (Blumcke et al. 2017;Blümcke et al. 2019;Slegers and Blumcke 2020). Surgical removal of LEAT itself does not always suffice to achieve favorable seizure outcome (Englot et al. 2012;Lamberink et al. 2020). The epileptogenic zone often extends further than the visible lesion and the epileptogenic effect takes place both inside and outside the tumor (Thom et al. 2012;Giulioni et al. 2017).
Promising interictal electroencephalographic biomarkers for identifying epileptogenic tissue are high frequency oscillations (HFOs, Hz; fast ripples (FRs) 250-500 Hz) (Jacobs et al. 2008;Zijlmans et al. 2012;van 't Klooster et al. 2017). HFOs in intraoperative electrocorticography (ioECoG) can be recorded over and around the lesion and may be used to delineate the epileptogenic tissue (van 't Klooster et al. 2017). Resection of the HFO-rich area has been associated with good seizure outcome and the presence of postoperative FRs predicts seizure recurrence (Wu et al. 2010;Fujiwara et al. 2012;van 't Klooster et al. 2017). HFO generation involves multiple mechanisms. On a cellular level, it has been proposed that HFOs mainly reflect action potentials of neurons and could be generated by synchronized fast firing of interconnected hyperexcitable neurons (Jiruska et al. 2017).
Neuronal hyperexcitability can be caused by inflammation. Proinflammatory molecules, activated microglia and other inflammatory components are found in and around epileptic lesions and contribute to epileptogenesis by increasing neuronal excitability and decreasing seizure threshold (Boer et al. 2008;Vezzani and Friedman 2011;Vezzani et al. 2011aVezzani et al. , 2013. Activation of the concomitant interleukin-1 beta (IL1Β)/ toll-like receptor 4 (TLR4)/ high mobility group box 1 protein (HMGB1) pathway induces functionality changes in microglia, astrocytes and neurons as well as promoting neuronal hyperexcitability (Vezzani et al., 2011b;Shimada et al., 2014).
We hypothesized that the epileptogenic mechanism of neuroinflammation might correspond to the pathophysiology behind HFO generation (Fig. 1). Understanding the relation between HFOs and inflammation markers in brain tissue could provide insights in the pathophysiological mechanism behind HFO generation and epileptogenesis. In this pilot study, we investigated the correlation between HFOs and inflammation markers known to contribute to epileptogenesis in patients with LEATs who underwent epilepsy surgery.

Patient selection
Patients who underwent epilepsy surgery with ioECoG at 2048 Hz sampling frequency due to LEAT (pathology confirmed) at the University Medical Centre Utrecht between 2008 and 2016 were included if the following data were available: 1) ioECoG recordings before tumor resection, 2) documentation of ioECoG positions, 3) pre-and postoperative MRIs, and 4) pathology specimen available for analyses. Patients with ioECoG with a high level of high frequency noise were excluded. These patients were also included in a retrospective study on HFOs and seizure outcome (van Klink et al. 2021).

Medical ethical approval
The institutional ethical committee approved the study and waived the need for written informed consent because of the retrospective character of the study if data were coded and handled pseudo-anonymously. Tissue for additional pathology analysis was obtained and used in accordance with the Declaration of Helsinki and the Research Code provided by the Medical Ethics Committee of both University Medical Center Utrecht and Amsterdam Medical Center. Tissue usage was approved by the review board of the University Medical Center Utrecht Biobank.

Intraoperative electrocorticography and clinical data
Epilepsy surgery was performed under general anesthesia with propofol. IoECoG was used to delineate the epileptogenic tissue and for functional cortical mapping when necessary. IoECoG was recorded directly on the cortex with 4x8 or 4x5 electrode grids and 1x6 or 1x8 electrode strips (Ad-Tech, Rachine, WI, USA). The silicone grids and strips embedded platinum electrodes with 4.2 mm 2 contact surface with 1 cm inter-electrode distance. Strip electrodes were used to record signals from the mesial and subtemporal area. The location of the grids and strips was captured with photographs. Signals were recorded with a 64-channel EEG system (Micromed, Veneto, Italy), at a 2048 Hz sample frequency with a low pass filter of 538 Hz. Propofol was stopped before each ioECoG recording to achieve a continuous background pattern with minimal propofol effect while the patients remained asleep. The surgical resection was determined based on MRI findings, interictal spikes and spike patterns in the ioECoG. Clinical data were available from medical reports. We collected seizure frequency from the most recent data available prior to surgery. If patients had different types of seizures with different seizure frequencies, the mean seizure frequency was calculated. We quantified semiquantitative descriptions, e.g. 'a couple of seizures' were quantified as 2-3 seizures. Seizure outcome was assessed according to the Engel classification at longest follow-up available. We classified patients with an Engel outcome 1A -1B as seizure free and those with 1C -4 as having recurrent seizures (Engel et al. 1993). Hypothesized relation between HFOs and neuroinflammation. Inflammation could promote neuronal excitability and decrease seizure threshold. HFOs are biomarkers for epilepsy and may be generated by synchronized fast firing of hyperexcitable neurons. Therefore, we hypothesized that HFOs may be associated with inflammatory markers. A. Inflammatory components investigated in this study: activated microglia, CD3-positive T-cells and the IL1b/TLR4/HMGB1 pathway. A short description of the IL1b/TLR4/HMGB1 pathway: IL1b binds to the interleukin-1 receptor-1; HMGB1 binds to TLR4. Activation of the concomitant IL1b/ TLR4/HMGB1 pathway induces transcriptional factors and consequently the expression of pro-inflammatory mediators. B. Neuroinflammation could lead to neuronal hyperexcitability. C. Fast firing of hyperexcitable neurons is associated with epileptogenic HFO generation and epileptogenesis. Seizures stimulate inflammatory processes and hereby form a positive feedback loop. Abbreviations: HFO = High frequency oscillation, CD3 = cluster of differentiation 3, IL1b = interleukin 1-beta, TLR4 = toll-like receptor 4, HMGB1 = high mobility group box 1 protein.

Electrode position
IoECoG electrode positions were reconstructed on the rendering of the cortical surface with 3D Slicer (V4.5.0-1). The cortex rendering was based on presurgical 3D T1 MRI with SPM 12 and 3D Slicer. We segmented the tumor in a presurgical 3D FLAIR MRI (1x1x1 mm) and the resected area in a postsurgical MRI (usually 3D T1, 0.6x0.6x0.6 mm). The presurgical and postsurgical MRIs were merged to determine the overlap of the tumor and the resected area. The ioECoG electrodes were reconstructed on the merged MRI rendering. The location of the electrodes on the cortex were determined by matching the gyral pattern of intraoperative photos of the grids to the cortical rendering (Fig. 2). We determined which contacts of the grids and strips were positioned on the tumor. Bipolar montages of the electrodes were defined as being located on tumor, on non-tumoral tissue that was resected, or outside the resection area (non-resected). For mesial temporal located tumors, the contacts covering the tumor were located on a strip that was slid under the temporal lobe and were out of sight for the surgeon. The first three strip electrodes were defined as being on the tumor.

HFO analyses
We selected one-minute epochs of each ioECoG preceding the resection that were free from large artefacts and with minimal propofol effects. Data were analyzed in a bipolar montage using an automated HFO detection algorithm (Burnos et al. 2014) adapted for our own ioECoG data. Every detected HFO was visually checked in Stellate Harmonie Reviewer (v7.0, Montreal, QC, Canada) by two out of the three reviewers (NvK/WZ/MZ). We used a split screen to visualize ripples (high pass FIR filter at 80 Hz, gain of 5 mV/mm, 0.4 s/page) and FRs (high pass finite impulse response filter at 250 Hz, gain of 1mVmm and 0.4 s/page). We analyzed all recordings preceding the resection. Multiple grid placements could overlap. If the same brain area is measured multiple times, we calculated the mean number of events for that location. HFO rate was defined as the number of ripples or FRs per minute. The total HFO rate was determined for each patient over all locations (R total, FR total ). Then we classified the location of the HFOs according to electrode positions determined on the MRI reconstruction: (a) HFO rate over the tumor site (R tumor , FR tumor ), (b) HFO rate over resected area which was not tumorous according to the MRI (R resection , FR resection ) and (c) HFO rate outside the resected area (R non-resected , FR non-resected ).
Sections of all specimens were processed for hematoxylin and eosin (HE), and for immunocytochemical stainings for several neuronal and glial markers, and Ki67 (as marker of cell proliferation) to confirm the diagnosis of ganglioglioma or dysembryoplastic neuroepithelial tumors. All labelled tissue sections were evaluated by two independent observers blinded to clinical data for the presence or absence of various histopathological parameters and specific immunoreactivity (IR) for the different markers. We semiquantitatively evaluated the IR for CD3. The intensity of HLA-DR (MHC-II), IL-1b, TLR4 and HMGB1 immunoreactive staining was evaluated using a scale of 0-3 (0: -, no; 1: +/-, weak; 2: +, moderate; 3: ++, strong staining). All areas of the tumors and the peritumoral cortex were examined and the score represents the predominant cell staining intensity found in each case. The expression of IL1b and TLR4 was analyzed both in neuronal and glial cells. The frequency of HLA-DR, IL1b, TLR4 and HMGB1 positive cells [(1) rare; (2) sparse; (3) high] was evaluated to give information about the relative number of positive cells within and around the tumor. As proposed before (Prabowo et al. 2015), the product of these two values (intensity and frequency scores) was taken to give the overall score (total score; total score; immunoreactivity score; IRS). A) Example of an MRI reconstruction from a patient included in this study. Red indicates the tumor and yellow indicates the resected tissue. Two grid recordings were performed and the electrode locations were reconstructed. B) Schematic representation of the MRI reconstruction. The tumor (red) located in the right frontal area. Peritumoral inflammation is around the tumor (green). Yellow represents the resected brain tissue that appeared normal on the MRI (resected area). The non-resected area is marked in blue. C) The ioECoG were divided into three areas based on their coverage: tumor (red), resected non-tumoral area (yellow) and non-resected tissue (blue). If multiple grids overlapped in position, we calculated the mean number of events for every location. D) Inflammation status was assessed for two areas: the tumor tissue (red) or the peritumoral area (green). Abbreviations: ioECoG = intraoperative electrocorticography; MRI = magnetic resonance imaging.

Statistical analyses
We used non-parametric statistical methods, because our data were not normally distributed. We reported continuous variables as medians with interquartile ranges (IQR) and categorical variables as frequencies and percentages (%). Differences between two continuous variables were tested with the Wilcoxon Signed Rank test (2 related samples) or Mann-Whitney U test (2 independent samples). We performed Spearman's Rho correlation test correlation analyses for the following variables: -Inflammation markers Â inflammation markers -Ripple rate or FR rate Â inflammation markers in tumor tissue -Ripple rate or FR rate Â inflammation markers in peritumoral tissue We performed standard multiple regression analyses if the assumptions for this analysis were met. A p-value < 0.05 was considered statistically significant. Statistical analysis was performed in IBM SPSS Statistics 26 (IBM Corp., Armonk, NY). We did not correct for multiple comparison because we deemed that the relevance of reporting all potential effects transcends the importance of avoiding type I errors in this explorative study.

Patient population
Thirty-two patients who underwent epilepsy surgery due to LEAT (confirmed by pathology) with ioECoG (sampled at 2048 Hz) were included (Table 1). All lesions were macroscopically completely resected. Twenty-four patients were seizure free after surgery and eight patients had recurrent seizures. All 32 surgical specimens contained sufficient tumoral tissue for analyses. Eighteen specimens contained sufficient peritumoral tissue (normalappearing cortex/white matter adjacent to the tumor). Based on the pathological hallmarks diagnosis was set for ganglioglioma in 26 patients, desmoplastic infantile ganglioglioma in one patient and desembryoplastic neuroepithelial tumor in 5 patients. The HFO rates were not significantly different between the pathologies. All tumors were IDH wild type. 18 tumors had BRAF V600E mutations. The HFO rates were not significantly different between BRAF V600E mutated and BRAF V600E wild type tumors.

IoECoG analyses
1326 bipolar electrodes in ioECoG were analyzed. Ripples were present in the ioECoG of 30 patients with a median ripple rate in this ripple-positive group of 21.5/min, ]. The ripple rate was the highest outside the resected area which was significantly higher than in the tumor (7.8/min vs. 1.3/min, Z = -2.54, p = 0.01) (Fig. 3A). FRs were found in the ioECoG of 10 patients with a median FR rate of 14.5/min, ]. Most FRs were found in the resected area. The FR rate in the resected area was similar to the FR rate in the tumor but significantly higher than the FR rate outside the resection (8.0/min vs. 0.0/min, Z = -2.38, p = 0.02) (Fig. 3B). Preoperative HFO rates were not significantly different between seizure free patients and patients with recurrent seizures.

Inflammatory status
The IRS of IL1b, TLR4, HMGB1 and HLA-DR and the number of CD3-positive cells is presented in the Supplement (Supplementary Table 1). The tumor tissue expressed significantly higher IRS of IL1b, TLR4, HMGB1 and HLA-DR than the peritumoral tissue (all p < 0.001, Fig. 4). The number of CD3-positive cells was higher in the tumor tissue than the peritumoral tissue (U = -3.76, p < 0.001, Fig. 4). IRS of TLR4 in the peritumoral neurons correlated positively with presurgical seizure frequency (Spearman's rho = 0.54, p = 0.02). Other inflammatory markers did not correlate with seizure frequency. The expression of the inflammatory markers in the tumor and the peritumoral tissue was not significantly different between the seizure free and the seizure recurrent group.

HFO rate and inflammation
Patients with ripples in their ioECoG had more CD3-positive cells infiltrating the tumor tissue than patients without ripples (U = 3.5, p = 0.03). When looking at different locations in ioECoG, more CD3-positive cells were found in the tumors that showed ripples than the tumors without ripples (U = 66.5, p = 0.023). The number of CD3-positive cells was not significantly different between patients with and without FRs. Patients with ripples outside the resected area had higher immunoreactivity of IL1b peri (U = 5.0, p = 0.037), TLR4 peri (U = 2.00, p = 0.014) and HMGB1 peri (U5.5, p = 0.041) than those without ripples. The correlation analyses demonstrated that increased ripple rate outside the resected area correlated with increased immunoreactivity of IL1b peri (Spearman's rho = 0.57, p = 0.01) and TLR4 peri (Spearman's rho = 0.62, p = 0.007) in the peritumoral glia cells (Fig. 5B and 6). Increased ripple rates outside the resected area were correlated with increased immunoreactivity of HMGB1 peri in de peritumoral tissue (Spearman's rho = 0.58, p = 0.01, Fig. 5B and Fig. 6). FRs did not correlate with inflammatory markers (Fig. 5C and D).

Discussion
We investigated the correlation between HFOs in ioECoG and inflammation markers involved in epileptogenesis (IL1b, TLR4, HMGB1, HLA-DR and CD3). In the ioECoG of patients who underwent epilepsy surgery due to low-grade tumors, ripples were present in 30 patients and FRs in ten patients. The HFO rate were determined in the ioECoG covering the tumor, the resected nontumorous tissue and outside the resected area. Tumor tissue that showed ripples was infiltrated by more CD3-positive T-cells than tumor tissue without ripples. Outside the resected area, increased ripple rate was positively correlated with increased immunoreactivity of IL1Β, TLR4 and HMGB1 in the peritumoral tissue. In other words, the higher the ripple rate outside the resected area, the more the concomitant IL1b/HMGB1/TLR4 pathways were active in the peritumoral tissue. Other studies suggested a link between ripples and epileptogenicity and a link between inflammation and epileptogenicity (Zijlmans et al., 2012;Vezzani et al., 2013;van Vliet et al., 2018). In our data, the immunoreactivity of TLR4 in peritumoral neurons correlated positively with presurgical seizure frequency. We found a link between ripples and inflammation which could either represent a direct relationship or indirectly support their mutual association with epileptogenicity. FR rate did not correlate with inflammation, but the number of patients with fast ripples was too small to draw conclusions.
We expected a positive correlation between ripple rate and IL1b/HMGB1/TLR4 because IL1b/HMGB1/TLR4 pathway activation promotes neuronal hyperexcitability which corresponds to the pathophysiology behind ripple generation. We found that only ripple rates in the non-resected area were positively correlated with the immunoreactivity of IL1b/HMGB1/TLR4 in the peritumoral tissue. The non-resected area and the peritumoral tissue did not overlap in precise location. The associations might be explained in two ways: 1) the expression of the inflammatory markers outside the resected area cannot be measured. The IRS of IL1b in the peritu-moral tissue does not preclude a diffusion of this cytokine towards the areas where the ripples are found as the IRS can only detect the cells producing interleukins and not the soluble and extracellular interleukins. 2) Ripples outside the resected tissue could be propagated from the tumor and resected area, the areas where the inflammatory markers were analyzed. In that case, one would also Abbreviations: M = male, F = female, age = age in years, DIG = desmoplastic infantile ganglioglioma, GG = ganglioglioma, DNET = dysembryoplastic neuroepithelial tumor, HFO = high frequency oscillation, R = ripple, FR = fast ripple, IDH = isocitrate dehydrogenase, WT = wild type, BRAF V600E = v-raf murine sarcoma viral oncogene homolog B1 p. Val600Glu, Y = yes, N = no. 1 = seizure outcome according to the Engel classification (Engel et al. 1993), 2 = follow-up period in months, 3 = number of high frequency oscillations per minute.
D. Sun, Nicole E.C. van Klink, A. Bongaarts et al. Clinical Neurophysiology 133 (2022) 165-174 expect an association between ripple rate in the tumor and the resected area with inflammation. The absence of such association might be because the ripple rates in these areas were less distributed between patients and thus limited the power of the correlation analyses. The combination of these two factors might explain why association exists between ripple rate and inflammation in two different areas in the brain.
High number CD3-positive T-cell infiltrating in the tumor was also associated with presence of ripples in the tumor. This finding might suggest that the activation of the adaptive immune system could support epileptogenesis and ripple generation. The immunoreactivity of HLA-DR did not relate to the ripple rates, despite the role of microglia in the IL1Β/HMGB1/TLR4 pathway. This could be due to the versatile role of microglia in LEAT related  epilepsy; activated microglia are not only important immune players, but also regulators of tumor proliferation and invasion (Graeber et al. 2002;Aronica et al. 2005).
FR rates did not correlate with inflammatory markers. FRs were recorded in a small group of patients and the FR rates were gener-ally low. Therefore, our observations could not exclude any association between FRs and inflammatory markers with certainty. In addition, FRs and ripples have different pathophysiological mechanisms. FR generation does not rely on synchronized firing of neurons alone, but also involves out-of-phase firing of neuronal   (Ferrier et al. 2006;van Klink et al. 2021;Peng et al., 2021). Differences in the number of included bipolar channels could also partially have contributed to the differences in the recorded number of FRs.
Our retrospective study design has several methodological limitations. First, the definition of the location of HFOs being in the tumor or in the resected tissue is not a 100% sure. We defined HFOs in tumor being HFOs recorded in the ioECoG covering the tumor site, but these HFOs are not 'purely' HFOs generated by the tumor. The ioECoG measures cortical activity in a two-dimensional way. In case of a superficial tumor, the ioECoG records mostly the signals of the neoplastic neurons. However, in a deep and more medially located tumor, the ioECoG would consist of signals from the overlaying cortex mixed with signals from the tumor itself. HFOs in the resected tissue outside the tumor should represent the non- Fig. 6. Scatterplot of ripple rate outside the resected area vs. IL1b, TLR4 and HMGB1 in the peritumoral tissue. The immunoreactivity of inflammatory markers on the x-axis and ripple rates were expressed in number of ripples per minutes on y-axis. Abbreviations: IRS = immunoreactivity score, CD3 = cluster of differentiation 3, IL1b = interleukin 1-beta, TLR4 = toll-like receptor 4, HMGB1 = high mobility group box 1 protein.
D. Sun, Nicole E.C. van Klink, A. Bongaarts et al. Clinical Neurophysiology 133 (2022) 165-174 neoplastic tissue which was epileptogenic as it had to be removed during surgery. It is, however, also possible that some of this brain tissue is resected to gain access to the suspected epileptogenic zone. We cannot differentiate these two scenarios. Secondly, we did not correct for the extent of the ioECoG coverage. We choose to analyze the overall ripple rates because the extensiveness of the electrode coverage was at least partially determined by the epileptogenic signal recorded and thus intrinsic to the extent of the epileptogenic zone. To limit potential bias, we calculated the mean number of HFOs if the same brain area was recorded multiple times. Lastly, we did not correct for clinical factors that might have acted as confounders such as seizure duration and higher seizure frequency (Aronica et al. 2005;Boer et al. 2008;Vezzani et al. 2011b) because our data was not suited for multivariate regression models. This research provides arguments for further investigations into the relationship between HFOs and inflammation. Extending these findings to IDH mutated (diffuse) gliomas, with its specific epileptogenic peritumoral environment (including the presence of 2hydroxyglutarate), may offer further pathophysiological insights. In future prospective studies, tissue for immunohistological analyses should be sampled at sites where HFOs are recorded to assure a close relation between HFOs and local inflammation. If possible, tissue should also be sampled from a site without HFOs to serve as control sample. The electrode positions have to be marked on the removed brain tissue, e.g. with ink markings or by registration of coordinates with the neuronavigation system. By including the depth of tumors with respect to the grid positions in the MRI reconstructions one could improve accuracy of HFO localization. Intraoperative recordings with high resolution grids could be considered to improve FR sampling. Additional spike analyses might add value in distinguishing between physiological and pathological ripples. Distinction between temporal and extratemporal surgeries is recommended in a larger group of patients.

Conclusion
Our work showed that increased ripple rate outside the resected area is correlated with increased immunoreactivity of IL1b/HMGB1/TLR4 in the peritumoral glia cells. CD3-positive Tcell infiltration in the tumor was associated with the local presence of ripples. We found no correlations between fast ripple rates and inflammatory markers, but association between the two could not be excluded due to low numbers of patients with fast ripples. This is the first study to directly compare HFOs and inflammatory markers that contribute to epileptogenesis and provides grounds for further investigations into the relation between HFOs and inflammation. To conclude, HFO rate in the ripple band is associated with inflammatory components such as T-cells and the IL1b/ HMGB1/TLR4 pathway. These findings might suggest that synchronized fast firing of hyperexcitable neurons may be a part of the underlying mechanism of ripple generation.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.