Neuro-glioma activity-dependent growth mechanisms: an actionable circuit from NLGN3-ADAM10 to AMPA synapses

Introduction

High-grade gliomas (HGGs), especially isocitrate dehydrogenase (IDH)-wild-type glioblastoma (GBM), remain among the most lethal primary brain tumors: under the current Stupp regimen, median survival rarely exceeds 2 years and long-term survivors are uncommon (1). GBM also accounts for the majority of malignant central nervous system (CNS) tumors and CNS tumor-related deaths (2). This disconnect between intensive, predominantly tumor cell-centric therapies and only modest survival gains has pushed attention toward the host brain milieu. Within the emerging field of “cancer neuroscience”, neurons have shifted from passive bystanders to active drivers whose electrical activity and secreted factors can shape glioma growth, invasion and epileptogenesis (3-5). A converging body of work defines a core “activity-dependent growth” circuit: neuronal activity triggers a disintegrin and metalloproteinase domain-containing protein 10 (ADAM10)-mediated shedding of neuroligin-3 (NLGN3), and soluble NLGN3 (sNLGN3) induces a pro-proliferative, synaptic gene program in glioma cells; in parallel, neuron-glioma α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor (AMPAR) synapses provide bona fide excitatory input to tumor cells, which in turn amplify local network hyperexcitability and seizures, thereby further increasing host neuronal firing and ultimately closing the “activity-shedding-signaling-synapse-reactivation” loop (6). At the tumor-intrinsic level, a “high-neural” GBM subtype with neuron-like epigenetic and transcriptional features, dense neuron-glioma synapses, strong infiltrative behavior and poor survival directly links circuit integration to clinical outcome (7). Cross-species single-cell analyses, however, reveal marked species- and subtype-specific synaptic gene and long non-coding RNA (lncRNA) modules, with primate-specific regulators, underscoring the limits of mouse-based circuit maps for human disease (8).

Neuronal activity also interfaces with the immune microenvironment: glutamate, NLGN3 and AMPAR-mediated signaling can drive an “immune-excluded” phenotype and dampen responses to immunotherapy, situating the neuro-immune-tumor axis as an additional layer of activity-dependent control (9). Key nodes of this circuit—NLGN3 shedding, downstream phosphoinositide 3-kinase-mechanistic target of rapamycin (PI3K-mTOR) amplification, Ca²⁺-permeable AMPARs and network-level hyperexcitability—are now relatively well defined (3,6), but major gaps remain regarding NLGN3 receptor/co-receptor repertoires, quantitative relationships between firing patterns, shedding thresholds and synaptic plasticity, and the clinically actionable “activity window” for intervention. Against this backdrop, the present review aims to integrate the NLGN3-ADAM10 axis and neuron-glioma AMPA synapses into a unified circuit framework, and to discuss therapeutic concepts centered on “activity dampening, shedding inhibition and synapse blockade” (10). Epilepsy—both the most common symptom of glioma and the most accessible readout of network hyperexcitability—provides a clinical anchor for these strategies, and supports combined targeting of glutamatergic signaling, IDH-dependent metabolic states in which relevant and circuit connectivity to improve tumor control and seizure management (10). Complementary lines of evidence from whole-brain rabies tracing and neuromodulation show that patient-derived GBM rapidly integrates into diverse neuronal populations, with cholinergic neurons playing a prominent pro-invasive role; silencing these tumor-associated neurons augments radiotherapy and can arrest tumor growth in vivo (11). Human induced pluripotent stem cell (iPSC)-derived cholinergic neurons form functional synapses with GBM via cholinergic receptor muscarinic 3 (CHRM3), enhancing Ca²⁺ dynamics and proliferative transcriptional programs, suggesting that non-glutamatergic neurotransmitter systems are also embedded in the activity-dependent growth circuit (12). Finally, AMPARs themselves have emerged as druggable nodes: clinical and experimental data on non-competitive AMPAR antagonists such as perampanel demonstrate the feasibility of pharmacologically modulating this circuit, while leaving open the key design question of how best to embed such agents within a broader strategy of initial “noise reduction”, subsequent loop disruption and long-term consolidation (13).

Evidence base for activity-dependent growth

Preclinical and cellular evidence for neuron-glioma interactions

A substantial preclinical literature provides causal evidence, in experimental models, that neuronal activity can promote glioma growth and invasion. Broadly, these studies adopt two complementary strategies. First, defined neuronal populations are driven to fire at higher rates in vivo using optogenetic, chemogenetic, or electrical stimulation, and tumor burden is quantified using volumetrics, proliferation markers, and invasion metrics in xenograft or genetically engineered models. Second, network activity is manipulated exogenously in acute brain slices or co-culture systems while directly measuring glioma proliferation, migration, membrane potential changes, and Ca²⁺ dynamics, thereby linking neuronal activity to tumor cell behavior under controlled conditions. Recent integrative reviews emphasize that, across adult and pediatric HGG, diffuse midline glioma (DMG), and low-grade optic pathway glioma models, neuronal activity supports tumor initiation, progression, and treatment resistance through two converging routes: activity-regulated paracrine factors and functional neuron-glioma synapses (14,15).

At the level of activity manipulation, Huang-Hobbs et al. used genetically engineered mouse models and optogenetic stimulation of callosally projecting neurons to demonstrate that excitatory activity in the contralateral hemisphere can accelerate growth of a primary GBM focus (16). Chronic high-frequency stimulation increased tumor volume and was associated with transcriptional changes and altered invasion trajectories, consistent with a role for long-range axonal projections and plasticity-related programs, including axon guidance components such as semaphorin 4F (SEMA4F). In related experiments, direct stimulation of light-gated ion channels expressed on tumor cell membranes indicated that membrane depolarization itself can increase proliferation indices, supporting an electrophysiological drive component in addition to paracrine signaling.

Beyond cortical projection neurons, specific neurotransmitter systems have been dissected in a projection- and region-dependent manner. In a DMG model, Drexler et al. showed that activity of midbrain cholinergic neurons regulates tumor proliferation with nucleus-target specificity: stimulation of the pedunculopontine nucleus preferentially promoted growth of pontine tumors, whereas stimulation of the laterodorsal tegmental nucleus accelerated expansion of thalamic lesions (17). These data extend the activity-dependent framework from local excitatory circuits to long-range modulatory systems and indicate that projection topology can constrain the anatomical map of tumor progression.

Whereas the studies above prioritize which neuronal populations are active, work centered on NLGN3 addresses which activity-regulated factors are released and how they instruct tumor state. Hua et al. summarized model experiments showing that optogenetic or electrical stimulation of cortical pyramidal neurons increases sNLGN3 in the peritumoral space and enhances growth and invasion of implanted gliomas, whereas the same tumor cells fail to form progressive lesions in Nlgn3-knockout hosts, underscoring dependence on microenvironmental NLGN3 in these settings (18). At a broader systems level, Wang et al. synthesized evidence for activity-regulated secreted factors—including NLGN3, Brain-derived neurotrophic factor (BDNF), and IGF-1—and noted that these cues activate proliferative signaling [for example, PI3K-mTOR and mitogen-activated protein kinase (MAPK)] while inducing synapse- and axon guidance-associated transcriptional programs, thereby priming subsequent establishment and strengthening of neuron-glioma synapses (19). Collectively, across diverse lineages and anatomical contexts, preclinical data support a convergent conclusion: increasing relevant network activity can accelerate tumor progression via paracrine and electrophysiological routes, whereas disrupting these routes can attenuate growth in experimental models.

Extending these findings from exogenous manipulation to endogenous integration, tissue- and cellular-level investigations address whether glioma cells are structurally and functionally integrated into neural circuits. Venkataramani et al. combined electron microscopy, immunofluorescent co-labeling, and in vivo imaging to resolve ultrastructural features of functional neuron-glioma synapses (20). In murine and patient-derived GBM models, they reported postsynaptic density-like specializations on tumor cell somata and on termini of tumor microtubes, presynaptic boutons from pyramidal neuron axons containing synaptic vesicles, and enrichment of glutamate receptors and synaptic adhesion molecules at the synaptic interface. Synapse density was higher in tumors with a more pronounced neural-like program and was associated with greater invasiveness and poorer survival, supporting biological and potential clinical relevance of circuit integration.

Functionally, Taylor et al. used patch-clamp recordings in acute brain slices with high-resolution live-cell imaging to show that glioma cells exhibit fast AMPAR-mediated excitatory postsynaptic currents (EPSCs) in response to neuronal glutamate release and that current amplitude and frequency can change with neuronal firing patterns, consistent with synaptic plasticity (21). Using TrkB deletion, BDNF blockade, and AMPAR subunit analyses, they further showed that BDNF-TrkB signaling increases AMPAR surface expression and postsynaptic clustering on tumor cells, strengthening neuron-glioma synapses at both structural and functional levels. These findings imply that, within a single tumor, the degree of electrical integration can evolve over time and across regions in response to activity patterns and microenvironmental cues.

Bridging scales from single cells to networks, Picart and Hervey-Jumper integrated intraoperative electrophysiology, functional imaging, and anatomical observations into a framework spanning unit physiology to circuit remodeling (22). In this view, glioma cells can form electrically coupled networks via tumor microtubes and gap junctions and may include pacemaker-like elements, while AMPAR-mediated synaptic inputs and tumor-derived glutamate efflux disrupt local excitation-inhibition balance and drive reorganization of cortical circuits, including in eloquent regions. Neuron-glioma synapses thus provide a cellular-physiological substrate that links microscopic electrical events to macroscopic network dysfunction and clinical phenotypes such as seizures and cognitive impairment.

Patient-related evidence

The concept of activity-dependent growth enters the clinical arena primarily through studies based on human samples and intraoperative recordings. In patients with GBM undergoing awake craniotomy for language mapping, Krishna et al. obtained intracortical electrophysiological recordings alongside spatially matched tumor biopsies and found that HGGs can markedly remodel long-range functional connectivity (23). Task-related neural activity was no longer confined to classical language areas but extended into broad swaths of tumor-infiltrated cortex. Tumor cells sampled from highly connected regions displayed greater synaptogenic capacity and neurotrophin secretion, and were more invasive in xenograft models. Clinically, patients with higher functional connectivity had shorter overall survival (OS) and worse performance on language tasks, suggesting that the more tightly a tumor is wired into the brain network, the more aggressive its biological behavior.

Approaching the problem from the visible phenotype of epilepsy, Rilinger et al. examined the relationship between tumor-related epilepsy (TRE) and survival in a large retrospective cohort of patients with HGGs (24). They found that patients who presented with seizures had longer OS even after multivariable adjustment. This observation is consistent with multiple prior systematic reviews reporting that patients with seizures appear to live longer. However, the authors emphasized that this apparently favorable association may partly reflect the tendency of seizures to arise in tumors that are smaller, better circumscribed, and somewhat less aggressive. At the same time, epilepsy itself imposes a substantial burden on quality of life and functional status, and late-onset seizures often signal disease progression; the epilepsy-survival relationship therefore warrants cautious interpretation. At a mechanistic level, Chen et al. systematically reviewed recent basic and clinical studies on glioma-associated epilepsy (GAE) and highlighted a vicious cycle linking tumor, microenvironment, and network (25). Gliomas release glutamate, downregulate glutamate uptake by surrounding astrocytes, injure fast-spiking inhibitory interneurons, and shift the polarity of GABAergic signaling, thereby creating a highly hyperexcitable network. Recurrent epileptiform discharges, in turn, promote tumor growth and invasion via Ca²⁺ loading, synaptic plasticity, and activity-dependent transcriptional programs. From this vantage point, epilepsy is not merely a “complication” but rather a clinical window and readout of the underlying activity-dependent growth circuit.

Methodological underpinnings: distinguishing correlation from causality

Interrogating activity-dependent growth across scales requires a coordinated methodological toolkit. A central challenge in this emerging field is to distinguish correlation—observing that neuronal activity accompanies tumor progression—from causation, or demonstrating that activity directly drives tumor behavior. The studies reviewed in Sections 2.1–2.3 illustrate how different methodological approaches contribute to establishing causality at multiple levels of biological organization.

At the preclinical level, causal inference derives primarily from interventional designs. Optogenetic, chemogenetic and electrical stimulation enable researchers to actively manipulate neuronal firing while measuring tumor outcomes, thereby establishing directionality. Loss-of-function experiments such as Nlgn3 knockout models, which demonstrate failure of tumor formation despite stimulation, provide complementary evidence by showing necessity. These interventional approaches in animal models represent the strongest available evidence for causality in the activity-dependent growth paradigm.

At the tissue and cellular level, causality is established through ex vivo manipulation combined with real-time measurement. Acute brain slice preparations allow researchers to apply specific stimuli while recording tumor cell responses, including patch-clamp demonstration of EPSCs time-locked to presynaptic stimulation (21). Pharmacological interventions such as AMPAR antagonists blocking neuron-induced Ca²⁺ signals further establish mechanistic necessity.

Patch-clamp electrophysiology and acute brain-slice recordings remain the reference standards for measuring single-cell excitability and local-circuit dynamics. Chen et al. provided standardized procedures for whole-cell patch-clamp and extracellular field recordings in mouse brain slices, including acute slice preparation, targeted recordings from neurons and astrocytes, induction of long-term potentiation, and key analysis steps (26). Such reproducible workflows form the foundation for quantifying neuron-glioma synaptic currents, tumor cell depolarization, and focal network discharges in xenograft and genetically engineered glioma models.

Approaches that explicitly couple functional recordings with molecular profiling are also emerging as tools to link cellular phenotypes to molecular mechanisms. Rodriguez et al. introduced a patch-clamp single-cell proteomics (patch-SCP) framework linking whole-cell recordings in acute slices to downstream single-cell mass spectrometry (27). Neurons are harvested while maintaining electrophysiological integrity, enabling simultaneous measurement of electrical phenotypes and proteomic states. Although still being validated in basic neuroscience, this concept is well suited to tumor-brain interfaces, for example by comparing receptor, channel, and synaptic protein repertoires between highly connected functional hotspot neurons and more typical peritumoral neurons, thereby constraining hypotheses about which circuit elements drive tumor behavior.

At the human tissue level, establishing causality is inherently more challenging due to ethical constraints on experimental manipulation. Here, the field relies on integrating multiple lines of correlative evidence that collectively support causal inference. Spatial transcriptomics can localize synaptic gene expression to specific tumor regions, but this alone demonstrates correlation. When combined with intraoperative electrophysiology showing hyperexcitability in those same regions, and with patient outcome data linking connectivity to survival, the converging evidence strengthens causal interpretation. Longitudinal studies tracking seizure burden or electroencephalography (EEG) changes alongside tumor progression provide temporal evidence supporting directionality. Causal inference in human studies therefore depends on what philosophers of science call consilience, meaning the convergence of multiple independent lines of evidence.

Multi-omics and epigenetic analyses extend the paradigm from correlative signaling to durable causal encoding. Chakraborty et al. reviewed how tumor cells translate neuronal cues such as glutamate, NLGN3, and gamma-aminobutyric acid (GABA) into stable transcriptional programs via enhancer reprogramming, chromatin remodeling, and three-dimensional genome organization (28). These data underscore that activity-linked signals can imprint persistent neural-like states rather than only transiently modulating ion channels and receptors. Demonstrating that blocking neuronal activity prevents these epigenetic changes, and that these changes in turn mediate tumor phenotypes, provides a causal chain from acute neuronal signal to long-term tumor behavior.

Computational modeling serves as a bridge between correlation and causation by generating testable predictions. Multiscale mathematical models incorporating neuronal activity as a driver can predict invasion patterns and recurrence trajectories that can be validated experimentally. When model predictions match experimental outcomes, this strengthens causal interpretation; when they diverge, it suggests missing mechanistic elements.

Together, electrophysiology, acute-slice Ca²⁺ imaging, transgenic and knockout models, neuromodulatory tools, and integration with single-cell and spatial multi-omics constitute a methodological toolkit that enables the field to move progressively from observing correlations to establishing causation. The strongest causal claims emerge when multiple approaches—interventional animal studies, mechanistic ex vivo experiments, and convergent human observational data—point to the same conclusion.

The NLGN3-ADAM10 axis: shedding, secretion, and downstream signaling

Activity-dependent shedding

Within the framework of cancer neuroscience, the NLGN3-ADAM10 axis is regarded as one of the prototypical pathways through which neuronal activity is encoded into tumor-promoting signals. Recent systematic reviews of tumor-neural circuits have pointed out that neuronal adhesion molecules originally devoted to establishing and maintaining synaptic contacts—such as the neuroligin family—are repurposed in multiple solid tumors as key signaling nodes that drive tumor initiation, stemness maintenance, and therapeutic resistance, with NLGN3 being particularly prominent in HGGs (29). In parallel, ADAM10, a transmembrane zinc-dependent protease, is capable of cleaving the ectodomains of dozens of substrates, including Notch, APP, E-cadherin, and immune co-stimulatory molecules, thereby mediating ectodomain shedding. Comprehensive reviews have shown that ADAM10 expression is upregulated in many cancers and correlates with higher malignancy and poorer prognosis. In brain tumors, ADAM10 has repeatedly been identified as the critical enzyme required for NLGN3 cleavage and is thus increasingly viewed as a candidate therapeutic target (30). Functionally, ADAM10 is not a simple on-off “switch”, but rather a hub that shapes multi-channel communication between tumor cells and their microenvironment. In a recent study using CRISPR-mediated ADAM10 knockout in human U251 GBM cells, in vitro proliferation was only moderately affected, whereas loss of ADAM10 markedly delayed tumor growth in xenografts in immunodeficient mice (31). Proteomic analyses revealed substantial reductions in the shedding or membrane expression of several known ADAM10 substrates (such as Notch/Eph receptors and their ligands) and of molecules involved in adhesion, migration, metabolic reprogramming, and angiogenesis. These findings suggest that ADAM10-mediated proteolysis does not merely tweak signaling through individual receptors; it globally rewires the “language” through which tumor cells communicate with surrounding cells. Against this backdrop, reviews on “nervous system hijacking” by tumors further emphasize that in multiple solid malignancies, including gliomas, neuronal activity modulates ADAM10 activity and thereby drives activity-dependent shedding of a cohort of synapse-associated molecules, among which NLGN3 stands out as a representative soluble pro-tumor factor in HGG (32). Across various animal models and human brain specimens, neuronal and oligodendrocyte precursor cell (OPC) firing is accompanied by detectable increases in sNLGN3 levels, whereas genetic or pharmacological inhibition of ADAM10 significantly suppresses this activity-related release and is associated with slower tumor progression.

Taken together, the activity-dependent shedding of NLGN3 via the NLGN3-ADAM10 axis displays several key informational features. First, shedding is tightly coupled to neuronal firing patterns and Ca²⁺ dynamics, exhibiting sensitivity to frequency and temporal windows. Second, proteolysis converts NLGN3 from a postsynaptic membrane protein into a soluble factor that can diffuse within cerebrospinal fluid (CSF) and the extracellular matrix, greatly expanding the spatial reach of signals initiated at individual synapses. Third, both neurons and OPCs serve as substrate sources, such that activity from distinct cellular lineages converges into a common pool of NLGN3. At this level, ADAM10 functions not only as the “scissors” but also as a crucial “encoder” that translates neuronal electrical activity into tumor-readable chemical signals.

Signaling pathways and feed-forward amplification

Early work established that neuron-derived sNLGN3 can rapidly activate pro-proliferative signaling in glioma cells, including focal adhesion kinase-PI3K-mTOR (FAK-PI3K-mTOR) and extracellular signal-regulated kinase/MAPK (ERK/MAPK) cascades, and can induce transcriptional programs enriched for synaptic components, axon guidance molecules, and migration-related genes. This activity-linked reprogramming provides a molecular basis for subsequent synaptic integration into local neural networks. More recent mechanistic studies suggest that NLGN3 signaling may also operate through tumor-intrinsic, autocrine reinforcement. In human glioma systems, Dang et al. reported that NLGN3 is overexpressed across multiple glioma cell lines and in HGG tissues, with higher expression correlating with pathological grade (33). siRNA-mediated knockdown of endogenous NLGN3 reduced proliferation, clonogenicity, migration, and invasion in U87 and U251 cells and attenuated activation of PI3K-AKT, ERK1/2, and the Src-family kinase LYN, whereas NLGN3 overexpression produced reciprocal phenotypes.

Importantly, this study proposed a feed-forward architecture in which LYN acts not only as a downstream effector but also as an upstream amplifier by increasing ADAM10 expression and activity, thereby enhancing NLGN3 shedding. Regarding the relationship among these signaling components, the LYN-PI3K-mTOR axis and the chondroitin sulfate proteoglycan 4 (CSPG4)-Piezo-type mechanosensitive ion channel component 1 (PIEZO1) mechanosensitive pathway described in Section 3.3 operate in parallel rather than in a strict linear sequence. Available evidence suggests that NLGN3 engagement simultaneously activates both routes: the LYN-PI3K-mTOR cascade promotes proliferation and transcriptional reprogramming, whereas the CSPG4-PIEZO1 module mediates rapid membrane depolarization and Ca²⁺ influx that further reinforce tumor cell excitability and invasive capacity. These two axes are not mutually exclusive; instead, they converge on common downstream effectors such as Ca²⁺-dependent transcriptional regulators and cytoskeletal remodeling factors, creating potential points of cross-talk. Whether one pathway predominates over the other likely depends on the specific cellular context, including tumor subtype, the composition of surface receptor complexes, and the intensity or pattern of neuronal activity. Temporally, the LYN-PI3K-mTOR cascade may initiate transcriptional changes that sustain synaptic integration over hours to days, while the CSPG4-PIEZO1 pathway provides more immediate electrophysiological responses that can rapidly modulate tumor cell behavior. The coexistence of parallel signaling routes has important implications for combination therapy: targeting only one axis may leave the other intact, potentially limiting therapeutic efficacy. Therefore, rational combination strategies may need to simultaneously disrupt both the LYN-PI3K-mTOR proliferative arm and the CSPG4-PIEZO1 mechanoelectrophysiological arm to achieve full circuit interruption.

Pharmacologic ADAM10 inhibition, including with GI254023X, suppressed glioma cell proliferation and invasion in vitro, and higher ADAM10 expression correlated with higher grade and poorer prognosis in clinical samples. Together, these data support a multi-compartment amplification model: host neuronal and OPC activity initiates ADAM10-dependent NLGN3 release; sNLGN3 activates kinase cascades (PI3K-AKT, ERK, FAK) and synapse-associated transcriptional programs in tumor cells; and tumor-intrinsic LYN-ADAM10 coupling further increases ADAM10 activity and NLGN3 availability within the tumor compartment, potentially converting tumor cells into secondary contributors to the ligand pool. This amplification is what elevates the NLGN3-ADAM10 axis from a paracrine pathway into a circuit-level growth controller (see Figure 1).

Figure 1 Convergence of mechanoelectrical and kinase signaling pathways drives a synapse-competent state and glioma progression. Supported and produced by Figdraw.

Putative receptors/co-receptors: from CSPG4 to mechanosensitive pathways

Despite robust evidence for the pro-tumor effects of NLGN3, the “receptor side” of this pathway in glioma cells has long remained a black box. A study published in 2025 provided the first clear molecular clue: Gillespie et al. used biochemical coupling, structural analyses, and functional assays to demonstrate that sNLGN3 forms a specific complex with chondroitin sulfate proteoglycan 4 (CSPG4, also known as NG2) (34). This interaction occurs on both normal OPCs and glioma cells and promotes ADAM10-mediated cleavage of CSPG4. Cleavage of CSPG4 alters membrane tension and activates the mechanosensitive ion channel PIEZO1, leading to membrane depolarization and Ca²⁺ influx, which in turn maintains OPCs in an undifferentiated state and enhances glioma cell proliferation. This work incorporates the “CSPG4-PIEZO1” axis into the NLGN3 signaling network and suggests that downstream of NLGN3 lie not only classical kinase cascades but also mechanosensory and electrophysiological modules. Within this framework, CSPG4 can be viewed as a critical co-receptor or “signal amplifier” for NLGN3. On the one hand, OPC populations with high CSPG4 expression are more readily kept in a proliferative/undifferentiated state by NLGN3, providing a fertile “soil” for gliomagenesis. On the other hand, in glioma cells, CSPG4 shedding and its functional coupling to PIEZO1 create a bridge from surface adhesion to membrane potential remodeling. This NLGN3-triggered mechano-electrophysiological conversion helps explain why tumor cells in NLGN3-enriched regions are more likely to form neuron-glioma synapses and to exhibit heightened invasiveness.

At a broader level, NLGN3 expression in the brain shows pronounced regional and synapse-type specificity. As summarized by Uchigashima et al., NLGN3 exists in multiple splice isoforms with distinct distributions and functions across excitatory versus inhibitory synapses and across brain regions; its intracellular tail harbors multiple phosphorylation sites and binding motifs for scaffold proteins, allowing it to serve as both a “marker” and a “modulator” of circuit-specific connectivity (35). Transposed into tumor biology, this physiological perspective implies that the receptor/co-receptor repertoire on glioma cells is unlikely to be limited to a single molecule. Instead, it probably consists of “receptor islands” composed of CSPG4, integrins, receptor tyrosine kinases, and other molecules, with specific combinations determined by tumor subtype, cell state, and the intrinsic properties of the surrounding circuit. Incomplete characterization of this receptor-level architecture remains one of the most critical knowledge gaps for circuit modeling and rational drug design targeting the NLGN3-ADAM10 axis.

Points of intervention: inhibiting shedding, sequestering ligand, and blocking effectors

Therapeutic strategies targeting the NLGN3-ADAM10 axis can be broadly grouped into three categories: (I) directly suppressing ADAM10-mediated shedding to reduce NLGN3 supply; (II) sequestering sNLGN3 through neutralization or decoy approaches; and (III) blocking downstream effectors, such as the LYN-ADAM10 feed-forward loop, FAK-PI3K-mTOR cascades, or the CSPG4-PIEZO1 mechanosensory pathway. Given that ADAM10 also processes physiological substrates including Notch and APP in the normal brain, designing ADAM10-targeted interventions inevitably raises the question of how to achieve sufficient anti-tumor activity while preserving an acceptable safety window. On the shedding-inhibition front, the most representative clinical effort to date is the PBTC-056 trial built around the ADAM10/17 inhibitor INCB007839 (aderbasib). Lenzen et al., on behalf of the Pediatric Brain Tumor Consortium, reported preliminary results from this phase I study in children and adolescents aged 3–21 years with recurrent or progressive HGGs, including diffuse intrinsic pontine glioma (DIPG), DMG, GBM, and anaplastic astrocytoma (36). The trial employed a twice-daily oral, fixed-dose-escalation design with safety and tolerability as primary endpoints. Unexpected thrombotic toxicities observed in early cohorts prompted protocol modifications and the introduction of routine low-molecular-weight heparin prophylaxis. Although the study remains in the safety-exploration stage and efficacy data are not yet mature, its significance lies in translating the concept of “targeting microenvironmental NLGN3 shedding” into a real-world pediatric HGG population, thereby laying the groundwork for future phase II/III designs centered on biomarker-driven stratification and rational drug combinations. At the level of ligand sequestration, in addition to conventional neutralizing antibodies and soluble receptor constructs, newer delivery and materials-based strategies are emerging. Ding et al. designed an “optogenetic-like” liposomal nanoplatform functionalized with NLGN3-binding modules to selectively accumulate in the local tumor microenvironment and used light-controlled release or structural modulation to disrupt NLGN3-dependent signaling (37). In vivo, this platform showed synergistic tumor suppression when combined with radiochemotherapy or other targeted agents. Such work suggests that NLGN3 need not be merely blocked; it can be “rewired” in space and time, opening up a wide design space for combined pharmacologic and device-based interventions.

From a circuit-level perspective, interventions targeting the NLGN3-ADAM10 axis are unlikely to be most effective in isolation. In patients with prominent epileptic phenotypes or overt network hyperexcitability, ADAM10 inhibition may need to be combined with modulation of neuronal activity to jointly dampen the upstream firing drive and the midstream shedding amplifier. In tumors with highly active NLGN3 autocrine signaling and a robust CSPG4-PIEZO1 mechanosensory axis, ligand neutralization/decoy strategies and mechanosensitive channel blockade may be particularly rational. Moreover, in clinical contexts where downstream PI3K-mTOR inhibitors are already available, combining NLGN3-ADAM10-targeted approaches with established small-molecule inhibitors may broaden coverage of the overall activity-dependent growth circuit without substantially increasing toxicity. How best to stratify patients—based on NLGN3 expression, ADAM10 activity, CSPG4/PIEZO1 status, and epilepsy/network metrics—and how to sequence or combine interventions accordingly are key questions that future clinical trial designs will need to address.

AMPA synapses: electrochemical coupling between neurons and glioma

Bona fide synapses

In the preceding discussion of the NLGN3-ADAM10 axis, the emphasis was on how neuronal “chemical language” is encoded and amplified. At AMPA receptor-mediated neuro-glioma synapses, this chemical language is further organized into electrical signaling units with precise temporal and spatial structure. Recent reviews have systematically compared neuron-glioma synapses with physiological neuron-OPC synapses and highlighted their structural and functional isomorphism: glioma cells express a full complement of postsynaptic density proteins and glutamate receptors and exhibit fast EPSCs time-locked to neuronal action potentials, thereby fulfilling both the morphological and physiological criteria for bona fide synapses (38). This perspective reframes gliomas from passive cell populations bathed in glutamate to aberrant neuron-like elements that actively participate in network information flow.

At the molecular level, a recent study of glioma-derived SPARCL1 has provided high-resolution structural and functional evidence. In neuron-GBM co-cultures and xenograft models, Li et al. showed that tumor-secreted SPARCL1 markedly increases the density of neuron-glioma synapses in peritumoral regions: presynaptic synapsin puncta and postsynaptic markers on glioma membranes exhibit high degrees of colocalization, and electron microscopy reveals synaptic clefts with the ultrastructural features of canonical excitatory chemical synapses (39). Tumor cells in these regions display frequent AMPAR-mediated excitatory events on patch-clamp recordings, which are highly sensitive to AMPAR antagonists. Genetic or antibody-based inhibition of SPARCL1 not only reduced synapse density but also attenuated focal cortical high-frequency discharges and tumor growth, indicating that these “extracellular matrix-bridged” contacts are not merely structural curiosities but functionally competent units of pathological electrochemical coupling. Beyond synapse stabilization driven by tumor-secreted factors, the intrinsic “synaptic gene program” of glioma cells is equally important for the formation of bona fide neuron-glioma synapses. In H3.1K27M DMG, Zhang et al. found that the chromatin remodeler chromodomain helicase DNA-binding protein 2 (CHD2), in cooperation with the transcription factor FOSL1, strongly upregulates axon guidance and synapse-related genes (40). CHD2 knockdown reduced the number of postsynaptic density protein 95 (PSD95)-synapsin co-labeled synapse-like structures formed by tumor cells in neuron co-cultures and blunted neuron activity-evoked Ca²⁺ transients and proliferative responses in glioma cells. Crucially, in these co-culture experiments, the AMPAR antagonist NBQX significantly suppressed neuron-induced Ca²⁺ signals and activity-dependent proliferation, further confirming that these CHD2-dependent neuron-glioma contacts function as classical AMPAR-mediated excitatory synapses.

Taken together, structural and electrophysiological data support the conclusion that in HGGs—particularly pediatric DMGs and a subset of adult GBMs—axon terminals of neurons form bona fide synapses with tumor cells that meet classical morphological and functional definitions. These synapses tend to cluster at the invasive front and in hyperexcitable cortical regions, providing tumors with rapid, high-fidelity electrical input and furnishing the microscopic substrate for subsequent AMPAR subunit remodeling and network-level plasticity.

Receptor properties: a Ca²⁺-permeable AMPAR phenotype

Synapses provide the conduit, but the functional consequences are strongly shaped by postsynaptic receptor composition. In the adult brain, edited GluA2-containing AMPARs that are relatively Ca²⁺-impermeable predominate. In contrast, multiple reviews highlight that glioma postsynaptic sites often show enrichment of GluA1 and GluA4 with relative paucity of GluA2 (41). This subunit bias yields Ca²⁺-permeable AMPARs, such that even individual synaptic events can produce substantial Ca²⁺ influx and downstream signal amplification. Functionally, this receptor architecture provides an efficient route by which neuronal spikes are converted into intracellular Ca²⁺-dependent programs that support migration, invasion, and state transitions.

The GluA2 deficiency that underlies Ca²⁺-permeable AMPAR expression in gliomas represents a potentially attractive tumor-specific therapeutic target. In normal mature neurons, AMPARs predominantly contain edited GluA2 subunits, rendering them Ca²⁺-impermeable and limiting their contribution to pathological excitotoxicity under physiological conditions. By contrast, glioma cells exhibit downregulation of GluA2 expression or altered RNA editing, leading to enrichment of GluA2-lacking, Ca²⁺-permeable assemblies that are rarely found in healthy surrounding brain tissue. This differential expression pattern creates a potential therapeutic window: agents that preferentially block Ca²⁺-permeable AMPARs may disrupt tumor-promoting synaptic inputs while largely sparing physiological neurotransmission. Whether this Ca²⁺-permeable phenotype is strictly restricted to tumor cells or also occurs in reactive glia or other stromal components within the tumor microenvironment remains incompletely characterized. Limited evidence suggests that peritumoral astrocytes may undergo reactive changes that include altered glutamate receptor expression, but the degree to which they recapitulate the GluA2-deficient, Ca²⁺-permeable AMPAR configuration seen in glioma cells is not well established. Similarly, tumor-associated microglia and infiltrating immune cells have not been systematically evaluated for AMPAR subunit composition. Addressing these gaps is important because off-tumor target engagement could contribute to toxicity or modulate the microenvironment in ways that either support or oppose antitumor efficacy.

Mechanistic work supports active maintenance of this Ca²⁺-permeable configuration. Ramaswamy et al. showed in U87MG GBM cells that MEK-ERK1/2 signaling regulates AMPAR subunit composition at the transcriptional level (42). ERK inhibition reduced GluA1 and GluA4 while increasing GluA2, shifting the overall receptor phenotype toward lower Ca²⁺ permeability and coinciding with reduced invasiveness. This argues that Ca²⁺-permeable AMPARs are not merely epiphenomena of transformation but can function as regulated effectors downstream of oncogenic signaling that sustain invasive behavior (42).

At the network level, Ca²⁺-permeable AMPARs appear to couple synaptic input to tumor-wide propagation of Ca²⁺ signals. Jang and Park emphasized that tumor cells can integrate synaptic glutamatergic inputs while also propagating Ca²⁺ waves through tumor microtube networks and gap junctions (43). In several models, enforced expression of a dominant-negative GLUA2 construct (GLUA2-DN) that reduces Ca²⁺ permeability slowed tumor growth, supporting the notion that modulating subunit composition alone can shift disease trajectory in vivo (43).

Overall, the pathological impact of neuron-glioma AMPAR synapses is driven not only by their existence but by a recurring bias toward Ca²⁺-permeable receptor assemblies (41-43). This bias amplifies the biological consequences of each presynaptic spike and provides a mechanistic rationale for AMPAR-directed therapies and for molecular subtyping based on synaptic receptor state. The preferential expression of Ca²⁺-permeable AMPARs in glioma cells relative to adjacent normal neurons underscores the potential for developing selective inhibitors that exploit this differential expression, though further work is needed to confirm the specificity of this phenotype and to rule out significant expression in other cell types within the tumor microenvironment.

Network shaping: from focal synapses to circuit reprogramming

Once AMPAR-mediated electrical input at individual synapses has been established, an obvious question arises: how do these microscopic events summate to produce network remodeling detectable by EEG, magnetoencephalography (MEG), or intraoperative electrocorticography? In a Trends in Cancer review, Gonzales et al. conceptualized malignant gliomas as lesions that actively reprogram neural circuits (44). Through neuron-glioma synapses, tumor microtube networks, and glutamate efflux, gliomas disrupt local excitation-inhibition balance, inducing high-frequency discharges and network synchrony. On this foundation, cortical functional reorganization and long-range connectivity remodeling emerge, providing low-resistance pathways for tumor infiltration and the anatomical substrate for epileptiform activity and cognitive impairment. Within this framework, Ca²⁺-permeable AMPAR-mediated synaptic currents constitute the critical first step from “single-cell plasticity” to “circuit plasticity”. Deeper layers of remodeling are evident at the genetic and epigenetic levels. Using human frontal cortex specimens, Kassaeyan et al. analyzed DNA methylation patterns in gene sets involved in neuro-glioma communication—including synaptic components, axon guidance molecules, and glutamate receptor families—and found that glioma-associated tissues display methylation and co-methylation networks that differ strikingly from those in control tissue (45). These differences extend beyond individual promoters to clusters of synapse- and circuit-related gene modules, suggesting that chronic interactions between tumors and surrounding neural networks give rise to stable “epigenetic circuit fingerprints” that encode memory traces of sustained hyperexcitability and synaptic remodeling. At an even broader scale, Mondal and Huse proposed the concept of a “neurotransmitter power play” (46). Glutamate, GABA, acetylcholine, noradrenaline, and other transmitters, along with their respective receptor combinations and plasticity rules, collectively form a “synaptic communication hub” within brain tumors. Within this hub, excitatory synapses centered on Ca²⁺-permeable AMPARs are regarded as one of the dominant forces driving tumor progression. This perspective emphasizes that neuron-glioma synapses are not isolated phenomena; they are embedded within a larger neurotransmitter network. Circuit remodeling is reflected not only in strengthened local excitatory input but also in cross-modal and cross-regional reconfiguration of functional networks. This multiscale view will be essential for designing “circuit-breaking” intervention strategies in subsequent sections.

Therapeutic strategies: from AMPAR blockade to network remodeling

Within the activity-dependent growth circuit, AMPA receptors act as gatekeepers that convert neuronal firing into tumor electrophysiological responses and Ca²⁺ signals. They therefore represent one of the most intuitive and practically accessible nodes for pharmacologic intervention. In a 2025 review, Radin summarized the preclinical and clinical evidence for AMPAR-targeted strategies in HGG (47). Preclinical studies have shown that AMPAR antagonists suppress tumor proliferation, migration, and glutamate release in vitro and in vivo, and that blocking Ca²⁺-permeable AMPARs weakens neuro-tumor synaptic drive and prolongs survival in animal models. The review also emphasized that perampanel, a third-generation noncompetitive AMPAR antagonist approved as an antiseizure medication, combines favorable blood-brain barrier (BBB) penetration, an established safety profile, and extensive clinical experience, making it an attractive candidate for rapid translation of AMPAR-targeted approaches into glioma trials. On the clinical translation front, the NOA-30 PerSurge trial initiated by Heuer et al. marks an important milestone in advancing the concept of “closing neuro-tumor synapses” into patients with recurrent or progressive GBM (48). This multicenter, phase IIa, randomized, double-blind, placebo-controlled study enrolls patients scheduled for resection of progressive GBM and assigns them 1:1 to perampanel or placebo. Treatment begins approximately 30 days before surgery and continues through the postoperative observation period. The trial is designed to assess changes in the density and functional properties of neuro-tumor synapses in resected tissue, alongside radiographic growth rates and seizure burden. Unlike conventional trials that focus solely on radiographic response or survival, PerSurge builds the concept of “disconnecting the neuro-tumor network” into its primary research question and explores AMPAR antagonists as genuine network-targeting therapies.

Complementing such mechanistic trials is experience from real-world epilepsy cohorts. In a retrospective study of postoperative brain tumor patients, Hino et al. compared perampanel monotherapy with classic levetiracetam monotherapy and found comparable seizure control between groups, with a suggestion of better radiographic tumor control in a subset of perampanel-treated patients and no increase in severe toxicity (49). Although limited in size and not restricted to GBM, this study provides important prior information for future trials that explicitly incorporate dual endpoints of seizure control and tumor control. At the preclinical level, Nishide et al. further combined AMPAR antagonism with standard chemotherapy (50). In mouse GBM models, perampanel plus temozolomide more effectively suppressed tumor growth and prolonged survival than either agent alone and, at the cellular level, showed synergistic induction of cell death and reduced glutamate release. These results support a “synapse blockade + cytotoxic” strategy and could, in future clinical trials, be incorporated into a broader “noise reduction, loop closure, and consolidation” paradigm: first dampening circuit drive with an AMPAR antagonist, then consolidating tumor control with cytotoxic agents or radiotherapy. More broadly, the pharmacologic positioning of AMPAR blockade benefits from a cancer neuroscience perspective. Colobon and Lee have proposed the concept of “neural vulnerability”, advocating repurposing existing neuroactive drugs—including AMPAR antagonists, GABAergic modulators, and calcium channel blockers—to exploit GBM’s dependence on neural circuits (51). Within this framework, perampanel is no longer viewed merely as an antiseizure drug, but as a network-modulating tool that engages the core of the activity-dependent growth circuit. In parallel, Beichert et al. reviewed innovative strategies targeting the “network architecture” of GBM, including AMPAR antagonists, network-informed radiotherapy paradigms, transcranial or cortical stimulation, and tumor-treating fields, and argued for constructing “orthogonal combinations” of molecular and network-directed interventions to weaken tumor reliance on neuronal activity across multiple dimensions (52).

Overall, intervention strategies centered on AMPA synapses are evolving from simple receptor blockade toward integrated network-level modulation. Designing individualized regimens that both effectively reduce pathological “noise” and avoid excessive suppression of normal network function will require stratifying patients by AMPAR subunit composition, neuro-tumor synapse density, epilepsy phenotype, and NLGN3-ADAM10 axis activity.

From NLGN3 to AMPA: an actionable circuit model

Circuit architecture

Integrating the evidence summarized in Sections 2–4, neuro–glioma interactions can be conceptualized as an activity-dependent positive feedback circuit with defined directionality and amplification. First, local and even remote neuronal firing activates ADAM10 through Ca²⁺ dynamics and intracellular signaling, driving ectodomain shedding of NLGN3 from neuronal and OPC membranes and releasing sNLGN3 into the CSF and extracellular matrix. Second, sNLGN3 functions as a potent pro-tumor ligand that activates FAK-PI3K-mTOR, ERK, and related cascades in glioma cells, inducing upregulation of a “neural/synaptic” gene program. Third, NLGN3-reprogrammed tumor cells accumulate Ca²⁺-permeable AMPARs and postsynaptic density proteins at their membrane, rendering them more prone to form bona fide neuron-glioma synapses with surrounding excitatory axons. Fourth, integration of AMPA synapses provides glioma cells with phase-locked excitatory input, triggering intercellular Ca²⁺ waves and electrical coupling within the tumor network, thereby exacerbating local hyperexcitability and epileptiform discharges. Ultimately, seizures and subclinical high-frequency activity further elevate neuronal firing, driving additional NLGN3 shedding and AMPA synaptic plasticity and closing the “activity-shedding-signaling-synapse-re-activity” loop (53). This loop is not restricted to a small cortical patch but is embedded within the broader hierarchy of CNS neuromodulation. Systematic reviews indicate that neuronal activity across cortico-thalamic circuits, limbic structures, white matter tracts, and deep nuclei influences the initiation, progression, and dissemination of multiple CNS tumors through endocrine, neurotransmitter, and axonal-projection pathways (54). In gliomas, such “neural influence” is particularly evident in activity-dependent secretion of soluble factors, synaptic integration, and selective remodeling of regional plasticity, biasing tumor expansion toward highly connected, hyperexcitable networks. Thus, the NLGN3-ADAM10 axis and AMPA synapses are not isolated “molecular events” but core interfaces embedded in multiscale neural circuit regulation.

From a broader cancer neuroscience perspective, neuro-tumor crosstalk appears to be a general phenomenon characterized by three recurrent themes: neural networks sculpt the tumor microenvironment; synapses and neurotransmitters modulate tumor immunity and metabolism; and systemic axes involving stress, circadian rhythms, and neurotransmitter spectra shape a tripartite neuro-tumor-immune relationship (55). In the brain, many of these regulatory pathways converge on high-bandwidth interfaces exemplified by NLGN3 shedding and AMPAR-mediated synapses. Integrating these processes into a single actionable circuit model facilitates the search for “break points” across molecular, cellular, network and systems levels, rather than treating isolated manifestations in a piecemeal fashion, and this activity-shedding-signaling-synapse-reactivation loop is schematically depicted in Figure 2.

Figure 2 Activity-dependent NLGN3-ADAM10-AMPA circuit in high-grade glioma. Neuronal and OPC activity drives ADAM10-mediated NLGN3 shedding, tumor-intrinsic signaling and “high-neural” reprogramming, formation of Ca²⁺-permeable neuron-glioma AMPA synapses, and network hyperexcitability, closing a positive feedback loop of activity-dependent glioma growth. AMPA, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid; OPC, oligodendrocyte precursor cell.

Key parameters: from firing frequency to spatial topology

To transform this conceptual circuit into a computable and stratifiable “therapeutic coordinate system”, several key parameters that determine circuit strength and stability must be specified. The first set of parameters relates to the temporal structure of the driving signal: average neuronal firing rate, burst versus tonic firing patterns, discharge duration, and the degree of synchrony across neuronal populations. Multiscale mathematical models have shown that, for a given intrinsic proliferative and migratory capacity of the tumor, small differences in local metabolic rate and nutrient supply can substantially alter the speed and morphology of the invasive front. When “neuronal activity” is incorporated as an additional driving term, model sensitivity to parameter perturbations is further amplified, with “subcritical-supercritical” transitions in growth dynamics emerging around specific activity thresholds (56). A second group of key parameters resides at the molecular and cellular amplification levels. Multiscale modeling from receptor binding to cell migration indicates that receptor-ligand affinity, downstream amplification factors, and directional bias in cell motility can be systematically mapped onto flux coefficients in macroscopic partial differential equations, thereby explicitly linking receptor-scale parameters to invasion front velocity and anisotropic diffusion tensors (57). Another class of three-dimensional multiscale moving-boundary models emphasizes phenotypic diversity among edge cells and microenvironmental heterogeneity (58). With total tumor volume and mean proliferation rate held constant, simply altering the migratory bias of “frontline cells” and the strength of microtube-mediated connectivity yields strikingly different patterns of postoperative recurrence and cavity regrowth. Once “neurogenic drive” is introduced into such models—for example, by using NLGN3 concentration or excitatory synapse density as local weights on proliferation and migration—it becomes theoretically possible to estimate how different combinations of circuit parameters shape tumor expansion and its dependence on resection margins. A third class of critical parameters arises from cross-scale, data-driven computational frameworks. Studies coupling intraoperative electrophysiology, radiomics, single-cell omics, and numerical modeling have shown that machine learning and reinforcement learning methods can, even in relatively small cohorts, support individualized predictions of invasion speed, recurrence location, and treatment response, and can inform optimized radiotherapy or drug scheduling (59). Within such frameworks, NLGN3 levels, electrophysiologically defined hyperexcitable regions, the proportion of Ca²⁺-permeable AMPARs, and OPC spatial distributions can all serve as feature variables for model fitting. This allows the originally qualitative notion of “high vs. low circuit activity” to be converted into quantitative risk scores and estimates of therapeutic windows.

A circuit-breaking matrix: from activity dampening to network closure

Within the NLGN3-ADAM10-AMPA loop, nearly every arrow corresponds to at least one class of potentially actionable intervention. To avoid compensatory mechanisms and toxicity associated with single-point attacks, one can construct a “circuit-breaking matrix” organized by disease stage and phenotype. Vertically, the matrix is structured by circuit stages. Horizontally, by intervention modalities. Each cell represents a category of combination entry points. Recent reviews of current and emerging surgical, radiotherapeutic, and systemic approaches in GBM have highlighted tumor-treating fields, intraoperative electrophysiological navigation, and adapted radiotherapy fractionation schemes as potential “network-level” interventions that could be layered with NLGN3/AMPAR-targeted agents to achieve multiscale circuit disruption (60). From a trial design standpoint, adaptive platform studies such as GBM AGILE provide a natural framework (61). Through Bayesian response-adaptive randomization and a master protocol, multiple drug combinations can be evaluated concurrently across different molecular subtypes and clinical subgroups, with enrollment probabilities adjusted dynamically based on interim results. If “neuro-tumor circuit activity” were incorporated as a stratification factor—operationalized, for example, as a composite score integrating imaging, electrophysiology, and NLGN3 measurements—one could envisage distinct branches within the same platform: in the high-activity branch, combinations of “activity dampening + shedding inhibition + synapse blockade” would be prioritized, whereas in the low-activity branch, strategies centered on immune or metabolic pathways might take precedence. From a target-centric viewpoint, recent cancer neuroscience reviews have explicitly proposed “reprogramming neural-tumor crosstalk” as a core design principle: shifting the focus from suppressing tumor-intrinsic pathways alone to reducing neural drive, disrupting key synapses, and interfering with neuro-immune coupling, thereby converting tumors from a “neural-dependent” to a state more responsive to conventional cytotoxic or immunotherapies (62). In the circuit model of this review, these strategies map naturally onto five broad categories: “activity dampening”, “shedding inhibition” (ADAM10/17 blockade), “signal disruption” (LYN-ADAM10, FAK-PI3K-mTOR), “synapse blockade” (AMPAR and related receptor antagonism), and “network intervention”. Together they define a flexible circuit-breaking matrix that can be deployed differentially across disease stages and phenotypic profiles.

Combination and timing: from “noise reduction first” to “consolidation later”

In practical treatment design, it is not sufficient to enumerate which links to break; it is equally crucial to determine the order in which they should be targeted and how these interventions should be integrated with existing radiochemotherapy and immunotherapies along the time axis. Drawing on the evidence reviewed above and new trial concepts emerging from cancer neuroscience, one can propose a relatively operational sequence: noise reduction first, loop closure second, consolidation last. “Noise reduction first” refers to prioritizing reduction of neuronal activity and network synchrony in patients with high seizure burden, widespread high-frequency discharges on EEG/MEG, or functional imaging evidence of marked hyperexcitability. This can be approached through antiseizure medications (such as AMPAR antagonists and GABAergic modulators) and/or non-invasive or invasive neuromodulation. “Loop closure” denotes the subsequent introduction—once activity levels have been stabilized—of more mechanism-focused agents such as ADAM10 inhibitors, NLGN3 decoys, or AMPAR/CSPG4-PIEZO1 blockers to interrupt midstream shedding and synaptic amplification. “Consolidation” corresponds to phases in which radiotherapy, temozolomide, tumor-treating fields, or immunotherapies are employed to capitalize on a window of reduced neural dependence and to reinforce cytotoxic or immune-mediated tumor control (63).

Real-world implementation of such sequential or parallel strategies will require refined biomarkers and appropriate trial designs. Statistical and methodological work has already outlined “biomarker-guided adaptive designs” for early-phase oncology trials, in which interim analyses in phase I/II can be used to dynamically enrich the study population—for example, steering from an all-comers cohort toward a subgroup with high NLGN3 levels and pronounced network hyperexcitability—while preserving overall efficiency (64). Combining such designs with cancer neuroscience-specific readouts—including CSF/plasma NLGN3, EEG/MEG synchrony indices, functional magnetic resonance imaging (MRI) connectivity, and single-cell/spatial omics-derived measures of AMPAR/NLGN3 axis activity—could enable genuinely circuit-driven clinical trials. In these studies, the temporal logic of “noise reduction first, loop closure second, consolidation last” would be reflected not only in the treatment schema but also in stratification criteria and adaptive decision rules that fully leverage information on neuro-tumor interactions. These considerations can be organized into a circuit-breaking framework spanning the NLGN3-ADAM10-AMPA axis, as illustrated in Figure 3.

Figure 3 Circuit-breaking framework targeting the NLGN3-ADAM10-AMPA axis. Therapeutic strategies are mapped onto the activity-shedding-signaling-synapse-reactivation circuit, including activity dampening, inhibition of NLGN3 shedding, blockade of downstream signaling and Ca²⁺-permeable AMPA synapses, and network-level interventions, and are organized along a three-step treatment sequence of noise reduction, loop closure, and consolidation. AMPA, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid.

Translational and clinical progress

ADAM10 inhibitors: from concept to PBTC-056/NCT04295759

From a translational perspective, the most direct pharmacologic entry point into the NLGN3-ADAM10 axis is inhibition of ADAM10. In an updated review on adult glioma management, Kirby and Finnerty highlighted blocking ADAM10-mediated NLGN3 shedding as a leading neuro-tumor strategy, framing it as a third therapeutic dimension that is relatively independent of canonical tumor-intrinsic axes such as IDH and O⁶-methylguanine-DNA methyltransferase (MGMT), and therefore potentially complementary to radiochemotherapy and intracellular targeted agents (65). In parallel, reviews focused on brain tumor networks emphasize that NLGN3 shedding and neuron-glioma synapses together act as an upstream control point for diffuse glioma network integration and invasion, positioning ADAM10 inhibition as an attempt to attenuate upstream neural drive and downstream synaptic coupling in the same causal chain (66,67).

Against this backdrop, the oral ADAM10/17 inhibitor INCB7839/INCB007839 (aderbasib) has entered pediatric and adult HGG trials. The pediatric PBTC-056 study (NCT04295759), conducted by the Pediatric Brain Tumor Consortium, is the first phase I trial explicitly designed to target microenvironmental NLGN3 in children and adolescents aged 3 to 21 years with recurrent or progressive HGGs, including DIPG, DMG, GBM, and anaplastic astrocytoma. The trial employs a twice-daily oral fixed-dose-escalation design with a rolling-6 enrollment scheme to determine the maximum tolerated dose and recommended phase II dose. Key eligibility criteria include measurable disease by imaging, failure of standard therapy, and adequate organ function. Primary endpoints are safety, tolerability, and pharmacokinetics, with secondary endpoints encompassing preliminary antitumor activity and pharmacodynamic assessment of target engagement. A notable feature of the design is the requirement for routine low-molecular-weight heparin prophylaxis, implemented after thrombotic toxicities observed in early cohorts, highlighting the class-effect safety considerations associated with ADAM10 inhibition. While efficacy outcomes are not yet mature, the study establishes clinical feasibility of targeting microenvironmental NLGN3 shedding in a real-world pediatric HGG population and sets the stage for future biomarker-informed phase II/III designs and combination strategies.

A critical challenge for ADAM10-directed therapies is achieving sufficient drug exposure within the brain parenchyma while minimizing systemic toxicity. The BBB and the heterogeneous blood-tumor barrier (BTB) present substantial obstacles to CNS delivery of small-molecule inhibitors. In the case of INCB7839, preclinical data suggest variable brain penetration, and clinical experience from adult solid tumor trials indicates that dose escalation is often limited by peripheral toxicities such as deep vein thrombosis, rather than by inadequate CNS exposure. This creates a paradox: doses that achieve meaningful target engagement in the brain may exceed the tolerability limits defined by systemic adverse events. Strategies to overcome this limitation include rational combination with BBB-disrupting agents, convection-enhanced delivery, or nanoparticle formulations that preferentially accumulate in tumor tissue. Additionally, CSF proteomics may serve as a surrogate for intratumoral drug concentration, though the relationship between CSF levels and target occupancy within the tumor core remains to be validated.

From a broader oncology standpoint, ADAM proteases exhibit multi-substrate, multi-phenotype effects across tumor types. Reviews emphasize that ADAM10 cleaves substrates spanning Notch, EGFR ligands, and immune regulatory molecules, and may contribute to tumor stemness programs and immune evasion (70). This underscores both the pharmacologic weight of the target and the risk that systemic inhibition could produce multi-organ toxicities involving immune, cardiovascular, and nervous systems.

In light of dose-limiting deep vein thrombosis reported in prior INCB7839 trials in breast cancer and lymphoma (20), advancing ADAM10 inhibition in pediatric HGG will likely require careful balancing of efficacy and safety across dose, treatment duration, concomitant anticoagulation, and combination with radiochemotherapy. Beyond systemic safety, the brain delivery challenge further complicates dose selection: a dose sufficient to achieve therapeutic levels in the tumor may exacerbate peripheral toxicities, while a dose limited by safety may fail to engage the target within the brain. Biomarker-driven enrichment, such as selection of tumors with higher NLGN3 expression and stronger neural integration, is likely to be important for identifying patients most likely to benefit from ADAM10 inhibition, thereby improving the risk-benefit ratio.

AMPAR antagonism: the clinical trajectory of perampanel

Within the activity-dependent growth circuit, AMPA receptors occupy the key gate through which neuronal firing is translated into tumor membrane depolarization and Ca²⁺ influx. On this basis, the noncompetitive AMPAR antagonist perampanel has gradually been repositioned from a pure antiseizure drug to a potential network-targeted antitumor agent. In an open-label controlled study of newly diagnosed, pre-chemoradiation HGG, Tobochnik et al. compared preoperative loading plus maintenance perampanel with standard levetiracetam on peritumoral hyperexcitability and clinical outcomes (71). The two groups exhibited similar levels of intraoperative high-frequency oscillations (HFOs) as a marker of cortical hyperexcitability, but over a median follow-up of 281 days, the incidence of clinical seizures was substantially lower in the perampanel arm (14% vs. 50%), with good tolerability and drug levels sufficient to reach effective brain concentrations. Although the sample size was limited and no significant survival benefit was observed, the study demonstrated the feasibility and tolerability of a perioperative high-dose perampanel strategy in HGG and provided dose and scheduling guidance for larger phase II trials such as NOA-30 PerSurge.

In the broader population of patients with brain TRE (BTRE), systematic reviews summarizing retrospective series and prospective cohorts indicate that, as add-on therapy, perampanel achieves ≥50% seizure reduction in approximately 60–70% of BTRE patients, with some small series reporting seizure freedom in over half of treated individuals (72). Overall tolerability is good; the main adverse events are dizziness, irritability, and fatigue, which are usually manageable with slow titration and dose adjustment. A multicenter retrospective analysis further suggests that adherence and discontinuation rates for perampanel are comparable to those of other newer antiseizure medications, and that seizure control may be particularly favorable in certain molecular subtypes, such as IDH-mutant low-grade gliomas, hinting at possible subtype-specific efficacy (73).

With respect to antitumor activity, in vitro studies show that perampanel inhibits cell viability and migration in several malignant glioma cell lines in a dose-dependent manner, induces apoptosis, and displays synergistic effects with temozolomide in selected models (74). These effects are associated with AMPAR blockade and downregulation of multiple genes involved in adhesion and invasion. However, animal studies and early clinical data have yet to demonstrate a clear survival benefit attributable to direct antitumor effects. At present, the most cautious positioning is to regard perampanel as a “dual-function” agent: it reduces peritumoral hyperexcitability and seizure burden through AMPAR blockade, thereby indirectly weakening the activity-dependent growth loop, and it exerts modest anti-invasive and sensitizing effects at the cellular level that may be leveraged in combination with radiochemotherapy or ADAM10 inhibitors as part of a “synapse blockade + cytotoxic/upstream inhibition” strategy. Whether NOA-30 PerSurge and other phase II/III trials can demonstrate more definitive benefits on the dual endpoints of tumor control and seizure control will be critical for defining the role of AMPAR antagonism within the overall circuit.

Biomarkers: from sNLGN3 to network readouts

Implementing circuit-driven precision therapy requires reproducible biomarkers that quantify the activity state of the NLGN3-ADAM10-AMPA axis and report pharmacodynamic effects. On the fluid side, circulating extracellular vesicles and CSF proteomics are increasingly explored as liquid-biopsy approaches. Systematic reviews of plasma and serum exosomes in glioma indicate that exosomal protein and nucleic acid signatures can reflect tumor grade, molecular subtype, and proliferative state, including enrichment of markers such as EGFR and transcripts including NLGN3 and PTTG1 (75). These observations support the feasibility of incorporating mediators of neuro-tumor crosstalk into liquid biomarker panels. CSF proteomic studies provide complementary leverage. Using high-resolution mass spectrometry, Schmid et al. profiled CSF from patients with malignant brain tumors and reported enrichment of proteins linked to angiogenesis, extracellular matrix remodeling, and immune regulation in glioma CSF, with quantitative associations to tumor burden and histologic subtype, and proposed CSF signatures as diagnostic, monitoring, and pharmacodynamic readouts for emerging therapies (76). A subsequent mini-review further noted that multiple proteases and their substrates, including matrix metalloproteinases (MMPs) and ADAM family members, associate with recurrence and invasion, creating a methodological basis for incorporating NLGN3 or cleavage products into CSF panels (77).

Regarding the detection of sNLGN3 in biofluids, current evidence remains preliminary but promising. Several groups have reported measurable sNLGN3 levels in conditioned media from neuron-glioma co-cultures and in CSF from glioma-bearing mice following optogenetic stimulation, with levels correlating with neuronal firing intensity (18,34). In human studies, a small-scale CSF proteomic analysis identified NLGN3 peptides in a subset of HGG patients, though the sensitivity and specificity of these measurements require further validation in larger cohorts. Ongoing clinical trials, including PBTC-056, have incorporated exploratory CSF and plasma collections to assess whether sNLGN3 levels correlate with target engagement of ADAM10 inhibitors and whether changes in sNLGN3 serve as pharmacodynamic indicators. However, several technical challenges remain: the lack of standardized immunoassays for sNLGN3, the instability of the cleaved ectodomain in biofluids, and the potential confounding contribution of sNLGN3 derived from normal neurons and OPCs, which limits the specificity for tumor-driven shedding. Future efforts to develop ultrasensitive assays, such as digital enzyme-linked immunosorbent assay (ELISA) or immunoprecipitation-mass spectrometry platforms, will be critical for advancing sNLGN3 as a clinically actionable biomarker.

An even more forward-looking direction is the use of network-level electrophysiological readouts as pharmacokinetic and pharmacodynamic markers. Unlike soluble biomarkers that reflect steady-state pathway activity, electrophysiological measures such as EEG-derived HFOs, resting-state functional connectivity, and peritumoral hyperexcitability indices can capture real-time changes in circuit dynamics in response to therapeutic interventions. The rationale for using network readouts as pharmacokinetic markers rests on the premise that drugs targeting the NLGN3-ADAM10-AMPA axis exert their effects directly on neuronal excitability and synaptic coupling. For instance, AMPAR antagonists such as perampanel produce measurable reductions in cortical HFOs within hours of administration, and these changes correlate with plasma drug levels in patients. Similarly, ADAM10 inhibitors would be expected to reduce activity-dependent NLGN3 shedding and consequently dampen network hyperexcitability, a change that could be tracked longitudinally with EEG or MEG. The NOA-30 PerSurge trial has taken initial steps in this direction by incorporating perioperative electrocorticography to assess synaptic activity before and after perampanel treatment. Future trial designs could integrate continuous EEG monitoring or wearable devices to capture dynamic changes in network activity as surrogate endpoints for target engagement. A major advantage of such network readouts is their non-invasive nature, enabling repeated sampling without the need for lumbar punctures or blood draws. However, several questions remain to be addressed: whether network changes precede or follow radiographic response, how to distinguish drug-induced effects from tumor-driven circuit remodeling, and whether baseline network activity predicts therapeutic sensitivity. Addressing these questions will require integration of pharmacokinetic sampling, pharmacodynamic biomarker measurement, and continuous electrophysiological monitoring in early-phase trials, ultimately enabling the development of closed-loop adaptive dosing strategies that adjust therapy based on real-time circuit state.

Functional biomarkers rely primarily on electrophysiology and network imaging. In a prospective study, Tobochnik et al. reported that widespread, early high-frequency discharges and hyperexcitability on preoperative EEG in newly diagnosed IDH-wildtype glioma were strongly associated with poorer OS (78). This suggests that the intensity of the activity-linked circuit state can be captured with clinically deployable electrophysiological metrics. Multidisciplinary reviews further argue for systematic integration of EEG, MEG, and functional MRI connectivity with behavioral assessments, and propose degree of network remodeling as a dimension for prognostication and treatment-response evaluation (79,80). Within this framework, peritumoral hyperexcitability indices, network synchrony, and functional connectivity can be interpreted as macroscopic projections of NLGN3-AMPAR circuit activity.

At the molecular-cellular-spatial level, single-cell and spatial transcriptomics provide high-resolution maps of synapse-enriched and neural-like tumor programs. A systematic review of spatial transcriptomics in GBM emphasized that platforms such as Visium and GeoMx, integrated with single-cell RNA-seq, can resolve regional states within tumors, including neurodevelopmental-like, spatial OPC-like, and reactive hypoxia programs, and can localize synaptic, axon guidance, and immunosuppressive modules to defined microregions (81). Complementing these approaches, Curry et al. applied patch-seq to pair patch-clamp recordings with single-cell transcriptomes from human glioma cells, identifying subsets with hybrid phenotypes that can fire action potentials while expressing synapse-related genes alongside interneuron- and OPC-associated markers (82). Together, these data support construction of multimodal biomarker systems combining CSF or blood signatures, EEG/MEG network metrics, and single-cell or spatial molecular readouts in the same patient to quantify axis activity and therapeutic engagement.

Developmental challenges: selectivity, safety windows, and endpoint design

Despite the strong mechanistic appeal of ADAM10 inhibition and AMPAR antagonism, several key challenges must be addressed before they can realistically become components of standard care. The first concerns selectivity and the therapeutic window. Recent functional reviews of ADAM10 underscore its broad involvement in neurodevelopment, immune regulation, and tumor progression, with substrates spanning Notch, EGFR ligands, PD-L1, NKG2D ligands, and other critical signaling axes (70,83). Systemic ADAM10 inhibition therefore carries potential risks of developmental toxicity, immune suppression or hyperactivation, and cardiovascular and metabolic complications. The observation that deep vein thrombosis was dose-limiting for INCB7839 in other solid tumor trials further suggests that routine anticoagulation and close monitoring of coagulation status will be required in pediatric HGG studies (68). How to balance sufficient suppression of NLGN3 shedding against preservation of physiological ADAM10 functions remains a central question for dose selection and population enrichment. A second challenge lies in BBB/BTB penetration and pharmacodynamic readouts. CSF proteomics studies indicate that glioma patients’ CSF is enriched with numerous blood-derived and tumor-derived proteins, reflecting a state of incomplete but highly heterogeneous BBB/BTB disruption (76). This heterogeneity offers an opportunity for drug entry but also implies substantial interpatient variability in CNS drug exposure. In the absence of direct measurements of intratumoral drug concentrations, trial designs will need to rely on CSF/blood biomarkers and functional readouts to indirectly infer whether ADAM10 inhibition or AMPAR blockade has reached an effective level in the brain. A third major issue concerns endpoint selection and companion diagnostics. Reviews on the management of BTRE and GAE point out that most glioma trials still use OS and progression-free survival (PFS) as primary endpoints, while seizure control, ≥50% seizure reduction rates, antiepileptic drug burden, and quality-of-life metrics—outcomes closely linked to network hyperexcitability—are rarely collected or reported in a standardized fashion (81,84). Recent systematic assessments argue that seizure outcomes are “seriously underreported” in glioma trials and recommend incorporation of standardized epilepsy-specific endpoints into the design of HGG studies to better capture the aspects of disease burden that matter most to patients (82). Consensus statements for newly diagnosed and recurrent GBM by SNO/EANO likewise emphasize that evaluation of new therapies should combine radiographic response, neurological function, neurocognition, and quality of life, and that “network remodeling” and “seizure control” should be considered key auxiliary endpoints where appropriate (85). Finally, the relative weight of neurocognition and long-term functional preservation in treatment decision-making is steadily increasing. Recent work shows that patients with HGG experience substantial cognitive decline throughout the disease course, driven by the combined effects of tumor burden, location, and treatment-related injury (86). In this context, introducing ADAM10 inhibitors and AMPAR antagonists into therapeutic regimens raises not only the questions “Can we slow tumor growth and reduce seizures?” but also “Do we do so at the expense of normal network plasticity and cognitive function?” This underscores the need to incorporate neurocognitive batteries and network imaging into core outcome sets from the outset of trial design, rather than relegating them to optional exploratory endpoints.

Overall, translational efforts centered on the NLGN3-ADAM10-AMPA axis have progressed from conceptual proposals to early clinical exploration, but routine clinical implementation will require advances along four dimensions: target selectivity, drug delivery, biomarker systems, and endpoint design.

Conclusions

Taking “activity-dependent growth” as its central theme, this review has assembled, across lines of evidence ranging from experimental models to molecular biology and systems neuroscience, a coherent circuit model of neuro-glioma interaction. Firing of host neurons and OPCs activates ADAM10 in a Ca²⁺- and activity-dependent manner, leading to proteolytic shedding of NLGN3 from the cell surface and thereby translating transient electrical events into a diffusible pro-tumor factor. sNLGN3 then engages multiple pathways, including FAK-PI3K-mTOR, and drives glioma cells into a “neural/synaptic” transcriptional state. In this state, tumor cells further upregulate NLGN3 and ADAM10, establishing an autocrine feed-forward amplifier. On this basis, glioma cells accumulate Ca²⁺-permeable AMPARs and postsynaptic density proteins at the membrane, form bona fide neuron-glioma synapses with excitatory axons, and acquire rhythmic electrical input and pronounced Ca²⁺ oscillations. Local and long-range networks, in turn, develop high-frequency discharges, epilepsy, and remodeled functional connectivity, which feed back to increase neuronal activity and close the loop of “activity-shedding-signaling-synapse-re-activity”. This circuit is not a purely conceptual construct but is underpinned by convergent, multi-level evidence. Animal and in vitro studies provide causal support; human tissue and intraoperative electrophysiology reveal genuine neuro-glioma electrical coupling and network reorganization; single-cell and spatial omics delineate “high neural-integration/synapse-enriched” tumor states; and mathematical and computational models indicate that modest perturbations of circuit parameters can reshape invasion patterns and recurrence trajectories. In this sense, NLGN3-ADAM10-mediated shedding and AMPA synapses jointly define the dual core of activity-dependent growth in HGG and represent some of the most accessible targets within neuro-tumor crosstalk for pharmacologic and device-based intervention.

From a clinical standpoint, the NLGN3-ADAM10-AMPA axis is not envisioned as a replacement for surgery, radiochemotherapy, or existing targeted therapies, but rather as an additional therapeutic dimension for selected subgroups of patients. It is particularly relevant in individuals who present with a pronounced epileptic phenotype accompanied by EEG or MEG evidence of widespread high-frequency discharges or network hypersynchrony, in those whose tumor tissue or liquid biopsies indicate high NLGN3 expression with enhanced ADAM10 activity, and in tumors in which single-cell or spatial transcriptomic or proteomic analyses demonstrate AMPAR enrichment together with activation of synaptic and axon-guidance gene modules. In such settings, a strategy that combines activity dampening, inhibition of NLGN3 shedding, and blockade of AMPA synapses can be considered as a preferential add-on to standard radiochemotherapy and any necessary intracellular targeted agents. More concretely, patients with a substantial seizure burden and clear evidence of network hyperexcitability may first benefit from interventions that attenuate upstream circuit drive, such as AMPAR antagonists and other antiseizure medications, complemented when appropriate by neuromodulatory techniques. In tumors with strongly active NLGN3/ADAM10 signaling, ADAM10 inhibition or NLGN3 decoy approaches may be explored within an acceptable safety window. In cases with marked AMPAR enrichment and convincing structural or functional evidence of neuron-glioma synapses, AMPAR blockade can be regarded as a key step in “closing the loop”, ideally combined with radiotherapy, temozolomide, tumor-treating fields, or other local and systemic therapies. In other words, this circuit is best conceived as a finely stratified, modular addition to existing treatment, rather than a universal strategy to be applied indiscriminately to all patients with HGG.

Several major lines of investigation emerge from this actionable circuit. One concerns the mapping of receptors and co-receptors and the construction of quantitative network models. On the NLGN3 effector side, the CSPG4-PIEZO1 mechanosensory axis is unlikely to be unique; integrins, receptor tyrosine kinases, glypicans, and other molecules likely assemble into higher-order complexes that need to be dissected by proteomics, structural biology, and functional genetics. On the AMPAR side, splice isoforms, RNA editing status, and auxiliary subunit composition need to be integrated into a “parameter map” of Ca²⁺-permeable receptors across tumor subtypes and brain regions. Coupling such molecular data to multiscale mathematical models and network simulations could yield refined indices of circuit activity and more accurate tools for predicting disease course. A second line of work relates to selective ADAM10 inhibition and brain-directed delivery platforms. Existing ADAM10 inhibition remains largely systemic and broad-spectrum. Achieving conformational or substrate selectivity for NLGN3-related cleavage through medicinal-chemical optimization, or constraining drug exposure spatially through nanocarriers, convection-enhanced delivery, or local implants, will be central to determining whether this strategy can achieve an acceptable therapeutic window. A third priority is precise targeting of Ca²⁺-permeable AMPARs. Ideally, pharmacologic agents would preferentially affect GluA2-lacking or unedited receptor assemblies, or specific auxiliary subunits, thereby sparing physiological synaptic transmission as much as possible. Compared with global AMPAR antagonism, such approaches would come closer to an ideal “razor-edge” synaptic intervention. Finally, there is a need for biomarker-driven adaptive clinical trials. Early-phase studies should embed, from the outset, biomarkers reflecting NLGN3-ADAM10-AMPAR axis activity, physiological network measures, and functional connectivity imaging, and use them adaptively to guide cohort enrichment and treatment allocation. Moving from traditional designs organized around static mutations and radiographic endpoints to trials explicitly structured around circuit interruption will require that NLGN3/ADAM10/AMPAR readouts, electrophysiological indices, and connectivity measures become integral components of eligibility, stratification, and decision rules rather than optional exploratory variables.

Translating the NLGN3-ADAM10-AMPA circuit into clinical practice will require an implementation pathway that is both conceptually coherent and practically feasible. At the stages of diagnosis and stratification, conventional histopathology and molecular testing will need to be progressively complemented by assessments centered on circuit activity. These may include baseline EEG or MEG evaluation of hyperexcitability and network synchrony, measurement of NLGN3 and related proteins or exosomal cargo in CSF and blood, immunohistochemical and transcriptomic profiling of AMPARs and synaptic gene modules, and, in specialized centers, targeted use of single-cell and spatial omics to identify patients with highly active circuits. During follow-up and response assessment, a corresponding panel of companion measures will be necessary. Longitudinal tracking of NLGN3 levels, seizure burden, network metrics, and cognitive performance can help determine whether ADAM10 inhibition or AMPAR antagonism is genuinely interrupting the circuit rather than merely dampening overt symptoms. At the same time, aggregating these data at the cohort level will feed back into mathematical and machine-learning models, progressively refining risk stratification and informing treatment decisions. Ultimately, sustained progress in this field will depend on multidisciplinary “cancer neuroscience” teams. Neuro-oncologists, neurosurgeons, epileptologists and electrophysiologists, neuroimaging specialists, neuropsychologists, computational biologists, and basic neuroscientists will need to work together on a shared platform, building a closed-loop translational pipeline that spans basic mechanisms, animal models, clinical trials, and real-world cohorts. Only within such a framework can the NLGN3-ADAM10-AMPA circuit move beyond elegant pathophysiological diagrams to become a set of concrete strategies capable of improving survival and quality of life for patients.

Acknowledgments

None.

Footnote

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2908/prf

Funding: This work was supported by the National Natural Science Foundation of China (grant No. 82160512) and the NHC Key Laboratory of Drug Addiction Medicine (Nos. K1323303 and KN202417).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2908/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Tan AC, Ashley DM, López GY, et al. Management of glioblastoma: State of the art and future directions. CA Cancer J Clin 2020;70:299-312. [Crossref] [PubMed]
2. Ostrom QT, Price M, Neff C, et al. CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2016-2020. Neuro Oncol 2023;25:iv1-iv99. [Crossref] [PubMed]
3. Pan Y, Monje M. Neuron-Glial Interactions in Health and Brain Cancer. Adv Biol (Weinh) 2022;6:e2200122. [Crossref] [PubMed]
4. Taylor KR, Monje M. Neuron-oligodendroglial interactions in health and malignant disease. Nat Rev Neurosci 2023;24:733-46. [Crossref] [PubMed]
5. D'Alessandris QG, Menna G, Izzo A, et al. Neuromodulation for Brain Tumors: Myth or Reality? A Narrative Review. Int J Mol Sci 2023;24:11738.
6. Goethe EA, Deneen B, Noebels J, et al. The Role of Hyperexcitability in Gliomagenesis. Int J Mol Sci 2023;24:749. [Crossref] [PubMed]
7. DrexlerRKhatriRSauvignyTEpigenetic neural glioblastoma enhances synaptic integration and predicts therapeutic vulnerability.bioRxiv 2023;2023.08.04.552017.
8. Xiong W, Zhang X, Peng B, et al. Pan-glioma analyses reveal species- and tumor-specific regulation of neuron-glioma synapse genes by lncRNAs. Front Genet 2023;14:1218408. [Crossref] [PubMed]
9. Xu L, Chen S, Fu Y, et al. Neuro-immune-tumor axis in gliomas: a review of mechanisms, models, and translational opportunities. Front Immunol 2025;16:1682322. [Crossref] [PubMed]
10. Grimi A, Bono BC, Lazzarin SM, et al. Gliomagenesis, Epileptogenesis, and Remodeling of Neural Circuits: Relevance for Novel Treatment Strategies in Low- and High-Grade Gliomas. Int J Mol Sci 2024;25:8953. [Crossref] [PubMed]
11. Tetzlaff SK, Reyhan E, Layer N, et al. Characterizing and targeting glioblastoma neuron-tumor networks with retrograde tracing. Cell 2025;188:390-411.e36. [Crossref] [PubMed]
12. Sun Y, Wang X, Zhang Z, et al. Cholinergic neuron-to-glioblastoma synapses in a human iPSC-derived co-culture model. Stem Cell Reports 2025;20:102534. [Crossref] [PubMed]
13. Charsouei S, Jabalameli MR, Karimi-Moghadam A. Molecular insights into the role of AMPA receptors in the synaptic plasticity, pathogenesis and treatment of epilepsy: therapeutic potentials of perampanel and antisense oligonucleotide (ASO) technology. Acta Neurol Belg 2020;120:531-44. [Crossref] [PubMed]
14. Monje M. The neuroscience of brain cancers. Neuron 2025;113:2734-9. [Crossref] [PubMed]
15. Zhang C, Zhang H, Cao J, et al. Neuroscience in glioma biology Oncol Rep 2025;54:166. (Review). [Crossref] [PubMed]
16. Huang-Hobbs E, Cheng YT, Ko Y, et al. Remote neuronal activity drives glioma progression through SEMA4F. Nature 2023;619:844-50. [Crossref] [PubMed]
17. Drexler R, Drinnenberg A, Gavish A, et al. Cholinergic neuronal activity promotes diffuse midline glioma growth through muscarinic signaling. Cell 2025;188:4640-4657.e30. [Crossref] [PubMed]
18. Hua T, Shi H, Zhu M, et al. Glioma-neuronal interactions in tumor progression: Mechanism, therapeutic strategies and perspectives Int J Oncol 2022;61:104. (Review). [Crossref] [PubMed]
19. Wang Y, Wang Y, Bao H, et al. Mechanisms and therapeutic implications of glioma-neuron interactions. Cancer Lett 2025;629:217884. [Crossref] [PubMed]
20. Venkataramani V, Yang Y, Schubert MC, et al. Glioblastoma hijacks neuronal mechanisms for brain invasion. Cell 2022;185:2899-2917.e31. [Crossref] [PubMed]
21. Taylor KR, Barron T, Hui A, et al. Glioma synapses recruit mechanisms of adaptive plasticity. Nature 2023;623:366-74. [Crossref] [PubMed]
22. Picart T, Hervey-Jumper S. Central nervous system regulation of diffuse glioma growth and invasion: from single unit physiology to circuit remodeling. J Neurooncol 2024;169:1-10. [Crossref] [PubMed]
23. Krishna S, Choudhury A, Keough MB, et al. Glioblastoma remodelling of human neural circuits decreases survival. Nature 2023;617:599-607. [Crossref] [PubMed]
24. Rilinger RG, Guo L, Sharma A, et al. Tumor-related epilepsy in high-grade glioma: a large series survival analysis. J Neurooncol 2024;170:153-60. [Crossref] [PubMed]
25. Chen X, Yang JZ, Kong LY, et al. Research progress in glioma-related epilepsy Biomed Rep 2025;23:167. (Review). [Crossref] [PubMed]
26. Chen H, Shi X, Guo F. Whole-cell patch clamp and extracellular electrophysiology recordings in mouse brain slices. STAR Protoc 2025;6:104008. [Crossref] [PubMed]
27. RodriguezLDiedrichJSunLPatch-Clamp Single-Cell Proteomics in Acute Brain Slices: A Framework for Recording, Retrieval, and Interpretation.bioRxiv 2025;2025.09.15.675920.
28. Chakraborty C, Nissen I, Remeseiro S. What Epigenetics Teaches Us About Neuron-Glioma Interactions. Bioessays 2025;47:e70043. [Crossref] [PubMed]
29. Yuan M, Xi R, Kang Y, et al. Cancer Neuroscience: Decoding Neural Circuitry in Tumor Evolution for Targeted Therapy. Adv Sci (Weinh) 2025;12:e06813. [Crossref] [PubMed]
30. Smith TM Jr, Tharakan A, Martin RK. Targeting ADAM10 in Cancer and Autoimmunity. Front Immunol 2020;11:499. [Crossref] [PubMed]
31. Yan H, Arora S, Hii L, et al. ADAM10 Knockout from Human Glioblastoma and Colon Cancer Cells Modulates Diverse Signalling Networks and Inhibits Tumour Growth In Vivo. Int J Mol Sci 2025;26:10684. [Crossref] [PubMed]
32. Zhang Y, Liao Q, Wen X, et al. Hijacking of the nervous system in cancer: mechanism and therapeutic targets. Mol Cancer 2025;24:44. [Crossref] [PubMed]
33. Dang NN, Li XB, Zhang M, et al. NLGN3 Upregulates Expression of ADAM10 to Promote the Cleavage of NLGN3 via Activating the LYN Pathway in Human Gliomas. Front Cell Dev Biol 2021;9:662763. [Crossref] [PubMed]
34. GillespieSMSeok KimYGeraghtyACNeuroligin-3 interaction with CSPG4 regulates normal and malignant glial precursors through PIEZO1.bioRxiv 2025;2025.07.12.664340.
35. Uchigashima M, Cheung A, Futai K. Neuroligin-3: A Circuit-Specific Synapse Organizer That Shapes Normal Function and Autism Spectrum Disorder-Associated Dysfunction. Front Mol Neurosci 2021;14:749164. [Crossref] [PubMed]
36. Lenzen A, Partap S, Billups C, et al. CNSC-02. Pediatric brain tumor consortium (PBTC)-056: a phase 1 study of the ADAM-10 inhibitor, INCB007839, in children with recurrent/progressive high-grade gliomas to target microenvironmental neurologin-3. Neuro Oncol 2024; [Crossref]
37. Ding M, Liu P, Yuan X, et al. Photogenetic-Like Liposomes Disrupt Neuroligin-3 Dependency to Enhance Glioma Treatment. Adv Mater 2025;37:e2503631. [Crossref] [PubMed]
38. Feng J, Yang J. Glioma-neuron interactions: insights from neural plasticity. Front Oncol 2025;15:1661897. [Crossref] [PubMed]
39. Li Y, Wang Y, Han X, et al. Glioma-derived SPARCL1 promotes the formation of peritumoral neuron-glioma synapses. J Neurooncol 2025;173:515-26. [Crossref] [PubMed]
40. Zhang X, Duan S, Apostolou PE, et al. CHD2 Regulates Neuron-Glioma Interactions in Pediatric Glioma. Cancer Discov 2024;14:1732-54. [Crossref] [PubMed]
41. Zhang L, Wang Y, Cai X, et al. Deciphering the CNS-glioma dialogue: Advanced insights into CNS-glioma communication pathways and their therapeutic potential. J Cent Nerv Syst Dis 2024;16:11795735241292188. [Crossref] [PubMed]
42. Ramaswamy P, Dalavaikodihalli Nanjaiah N, Prasad C, et al. Transcriptional modulation of calcium-permeable AMPA receptor subunits in glioblastoma by MEK-ERK1/2 inhibitors and their role in invasion. Cell Biol Int 2020;44:830-7. [Crossref] [PubMed]
43. Jang HJ, Park JW. Microenvironmental Drivers of Glioma Progression. Int J Mol Sci 2025;26:2108. [Crossref] [PubMed]
44. Gonzales CN, Negussie MB, Krishna S, et al. Malignant glioma remodeling of neuronal circuits: therapeutic opportunities and repurposing of antiepileptic drugs. Trends Cancer 2024;10:1106-15. [Crossref] [PubMed]
45. Kassaeyan V, Arashlow FT, Ahmadi S, et al. Do Glioma Cells Rewire Neural Circuits through Epigenetic Changes? DNA Methylation Analysis of Genes Involved in Neuron-Glioma Communication in the Human Frontal Cortex. J Mol Neurosci 2025;75:155.
46. Mondal J, Huse JT. Neurotransmitter power plays: the synaptic communication nexus shaping brain cancer. Acta Neuropathol Commun 2025;13:85. [Crossref] [PubMed]
47. Radin DP. AMPA Receptor Modulation in the Treatment of High-Grade Glioma: Translating Good Science into Better Outcomes. Pharmaceuticals (Basel) 2025;18:384. [Crossref] [PubMed]
48. Heuer S, Burghaus I, Gose M, et al. PerSurge (NOA-30) phase II trial of perampanel treatment around surgery in patients with progressive glioblastoma. BMC Cancer 2024;24:135. [Crossref] [PubMed]
49. Hino U, Tamura R, Kosugi K, et al. Optimizing perampanel monotherapy for surgically resected brain tumors. Mol Clin Oncol 2024;20:42. [Crossref] [PubMed]
50. Nishide T, Yamamuro S, Sano E, et al. Synergistic effect of perampanel with temozolomide on glioblastoma cells in vivo. Heliyon 2025;11:e43167.
51. Colobon PO, Lee S. Neural Vulnerabilities in Glioblastoma: Rethinking Therapy Through Neuroactive Drug Repurposing. Brain Tumor Res Treat 2025;13:130-9. [Crossref] [PubMed]
52. Beichert J, Hoffmann DC, Winkler F, et al. Innovative Therapeutic Strategies Targeting the Network Architecture of Glioblastoma. Clin Cancer Res 2025;31:2864-71. [Crossref] [PubMed]
53. Liu J, Shi W, Lin Y. Interactive relationship between neuronal circuitry and glioma: A narrative review. Glioma 2022;5:43-9.
54. Lv W, Wang Y. Neural Influences on Tumor Progression Within the Central Nervous System. CNS Neurosci Ther 2024;30:e70097. [Crossref] [PubMed]
55. Huang Q, Hu B, Zhang P, et al. Neuroscience of cancer: unraveling the complex interplay between the nervous system, the tumor and the tumor immune microenvironment. Mol Cancer 2025;24:24. [Crossref] [PubMed]
56. Amereh M, Shojaei S, Seyfoori A, et al. Insights from a multiscale framework on metabolic rate variation driving glioblastoma multiforme growth and invasion. Commun Eng 2024;3:176. [Crossref] [PubMed]
57. Dietrich A, Kolbe N, Sfakianakis N, et al. Multiscale modeling of glioma invasion: From receptor binding to flux-limited macroscopic PDEs. Multiscale Model Simul 2022;20:685-713.
58. Suveges S, Hossain-Ibrahim K, Steele JD, et al. Mathematical modelling of glioblastomas invasion within the brain: A 3D multi-scale moving-boundary approach. Mathematics 2021;9:2214.
59. Bhargav AG, Domino JS, Alvarado AM, et al. Advances in computational and translational approaches for malignant glioma. Front Physiol 2023;14:1219291. [Crossref] [PubMed]
60. Valerius AR, Webb LM, Thomsen A, et al. Review of Novel Surgical, Radiation, and Systemic Therapies and Clinical Trials in Glioblastoma. Int J Mol Sci 2024;25:10570. [Crossref] [PubMed]
61. Cloughesy TF, Alexander BM, Berry DA, et al. GBM AGILE: A global, phase 2/3 adaptive platform trial to evaluate multiple regimens in newly diagnosed and recurrent glioblastoma. J Clin Oncol 2022;40:TPS2078.
62. Liu QQ, Dong ZK, Wang YF, et al. Reprogramming neural-tumor crosstalk: emerging therapeutic dimensions and targeting strategies. Mil Med Res 2025;12:73. [Crossref] [PubMed]
63. Jones G, Anderson JL, Nguyen PTT, et al. Novel approaches to clinical trial design in cancer neuroscience. Neuron 2025;113:2791-813. [Crossref] [PubMed]
64. Serra A, Saint-Hilary G, Guilleminot S, et al. Adaptive Biomarker-Based Design for Early Phase Clinical Trials. Stat Med 2025;44:e70275. [Crossref] [PubMed]
65. Kirby AJ, Finnerty GT. New strategies for managing adult gliomas. J Neurol 2021;268:3666-74. [Crossref] [PubMed]
66. Yang Y, Schubert MC, Kuner T, et al. Brain Tumor Networks in Diffuse Glioma. Neurotherapeutics 2022;19:1832-43. [Crossref] [PubMed]
67. Venkataramani V, Yang Y, Ille S, et al. Cancer Neuroscience of Brain Tumors: From Multicellular Networks to Neuroscience-Instructed Cancer Therapies. Cancer Discov 2025;15:39-51. [Crossref] [PubMed]
68. Stanford University. A phase I study of the ADAM-10 inhibitor, INCB7839, in children with recurrent/progressive high-grade gliomas to target microenvironmental neuroligin-3 (NCT04295759). Stanford Clinical Trials. Retrieved November 29, 2025. Available online: https://clinicaltrials.stanford.edu/trials/i/NCT04295759.html
69. National Cancer Institute. Testing INCB007839 in patients with recurrent, progressive high-grade glioma (NCI-2019-08964). Retrieved November 29, 2025. Available online: https://www.cancer.gov/research/participate/clinical-trials-search/v?id=NCI-2019-08964
70. Arora S, Scott AM, Janes PW. ADAM Proteases in Cancer: Biological Roles, Therapeutic Challenges, and Emerging Opportunities. Cancers (Basel) 2025;17:1703. [Crossref] [PubMed]
71. Tobochnik S, Regan MS, Dorotan MKC, et al. Pilot Trial of Perampanel on Peritumoral Hyperexcitability in Newly Diagnosed High-grade Glioma. Clin Cancer Res 2024;30:5365-73. [Crossref] [PubMed]
72. Tabaee Damavandi P, Pasini F, Fanella G, et al. Perampanel in Brain Tumor-Related Epilepsy: A Systematic Review. Brain Sci 2023;13:326. [Crossref] [PubMed]
73. Rossi J, Cavallieri F, Bassi MC, et al. Efficacy and Tolerability of Perampanel in Brain Tumor-Related Epilepsy: A Systematic Review. Biomedicines 2023;11:651. [Crossref] [PubMed]
74. Tatsuoka J, Sano E, Hanashima Y, et al. Anti-tumor effects of perampanel in malignant glioma cells. Oncol Lett 2022;24:421. [Crossref] [PubMed]
75. Nikoobakht M, Shamshiripour P, Shahin M, et al. A systematic update to circulating extracellular vesicles proteome; transcriptome and small RNA-ome as glioma diagnostic, prognostic and treatment-response biomarkers. Cancer Treat Res Commun 2022;30:100490. [Crossref] [PubMed]
76. Schmid D, Warnken U, Latzer P, et al. Diagnostic biomarkers from proteomic characterization of cerebrospinal fluid in patients with brain malignancies. J Neurochem 2021;158:522-38. [Crossref] [PubMed]
77. Xiao F, Li C, Zhang J, et al. Cerebrospinal fluid biomarkers for brain tumor detection. Oncol Lett 2020;20:1-12.
78. Tobochnik S, Lapinskas E, Vogelzang J, et al. Early EEG hyperexcitability is associated with decreased survival in newly diagnosed IDH-wildtype glioma. J Neurooncol 2022;159:211-8. [Crossref] [PubMed]
79. Maas DA, Douw L. Multiscale network neuroscience in neuro-oncology: How tumors, brain networks, and behavior connect across scales. Neurooncol Pract 2023;10:506-17. [Crossref] [PubMed]
80. Martínez Lozada PS, Pozo Neira J, Leon-Rojas JE. From Tumor to Network: Functional Connectome Heterogeneity and Alterations in Brain Tumors-A Multimodal Neuroimaging Narrative Review. Cancers (Basel) 2025;17:2174. [Crossref] [PubMed]
81. Shireman JM, Cao J, Shih DJH. Spatial transcriptomics in glioblastoma: Is knowing the right zip code the key to the next therapeutic breakthrough? Neurooncol Adv 2023;5:vdad128.
82. Curry RN, Ma Q, McDonald MF, et al. Integrated electrophysiological and genomic profiles of single cells reveal spiking tumor cells in human glioma. Cancer Cell 2024;42:1713-1728.e6. [Crossref] [PubMed]
83. Avila EK, Chamberlain MC, Reijneveld JC, et al. Brain tumor-related epilepsy management. Neurol Res Pract 2023;5:20.
84. Haile H, Shlobin NA, Adapa AR, et al. Seizure outcomes as an understudied metric in glioma clinical trials: A review of the ClinicalTrials.gov database. Neurooncol Adv 2025;8:vdaf237.
85. Wen PY, Weller M, Lee EQ, et al. Glioblastoma in adults: a Society for Neuro-Oncology (SNO) and European Society of Neuro-Oncology (EANO) consensus review on current management and future directions. Neuro Oncol 2020;22:1073-113. [Crossref] [PubMed]
86. Ghadimi K, Abbas I, Karandish A, et al. Cognitive Decline in Glioblastoma (GB) Patients with Different Treatment Modalities and Insights on Untreated Cases. Curr Oncol 2025;32:152. [Crossref] [PubMed]