Molecular characterization of high-grade glioma-associated seizures

Affiliations:

Authors: Lydia Guo^{1,2,^}, Rowan Barker-Clarke^{3,^}, Ryan G. Rilinger^1,2, Akshay Sharma², Nicolas R. Thompson^4,5, Josephine Volovetz², Mina Lobbous^1,3, Andrew Dhawan^1,3,6, Matthew M. Grabowski^2,3,6*

¹Cleveland Clinic Lerner College of Medicine, Case Western Reserve University, Cleveland, Ohio, United States

²Department of Neurosurgery, Cleveland Clinic, Cleveland, Ohio, United States

³Rose Ella Burkhardt Brain Tumor and Neuro-Oncology Center, Cleveland Clinic, Cleveland, Ohio, United States

⁴Department of Quantitative Health Sciences, Cleveland Clinic Research, Cleveland, Ohio, United States

⁵Neurological Institute Center for Outcomes Research & Evaluation, Cleveland Clinic, Cleveland, Ohio, United States

⁶Department of Cancer Biology, Cleveland Clinic Research, Cleveland, Ohio, United States

Running title: Molecular markers in glioma-associated seizures

* Corresponding author: Matthew Grabowski, 9500 Euclid Ave, Cleveland, OH 44195, 216-444-4390, grabowm2@ccf.org

^ Indicates co-first authors

Total manuscript word count

6,003

Abstract

Seizures occur in nearly half of all patients with high-grade gliomas, but few molecular markers have been identified as prognostic for glioma-associated seizures. We sought to examine the relationship between tumor molecular markers and glioma-associated seizures in patients with WHO grade 4 gliomas (glioblastoma, IDH-mutant astrocytoma). Amongst 950 patients diagnosed with grade 4 gliomas between 1999 and 2023, 414 (44%) patients experienced seizures. Tumor genomic characteristics were correlated with seizure incidence (before or after glioma diagnosis) and frequency in multivariable analyses. In multivariable analyses, chromosome 1p deletion (OR = 2.7, 95% CI [1.6, 4.4], p < 0.001), pathogenic IDH1 variants (OR = 3.1, 95% CI [1.4, 7.1], p = 0.033), and EGFR amplification (OR = 1.6, 95% CI [1.1, 2.2], p = 0.039) were all significantly associated with increased odds of seizures before glioma diagnosis. For an exploratory subset of 83 patients, we conducted whole exome sequencing of the tumor, but no specific variants were associated with seizure occurrence. In conclusion, chromosome 1p deletion, pathogenic IDH1 status, and EGFR amplification were significantly associated with seizures before glioma diagnosis. Future work to identify additional molecular markers for patients at greatest risk for tumor-associated epilepsy may improve morbidity in high-grade glioma.

Words: 199

Introduction

Gliomas are the most common primary malignant intra-axial tumors of the central nervous system in adults, and many patients experience seizures secondary to the glioma, termed glioma-associated seizures.¹ Around 45% of patients with WHO grade 3 or 4 gliomas present with seizures before glioma diagnosis and 20% present with seizures after diagnosis.² Seizures are highly detrimental to a patient’s quality of life and functional status, and anti-seizure medications (ASMs) can have side effects. Thus, new insights into the pathogenesis, risk-stratification, and management of glioma-associated seizures are needed.

Although nearly all patients receive perioperative ASM prophylaxis for craniotomies, there is a lack of standardized guidelines delineating duration of ASM prophylaxis and discontinuation in patients at risk for seizures.³ Molecular profiles obtained from glioma biopsies provide insight into therapeutic approaches, but these profiles provide little information on the prognosis of glioma-associated seizures.⁴ As such, there is an unmet need to better understand the prognostic potential of molecular markers in glioma-associated seizures.

Similar pathways likely drive both tumor progression and related seizures, as glioma progression is often accompanied by worsening seizures, but few genetic markers have been implicated in both glioma and seizure pathogenesis.^5,6 Pathogenic variants in isocitrate-dehydrogenase 1 (IDH1) may be involved in both glioma pathogenesis and associated seizures.^7–9 Indeed, preclinical data has identified a byproduct of pathogenic IDH1 variant metabolism that resembles glutamate, promoting glioma cell infiltration and excitatory conduction.⁸ Furthermore, IDH1 inhibitors have recently been shown to have antiepileptic properties both in preclinical models and low-grade gliomas.^10–12

However, relationships between other genomic alterations in gliomas and glioma-associated seizures remain inconclusive.⁴ According to a recent meta-analysis, pathogenic IDH1 variants but not MGMT promoter methylation nor loss of chromosome 1p/19q were associated with seizures before glioma diagnosis.⁴ Additionally, the relation of p53 expression and EGFR amplification to glioma-associated seizures were not assessed in the meta-analysis due to the limited number of studies.⁴ Furthermore, few studies have focused solely on the association between grade 4 gliomas and related seizures, especially since grade 4 gliomas are uniquely aggressive compared to other lower grades.^4,13

Therefore, we sought to characterize the association of the latest molecular markers to seizures in patients with grade 4 gliomas through a retrospective cohort study. While we have previously shown that seizure activity is associated with improved survival in patients with high-grade gliomas,⁹ it remains unclear as to which molecular factors drive seizure presentation. For this patient cohort, we analyzed how IDH1 mutation status, MGMT promoter methylation status, presence of EGFR amplification, deletion of chromosome 1p/19q, p53 expression, and Ki-67 immunohistochemistry were associated with both seizure incidence and frequency. For a subset of patients, we paired whole exome sequencing data with patient characteristics to examine common exome variants (copy number alterations, point mutations, gene fusions) involved in glioma-associated seizures. By doing so, we aim to provide an improved understanding of prognostic factors in glioma-associated seizures to guide postoperative care.

Materials and Methods

Patient Cohort

Patients (N = 950) included were diagnosed with grade 4 gliomas by WHO 2021 criteria from 1999 to 2023 at the Cleveland Clinic in Cleveland, Ohio, USA. Diagnosis was based on histopathological analysis of tumor tissue biopsy by board-certified pathologists. Data were collected via retrospective review of patient charts from the electronic health record system.

The Institutional Review Board (IRB) of Cleveland Clinic approved this study (IRB #18–937).

Informed consent of participants was waived due to the retrospective and non-invasive design.

Our research was conducted in accordance with the Declaration of Helsinki. Clinical trial number: not applicable. Additional details can be found in Rilinger et al., 2024.⁹

For patients diagnosed at the Cleveland Clinic, board-certified neuropathologists examined fresh, frozen tissue with hematoxylin and eosin staining for initial diagnosis and subsequent analyses were done on formalin-fixed, paraffin-embedded tissue. Immunohistochemical stains were used to examine for the common pathological IDH1 variant (R132H), p53 nuclear positivity, and Ki-67 proliferative index. Molecular studies using fluorescence in situ hybridization (FISH) were used to examine for loss of chromosome 1p/19q and EGFR amplification. Bisulfite pyrosequencing was used to determine MGMT promoter methylation status.

Seizure Characteristics

Glioma-associated seizures were defined as any occurrence of seizure related to glioma. Electroencephalogram (EEG) data was not utilized for diagnosis. All data on seizure activity including frequency were acquired from retrospective review of notes from board-certified neurologists. Glioma diagnosis was defined as the date of first surgery for tissue sampling related to glioma.

Patients were stratified into three groups: 1) None, 2) Early Seizure (first seizure occurred within 30 days of glioma diagnosis), and 3) Late Seizure (first seizure occurred after 30 days from glioma diagnosis). For patients who progressed from lower grade (grade 1–3) gliomas, first surgery was defined as first biopsy that revealed grade 4 glioma.

Initial seizure frequency was defined as number of seizures that occurred within 30 days of date of first surgery. Seizure frequency was captured over the first six months after diagnosis and quantified by dividing the total number of seizures by six (average number of seizures / month).

Statistical Analysis

The patient sample was summarized by descriptive statistics both overall and stratified by seizure incidence (No Seizure, Early Seizure, Late Seizure). Mean with standard deviation or median with interquartile range was used for continuous variables and frequency with percentage was used for categorical variables. Group comparisons used one-way analysis of variance (ANOVA) or Kruskal-Wallis test for continuous variables and chi-squared or Fisher’s exact test for categorical variables.

Multivariable analyses were used to examine seizure occurrence (No Seizure, Early Seizure, Late Seizure). In all models, covariates included age, sex, gross total resection (vs. all other surgery types), laterality, location (frontal lobe, parietal lobe, temporal lobe, occipital lobe, other), Karnofsky performance status (KPS) at diagnosis, radiation therapy, chemotherapy (cytotoxic), and chemotherapy (biological target). Variance inflation factors > 5 were used to indicate multicollinearity.

To explore the relationship between molecular markers and glioma-associated seizures, four outcomes were examined using multivariable models:

Early Seizure occurrence (binary): All patients were analyzed with a multivariable logistic regression model.

Initial seizure frequency (binary): Patients in Early and Late Seizure groups were analyzed with a multivariable logistic regression model. The dependent variable was any seizure frequency (0 vs. 1+).

Average seizure frequency over first six months after diagnosis (binary): Patients in Early and Late Seizure groups were analyzed with a multivariable logistic regression model. The dependent variable was any seizure frequency (0 vs. 1+).

Late Seizure occurrence (time dependent outcome): Patients in No and Late Seizure groups were analyzed with a multivariable cause-specific Cox proportional hazard model. The dependent variable was time from glioma diagnosis to first seizure occurrence, and death was treated as a competing risk. Patients who died without experiencing seizure were censored at date of death. Surviving patients who did not experience seizure were censored at date of last follow-up.

For all outcomes, each molecular marker was examined as the independent variable of interest in separate models. The molecular markers examined included IDH1 (R132H), p53 nuclear positivity, Ki-67 proliferative index, chromosome 1p/19q, EGFR amplification, and MGMT promoter methylation. Multivariable models were also fit including all molecular markers as predictors in the same model (Supplementary Table 2). Missingness of the molecular markers ranged from 4.1% (Ki-67) to 40.2% (MGMT promoter methylation). Missing data were handled in (Supplementary) Table 2A without imputation and in (Supplementary) Table 2B with multiple imputation.

Computations were conducted in R (version 4.3.1). All statistical tests were two-sided and p-values < 0.05 were considered statistically significant. For multivariable models, we used complete case analysis and corrected for multiple comparisons using Holm’s method. Univariable analyses were not corrected for multiple testing.

Next-Generation Sequencing (NGS)

For a subset of 101 patients diagnosed with grade 4 gliomas between 2015 and 2023, fresh, frozen tissue was sent for NGS (Caris Life Sciences, Phoenix, AZ). Of 101 samples, the quality of 18 samples was insufficient for NGS sequencing. Whole exome sequencing (WES) was reported on the remaining 83 samples for a proprietary brain cancer gene panel. Each report describes copy number alterations (none, intermediate, amplified, deleted), pathogenic variants (none, benign, likely benign, variant of uncertain significance [VUS], likely pathogenic, pathogenic), and fusions (none, unclassified, pathogenic isoform, pathogenic fusion). Copy number alterations (CNAs) were reported for a set of 138 genes, single nucleotide variants (SNVs) for a set of 140 genes, and gene fusion events for 334 genes. Scores were also reported for genomic loss of heterozygosity (gLOH) (high, low, indeterminate), microsatellite instability (MSI) (high, stable, indeterminate), and estimates of tumor mutation burden (TMB) (n per mb). Further details on Caris scoring methods are described elsewhere.¹⁴

For this cohort, we described specific exome variants and gene-level variant frequencies (CNAs, SNVs, and fusions). Singleton and doubleton masks were used for rare variant burden analysis.¹⁵ Singleton variant refers to an exome variant that only appeared in one sample, and doubleton variant refers to an exome variant that appeared in two samples. R (version 4.4.0) was used for the statistical analysis and visualization of NGS results. Spearman’s rank correlation, ρ, was computed to test for rank correlations between exome variants (CNAs, fusions, SNVs), tumor mutational burden (TMB), and patient characteristics (age, KPS at diagnosis, average seizure frequency six months after diagnosis).

Hierarchical clustering and linear discriminant analysis with bootstrapping were used to test for associations between exome variants and seizure incidence. Rare variant burden testing was carried out using Fisher’s exact test for categorical variables. The R packages ComplexHeatmap (version 2.20.0), ggcorrplot (version 0.1.4.1), GGally (version 2.2.1) and gpairs (version 1.3.3) were used for further visualization.¹⁶ Lollipop mutation plots were generated using the Lollipops package (version 1.7.2).¹⁷ The code for this analysis is archived at DOI: 10.5281/zenodo.16990424.

Results

Baseline patient characteristics

Of 950 patients with grade 4 gliomas, 536 (56.4%) had No seizures, 261 (27.5%) had Early seizures (within 30 days of diagnosis), and 153 (16.1%) had Late seizures (beyond 30 days from diagnosis), (Table 1). The average age was 61 years, and 37% of patients identified as female. Patients with No seizures were older than patients with Early and Late seizures (mean 64 vs. 58 vs. 57 years respectively, p < 0.001). Additionally, patients with Early seizures had greater KPS at diagnosis than patients with No and Late seizures (median 90 vs. 80 vs. 80 respectively, p < 0.001).

Table 1

Baseline characteristics for study cohort of patients diagnosed with WHO grade 4 gliomas. Patients were stratified by seizure status, with patients in No Seizure (n = 536), Early Seizure (n = 261), or Late Seizure (n = 153) groups. Differences between groups were analyzed using one-way ANOVA or Kruskal-Wallis tests. Categorical variables were assessed by χ² or Fisher’s exact tests. Significant p-values determined at α = 0.05 were denoted by *. See Supplementary Table 1 for full baseline characteristics. Abbreviations: IDH, Isocitrate dehydrogenase. MGMT, O⁶-methylguanine DNA methyltransferase. EGFR, Epidermal growth factor receptor.
	All Patients		No Seizures		Early Seizures		Late Seizures		P-value
	N	Statistic	N	Statistic	N	Statistic	N	Statistic
Deletion 1p	873	89 (10.2%)	488	44 (9.0%)	244	37 (15.2%)	141	8 (5.7%)	0.005*
Deletion 19q	872	103 (11.8%)	487	59 (12.1%)	244	30 (12.3%)	141	14 (9.9%)	0.749
Pathogenic IDH1 variant	600	37 (6.2%)	339	14 (4.1%)	150	19 (12.7%)	111	4 (3.6%)	< 0.001*
MGMT promoter methylation
No	573	333 (58.1%)	321	191 (59.5%)	153	83 (54.2%)	99	59 (59.6%)	0.383
Yes		235 (41.0%)		126 (39.3%)		69 (45.1%)		40 (40.4%)
Borderline		4 (0.7%)		4 (1.2%)		0 (0.0%)		0 (0.0%)
Indeterminate		1 (0.2%)		0 (0.0%)		1 (0.7%)		0 (0.0%)
ki-67 expression	905	30 (20, 50)	508	30 (20, 50)	253	30 (19, 50)	144	30 (20, 50)	0.968
EGFR amplification	856	332 (38.8%)	481	172 (35.8%)	239	110 (46.0%)	136	50 (36.8%)	0.025*
p53 expression	858	15 (10, 50)	484	15 (10, 50)	240	10 (10, 50)	134	15 (10, 47.5)	0.570
Initial seizure frequency
Median (IQR)	407	1 (0, 1)			255	1 (1, 2)	152	0 (0, 0)	< 0.001*
0	407	149 (36.6%)			255	18 (7.1%)	152	131 (86.2%)	< 0.001*
1		165 (40.5%)				154 (60.4%)		11 (7.2%)
2+		93 (22.9%)				83 (32.5%)		10 (6.6%)

	Early Seizure (Yes vs. No) Logistic Regression			Time to Late Seizure Cox Proportional Hazards			Initial Seizure Frequency (1 + vs. 0) Logistic Regression			Average Seizure Frequency Over 1st 6 months (Any vs. 0) Logistic Regression
	N	Odds Ratio (95% CI)	P-value	N	Hazard Ratio (95% CI)	P-value	N	Odds Ratio (95% CI)	P-value	N	Odds Ratio (95% CI)	P-value
Deletion 1p	831	2.66 (1.61, 4.37)	< 0.001*	592	0.68 (0.31, 1.49)	1.000	369	1.84 (0.89, 4.04)	0.659	345	0.67 (0.33, 1.34)	1.000
Deletion 19q	830	1.04 (0.64, 1.67)	1.000	591	0.77 (0.45, 1.32)	1.000	369	1.25 (0.62, 2.61)	1.000	345	1.01 (0.51, 2.00)	1.000
Pathogenic IDH1 variant	555	3.12 (1.40, 7.05)	0.033*	409	0.53 (0.18, 1.52)	1.000	250	2.24 (0.75, 7.19)	0.798	232	0.19 (0.06, 0.58)	0.037*
MGMT promoter methylation	526	1.21 (0.80, 1.83)	1.000	378	0.75 (0.47, 1.19)	1.000	240	0.78 (0.44, 1.38)	1.000	223	0.83 (0.47, 1.48)	1.000
Ki-67 expression (per 10% increase)	853	1.00 (0.93, 1.07)	1.000	607	1.02 (0.94, 1.10)	1.000	379	0.93 (0.84, 1.03)	0.798	353	0.93 (0.84, 1.03)	0.967
EGFR amplification	810	1.56 (1.12, 2.17)	0.039*	576	0.85 (0.58, 1.23)	1.000	358	1.54 (0.96, 2.47)	0.514	336	1.18 (0.75, 1.86)	1.000
p53 expression (per 10% increase)	808	0.95 (0.89, 1.00)	0.256	575	0.97 (0.91, 1.03)	1.000	356	0.97 (0.89, 1.06)	1.000	333	1.01 (0.92, 1.09)	1.000
A)
B)

10% of all patients had chromosome 1p deletion, 12% had chromosome 19q deletion, 6% had the pathogenic IDH1 variant, 41% had MGMT promoter methylation, and 39% had EGFR amplified. For all gliomas, median expression for Ki-67 was 30% and for p53 was 15%. Seizure incidence also differed depending on the patient’s tumor location, history of radiation therapy, and history of chemotherapy treatment (Supplementary Table 1).

Chromosome 1p deletion, pathogenic IDH1 variants, and EGFR amplification were associated with Early seizure incidence

In the full cohort, we identified molecular markers associated with seizure incidence via univariable analyses (Table 1). Chromosome 1p deletion was significantly more common among patients with Early seizures (15%) than Late (6%) or No (9%) seizures (p = 0.005). We also identified more pathogenic IDH1 variants in patients with Early seizures (13%) versus Late (4%) or No (4%) seizures (p < 0.001). EGFR amplification was also more frequent among patients with Early seizures (46%) than Late (37%) or No (36%) seizures (p = 0.025). We did not find that seizure incidence differed depending on molecular status of chromosome 19q deletion, MGMT promoter methylation, Ki-67 expression, and p53 expression.

We also examined associations between common pathological variants and seizure incidence or frequency via multivariable models. For all outcomes, pathological variant association was examined separately in individual models (Table 2) and together (Supplementary Table 2). We handled missing data without imputation (Table 2A, Supplementary Table 2A) and with multiple imputation (Table 2B, Supplementary Table 2B).

Table 2

Multivariable analyses examining the association between pathological variants and seizure incidence or seizure frequency. Separate models were fit for each pathological variant. All models were adjusted for the following covariates: age, sex, gross total resection, laterality, location (frontal lobe, parietal lobe, temporal lobe, occipital lobe, other), KPS at diagnosis, radiation therapy, chemotherapy (cytotoxic), and chemotherapy (biological target). Significant p-values were determined at α = 0.05 denoted by *. Holm’s method was used to correct for multiple comparisons. Missingness of the molecular markers ranged from 4.1% (Ki-67) to 40.2% (MGMT promoter methylation). Missing data were handled in Table 2A) without imputation and 2B) with multiple imputation.
		Early Seizure (Yes vs. No) Logistic Regression			Time to Late Seizure Cox Proportional Hazards			Initial Seizure Frequency (1 + vs. 0) Logistic Regression			Average Seizure Frequency Over 1st 6 months (Any vs. 0) Logistic Regression
	N	Odds Ratio (95% CI)	P-value	N	Hazard Ratio (95% CI)	P-value	N	Odds Ratio (95% CI)	P-value	N	Odds Ratio (95% CI)	P-value
Deletion 1p	950	2.50 (1.53, 4.08)	0.002*	689	0.72 (0.35, 1.48)	1.000	950	1.50 (0.89, 2.54)	0.759	950	0.83 (0.50, 1.40)	1.000
Deletion 19q	950	0.99 (0.62, 1.60)	1.000	689	0.76 (0.44, 1.31)	1.000	950	1.01 (0.62, 1.65)	1.000	950	0.95 (0.56, 1.62)	1.000
Pathogenic IDH1 variant	950	2.69 (1.25, 5.80)	0.059	689	0.53 (0.20, 1.46)	1.000	950	1.60 (0.74, 3.48)	0.916	950	0.27 (0.09, 0.75)	0.092
MGMT promoter methylation	950	1.20 (0.82, 1.74)	1.000	689	0.73 (0.46, 1.15)	1.000	950	1.04 (0.72, 1.49)	1.000	950	0.93 (0.64, 1.35)	1.000
ki-67 expression (per 10% increase)	950	0.99 (0.92, 1.07)	1.000	689	1.01 (0.94, 1.09)	1.000	950	0.98 (0.91, 1.05)	1.000	950	0.97 (0.90, 1.05)	1.000
EGFR amplification	950	1.55 (1.12, 2.15)	0.049*	689	0.92 (0.64, 1.32)	1.000	950	1.51 (0.99, 2.28)	0.382	950	1.13 (0.79, 1.61)	1.000
p53 expression (per 10% increase)	950	0.95 (0.90, 1.01)	0.465	689	0.97 (0.91, 1.03)	1.000	950	0.95 (0.90, 1.02)	0.759	950	0.99 (0.93, 1.06)	1.000

Similar to univariable findings, patients with Early seizures had greater odds of chromosome 1p deletion (OR = 2.7, 95% CI [1.6–4.4], p < 0.001), pathogenic IDH1 variant (OR = 3.1, 95% CI [1.4–7.1], p = 0.033), and EGFR amplification (OR = 1.6, 95% CI [1.1–2.2], p = 0.039) via multivariable logistic regression (Table 2A). Furthermore, patients with pathogenic IDH1 variants were significantly less likely to have seizures over their first six months after glioma diagnosis via multivariable logistic regression (OR = 0.2, 95% CI [0.06–0.6], p = 0.037), (Table 2A). No other associations were found between other molecular markers and seizure-related outcomes.

Whole exome sequencing revealed diverse variant landscape within grade 4 gliomas

We next aimed to characterize a subset of 83 patients via whole exome sequencing (WES) to identify molecular variants within exonic gene regions. First, we explored how baseline patient characteristics for this subset differed by seizure incidence (Table 3). Male patients were more likely to have Early (79%) seizures than No (50%) or Late (39%) seizures, and female patients were more likely to have Late (61%) or No (50%) seizures than Early (21%) seizures, and these differences were statistically significant by Pearson’s Chi-squared test (p = 0.011). Otherwise, patients did not differ in other baseline characteristics by seizure incidence.

Table 3

Baseline characteristics for patients with NGS. Patients with whole exome sequencing (n = 83) were stratified by seizure incidence (No Seizure, Early Seizure, Late Seizure). Data reported as median with interquartile range or counts with percentages. Continuous variables were analyzed with Kruskal-Wallis rank sum test, and categorical variables were analyzed with Pearson’s Chi-squared test or Fisher’s exact test. Significant p-values were determined at α = 0.05 denoted by *. Abbreviations: CNA, Copy number alteration. IDH, Isocitrate dehydrogenase. KPS, Karnofsky performance status. LOH, Loss of heterozygosity. MGMT, O⁶-methylguanine DNA methyltransferase. MSI, Microsatellite instability. SNV, Single nucleotide variant.
Characteristic	N	No Seizure, N = 36¹	Early Seizure, N = 29¹	Late Seizure, N = 18¹	p-value²
Genes with Fusions	83	2.0 (0.8, 4.0)	2.0 (1.0, 5.0)	2.0 (0.0, 3.8)	0.6
Genes with CNAs	83	2 (0, 5)	4 (2, 8)	1 (0, 6)	0.13
Genes with SNVs	83	4 (3, 5.25)	5 (3, 6)	5 (3.25, 6)	0.7
MSI	83				> 0.9
High		0 (0%)	1 (3.4%)	0 (0%)
Indeterminate		2 (5.6%)	1 (3.4%)	1 (5.6%)
Stable		34 (94%)	27 (93%)	17 (94%)
LOH	83				> 0.9
High		1 (2.8%)	1 (3.4%)	1 (5.6%)
Indeterminate		1 (2.8%)	0 (0%)	0 (0%)
Low		34 (94%)	28 (97%)	17 (94%)
TMB Numeric	80	3.00 (2.00, 5.00)	3.00 (2.00, 5.25)	4.00 (3.00, 5.00)	0.5
Age	83	58 (52, 70)	59 (53, 65)	55 (45, 68)	0.5
Sex	83				0.011*
Male		18 (50%)	23 (79%)	7 (39%)
Female		18 (50%)	6 (21%)	11 (61%)
KPS at Diagnosis	79	80 (70, 90)	85 (80, 90)	80 (70, 90)	0.5
Race	83				0.5
White		31 (86%)	28 (97%)	15 (83%)
Black		1 (2.8%)	1 (3.4%)	1 (5.6%)
Asian		1 (2.8%)	0 (0%)	1 (5.6%)
Other		3 (8.3%)	0 (0%)	1 (5.6%)
Pathogenic IDH1 Variant	82				> 0.9
No		32 (89%)	25 (89%)	16 (89%)
Yes		4 (11%)	3 (11%)	2 (11%)
MGMT Promoter Methylated	80				0.2
No		20 (57%)	9 (32%)	7 (41%)
Yes		15 (43%)	18 (64%)	10 (59%)
P53% Reactivity	66	18 (10, 76)	14 (10, 39)	18 (12, 50)	0.9
Unknown		8	7	2
Location (Frontal Lobe)	83				0.4
No		23 (64%)	14 (48%)	9 (50%)
Yes		13 (36%)	15 (52%)	9 (50%)
Location (Parietal Lobe)	83				> 0.9
No		25 (69%)	20 (69%)	13 (72%)
Yes		11 (31%)	9 (31%)	5 (28%)
Location (Occipital Lobe)	83				0.3
No		29 (81%)	27 (93%)	17 (94%)
Yes		7 (19%)	2 (6.9%)	1 (5.6%)
Location (Temporal Lobe)	83				0.4
No		20 (56%)	20 (69%)	13 (72%)
Yes		16 (44%)	9 (31%)	5 (28%)
Location (Other)	83				0.040*
No		29 (81%)	29 (100%)	15 (83%)
Yes		6 (17%)	0 (0%)	3 (17%)
¹Median (IQR); n (%)
²Kruskal-Wallis rank sum test; Fisher's exact test; Pearson's Chi-squared test

Next, we examined the mutational landscape of exome variants (Table 4). Within 83 samples, we observed 1,456 exome variants including 488 SNVs (404 unique SNVs), 618 CNAs (165 unique changes), and 350 gene fusions (301 unique fusions). The most common SNVs were missense (61%) followed by splice site variants (19%). While patient tumors were highly varied in frequency of exome variants (Supplementary Fig. 1), most tumors were microsatellite-stable (94%), had low levels of loss of heterozygosity (95%), and had low tumor mutational burden (Table 3). Across samples, SNVs were observed in 123 of 140 targeted genes, and CNAs were observed in 132 of 138 targeted genes. Most of these changes were unique to a specific patient, as there were 396 singleton SNVs, 298 singleton fusion variants, and 59 singleton CNAs. No collinearity was observed between exome variant frequencies.

Table 4

Mutational landscape of variants in whole-exome sequencing analysis. 83 glioma samples were analyzed via whole-exome sequencing with 1456 total exome variants observed. Gene frequency was stratified as singletons (appearing in 1 patient), doubletons (appearing in 2 patients), or present in 3 or more samples. Most frequently altered genes (by sample count) were identified within variant subtypes of CNAs, SNVs, and gene fusions.
	Total Variants					Singleton Genes (%)			Doubleton Genes(%)		Common Genes (≥ 3 samples)
Copy number alteration			618	133		41 (30.8)			15 (11.2)		77
Fusion			350	252		244 (96.5)			6 (2.7)		2
Single nucleotide variant			488	123		57 (46.0)			24 (18.5)		43
Genes with most copy number alterations (by sample count)
Gene	Amplified				Intermediate			No Change				Deletion
EGFR	18				6			59				-
STK11	3				13			66				1
MEF2B	2				13			68				-
MAP2K2	-				13			70				-
PDCD1	2				11			70				-
Most common genes with single nucleotide variants (by sample count)
Gene		Positive					VUS			No mutation
TERT		59					1			23
TP53		33					-			50
PTEN		20					6			57
EGFR		14					3			66
NF1		8					5			70
Classes of single nucleotide variants
Frameshift		12
Indel		42
Missense		299
Nonsense		37
Synonymous		7
Splice Site		91
Most frequent gene fusions (by sample)
Gene		Pathogenic Isoform					Unclassified			No fusion detected
EGFR		19					-			64
TIMM23B		-					17			66
Abbreviations: CNA, Copy number alteration. SNV, Single nucleotide variant.
Abbreviations: CNA, Copy number variant. MGMT, O⁶-methylguanine DNA methyltransferase. SNV, Single nucleotide variant. TMB, tumor mutational burden. VUS, Variant of uncertain significance.
Abbreviations: CNA, Copy number variant. KPS, Karnofsky performance status. SNV, Single nucleotide variant. TMB, tumor mutational burden. VUS, Variant of uncertain significance.

Pathogenic variants in EGFR were common

We depicted the top exome variants with a heatmap (Fig. 1). The most common CNAs were EGFR, STK11, MEF2B, MAP2K2, and PDCD1, and most CNAs resulted in either amplification or intermediate changes with few deletions (Table 4). The most common genes with SNVs were TERT, TP53, PTEN, EGFR, and NF1, which were classified as positive (previously pathogenic variant) or VUS. In our cohort, the two most common SNVs were mutually exclusive mutations in the TERT promoter region: 1) c.-124C > T occurred in 51 samples (61%) and 2) c.-146C > T occurred in 10 samples (12%). We also illustrated select common amino acid substitutions with lollipop genomic plots including R273C/H in TP53 (n = 4) and A289D/V in EGFR (n = 4), (Supplementary Fig. 2).

Fig. 1

Complex landscape of whole exome sequencing variants. The most common (top 10) genes with CNAs, mutations (SNVs), and fusions present in three or more samples are shown in the heatmap. Gene names (x-axis) are stratified by exome variant type and plotted against patient samples (y-axis) which are stratified by patient seizure presentation (Early vs. Late vs. No Seizure groups). Most CNAs were gains (intermediate or amplification). CNAs, SNVs, and pathogenic gene fusion isoforms in EGFR were common.

We identified two recurrent gene fusions in our cohort: EGFR and TIMM23B. All 19 tumors (23%) with EGFR fusions exhibited pathogenic EGFRvIII fusions and some had additional EGFR gene fusions. TIMM23B fusions were present in 17 tumors (20%) as an unclassified isoform between TIMM23B and exon 2 or exon 3 of TIMM23.

Most common single nucleotide variants were mutually exclusive

Mutual exclusivity of variants can reveal distinct evolutionary pathways, and several SNVs were found to be mutually exclusive with other pathogenic variants. Starting with IDH1, pathogenic IDH1 SNVs (R132H/C) occurred in 10 of 83 samples (12%). These IDH1 SNVs were significantly associated with the presence of pathogenic TP53 SNVs (p = 0.002). Furthermore, IDH1 SNVs were mutually exclusive with EGFR SNVs, fusions, and CNAs. IDH1 SNVs were also mutually exclusive with other pathogenic SNVs in TERT (p < 0.00001), PTEN, and NF1. Pathogenic NF1 SNVs were mutually exclusive with pathogenic TERT SNVs (p = 0.020). While EGFR SNVs were associated with decreased odds of EGFR amplification (p = 0.006), EGFR fusions were associated with increased odds of EGFR amplification (p < 0.00001).

Individual exome variants were not associated with seizure incidence

To investigate whether molecular variants are associated with seizure presentation, we performed combinatorial linear discriminant analysis (Supplementary Fig. 3). However, we found that no combinations of two common exome variants were significantly associated with seizure incidence (No Seizure, Early Seizure, Late Seizure). This finding was confirmed via Pearson’s Chi-squared test (p = 0.2).

We then tested whether the number of exome variants was associated with seizure incidence. We found that patients with Early seizures had more CNAs than patients with Late and No seizures (median 4 vs. 1 vs. 2 respectively), (Table 3). We also compared exome variant frequency based on seizure incidence (Supplementary Fig. 1).

Tumor mutational burden was associated with patient clinical characteristics

While we did not find an association between molecular variants and seizure presentation, we aimed to investigate whether molecular variants were associated with seizure frequency and other clinical characteristics. We used Spearman’s rank correlation to compare exome variants with select patient characteristics both in aggregate and stratified by seizure incidence (Fig. 2). Among patients with Late seizures, older age (ρ = 0.65, p < 0.01) and fewer SNVs (ρ = -0.54, p < 0.05) were associated with more frequent seizures in the first 6 months after glioma diagnosis.

Fig. 2

Seizure incidence groupings show distinct relationships between variant burden and clinical characteristics. Variables included exome variants (CNAs, mutations or SNVs, gene fusions), TMB, and patient characteristics (age, KPS at diagnosis, average seizure frequency six months after diagnosis). Patients were stratified by seizure incidence and color-coded (No Seizure (0) = red, Early Seizure (1) = green, Late Seizure (2) = blue). Numerical variables log transformed for visualization. Comparison variables were listed across top and right figure borders. Density plots, scatter plots, and histograms depict data distributions. Significant p-values were determined at α = 0.05 denoted by * (p < 0.05), ** (p < 0.01), *** (p < 0.001).

When examining the relationship between variant frequencies and other clinical variables, estimated TMB was, as expected, significantly positively correlated with SNVs in targeted genes (ρ = 0.63, p < 0.001). While increased SNVs were negatively correlated with CNAs (ρ = -0.24, p < 0.05) for all patients, this overall relationship was primarily driven by a stronger association in the No Seizure group (ρ = -0.53, p < 0.01).

Higher KPS at diagnosis was significantly associated with lower TMB (ρ = -0.24, p < 0.05). This finding was especially prominent among patients with Early seizures (ρ = -0.5, p < 0.01). For patients without seizures, older age was significantly correlated with higher TMB (ρ = 0.44, p < 0.01).

Discussion

Our retrospective cohort study examined 950 patients with grade 4 gliomas to comprehensively characterize the association between tumor molecular markers and glioma-associated seizures. We demonstrated that pathogenic IDH1 variants, EGFR amplification, and chromosome 1p deletion were associated with Early seizure incidence (Table 2A). Although our WES analysis of 83 patients did not identify any specific exome variants associated with seizure activity (Supplementary Fig. 3), we observed high frequencies of pathogenic variants in genes previously implicated in grade 4 glioma pathogenesis including TERT, TP53, PTEN, EGFR, NF1, and IDH1 (Table 3).^18–20 While other studies have similarly examined genes implicated in glioma-associated seizures,^21–23 our study contributes further information on how exome variants relate to patient clinical characteristics including seizure activity.

Variants in IDH1 have been extensively characterized in the context of patient survival, but recent studies also show an association between IDH1 (R132H) and seizure activity. Such studies found that pathogenic IDH1 variants in low-grade gliomas were associated with Early seizures, but evidence for high-grade gliomas remained inconclusive.^8,24–26 Our study demonstrates that pathogenic IDH1 variants are indeed associated with Early seizures in grade 4 gliomas (Table 2A). Additionally, we found that IDH1 (R132H) was associated with fewer seizures over the first six months after glioma diagnosis, suggesting that IDH1 (R132H) may be implicated in early seizure activity but not later in the disease course (Table 2A).

EGFR amplification occurs in around half of glioblastomas and has been well-studied as a driver of oncogenesis.¹⁹ Interestingly, our study found a strong association between EGFR amplification and Early seizure activity that has not yet been characterized in grade 4 gliomas (Table 2). One retrospective study has similarly linked EGFR amplification with Early seizures though only in grade 3 gliomas.²⁷ A preclinical study found that EGFR is involved in glutamate-driven cell proliferation in glioblastoma cells, which may signify a role for EGFR in both cell proliferation and neuronal excitability.²⁸ In our WES cohort, we identified EGFR as one of the most common exome variants including EGFR A289D/V and EGFRvIII fusions (Table 4), (Supplementary Fig. 2D). Both EGFR A289D/V and EGFRvIII fusions have been well-characterized as drivers of glioblastoma pathogenesis.²⁹ Our study corroborates EGFR A289D/V and EGFRvIII fusions as key players in grade 4 gliomas and adds further evidence that EGFR amplification is associated with Early seizure activity.

In addition to IDH1 (R132H) and EGFR amplification, our study found that chromosome 1p deletion was associated with Early seizure incidence (Table 2). This finding opposes prior evidence that chromosome 1p/19q co-deletion has no association with glioma-related seizures.^30,31 As chromosome 1p/19q co-deletion is characteristic of oligodendrogliomas, the importance of chromosome 1p deletion in grade 4 gliomas remains uncertain. Our study also contrasts with prior literature as we did not find an association between seizure activity and chromosome 19q deletion, MGMT promoter methylation, Ki-67 proliferative index, nor p53 expression (Table 2).⁴

Our study identified SNVs that corroborate existing literature findings regarding their role in glioma pathogenesis.^32–35 These findings emphasize our cohort as representative of grade 4 glioma patients in literature. However, our study uniquely identified fusions between TIMM23B and exon 2 or exon 3 of TIMM23 as a common gene fusion in our WES cohort (Table 4). Intriguingly, translocase of inner mitochondrial membrane 23 (TIMM23B) has not yet been characterized in gliomas. A recent study shows that TIMM23B is essential for mitochondrial function and is regulated by the GABP transcription factor.³⁶ A related gene, TIMM44, has been found to be upregulated in mitochondria of glioma cells, and downregulation reduced glioma cell proliferation in a murine model.³⁷ As such, TIMM23B represents a new target for future preclinical studies to examine in the context of glioma pathogenesis.

When exome variants were compared with patients’ clinical characteristics, we found that lower KPS at diagnosis was associated with higher TMB (Fig. 2). Interestingly, higher TMB has previously been associated with shorter survival in patients with gliomas.³⁸ Furthermore, different associations between age and TMB were identified when stratified by seizure incidence, suggesting that factors intrinsic to the tumor microenvironment may differ among patients in the No, Early, and Late seizure groups. Studies have found common drivers of tumorigenesis and epileptogenicity, and ASMs may contribute to differences observed between patients with and without seizures.³⁹ Additionally, postoperative changes may contribute to the pathogenesis of Late seizures that do not apply to Early seizures.⁴⁰

While WES provides unique insight into exome variants within our cohort, there remains great heterogeneity in molecular characterization, and single-cell techniques may offer even stronger resolution into the molecular drivers of grade 4 glioma pathogenesis and associated seizures. Due to the sample size of our WES cohort, we had very limited power to detect an association between specific exome variants and seizure activity. Furthermore, we did not distinguish between somatic and germline mutations in the WES analysis, as we did not have access to matched non-tumor samples from patients. Another limitation is that we did not account for ASM use and patient compliance in our assessment of seizure activity, so our estimation of seizure activity may be confined in accuracy. Of consideration, nearly all patients receive perioperative ASM prophylaxis for craniotomies,⁴¹ and such perioperative ASM administration may contribute to postoperative seizure control even for patients in the No Seizure group.

A particular strength of our study is the large sample size, but due to the time frame of our study (1999–2023), we do not have information on IDH1 status for all samples in the larger cohort (N = 950), as the clinical relevance of IDH1 was not well-known until the past decade. As the WHO updated glioma classification guidelines in 2021, our study can only be generalized to all grade 4 gliomas when considering the full cohort.⁴² Furthermore, our study was conducted at a single institution with the majority of patients identifying as White, which limits our study’s generalizability to other racial and ethnic groups.

Overall, as a large-scale retrospective analysis of grade 4 glioma-associated seizures, our study offers strong evidence that suggests an association between patients’ seizure activity with their tumor molecular profiles. These findings may be validated in future prospective studies aimed at identifying predictors of seizure occurrence and burden in patients with high-grade gliomas.

Words: 3,967

FUNDING:

There were no sources of funding for this research.

CONFLICT OF INTEREST:

There are no conflicts of interest to be reported.

AUTHORSHIP:

L.G., R.B.-C., A.S., M.L., A.D., and M.M.G. conceived of the project goals and design. R.B.-C. and N.T. conducted formal analysis and visualization of data. L.G., R.G.R., A.S., J.V. performed retrospective review of patient records and curated data. L.G. prepared the initial manuscript draft. L.G., R.B.-C., R.G.R., A.S., N.T., M.L., A.D., and M.M.G. contributed to subsequent drafts and manuscript revisions. M.L., A.D., M.M.G. provided supervision of project planning and administration.

All authors read and approved of the final manuscript.

Data Availability

Code used to produce figures is available at github.com/rbarkerclarke. Deidentified exome and clinical data will be made available upon reasonable request.

Acknowledgement

We would like to thank the Neurological Institute CORE at the Cleveland Clinic Foundation for their biostatistical support. Part of the data in this manuscript was presented at the 2023 SNO/ASCO CNS Cancer Conference and published as an abstract in Neuro-Oncology Advances. Another part of the data was presented at the 2025 American Academy of Neurology Annual Meeting and published as an abstract in Neurology.

Electronic Supplementary Material

Below is the link to the electronic supplementary material

Supplementary Material 1

Author Contribution

REFERENCES

Fisher, R. S. et al. ILAE official report: a practical clinical definition of epilepsy. Epilepsia 55, 475–482 (2014).

Englot, D. J., Chang, E. F. & Vecht, C. J. Epilepsy and brain tumors. Handb Clin Neurol 134, 267–285 (2016).

Greenhalgh, J., Weston, J., Dundar, Y., Nevitt, S. J. & Marson, A. G. Antiepileptic drugs as prophylaxis for postcraniotomy seizures. Cochrane Database Syst Rev 2020, CD007286 (2020).

Song, L., Quan, X., Chen, C., Chen, L. & Zhou, J. Correlation Between Tumor Molecular Markers and Perioperative Epilepsy in Patients With Glioma: A Systematic Review and Meta-Analysis. Frontiers in Neurology 12, (2021).

Snijders, T. J., Berendsen, S., Seute, T. & Robe, P. A. Glioma-associated epilepsy: toward mechanism-based treatment. Translational Cancer Research 6, (2017).

BERNTSSON, S. G., MALMER, B., BONDY, M. L., QU, M. & SMITS, A. Tumor-associated epilepsy and glioma: Are there common genetic pathways? Acta Oncol 48, 955–963 (2009).

Cohen, A., Holmen, S. & Colman, H. IDH1 and IDH2 Mutations in Gliomas. Curr Neurol Neurosci Rep 13, 345 (2013).

Chen, H. et al. Mutant IDH1 and seizures in patients with glioma. Neurology 88, 1805–1813 (2017).

Rilinger, R. G. et al. Tumor-related epilepsy in high-grade glioma: a large series survival analysis. J Neurooncol https://doi.org/10.1007/s11060-024-04787-z (2024) doi:10.1007/s11060-024-04787-z.

10.

Vo, A. H., Ambady, P. & Spencer, D. The IDH1 inhibitor ivosidenib improved seizures in a patient with drug-resistant epilepsy from IDH1 mutant oligodendroglioma. Epilepsy Behav Rep 18, 100526 (2022).

11.

Drumm, M. R. et al. Postoperative risk of IDH-mutant glioma–associated seizures and their potential management with IDH-mutant inhibitors. J Clin Invest 133, (2023).

12.

Peters, K. B. et al. Use, access, and initial outcomes of off-label ivosidenib in patients with IDH1 mutant glioma. Neurooncol Pract 11, 199–204 (2024).

13.

Stritzelberger, J. et al. Time-dependent risk factors for epileptic seizures in glioblastoma patients: A retrospective analysis of 520 cases. Epilepsia 64, 1853–1861 (2023).

14.

Akinjiyan, F. A. et al. ARID2 mutations may relay a distinct subset of cutaneous melanoma patients with different outcomes. Sci Rep 14, 3444 (2024).

15.

Li, B. & Leal, S. M. Methods for detecting associations with rare variants for common diseases: application to analysis of sequence data. Am J Hum Genet 83, 311–321 (2008).

16.

Gu, Z. Complex heatmap visualization. iMeta 1, e43 (2022).

17.

Jay, J. J. & Brouwer, C. Lollipops in the Clinic: Information Dense Mutation Plots for Precision Medicine. PLOS ONE 11, e0160519 (2016).

18.

Brennan, C. W. et al. The somatic genomic landscape of glioblastoma. Cell 155, 462–477 (2013).

19.

López-Ginés, C. et al. Whole-exome sequencing, EGFR amplification and infiltration patterns in human glioblastoma. Am J Cancer Res 11, 5543–5558 (2021).

20.

Sakthikumar, S. et al. Whole-genome sequencing of glioblastoma reveals enrichment of non-coding constraint mutations in known and novel genes. Genome Biology 21, 127 (2020).

21.

Tobochnik, S. et al. Glioma genetic profiles associated with electrophysiologic hyperexcitability. Neuro Oncol 26, 323–334 (2024).

22.

Soeung, V., Puchalski, R. B. & Noebels, J. L. The complex molecular epileptogenesis landscape of glioblastoma. Cell Rep Med 5, 101691 (2024).

23.

Lim-Fat, M. J. et al. Clinical and Genomic Predictors of Adverse Events in Newly Diagnosed Glioblastoma. Clin Cancer Res 30, 1327–1337 (2024).

24.

Phan, K. et al. Association Between IDH1 and IDH2 Mutations and Preoperative Seizures in Patients with Low-Grade Versus High-Grade Glioma: A Systematic Review and Meta-Analysis. World Neurosurg 111, e539–e545 (2018).

25.

Li, Y. et al. IDH1 mutation is associated with a higher preoperative seizure incidence in low-grade glioma: A systematic review and meta-analysis. Seizure 55, 76–82 (2018).

26.

Yang, Y. et al. An analysis of 170 glioma patients and systematic review to investigate the association between IDH-1 mutations and preoperative glioma-related epilepsy. J Clin Neurosci 31, 56–62 (2016).

27.

Yang, P. et al. Correlation of preoperative seizures with clinicopathological factors and prognosis in anaplastic gliomas: A report of 198 patients from China. Seizure 23, 844–851 (2014).

28.

Schunemann, D. P. et al. Glutamate promotes cell growth by EGFR signaling on U-87MG human glioblastoma cell line. Pathol Oncol Res 16, 285–293 (2010).

29.

Binder, Z. A. et al. Epidermal Growth Factor Receptor Extracellular Domain Mutations in Glioblastoma Present Opportunities for Clinical Imaging and Therapeutic Development. Cancer Cell 34, 163–177.e7 (2018).

30.

Mulligan, L. et al. Genetic features of oligodendrogliomas and presence of seizures. The relationship of seizures and genetics in LGOs. Clin Neuropathol 33, 292–298 (2014).

31.

Feyissa, A. M. et al. Potential influence of IDH1 mutation and MGMT gene promoter methylation on glioma-related preoperative seizures and postoperative seizure control. Seizure 69, 283–289 (2019).

32.

Elliott, K. et al. Mechanistic basis of atypical TERT promoter mutations. Nat Commun 15, 9965 (2024).

33.

Pierini, T. et al. New somatic TERT promoter variants enhance the Telomerase activity in Glioblastoma. acta neuropathol commun 8, 145 (2020).

34.

Zhang, J. et al. TP53 R273C Mutation Is Associated With Poor Prognosis in LGG Patients. Front Genet 13, 720651 (2022).

35.

Joruiz, S. M., Von Muhlinen, N., Horikawa, I., Gilbert, M. R. & Harris, C. C. Distinct functions of wild-type and R273H mutant ∆133p53α differentially regulate glioblastoma aggressiveness and therapy-induced senescence. Cell Death Dis 15, 1–11 (2024).

36.

Prieto-Ruiz, J. A. et al. Expression of the human TIMM23 and TIMM23B genes is regulated by the GABP transcription factor. Biochim Biophys Acta Gene Regul Mech 1861, 80–94 (2018).

37.

Guo, Y. et al. TIMM44 is a potential therapeutic target of human glioma. Theranostics 12, 7586–7602 (2022).

38.

Wang, L. et al. Tumor mutational burden is associated with poor outcomes in diffuse glioma. BMC Cancer 20, 213 (2020).

39.

Aronica, E. et al. Epilepsy and brain tumors: Two sides of the same coin. Journal of the Neurological Sciences 446, (2023).

40.

Armstrong, T. S., Grant, R., Gilbert, M. R., Lee, J. W. & Norden, A. D. Epilepsy in glioma patients: mechanisms, management, and impact of anticonvulsant therapy. Neuro-Oncology 18, 779–789 (2016).

41.

Rahman, M. et al. Duration of Prophylactic Levetiracetam After Surgery for Brain Tumor: A Prospective Randomized Trial. Neurosurgery 92, 68 (2023).

42.

Louis, D. N. et al. The 2021 WHO Classification of Tumors of the Central Nervous System: a summary. Neuro Oncol 23, 1231–1251 (2021).

Yes

Abstract

Seizures occur in nearly half of all patients with high-grade gliomas, but few molecular markers have been identified as prognostic for glioma-associated seizures. We sought to examine the relationship between tumor molecular markers and glioma-associated seizures in patients with WHO grade 4 gliomas (glioblastoma, IDH-mutant astrocytoma). Amongst 950 patients diagnosed with grade 4 gliomas between 1999 and 2023, 414 (44%) patients experienced seizures. Tumor genomic characteristics were correlated with seizure incidence (before or after glioma diagnosis) and frequency in multivariable analyses. In multivariable analyses, chromosome 1p deletion (OR = 2.7, 95% CI [1.6, 4.4], p 0.001), pathogenic IDH1 variants (OR = 3.1, 95% CI [1.4, 7.1], p = 0.033), and EGFR amplification (OR = 1.6, 95% CI [1.1, 2.2], p = 0.039) were all significantly associated with increased odds of seizures before glioma diagnosis. For an exploratory subset of 83 patients, we conducted whole exome sequencing of the tumor, but no specific variants were associated with seizure occurrence. In conclusion, chromosome 1p deletion, pathogenic IDH1 status, and EGFR amplification were significantly associated with seizures before glioma diagnosis. Future work to identify additional molecular markers for patients at greatest risk for tumor-associated epilepsy may improve morbidity in high-grade glioma.