Neuroligin-3–CSPG4 interaction maintains oligodendrocyte precursor cell progenitor state and promotes glioma proliferation through mechanotransduction

Abstract

Glioma pathophysiology is robustly regulated by interactions with neurons. Key to these interactions is the role of neuroligin-3 (NLGN3), a synaptic adhesion molecule shed in response to neuronal activity that functions as a paracrine factor crucial for glioma growth. Here we elucidate the mechanistic pathway whereby shed NLGN3 interacts with glioma and their normal glial counterparts. NLGN3 binds to chondroitin sulfate proteoglycan 4 (CSPG4, also known as NG2) on both glioma and healthy oligodendrocyte precursor cells (OPCs), facilitating CSPG4 shedding by ADAM10. NLGN3–CSPG4 interactions alter membrane tension, thereby activating mechanotransducers, primarily PIEZO1, leading to membrane depolarization and subsequent ADAM10-mediated CSPG4 shedding. The NLGN3–CSPG4–PIEZO1 pathway maintains OPCs in an undifferentiated, stem-like state and promotes glioma proliferation, underscoring its dual roles in healthy and malignant contexts.

This study finds that shed NLGN3 interacts with the glycosylated surface protein CSPG4 on both normal and malignant glial cells, activating mechanotransduction pathways that drive glioma growth and maintain oligodendroglial progenitor states.

Main

Gliomas are a devastating group of brain cancers. High-grade gliomas such as glioblastoma (GBM) and H3K27M-altered diffuse midline gliomas (DMGs, DMG originating in the brainstem is also called diffuse intrinsic pontine glioma (DIPG)) are the leading cause of primary brain tumor-related death in both adults and children, while low-grade gliomas cause substantial neurological morbidity. Many forms of glioma arise from oligodendrocyte precursor cells (OPCs), and OPC-like cancer cells comprise an important cellular subpopulation within gliomas such as GBM and DMG (for review, please see ref. ¹). Concordantly, many mechanistic parallels exist between healthy OPCs and malignant gliomas. A key similarity is the proliferation-promoting effect of neuronal activity. Glutamatergic neuronal activity robustly promotes the proliferation of both healthy OPCs² and glioma cells³. In OPCs, this activity-regulated proliferation results in the generation of new mature oligodendrocytes and adaptive changes in myelination that tune neural circuit function and contribute to brain functions like attention, learning and memory^2,4–6. Gliomas hijack these mechanisms of glial plasticity to promote growth^3,7–11 and invasion^9–11.

Neuron–glioma interactions are mediated by activity-regulated paracrine growth factors^3,7,8 and by neuron-to-glioma synapses^9,10. One important mechanism underlying activity-regulated glioma growth is the activity-regulated shedding of the synaptic protein neuroligin-3 (NLGN3)^3,7. OPCs are a major cellular source of shed NLGN3 (ref. ⁷). For context, it is important to note that OPCs receive extensive synaptic inputs from neurons^12–14, although the function of neuron-to-OPC synapses is still incompletely understood. NLGN3 shedding occurs through neuronal activity regulation of ADAM10 protease, which cleaves NLGN3 to release the large N-terminal ectodomain into the extracellular microenvironment. Shed NLGN3 promotes the proliferation of both high-grade^3,7,15 and low-grade⁸ gliomas through stimulation of proliferation signaling pathways PI3K-mTOR^3,7, RAS and SRC⁷. Genetic knockout (KO) of Nlgn3 gene from the cells in the brain microenvironment or pharmacological blockade of NLGN3 shedding through ADAM10 inhibition markedly slows glioma growth and progression in preclinical models of high-grade⁷ and low-grade gliomas⁸. The stark dependency of gliomas on NLGN3 signaling can be understood in part by observations that NLGN3 exposure increases glioma gene expression related to multiple neuron–glioma interaction mechanisms; notably, synapse-related gene expression⁷ and concordantly neuron-to-glioma synapses are reduced in the absence of NLGN3 (ref. ¹⁶). Taken together, these observations highlight the therapeutic potential of disrupting NLGN3 signaling in glioma cells. However, a number of pressing questions remain that include mechanisms by which NLGN3 interacts with and signals to the glioma cell, whether similar interactions occur on healthy OPCs, and what role NLGN3 shedding may have in healthy oligodendroglial biology. Here we show that shed NLGN3 engages a shared cell-surface interacting partner on glioma cells and healthy OPCs to activate a mechanotransductive signaling pathway. This interaction increases membrane tension, activates mechanosensitive ion channels, chiefly PIEZO1, and promotes ADAM10-dependent ectodomain shedding. In glioma cells, this signaling mechanism drives proliferation, whereas in healthy OPCs it maintains an undifferentiated progenitor state. These results reveal a shared mechanism through which neuron–glial interactions regulate both malignant glioma growth and normal OPC biology.

Results

Unbiased biochemical screening for NLGN3 binding partners

To screen for putative interaction partners of shed NLGN3, we performed affinity capture chromatography using an immobilized NLGN3 (Extended Data Fig. 1a) column^16,17 and solubilized membranes from patient-derived DIPG/DMG and adult GBM cultures (Fig. 1a,b). Mass spectrometry (Methods) of column eluates identified several putative NLGN3-interacting partners from both cell cultures, prominently including chondroitin sulfate proteoglycan 4 (CSPG4; Fig. 1b). CSPG4 contains two laminin-neurexin-sex-hormone binding globulin domains (Fig. 1c)^18–20, similar to those in neurexins (mediating well-characterized interaction with postsynaptic neuroligins^21,22), highlighting CSPG4 as a likely interaction partner. CSPG4 (NG2) is a canonical marker of OPCs (which were previously referred to as ‘NG2 cells’) and is also highly expressed in a variety of gliomas, particularly in the OPC-like cellular populations^23,24. Of note, the canonical NLGN3 binding partners in neurons—neurexins²⁵—were not found to be expressed in these patient-derived glioma cell models. NRXN gene expression was similarly absent or low in the patient-derived models, and variable but generally low in primary patient samples of DMG and GBM (Extended Data Fig. 1d,e).

Extended Data Fig. 1. Biochemical analysis of NLGN3 interaction with chondroitin sulfate.

a, Coomassie-stained SDS–PAGE gel showing purified recombinant NLGN3 protein. The initial eluted fraction, which contains minor contaminants, was excluded to ensure high-purity material for assays. b, Schematics of size-exclusion chromatography (SEC)-based coupling assay. NLGN3 and its substrates are mixed, run through the column, and the left-shift of the peak will be assessed. c, Binding assay of the NLGN3 with chondroitin sulfate (CS) and heparin. Both heparin and CS are binding to NLGN3, causing the peak to shift to the left. d, Dot plots showing scaled expression levels of NRXN1, NRXN2, and NRXN3 across various cell types and tumor subtypes, including H3K27M diffuse midline gliomas (DMGs), IDH1-mutant glioblastomas, IDH1 wild-type glioblastomas, and normal oligodendrocytes. Datasets were extracted from previously published studies^68,69,73. Individual values as dots, with mean and range indicated (Wilcoxon rank-sum test, *** P < 0.001, **** P < 0.0001). e, Bulk RNA-seq analysis comparing wild-type (WT) and PIEZO1 knockout (KO) MGG8 glioma cells. Expression of NRXN1–3 is minimal or undetectable in both conditions, whereas CSPG4 is robustly expressed (n (independent biological batches) = 2, mean with range). f, Dose-dependent interaction of NLGN3 and CS. g, Quantification of NLGN3-CS interaction shown in c.

Source data

Fig. 1. Unbiased screening identifies CSPG4 as a putative interacting partner of shed NLGN3.

a, Schematic of glioma membrane isolation, solubilization and affinity capture by immobilized NLGN3 ectodomains. NLGN3 affinity column is prepared by immobilizing biotinylated (C-terminal AviTag) NLGN3 ectodomains within a streptavidin resin. b, Selected top hits from the NLGN3 affinity chromatography assay. Purified membrane proteins captured through NLGN3 affinity chromatography were analyzed by mass spectrometry. Please note that both pediatric (DIPGXIII) and adult (MGG8) brain cancers show CSPG4 as a top hit. c, Cartoon representation of CSPG4 functional domains. There are three subdomains in the extracellular region, D1, D2 and D3. The laminin-G (LNS) domains of D1 are analogous to those of neurexins (NRXNs), the canonical binding partners to NLGN3, followed by the signature chondroitin sugar-binding site in the middle of core repeat modules. Numbers in parentheses indicate the amino acid length in each subdomain. d, WB of human CSF with anti-CSPG4 N-term and anti-NLGN3 antibodies. Control patient CSFs are all from patients with ALL. Only CSF samples from DMG-bearing patients show both CSPG4 and NLGN3. e, Quantification of bands in d (n = the number of patients = 4 per group, data represented are individual data points). LNS, laminin-neurexin-sex-hormone binding globulin; ALL, acute lymphocytic leukemia.

Source data

We then tested whether NLGN3 binds chondroitin sulfate (CS), the glycan substrate covalently linked to CSPG4 (Fig. 1c). Using fluorescence size-exclusion chromatography (FSEC; Extended Data Fig. 1b) with recombinant NLGN3 ectodomain, we found that both CS and the established NLGN-interacting partner heparin²⁶ showed a notable binding-induced left shift in FSEC (Extended Data Fig. 1c). Although the CS-induced shift was smaller, CS binding was robustly dose dependent (Extended Data Fig. 1f,g). Together, these biochemical data support CSPG4 as a candidate NLGN3 interactor.

Because both CSPG4 ectodomains and NLGN3 are shed in response to neuronal activity²⁷, we asked whether they are detectable in cerebrospinal fluid (CSF) of human glioma patients. Western blotting of CSF from patients with DIPG/DMG or leukemia control patients detected both NLGN3 and CSPG4 in two of four glioma patients, but in none of the leukemia patients (Fig. 1d,e)⁷. These findings support CSPG4 as a candidate interacting partner for shed NLGN3, while leaving unresolved how these proteins interact and whether this interaction is linked to activity-dependent shedding^3,7,8,28.

NLGN3 exposure augments ADAM10-mediated CSPG4 proteolysis

Previous studies showed that neuronal activity induces ADAM10-dependent shedding of both NLGN3 and CSPG4 ectodomains^7,27. Here we found that NLGN3 shedding precedes CSPG4 shedding after neuronal activity, suggesting that shed NLGN3 may subsequently promote CSPG4 cleavage in glioma cells and OPCs. To test this, we treated glioma cultures with recombinant NLGN3 ectodomain and analyzed conditioned medium for shed CSPG4 by western blotting (Fig. 2a).

Fig. 2. Shed NLGN3 treatment augments ADAM10-dependent CSPG4 cleavage.

a, Experimental workflow for 1-h recombinant sNLGN3 treatment of glioma cells or OPCs, followed by conditioned medium concentration and western blotting for shed CSPG4. b, WB and quantification of CSPG4 shedding from SU-DIPG13 cells treated with vehicle or 100 nM sNLGN3, with or without 5 µM ADAM10 inhibitor. sNLGN3 increased CSPG4 shedding (P = 0.0033), which was blocked by ADAM10 inhibition (P < 0.0001). Chondroitinase ABC confirmed variable CS glycosylation, and C-terminal CSPG4 signal was absent from conditioned medium, confirming ectodomain shedding; n = 4 independent biological replicates, mean ± s.e.m., ordinary one-way ANOVA, P values are shown in b. c, Left, WB of CSPG4 shedding from IDH1 WT GBM model conditioned medium after exposure to 100 nM sNLGN3 for 1 h in the presence or absence of 5 µM ADAM10 inhibitor. NLGN3 treatment increases shedding relative to vehicle controls (P = 0.0013). NLGN3-induced shedding of CSPG4 is blocked by pretreating cells with 5 µM ADAM10 inhibitor (P = <0.0001). Right, quantification of protein band intensities relative to vehicle-treated controls; n (independent biological replicates of WB) = 4, mean ± s.e.m., ordinary one-way ANOVA, P values are shown in c. d, Left, WB of CSPG4 shedding from NF1-mutant optic pathway glioma model conditioned medium after exposure to 100 nM sNLGN3 for 1 h in the presence or absence of 5 µM ADAM10 inhibitor. NLGN3 treatment increases shedding relative to vehicle controls (P = 0.0066). NLGN3-induced shedding of CSPG4 is blocked by pretreating cells with 5 µM ADAM10 inhibitor (P = 0.0004). Right, quantification of protein band intensities relative to vehicle-treated controls; n (independent biological replicates of WB) = 4, mean ± s.e.m., ordinary one-way ANOVA, P values are shown in d. e, Left, WB of CSPG4 shedding from primary mouse OPC conditioned medium after exposure to 100 nM sNLGN3 for 1 h in the presence or absence of 5 µM ADAM10 inhibitor. NLGN3 treatment increases shedding relative to vehicle controls (P = 0.0011). NLGN3-induced shedding of CSPG4 is blocked by pretreating cells with 5 µM ADAM10 inhibitor (P = 0.0003). Right, quantification of protein band intensities relative to vehicle-treated controls; n (independent biological replicates of WB) = 3, mean ± s.e.m., ordinary one-way ANOVA, P values are shown in e. f, Model of NLGN3-induced, ADAM10-dependent CSPG4 ectodomain shedding from glioma cells. Red denotes ADAM10; green denotes shed NLGN3; blue denotes CSPG4 ectodomain. Illustrations in a created in BioRender; Kim, Y. https://biorender.com/ftk7bg7 (2026). ANOVA, analysis of variance.

Source data

We found that glioma cells constitutively shed CSPG4 (Fig. 2b), a process inhibited by the ADAM10 inhibitor GI254023X (5 µM, 15 min; Fig. 2b). Notably, NLGN3 treatment significantly increased CSPG4 shedding (Fig. 2b). The absence of CSPG4 C-terminal signal in conditioned medium confirmed ectodomain-specific shedding, while chondroitinase ABC treatment revealed variable glycosylation of shed CSPG4 (Fig. 2b).

We found that NLGN3-induced CSPG4 shedding occurs across a spectrum of glioma subtypes, including human IDH1 wild-type (WT) GBM (Fig. 2c) and a mouse model of NF1-associated optic glioma⁸ (Fig. 2d). Furthermore, we confirmed that this is not exclusive to malignant cells, as primary OPCs readily shed CSPG4 ectodomains in response to NLGN3 (Fig. 2e). Consistent with previous findings²⁷, NLGN3-induced shedding of CSPG4 was inhibited by pretreating OPCs with an ADAM10 inhibitor. These results collectively demonstrate that NLGN3 exposure is sufficient to enhance ADAM10-dependent shedding of CSPG4 ectodomains from OPCs and glioma cells (Fig. 2f).

Membrane depolarization disinhibits ADAM10 and increases CSPG4 shedding

We then examined how NLGN3 augments ADAM10-dependent CSPG4 shedding. ADAM10-mediated shedding of proteins, like NLGN3 and CSPG4, is regulated in part by an auto-inhibitory c-loop mechanism that obstructs the active site²⁹; an inhibition relieved by membrane depolarization³⁰. We therefore conducted a CSPG4 shedding assay in which cells were exposed to various depolarizing stimuli (Fig. 3)²⁷. First, we visualized depolarization by GCaMP6s imaging in glioma cells^9,10,15; cells were treated with either vehicle or 200 nM NLGN3, followed by wide-field calcium imaging over 20 min (2-min baseline, 18-min post-treatment; Fig. 3a). While spontaneous activity was rare in controls, NLGN3-treated cells showed markedly increased calcium transients (1.7% versus 21.9%; Fig. 3b), consistent with reports implicating NLGN3 and membrane depolarization in glioma progression^3,7,9,10. To further verify the role of membrane depolarization, we treated glioma cells with the voltage-gated K⁺-channel blocker, 4-aminopyridine (4-AP; 2.5 mM)^31–33, which elicited prominent calcium responses (with 98.9% of cells responding), comparable to a previously study³³ (Fig. 3c,d) and increased CSPG4 shedding (Fig. 3e,f, left), without changing glioma cell death (Extended Data Fig. 2a); CSPG4 shedding was markedly attenuated by cotreatment with an ADAM10 inhibitor (Fig. 3f, right), indicating that depolarization-induced CSPG4 shedding is ADAM10 dependent.

Fig. 3. ADAM10-dependent CSPG4 shedding is driven by membrane depolarization.

a, Schematic of the unbiased wide-field calcium imaging workflow. b, Heatmaps of calcium responses in glioma cells treated with vehicle or NLGN3. Red dashed line indicates the time point (2 min) of the treatments; n (number of cells imaged, from three independent biological replicates) = 743 cells for vehicle; n = 3,994 cells for NLGN3. c,d, Calcium imaging of glioma cells treated with 4-AP. c, Heatmap of calcium responses. Red dashed line indicates the time point of the 4-AP treatment (2 min). d, Averaged calcium transients over time (mean ± s.e.m.; n (number of cells imaged, from three independent biological replicates) = 2,378 cells). e, Workflow depicting treatment of IDH1 WT GBM (MGG8) cells for 1 h with the potassium channel blocker 4-AP followed by culture medium concentration and WB to detect shedding of CSPG4. f, Left, representative WB of shed CSPG4 ectodomains from experiment depicted in a. 4-AP treatment increases shedding relative to vehicle controls. Please note that 4-AP-induced shedding (P < 0.0001) of CSPG4 is blocked by pretreating cells with 5 µM ADAM10 inhibitor (P < 0.0001). Right, quantification of WBs; n (independent biological replicates of WB) = 4, mean ± s.e.m., ordinary one-way ANOVA, P values are shown in e. g, Schematic of the optogenetic stimulation experiment. h, Left, representative WB showing CSPG4 ectodomain levels in the culture medium after optogenetic stimulation of increasing durations (470 nm, 1 mW mm⁻², 0–60 s, every 15 min for 1 h). Right, quantification of WB demonstrates a time-dependent increase in CSPG4 shedding (n = 3 independent experiments; data are mean ± s.e.m.; Kruskal–Wallis test, *P < 0.05). i, Schematic of the optogenetic experimental workflow. SU-DIPG6 cells expressing ChR2 were stimulated with blue light using the same parameters described in g. Total membrane depolarization was quantified by integrating the photocurrent over time (pA × ms). j, Left, representative whole-cell patch-clamp recordings of ChR2-expressing glioma cells under the same illumination parameters used in h. Right, quantification of photocurrent integrals (area under the current trace) showing increased current with longer illumination durations (n = 3 independent cells; mean ± s.e.m.; Kruskal–Wallis test, *P < 0.05). k, Schematic workflow of the wide-field calcium imaging assay (left) and the heatmap (right) of calcium responses in glioma cells treated with cytochalasin D. l, Summary of the calcium activity in glioma after different treatments. Calcium activity is measured as AUC of F/F₀; n (number of independent biological replicates) = 3; mean ± s.e.m., two-tailed t-test, **P < 0.01, ***P < 0.001. m, Left, representative WB of shed CSPG4 ectodomains from CSPG4 shedding assay with cytochalasin D treatment. Right, quantification of WBs. Please note that cytochalasin D treatment increases shedding relative to vehicle controls (P = 0.0175). Cytochalasin D-induced shedding of CSPG4 is blocked by pretreating cells with 5 µM ADAM10 inhibitor (P < 0.0001); n (independent biological replicates of WB) = 7, mean ± s.e.m., ordinary one-way ANOVA, P values are mentioned in m. n, Cartoon diagram summarizing the treatments of 4-AP, ChR2 and cytochalasin D that are inducing membrane depolarization. o, Model depicting CSPG4 shedding through NLGN3 interaction that implicates an ion channel that is sensitive to membrane stress to depolarize the membrane and activate ADAM10. Illustrations in a, c, e, g and k created in BioRender; Kim, Y. https://biorender.com/8z1i0ym (2026). AUC, area under the curve; stim, stimulation.

Source data

Extended Data Fig. 2. Validation of pharmacological and optogenetic perturbations in glioma cells.

a, Representative flow cytometry plots assessing early and late apoptosis following pharmacological stimulation. SU-DIPG13pons cells were stained with annexin V–568 (x-axis) and DAPI (y-axis) after 1-h treatment with vehicle, 4-aminopyridine (4-AP, 2.5 mM), or neuroligin-3 (NLGN3, 200 nM). b, Top, representative current of ChR2-expressing OPCs. (470 nm, 1.0 mW mm⁻², 1 s). Bottom, representative western blot of shed CSPG4 ectodomains from CSPG4 shedding assay (470 nm, 1.0 mW mm⁻²). c, Representative flow cytometry plots of apoptosis following optogenetic stimulation. SU-DIPG6 cells expressing channelrhodopsin-2 (ChR2) were stimulated with blue light (470 nm, 1 mW mm⁻²) for the indicated durations and subsequently stained with annexin V–568 and DAPI. d, Representative fluorescence images of SU-DIPG13pons cells treated with vehicle (HBSS) or cytochalasin D (2 µM) for 1 h. Cytochalasin D treatment results in actin depolymerization and aggregation, leading to loss of filamentous cytoskeletal structure. e, Summary of calcium activity in glioma cells after pharmacological treatments, across three independent biological replicates, displayed at the level of technical replicates (individual cells). n (number of cells imaged, in the order of vehicle, 4-AP, CytoD and NLGN3) = 301, 1,018, 287, 1,038 for sample 1; 208, 588, 790, 824 for sample 2; and 234, 772, 637, 2,132 for sample 3. Data are shown as individual dot plots. Two-tailed t-test; ****P < 0.0001.

We then investigated whether ADAM10-mediated shedding of CSPG4 is contingent upon the duration of depolarization; to address this question, we used optogenetics—a method developed for precise control of membrane potential using light stimuli³⁴. We achieved this by expressing the light-gated cation channel channelrhodopsin-2 (ChR2) in glioma cells and OPCs and by illuminating blue light (470 nm, 1 mW mm⁻², 0–60 s, every 15 min, 1 h) to induce controlled depolarization (Fig. 3g and Extended Data Fig. 2b, top). We observed a robust increase in CSPG4 shedding with optogenetic depolarization of glioma cells (Fig. 3h; *P < 0.05). Notably, the magnitude of shedding was contingent upon the stimulation duration and consequent magnitude of membrane depolarization, with longer light stimulation duration resulting in larger membrane depolarization amplitude (Fig. 3i,j and Extended Data Fig. 2b, bottom; *P < 0.05), indicating that depolarization directly enhanced ADAM10-mediated CSPG4 shedding, without any changes in cell death (Extended Data Fig. 2c). These results directly support the notion that membrane depolarization enhances ADAM10-mediated CSPG4 shedding.

Intriguingly, CSPG4 ectodomains have also been reported to be shed in response to actin depolymerizing agents^18,28,35. Because disruption of actin microfilaments has been shown to induce membrane depolarization^36,37, we hypothesized that actin depolymerization may promote CSPG4 shedding through depolarization-mediated ADAM10 activation. To test this, we treated glioma cells with cytochalasin D (2 µM, 18 min), an actin depolymerase and performed whole-field calcium imaging (Fig. 3k, left). As expected, cytochalasin D treatment caused marked depolymerization of actin filaments (Extended Data Fig. 2d) and triggered robust calcium transients (20.8% responding cells; Fig.3k, right), comparable to NLGN3 and 4-AP treatments (Fig. 3l and Extended Data Fig. 2e), consistent with prior findings^36,37.

To directly assess the effect of actin depolymerization on CSPG4 shedding, we treated glioma cells with cytochalasin D (2 µM, 15 min) and subsequently analyzed conditioned medium for CSPG4 ectodomains (Fig. 3m, left). We observed a significant increase in CSPG4 shedding, which was completely blocked by ADAM10 inhibition (Fig. 3m, right), reinforcing the conclusion that actin depolymerization promotes ADAM10-mediated shedding through depolarization.

Collectively, our results demonstrate that depolarization enhances ADAM10-mediated shedding of NLGN3 (Fig. 3f,h,j,m,n). These findings also suggest that NLGN3 exposure results in a membrane depolarization that enhances ADAM10-mediated CSPG4 shedding (Fig. 3o). We therefore wondered if the membrane-bound actuators responsible for translating NLGN3 exposure into enhanced ADAM10 activity in glioma cells and OPCs were cation-conducting channels sensitive to disruptions in the actin cytoskeleton.

NLGN3 disinhibits ADAM10 via mechanotransduction

Actin depolymerization, which mimics NLGN3 treatment in inducing membrane depolarization, ADAM10 activation and CSPG4 cleavage (Fig. 3), is also known to increase shear stress and membrane tension^38–40. To explore whether NLGN3 itself alters tension, we assessed changes in membrane tension using Flipper-TR (Spirochrome), a widely used method to measure mechanical changes in live cells (Fig. 4a)^41,42. Remarkably, we observed a significant increase in membrane tension in glioma cells after treatment with NLGN3 (Fig. 4a,b, left). Notably, this increase was dependent on the presence of CSPG4, as we did not observe changes in tension in CSPG4 KO cells (Fig. 4a,b, right). Given these results, we formulated a model in which the force of NLGN3 interacting with CSPG4 causes increased membrane tension that is sensed by mechanotransducers, depolarization-mediated disinhibition of ADAM10 and ultimately shedding of CSPG4 (refs. ^{30,38–40,43,44}).

Fig. 4. NLGN3 induces ADAM10-dependent CSPG4 cleavage through mechanotransduction.

a, Representative images of glioma cells colored based on the lifetime of fluorescence tension sensor Flipper-TR. Scale bar = 20 µm. b, Quantification of the lifetimes measured from the experiment shown in a; n (number of independent biological samples used for imaging) = 6–10, mean ± s.e.m., two-tailed t-test, ****P < 0.0001. c, Scatterplots depicting tumor architecture of scRNA-seq produced from molecularly distinct patient glioma samples. From left to right are histone H3K27M mutated DMG and IDH1 WT GBM. PIEZO1 expression displayed in red across two classes of glioma. d, Left, grayscale channels of PIEZO1 and SU-DIPG13-FL marker GFP shown separately (scale bar = 10 µm). Middle, merged confocal image of immunostaining in SU-DIPG13-FL xenografted mouse brain slice (PIEZO1 = magenta, GFP (marker for glioma cells) = green, scale bar = 10 µm). Right, 3D reconstructions of glioma cell surface showing PIEZO1 localization. Patient-derived DIPG cells (SU-DIPG13-FL; GFP, green) colocalize with PIEZO1 puncta (magenta). Scale bar = 3 µm. e, WB of shed CSPG4 ectodomains from experiment depicted in c. Treatment with Yoda1 (PIEZO1 agonist), increases shedding relative to vehicle controls (P = 0.0464). In contrast, treating cells with MSC blockers gadolinium (Gd³⁺) or the spider toxin GsMTx4 prevents NLGN3-induced shedding of CSPG4 (P = 0.0008 and P < 0.0001, respectively). Right, quantification of WBs; n (independent biological replicates of WB) = 11, mean ± s.e.m., ordinary one-way ANOVA, P values are shown in e. f, Representative WB of shed CSPG4 ectodomains from WT and PIEZO1 KO DIPG cells. g, Quantification of WBs in vehicle and NLGN3-treated conditions (n = 6, mean ± s.e.m., ordinary one-way ANOVA with multiple comparisons, *P < 0.05, **P < 0.01). h, Schematic workflow of the wide-field calcium imaging assay. i,j, Heatmaps (h) and quantification of calcium responses (i) in glioma cells treated with NLGN3, with or without GsMTx4. Calcium activity is measured as AUC of F/F₀; n (number of the cells imaged, from three independent biological replicates) = 1,428 for NLGN3 group and 871 for the NLGN3 + GsMTx4 group. Data are shown as individual dot plots. Two-tailed Mann–Whitney test; **P < 0.01. k,l, Heatmaps (j) and quantification of calcium responses (k) in glioma cells treated with cytochalasin D, with or without GsMTx4; n (number of the cells imaged, from three independent biological replicates) = 435 for CytoD and 476 for CytoD + GsMTx4 group. Data shown are individual dot plots; two-tailed Mann–Whitney test; ****P < 0.0001. m, Schematic of whole-cell patch-clamp recording setup to measure mechanosensitive currents in glioma cells. n, Representative traces of mechanosensitive currents recorded from WT and PIEZO1 KO glioma cells. o, Summary of the current amplitude data represented in b; n (number of independent cells recorded) = 5, mean ± s.e.m. p, Left, schematics of cell-attached patch-clamp recording of mechanosensitive currents from glioma cells. Right, the representative waveforms of mechanosensitive currents measured with 10 mmHg and 30 mmHg pressure inputs. q, Representative traces of mechanosensitive currents (pressure input = 40 mmHg, 1 s) with vehicle (black), PIEZO1 agonist Yoda1 (cyan), NLGN3 (red) and mechanosensitive channel blocker GsMTx4 (brown). r, Summary of mechanosensitive currents shown in b; n (number of independent biological samples) = 6–8, mean ± s.e.m., Kruskal–Wallis test, **P < 0.01, ***P < 0.001. s, Model depicting CSPG4 shedding through ADAM10 activation that is downstream of NLGN3 interaction and ensuing membrane depolarization through PIEZO1 channel activation. Red circular sector, ADAM10; green rectangle, shed NLGN3; blue,CSPG4 ectodomain; light blue, PIEZO1. Illustrations in h and s created in BioRender; Kim, Y. https://biorender.com/8z1i0ym (2026). NS, not significant; AC like, astrocyte-like cells; OC like, oligodendrocyte-like cells; MES like, mesenchymal cell-like cells; NPC like, neural progenitor-like cells; 3D, three-dimensional.

Source data

To validate our model, we endeavored to identify actuators for sensing and responding to the NLGN3–CSPG4 interaction and the increase in membrane tension. Increased shear stress and membrane tension are known to activate mechanotransducers such as PIEZO and transient receptor potential channels that are associated with the cytoskeleton^{38–40,43,44}. Thus, we interrogated multiple single-cell RNA-seq (scRNA-seq) datasets from pediatric DMG and adult IDH1 WT GBMs to curate a list of mechanotransducers expressed in OPC-like cells also expressing CSPG4 (Fig. 4c and Extended Data Fig. 3a), finding several including PIEZO1, PIEZO2 and several transient receptor potential channels (Extended Data Fig. 3b)^45,46. The expression level of the mechanotransducer PIEZO1 was the highest (Extended Data Fig. 3b), and immunostaining (Fig. 4d) and western blot (WB; Extended Data Fig. 3c) analyses indicated high levels of PIEZO1 protein in glioma cells. In addition, several studies have linked PIEZO1 to cancer progression and poor patient prognosis, including in brain tumors, where its activity promotes proliferation and invasion^47–49.

Extended Data Fig. 3. sNLGN3 induces ADAM10-dependent CSPG4 cleavage via activation of PIEZO1 channels.

a, Scatterplots depicting tumor architecture of scRNA-seq produced from molecularly distinct patient glioma samples. From left to right are Histone H3K27M mutated diffuse midline glioma (DMG) and IDH1 wild-type glioblastoma (GBM). CSPG4 expression displayed in red across two classes of glioma. PIEZO1 and CSPG4 are both detected within the OPC-like fraction of these distinct glioma types. b, Dot plots of scaled gene expression levels of mechanotransducers PIEZO1, PIEZO2, TRPC1 and TRPC6 in multiple patient biopsies of H3K27M mutant diffuse midline gliomas profiled by smart-seq2. All six biopsies are from ref. ⁶⁸. Counts were downloaded from GEO (GSE102130) and plotted following log normalization and scaling functions in Seurat (https://satijalab.org/seurat/). c, Western blot of PIEZO1 protein expression from lysates of SU-DIPG6 PIEZO1 WT and KO cells. d, Workflow depicting treatment of pediatric cortical GBM cells with recombinant NLGN3 ectodomains in the presence or absence of inhibitors of mechanotransducers (MSC) followed by culture medium concentration and western blot to detect shedding of CSPG4. Illustrations in d created in BioRender; Kim, Y. https://BioRender.com/8z1i0ym (2026). e, Western blot of shed CSPG4 ectodomains from pediatric cortical GBM experiments depicted in d. f, Quantification of western blots (n (independent biological replicates of western blot) = 3, mean ± s.e.m., ordinary one-way ANOVA with multiple comparisons, **P < 0.01, ****P < 0.0001). g–i, Juxtaposition of N-terminal and C-terminal CSPG4 western blots from representative CSPG4 shedding assays shown in Fig. 2d (g), Fig. 3f (h), and Fig. 4f (i). Please note that C-terminal CSPG4 is not detected under all experimental conditions.

Source data

To assess whether mechanotransduction is necessary for NLGN3-induced ADAM10 activity, we performed CSPG4 shedding assays using the mechanosensitive channel blockers gadolinium (Gd³⁺) and GsMTx4 (Extended Data Fig. 3d)⁵⁰. Remarkably, blockade of channel conduction proved to be as effective as the ADAM10 inhibitor in preventing NLGN3-induced shedding of CSPG4 (Fig. 4e and Extended Data Fig. 3e,f). In contrast, treatment of cells with the PIEZO1 agonist Yoda1 resulted in enhanced shedding (Fig. 4e and Extended Data Fig. 3e,f). Consistent with other assays (Fig. 2b and Extended Data Fig. 3g,h), only the N-terminal ectodomain was released into the conditioned medium (Extended Data Fig. 3i).

Using CRISPR–Cas9, we generated PIEZO1 KO glioma cells and observed a marked reduction in NLGN3-induced CSPG4 shedding (Fig. 4f,g), confirming the necessity of PIEZO1. Wide-field calcium imaging confirmed that calcium transients induced by both NLGN3 and cytochalasin D were significantly diminished by GsMTx4, indicating that these responses are mediated by mechanotransducer activity (Fig. 4h–l and Extended Data Fig. 4a). Furthermore, to directly assess the biophysical properties of mechanosensitive currents, we performed whole-cell patch-clamp recordings in WT and PIEZO1 KO glioma cells using hypertonic solution-induced stress (Fig. 4m). Previous studies^47,51 demonstrated that osmotic stress can preserve cellular morphology and membrane functionality. We detected large inward currents after stimulation with hypertonic solution. Notably, these responses were significantly decreased in PIEZO1 KO, confirming that the current mainly arises from PIEZO1 (Fig. 4n,o). Notably, NLGN3 treatment slowed channel inactivation (Extended Data Fig. 4b).

Extended Data Fig. 4. Biochemical and electrophysiological characterization of PIEZO1 signaling in glioma cells.

a, Summary of calcium activity shown in Fig. 4j,l, across three independent biological replicates, displayed at the level of technical replicates (individual cells). n (number of cells imaged, in the order of cytoD, cytoD + G4, NLGN3, NLGN3 + G4) = 133, 127, 293, 360 for sample 1; 243, 249, 320, 289 for sample 2; and 59, 100, 815, 272 for sample 3. Data are shown as individual dot plots. Two-tailed Mann–Whitney test; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. b, Top, representative currents from whole-cell patch-clamp experiment in glioma cells with the vehicle and NLGN3 treated conditions. Bottom, summary of the off kinetics of the cells as shown at the top. (n (number of independent cells recorded) = 3 for the vehicle and 4 for NLGN3, mean ± s.e.m., two-tailed Student’s t-test, **P < 0.01). c, Representative traces of mechanosensitive currents recorded with cell-attached recording from WT pcGBM2 glioma cells versus CSPG4 KO cells, with and without NLGN3 treatment. d, Summary of cell-attached experiment performed in a (n (number of independent cells recorded) = 5, mean ± s.e.m.).

We then used cell-attached recordings under stepwise negative pressure to investigate mechanosensitive currents^50–52. Initially, we observed that a sequence of stimulations resulted in significant inward currents (Fig. 4p). These currents were markedly attenuated by the presence of GsMTx4 (7.9 ± 1.9 versus 2.8 ± 0.23 pA, at −40 mmHg) and augmented by the PIEZO1 agonist Yoda1 (ref. ⁵¹; 7.9 ± 1.9 versus 23.6 ± 6.0 pA, at −40 mmHg), confirming that these currents are elicited by the activation of the mechanosensitive channels, primarily PIEZO1 (Fig. 4q,r). Furthermore, we observed a notable increase in mechanosensitive currents with NLGN3 (Fig. 4r; 7.9 ± 1.9 versus 17.1 ± 5.0 pA, at −40 mmHg), similar to Yoda1. This finding is consistent with earlier observations (Fig. 4a,b) and indicates that mechanotransduction is directly modulated by NLGN3. To further validate, we conducted cell-attached recordings in CSPG4 KO glioma cells. Surprisingly, we observed minimal currents after a series of negative pressures with or without NLGN3 (Extended Data Fig. 4c,d).

Together, these findings support a model in which NLGN3–CSPG4 engagement increases membrane tension, activating mechanotransducers that drive depolarization, ADAM10 activation and ectodomain shedding (Fig. 4s). While PIEZO1 likely serves as a major transducer in this pathway, the efficacy of broad-spectrum inhibitors and the partial nature of the KO phenotype suggest that other mechanosensitive channels may also contribute⁵².

Functional consequences in NLGN3–mechanotransduction-mediated pathways

To explore downstream signaling cascades, we examined the PI3K–mTOR pathway^3,7. In glioma cells treated with NLGN3 or Yoda1, we observed robust Ser473 phosphorylation of AKT (Fig. 5a)^7,48,49. In contrast, NLGN3-induced pAKT was markedly reduced, although not completely abrogated, as the Yoda1-induced phosphorylation, in GsMTx4 treated cells (Fig. 5a) and PIEZO1 KO cells (Fig. 5b,c). These results suggest that NLGN3-induced activation of the PI3K–mTOR pathway is mediated primarily through PIEZO1, while the observed residual signals suggest that other mechanotransducers may also act in concert with PIEZO1, reiterating a broader role for mechanotransduction as a central pathway for translating NLGN3 signaling into proliferation and growth in glioma^3,7,8,15.

Fig. 5. Blockade of CSPG4 shedding by mechanotransduction prevents NLGN3-induced changes to glioma cells.

a,b, Representative flow cytometry plots from phospho-flow cytometry assays showing pAKT (Alexa Fluor 647) staining in WT (a) and PIEZO1 KO (b) cells treated for 1 h with vehicle, Yoda1, or NLGN3. In the WT condition, GsMTx4 was included to inhibit mechanotransducer activity. The experiment was performed in three independent biological replicates. c, Histograms corresponding to the flow cytometry data in a, illustrating the distribution of pAKT signal intensities. d, Schematics of xenografting of SU-DIPG6 WT/PIEZO1 KO into immunocompromised mice followed by assessment of proliferation with Ki67⁺/HNA staining. e, Representative confocal images of WT and PIEZO1 KO brain slices explained in d (white, Ki67; red, HNA; white arrows denote the overlapping cells with both Ki67 and HNA stained). f, Summary of tumor proliferation (g, %Ki67⁺) in WT and PIEZO1 KO samples; n (number of independent biological samples) = 5, mean ± s.e.m., two-tailed t-test, *P < 0.05. g, Volcano plot of differentially expressed genes comparing WT and PIEZO1 KO MGG8 cells. Genes enriched in PIEZO1 KO relative to WT cells are plotted with positive log₂ fold change. Conversely, genes that are depleted in PIEZO1 KO cells relative to WT are plotted as negative log₂ fold change. Genes are colored according to log₂ fold change and adjusted P value thresholds (gray, nonsignificant by either metric; green, log₂ fold change of 1 or greater; blue, adjusted P value of 10 × 10⁻⁶ or greater; pink, gene exceeds thresholds for fold change and P value). h, GO analysis of genes that were significantly depleted in PIEZO1 KO cells relative to WT. Enriched programs all center around sterol biosynthesis and metabolism. Of note, FABP7 is among the most strongly downregulated genes in PIEZO1 KO cells and is reported to function in the maintenance of both glioma stem cells and healthy OPCs⁶⁵. i, Cartoon model depicting mechanism of NLGN3-induced CSPG4 shedding and functional consequences. Red circular sector, ADAM10; green rectangle, shed NLGN3; blue, CSPG4 ectodomain; light blue, PIEZO1. Illustrations in d and i created in BioRender; Kim, Y. https://biorender.com/8z1i0ym (2026). GO, Gene Ontology; HNA, human nuclear antigen.

Source data

Having established that PIEZO1 is a primary actuator responsible for NLGN3-mediated shedding of CSPG4, we hypothesized that blocking the PIEZO1 response in glioma would mimic loss of microenvironmental NLGN3 and affect glioma growth in vivo^7,8. To test this hypothesis, we orthotopically xenografted the PIEZO1 KO glioma (DIPG) cells into pons region of mouse brains (Fig. 5d). Notably, after 4 weeks, mice bearing PIEZO1 KO glioma xenografts showed a significant reduction in glioma cell proliferation (Fig. 5e,f) compared to those bearing PIEZO1 WT glioma cells from the same patient. We further confirmed that administration of an ADAM10 inhibitor, which would block both NLGN3 secretion and NLGN3-mediated CSPG4 shedding, also resulted in a significant decrease in tumor burden and proliferation rate in patient-derived glioma xenografts, as expected⁷ (Extended Data Fig. 5).

Extended Data Fig. 5. Treatment with ADAM10 inhibitor decreases tumor burden in xenografted mouse model.

a, Paradigm for in vivo treatment with ADAM10 inhibitor aderbasib in immunodeficient mice xenografted with a patient-derived glioblastoma cell line (‘SF-232-HFC’) and the counting model for tumor proliferation and burden. Illustrations in a created in BioRender; Kim, Y. https://BioRender.com/8z1i0ym (2026). b, Proliferation index in vehicle- and aderbasib-treated mice (n = 5 per group) in the xenografted area of the secondary motor cortex (‘M2’), quantified by confocal microscopy of Ki67⁺/HNA⁺ cells. Data are presented as mean ± range. Two-tailed Student’s t-test, ***P < 0.001. c, Proliferation index in vehicle- and aderbasib-treated mice (n = 5 per group) of migrated glioblastoma cells into the ipsilateral corpus callosum (‘CC’), quantified by confocal microscopy of Ki67⁺/HNA⁺ cells. Data are presented as mean ± range. Two-tailed Student’s t-test, ns = not significant. d, Tumor burden in vehicle- and aderbasib-treated mice (n = 5 per group) in the xenografted area of the secondary motor cortex (‘M2’), quantified by confocal microscopy as the number of HNA⁺ cells. Data are presented as mean ± range. Two-tailed Student’s t-test, **P < 0.01. e, Tumor burden in vehicle- and aderbasib-treated mice (n = 5 per group) in the xenografted area of the ipsilateral corpus callosum (‘CC’), quantified by confocal microscopy as the number of HNA⁺ cells. Data are presented as mean ± range. Two-tailed Student’s t-test, ***P < 0.001. f, Representative images of xenografted glioblastoma cells in the secondary motor cortex (‘M2’) in vehicle- and aderbasib-treated mice. Gray denotes myelin sheaths labeled by MBP; green denotes HNA⁺ glioma cells; red denotes EdU. Scale bar = 100 μm. g, Representative images of migrated glioblastoma cells into the ipsilateral corpus callosum (‘CC’) in vehicle- and aderbasib-treated mice. Gray denotes myelin sheaths labeled by MBP; green denotes HNA⁺ glioma cells; red denotes EdU. Scale bar = 100 μm.

To understand the consequences of PIEZO1 depletion, we performed bulk RNA-seq of a patient-derived GBM model with or without PIEZO1. Differential gene expression analysis revealed that, upon PIEZO1 KO (Fig. 5g), GBM cells dramatically increase expression of several stress/apoptosis-related genes including CASP8, UBB and GSN. Intriguingly, downregulated genes in PIEZO1-depleted cells were related to sterol metabolism (Fig. 5h), known to be highly active in GBM cancer stem cells (CSCs)⁵³. Furthermore, several downregulated genes are markers of GBM CSCs including FABP7, TF and SLC2A3 (aka GLUT3), suggesting that PIEZO1 may influence the maintenance of the CSC state (Fig. 5g,i)^54,55.

Blockade of CSPG4 shedding precludes NLGN3-induced changes in glioma

We then investigated whether the NLGN3–CSPG4 interaction is necessary for the NLGN3-induced glioma proliferation and growth^3,7. We first tested the proliferation of glioma cells with or without NLGN3 (Extended Data Fig. 6a); consistent with prior work^3,7,8, we found that NLGN3 increased glioma proliferation (Extended Data Fig. 6b). Notably, this NLGN3-induced proliferation was attenuated by ADAM10 inhibitor pretreatment, even with NLGN3 exposure (Extended Data Fig. 6b). Next, we investigated cell growth in CSPG4 KO glioma cells using the CellTiterGlo assay (Extended Data Fig. 6c), finding a significant growth reduction (Extended Data Fig. 6d).

Extended Data Fig. 6. Genetic depletion of CSPG4 disrupts NLGN3-induced growth of glioma cells in vitro.

a, Workflow depicting NLGN3-induced EdU incorporation assay in the presence or absence of CSPG4 shedding blocker, ADAM10i. b, Quantification of NLGN3-induced EdU incorporation from mouse NF1-associated optic glioma cells. NLGN3 treatment increases EdU incorporation relative to vehicle controls (n (number of independent cells, in the order of Veh, NLGN3, ADAM10i and combo) = 11, 11, 5, 6, mean ± s.e.m., *P = 0.03, one-way ANOVA with Brown–Forsythe and Welch’s test). c, Schematics of cell proliferation assay using CellTiterGlo assay. d, Quantification of CellTiterGlo proliferation assay of WT and CSPG4 KO SU-DIPG13pons cells (n (number of independent biological trials) = 3, mean ± s.e.m., two-tailed Student’s t-test, *P < 0.05, ***P < 0.001, ****P < 0.0001). Please note that from day 3 the proliferation rate (assessed by the raw luminescence) is significantly lower in the CSPG4 KO group. Illustrations in a and c created in BioRender; Kim, Y. https://BioRender.com/8z1i0ym (2026).

NLGN3–CSPG4 interaction impedes differentiation of OPCs

Because gliomas hijack developmental OPC programs, we then asked the role of NLGN3–CSPG4–PIEZO1 signaling in normal OPCs. Having confirmed that NLGN3 induces CSPG4 shedding from OPCs (Fig. 2), we investigated the activity of PIEZO1 in OPCs in response to NLGN3. Previous reports have demonstrated that PIEZO1 is a key mediator of OPC differentiation in the brain⁵⁶. To confirm functional PIEZO1 expression, we conducted cell-attached recordings in OPCs (Extended Data Fig. 7a). We observed minuscule inward mechanosensitive currents from OPCs, which were notably augmented by NLGN3 (Extended Data Fig. 7b). Moreover, these currents were pressure dependent (Extended Data Fig. 7c), similar to PIEZO1 currents in glioma cells (Fig. 4).

Extended Data Fig. 7. In vivo characterization of OPC proliferation state in NLGN3 KO mice.

a, Schematics for in vitro OPC electrophysiology. b, Representative traces of mechanosensitive currents measured from OPCs. c, Summary of mechanosensitive currents shown in b. (n (number of OPCs per group) = 5 for vehicle and 6 for NLGN3, mean ± s.e.m.). d, qPCR results for the rat OPCs post 48 h after PDGFα withdrawal (n (number of independent trials) = 2) with and without NLGN3 treatment. Data are mean with range. e, Schematics of NLGN3 deletion from OPCs in vivo. Mice expressing cell type-specific (PDGFα), inducible Cre drivers were bred to Nlgn3fl/fl mice. Tamoxifen was administered for 5 days beginning at P21. Illustrations in e created in BioRender; Kim, Y. https://BioRender.com/8z1i0ym (2026). f, Representative confocal images of Pdgfra positive (white) and Olig2 positive cells in WT and NLGN3 KO animals. Scale bar = 50 µm. g, Quantification of total OPC counts in WT and NLGN3 groups, as shown in e. (n (number of independent biological replicates of animals) = 4, mean ± s.e.m., two-tailed Student’s t-test, *P < 0.05). h, PIEZO1 gene expression across different OPC age groups, retrieved from scRNA-seq data reported in ref. ⁶¹. (n (independent biological trials) = 4). For box plots, the centerline indicates the median, box bounds indicate the 25th and 75th percentiles, and whiskers indicate the minimum and maximum values. Individual data points are shown as dots.

We then investigated whether NLGN3 exposure influences OPC proliferation, both in place of, and in addition to, the canonical OPC mitogen PDGFAA (Fig. 6a)^3,8. In contrast to robust proliferation induced by PDGFAA, OPCs did not proliferate in response to NLGN3 (Fig. 6b,c). PDGFAA withdrawal triggers oligodendrocyte differentiation marked by dramatic and stereotypical morphological changes. NLGN3 exposure impeded these morphological transitions (Fig. 6d,e); OPCs in medium containing PDGFAA typically extend single processes from opposite sides of their somas, while differentiating oligodendroglial cells become more complex and arborized (Fig. 6e,f)⁵⁷. We reasoned that if maintenance of the OPC progenitor state depended on NLGN3-mediated cleavage of CSPG4, ADAM10 inhibition would render OPCs insensitive to NLGN3 and PDGFAA withdrawal would trigger differentiation. Indeed, ADAM10 inhibition partially restored the morphological progression of OPC differentiation (Fig. 6g,h)⁵⁸.

Fig. 6. sNLGN3-induced CSPG4 shedding regulates OPC state.

a, Workflow depicting EdU incorporation and detection in rat OPCs after treatment with NLGN3. b, Representative images of EdU incorporation in primary rat OPCs after 4-h treatments. c, Quantification of cell proliferation depicted in a; n (independent biological replicates) = 3; mean ± s.e.m., one-way ANOVA with Tukey’s multiple comparisons correction; (P < 0.0001). d, Schematic depicting differentiation and morphological changes of OPCs upon mitogen withdrawal in vitro. e, Representative images of OPCs after 48-h mitogen withdrawal in the presence or absence of supplemental sNLGN3. f, Quantification of cells at each morphological stage after 48-h treatment; n (number of independent trials) = 4, two-way ANOVA with multiple comparisons and Bonferroni correction, NS. Please note that PDGFAA-treated and NLGN3-treated cells are not significantly different regarding progression to stage 3 morphology as assessed by two-way ANOVA with multiple comparisons and Bonferroni correction. g, Representative images of OPCs after 48-h mitogen withdrawal in the presence or absence of supplemental sNLGN3 and ADAM10 inhibition. h, Quantification of OPC (CSPG4⁺) to OL (MBP⁺) ratio in cells depicted in g; n (independent biological replicates) = 3; mean ± s.e.m., one-way ANOVA with Tukey’s multiple comparisons correction. P values are shown in h. i, PCA plot of bulk RNA-seq produced from rat OPCs that were cultured for 48 h under PDGFAA withdrawal and supplementation with NLGN3 and/or inhibitors of CSPG4 shedding. Vehicle and NLGN3-treated cells are distinct in PC space from the inhibitor-treated samples. j, Volcano plot of differentially expressed genes comparing NLGN3-treated rat OPCs against PDGFAA withdrawal alone. Genes enriched in NLGN3-treated relative to WT cells are plotted with positive log₂ fold change. Conversely, genes that are depleted in NLGN3-treated cells relative to WT are plotted as negative log₂ fold change. Genes are colored according to log₂ fold change and adjusted P value thresholds. k, Heatmap of NLGN3-induced genes from ref. ⁷. Primary rat OPCs were cultured for 48 h without PDGFAA to initiate differentiation and NLGN3 was supplemented in half the cultures. Duplicate vehicle and NLGN3-treated samples were profiled by bulk RNA-seq. TPM values for each condition were averaged and then the log₂ fold change was calculated between NLGN3-treated and vehicle control samples. These values were plotted such that zero appears as white on a symmetrical axis that ranges from −4 in blue and 4 in red. l, Model depicting CSPG4 shedding through PIEZO1 opening and ADAM10 activation. These events impede OPC differentiation in vitro upon growth factor withdrawal. Red circular sector, ADAM10; green rectangle, shed NLGN3; blue, CSPG4 ectodomain; light blue, PIEZO1. Illustrations in a and l created in BioRender; Kim, Y. https://biorender.com/8z1i0ym (2026). TPM, transcripts per million.

We then assessed transcriptomic changes through bulk RNA-seq of OPCs undergoing differentiation in the presence or absence of NLGN3 and modulators of CSPG4 shedding. We determined that OPCs treated with NLGN3 clustered apart from cells exposed to vehicle or pretreated with inhibitors of CSPG4 shedding before NLGN3 exposure (Fig. 6i). Furthermore, our differential gene expression analysis comparing NLGN3-treated OPCs and vehicle-treated OPCs clearly demonstrates an enrichment of genes associated with stemness and synaptic communication (Pdgfra, Cacng4 and Kcnip3) and a suppression of genes related to differentiation (Mbp, Plp, Enpp6; Fig. 6j). Of note, we found that NLGN3 treatment of OPCs resulted in upregulation of several genes that also increased in expression when human glioma cells were treated with NLGN3 in a previous study⁷, demonstrating the functional similarities of Nlgn3 in OPCs and glioma cells (Fig. 6k). Finally, using RT-qPCR analyses in OPCs, we verified that NLGN3 treatment prevented the expression of myelin-specific transcripts Plp and Mbp and maintained the expression of the differentiation inhibitor Sox11 relative to OPCs exposed only to vehicle control after growth factor withdrawal (Extended Data Fig. 7d)⁵⁹.

Previous in vivo genetic mouse modeling experiments demonstrated that NLGN3 is chiefly shed from OPCs, indicating that OPC-derived NLGN3 shedding may have a role in oligodendroglial biology. PIEZO1 is expressed in OPCs and expression remains stable across OPCs of different ages (Extended Data Fig. 7h). The in vitro studies described above implicate NLGN3 signaling in maintenance of a self-renewing, undifferentiated state in OPCs, raising the possibility that NLGN3 shed from OPCs may contribute to OPC stemness and population maintenance. We therefore tested the effects of Nlgn3 loss on OPC population homeostasis using a conditional KO mouse model, PDGFRa::Cre^ERT;Nlgn3^fl/fl, where Nlgn3 is specifically knocked out from OPCs upon tamoxifen administration (Extended Data Fig. 7e). Remarkably, acute deletion of NLGN3 led to an approximately 30% reduction in the number of OPCs in the corpus callosum (Extended Data Fig. 7f,g and Fig. 6l). Taken together, these findings indicate that NLGN3–CSPG4–PIEZO1 signaling contributes to OPC population maintenance by promoting maintenance of OPCs in a progenitor state.

Discussion

In this study, we uncover a mechanotransduction-based mechanism linking activity-regulated NLGN3 signaling to both glioma proliferation and OPC state regulation. We identify CSPG4 as a prominent NLGN3-interacting partner on glioma cells and healthy OPCs and show that NLGN3–CSPG4 interaction increases membrane tension, activates mechanosensitive ion channels, chiefly PIEZO1, and promotes ADAM10-dependent CSPG4 ectodomain shedding. Through this pathway, NLGN3 supports glioma cell proliferation while maintaining healthy OPCs in an undifferentiated progenitor state. These findings establish the NLGN3–CSPG4–PIEZO1 pathway as a shared mechanism between malignant and normal glial biology and suggest that gliomas exploit activity-regulated mechanisms of OPC state control to promote tumor growth.

Using unbiased biochemical screening, we identified CSPG4 as a prominent NLGN3-interacting partner. CSPG4 was prioritized based on its reproducible enrichment in affinity chromatography–mass spectrometry experiments, strong expression in both OPCs and glioma cells, and established roles in progenitor maintenance and oncogenic signaling^23,24,59,60. Building on these findings, we uncovered a mechanism whereby NLGN3 interacts with CSPG4, thus altering membrane tension and activating the mechanotransducers, primarily PIEZO1, leading to membrane depolarization and depolarization-induced activity of the ADAM10, a sheddase that processes multiple substrates including CSPG4 ectodomain (but not C-terminal CSPG4) and NLGN3 (refs. ^7,29,39).

Beyond its implications for glioma biology, these findings shed new light on roles for NLGN3 in the regulation of OPC states, elucidating a newly appreciated role for activity-regulated maintenance of homeostatic OPC population dynamics through the NLGN3–CSPG4–mechanotransduction pathway. NLGN3–CSPG4–mechanotransduction signaling thus joins a growing list of mechanistic parallels between neuron–glial and neuron–glioma interactions, supporting a stem-like, undifferentiated state in both healthy and malignant glial cells. In healthy OPCs, this stem-like state appears to prevent differentiation rather than directly driving proliferation, suggesting that NLGN3 may contribute to OPC maintenance through inhibition of differentiation rather than expansion of the population through proliferation. This interpretation is supported by our observation that OPC numbers decline after NLGN3 deletion in vivo, despite no significant effect on OPC proliferation in vitro, indicating that NLGN3 signaling may preserve OPC homeostasis by sustaining a pool of undifferentiated cells rather than inducing active cell division.

Our findings also suggest that activation of PIEZO1 by NLGN3–CSPG4 signaling selectively prevents OPC differentiation without significantly affecting proliferation, whereas it was reported that PIEZO1 activation in aged OPCs inhibits both proliferation and differentiation⁵⁶. These differing outcomes may reflect the influence of developmental context, as neonatal and aged OPCs likely interpret mechanical cues differently—consistent with age-related transcriptional shifts reported in ref. ⁶¹. In addition, variations in the mode of PIEZO1 activation may contribute to the observed differences. Mechanical stress is identified in ref. ⁵⁶ from increased extracellular matrix stiffness as the activating stimulus, while in our study, PIEZO1 is engaged through protein-mediated interactions between NLGN3 and CSPG4(refs.^3,7,56). Mechanotransducers like PIEZO1 can respond to a variety of distinct modes of activation^38,43,51, which may influence channel gating, localization and downstream signaling in context-specific ways. Moreover, while our data support a central role for mechanotransduction, primarily through PIEZO1, in mediating the effects of NLGN3, residual pAKT signaling after PIEZO1 KO and mechanosensitive channel inhibition suggests that additional, nonmechanosensitive pathways may also contribute.

While our findings provide a mechanistic framework linking NLGN3, CSPG4 and mechanotransduction signaling, several limitations of the current study should be noted. Although our in vivo data indicate that PIEZO1 KO attenuates the NLGN3/CSPG4-dependent glioma phenotype, the precise mechanisms by which PIEZO1 integrates NLGN3-mediated and CSPG4-mediated signaling in vivo remain to be elucidated. Furthermore, while our data indicate that NLGN3 binding to CSPG4 enhances membrane tension and mechanosensitive channel activity, the absence of mechanotransduction in CSPG4-deficient cells suggests that CSPG4 is also required for baseline mechanosensitivity and membrane organization. This implies that CSPG4 not only serves as an interacting partner for NLGN3 but also may contribute to the structural context necessary for mechanotransduction through its linkage to the actin–membrane network, as the complete loss of response in CSPG4 KO cells may reflect a disruption of mechanical coupling rather than a selective loss of NLGN3 responsiveness. Consistent with our findings, previous studies have shown that CSPG4 promotes glioma proliferation, invasion and tumor growth in vivo, and that its loss or inhibition suppresses malignancy^62,63, further supporting our data that the NLGN3–CSPG4 signaling pathway may contribute to glioma progression through both mechanotransductive and proliferative mechanisms. Together, these findings highlight the complexity of NLGN3–CSPG4–mechanotransduction across developmental stages and activation contexts, and underscore the importance of considering both biochemical and biomechanical inputs when studying mechanotransduction in glioma and OPC biology.

While our findings define a NLGN3–CSPG4–mechanotransduction pathway critical for glioma proliferation and OPC state maintenance, they also raise important considerations for therapeutic targeting. The substantial reduction in OPC numbers after NLGN3 deletion underscores that this signaling pathway may have a role in sustaining the OPC pool in the healthy brain. Given the molecular parallels between glioma cells and OPCs, therapeutic interventions aimed at disrupting this pathway could pose risks to healthy glial function, including myelination^{12,14,58,59,64}. Although our study did not assess the long-term functional consequences of the observed OPC loss, these findings emphasize the need for further study and targeted delivery strategies or transient inhibition approaches to minimize potential toxicities.

In summary, the findings presented here underscore a critical role for an NLGN3–CSPG4–mechanotranduction pathway in the interplay between neuronal activity and glioma progression, advancing understanding of a key mechanistic pathway by which gliomas exploit neuronal signals for growth. The identification of mechanotransduction as an important mediator in this process suggests new therapeutic opportunities, highlighting the potential for targeting mechanosensitive pathways in glioma and possibly in other cancers.

Methods

Mice and housing conditions

All animal experiments were conducted in accordance with protocols approved by the Stanford University Institutional Animal Care and Use Committee and performed in accordance with institutional guidelines. Animals were housed according to the standard guidelines with unlimited access to water and food, under a 12-h light/12-h dark cycle. For brain tumor xenograft experiments, the Institutional Animal Care and Use Committee has a limit on indications of morbidity (as opposed to tumor volume). Under no circumstances did any of the experiments exceed the limits indicated and mice were immediately killed if they showed signs of neurological morbidity or if they lost 15% or more of their initial body weight.

Recombinant NLGN3 purification and NLGN3 capture assay

DNA encoding the signal peptide and ectodomain of human NLGN3 (Sino Biological, HG11160-UT) was cloned into pD649 with C-terminal AviTag and 6xHis tags. Plasmids were transfected into Expi293 cells (Thermo Fisher Scientific, A14635), induced 18–24 h post-transfection and media were collected after 72–96 h, filtered through a 0.22-micron Steriflip, and purified using HisPur Cobalt Resin (Thermo Fisher Scientific, 89964). Protein was washed with 100 mM NaCl, 20 mM HEPES, pH 7.5, 20 mM imidazole, eluted with 100 mM NaCl, 20 mM HEPES, pH 7.5, 250 mM imidazole, further purified by gel filtration in 100 mM NaCl, 20 mM HEPES, pH 7.5, and concentrated to 3–6 mg ml⁻¹. For NLGN3 affinity capture, 30 million SU-DIPG13pons or MGG8 neurosphere cells were pelleted without dissociation (200g, 3 min, at room temperature), washed in calcium-free and magnesium-free HBSS, snap frozen in dry ice/ethanol and stored at −80 °C. Pellets were thawed on ice, resuspended in 5 ml lysis buffer with protease inhibitors (10 mM HEPES, pH 7.5, 10 mM MgCl₂, 20 mM KCl, 1× Halt protease inhibitor cocktail; Thermo Fisher Scientific, 78429), and centrifuged at 20,000g for 20 min at 4 °C. Pellets were resuspended in 1 ml lysis buffer, homogenized with 20 strokes using a 2-ml glass Dounce, diluted to 4 ml and spun at 20,000g for 20 min at 4 °C; this was repeated thrice. Pellets were then washed in 2.5-ml salt wash buffer (lysis buffer + 150 mM NaCl) until no protein was detected in the cleared supernatant by Bradford assay. Purified membrane fractions were solubilized in 1% L-MNG, 150 mM NaCl, 20 mM HEPES, 10% glycerol and 1× Halt protease inhibitor cocktail. NLGN3 ectodomains were biotinylated using the BirA biotin–protein ligase standard reaction kit (Avidity, EC 6.3.4.15), purified and concentrated using a 30-kDa Amicon filter (MilliporeSigma, UFC503096), and immobilized on streptavidin resin (Pierce ProFound Biotinylated-Protein Interaction Pull-Down Kit; Thermo Fisher Scientific, 21115). Solubilized membrane proteins were incubated with NLGN3 resin overnight at 4 °C with rotation, washed thrice with supplied acetate buffer plus 150 mM NaCl and eluted in LDS sample buffer + β-mercaptoethanol (BME) at 42 °F for 15 min. Proteins were resolved on a 3–8% Tris-acetate gel at 120 V for 15 min, and upper gel slices were excised for mass spectrometry. Protein abundance was normalized to actin, and the enrichment index was calculated as the ratio of each protein detected with NLGN3 capture to the same protein detected without NLGN3 capture.

WBs from human CSF

Human CSF samples were obtained under IRB14511 from the Bass Center Leukemia Tissue Bank for nonglioma acute lymphocytic leukemia patient CSF samples, and from the GD2-CAR-T cell clinical trial for glioma patient CSF samples provided by Sneha Ramakrishna. Samples were concentrated 20-fold using Amicon 50 K filters.

Samples were prepared with 4× LDS sample buffer (Thermo Fisher Scientific, NP0007) and BME (Sigma-Aldrich, M6250), then heated at 70 °C for 15 min. Proteins were resolved on 3–8% Tris-acetate gels at 100 V for 10 min followed by 120 V for 45 min, and transferred to PVDF membranes using the iBlot2 system (Thermo Fisher Scientific) with a high-molecular weight transfer program—20 V for 4 min, 23 V for 4 min and 25 V for 2 min. Membranes were blocked in 3% BSA in 0.1% TBS-T for 1 h at room temperature, then incubated overnight at 4 °C with primary antibodies diluted 1:1,000 in blocking buffer—anti-NLGN3 ectodomain (Abcam, 90080) or anti-CSPG4 N terminal (Cell Signaling Technology, 52635). After washing 3× 5 min in TBS-T, membranes were incubated for 1 h at room temperature with HRP-conjugated secondary antibody Cell Signaling Technology, 7074; 1:10,000), washed again 3× 5 min, developed with SuperSignal West Femto chemiluminescent substrate (Thermo Fisher Scientific, 34580) and imaged using the Bio-Rad Gel Doc system. Band and smear intensities were quantified in Fiji (ImageJ) using consistent exposure times across conditions. Only nonsaturated images were used for quantification.

NLGN3 coupling assay using FSEC

For initial coupling analysis, purified NLGN3 (320 µg ml⁻¹) was mixed with 6 mg ml⁻¹ of CS (Sigma-Aldrich) or 400 µg ml⁻¹ of heparin (heparan sulfate, US Biological). Coupling was assessed by the left shift in FSEC⁶⁶, using tryptophan fluorescence for detecting NLGN3 signal.

For quantification, we mixed the same amount of NLGN3 with 0, 0.75, 1.5, 3 and 6 mg ml⁻¹ of CS and quantified the left shift of the NLGN3 peak in FSEC. Experiments were conducted thrice, with independently purified batches of NLGN3.

Glioma and primary OPC culture

For glioma cell culture, DIPG or GBM neurospheres (SU-DIPG6, SU-DIPG13pons, SU-DIPG21, MGG8) were maintained in neurobasal/F12 medium (1:1) supplemented with GlutaMAX, penicillin–streptomycin, MEM nonessential amino acids, sodium pyruvate, HEPES, B-27 without vitamin A, heparin and growth factors (EGF, FGF, PDGF-AA, PDGF-BB) at 37 °C with 5% CO₂. Primary oligodendrocyte precursors were purified by immunopanning from P5–P7 Sprague Dawley rat and P6–P7 transgenic mouse brains as previously described⁶⁷, seeded at 150,000–250,000 cells per 10-cm dish, recovered for 4 days and trypsinized for assays. Plasticware was coated with 0.01 mg ml⁻¹ poly-D-lysine (PDL; Sigma-Aldrich, P6407), and glass coverslips with 0.01 mg ml⁻¹ PDL in 150 mM boric acid, pH 8.4. For proliferation, cells were cultured in serum-free DMEM-SATO with 4.2 μg ml⁻¹ forskolin (Sigma-Aldrich, F6886), 10 ng ml⁻¹ PDGF (Peprotech, 100-13A), 10 ng ml⁻¹ CNTF (Peprotech, 450-02) and 1 ng ml⁻¹ NT-3 (Peprotech, 450-03) at 37 °C with 10% CO₂. For differentiation, DMEM-SATO was supplemented with 4.2 μg ml⁻¹ forskolin, 10 ng ml⁻¹ CNTF, 40 ng ml⁻¹ T3 (Sigma-Aldrich, T6397) and 1× N21-MAX (R&D Systems, AR008).

CSPG4 shedding assay

Glioma neurospheres were pelleted, washed and dissociated into single cells using 5 mM ethylenediaminetetraacetic acid in calcium-free and magnesium-free HBSS with gentle trituration. Cells were resuspended in HBSS containing calcium and magnesium, counted and plated in 12-well plates. For compound treatments, cells were incubated at 37 °C with 5% CO₂ for 15 min with 5 µM ADAM10 inhibitor GI254023X, 2.5 mM 4-AP, 2 µM cytochalasin D, 10 µM Yoda1, 10 µM gadolinium chloride, 2.5 µM GsMTx4 or DMSO vehicle. For recombinant NLGN3 experiments, cells were treated with 200 nM NLGN3 ectodomain for 1 h at 37 °C with 5% CO₂.

For optogenetic stimulation, glioma cells and OPCs were transduced with lentivirus encoding ChR2, dissociated and plated as above, then stimulated with blue light at 470 nm, 1 mW mm⁻², for 0, 15, 30 or 60 s every 15 min for 1 h.

After incubation, media were collected, centrifuged at 10,000g for 10 min at 4 °C, and concentrated to approximately 40 µl using 30-kDa MWCO Amicon filters (MilliporeSigma, UFC503096) by centrifugation at 14,000g for 7 min at 4 °C. LDS sample buffer with BME (Thermo Fisher Scientific, NP0007; Sigma-Aldrich, M6250) was added, and samples were heated at 70 °C for 15 min. Proteins were separated on 3–8% Tris-acetate gels at 100 V for 10 min followed by 120 V for 45 min, and transferred to PVDF membranes using an iBlot2 high-molecular weight program—20 V for 4 min, 23 V for 4 min and 25 V for 2 min.

Membranes were blocked in 3% BSA in 0.1% TBS-T for 1 h at room temperature and incubated overnight at 4 °C with N-terminal CSPG4 antibody (Cell Signaling Technology, 52635; 1:1,000). After 3× 5 min washes in TBS-T, membranes were incubated with HRP-conjugated rabbit secondary antibody (Cell Signaling Technology, 7074; 1:10,000) for 1 h at room temperature, washed again, developed with SuperSignal West Femto substrate (Thermo Fisher Scientific, 34094) and imaged using a Bio-Rad Gel Doc system. For Extended Data Fig. 3g–i, membranes were reprobed overnight at 4 °C with C-terminal CSPG4 antibody (Abcam, ab139406; 1:1,000), followed by the same secondary antibody and detection steps.

CSPG4 band and smear intensities were quantified in Fiji using box, line and magic wand tools and normalized to vehicle controls. Only nonsaturated exposures were used for quantification, while longer exposures were used for visualization. Mouse optic pathway glioma cells were dissociated with Accutase and plated overnight on fibronectin-coated plates because they did not tolerate ethylenediaminetetraacetic acid. Normal mouse and rat OPCs were isolated using CD140a MACS microbeads or immunopanning, plated at equal densities and allowed to recover overnight due to CSPG4 sensitivity to trypsin. OPCs were washed twice the next morning with HBSS containing calcium and magnesium before assays. All experiments were repeated at least thrice.

Wide-field GCaMP6s Ca²⁺ imaging of glioma cells

To generate a stable GCaMP6s-expressing glioma cell line, cells were lentivirally transduced with a virus encoding pLV-Ef1a-GCaMP6s-p2A-nls-tdTomato^10,15. GCaMP6s-expressing glioma cells were enriched by sorting transduced cells based on the tdTomato fluorescence. For calcium imaging, cells were plated on poly-l-lysine/laminin-coated glass-bottom 24-well plates at least 24 h before imaging. On the day of imaging, cells were washed twice with 1 ml Tyrode’s solution (150 mM NaCl, 10 mM HEPES-NaOH, pH 7.3–7.4, 10 mM glucose, 4 mM KCl, 2.5 mM CaCl₂, 1.5 mM MgCl₂; 320–330 mOsm), incubated for 15 min at 37 °C with 5% CO₂ and washed once more immediately before imaging. Drug treatments were prepared as 100× stocks and added directly to wells at 1× final concentration (10 µl per well). Final compound concentrations matched those used in the shedding assay, with calcium-free and magnesium-free HBSS as vehicle. Wide-field calcium imaging was performed on a confocal microscope using GCaMP6/GFP filters. Cells were localized at 10% laser power to minimize photobleaching, and time-lapse imaging was acquired at 50% laser power. After baseline acquisition, imaging was paused, 10 µl drug stock was added and gently mixed with a 200-µl pipette tip and imaging was resumed. Cells were classified as active if ΔF/F₀ exceeded 20% at any poststimulation time point, and the active fraction was calculated as the proportion of total cells meeting this criterion. Experiments were performed in triplicate under matched culture time, temperature and imaging conditions.

Apoptosis assay by annexin V/DAPI costaining

Apoptosis was assessed using annexin V–Alexa Fluor 568 (Thermo Fisher Scientific) and DAPI. Cells were collected, washed twice with cold Ca2⁺/Mg2⁺-free HBSS and resuspended in 1× binding buffer (10 mM HEPES, 140 mM NaCl, 2.5 mM CaCl₂, pH 7.5). Cells were plated in 96-well V-bottom or U-bottom plates at 100 µl per well, stained with 5 µl annexin V–Alexa Fluor 568 and 0.5 µl DAPI (5 µg ml⁻¹ final) and incubated for 15 min at room temperature in the dark. After adding 400 µl 1× binding buffer, samples were analyzed by flow cytometry within 1 h. annexin V⁺/DAPI⁻ cells were classified as early apoptotic, annexin V⁺/DAPI⁺ as late apoptotic and annexin V⁻/DAPI⁻ as viable.

F-actin staining after cytochalasin D treatment

F-actin polymerization was assessed using an F-actin visualization kit (Cytoskeleton, BK005) according to the manufacturer’s instructions. Glioma cells were cultured on PDL-coated glass coverslips and treated with HBSS vehicle or 2 µM cytochalasin D for 15 min at 37 °C with 5% CO₂. Cells were washed, fixed in 1× fixative solution for 10 min at room temperature, permeabilized for 5 min and stained with 100 nM rhodamine phalloidin for 30 min in the dark. Coverslips were washed thrice, mounted with the supplied mounting medium, cured for 30 min in the dark and imaged by confocal microscopy using rhodamine-compatible settings (excitation = 535 nm, emission = 585 nm).

Phospho-AKT flow cytometry analysis

Cells were dissociated and treated as described for the shedding assay. After incubation, media were removed, and cells were detached with 100 µl TrypLE (Gibco) for 5 min at 37 °C. Cells were transferred to 96-well U-bottom plates containing 15 µl freshly opened 16% paraformaldehyde (PFA) and fixed for 10 min at room temperature on a shaker, protected from light. Cells were centrifuged at 1,600g for 3 min, resuspended in 200 µl ice-cold 100% methanol, mixed and incubated overnight at −20 °C. After fixation, cells were centrifuged at 1,600g for 4 min and washed twice with 200 µl wash buffer (1% BSA, 0.05% sodium azide in PBS or commercial FACS/MACS buffer). Cells were incubated with 50 µl phospho-AKT Ser473 rabbit mAb (Cell Signaling Technology, 4075; 1:50) in wash buffer for 1 h at room temperature, protected from light. After two washes, cells were resuspended in 100 µl wash buffer and analyzed immediately by flow cytometry.

Live cell membrane tension measurement

Primary cultured cells were plated on PDL-coated or poly-L-lysine-coated coverslips and stained with 1 µM Flipper-TR (Spirochrome) in imaging solution (150 mM NaCl, 10 mM HEPES, pH 7.3, 10 mM glucose, 2 mM CaCl₂, 2 mM MgCl₂, 4 mM KCl) for 30 min at 37 °C with 5% CO₂. For treatment, 400 nM NLGN3 was added during staining; 400 nM BSA or mouse IgG served as negative controls. No effect on cell viability was observed after 2–3 days. Cells were imaged by FLIM microscopy using a 488 nm pulsed laser and a 600/50 nm emission filter. Photon histograms from regions of interest or single pixels with sufficient counts were fitted with a double-exponential model.

Electrophysiology

For whole-cell recording experiments, primary cultured cells were recorded using the methods described previously⁶⁶. Cells were recorded in extracellular medium containing 150 mM NaCl, 4 mM KCl, 2 mM CaCl₂, 2 mM MgCl₂, 10 mM HEPES, pH 7.3–7.4, and 10 mM glucose, adjusted to 335–345 mOsm. Borosilicate patch pipettes (Harvard Apparatus, 5–6 MΩ) were filled with intracellular medium containing 140 mM potassium gluconate, 10 mM EGTA, 2 mM MgCl₂ and 10 mM HEPES, pH 7.2. Signals were acquired using a MultiClamp 700B amplifier and Digidata 1400 digitizer (Molecular Devices), with cells visualized on a Leica DM LFSA microscope. Whole-cell voltage-clamp recordings were performed at −70 mV. For optogenetic experiments, 460 nm light at 1.0 mW mm⁻² was delivered for 1 s, and photocurrents were measured. For mechanosensitive current recordings, cells were stimulated with isotonic extracellular medium or hypertonic medium containing 300 mM sucrose to evoke osmotic stretch-induced currents. Cell-attached recordings were performed as previously described. The external solution used to zero membrane potential contained 140 mM KCl, 1 mM MgCl₂, 10 mM glucose and 10 mM HEPES, pH 7.3, with KOH. Patch pipettes (3–6 MΩ) contained 130 mM NaCl, 5 mM KCl, 1 mM CaCl₂, 1 mM MgCl₂, 10 mM TEA-Cl and 10 mM HEPES, pH 7.3, with NaOH, with 30 µM Yoda1, 10 µM GsMTx4 or 500 nM NLGN3 added where indicated. Membrane patches were stimulated with 500-ms or 1-s negative pressure pulses using a Clampex-controlled HSPC-1 pressure clamp device (ALA Scientific), with pressure steps from 0 to −60 mmHg in −10 mmHg increments applied every 30 s. All recordings were performed at room temperature and analyzed using pClamp10 (Molecular Devices). Mechanically activated currents were quantified from peak currents measured from pressure-step onset to 1.0 s to capture late or secondary currents.

Visualization of PIEZO1 and CSPG4 expression from scRNA-seq of high-grade gliomas

Gene expression matrices corresponding to refs. ^68,69 were downloaded from GEO and first analyzed through the Seurat software package to identify and remove normal microglia and oligodendrocytes from the datasets⁷⁰. Cells were clustered according to the standard Seurat workflow and scored for microglia and oligodendrocyte gene expression programs. Both cell types form clear and discrete clusters that are removed by subsetting the Seurat object for further analysis. Remaining cells were then scored according to gene expression modules that were identified in each paper (stem like, AC like, OC like, OPC like, MES like, NPC like), then visualized by ggplot2 according to these scores. The color of each cell corresponds to a scaled expression value for PIEZO1 (top row) or CSPG4 (bottom row). Furthermore, cells were first ranked according to the PIEZO1 expression before plotting such that cells with the highest expression were plotted last.

CRISPR KO of CSPG4 from high-grade glioma cells

sgRNA oligos targeting human CSPG4 were designed using the CRISPOR pipeline (http://crispor.tefor.net/crispor.py).

guideRNA1rvU6senseMLM3636 ACACCGTTGGCCAGACTTGCATCCGCG

guideRNA1rvU6antisenseMLM3636 AAAACGCGGATGCAAGTCTGGCCAACG

Oligos were annealed and cloned into pL-CRISPR.EFS.GFP (Addgene, 57818) using BsmbI according to the Juong lab protocol available on Addgene. Plasmids containing CSPG4-targeting sgRNA were verified by Sanger sequencing. This plasmid was transfected with helper plasmids in 293T cells to produce replication deficient lentivirus. An empty pL-CRISPR.EFS.GFP plasmid was used to produce nontargeting lentivirus for controls. Collected viral supernatant was spun at 500g for 5 min, then filtered through a 0.45-micron filter before being concentrated using the Takara Lenti-X Concentrator (Takara, 631231) reagent according to the manufacturer’s instructions. Concentrated virus titer was approximated using the Takara Lenti-X GoStix Plus (631280) system according to the manufacturer’s instructions. Glioma neurospheres were dissociated to single cells and 1.5 million cells were plated in 2 ml of complete medium in one well of a six-well plate and allowed to recover for 2 h in the incubator before being infected with control or CSPG4-targeting lentivirus. After 96 h, CSPG4 KO cells were isolated through FACS. After FACS isolation, cells were returned to regular medium for propagation.

FACS of CSPG4 KO high-grade glioma cells

Ninety-six hours after infection with nontargeting or CSPG4-targeting sgRNA and Cas9, high-grade glioma neurospheres were dissociated to single-cell suspensions and stained with an Alexa Fluor 647-labeled anti-CSPG4 antibody (BD Biosciences, 562414) or an isotype-matched control antibody. CSPG4-poisitive cells were sorted into a six-well tissue culture plate containing complete medium using a BD Aria and following BL2⁺ containment practices.

CRISPR KO of PIEZO1 from high-grade glioma cells

DNA plasmid encoding a gRNA targeting human PIEZO1 was purchased from Genscript (PIEZO1 CRISPR guide RNA 3_pLentiCRISPR v2, CATCCCCAACGCCATCCGGC) and transfected with helper plasmids in 293T cells to produce replication deficient lentivirus. An empty pLentiCRISPR v2 plasmid was used to produce nontargeting lentivirus for controls. Collected viral supernatant was spun at 500g for 5 min, then filtered through a 0.45-micron filter before being concentrated using the Takara Lenti-X Concentrator (Takara, 631231) reagent according to the manufacturer’s instructions. Concentrated virus titer was approximated using the Takara Lenti-X GoStix Plus (631280) system according to the manufacturer’s instructions. Glioma neurospheres were dissociated to single cells and 1.5 million cells were plated in 2 ml of complete medium in one well of a six-well plate and allowed to recover for 2 h in the incubator before being infected with control or PIEZO1 targeting lentivirus. After 48 h, 0.5–1 μg ml⁻¹ of puromycin was added to culture medium to select for successfully infected cells. Control cells without puromycin were monitored to ensure loss of viability. Depletion of PIEZO1 protein levels from cells was confirmed through WB. After puromycin selection, cells were returned to regular medium for propagation.

Bulk RNA of glioma cells and rat OPCs

Glioma neurospheres were collected and dissociated to a single-cell suspension using Accutase (Stem Cell Technologies, 07920). Cells were counted and 400,000 were pelleted. Cell pellets were lysed in 350 μl of buffer RLT (Qiagen, 74134) and total RNA isolated following the manufacturer’s instructions for the Rneasy Mini kit. In addition, 200 ng of total RNA for each sample was used as input for reverse transcription with the Template Switching RT Enzyme Mix (NEB, M0466L). The reverse transcription primer, template switching oligo (TSO) and cDNA amplification primer (ISPCR) were sourced from the Smart-seq2 protocol⁷¹.

Oligo-dT30VN (5′–AAGCAGTGGTATCAACGCAGAGTACT30VN-3′)

TSO (5′-AAGCAGTGGTATCAACGCAGAGTACATrGrG+G-3′)

ISPCR (5′-AAGCAGTGGTATCAACGCAGAGT-3′)

cDNA was amplified using ISPCR primer and 2× KAPA HiFi ReadyMix (Roche, KK2102), cleaned by 2× SPRI bead purification and quantified with a Qubit 2.0 fluorometer. Samples were normalized to 0.2 ng µl⁻¹, and 0.8 ng cDNA was used for tagmentation according to the manufacturer’s instructions (Illumina, FC-131-1096). Tagmented DNA was indexed with dual unique indexes (Illumina, 20027213), PCR amplified, cleaned with 0.6× SPRI beads and analyzed using the high sensitivity DNA Bioanalyzer kit (Agilent, 50674626). DNA between 200 and 1,000 bp was quantified from Bioanalyzer traces, pooled to 2 nM, diluted to 650 pM and sequenced using a P2 kit (Illumina, 20046812) on an Illumina NextSeq 2000 with 2× 100 bp reads and 2× 10 bp indices. Rat OPCs were processed similarly with minor modifications. Briefly, 85,000 rat OPCs were plated on PDL-coated 12-well plates and cultured for 48 h without PDGF-AA, with 400 nM NLGN3, 5 µM ADAM10 inhibitor, 5 µM GsMTx4 or indicated combinations. After 48 h, medium was aspirated and cells were lysed directly in 350 µl buffer RLT. For the alignment and quantification of bulk RNA-seq from glioma cells and rat OPCs, demultiplexed FASTQ files were first preprocessed for quality through fastp (https://academic.oup.com/bioinformatics/article/34/17/i884/5093234) then aligned to the human or rat transcriptome through Salmon (v0.12.0)⁷². Quantifications from Salmon were then passed through tximport and summarized at the gene level according to the standard DeSeq2 workflow (https://www.bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#transcript-abundance-files-and-tximport-tximeta).

scRNA-seq

Differential gene expression analysis of PIEZO1 WT and KO glioma cells and rat OPCs

Differential gene expression analysis was performed using the DESeq2 package with references set for each experiment separately (PIEZO1 WT glioma cells and vehicle-treated rat OPCs). The results table was downloaded and used to construct a volcano plot of significantly upregulated and downregulated genes with the EnhancedVolcano R package (https://bioconductor.org/packages/release/bioc/html/EnhancedVolcano.html).

Scoring of rat OPCs for NLGN3-induced genes

The log₂-transformed transcript per million expression values for genes that were significantly upregulated in pcGBM2 after NLGN3 treatment were extracted from the rat OPC samples treated with NLGN3 or vehicle and plotted as a heatmap using the pHeatmap R package (https://www.rdocumentation.org/packages/pheatmap/versions/1.0.12/topics/pheatmap).

Gene Ontology analysis of PIEZO1 WT and KO glioma cells

Differentially expressed genes identified by DESeq2 were used as input for Gene Ontology analysis with the enrichGO R package (https://www.rdocumentation.org/packages/clusterProfiler/versions/3.0.4/topics/enrichGO).

EdU incorporation assay

Optic glioma cells (10,000 cells per well) were seeded onto fibronectin-coated (10 µg ml⁻¹) eight-chamber slides (Thermo Fisher Scientific, 154534PK) and cultured for 48 h in NSC medium lacking N2, FGF and EGF. Cells were treated with vehicle (150 mM NaCl, 50 mM HEPES) or recombinant human NLGN3 ectodomain (100 nM, produced in house) at plating and again 24 h later. EdU incorporation was measured using the Click-iT EdU Alexa Fluor 594 Imaging Kit (Thermo Fisher Scientific, C10339) according to the manufacturer’s instructions. EdU (10 µM) was added 12 h before fixation in 4% PFA. Images were acquired using a Zeiss Axio Imager M2 with Stereo Investigator software (MBF Bioscience, v2019). Proliferation was quantified as the fraction of EdU⁺ cells among DAPI⁺ cells using CellProfiler (v3.1.9). Each data point represents the average of two independent wells and three images per well.

For rat OPCs, cells were first expanded in PDGF-AA-containing medium, lifted by trypsin, and replated at 10,000 cells per 100 µl bubble on dried PDL-coated glass coverslips. After 30 min adherence, 400 µl base medium was added, and cells were incubated overnight at 37 °C with 10% CO₂. The next morning, 100 µl base medium containing EdU was added to a final concentration of 10 µM, together with NLGN3, ADAM10 inhibitor and/or PDGF-AA where indicated. After 4 h, cells were fixed in 4% PFA, stained using the Click-iT EdU Alexa Fluor 594 Imaging Kit and DAPI, and imaged on an LSM700 by a blinded investigator. Fields of view were selected using the DAPI channel to avoid counting bias.

High-grade glioma xenografts of WT and PIEZO1 KO cells

For all xenograft studies, NSG mice (NOD-SCID-IL2Rγ chain deficient, Jackson Laboratory) were used. Male and female mice were used equally. For all experiments, SU-DIPG6, SU-DIPG13pons or pcGBM2 neurospheres were dissociated using 1× Accutase. Cells were counted with Trypan Blue. Cells were pelleted and resuspended in a volume such that 100,000 cells were suspended in 1 µl of HBSS and placed on ice. Animals at postnatal day (P) 28 were anesthetized with isoflurane (1–4%) and then placed in a stereotactic apparatus. The cranium was exposed through midline incision under aseptic conditions. Approximately 300,000 cells in 3 µl sterile HBSS were stereotactically implanted into cortex (0.5 mm lateral to midline, 1 mm anterior to bregma, −1.0 mm deep to cranial surface) or the pons (1.0 mm lateral to midline, −0.8 mm posterior to lambda, −5.0 mm deep to cranial surface) through a burr hole, using a digital pump at infusion rate of 0.4 µl min⁻¹ and 26-gauge Hamilton syringe. At the completion of infusion, the syringe needle was allowed to remain in place for a minimum of 2 min, then manually withdrawn at a rate of 0.875 mm min⁻¹ to minimize backflow of the injected cell suspension.

Immunohistochemistry of PIEZO1 KO/WT SU-DIPG6 mice

Mice were anesthetized with intraperitoneal avertin and transcardially perfused with 20 ml PBS. Brains were fixed overnight in 4% PFA at 4 °C, cryoprotected in 30% sucrose, embedded in Tissue-Tek O.C.T. (Sakura), and sectioned coronally at 40 µm using a sliding microtome (AO 860, American Optical). Coronal or sagittal sections were blocked for 30 min at room temperature in 3% normal donkey serum and 0.3% Triton X-100 in TBS, then incubated overnight at 4 °C with rabbit anti-Ki67 antibody (Abcam, ab15580; 1:500) diluted in 1% normal donkey serum and 0.3% Triton X-100 in TBS. Sections were rinsed with TBS, stained with DAPI (1 µg ml⁻¹ in TBS; Thermo Fisher Scientific), rinsed again and incubated overnight at 4 °C with Alexa 594 donkey antirabbit IgG (Jackson ImmunoResearch; 1:500). Sections were washed thrice with TBS and mounted with ProLong Gold mounting medium (Life Technologies). Images were acquired using a Zeiss Airyscan 800 or Zeiss Airyscan LSM980 with Zen 2011 (v8.1). Proliferation index was calculated as the fraction of Ki67⁺ cells among human nuclear antigen-positive cells in ×20 confocal images by a blinded investigator.

Confocal Imaging of SU-DIPG13pons mice

For three-dimensional reconstruction, SU-DIPG13pons xenografted mice were treated in the same manner as for immunohistochemistry of SU-DIPG6 mice, using a 1:500 dilution of rabbit PIEZO1 antibody (Proteintech, 15939-1-AP) and chicken anti-GFP antibody (Abcam, ab13970). This was followed by the application of antichicken 488 (Jackson ImmunoResearch, 703-545-155) and antirabbit 594 secondary antibodies (Jackson ImmunoResearch, 111-585-144). Images were processed using IMARIS (v10.0.0) software.

OPC differentiation assay

Rat OPCs were initially plated and expanded in PDGFAA-containing medium as described⁶⁷. Cells were lifted with trypsin and replated at 10,000 cells per 100 µl drop on dried PDL-coated glass coverslips. After 20 min of adherence, 400 µl base medium was added and supplemented with 5 µM DMSO vehicle, 20 ng ml⁻¹ PDGF-AA, 400 nM NLGN3, 5 µM ADAM10 inhibitor, 5 µM GsMTx4 or specified combinations. Cells were incubated at 37 °C with 10% CO₂ for 24 h, supplemented again with NLGN3 and fixed with 4% PFA at 48 h. Cells were blocked and permeabilized in 3% normal donkey serum and 0.5% Triton X-100 for 30 min at room temperature, then incubated overnight at 4 °C with anti-MBP (Abcam, ab7349), CSPG4 (gift from Jacqueline Trotter) and OLIG2 (R&D Systems, AF2418), each diluted 1:1,000 in 1% normal donkey serum and 0.3% Triton X-100. Coverslips were washed 3 × 5 min in TBS, incubated with fluorescence secondary antibodies (Jackson ImmunoResearch; 1:500) for 1 h at room temperature, washed again 3 × 5 min in TBS, dried in the dark and mounted with ProLong Gold (P36930). Cells were imaged on an LSM700 confocal microscope with a ×20 objective. For all treatment groups, fields of view were selected using the DAPI channel to avoid bias, with the goal of capturing evenly spaced nuclei. Two-dimensional images were acquired per coverslip across four channels, with two coverslips processed per experiment, and the experiment was replicated. Cell classification was performed in Fiji using the Cell Counter plugin. Total DAPI⁺ nuclei were first counted, followed by OLIG2⁺/DAPI⁺ nuclei. Cells were then classified by morphology using CSPG4 staining into OPC-like, star-like or spider-like categories, while MBP staining was used to identify arborized to lamellar differentiated cells. Data were reported as the total number of cells in each differentiation stage, allowing simultaneous assessment of differentiation state and changes in cell number across conditions.

Quantification of CSPG4-to-MBP ratio in primary rat OPCs

Rat OPCs from the differentiation assay were analyzed for CSPG4 and MBP expression as markers of OPCs and differentiated oligodendrocytes, respectively. For each experiment, treatment conditions were replicated across three coverslips, with eight z-stacks acquired per coverslip and converted to maximum-intensity projections. OLIG2⁺ nuclei, CSPG4⁺ cells and MBP⁺ cells were counted per field, and differentiation status was reported as the CSPG4:MBP-positive cell ratio.

qRT–PCR of rat OPCs undergoing spontaneous differentiation

Rat OPCs used in gene expression studies were identically manipulated as above. Cells were directly collected from coverslips in 24-well plates. Cells were first differentiated as described above in the presence or absence of NLGN3 and then washed with HBSS before being lysed in buffer RLT from the Qiagen RNeasy mini kit. Total RNA was isolated and 100 ng from each experimental condition was added to reverse transcription reaction using the New England Biolabs Protoscript II reverse transcriptase enzyme. Equal volumes of cDNA were added to each qRT–PCR reaction. Expression primers were previously validated^59,73. Fold change in gene expression was calculated using the delta–delta Ct method relative to GAPDH expression.

Statistics and reproducibility

No statistical methods were used to predetermine sample sizes. Sample sizes were chosen based on those commonly used in comparable studies and were equal to or larger than those reported in such studies. All experiments were performed at least thrice independently to obtain representative images and graphs. No data were excluded from the analyses. Where applicable, experiments were randomized, and investigators were blinded to allocation during experiments and outcome assessment, as described in the Methods. For calcium imaging, we performed experiments in three independent biological replicates, defined as experiments conducted on different dates with independently isolated, treated and imaged cells. Individual cell measurements within each experiment were considered technical replicates nested within each biological replicate. For western blot experiments in Fig. 3, the NLGN3 control in Fig. 3f,m is from the same run (first run, n = 4, all six conditions). Fig. 3m adds an independent second run (n = 4) for CytoD-related conditions (n = 8 total). For each condition, cell-level responses were averaged to generate one replicate-level summary value for each biological replicate. Independent calcium imaging datasets are presented separately in Extended Data Fig. 2e, corresponding to Fig. 3, and Extended Data Fig. 4a, corresponding to Fig. 4. Where applicable, statistical analyses were performed at the biological replicate level using replicate-level means. Power analyses were conducted using a two-sided two-sample t-test framework, with α = 0.05, pooled inter-replicate variance, Cohen’s d effect sizes and standard normal approximations to estimate the number of biological replicates required. For Fig. 3, statistical testing was performed directly using n = 3 independent biological replicates per condition, and replicate-level statistics are reported herein. For Fig. 4, power analyses indicated that substantially larger numbers of biological replicates would be required for some comparisons. Therefore, replicate-level statistical testing was not emphasized, and the results were interpreted in the context of limited statistical power.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41593-026-02397-8.

Author contributions

Y.S.K., S.M.G., A.C.G., B.Y., C.M. and Y.P. designed, conducted and analyzed experiments. R.M., A.E.I., J.R., R.D., Y.G.B., J.H., M.Q., K.M., P.W., A.Y. and M.L. conducted and analyzed experiments. Y.S.K., S.M.G. and M.M. wrote the manuscript. J.B.Z. and J.T. provided reagents and insights. M.M. conceived the study and supervised all aspects of the work.

Peer review

Peer review information

Nature Neuroscience thanks the anonymous reviewer(s) for their contribution to the peer review of this work.

Funding

The authors acknowledge funding from the National Institute of Neurological Disorders and Stroke (R01NS092597 to M.M.), NIH Director’s Pioneer Award (DP1NS111132 to M.M.), National Cancer Institute (P50CA165962, R01CA258384 and U19CA264504 to M.M.), Gatsby Charitable Foundation (Gatsby Initiative in Brain Development and Psychiatry; to M.M.), Oscar’s Kids Foundation (to M.M.), McKenna Claire Foundation (to M.M.), Kyle O’Connell Foundation (to M.M.), Yuvaan Tiwari Foundation (to M.M.), Virginia and D.K. Ludwig Fund for Cancer Research (to M.M.), Waxman Family Research Fund (to M.M.), Will Irwin Research Fund of the Pediatric Cancer Research Foundation (to M.M.), Austin Strong Foundation (to M.M.), Avery Huffman DIPG Foundation (to M.M.), Chadtough Defeat DIPG (to M.M.), Jacob Van de Roovaart Glioblastoma Research Fund (to M.M.), The William E. and Carolyn Reller Oligodendroglioma Research Fund, Botha-Chan Family Research Fund (to M.M.), Shoshana Levy Early Career Fellowship (to B.Y.), TR231/10-1 from the Deutsche Forschungsgemeinschaft (to J.T.), Cancer Prevention and Research Institute of Texas (RR210085 to Y.P. as the CPRIT scholar in Cancer Research), Walter V. and Idun Berry Fellowship (to Y.S.K.) and Chadtough Defeat DIPG Fellowship (to Y.S.K.).

Data availability

The source data for this manuscript are available with the manuscript. Any additional information required to reanalyze the data reported in this paper is available from the corresponding authors upon request. Bulk RNA-seq alignments for MGG8 cells and primary rat OPCs are available to download at https://doi.org/10.6084/m9.figshare.32609211. Source data are provided with this paper.

Code availability

Custom R notebooks for the MGG8 PIEZO1 KO and the rat OPC analyses are available using figshare at https://doi.org/10.6084/m9.figshare.32609211.

Competing interests

M.M. holds equity in MapLight Therapeutics and Stellaromics, and is on the SAB of S1 Oncology and TippingPoint Biosciences. The remaining authors declare no completing interests.

Extended data

is available for this paper at https://doi.org/10.1038/s41593-026-02397-8.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41593-026-02397-8.