Abstract

   Glioblastoma (GBM) is the most aggressive and common type of primary
   malignant brain cancer in adults. GBM often recurs locally near the
   resection cavity (RC) following the surgical removal of primary tumors.
   Recent research has reported that neural stem cells (NSCs) in the
   subventricular zone (SVZ) harboring cancer-driving mutations serve as
   the cells of origin for human GBM. However, the pathological role of
   tumor-initiating NSCs in the SVZ in tumor recurrence remains to be
   elucidated. Here, we explore the potential contribution of
   mutation-harboring NSCs in the SVZ to tumor recurrence around the RC
   following surgical resection. Our hypothesis emerged from performing
   deep sequencing of longitudinal tissues from 10 patients with GBM,
   including (i) tumor-free SVZ tissue, (ii) primary tumor tissue, (iii)
   recurrent tumor tissue, and (iv) blood. As a result of this sequencing,
   we observed evidence suggesting that recurrent tumors show genetic
   links to the SVZ in 60% (6/10) of patients, which are distinct from the
   primary tumors. Using a genome-edited mouse model, we further
   identified that mutation-harboring NSCs appeared to migrate to the RC
   through the aberrant growth of oligodendrocyte progenitor cells,
   potentially contributing to the reconstruction of high-grade malignant
   gliomas in the RC. This process was associated with the CXCR4/CXCL12
   axis, as supported by RNA sequencing data from human recurrent GBM.
   Taken together, our findings suggest that NSCs in human SVZ tissue may
   play a role in GBM recurrence, potentially highlighting a novel distant
   contributor of recurrence.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12943-025-02273-2.

   Keywords: Glioblastoma, Recurrence, Subventricular zone, Neural stem
   cells

Introduction

   Glioblastoma (GBM) is the most aggressive diffuse glioma, and it
   accounts for 54% of all gliomas and 16% of all primary brain tumors
   [[62]1, [63]2]. The current standard of care for newly diagnosed GBM is
   maximal surgical resection followed by concurrent radiotherapy and
   temozolomide [[64]3, [65]4]. Although there have been advancements in
   diagnosis and treatment, the prognosis of patients with GBM has not
   significantly improved, as it has a median survival of 14–20 months
   [[66]5, [67]6]. The poor prognosis of GBM can be attributed to several
   interrelated factors, including infiltrative growth, complex genetic
   and molecular heterogeneity, therapy resistance, and a pro-tumorigenic
   microenvironment, all of which contribute to a high tumor recurrence
   rate nearing 100% [[68]7, [69]8]. Recurrence has been shown to occur in
   approximately 80% of aggressively treated patients within the resection
   cavity (RC), even after gross total resection of the primary tumor
   [[70]9, [71]10]. Thus, we hypothesize that alternative cellular
   origins, in addition to residual tumor cells at the resection margin,
   may be responsible for tumor recurrence.

   Recent studies have shown that neural stem cells (NSCs) in the
   subventricular zone (SVZ) harbor cancer-driving mutations and
   potentially serving as the cellular origin for human GBM
   [[72]11–[73]15]. These cells are hypothesized to play a role in tumor
   recurrence due to their retained tumor-initiating capacity and
   migratory potential [[74]16]. Adult NSCs in the SVZ typically migrate
   along the rostral migratory stream to the olfactory bulb [[75]17];
   however, these cells can change their path to move toward damaged areas
   following brain injury [[76]18]. This altered migration is guided by
   chemokines and cytokines that are released from the injured brain
   region, including tumor necrosis factor-alpha (TNFα), interferon-gamma
   (IFNγ), monocyte chemotactic protein-1 (MCP-1), and
   stromal-derived-factor-1 (SDF-1) [[77]19–[78]21]. Although this
   migration diminishes with age [[79]22, [80]23], adult NSCs still
   possess the ability to migrate and differentiate following ischemic
   injury [[81]24, [82]25]. However, the specific association between
   mutation-harboring NSCs in SVZ and GBM recurrence after maximal
   surgical resection remains poorly understood. To address this, we
   established a genome-edited mouse model by labeling mutation-harboring
   SVZ NSCs with a different reporter protein than that used for tumor
   cells, thus allowing us to trace the origin of GBM recurrence. We also
   conducted deep sequencing of longitudinal tissues from 10 GBM patients,
   with the results providing evidence indicating that mutation-harboring
   NSCs in SVZ are involved in the process of tumor reformation within the
   RC following complete surgical excision of the primary tumor.

Results

Genetic profiles and clonal relationship linking the SVZ with primary and
recurrent tumors

   To investigate whether SVZ NSCs are clonally connected to primary and
   recurrent tumors, we analyzed the mutation patterns in paired primary
   and recurrent tumor tissues, along with tumor-free SVZ samples.
   Recurrence samples were prospectively collected between 2013 and 2022
   from patients exhibiting features of clonal evolution at the time of
   their primary tumors after gross total resection, defined by (i)
   low-frequency cancer-driving mutations in the SVZ, and (ii) the
   emergence of tumor-private mutations in the primary tumor. We initially
   performed deep sequencing on three matched tissue samples from each
   patient with IDH-wildtype GBM, including (i) tumor-free SVZ tissue,
   (ii) primary tumor tissue, and (iii) blood. Deep hybrid capture
   sequencing (median depth: 1063x) was conducted to detect somatic base
   substitution and short indels in genes commonly mutated in GBM. From
   this initial cohort, 10 patients who experienced tumor recurrence after
   gross total resection of the primary tumors were selected for further
   analysis through deep sequencing of their recurrent tumor samples
   (Fig. [83]1a).

Fig. 1.

   [84]Fig. 1
   [85]Open in a new tab

   Evidence of SVZ NSC involvement in glioblastoma recurrence with
   cancer-driving mutations. a, Schematic diagram of sampling for human
   longitudinal paired primary, recurrent tumors and SVZ tissues, which
   was followed by application for deep sequencing. Created with
   Biorender.com. b, Oncoplot displaying DNA panel sequencing results from
   longitudinal paired primary and recurrent tumors and SVZ tissues from
   10 patients. Mutations are categorized as SVZ-Tumor-shared (green),
   Inter-Tumor-shared (black), Primary-private (blue), and
   Recurrent-private (red). An asterisk (*) denotes a distinct PTEN
   mutation present in the recurrent tumor of GBM-1250 (g.87933010 C > G)
   compared to the primary tumor (p.Asp252Val). Double asterisks (**)
   indicate a distinct SPTA1 mutation present in the recurrent tumor of
   GBM-1250 (g.158680532 C > T) compared to the primary tumor
   (g.158666421 A > C). A dagger (†) marks a TP53 mutation found in the
   recurrent tumor of GBM-1938 (c.524 C > T) compared to the primary tumor
   (c.613 A > G). A double dagger (‡) marks PIK3CA mutations found in the
   recurrent tumor of GBM-1938 (p.Gly118Asp) compared to the primary tumor
   (g.179239627 A > T). A section sign (§) highlights a distinct PTEN
   mutation in the recurrent tumor of GBM-2488 (g.87864488-87864493del)
   compared to the primary tumor (g.87961042_87961045del). Annotation bars
   indicate tissue type and recurrence pattern, classified as
   SVZ-associated, Primary-associated, or Unknown. c, MRI image of the
   sampling sites in GBM-1932 patient, with the red box representing the
   paired tumors and the yellow box representing the SVZ. d, VAF
   scatterplot of mutations from triple-matched samples of GBM-1932. Each
   point represents an individual somatic mutation identified in the
   samples. Mutations that are private to the SVZ (x-axis), private to
   tumor (y-axis), or shared between the two tissues are plotted according
   to their VAF. Blue dots indicate mutations found in the primary tumor,
   while red dots represent mutations in the recurrent tumors. e,
   Phylogenetic evolutionary tree of the evolutionary process of primary
   and recurrence in patient GBM-1932. Created with Biorender.com. f, MRI
   image of the sampling sites in GBM-244 patient, with the red box
   representing the paired tumors and the yellow box representing the SVZ.
   g, VAF scatterplot of mutations from triple-matched samples of GBM-244.
   Each point represents an individual somatic mutation that is identified
   in the samples. Mutations that are private to the SVZ (x-axis), private
   to tumor (y-axis), or shared between the two tissues are plotted
   according to their VAF. Blue dots indicate mutations found in the
   primary tumor, while red dots represent mutations in the recurrent
   tumors. h, Phylogenetic evolutionary tree of the evolutionary process
   of primary and recurrence in patient GBM-244. i, Schematic diagram of
   transplanted primary tumor removal model. First, primary glioblastoma
   cells were dissected from the brain of mice bearing tdTomato-positive
   primary tumor and cultured as suspension cells. Four weeks after
   electroporation, these cells were implanted into the cerebral cortex of
   mice with GFP-positive SVZ NSCs. One week later, a bulky tumor formed
   and was surgically removed. Recurrent tumors were shown in the RC
   post-resection. Created with Biorender.com. j, Bar chart showing the
   proportion of recurrent tumors in GFP, tdTomato, and Mix (mixture of
   GFP and tdTomato). k, Representative coronal sections of mice stained
   with DAPI showing sections from tumor-implanted mice at seven days
   after implantation as well as 0, 14, and 28 days after surgical removal
   of the primary tumor. The asterisk indicates the RC. The white dashed
   line indicates the SVZ and RC sites. Scale bars, 500 μm. l,
   Kaplan-Meier survival analysis of mice (n = 5 per group, log-rank test)
   in the control group and SVZ-mutated group with or without primary
   tumor resection

   In the 10 patients with recurrence samples, TERT promoter mutations
   were consistently shared between the tumors and their paired SVZ,
   indicating a clonal lineage link (Fig. [86]1b). Additionally, two
   patients (20%)―Patient IDs 1932 and 1492―exhibited additional mutations
   shared between the primary and recurrent tumors that were distinct from
   those found in the SVZ (Fig. [87]1c-e, Supplementary Fig. [88]1). We
   categorized this recurrence pattern as “primary-associated,” as the
   recurrent tumors appeared to evolve from dominant clones of the primary
   tumor within the same clonal lineage, suggesting local recurrence
   likely driven by residual infiltrative tumor cells near the RC. For two
   other patients—Patient IDs 2455 and 2437—we were unable to determine
   the clonal lineage of the recurrence and, therefore, categorized them
   as “unknown.”

   In contrast, the other six patients (60%)―Patient IDs 1250, 1938, 2488,
   2441, 260, and 244―did not exhibit any inter-tumor shared mutation
   between the primary and recurrent tumors excepted the mutations shared
   between SVZ and tumors (Fig. [89]1b, Supplementary Fig. 2). TERT
   promoter mutations were consistently shared among the SVZ, primary, and
   recurrent tumors in all cases. Notably, GBM-1938 shared three
   mutations, including TERT promoter, PDGFRA, and PTEN (Supplementary
   Fig. [90]2d-f). We categorized this recurrence pattern as
   “SVZ-associated,” as the high-allele frequency, tumor-private mutations
   of the primary tumors (primary-private mutations) became extinct in the
   recurrent tumors in these patients, indicating a lack of direct genetic
   linkage between the primary and recurrent tumors. Additionally,
   recurrent tumors exhibited de novo genetic variations
   (recurrent-private mutations) or reversion to normal alleles due to the
   loss of primary-private mutations. These findings suggest that the
   recurrent tumors may undergo independent clonal evolution and maintain
   genetic links to the SVZ, distinct from the primary tumor, especially
   following the complete resection of the primary tumors. To further
   investigate the dynamics of primary-private and/or recurrent-private
   somatic mutations, we performed whole genome sequencing (WGS) followed
   by deep amplicon sequencing for validation in a single patient who
   harbored the TERT promoter mutation (C228T) shared among the SVZ,
   primary, and recurrent tumors (Fig. [91]1f). In this case, we
   identified 10 passenger mutations in addition to the EGFR mutation.
   Notably, five primary-private mutations were absent in the recurrent
   tumor, while five recurrent-private mutations were not present in the
   primary tumor (Fig. [92]1g, h). Additionally, the copy number variation
   profile shifted from the primary to the recurrent tumors (Supplementary
   Fig. [93]3).

   As a result, 60% of patients demonstrated no direct clonal linkage
   between primary and recurrent tumors, except for the shared mutations
   involving the SVZ. The distinct mutational patterns observed in
   recurrent tumors suggest that recurrence may originate from early
   progenitor clones in tumor evolution—harboring only shared
   mutations—rather than evolving directly from the dominant primary tumor
   clone. This model, in which recurrence emerges through a branched
   evolutionary trajectory from an early progenitor, aligns with previous
   studies comparing primary and recurrent tumors [[94]26–[95]28].
   Notably, the genetic profiles of the SVZ, characterized by the presence
   of shared mutations without tumor-private mutations, suggest that the
   SVZ may serve as such an early progenitor driving recurrence
   independently of the primary tumor. Alternatively, other early
   progenitor clones responsible for recurrence may have remained
   undetected in primary tumor sequencing due to spatial heterogeneity and
   the effects of treatment on clonal distribution in GBM. Given the
   limitations of cross-sectional human genomic data in resolving these
   hypotheses, we propose in vivo modeling and time-sequenced longitudinal
   observations to investigate whether the SVZ has the potential to drive
   independent tumor recurrence.

Multiple cellular origins for tumor recurrence

   To test the hypothesis that GBM recurrence directly originates from
   NSCs in the SVZ, we developed a mouse model of tumor recurrence
   (Fig. [96]1i). First, we used postnatal electroporation to generate
   primary tumors by introducing Trp53, Pten and Egfr mutations into NSCs
   of the SVZ [[97]12]. This single vector expressed Cre recombinase along
   with highly efficient single-guide RNAs (sgRNAs) targeting Trp53 and
   Pten in the lateral ventricle of LSL-EGFRviii; LSL-Cas9-GFP mice
   (Supplementary Fig. [98]4a-c). At four weeks of age, we transplanted
   tdTomato-expressing mouse GBM cells into the cerebral cortex. To
   determine the origin of tumor recurrence, we isolated tumor cells from
   the same mouse model, which had been labeled with a different reporter
   protein and cultured them in vitro as tumorspheres prior to
   transplantation. These cells formed a solid tumor mass within one week,
   which we then removed through maximal surgical resection. Recurrent
   tumors were observed four weeks post-surgery, where the color of
   fluorescent proteins in tumor cells indicated the cellular origin of
   recurrent tumor.

   Interestingly, all observed recurrent tumors developed within the RC,
   with 53.9% of recurrent tumors (7 out of 13 cases) being GFP-positive.
   This indicates that GFP-labeled SVZ cells had migrated and
   reconstructed secondary tumors, which was identified as recurrence
   within the RC according to the clinical timeline (Fig. [99]1j,
   Supplementary Fig. [100]7b). Migration of GFP-positive SVZ cells to the
   RC was observed even two weeks post-surgery (Fig. [101]1k). Despite the
   extensive resection, residual tdTomato-labeled tumor cells repopulated
   at the surgical margin, thus generating tdTomato-positive tumors
   (30.8%), ultimately suggesting direct recurrence from primary residual
   tumor cells. The remaining 15.4% tumors comprised both cell types,
   suggesting a mixture of both origins. Survival analysis for our mouse
   model for GBM recurrence revealed that recurrent tumors originating
   from mutation-harboring SVZ cells were associated with the worst
   prognosis, with a median survival time of approximately 8 weeks, when
   compared to recurrent tumors derived from residual primary tumors
   (Fig. [102]1l). These findings prompted further exploration of the
   possible mechanisms underlying GBM recurrence originating from
   mutation-harboring SVZ NSCs.

Mouse modelling of post-resection tumor development driven by NSCs

   To eliminate the potential influence of residual tumor cells in
   generating recurrent tumors from our analysis, we optimized our mouse
   model by performing a surgical resection of the cortex, where we
   simulated the post-resection status following a gross total resection
   (GTR) of the primary tumor (GTR model) (Fig. [103]2a). Briefly, we
   performed postnatal electroporation of a vector expressing Cas9 and Cre
   recombinase, along with sgRNAs targeting Trp53 and Pten into the SVZ of
   LSL-EGFRviii; LSL-tdTomato mice (Supplementary Fig. [104]4d-f). At four
   weeks of age, we performed a surgical resection of the distant cortex,
   while ensuring minimal disturbance to the adjacent white matter and
   observed the behavior of mutation-harboring SVZ cells migrating to the
   RC (Fig. [105]2b). Aligning with these expected findings, 64.7% of the
   mice (11 out of 17) developed brain tumors surrounding the RC four
   weeks after surgery. Histological examination of these tumors revealed
   the classical features of highly proliferative, high-grade glioma
   (Fig. [106]2c); by contrast, control mice with SVZ NSCs targeted by
   sgLacZ exhibited no tumor formation.

Fig. 2.

   [107]Fig. 2
   [108]Open in a new tab

   Glioma reconstruction at RC in genome-edited mice carrying cancer
   mutations in NSCs from the SVZ. a, An experimental model used in an
   attempt to replicate the local reconstruction of glioma at the RC.
   Initially, a plasmid containing single guide RNA (sgRNA) targeting
   towards the p53 and Pten genes, along with the expression of Cas9 and
   Cre recombinase, was injected into the SVZ of a LSL-EGFRviii f/+;
   LSL-tdTomato f/+ mouse, resulting in the creation of an SVZ-mutated
   model. A control group was also established, wherein a
   Cas9/Cre-expressing plasmid containing sgRNA for the lacZ gene was
   injected into the SVZ of a LSL-tdTomato f/+ mouse. Subsequently, a
   surgical resection was carried out 28 days post-injection, with the aim
   of excising distant cortical tissue while minimizing any possible
   damage to the adjacent white matter. Then, the brains of both the
   experimental and control groups were harvested, and tissue sections
   were collected via cryosectioning at various time points following the
   surgical resection, specifically 2, 14, and 28 days post-resection,
   referred to as POD2, POD14, and POD28, respectively. (lower panel)
   Representative MRI images of local recurrent tumor development in RC at
   POD14 and POD28. Created with Biorender.com. b, Representative coronal
   sections of mice stained with DAPI at 2, 14, and 28 days post-resection
   in mice of each group (n = 3). The asterisk indicates the RC. Images
   from SVZ-mutated mice show that tdTomato-positive cells were initially
   localized in the rostral SVZ, where mutations were introduced. Over
   time, these cells migrate to the RC and proliferate to form a tumor;
   however, there is no tumor formation in the control group. The white
   dashed line indicates the SVZ and RC sites. Scale bars, 500 μm. c,
   Representative images of mouse tumor sections stained with H&E. Scale
   bars, 200 μm. d-e, Quantification of tdTomato immunofluorescence
   intensity (relative to DAPI) in ROI1-3 (Region of Interest) regions at
   different timepoints post resection of the SVZ-mutated group, compared
   with the control group. ROI 1 was defined as any random area of the
   brain excluding the SVZ or RC, ROI 2 as the SVZ region, and ROI 3 as
   the RC site. *P = 0.0423, **P = 0.0017, **** P<0.0001 (n = 3 for
   control and mutant mice at each time point). Error bars represent
   mean ± s.e.m. f, Representative immunostaining images of RC at POD14,
   stained with GFAP, Olig2, PDGFRa and Neun. Scale bars, 20 μm. g,
   Quantification of lineage marker-positive tdTomato-positive cells at RT
   at POD14 in mutation-harboring SVZ group. *P = 0.0012(GFAP vs. Olig2).
   *P = 0.0427 (GFAP vs. PDGFRa). Student’s two-tailed t-test. Error bars
   represent mean ± s.e.m. h, Comparison of GFP positive cell number in RC
   with virus injection, SVZ, and cortex with virus injection,
   respectively. *P = 0.0295 (left panel), *P = 0.0181(right panel)
   (n = 3). Student’s two-tailed t-test. Error bars represent mean ± s.e.m

   To monitor the proliferation and migration of tdTomato-expressing SVZ
   cells, we selected three regions of interest (ROI), i.e., the SVZ, RC,
   and distant ventral region to analyse the fluorescent intensity
   detected in each ROI. Increased proliferation of SVZ cells was detected
   two days after cortical resection, with these cells appearing within
   the RC starting at 14 days in both SVZ cells with and without driver
   mutations (Fig. [109]2d, e). However, only the mutation-harboring SVZ
   cells were able to reconstruct tumors within the RC by 28 days
   (Fig. [110]2b). To rule out the influence of spatial proximity between
   the surgical site and the SVZ, we performed surgical resection in the
   distant caudal brain of the parietotemporal lobe, after which we still
   observed tumor occurrence at the distant RC (Supplementary
   Fig. [111]5). We also conducted immunostaining analyses to characterize
   the cellular subtypes of the migrating cell population derived from the
   SVZ (Fig. [112]2f, g). Notably, Olig2 + oligodendrocyte progenitor
   cells (OPCs) emerged as the predominant cellular subtype among
   tdTomato-positive cells migrating to the RC at 14 days post-surgery,
   suggesting that aberrant proliferation and differentiation into the OPC
   lineage of mutation-harboring NSCs are essential for recurrent tumor
   formation (Fig. [113]2h). Overall, our findings provide valuable
   insights into the spatiotemporal dynamics underlying GBM recurrence
   after surgical resection.

Upregulation of CXCR4 expression in migrating SVZ cells

   Next, to elucidate the factors involved in the migration and tumor
   formation of mutation-harboring NSCs in the SVZ, we analysed the
   transcriptomic gene signatures of the RC. To this end, we collected the
   tissues surrounding the RC on post-operative days (POD) 2, 14, and 28,
   and assessed them using RNA-seq analyses. We focused on genes that are
   consistently upregulated in the RC compared to the normal cortex, and
   we identified 106 co-upregulated genes across all time points
   (Fig. [114]3a). Subsequent KEGG analysis revealed that these genes were
   enriched in the cytokine cytokine-receptor interaction pathway
   (Fig. [115]3b, c). Notably, we identified eight cytokine genes that
   were significantly co-upregulated in our model: Cxcr4, Ltbr, Ccl22,
   Ccr7, Ccl4, Cxcl16, Tnfrsf1a, and Tnfrsf11b (Fig. [116]3d). Among them,
   CXCR4 expression was found to be highly elevated in migrating
   tdTomato-positive cells at the RC, and this was validated through
   staining of the RC and normal cortex (Fig. [117]3e, f, Supplementary
   Fig. [118]6).

Fig. 3.

   [119]Fig. 3
   [120]Open in a new tab

   Cxcr4 was highly expressed at the RC after surgical resection and
   correlated with migration of SVZ-mutated NSCs. a, Venn diagram
   demonstrating the overlap between upregulated differentially expressed
   genes (DEGs) detected via the RNA-sequencing analysis of samples from
   the POD2, POD14, and tumor groups versus the control group samples. b,
   Bubble plot showing the overlap between enriched KEGG pathways detected
   via the RNA-sequencing analysis of samples from the POD2, POD14, and
   tumor groups versus the control group samples. c, Heatmap showing the
   expression of the geneset of the cytokine-cytokine receptor interaction
   pathway in the control, POD2, POD14, and tumor groups. d, Comparison of
   the expression levels of eight genes filtered out through panel d among
   POD2, POD14, and tumor groups versus control group samples. e,
   Representative immunostaining images of CXCR4 in the mouse SVZ at POD2
   after surgical resection. The white arrowheads represent CXCR4-positive
   cells colocalizing with tdTomato + cells in the SVZ-mutated group.
   Scale bars, 50 μm. f, Representative immunostaining images of CXCR4 in
   the RC on POD14. The white arrowheads represent CXCR4-positive cells
   colocalizing with tdTomato + cells in the SVZ-mutated group. Scale
   bars, 50 μm. g, Boxplot showing the proportion of endothelial cells in
   control, POD2, and POD14 groups after deconvolution of RNA-seq from a
   mouse model. h, t-SNE plots from the public dataset ([121]GSE160763)
   showing upregulated expression of representative marker genes of
   endothelial cells (Ly6c1 and Cldn5) as well as Cxcl12 in mice in the
   traumatic brain injury group. i, Representative immunostaining images
   of CD31 and CXCL12 in the RC on POD2 after surgical resection. Scale
   bars, 20 μm, and the white arrowheads represent CD31-positive
   endothelial cells colocalizing with CXCL12. Scale bars, 20 μm

   To identify the potential source of CXCL12, we performed deconvolution
   analysis of our RNA-seq data along with single-cell RNA-seq analysis.
   Studies have shown that CXCL12 is released by multiple cell types in
   the brain, including neuronal, glial, and endothelial cells.
   Deconvolution analysis indicated that the proportion of endothelial
   cells around the RC was significantly increased at POD2 (Fig. [122]3g).
   Single-cell RNA-seq analysis of a brain trauma model ([123]GSE160763)
   revealed that Cxcl12 expression was specific to Cldn5-expressing
   endothelial cells (Fig. [124]3h), and this was further confirmed by the
   findings of immunostaining analysis showing augmented secretion of
   CXCL12 in endothelial cells (Fig. [125]3i). This increased release of
   CXCL12 surrounding the RC in turn likely facilitates the migration of
   CXCR4-positive NSCs to this site, thus contributing to tumor
   recurrence.

Clinical implications of the CXCR4 pathway in tumor recurrence

   To further investigate the involvement of CXCR4 in recurrent tumor
   development, we conducted a comparative transcriptomic analysis between
   primary and recurrent tumors in our mouse model of GBM. Differentially
   expressed gene analysis identified 576 up-regulated and 531
   down-regulated genes in recurrent tumors (Fig. [126]4a). Based on the
   results of our time-course transcriptome analysis, we hypothesized that
   the CXCR4 pathway might be enhanced in recurrent malignancies. Indeed,
   the expression of genes in the CXCR4 pathways was found to be elevated
   in recurrent tumors (Fig. [127]4b). Immunostaining analysis also
   demonstrated a significant increase in CXCR4 levels in recurrent tumors
   (P < 0.0001) (Fig. [128]4c, d). Moreover, our GBM recurrence model
   revealed increased expression of Cxcr4 and enrichment of CXCR4 pathways
   in GFP-positive and mixed recurrent tumors, suggesting that CXCR4 plays
   a critical role as a key regulator in SVZ-derived tumor recurrence
   (Supplementary [129]7).

Fig. 4.

   [130]Fig. 4
   [131]Open in a new tab

   CXCR4 expression in the GBM reconstructed after surgical resection in
   mice and human GBM patients was increased when compared to primary
   tumors. a, Heatmap of the top 50 differentially expressed genes between
   primary and recurrent GBM groups. b, Heatmap depicting the expression
   levels of the 18 signature genes between recurrent and primary tumors.
   c, Immunostaining images of Cxcr4 expression in primary and recurrent
   GBM tissues of mice, with the lowest panel showing the tumor tissues
   from the recurrent GBM after primary tumor removal. d, Quantification
   of Cxcr4 expression levels in each group. Statistical analysis was
   performed using a two-sided Student’s t-test (****P < 0.0001). Data are
   presented as mean ± SEM. e, Schematic illustration of tumor recurrence
   patterns following gross total resection (GTR) and subtotal resection
   (STR). Cases of local recurrence possibly originating from neural stem
   cells were classified as the LR group, while other cases were
   categorized as the non-LR group. Created with Biorender.com. f, Heatmap
   showing the expression of the CXCR4 pathway geneset in the LR and
   non-LR groups. g-h, Comparison of CXCR4 expression levels (g) and CXCR4
   pathway enrichment scores (h) in the LR and non-LR tumor samples.
   Statistical analysis was performed using a two-sided Student’s t-test
   (*P < 0.05). Data are presented as mean ± SEM. i, Representative
   immunostaining images of CXCR4 expression in recurrent GBM tissues from
   human patients (#SNU-F3T2, LR group; #SNU-F1T2, non-LR group). Scale
   bars, 50 μm. j, Overall survival curve of recurrent GBM patients in the
   LR group stratified by CXCR4 expression

   To further explore the potential link between CXCR4 signaling and the
   SVZ-driven recurrence phenotype in GBM, we closely examined the pattern
   of tumor recurrence and conducted RNA-seq analysis on 55 recurrent
   tissue specimens. Given that CXCR4 is a key regulator of the migration
   and repopulation of mutation-harboring SVZ cells in the RC, we
   hypothesized that CXCR4 expression might be particularly elevated in
   cases of local recurrence (LR) within the RC following total removal of
   primary tumors via GTR (Fig. [132]4e). Patients who experienced tumor
   recurrence at the RC after complete resection of primary tumor were
   designated as the LR group, with this subgroup used to investigate
   tumors potentially driven by SVZ NSCs. Conversely, other types of
   recurrence, including those from the residual primary tumor after
   subtotal resection, were classified into the non-LR group.
   Comprehensive transcriptome analysis revealed that CXCR4
   pathway-related genes were highly expressed in the LR group
   (Fig. [133]4f-h). CXCR4 expression was significantly elevated in the LR
   group, both in RNA-seq data and immunostaining of recurrent tumor
   tissues (Fig. [134]4i). Notably, we observed a significant correlation
   between high CXCR4 expression levels and poorer overall survival (OS)
   outcomes in LR patients (P = 0.018) (Fig. [135]4j), thus indicating the
   prognostic relevance of CXCR4 in distinguishing distinct recurrence
   patterns and their associated clinical trajectories. This finding
   underscores the potential utility of CXCR4 as a prognostic biomarker
   for stratifying GBM patients based on their recurrence risk as well as
   for guiding personalized treatment strategies.

Therapeutic potential of targeting the CXCR4/CXCL12 pathway

   Following complete resection of GBM, the CXCR4/CXCL12 pathway has
   emerged as a promising therapeutic target to improve patient outcomes,
   particularly in preventing SVZ-driven recurrence [[136]29]. Although
   radiotherapy and temozolomide have shown efficacy in treating GBM, the
   high rate of local recurrence suggests a need for additional novel
   therapeutic approaches. To investigate the effects of pharmacological
   intervention targeting the CXCR4/CXCL12 chemokine axis, we treated our
   modified mouse model for GBM recurrence with the CXCR4 blocker
   Plerixafor (AMD3100). Following cortical resection, AMD3100 was
   administered twice at a dose of 1.25 mg/kg via intraperitoneal
   injection. Remarkably, the treated group did not exhibit tumor
   formation within the RC at POD28 (Fig. [137]5a). Although AMD3100
   treatment did not completely inhibit the migration of SVZ cells, it
   significantly reduced the number of mutation-harboring SVZ cells
   expressing CXCR4 and decreased the differentiation of these cells into
   Olig2 + OPCs (Fig. [138]5b-d). Notably, blockade of the CXCR4/CXCL12
   axis led to reduced incidence of tumor recurrence and prolonged
   survival rates in both GBM recurrence and GTR models (Fig. [139]5e-h).

Fig. 5.

   [140]Fig. 5
   [141]Open in a new tab

   Treatment via CXCL12/CXCR4 blockade decreased the number of immigrating
   OPC lineage cells and improved survival in mice. a, Representative
   images of local reconstructed tumor development treated with AMD3100, a
   CXCR4 inhibitor, while the control group was treated with PBS on POD28
   following surgical resection. Scale bars, 500 μm. b, Representative
   images of GFP + cells around RC after their migration from the SVZ at
   POD14 of surgical resection. The white arrowheads represent CXCR4
   colocalizing with GFP + cells. Scale bars, 50 μm. c, Quantification of
   the number of CXCR4 + cells in immigrating GFP + cells around the RC at
   POD14 of surgical resection in AMD3100 treatment versus PBS group
   (n = 3 per group). Statistical analysis was performed using a two-sided
   Student’s t-test (*P < 0.05). Data are presented as mean ± SEM. d,
   Quantification of the number of CXCR4 + cells that showed
   double-staining with GFAP, OLIG2, and NeuN in AMD3100 treatment, versus
   PBS group at POD14 after surgical resection. e, Kaplan-Meier survival
   analysis of mice (n = 13 per group) in AMD3100 treatment, versus PBS
   group; P = 0.0356 (log-rank test). f, Cumulative incidence of tumor in
   mice (n = 13 per group) in AMD3100 treatment, versus PBS group;
   P = 0.0256 (log-rank test). g, Kaplan-Meier survival analysis of mice
   (n = 11 per group) without surgical resection (primary tumor), and of
   mice after surgical resection (recurrent tumor); P = 0.0155 (log-rank
   test). h, Kaplan-Meier survival analysis of mice (n = 7 per group) in
   AMD3100 treatment group after transplanted primary tumor removal versus
   PBS group; P = 0.0281 (log-rank test). i, Graphic illustration showing
   the mechanism by which CXCL12/CXCR4 axis-mediated tumor initiate NSCs
   in SVZ immigrating to RC in glioblastoma recurrence. Created with
   Biorender.com

   To further elucidate the impact of CXCR4 blockade on NSC
   differentiation, we generated neurospheres from the SVZ of LSL-EGFRviii
   f/+; LSL-Cas9-GFP f/+ mice in response to the introduction of
   cancer-driving mutations (Supplementary Fig. [142]8a). Cells were
   treated with AMD3100 for 7 days, followed by immunostaining and an
   analysis of lineage differentiation (Supplementary Fig. [143]8b).
   Interestingly, AMD3100 treatment inhibited OPC differentiation,
   particularly favoring differentiation toward astrocytes that are
   positive for glial fibrillary acidic protein (GFAP) (Supplementary
   Fig. [144]8c, d). This observation suggests that CXCR4 blockade may
   influence the differentiation capacity of mutation-harboring NSCs,
   which is a critical step in tumor reconstruction alongside migration
   (Fig. [145]5i). Altogether, our findings provide valuable insights into
   the molecular mechanisms underlying tumor progression while also
   highlighting the therapeutic potential of targeting the CXCR4/CXCL12
   axis in preventing local recurrence of GBM.

Discussion

   GBM is a primary brain tumor that continues to have a poor prognosis
   despite the range of treatment options available, including surgical
   resection, radiation therapy, and chemotherapy [[146]1, [147]4,
   [148]30]. Previous research has established that cancer-driving
   mutations in SVZ NSCs can be the cellular origin of GBM [[149]12,
   [150]14, [151]15]. Our findings provide novel evidence suggesting that
   GBM recurrence may result from the repetitive and cyclical evolution of
   the remaining tumor-initiating NSCs located in the distant SVZ after
   primary tumor resection. The persistence of mutation-harboring NSCs
   represents a potential therapeutic target for preventing recurrence of
   this challenging disease.

   Several longitudinal studies have utilized paired primary and recurrent
   tumor specimens from GBM patients. In terms of recurrence,
   approximately 40–60% of recurrences showed divergence of somatic
   mutations from primary tumors [[152]26–[153]28, [154]31]. The linear
   evolution model fails to explain this genetic divergence, supporting a
   branched evolutionary trajectory in which early progenitor
   clones—present even before diagnosis—drive recurrence independently of
   the dominant primary clones. We hypothesized that this branched
   evolution trajectory might be associated with the involvement of
   distant, mutation-harboring SVZ cells in tumor recurrence
   [[155]26–[156]28, [157]31]. To explore this possibility, we analyzed 10
   patients with four matched samples: blood, tumor-free SVZ, primary, and
   recurrent tumors. In 60% of cases, recurrent tumors shared a somatic
   mutation with both tumor-free SVZ and primary tumors, while
   primary-private mutations were not retained in the recurrent tumors.
   Instead, the recurrent tumor acquired additional oncogenic mutations
   independent of the primary tumors. These findings suggest that the
   mutation-harboring SVZ may serve as a potential cellular reservoir of
   the early progenitors that evolve into recurrent tumors, distinct from
   the primary dominant clones, alongside residual tumor cells. However,
   the inherent heterogeneity of GBM and the possibility of sampling bias
   must be carefully considered especially whether other early progenitors
   may drive the recurrence. The spatial and genetic diversity of GBM
   could result in either the underrepresentation or overrepresentation of
   certain subclonal populations within tumor samples, contributing to
   variability in observed clonal evolution patterns. To address these
   challenges, future studies should incorporate multi-sectional sampling
   strategies to better account for tumor heterogeneity and further
   clarify the genetic linkage between the SVZ and recurrent tumors.

   To elucidate the underlying mechanisms of the tumor-initiating NSCs in
   the SVZ responding to surgical resection, we collected time-course
   samples from the RC area in a genome-editing mouse model and performed
   RNA sequencing. Notably, cytokine levels began increasing as early as
   POD2 and remained elevated from them, which is consistent with previous
   reports indicating that chemokines play a crucial role in tumor
   progression and metastasis [[158]32, [159]33]. Among these genes, CXCR4
   was found to be the most upregulated chemokine-associated gene during
   recurrence CXCR4 is known to drive the migration of NSCs toward injury
   sites after traumatic brain injury [[160]19, [161]34]. Our study
   suggests that increased CXCR4 expression in mutation-harboring NSCs
   from the SVZ contributes to GBM recurrence. In addition to previous
   reports on CXCR4 expression associated with GBM [[162]35–[163]37], our
   results highlight that CXCR4-positive SVZ NSCs should be examined as a
   potential target for therapeutic intervention, particularly in the
   context of local recurrence after surgical resection.

   Our findings also indicate that CXCR4 is involved in the
   differentiation of mutated NSCs toward the OPC lineage. This is
   consistent with previous observations showing that CXCR4 promotes the
   differentiation of OPCs into oligodendrocytes during remyelination in
   injured adult CNS [[164]38]. The OPC lineage was found to be the
   dominant cell type involved in glioma reconstruction at the RC in our
   mouse model, suggesting that mutated NSCs migrate specifically to the
   RC through OPC lineage. Previous studies have reported that mutated
   NSCs in the SVZ contribute to glioma through aberrant growth of the OPC
   lineage [[165]32, [166]34, [167]39]. Thus, high CXCR4 expression is
   associated with NSC differentiation into OPCs and migration to the RC,
   ultimately leading to glioma reconstruction at this site. Moreover,
   CXCL12/CXCR4 blockade by AMD3100 reduced OPC lineage cell markers in
   vitro, supporting the idea that increased CXCR4 expression promotes NSC
   migration and differentiation. CXCL12/CXCR4 inhibition also reduced
   tumor development and mortality rates in mice with mutated residual
   NSCs in the SVZ. Notably, high CXCR4 levels were found in local
   recurrent tumors from GBM patients after GTR, underscoring that
   CXCL12/CXCR4 is a promising target for GBM treatment.

   Strong and specific CXCR4 expression was observed in recurrent tumor
   cells (Supplementary Fig. [168]9), suggesting its critical role in
   local recurrence. Nevertheless, immune cell-derived CXCR4 expression
   may contribute to the high expression of CXCR4 observed in the bulk RNA
   sequencing of the locally recurrent tumor samples [[169]25, [170]43,
   [171]44]. To elucidate the complex interplay between tumor cell-derived
   and immune cell-derived CXCR4 in local recurrence originating from SVZ
   NSCs, further investigations are warranted. Another important
   consideration is the interaction of CXCL12 with its alternative
   receptor, ACKR3, which is known to be signaling-defective and functions
   primarily as a scavenger receptor [[172]45, [173]46]. In addition to
   the role of CXCL12/CXCR4 signaling in tumor recurrence, ACKR3 may
   modulate CXCL12 availability or signaling dynamics within the tumor
   microenvironment, influencing recurrence pathways indirectly. Further
   investigation into this scavenger receptor and its potential impact on
   CXCL12-mediated mechanisms may provide additional insights into the
   complexity of tumor relapse processes.

   Our patient sequencing data could also support an alternative scenario,
   where an early GBM clone migrates to the SVZ, while low-level somatic
   mutations associated with various brain disorders have been
   consistently observed [[174]40–[175]42]. Although histological
   examination and deep sequencing were employed to detect low-frequency
   shared and private clones, these methods may not completely rule out
   the possibility of micro-invasion by early GBM clones, as indicated by
   our human genetic data analysis. Previous studies have reported SVZ can
   serve as a hidden reservoir, allowing primary tumor cells to migrate to
   the SVZ during tumor progression, particularly in patient-derived
   xenograft (PDX) models [[176]43–[177]45]. In addition to the
   mutation-carrying NSCs implicated in recurrence in this study, the
   subclone of primary tumor initially invading SVZ might contribute to
   local recurrence through bidirectional migration between SVZ and
   primary sites. While this hypothesis also remains plausible, it
   ultimately underscores the need for further studies to elucidate the
   role of the SVZ as a distant source and to explore strategies for
   suppressing recurrence.

Conclusion

   In summary, our study provides evidence showing that SVZ NSCs may serve
   as a viable target for preventing GBM recurrence following surgical
   resection, which could facilitate the development of novel treatment
   strategies for glioma patients. We emphasize the importance of
   understanding the residual origins of solid cancers after primary
   treatments and their interrelation with these treatments. Further
   research is needed to develop new therapies that effectively target
   these residual origins of cancer.

Materials and methods

Human biospecimens

   This study was approved by the Institutional Review Board of Seoul
   National University Hospital (IRB#H-1907-103-1048 and #H-0507-509-153)
   and Yonsei University Severance Hospital (IRB# 4-2021-1319). Recurrence
   samples were prospectively collected between 2013 and 2022 with at
   least three matched specimens obtained from each patient, including:
   (i) pathologically normal SVZ tissue, (ii) primary tumor tissue after
   gross total resection, and (iii) blood. Sample collection was conducted
   following our previously described protocol [[178]12]. H&E-stained
   slides were carefully reviewed by a neuropathologist to assess the
   presence or absence of tumor infiltration into the SVZ. Patients with
   evidence of suspicious SVZ infiltration were excluded following
   mutational analysis (Supplementary Fig. [179]1h, i). From this initial
   cohort, 10 patients who experienced tumor recurrence were selected for
   further analysis through deep sequencing of their recurrent tumor
   samples. Each selected patient exhibited features of clonal evolution
   at the time of their primary tumors, as described in the Results
   section. Recurrence samples were not collected from patients who either
   did not experience recurrence during follow-up or were ineligible for
   re-surgery due to the salvage nature of their treatment.

   For RNA sequencing, we collected recurrent GBM tissues from 55 patients
   with confirmed cases of tumor recurrence. Among these, 44 samples were
   obtained from Seoul National University Hospital, and 11 samples were
   collected from Yonsei University Severance Hospital. All samples were
   collected with the informed consent from the participants. To ensure
   the accuracy and validity of our findings, the extent of tumor
   resection was assessed using pre-operative and post-operative magnetic
   resonance imaging (MRI) scans. Survival times were calculated from the
   date of the initial surgery to evaluate the effectiveness of the
   surgical interventions and their impact on patient outcome.

Longitudinal analysis of four matched samples

   Genomic DNA was extracted from fresh frozen samples using the QIAamp
   DNA Mini Kit (Qiagen, #51304). For peripheral blood samples, DNA
   extraction was performed using the QIAamp DNA Blood Mini Kit (Qiagen,
   #51104). DNA yield was quantified using the Qubit dsDNA Quantification
   Assay Kit (Invitrogen, #Q332391) on a Quant 4 Fluorometer (Invitrogen).
   A custom NGS panel was designed to include 400 amplicons ranging from
   125 to 175 bp, which were distributed across two primer pools. This
   panel covered 100.08 kb and targeted a 200 bp region of the TERT
   promoter, as along with coding and untranslated regions of 14 cancer
   driver genes that are frequently mutated in GBM [[180]46, [181]47]. The
   targeted gene hybrid sequencing was designed and manufactured by
   Celemics Inc. (Seoul, Korea). The libraries were prepared according to
   the manufacturers’ protocol, and the final libraries were sequenced on
   a MiSeq Dx sequencer (Illumina, CA) by Sovargen (Daejeon, Korea).

   BAM files were generated from FASTQ files following the GATK Best
   Practices workflow established by the Broad Institute. Sequenced reads
   were mapped to the human reference genome (hg19) using the BWA-MEM
   algorithm. Duplicated reads were removed using Picard, and base quality
   score recalibration was applied to correct for systematic errors in the
   base quality score. Somatic base substitutions and short indels were
   identified using Mutect2, with matched blood samples used as controls;
   only variants that passed the “PASS” filter were included in our final
   analysis. All single nucleotide variations (SNVs) were annotated using
   the Ensemble Variant Effect toolset. To ensure accuracy and exclude
   false positive events, all mutations were visually inspected using the
   Integrative Genomics Viewer (IGV). Moreover, LICHeE
   ([182]https://github.com/viq854/lichee) was used to reconstruct
   longitudinal-sample cell lineage trees and infer the sub-clonal
   composition of the samples based on the variant allele frequencies of
   somatic single nucleotide variants.

Whole genome sequencing and targeted amplicon sequencing

   DNA libraries for whole-genome sequencing (WGS) were prepared from 1 µg
   of genomic DNA using the Truseq DNA PCR-Free Library Prep Kit
   (Illumina). Sequencing was carried out on the Illumina NovaSeq 6000
   platform, and average coverages of 60× for tumors and 30× for matched
   blood tissues were ultimately achieved. Reads were aligned to the human
   reference genome (GRCh37) using the BWA-MEM algorithm. Duplicated reads
   were filtered out using Picard, while base substitution and short
   indels were identified with MuTect2 and Strelka2. Additionally, segment
   copy-number profiles for the whole-genome sequenced samples were
   estimated using the Sequenza [[183]48] algorithms, with matched blood
   samples used as controls. To ensure accuracy and exclude false positive
   events, all mutations were visually inspected using the Integrative
   Genomics Viewer (IGV).

   For validation, deep targeted amplicon sequencing was performed using
   four matched specimens for each patient. Primers were designed using
   the Primer3 software ([184]https://primer3.ut.ee/), and the primer
   sequences are listed in Supplementary Table [185]1. The target region
   was amplified by PCR and high fidelity PrimeSTAR GXL DNA polymerase
   (Takara, Shiga, Japan). The MiSeq Reagent Micro Kit v2 (300 cycles)
   were used for library preparation and sequenced on a MiSeq Dx sequencer
   (Illumina, CA) by Sovargen (Daejeon, Korea).

Cre-expressing constructs to model p53, Pten mutations

   AAV: ITR-U6-sgRNA(backbone)-pCBh-Cre-WPRE-hGHpA-ITR was provided as a
   gift from Feng Zhang (Addgene plasmid #60229;
   [186]http://n2t.net/addgene:60229; RRID: Addgene_60229).
   pAAV-U6sgp53-U6sgbbSapI-GFAPCre was provided as a gift from Sidi Chen
   (Addgene plasmid #100275; [187]http://n2t.net/addgene:100275; RRID:
   Addgene_100275). gRNAs targeting p53 (sgP53) and Pten (sgPTEN) were
   designed as previously described^[12]. The sequences for the targeting
   sgRNA are as follows: sgP53 5’-TAATAGCTCCTGCATGG-3’, sgPTEN
   5’-GGTCAAGATCTTCACAGA-3’. Oligonucleotides containing these sgRNA
   sequences were synthesized by Cosmogenetech and annealed in vitro using
   a thermocycler.

   To create a single vector containing sgRNAs targeting p53, Pten, and
   Cre recombinase under the control of the GFAP promoter, we first
   amplified the GFAP promoter from pAAV-U6sgp53-U6sgbbSapI-GFAPCre
   plasmid, after which replaced the pCBh promoter in the
   pU6-sgRNA(backbone)-pCBh-Cre-WPRE-hGHpA plasmid with the GFAP promoter.
   Next, pU6-sgP53 and pU6-sgPTEN were amplified, and the
   pU6-sgRNA(backbone)_pGFAP-Cre-WPRE-hGHpA was modified to include both
   sgP53 and sgPTEN, ultimately resulting in the
   pU6-sgP53-pU6-sgPTEN_pGFAP-Cre-WPRE-hGHpA plasmid (referred to as
   sgTP-GFAP-cre).We also generated the sgLacz-pGFAP-Cre plasmid by
   inserting the U6-sgRNA(backbone)_pGFAP-Cre-WPRE-hGHpA sequence.

Mouse care and information

   All experiments were approved by the Institutional Animal Care and Use
   Committee in Seoul National University Hospital (SNUH-IACUC). The
   animals were housed in a facility accredited by AAALAC International
   (#001169) and were maintained in accordance with the Guide for the Care
   and Use of Laboratory Animals 8th edition, NRC (2010). The LSL-tdTomato
   (007914), LSL-Cas9 (25263330), and LSL-EGFRviii (19196966) mice
   [[188]49] were purchased from the Jackson Laboratory and maintained on
   a C57BL/6 strain and FVB strain background. The mice were housed in
   isolator cages where they had free access to food and water in a quiet,
   specific-pathogen-free room that was maintained at a controlled
   temperature of 23 °C on a 12-hour light-dark cycle. The veterinarians
   and the researchers carried out routine health assessments.
   Disease-specific survival endpoints were defined as either the natural
   death of the mice or the meeting of euthanasia criteria as specified
   under the IACUC protocol.

In vivo electroporation

   Postnatal day 0 (P0) to P2 mice pups were anesthetized on ice for over
   5 min. The injection site was defined as the midpoint of an imaginary
   line connecting the lambda and the upper-left corner of the eye. A
   mouth-controlled microinjection capillary was assembled for use, and
   the capillary was slowly lowered 4 mm down into the right lateral
   ventricle. Plasmid solution supplemented with 1% Fast Green dye was
   injected, and the capillary was retracted 5 s after the completion of
   the injection. Following successful injection, 5-mm tweezer electrodes
   (45–0489, BTX-Harvard apparatus) were positioned on each side of the
   head, and five electrical pulses (100 V, 50 ms duration, 950 ms
   intervals) were delivered using an ECM830 electroporator (BTX-Harvard
   apparatus). After electroporation, the pups were placed on a 37 °C
   heating pad until they began to move, at which point they were returned
   to their mother.

Surgical resection

   To perform surgical resection in four-week-old mice, anesthesia was
   induced through intraperitoneal injection of a ketamine-xylazine
   mixture. The mice were then securely positioned in a stereotaxic frame
   (Stoelting) for precise localization of the surgical site. A
   longitudinal skin incision was made, and a hand-held drill was used to
   create a 2 mm diameter craniotomy centered at 2 mm posterior to the
   bregma and 2 mm lateral to the midline. Cortical injury was induced
   using a flat pipette tip attached to a vacuum pump (VACUSIP, INTEGRA
   Biosciences), while care was taken to preserve the integrity of the
   corpus callosum. Following the resection, the mice were placed on a
   heating plate set to 37 °C until they fully recovered from anesthesia.
   The animals were then returned to their cages for post-operative care
   and monitoring.

Stereotaxic injection of AAV5 into cortex after surgical resection

   To construct adeno-associated virus (AAV) targeting astrocytes
   [[189]50], we co-transfected HEK293T cells with the pAAV5 capsid
   plasmid, a virus assembly helper plasmid (pAd deltaF6, obtained from
   the UPENN Vector Core), and target plasmids using polyethylenimine
   (1 mg/ml). Before transfection, the fetal bovine serum (FBS;
   Gibco)-containing DMEM (Gibco) was replaced with serum-free media. Six
   hours after transfection, the media was replaced again with
   FBS-containing DMEM, and the transfected cells were incubated for 72 h.
   The cells were then harvested, resuspended in a solution of 50% fresh
   DMEM and 0.04% DNase I (Worthington) in nuclease-free water, and lysed
   through three freeze-thaw cycles. Cellular debris was separated from
   the AAV-containing supernatant by centrifugation, at which point the
   supernatant was collected. AAV particles were then purified using
   polyethylene glycol-mediated precipitation^[39, 50].

   For stereotaxic injection, four-week-old mice were anaesthetized
   through intraperitoneal injection of a mixture with ketamine and
   xylazine, after which they were positioned in a stereotaxic frame
   (Stoelting). Hamilton syringes were inserted through the craniotomy
   using the stereotaxic apparatus and positioned unilaterally at the
   brain’s surface. The injection coordinates were 0.8 mm caudal, -1 mm
   lateral, and − 1 mm ventral to the bregma. In total, 100 nL of virus
   solution was injected at a flow rate of 25 nL/min using a syringe pump
   (KD Scientific Inc.). Following injection, the syringe was retracted
   slowly after a 3-minute delay. Virus injections were performed at the
   RC margin either immediately after surgical resection or 2 weeks prior
   to surgery. After the procedure, mice were placed on a 37 °C heating
   pad until they fully recovered and then returned to their cages.

In vivo treatment with AMD3100

   CXCL12/CXCR4 chemokine axis was pharmaceutically inhibited using
   Plerixafor (AMD3100, Sigma, A5602), a specific CXCR4 blocker. Following
   surgical resection, mice received intraperitoneal injections of AMD3100
   (1.25 mg/kg) or PBS twice daily, with a 6-hour interval between
   injections. After the final injection, the mice were sacrificed, and
   their brains were harvested for further anaysis.

Sphere culture and in vitro analysis of differentiation

   GBM tumorspheres were generated from tumors induced in LSL-EGFRviii
   f/+; LSL-tdTomato f/+ mice using the
   pU6-sgP53-pU6-sgPTEN_CBh-Cas9-P2A-Cre plasmid. These tumorspheres were
   cultured in Neural Stem Cell media, consisting of DMEM/F12 (Corning,
   0-090-CV) supplemented with B27 (50X, Gibco, 17504044),
   Penicillin-Streptomycin(Gibco, 15140122), and recombinant human EGF
   (Novoprotein, C046) and bFGF (Novoprotein, C029), each used at a final
   concentration of 20 ng/ml. Tumorspheres were dissociated with Accutase
   (Gibco, A1110501), plated at a density of 2 × 10^5 cells/ml, then
   passaged every 4–5 days.

   SVZ spheres (neurospheres) were generated from mutated SVZ induced in
   LSL-EGFRviii f/+; LSL-Cas9-GFP f/+ mice using the
   pU6-sgP53-pU6-sgPTEN_pGFAP-Cre-WPRE-hGHpA plasmid and cultured in
   Neural Stem Cell media. They were maintained under the same conditions
   as those described for the GBM tumorspheres.

   For neurosphere differentiation analysis, cells were dissociated with
   Accutase and plated with a density of 40,000 cells per dish, where each
   dish had been pre-coated with 20 µg/mL Poly-L-ornithine and 10 µg/mL
   Laminin (Sigma, L2020, and P4957) in Neural Stem Cell media with a
   reduced concentration of Fetal Bovine Serum (FBS, 1%). AMD3100 (25 µM)
   or PBS was added to the cells every 72 h. After 7 days of treatment,
   the cells were immunostained and processed for lineage differentiation
   analysis.

Orthotopic implantation of GBM tumorshphere cells and surgical removal of the
primary tumor

   The tdTomato + tumorsphere cells derived from LSL-EGFRviii f/+;
   LSL-tdTomato f/+ mice were in the culture for three to four days, and
   the cells were collected and dissociated into a single cell suspension.
   The cells were resuspended in a small volume of DMEM/F12 to a
   concentration of 10^5 cells/µl, and 1 µl of this suspension was used
   for stereotaxic injection. The tdTomato + tumor cells were
   orthotopically implanted into three-week-old LSL-EGFRviii f/+;
   LSL-Cas9-GFP f/+ mice that had been previously injected with
   sgTrp53-sgPTEN-GFAP-cre plasmid and electroporated at P0-P2. One week
   after implantation, bulky primary tumors developed and were surgically
   resected, as described previously. The mice were monitored daily for
   signs of tumor burden, and moribund mice were sacrificed. The brains
   were then processed for histological analysis.

Image analysis of mouse brain

   Mice under each condition were sacrificed and perfused with
   phosphate-buffered saline followed by 4% paraformaldehyde (PFA). The
   brains were then harvested, fixed in 4% PFA, and cryoprotected
   overnight in 30% sucrose. At this point, the tissues were embedded in
   OCT compound (3801480, Leica Biosystems) and stored as frozen blocks at
   − 80 °C. Cryostat sections were cut at a thickness of 25 μm and mounted
   on glass slides. For H&E staining, sections were cut at a thickness of
   4 μm. Immunohistochemistry was performed using the following
   antibodies: Mouse antibody to Nestin (1:200 dilution; MAB5326, Merck
   Millipore), rabbit antibody to GFAP (1:500 dilution; Z0334, DAKO),
   rabbit antibody to oligodendrocyte transcription factor 2 (OLIG2; 1:500
   dilution; AB9610, Merck Millipore), rat antibody to platelet-derived
   growth factor receptor α (PDGFRα; 1:200 dilution; 14-4321,
   eBioscience), rabbit antibody to S100β (1:500 dilution; ab52642,
   Abcam), rat antibody to myelin basic protein (MBP; 1:500 dilution;
   MAB386, Merck Millipore), rabbit antibody to Ki67 (1:500 dilution;
   ab15580, Abcam), rabbit antibody to NeuN (1:500 dilution; ab104225,
   Abcam), mouse antibody to C-X-C chemokine receptor type 4 (CXCR4; 1:500
   dilution; sc-53534, Santa Cruz Biotechnology), and rabbit polyclonal
   antibody to CXCL12 (1:500 dilution; ab25117, Abcam), Rabbit antibody to
   CD31 (1:500 dilution; ab222783, Abcam). Nuclear staining was performed
   using DAPI that had been included in the mounting solution
   ([190]P36931, ThermoFisher). Images were captured using a Zeiss LSM800
   confocal microscope with Z-stacks set at a step size of 1.5 μm.
   Fluorescence intensities, reflecting the distribution of fluorescent
   reporter-positive cells were quantified by converting them into grey
   values using ImageJ software ([191]http://rsbweb.nih.gov/ij/).

Bulk RNA sequencing

   Frozen brain tissue samples from patients with primary and recurrent
   GBM were dissected, and the tissue samples were rapidly flash-frozen in
   liquid nitrogen and stored at -80℃ to preserve their integrity. RNA was
   extracted from the tissue samples using the QIAGEN RNeasy Micro Kit.
   For sequencing and analysis, cDNA synthesis, library preparation, and
   sequencing were conducted by technicians at Macrogen. Sequencing was
   performed using the Illumina Novaseq 6000, and it ultimately yielded
   80 million paired-end reads per sample (40 million in each direction).
   Reads were aligned to human (UCSC hg38) and mouse (mm10) reference
   genomes using HISAT2 (v.2.2.0) [[192]51], with annotations provided by
   the Gencode GTF file.

   Gene expression quantification was performed using featureCounts from
   Subread (v1.6.4) [[193]39]. The resulting cell count files were
   combined into a matrix using a custom Python script (Python 3.6), while
   raw read counts were filtered to only include genes with at least ten
   counts across all samples. DESeq2 (v.1.22.2) [[194]52] was used to
   compute the size factors for each sample and perform variance
   stabilizing transformation. Batch correction was applied to account for
   both technical and biological variables. The variance
   stabilizing-transformed RNA-seq data served as input for clustering
   analysis. Differential gene expression was assessed using DESeq2, with
   batch status incorporated as a covariate in this analysis. Upregulated
   genes were identified based on a Log2Fold Change (FC) greater than 1
   and an FDR less than 0.01.

   Gene set enrichment analysis (GSEA) (v.3.0) [[195]53] was performed on
   genes ranked by differential expression, with pathways obtained from
   the Molecular Signatures Database (MSigDB). Pathways were filtered to
   include those with a minimum of 15 genes and a maximum of 500 genes.
   The R package “clusterProfiler” (v3.10.1) [[196]54] was also utilized
   for GSEA of the Gene Ontology (GO) terms, including biological
   processes (BP), cellular components (CC), and molecular functions (MF)
   of differentially expressed genes (DEGs). The five most significant
   biological processes, with an FDR less than 0.05 and a log2FC less than
   or equal to 1, were reported. KEGG pathway enrichment analysis of DEGs
   was also conducted, and the results were visualized in the form of bar
   plots showing the top 20 pathways.

Single cell RNA sequencing

   Single-cell RNA sequencing data from murine traumatic brain injury were
   obtained from the GEO database ([197]GSE160763). Clustering and cell
   type annotation were performed while following previously established
   protocols [[198]55]. Differential expression analysis for each cell
   type was conducted using the Wilcoxon rank-sum test, as implemented in
   the “FindMarkers” function of the Seurat package (v.4.0.4). The
   expression levels of Cxcl12 were visualized using tSNE plots to
   illustrate cellular distribution.

Deconvolution

   Cell fractions from the bulk RNA-seq data for mouse GTR model were
   enumerated using the CIBERSORTx online platform
   ([199]https://cibersortx.stanford.edu/). We constructed the reference
   dataset [[200]55] based on the expression profiles of 2000 highly
   variable genes in attempt to minimize noise stemming from low-quality
   genes. Using the single-cell reference file, we followed the
   instructions for CIBERSORTx while using default parameters to analyze
   the cellular composition.

Statistical analyses

   Data are presented as a mean ± s.e.m. Results were analyzed with a
   t-test, Manny-Whitney U test, or Fisher’s exact test where appropriate
   using GraphPad Prism version 9.4.1 (Graph-Pad Software, Inc.). Tumor
   accumulated incidence and survival data for mice were analyzed using
   Kaplan–Meier analysis. All P values less than 0.05 were considered to
   be statistically significant. All experiments related to the use of
   animals or the source of cells were subjected to randomization. Sample
   sizes were predetermined based on the variability found in preliminary
   and similar experiments. Researchers were not blinded to allocation
   during experiments or outcome analysis.

Electronic supplementary material

   Below is the link to the electronic supplementary material.
   [201]Supplementary Material 1^ (37.3MB, docx)

Author contributions

   Joo Ho Lee, S.G.K., and Jeong Ho Lee designed the study. Joo Ho Lee,
   L.X., and H.J.K. coordinated the experiments and drafted the
   manuscript. Joo Ho Lee, L.X., H.J.K., Y.H.L., and C.H.N. analyzed DNA
   and RNA sequencing data. Joo Ho Lee, T.M.K., S.H.C., S.T.L., J.B.P.,
   J.H.Y., J.H.C., S.H.K., J.K.W., S.H.P., J.H.P., and Y.S.J. contributed
   to data interpretation. J.H.Y., S.H.K., J.K.W., and S.H.P. performed
   the analysis of human brain MRI data and provided technical expertise
   for histological studies in both human and mouse tissues. Joo Ho Lee,
   L.X., H.J.K., Jeong Ho Lee, J.H.P., and H.G.W. contributed to the
   development of the mouse model. J.H.Y., J.H.C., S.G.K., and C.K.P.
   provided human tissue samples. Joo Ho Lee, L.X., H.J.K., J.H.Y.,
   J.H.C., S.G.K., and C.K.P. were involved in data collection and
   curation. Joo Ho Lee, S.G.K., and Jeong Ho Lee supervised the project,
   critically revised the manuscript, and secured funding for the study.
   All authors reviewed and approved the final manuscript.

Funding

   We are grateful to the members in the Joo Ho Lee’s lab from Seoul
   National University Hospital for engaging in for discussions and
   providing technical support. This study was funded by National Research
   Foundation (NRF) grants from the Ministry of Science and ICT
   (Information and Communications Technology), Korea (Grant No.
   NRF-2021R1A2C1095168, NRF-2020M2D9A2092373, NRF-2018R1D1A1A02085487,
   and RS-2023-00218623 to Joo Ho Lee), 2019R1A3B206661923 to Jeong Ho
   Lee, NRF-2022R1A2B5B03001199 and RS-2024-00437820 to S.G.K). It was
   also supported by a grant from the Korea Health Technology R&D Project
   through the Korea Health Industry Development Institute (KHIDI), funded
   by the Ministry of Health & Welfare, Korea (Grant No. HI23C1573
   (RS-2023-[202]KH140077) to Joo Ho Lee; RS-2024-00438443 to S.G.K).

Data availability

   No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

   Human sample collection: all samplings were approved by Seoul National
   University Hospital (IRB#H-1907-103-1048 and #H-0507-509-153) and
   Yonsei University Severance Hospital (IRB# 4-2021-1319). Mouse
   experiments: All procedures were conducted in accordance with the
   guidelines of the Institutional Animal Care and Use Committee at Seoul
   National University Hospital (SNUH-IACUC).

Competing interests

   Jeong Ho Lee is a co-founder and chief scientific officer of Sovargen
   Inc.

Footnotes

   Publisher’s note

   Springer Nature remains neutral with regard to jurisdictional claims in
   published maps and institutional affiliations.

   Xue Li and Hyun Jung Kim contributed equally to this work.

Contributor Information

   Jeong Ho Lee, Email: jhlee4246@kaist.ac.kr.

   Seok-Gu Kang, Email: seokgu9@gmail.com.

   Joo Ho Lee, Email: jooholee@snu.ac.kr.

References