Aperiodic neural dynamics define a novel signature of glioma-induced excitation-inhibition dysregulation

Abstract

Diffuse gliomas remodel neuronal circuits with prognostic and therapeutic significance for patients. Electrophysiologic measures of cortical excitability hold promise for monitoring disease progression and evaluating therapeutic responses. The power law exponent (aperiodic slope) reflects the balance between excitatory and inhibitory activity within neuronal networks, a critical aspect of normal brain function often disrupted in neurological conditions. Despite its potential, the significance of the aperiodic slope in glioma-infiltrated tissue and its underlying cellular processes has not been fully investigated. Here, we integrate multimodal electrophysiological analysis with transcriptomic profiling to analyze the aperiodic slope in both normal and glioma-infiltrated cortex. We determine that glioma infiltration induces a flattening of the aperiodic slope, indicating a shift toward excitation dominance that varies according to tumor subtype and correlates with impairments in semantic naming. Single-nucleus RNA sequencing revealed that cortical regions with flat aperiodic slope exhibit transcriptional programs enriched in glutamatergic signaling, membrane depolarization, and excitatory synaptic transmission. The aperiodic slope responds to pharmacologically induced changes in cortical inhibition during propofol administration, a GABA_A agonist. Our results establish the aperiodic slope as a robust biomarker of glioma-associated excitation-inhibition imbalance, with potential applications in tumor classification and treatment monitoring.

Keywords: Aperiodic slope; excitation-inhibition balance; glioblastoma; glioma; isocitrate dehydrogenase mutant glioma; snRNA-seq; transcriptomics.

Fig. 1.. Glioma-associated Alterations in Cortical Aperiodic Activity and Molecular Features.

The study cohort consists of 13 patients diagnosed with oligodendroglioma, IDH-mutant and 1p/19q-codeleted, CNS WHO grade 2 (n=3, 23.0%), astrocytoma, IDH-mutant, CNS WHO grade 2 and 3 (n=5, 38.5%), and glioblastoma, IDH-wildtype, CNS WHO grade 4 (n=5, 38.5%). Spectral analysis was performed on 660 electrodes, comprising of both normal-appearing (n=518, 78.5%) and glioma-infiltrated (n=142, 21.5%) sites. A, Oncoprint displaying the clinical, histopathological, and genetic features of patients in cohort. Clinical variables include age, sex, WHO grade, glioma subtype, and extent of resection. Histopathological variables include proliferation index (Ki-67) and MGMT promoter methylation index. B, Overlay of FLAIR-positive regions on preoperative MRI T2 FLAIR across all patients. The heatmap represents the normalized spatial overlap of FLAIRpositive areas indicative of glioma-infiltrated cortex, with warmer colors signifying regions of higher overlap. C, Spectral analysis and aperiodic exponent fit of all electrodes across all patients analyzed in this study during resting state. The black line illustrates the original neural power spectrum, and the red line displays the full model fit. The aperiodic fit is represented by the blue dashed line. D, Left: Log-log aperiodic fits analyzed in the gamma frequency range (30–50 Hz) for all glioma-infiltrated (red) and normal-appearing (blue) electrodes during resting state. Right: Bar plot of average aperiodic exponent magnitudes comparing normal-appearing and glioma-infiltrated electrodes at 30–50 Hz. E, Left: Log-log aperiodic fits analyzed in the high gamma frequency range (70–150 Hz) for all glioma-infiltrated and normal-appearing electrodes during resting state. Right: Bar plot of average aperiodic exponent magnitudes comparing normal-appearing 2.813 (range 1.24–4.71) and glioma-infiltrated 3.525 (range 0.8–4.37) electrodes. Glioma-infiltrated electrodes have significantly flatter aperiodic exponents than normal-appearing electrodes. Each dot represents an electrode; error bars represent the 95% confidence interval. Comparisons were conducted with linear mixed effects models incorporating a patient-level random effect **p≤0.01, ****p≤0.0001. F, Sigmoidal regression modeling of the relationship between aperiodic exponent and average propofol dose (log(μg/mL)) in a patient’s body during surgery at a given time-point. Dots represent average aperiodic exponent across all patients computed in their respective concentration bin. Dashed line represents regression fit. A significant positive correlation (sigmoidal fit R² = 0.9824, 95% CI Top: 3.33–3.37, Bottom: 2.68–2.90) suggests that a steeper aperiodic exponent correlates with a higher propofol dose. Abbreviations: WHO = World Health Organization, MGMT = O6-methylguanine-DNA methyltransferase, O = oligodendroglioma, A = astrocytoma, GBM = glioblastoma multiforme

Fig. 2.. Aperiodic exponent differences across glioma subtypes and classification performance.

A, Violin plots of ECoG aperiodic exponent values in the 70–150 Hz frequency range grouped by glioma subtype across patients: Dashed lines within violins represent median, first quartile, and third quartile. B, (Left) Bar plot comparisons of ECoG aperiodic exponent magnitudes across glioma subtypes at 70–150 Hz, (Right) Bar plot comparisons of aperiodic exponent magnitudes within patient-matched magnetoencephalography (MEG) validation set at 30–50 Hz. Individual dots represent an electrode. Linear mixed effects modeling of ECoG data revealed GBM, and astrocytoma had a significantly flatter average aperiodic exponent compared to oligodendroglioma (GBM $\overline{\text{x}}$ vs oligodendroglioma $\overline{\text{x}}$, ****p≤0.0001; astrocytoma $\overline{\text{x}}$ vs oligodendroglioma $\overline{\text{x}}$, p=0.0052). Further, GBM had a significantly flatter average aperiodic exponent compared to astrocytoma (GBM $\overline{\text{x}}$ vs astrocytoma $\overline{\text{x}}$; ****p≤0.0001). C, Log-log aperiodic fits for normal-appearing (blue) and glioma-infiltrated (red) electrodes across glioma subtypes. D, Comparison of ECoG average aperiodic exponent between normal-appearing (blue) and glioma-infiltrated (red) electrodes for each glioma subtype at 70–150 Hz. Across all glioma subtypes in this study, glioma-infiltrated electrodes were found to have significantly flatter mean aperiodic exponents compared to normal-appearing electrodes (O: 3.04 vs 3.71, A: 2.46 vs 3.70, GBM: 1.40 vs 3.29; p≤0.0001). Bars represent mean aperiodic exponent magnitudes, and error bars represent 95% confidence intervals. Linear mixed effects modeling, ****p≤0.0001. E, Comparison of average aperiodic exponent between normal-appearing and glioma-infiltrated regions within MEG analysis at 30–50 Hz. Across all glioma subtypes in this study, glioma-infiltrated electrodes were found to have significantly flatter mean aperiodic exponents compared to normal-appearing electrodes (O: 2.98 vs 3.21, A: 2.91 vs 3.00, GBM: 2.99 vs 2.71; p≤0.0001). Bars represent mean aperiodic exponent magnitudes, and error bars represent 95% confidence intervals. Linear mixed effects modeling, ****p≤0.0001. F, ROC curves for bootstrap random forest classifiers predicting electrode tissue type (left), glioma subtype (right: O vs A vs GBM) using aperiodic exponent values. A “leave one patient out” cross-validation approach was utilized for each classifier. G, Representative immunofluorescence images from oligodendroglioma WHO 2 (left), astrocytoma WHO 2 (middle), and GBM (right). Tumor cells were labeled by IDH1 R132H or SOX2 immunostaining (red), with DAPI counterstaining (blue). For each sample, tumor cell density (cells/μm²) was quantified from a 20x field of view and normalized to tissue area, scale bar 50 μm. H, Aperiodic exponents were assigned to each sample based on the nearest electrode using a Euclidean distance-matching approach. Tumor cell density correlated inversely with the aperiodic exponent (n = 12 samples, R² = 0.578, p = 0.004). This relationship was consistent across glioma subtypes. Abbreviations: O = oligodendroglioma, A = astrocytoma, GBM = glioblastoma multiforme

Fig. 3.. Relationship between aperiodic exponent and cognitive performance during picture naming (PN) and auditory naming (AN) language tasks.

A, Overview of cognitive assessment language tasks and the difference between correct and incorrect trials. Left: PN task, where participants name the object shown on a laptop screen (e.g. “lion”) at stimulus onset. Periods of interest include the baseline period (pre-stimulus) and the post-stimulus onset window (500 ms). Right: AN task, where participants name an object (e.g., “Cat”) following a sentence prompt (“What animal meows?”). Periods of interest include the baseline period (pre-sentence) and the post-stimulus window (500 ms). B, Average aperiodic exponent extracted at 500 ms post-stimulus windows during PN trials at 70–150 Hz for normal-appearing (blue) and glioma-infiltrated (red) electrodes comparing correct (dark) and incorrect (light) responses. Linear mixed effects modeling showcased glioma-infiltrated electrodes having significantly lower aperiodic exponents for incorrect responses compared to correct responses (****p≤0.0001), while normal-appearing electrodes showed no significant differences across trial response (ns, p=0.575). C, Average aperiodic exponent during AN trials, comparing correct and incorrect responses across glioma-infiltrated and normal-appearing electrodes. Linear mixed effects modeling yielded similar results as PN where the aperiodic exponent was significantly lower in incorrect trials compared to correct trials within glioma-infiltrated electrodes (**p≤0.001). There was no significant difference in aperiodic exponent between trial response within normal-appearing electrodes (ns, p=0.480). Error bars represent the 95% confidence interval. D, Ridgeline density plots showing the distribution of the aperiodic exponent values during resting state, correct trials, and incorrect trials across glioma-infiltrated and normal-appearing electrodes during the PN language task. Distributions of the aperiodic exponent are further stratified by glioma subtype (O, A, GBM). Glioma-infiltrated electrodes consistently exhibit a shift toward flatter exponents during task-related errors compared to resting state and correct trials indicating a change in the evoked response to stimulation. E, Density plots showing the distribution of the aperiodic exponent values during resting state, correct trials, and incorrect trials within the AN language task. Similar to PN, when stratified by subtype, glioma-infiltrated regions show shifts toward shallower exponents, particularly during task-related errors. This shift is most prominent in GBM. Dot represents the mean, error bars represent the 95% confidence interval (*p≤0.01). Abbreviations: PN = picture-naming, AN = auditory naming, ns = no significance, O = oligodendroglioma, A = astrocytoma, GBM = glioblastoma multiforme

Figure 4.. Transcriptional analysis reveal excitation dominant tone in flat exponent region neurons of normal-appearing cortex.

A, Uniform manifold approximation and projection (UMAP) of all nuclei from both glioma-infiltrated and normal-appearing cortex tissue, annotated by broad cell type categories, including neurons (purple), glial cells, and tumor-infiltrating populations. B, UMAP embedding of neuronal nuclei, colored by transcriptionally defined neuronal subclasses, illustrating the diversity of excitatory (glutamatergic) and inhibitory (GABAergic) neurons. Extended data figure 8 displays feature plots for each UMAP. C, Hierarchical classification of neuronal subtypes into excitatory (Glutamatergic, left) and inhibitory (GABAergic, right) classes. Neuronal subclass composition differs between cortical regions preselected for steep and flat exponent samples. Bar plots display relative proportions of neuronal subclasses in cortical regions with steep and flat exponent assignments. D, Top: Coronal gross anatomical brain slice illustration depicting an insular tumor and the overlying electrode grid positioned on adjacent normal-appearing cortex. Subsequent panels reflect single-nucleus RNA sequencing (snRNA-seq) data derived from steep and flat exponent samples collected from normal-appearing cortex. Bottom: violin plot showing a significant difference in excitatory marker expression between steep and flat exponent samples (p < 1 × 10⁻¹⁰, two-tailed Mann Whitney U Test), suggesting an association between excitation-dominant transcriptional signatures and flatter exponents. E, Heatmap illustrating the z-score normalized difference in excitatory module scores between steep and flat exponent normal-appearing cortex samples (x-axis), displayed across identified neuronal subtypes (y-axis). Higher z-scores indicate relative enrichment within that cell type. F, Volcano plot showing log₂ fold-change of gene expression between steep and flat exponent normal-appearing cortex samples (two-sided Wilcoxon rank-sum test). Glutamatergic receptor subunits (e.g. GRIN2A, GRM7, GRIA2), inhibitory synaptic genes (e.g. GABBR2), and synaptic plasticity regulators (e.g. HOMER1) are significantly differentially expressed. G-H, Gene ontology analysis of biological processes enriched in steep (G) versus flat exponent normal-appearing cortex (H). Top enriched pathways in each condition are displayed, with point size indicating the gene count. I, Representative immunofluorescence images of formalin-fixed and paraffin embedded normal-appearing cortex samples showing DAPI+ nuclei and NeuN+ mature neurons. Left: SOX2 was not detected across both steep and flat exponent samples, consistent with the absence of glioma stem-like or infiltrating tumor cells in this region. Right: Images showing the distribution of Homer1+ and NeuN+ positive cells between steep and flat exponent samples. Scale bar, 50 μm. J, Quantification of Homer1+ and NeuN+ cells as a percentage of NeuN+ cells. Bars represent mean ± SEM, and statistical comparisons were performed using two-tailed T-test (p < 0.001).

Figure 5.. Transcriptional analysis reveals excitation dominance in flat exponent cortical region neurons within glioma-infiltrated cortex.

A, Left: Coronal gross anatomical brain slice illustration depicting a cortically-projecting tumor and the overlying electrode grid positioned on adjacent glioma-infiltrated cortex. This panel provides the anatomical context for subsequent single-nucleus RNA sequencing (snRNA-seq) analyses in this figure. Right: volcano plot showing log₂ fold-change of gene expression between steep and flat exponent glioma-infiltrated cortex samples (two-sided Wilcoxon rank-sum test). B-C, Gene ontology analysis of biological processes enriched in steep (B) versus flat exponent glioma-infiltrated cortex (C). Top enriched pathways in each condition are displayed, with point size indicating the gene count. D, Gene set enrichment analysis (GSEA) of glioma-infiltrated cortex steep (n=6 samples) and flat exponent samples (n=6 samples), showing enrichment of glutamatergic synaptic pathways in flat versus steep exponent glioma-infiltrated cortex samples. Normalized enrichment scores (NES) and adjusted P values (false discovery rate correction) are indicated. E, Summary of neuronal biophysical mechanisms gene set enrichment analysis associated with steep versus flat exponent samples, highlighting key glutamatergic and inhibitory pathways. F, Violin plots of excitatory module scores in oligodendroglioma, astrocytoma, and glioblastoma, stratified by steep versus flat exponent samples. Center values represent median, dashed lines represent upper and lower quartile, and statistical significance was determined using a Mann-Whitney U test, ****p < 0.0001. G, Heatmap illustrating the z-score normalized difference in excitatory module scores between steep and flat exponent glioma-infiltrated cortex samples by glioma subtype (x-axis), displayed across identified neuronal subtypes (y-axis). Higher z-scores indicate relative enrichment within that cell type. H, Representative immunofluorescence images of formalin fixed and paraffin embedded glioma-infiltrated cortex samples showing the distribution of Homer1+ and NeuN+ positive cells between steep and flat exponent samples. Scale bar, 50 μm. I, Quantification of Homer1+ and NeuN+ cells as a percentage of NeuN+ cells of all samples combined by status (steep vs flat; left) and stratified by glioma subtype (right). Bars represent mean ± SEM, individual dots represent a snapshot used for quantification, and statistical comparisons were performed using two-tailed Mann-Whitney U test (*p < 0.05, **p < 0.001). J, Representative confocal microscopy images of organoid slices stained for MAP2 (neuronal marker), synapsin-1 (presynaptic marker), and Homer1 (postsynaptic marker) across glioma subtypes. Scale bar, 20 μm. K, Quantification of glioma-infiltrated organoids. Colocalization of Homer1 and Synapsin-1 is quantified per DAPI+ nucleus across glioma subtypes. Bars represent mean ± SEM, individual dots represent a snapshot used for quantification, and statistical comparisons were performed using one-way ANOVA (*p < 0.05, **p < 0.001). L, Multielectrode array analysis of glioma-infiltrated GBM organoids co-cultured with neurons, showing weighted mean firing rate (Hz) across conditions. Bars represent mean ± SEM. Statistical significance was determined using a two-tailed Mann-Whitney U test (*p < 0.05).