Analysis of intraoperative human brain tissue transcriptome reveals putative risk genes and altered molecular pathways in glioma-related seizures

Abstract

Background:

The pathogenesis of glioma-related seizures (GRS) is poorly understood. Here in, we aim to identify putative molecular pathways that lead to the development of GRS.

Methods:

We determined brain transcriptome from intraoperative human brain tissue of patients with either GRS, glioma without seizures (non-GRS), or with idiopathic temporal lobe epilepsy (iTLE). We performed transcriptome-wide comparisons between disease groups tissue from non-epileptic controls (non-EC) to identify differentially-expressed genes (DEG). We compared DEGs to identify those that are specific or common to the groups. Through a gene ontology analysis, we identified molecular pathways enriched for genes with a Log-fold change ≥1.5 or ≤−1.5 and p-value <0.05 compared to non-EC.

Results:

We identified 110 DEGs that are associated with GRS vs. non-GRS: 80 genes showed high and 30 low expression in GRS. There was significant overexpression of genes involved in cell-to-cell and glutamatergic signaling (CELF4, SLC17A7, and CAMK2A) and down-regulation of genes involved immune-trafficking (CXCL8, H19, and VEGFA). In the iTLE vs GRS analysis, there were 1098 DEGs: 786 genes were overexpressed and 312 genes were underexpressed in the GRS samples. There was significant enrichment for genes considered markers of oncogenesis (GSC, MYBL2, and TOP2A). Further, there was down-regulation of genes involved in the glutamatergic neurotransmission (vesicular glutamate transporter-2) in the GRS vs. iTLE samples.

Conclusions:

We identified a number of altered processes such as cell-to-cell signaling and interaction, inflammation-related, and glutamatergic neurotransmission in the pathogenesis of GRS. Our findings offer a new landscape of targets to further study in the fields of brain tumors and seizures.

1. Introduction

Seizures develop in 40–70 % of patients with gliomas, and approximately 30 % are pharmaco-resistant even after glioma resection (van Breemen et al., 2007). Seizures are devastating with a significant negative effect on a patient’s quality of life (Chaichana et al., 2009). The exact mechanisms of glioma-related seizures (GRS) remain poorly understood but are likely multifactorial (Armstrong et al., 2016). Among these, glutamate-induced excitotoxicity and disruption of intracellular communication have garnered the most attention (Huberfeld et al., 2011). Accumulating evidence also suggests that tumor growth stimulates seizures which in turn encourage tumor growth suggesting the two conditions may share common pathogenic mechanisms (Armstrong et al., 2016; Venkatesh et al., 2019; Venkataramani et al., 2019). One explanation for the close link between the two conditions is elevated glutamate in the tumor microenvironment (TME) due to an increased expression of the cystine-glutamate transporter with ensuing over-activity of glutamatergic signaling leading to seizures and oncogenesis (Armstrong et al., 2016; Robert et al., 2015).

Transcriptome profiling using high-throughput RNA sequencing (RNAseq) allows an unbiased broad mapping of molecular constituents present in cells and tissues (Stark et al., 2019). In the past two decades, aberrant gene expression has been reported in different epilepsy pathologies, including mesial temporal sclerosis (MTS), focal cortical dysplasia, status epilepticus, and epileptic encephalopathies (Dixit et al., 2016, 2018; Griffin et al., 2016; Hansen et al., 2014; Veeramah et al., 2013). However, such studies using an RNAseq approach analyzing transcriptome profiling in patients with GRS are lacking. The lack of transcriptomic data makes it difficult to perform mechanistic studies addressing the pathogenesis of GRS. To this end, we performed transcriptome profiling in brain tissues collected from fresh intraoperative surgical tissue of patients with GRS, glioma without seizures (non-GRS), idiopathic temporal lobe epilepsy (iTLE), and frozen brains of autopsied individuals without a history of seizure or glioma (non-epileptic controls). The non-GRS samples were included to identify genes and pathways involved in the epileptogenesis of glioma, while the iTLE cohort was included to provide insight into the common and novel genes and pathways in the genesis of GRS and iTLE. The non-epileptic control group was included to identify altered genes and pathways present in one group and not others or common to both the disease and non-epileptic control groups (Fig. 1A). Our goal was to test the hypothesis that alterations in inflammation-related processes and glutamatergic neurotransmission underlie the pathogenesis of GRS.

Fig. 1.

A. The DEG analysis outline is shown; DEGs were defined as genes with a|logFC|≥1.5 and p-Value ≤0.05. B. Pairwise comparison: GRS vs. non-GRS DEGS. Upregulated and downregulated DEGs were used to generate a Venn diagram showing unique and shared DEGs between GRS and non-GRS patients. C. Pairwise comparison: GRS vs. iTLE DEGS. Upregulated and downregulated DEGs were used to generate a Venn diagram showing unique and shared DEGs between GRS and iTLE patients. D. Tre-way comparison: GRS vs. non-GRS vs. iTLE DEGS. Upregulated and downregulated DEGs were used to generate Venn diagrams showing unique and shared DEGs between GRS, non-GRS, and iTLE patients.

2. Methods

2.1. Patients and tissue collection

Following institutionally review board-approved procedures and patient informed consent, we collected patient’s demographics and brain tissue samples during craniotomy for tumor resection or lobectomy from patients with either GRS (n = 9), glioma without seizures (non-GRS, n = 8), or with idiopathic temporal lobe epilepsy (iTLE, n = 7). All our tumor samples were obtained in bulk from contrast-enhancing regions of the tumor.

2.2. RNA sequencing and bioinformatics analyses

2.2.1. RNA isolation and library construction

Intraoperative surgical brain tissue specimens were processed within an hour from surgery, and 200 mg aliquots are stored in RNAlater solution (Thermo Fisher) at −80 °C until processing. Postmortem frozen tissue of subjects from the non-epileptic control group was obtained from the Mayo Clinic Brain Bank. Post mortem interval ranged from 3 to 18 h. None of the study cohort except for the iTLE group had undergone intracranial EEG.

RNA was isolated from homogenized frozen brain tissue samples using TRIzol® (Thermo Fisher) and DNase purification, followed by clean-up using Qiagen RNeasy columns. Samples were assessed for RNA quality and quantity using a 2100 Bioanalyzer (Agilent) with a RNA 6000 Nano Chip (Agilent) and diluted to a final concentration of 50 ng/μl. RNA library preparation and sequencing were performed by the Mayo Clinic Medical Genome Facility Gene Expression Core (GEC) using the TruSeq RNA Exome Sample Prep Kit (Illumina, San Diego, CA). The library concentration and size distribution were determined on a Bioanalyzer DNA 1000 chip (Agilent). Before RNA sequencing, samples were randomized across flow-cell lanes with respect to diagnosis, age, sex, tumor location, and IDH1 mutation status. Libraries were sequenced on an Illumina HiSeq 4000 (paired-end, 100bp reads), with 6 samples per flow cell lane (supplementary Fig. 1A).

2.3. Read alignment and sample outlier detection

Base-calling was performed using Illumina’s RTA, and FASTQ reads aligned to human genome build GRCh38 using the in-house MAP-RSeq pipeline version 3.0.1, (Sun et al., 2013) applying STAR (Murray et al., 2000) and featureCounts (Wilson et al., 2005) for read alignment and counting, respectively. On average, more than 50 (56.2 ± 2.0) million pairs of reads were mapped to the reference genome and more than 80 % (87.6 % ± 2.6 %) of these reads were mapped to known genes defined in Ensemble release 78. Quality measures were examined, including base calling quality, GC content, mapping statistics, and sex consistency, which required the consistency between recorded sex and the sex inferred from chromosome Y genes expression. Raw read counts were normalized using Conditional Quantile Normalization (CQN) (Liu and Murray, 2012) via the Bioconductor package cqn, accounting for sequencing depth, gene length, and GC content (Supplementary Fig. 1B). CQN normalized expression measures were assessed using principal component analysis (PCA) to identify outliers defined as greater than 4 standard deviations from the mean of the first two principal components. One sample (from the GRS cohort) was identified as an outlier and removed from downstream analysis. Final number per group is as follows: Control, n = 5, GRS, n = 9, nonGRS, n = 8, and iTLE n = 7. We also performed PCA plot using two sets (A and B) of genes independently. Set A includes genes with one or more reads in any sample and have a standard deviation of expression greater than zero, resulting in 35,656 genes. Set B includes the top 50 DEGs between GRS and non-GRS.

2.3.1. CQN and source of variation analysis

Before edgeR, conditional quantile regression (CQN) normalization was applied to raw read counts using R cqn package to generate an offset matrix that incorporated the adjustment of library size, gene length, and gene GC content. The raw read counts, the offset matrix, and the design matrix were provided to edgeR, which then estimated gene-wise dispersions and fitted the negative binomial generalized linear models using glmFit function. Likelihood ratio tests were performed for each diagnosis group vs. controls using glmLRT function. Association p-values were adjusted for multiple tests using false discovery rate (Benjamini-Hochberg), where appropriate. Statistical analysis was performed using R statistical software (R Foundation for Statistical Computing, version 3.2.3). After the source of variation analysis, age was included as a covariate (Supplementary Fig. 1C). Due to the limited sample size, the False discovery rate (FDR) adjusted p-value of 0.05 is too stringent for this exploratory study.

2.4. Differential expression analysis

Differential expression was performed using edgeR R package version 3.12.0. with negative binomial generalized linear model including age at the time of surgery (disease groups) or death (control group) as a covariate. CQN and source of variation analysis are described in Supplementary Fig. 1B–C Differentially expressed genes (DEGs) were selected for all 3 disease groups (GRS, iTLE, and nonGRS) based on log fold change (logFC) ≥1.5 or ≤− 1.5 and p-value < 0.05 compared to controls, as shown in the volcano plots in Supplementary Fig. 1D–F. Upregulated (logFC ≥1.5) and downregulated (logFC ≤− 1.5) DEGs for each disease group were then used for Venn diagrams generation showing common and unique DEGs between the selected gene lists (Fig. 1B–D). Genes with logFC between −1.5 and +1.5 were excluded from downstream analysis. Volcano plot and bar graphs were generated using GraphPad Prism version 8.0.0 for Windows, GraphPad Software, San Diego, California USA. Heatmap was created using function heatmap.2 in R gplots package, using the default method for clustering.

2.5. Pathway analysis

Gene ontology (GO) and pathway analysis were performed on DEG results from direct analysis of GRS vs. nonGRS and GRS vs. iTLE data sets, using Ingenuity Pathway Analysis (Qiagen, Hilden, Germany). Core analysis function was utilized on all diseased groups vs. non-epileptic controls and was utilized for pathway analysis for the GRS vs non-GRS vs iTLE comparison. Canonical pathways, diseases and functions, and upstream regulators were assessed for significant enrichment and directionality, utilizing a z score >±1.5 and p-value <0.05. IPA generates p values by Fisher’s exact test with p-value adjusted using Benjamini-Hochberg multiple test correction by comparing the number of DEGs relative to the total number of occurrences of these genes in all functional annotations stored in the Pathway Knowledge Base.

3. Results

3.1. Clinical characteristics and source of variation analysis

The clinical characteristics of the disease groups are summarized in Table 1. Representative histopathology slides from the three disease groups are also provided in Fig. 2. Additional information on these groups is provided in Supplementary Tables 1 and 2. The clinical characteristics of the non-epileptic control are provided in Table 2. RINs data is provided in Supplementary Fig. 1A. Before performing differential expression analysis, the source of variation analysis was conducted, which determined that among sex, age, RIN, and diagnosis group, the largest variation in the data was from the diagnostic grouping followed by age (Supplementary Fig. 1C).

Table 1.

Clinical characteristics of the three disease group.

	GRS cohort (N = 9)	Non-GRS cohort (N = 8)	iTLE cohort (N = 7)
Mean age, yrs. (Range)	50 (32–57)	56 (28–74)	46 (26–74)
Sex (n)	Male (8): Female (1)	Male (6): Female (2)	Male (2): Female (5)
Presenting symptom (n)	Seizure (9)	Incidental (1), Hearing problem (1), Headache (1), Dysphasia (2), Memory, complaint (2), Syncope (1)	Seizure (7)
Disease duration (mean)	4.5 months	4.7 months	191 months
	Focal aware (3)		Focal impaired aware (7)
Seizure type (n)	Focal impaired aware (4)	N/A	Focal to bilateral tonic-clonic (4)
	Focal to bilateral tonic-clonic (3)
Antiseizure drug therapy (n)	Levetiracetam (7), Lamotrigine (1), Lacosamide (1), Zonisamide (1)	None (7)	Lamotrigine (2), Levetiracetam (4) Lacosamide (1), Topiramate (2) Perampanel (1), Clobazam (1), Eslicarbazepine (1)
Resected tissue location (side, lobe) [n]	Left frontal (3), left temporal (2), right temporal (3), left parietal (1)	Levetiracetam (1)	Right temporal (5)
			Left temporal (2)
Pathology (n)	Diffuse astrocytoma, IDH1 mutant (1)	Right frontal (1), left frontal (1), left temporal (2), right temporal (1), left parietal (3)	Mild gliosis (2)
	Oligodendroglioma IDH1 mutant, 1p/19q co-deleted (3)		Marked gliosis (2)
	Anaplastic astrocytoma, IDH1 wild (3)		Chaslin gliosis (3)
	Glioblastoma, IDH1 wild, MGMT-(2)

GRS: glioma-related seizure; iTLE: idiopathic temporal lobe epilepsy; NA: not applicable; IDH1: isocitrate dehydrogenase 1; MGMT+: O⁶-methylguanine-DNA methyl-transferase- gene promoter methylation; MGMT-: MGMT gene promotor un- methylated.

Fig. 2.

Representative histopathology slides from the study cohort. A. Sample for a patient with idiopathic temporal lobe epilepsy. Chaslin gliosis with subpial gliosis (arrows) on the H&E and GFAP stains, respectively. B. Sample for a patient with diffuse astrocytoma, IDH-wildtype, glioma without seizures. Granular cell neurons of the hippocampus (arrow). The tumor is negative with the IDH1-R132H stain (insert). C. Sample from a patient with diffuse astrocytoma, IDH-mutant, glioma without seizures. The tumor is positive for the IDH1-R132H stain (insert). D. Sample from a patient with an oligodendroglioma, IDH-mutant, and 1p/19q co-deleted, glioma-related seizures. The tumor is positive for the IDH1-R132H stain (insert). E. Sample from a patient with anaplastic astrocytoma, IDH-mutant, glioma without seizures. The tumor is mitotically active (circles) and is positive for the IDH1-R132H stain (insert). F. Sample from a patient with anaplastic astrocytoma, IDH-wildtype, glioma-related seizures. The tumor is mitotically active (circle) and is negative for the IDH-132H stain (insert). G. Sample from a patient with glioblastoma, IDH-wildtype, glioma without seizures. The tumor has areas of necrosis (asterisk) and is negative IDH1-R132H (insert). H. Sample from a patient with glioblastoma, IDH-wildtype, glioma-related seizures. The tumor has areas of necrosis (asterisk) and is negative IDH1-R132H (insert).

Table 2.

Clinical characteristics of the non-epileptic control cohort.

Patient #	Sex	Age at death, yrs	Diagnosis	Cause of death	Tissue for pathology	Pathology
1	Male	63	Coronary artery disease; polymyalgia rheumatica	Cardiac	Left temporal	Normal for age (Braak stage I; Thal phase 1)
2	Female	72	Ischemic heart disease; myocardial infarction	Cardiac	Left temporal	Normal for age (Braak stage II; Thal phase 1)
3	Male	75	Ischemic heart disease	Cardiac	Left temporal	Normal for age (Braak stag II, Thal phase 0)
4	Male	76	Multisystem organ failure (acute myocardial infarction; pulmonary hemorrhage; acute renal failure)	Cardiopulmonary	Right temporal	Cerebrovascular disease (chronic) (Braak stage III, Thal phase 0)
5	Female	78	Depression	Dissecting aortic aneurysm	Left temporal	Normal for age (Braak stag I, Thal phase 1)

3.2. Differential gene expression analysis and PCA plots

Gene expression levels for the 3 groups were analyzed against control tissue. Supplementary Fig. 1D–F shows the distribution of logFC data for each group. Using a logFC ≥1.5 or ≤− 1.5 and p-value < 0.05, we identified 857 transcripts upregulated in GRS, 1694 transcripts in non-GRS, and 111 transcripts in iTLE; 338 downregulated transcripts were identified in GRS, 1342 in non-GRS, and 77 in iTLE. To identify genes that were differentially expressed in one disease but not another, we compared lists of DEG’s and illustrated “common” and “unique” genes using Venn diagrams (Fig. 1B–D). The PCA analysis showed that the GRS and non-GRS samples are more spread than non-epileptic-control and iTLE samples (Fig. 3A). It also showed that compared to the non-GRS, the GRS samples are closer to the iTLE and non-epileptic control samples (Fig. 3A and Supplementary Fig. 2A). As expected, we also observed that different glioma types (glioblastoma, diffusion astrocytoma, anaplastic astrocytoma, oligodendroglioma) tend to form different clusters (Fig. 3B and Supplementary Fig. 2B).

Fig. 3.

PCA analysis of the 50 DEGs between glioma-related and non-GRS (Table 3). A. Shows sampling clustering by grouping: non-epileptic controls (ctrl), glioma-related seizures (GRS), glioma without seizures (non-GRS), and idiopathic temporal lobe epilepsy (iTLE). B. shows sample clustering by histopathology (Ctrl = normal for age; iTLE = Chaslin gliosis).

3.2.1. Differences between GRS and non-GRS

To identify genes that may provide insights into dysregulated pathways specific to seizures associated with glioma, we first analyzed the DEGs unique for the GRS group. We identified 110 transcripts that met our criteria for dysregulation in patients with GRS than controls that did not meet these criteria in non-GRS. Among these, 80 genes were overexpressed, and 30 genes were underexpressed (Fig. 1B, Supplementary Fig. 3A). We also performed direct expression analysis between the GRS and non-GRS cohorts. The top 50 genes (ranked by logFC levels, average p-value <0.001, provided in Table 3) generated directly comparing GRS vs. non-GRS were then used to create a heat map that shows cqn normalized expression levels in all samples analyzed, as well as fold change values for the 3 disease groups vs. controls (Fig. 4).

Table 3.

Top 50 differentially expressed genes GRS vs non-GRS.

Upregulated Genes GRS vs nonGRS			Downregulated Genes GRS vs nonGRS
Gene	logFC	p-value	Gene	logFC	Pvalue
SNORD115–15	2.588	8.56E-03	H19	−5.862	2.91E-06
SNORD115–34	2.521	1.67E-02	MMP9	−5.862	3.01E-02
ZNF676	2.509	1.80E-03	ACEA_U3	−4.167	6.01E-05
CPLX2	2.398	9.41E-03	TRDC	−3.89	7.02E-07
F5	2.397	7.63E-03	SLC34A2	−3.782	5.27E-05
SNORD115–11	2.358	6.43E-03	CA9	−3.767	6.33E-04
HPCAL4	2.345	1.86E-03	LIF	−3.763	1.34E-03
ZNF99	2.329	3.28E-03	LOX	−3.644	1.40E-04
CELF4	2.319	8.76E-03	CXCL8	−3.599	1.55E-03
SNORD115–10	2.288	9.60E-03	STC2	−3.52	5.00E-05
FAM153C	2.250	2.51E-02	POSTN	−3.517	3.75E-03
SNORD115–40	2.218	7.54E-03	DES	−3.51	5.53E-06
NSFP1	2.204	3.35E-02	MIR6765	−3.468	2.18E-02
KRT17P1	2.171	1.47E-02	SNORD3B-1	−3.413	3.77E-04
SNCG	2.096	1.25E-02	DNAH11	−3.311	3.40E-05
FBXL16	2.071	7.42E-03	VEGFA	−3.035	2.95E-05
SNORD115–21	2.063	2.14E-03	HIST1H4L	−3.033	3.47E-03
DRC1	2.042	8.83E-03	TREM1	−3.006	5.30E-03
SNORD115–5	2.028	5.20E-03	PLAU	−3.003	1.38E-04
TCEAL6	1.976	1.50E-02	ADM	−2.87	1.46E-03
IGFN1	1.967	4.20E-02	IGFBP2	−2.805	2.30E-03
ZNF98	1.955	9.04E-03	HIST1H2AH	−2.766	1.96E-04
CRABP1	1.945	1.65E-03	HIST1H2AG	−2.703	3.74E-03
SNORD115–41	1.931	1.20E-02	HILPDA	−2.701	6.88E-07
SLC17A7	1.905	3.45E-02	IGF2BP3	−2.694	7.96E-04

Fig. 4.

Heatmaps of the TOP 50 DEGs selected in the GRS vs. non-GRS direct analysis. The bottom panel shows the relative expression levels (log2FC) of GRS, non-GRS, and iTLE versus control and of GRS samples versus non-GRS samples. The top panel depicts the cqn normalized expression levels of all samples analyzed. Red represents high expression, and blue represents low expression.

3.2.2. Differences between GRS and iTLE

To provide insight into the common pathways that are shared by iTLE and GRS, the second comparison focused on GRS and iTLE against control tissue. We identified 26 down-regulated DEGs and 69 upregulated DEGs shared by the two groups (Fig. 1C, Supplementary Fig. 3B). 1098 transcripts were dysregulated uniquely in the GRS samples, among which 786 genes were overexpressed, and 312 genes were underexpressed. The top 50 most significantly differentially overexpressed or underexpressed genes in samples with GRS vs. iTLE are provided in Table 4 (average p-value <0.001).

Table 4.

Top 50 differentially expressed genes GRS vs iTLE.

Upregulated Genes GRS vs iTLE			Downregulated Genes GRS vs iTLE
Gene	logFC	Pvalue	Gene	logFC	Pvalue
MYBL2	6.966	5.41E-10	SLC17A6	−2.585	2.71E-04
RRM2	6.837	4E-08	SPATA31C1	−2.557	1.46E-02
TOP2A	6.693	2.75E-09	SNORD115–5	−2.547	1.33E-02
H2AC14	6.210	3.09E-06	DCAF12L2	−2.447	2.19E-03
KRT18P14	5.717	1.31E-05	TYRP1	−2.433	1.44E-04
MKI67	5.710	9.99E-09	DES	−2.35	3.45E-04
H2AC16	5.672	9.77E-06	SERTM1	−2.346	5.66E-04
H1–5	5.530	2.07E-08	BDNF	−2.315	1.23E-04
KIF20A	5.378	8.74E-06	PROKR2	−2.275	4.18E-03
HJURP	5.239	1.71E-06	FOSB	−2.259	1.79E-03
GSC	5.178	0.0413	CARTPT	−2.247	5.38E-03
H3C2	5.105	2.3E-07	RTBDN	−2.198	1.56E-04
H3C3	5.039	3.09E-07	NLGN4Y	−2.193	7.04E-02
NCAPG	4.943	2.49E-07	HTR3B	−2.15	2.73E-03
H2BC17	4.873	6.11E-08	TDRD9	−2.13	3.20E-05
PBK	4.847	4.06E-05	CHAD	−2.116	5.20E-03
H3C12	4.821	3.24E-11	HAS1	−2.093	1.52E-03
H2AC17	4.724	7.21E-10	KIF12	−2.073	7.15E-03
LTF	4.580	0.0188	MT1A	−2.067	5.03E-04
NDC80	4.523	7.07E-06	KCNV1	−2.065	1.44E-02
KIFC1	4.385	5.99E-06	HTR2A	−2.046	1.10E-03
FCGBP	4.372	0.000622	CNGB1	−2.033	1.72E-03
ASPM	4.180	8.69E-07	FRMPD2B	−2.013	4.91E-03
H2AC13	4.063	6.06E-06	DUX4L27	−2.011	2.61E-02
DTL	4.054	1.39E-06	ZCCHC12	−2.003	2.34E-03

3.2.3. Three-way comparison

Finally, to provide insight into the dysregulated genes unique to GRS, we aimed to identify DEGs in the GRS groups that were not shared by non-GRS and iTLE. We identified 101 unique DEGs for GRS; among these, 74 genes were overexpressed, and 26 genes were underexpressed in the GRS samples only (Fig. 1D, Supplementary Fig. 3C). When we plotted the DEGs identified in the 3 disease groups against each other, we identified 11 genes shared among GRS and iTLE samples but not present in the non-GRS lists. Among these, 6 genes were overexpressed, and the remaining 5 were downregulated.

3.3. Gene ontology analyses

The top canonical pathways identified as enriched for genes significantly dysregulated in GRS than non-GRS (top genes listed in Table 3) are provided in Fig. 5A. This analysis revealed uniquely overexpressed genes involved in cell movement and cell-to-cell signaling and interaction in the GRS samples. The top canonical pathways identified as enriched for genes that were dysregulated in GRS than iTLE (top genes listed in Table 4) are provided in Fig. 5B. Among the genes that were upregulated only in the GRS samples, there was significant enrichment for genes involved in cellular motility and cell death and survival. Similarly, among the genes that were down-regulated only in the GRS samples, there was down-regulation of genes encoding for vesicular glutamate transporter 2, BDNF, DES, and FBJ murine osteosarcoma viral oncogene homolog B (FOSB). We also performed a pathway analysis of the common DEGs between GRS and iTLE identified in the Venn diagram analysis. Top canonical pathways are provided in Fig. 5C. GO enrichment analysis identified differential regulation of pathways distinguishing GRS from non-GRS, and iTLE is provided in Fig. 5D. Lastly, we assessed the DEGs in the iTLE vs. control analysis to determine if known pathways implicated in epilepsy were identified. GO analysis also revealed that the iTLE and GRS samples had several genes previously associated with epileptic seizures compared to the non-epileptic controls [Supplementary Table 3].

Fig. 5.

A. Top significant pathways identified by IPA in the pairwise comparison GRS vs. non-GRS. Bars represent the percentage of genes upregulated (red) or downregulated (blue) in each pathway. Black dots represent p-values. Top significant pathways identified by IPA. B. Pathways analysis of DEGs in the pairwise comparison GRS vs. non-GRS. C. Pathways analysis of DEGs common for GRS and iTLE. D. Pathways analysis of DEGs identified in the comparison GRS vs. nonGRS vs. iTLE. Bars represent the percentage of genes upregulated (red) or downregulated (blue) in each pathway. Black dots represent the p-value.

4. Discussion

This study used intraoperative human fresh brain tissue from different patients with different conditions and examined risk genes and molecular pathways involved in GRS using a high-throughput RNAseq analysis of human brain tissue transcriptomes. Our analysis provides a valuable list of study targets to the fields of neuro-oncology and epilepsy research.

4.1. Glioma with seizures vs. glioma without seizures

4.1.1. Relationship between overexpressed genes in GRS and epileptogenesis

Compared to the non-GRS samples, the GRS samples showed significant enrichment for genes involved in cell-to-cell signaling and interaction. While most of the identified genes were novel, altered expression of SLC17A7, CELF4, and CAMK2A genes have been reported in non-tumoral epilepsies (NTE) (van der Hel et al., 2009; Sun et al., 2013; Murray et al., 2000). Vesicular glutamate transporters (VGLUTs) are responsible for loading synaptic vesicles with glutamate and have been implicated in the regulation of quantal size and presynaptic plasticity (Wilson et al., 2005).TLE patients without MTS had an increase in VGLUT1 immunoreactivity. Besides, upregulated VGLUT1 reactivity was detected in the dentate gyrus of these patients (van der Hel et al., 2009). The overexpression of VGLUT1 could represent a higher vesicular glutamate storage capacity, which may increase glutamatergic transmission and contribute to higher extracellular glutamate levels and excitability (Wilson et al., 2005). Altered CELF4 has also been suggested to play a crucial role in seizure generation in a complex neurological disease model (Sun et al., 2013). The increase in the expression of the NMDA receptor Subunit 3A (GRIN3A) with its closely related CAMK2A gene in the GRS samples is surprising given that reduced expression of CAMK2 activity has been consistently reported in MTS (Murray et al., 2000) and various animal models of epilepsy (Liu and Murray, 2012).

4.1.2. Relationship between downregulated genes in GRS (vs non-GRS) and epileptogenesis

Some of the under-expressed genes in the GRS samples compared to the non-GRS samples, including BMP, DES, PBK, and H19, have also been previously reported as epilepsy-related genes (Han et al., 2018; Liu et al., 2016). Long non-coding RNA 19 (H19) mediates seizure-induced neuronal damage in a TLE rat model (Liu et al., 2016). Furthermore, reduced expression of BMP and DES genes has been reported in refractory epilepsy (Liu et al., 2016). Interestingly, we also found significant down-regulation of pro-inflammatory cytokines (CXCL8, LIF, TGFβ1, VEGFA, and IL1RAP) in the GRS samples. This is in contrast to previous studies that suggested an up-regulation of pro-inflammatory cytokines in NTE (Vezzani et al., 2008; Ravizza et al., 2008). It is possible that the downregulation of CXCL8 and unaltered IL1-beta in our study could be attributed to deregulation of the pathway, thus failing to counteract the pro-inflammatory state (Ravizza et al., 2008). Alternatively, it is conceivable that the down-regulation of pro-inflammatory cytokines in the GRS sample could be secondary to the anti-inflammatory effects of antiseizure medications (ASMs).23ASMs including levetiracetam and lamotrigine, have been previously shown to reduce levels of pro-inflammatory cytokines in vitro and in vivo studies (Stienen et al., 2011; Abu-Rish et al., 2018).

4.1.3. Relationship between dysregulated genes in GRS and oncogenesis

Some of the genes differentially expressed between the GRS and non-GRS samples identified in this study have been previously shown to be involved in glioma proliferation and migration (Lau and Xu, 2018; Shi et al., 2014; Hasan et al., 2019). This further supports the prevailing notion that seizures and oncogenesis may likely share a common mechanism and pathway (Armstrong et al., 2016; Huberfeld et al., 2011; Venkatesh et al., 2019; Venkataramani et al., 2019). Alternatively, the oncogenic genes could cause changes in the TME from tumor proliferation and invasion that are conducive for the genesis of seizures. For example, F-box proteins (FBS), including FBXL16, play an important role in cancer development and progression (Randle and Laman, 2016). Altered expression of long non-coding RNA H19 has also been shown to promote glioma survival, proliferation, and invasion (Shi et al., 2014). Further, emerging evidence has revealed the involvement of inflammatory mediators including CXCL8 in glioma growth by promoting tumor cell proliferation and neovascularization (Hasan et al., 2019). The upregulation of several genes involved in the glutamatergic signaling (VGLUT1, GRIN3A, and CAMK2A) along with the down-regulation of the inflammatory pathways (CXCL8, TGFβ1, and IL1RAP) may suggest a possible intersection between these pathways in the TME.

4.2. Glioma with seizures vs. idiopathic temporal lobe epilepsy

4.2.1. Relationship between overexpressed genes in GRS (vs. iTLE) and epileptogenesis

While most of the genes that were overexpressed in the GRS samples (vs. iTLE samples) are markers of oncogenesis (Niesen et al., 2013; Stangeland et al., 2015), some of these have also been reported to be associated with genetic and acquired epilepsies (ASPM, FCGBP, and PBK). Significantly, activation of the mTOR pathway by various proteins, including a serine/threonine-protein kinase (PBK), has been implicated as a mechanism by which diverse genetic mutations and acquired abnormalities lead to a final common pathway of seizures (Ostendorf and Wong, 2015). Besides, rapamycin and similar compounds inhibit mTOR complex 1 and decrease seizures, delay seizure development or prevent epileptogenesis in many animal models of mTOR hyperactivation (Huang et al., 2012). While ASPM mutations have been identified in patients with primary microcephaly and seizures (Shen et al., 2005), mutations in FCGBP gene have also been reported in idiopathic epilepsy (Rawat et al., 2019).

4.2.2. Relationship between downregulated genes in GRS (vs. iTLE) and epileptogenesis

Some of the under-expressed DEGs in the GRS samples (vs. iTLE) have been previously identified as epilepsy-related genes. For example, HAS1–3 knockout in mice results in alteration of neuronal activity and seizures (Arranz et al., 2014). In a large-scale microarray expression study on mesial TLE (MTLE) patients, the serotonin receptor (HTR2A) gene was suggested to play a role in the MTLE pathophysiology (Jamali et al., 2006). Significantly, among the genes under-expressed in the GRS samples (vs. iTLE), two genes are known to be involved in the glutamatergic neurotransmission: vesicular glutamate transporter-2 and FOSB. Altered protein expression of VGLUTs could affect the quantal size and glutamate release under both physiological and pathological conditions (El Mestikawy et al., 2011). In fact, VGLUT-2 protein expression was found to be significantly decreased while VGLUT-3 protein was significantly increased in the hippocampus of patients suffering from TLE due to MTS (Van Liefferinge et al., 2015). VGLUT-2 knockout mice were also shown to be more susceptible to seizures in animal models of epilepsy (Schallier et al., 2009). Further, increased hippocampal excitability was observed in conditional VGLUT-2 knockout mice (Schallier et al., 2009). In contrast to previous reports, we found down-regulation of FOSB and BDNF genes in the GRS samples (vs. iTLE) (Hiroi et al., 1998; Chen et al., 2000; You et al., 2017). FOSB increases the levels of the NMDA receptor and play a critical role in molecular, electrophysiological, and behavioral adaptations to motor seizures (Hiroi et al., 1998). Also, FOSB expression is increased in the hippocampus after seizure activity in various models of epilepsy (Chen et al., 2000). Seizures and epileptiform activity have also been shown to increase FOSB expression robustly (You et al., 2017). We consider that although there is likely disruption in the glutamatergic neurotransmission in both GRS and iTLE, the specific signaling pathways and key molecules that are altered could differ in the two conditions.

4.3. Limitations

Given the limitations caused by the relatively small cohort of patients, multiple clinical characteristics need to be considered as potential confounding factors. These include sex, age, tumor location, grade, molecular status, and individual treatment schedules, among others. A larger cohort of future patients will be necessary to perform a multivariate analysis that accounts for all these features. Given the known remarkable heterogeneity of gliomas (see Supplementary Tables 1 and Fig. 2), merging glioma samples (GRS or non-GRS) of different grades, cell-types, and molecular markers together also significantly impacts the findings of our study. Further, in the current study, the tumor samples were obtained in bulk from the contrast-enhancing regions of the tumor, and we cannot assume that they contain 100 % glioma cells (likely includes peritumoral cells). Future prospective studies, including single-cell RNA-sequencing, are needed to clarify the contribution of tumoral and peritumoral cells in the genesis of seizures. The use of single-cell RNA-sequencing will also allow us to attribute the observations in DEG to the specific cell population both from tumoral and peritumoral cells. A second aspect that needs to be considered in future prospective studies is incorporating intracranial electroencephalogram (iEEG) recordings in these studies. Previous studies using long-term iEEG recordings have suggested that the seizure onset zone in GRS could be outside the glioma margins (Sweet et al., 2013; Mittal et al., 2016; Feyissa et al., 2018). Finally, functional in vitro and in vivo studies in model systems are needed to evaluate the effects of the DEGs identified here.

5. Conclusions

This study identifies genes and biological pathways that are altered in GRS. Specifically, alterations in acute phase signaling, inflammation-related processes, and glutamatergic neurotransmission may underlie the pathogenesis of GRS. Our findings also indicate that altered genes and pathways in GRS may differ from iTLE. Functional studies with a larger cohort and advanced genomic approaches are needed to confirm our findings further and elucidate the mechanisms of GRS.