Glioblastoma immune landscape remodeling following treatment with CD70-targeted CAR-T cells Authors: Alexandros Kourtesakis, Dennis Alexander Agardy, Clara Tejido Dierssen, Ling Hai, Dirk Carsten Frieder Hoffmann, Normann Mussnig, Ralph Sinn, Manuel Fischer, Andy Schacht, Michael Breckwoldt, Lukas Bunse, Tim Sauer, Michael Platten, Wolfgang Wick, Tobias Kessler Published: 2026 ABSTRACT AbstractChimeric antigen receptor (CAR)-T cell therapy has shown limited efficacy in glioblastoma (GB), largely because of the immunosuppressive tumor microenvironment (TME). We have previously demonstrated that the ligand-based mCD70-directed construct mCAR27 specifically eliminates CD70⁺ GB cells in vitro, and induces durable tumor remission with prolonged survival in a syngeneic model of orthotopic GB. Despite the therapeutic promise of CD70 as a GB CAR-T cell target, the impact of CD70-directed CAR-T cells on the brain tumor immune landscape after target interception remains unexplored. Here, we independently confirmed a tumorigenic role for CD70 in murine GB by performing bulk RNA sequencing and in vivo experiments. Next, we applied single-cell RNA sequencing (scRNAseq) to decipher the transcriptome of GB-infiltrating immune cells in immunocompetent mice bearing CD70+ GB after intratumoral mCAR27 therapy. CAR-T cell treatment expanded adoptively transferred cells and was associated with increased endogenous Cd8⁺ T-cell infiltration, alongside a reduction in immunosuppressive populations, including Tregs and anti-inflammatory M2-like macrophages, but also subsets of dendritic and natural-killer cells. CD70 expression was detected on multiple GB-infiltrating immune subsets, notably Tregs and M2-like macrophages. Transcriptionally, mCAR27-mediated GB remission led to a coordinated downregulation of type-I interferon-stimulated genes and immunosuppressive mediators such as Arg1, Tgfbi, Cd274, Pdcd1lg2 and Lag3, coupled with upregulation of genes involved in inflammatory resolution, findings consistent with attenuation of GB-derived inflammatory stress and immunosuppression. Collectively, we showcase that CD70-directed CAR-T cell therapy remodels the post-treatment immune landscape towards resolution of inflammation and reduced immunosuppression, supporting sustained immune surveillance. Introduction While chimeric antigen receptor (CAR)-T cell therapy has revolutionized the treatment of several hematologic malignancies, its application in glioblastoma (GB) is hindered by significant challenges such as the notoriously immunosuppressive tumor microenvironment (TME). 1 Overcoming such obstacles is pivotal to achieving meaningful clinical benefit for GB patients treated with CAR-T cells. The development of single-cell methodologies has allowed researchers not only to analyze CAR-T cell products, 2 but also to comprehensively profile the TME, the molecular and functional dynamics of administered CAR-T cells, and their interplay with endogenous immunity. 3 , 4 Understanding these interactions represents a crucial step towards development of optimized novel immunotherapeutic approaches for GB treatment, a disease with a dismal prognosis for patients. 5 In a characteristic example, Alizadeh and colleagues performed single-cell RNA sequencing (scRNAseq) after treatment of a syngeneic GB mouse model with IL13Ra2-targeting CAR-T cells and demonstrated that CAR-T cell therapy promoted IFN-γ-driven activation of endogenous immune cells and induced immunological memory, while at the same time shifting the TME towards a less immunosuppressive state. 6 This study showcases how such pipelines can help decipher the immunological signatures following CAR-T cell therapy. CD70 is a member of the tumor necrosis factor receptor superfamily. Although not present in normal brain tissue, it is ectopically expressed in a fraction of GB patients, where it inflicts apoptosis of cytotoxic T-cells and is linked to increased prevalence of tumor-associated macrophages (TAM), contributing to the establishment of an immunosuppressive milieu. 7–0 Previously, we highlighted the specificity and potency with which a murine, ligand-based CAR-T cell construct (mCAR27) targets mCD70-expressing GB leading to durable tumor remission and remarkable survival prolongation in treated immunocompetent mice. 11 However, whether this sustained anti-tumor effect is the outcome of communication between therapeutic and endogenous immune cells, and whether CAR-T cells can reshape the immune landscape to promote long-lasting tumor surveillance remains elusive. To address these questions, we employed scRNAseq to analyze the effects of mCD70-directed CAR-T cell immunotherapy on endogenous immunity after intratumoral treatment. Administration of mCD70-targeting CAR-T cells led to a decrease in the incidence of immunosuppressive populations such as anti-inflammatory M2-like macrophages and Treg cells, and on the other hand increased frequency of therapeutic CAR-T cells and endogenous CD8a + cytotoxic T-cells. In several identified clusters, our cell therapy induced a transcriptional profile suggesting reduced immunosuppression, resolution of tumor-driven inflammation and restoration of immune homeostasis. Additionally, cell-to-cell communication analysis further revealed treatment-associated remodeling of predicted signaling networks involving endogenous CD8 T-cells, including altered communication from immunoregulatory populations. Finally, we demonstrate that Cd70 expression conferred oncogenic and tumorigenic properties in murine GB both in vitro and in a syngeneic mouse model, providing a rationale for our immunotherapeutic approach and its detailed characterization by scRNAseq. Conclusively, our study provides valuable insights into the functional and molecular effects of anti-CD70 CAR-T cell therapy on endogenous immunity and underscores the potential of single-cell methodologies to uncover the immune cell dynamics following CAR-T cell therapy in GB. Materials and methods CAR design The design of the mCAR27 construct has been described previously. 11 Briefly, the full length murine Cd27 sequence (RefSeq ID: NM_001033126.2) served as target recognition, transmembrane, and co-stimulatory domain, and was fused to the cytoplasmic domain of the murine Cd3ζ chain (RefSeq ID: NM_001113391.2). For detection purposes, the tdTomato fluorophore (FPbase ID: PGG5S) was introduced downstream, driven by a T2A self-cleavage site. All sequences were codon-optimized for expression in murine cells (Mus musculus) and were synthesized by VectorBuilder. Transduction of primary murine T-cells To produce ecotropic retrovirus, HEK293 cells were co-transfected with 4.75 μg of gag/pol/env-encoding plasmid pCL-Eco (Addgene #12371; http://n2t.net/addgene:12371; RRID:Addgene_12371, gift from Inder Verma) and 3 μg of tdTomato-encoding murine retroviral constructs pMSCV or mCAR27 using GeneJuice (#70967, Sigma-Aldrich). 11 SN was harvested 48 h post-transfection, filtered, and used to transduce activated murine T-cells. For murine T-cell isolation, 7–12 week-old female C57BL/6J mice (Janvier Labs, N = 5 animals) were sacrificed by cervical dislocation and the spleen and lymph nodes were aseptically obtained. They were meshed through a 70 μm strainer (#CNA0.1, Carl Roth) and washed twice with PBS. Meshed spleens were resuspended in ACK Lysing Buffer (#A1049201, Invitrogen) at RT for 1.5 min, washed with PBS, and meshed again. The spleen and LN preparation products were pooled, and murine T-cells were enriched using the MagniSort™ Mouse T cell Enrichment Kit (#8804-6820-74, Invitrogen) following the manufacturer's recommendations. Enriched murine T-cells were activated on 24-well plates, previously coated with 20 μg/mL anti-hamster IgG whole molecule (#55397, MP Biomedicals) at 37 °C for 4 h, followed by a PBS wash and coating with 4 μg/mL Ultra-LEAF™ Purified anti-mouse CD3 Antibody (#100340, Biolegend, RRID: AB_11149115) at 4 °C for 1 h. Activation took place in T-cell proliferation medium (TCPM), supplemented with 5 μg/mL Ultra-LEAF™ Purified anti-mouse CD28 Antibody (#102115, Biolegend, RRID: AB_11150408) and 100 IU/mL recombinant human IL-2 (#130-097-743, Miltenyi Biotec). TCPM comprised RPMI-1640 supplemented with 10% (v/v) FBS, 1% (v/v) Penicillin/Streptomycin, 2 mM L-Glutamine (#25030081, Gibco), 50 μM β-mercaptoethanol (#M6250, Sigma-Aldrich), 25 mM HEPES (#15630080, Gibco), 1 mM sodium pyruvate (#11360070, Gibco) and 0.1 mM MEM non-essential amino acids solution (#11140050, Gibco). Murine T-cell transduction took place 36 h after activation, as previously described. 11 Specifically, non-tissue culture-treated 24-well plates (#371147, Falcon) were coated O/N at 4 °C with a 7 μg/mL dilution of Retronectin (#T110A, Takara) in PBS. Plates were washed, received 1 mL of retroviral SN and were centrifuged at 2000 × g for 2 h at 32 °C. Next, 0.4 * 106 activated murine T-Cells were added per well. Spinoculation occurred at 500 × g for 5 min at 32 °C and transduced murine T-cells were expanded in TCPM supplemented with 100 IU/mL IL-2. Transduction efficiency was evaluated 96 h later by flow cytometry. Orthotopic implantation of cells For orthotopic implantation of GL261/CD70 cells, 7–12-week-old female C57BL/6J mice (Janvier Labs) were anesthetized with an intraperitoneal (i.p.) injection of a 100 mg/kg body-weight ketamine (#FS1670041, bela-pharm) and 20 mg/kg body weight xylazine (#PZN-01320422, Elanco) mixture, prepared in a 0.9% w/v NaCl solution. A total of 1 * 105 GL261/CD70 or GL261/MCS cells were loaded on a 10 μL Hamilton syringe (#80308, Hamilton) and stereotactically injected into the right hemisphere, 2 mm right lateral of the bregma and 1 mm anterior to the coronal suture, 3 mm below the dural surface, with an injection rate of 5 * 104 cells/min (N = 31 animals injected with GL261/CD70 cells, N = 12 animals injected with GL261/MCS cells). A total of N = 43 C57BL/6J animals were implanted with tumor cells, and in total N = 48 animals were used cumulatively in the present study. Tumor formation was verified two weeks post implantation by MRI. Animal assignment to different groups was performed to harmonize the tumor burden among the groups prior to treatment initiation. Investigators were blinded during the conduct of the experiment. One day later, mice were either treated with CAR-T cells or mock-transduced T-Cells (N = 8 animals per group) or were intracardially perfused (N = 24 animals) as described below. For CAR-T cell treatment purposes, effector cells were harvested, washed with PBS and resuspended in PBS at a concentration of 5 * 105 cells/μL. Intratumoral (ICT) administration of 1.5 * 106 CAR-T or mock-transduced non-targeting T-cells (N = 8 animals per group) took place as described above, at a rate of 5 * 105 cells/min. The therapeutic effect of administered CAR-T cells was assessed six days later by high field 9.4 Tesla MRI, as described previously. 11 , 12 In our study, a single animal served as a single experimental unit. The primary outcome measure of the study was to investigate the frequency of brain-infiltrating immune populations and their gene expression profile in treated mice. A secondary outcome was to evaluate the immunotherapeutic effect by MRI. No experimental units or individual data points were excluded from the analyses. Absence of detectable tumor growth by MRI two weeks after orthotopic glioblastoma cell implantation was defined as a priori exclusion criterion. Processing of brain tissue Seven days post-treatment or one day after tumor presence verification by MRI, animals were deeply anesthetized with a mixture comprising 120 mg/kg body-weight ketamine and 20 mg/kg body weight xylazine in a 0.9% (w/v) NaCl solution. After complete loss of the toe-reflex, animals were intracardially perfused with 20 mL of PBS and brains were extracted. The tumor-bearing hemispheres were minced on a sterile glass surface with single-use scalpels. Minced pieces were transferred into a 50 μg/mL Liberase (#5466202001, Sigma-Aldrich) solution diluted in Hank's Balanced Salt Solution (HBSS, #H9269, Sigma-Aldrich), briefly vortexed, and incubated at 37 °C for 30 min under rotation. Subsequently, they were washed with PBS and meshed through a 100 μm strainer (#130-098-463, Miltenyi Biotec), stacked on top of a 70 μm strainer (#130-110-916, Miltenyi Biotec). Samples were washed with PBS and spun at 500 × g at RT for 5 min. For myelin removal purposes, samples were resuspended in a 30% (v/v) Percoll (#17089101, Cytiva) solution in PBS and centrifuged at 314 × g at RT for 30 min without brake. The myelin layer was then removed to generate the tumor-bearing hemisphere single cell suspensions. MRI Animals were anesthetized with 2% (v/v) isoflurane and maintained under anesthesia at a concentration of 1%–1.5% (v/v). To ensure physiological stability, animals were placed on a heated pad (37 °C), and their respiratory pattern was continuously tracked using a sensor pad connected to a LabVIEW monitoring system (National Instruments Corporation). A 9.4 Tesla horizontal-bore small animal MR scanner (BioSpec 94/20 USR, Bruker BioSpin) featuring a four-channel phased-array surface coil was used for image acquisition. Tumor volume was determined using a three-dimensional T2-weighted rapid acquisition with refocused echoes (RARE) sequence. Acquisition parameters were as follows: echo time (TE) = 72.56 ms, repetition time (TR) = 1800 ms, flip angle = 90°, matrix dimensions = 200 × 100 × 120, one average, slice thickness = 100 μm, and isotropic resolution = 100 μm, resulting in a total scan duration of 3 min. After acquisition, images were stored in DICOM format and subsequently analyzed with Slicer software (version 4.11.20210226). Single-cell RNA sequencing Immune cell sorting from dissociated tissue Generated single-cell suspensions were washed with PBS and blocked with a 10 μg/mL a-mCD16/a-mCD32 mix (#14-0161-82, Thermo Fisher Scientific) in PBS for 10 min at RT. Samples were then split in two equal parts, each stained with an anti-mouse CD45-BV510 (#103138, Biolegend, RRID: AB_2563061) antibody, and for multiplexing purposes, a TotalSeq™-C anti-mouse Hashtag antibody (Biolegend) at 4 °C for 30 min in the dark. All hashing antibodies utilized are provided in Supplementary Table S1. Cells were then washed three times with 0.5% (w/v) BSA/2mM EDTA in PBS and finally resuspended in 0.04% (w/v) BSA in PBS. Sorting of CD45+ cells took place on a BD Aria Fusion instrument using a 100 μm nozzle and 4-way purity. For a given sample, the two equal parts stained with anti-CD45 and two different hash antibodies were pooled right before sorting. Cells were sorted in 0.04% (w/v) BSA in PBS. Library preparation and sequencing Gel beads-in-emulsion (GEM) generation and subsequent reverse transcription were performed using the 10× Genomics Chromium™ Next GEM Chip K Single Cell Kit (#1000287, 10× Genomics) and the Chromium™ Next GEM Single Cell 5′ Kit (#1000265, 10× Genomics) according to the manufacturer's recommendations. GEM composition is provided in Supplementary Table S2. Library preparation was performed according to the Chromium Next GEM Single Cell 5′ Reagent Kit v2 (Dual Index) with Feature Barcode technology for Cell Surface Protein & Immune Receptor Mapping (#CG000330 Rev F, 10× Genomics). DNA concentration was measured with the Qubit Broad Range dsDNA Quantitation kit (#Q32850, Thermo Fisher Scientific). Library fragment size and overall DNA quality were assessed with the D1000 ScreenTape Assay (#5067-5582, Agilent) loaded on a 4150 TapeStation System (Agilent). Generated libraries were sequenced on a NovaSeq6000 instrument (Illumina) using the NovaSeq 6000 S2 Reagent Kit v1.5 (300 cycles, #20028314, Illumina). Data analysis The Cell Ranger workflow (version 8.1.0, 10× Genomics) was used to align sequencing reads to the reference genome by executing the Cell Ranger mkref pipeline. The reference genome comprised the murine genome GRCm38 (release 2020-A) which was further appended to include the full length tdTomato DNA sequence, to allow identification of exogenously administered effector cells. Subsequently, the Cell Ranger multi command was employed according to the developer's recommendations to process gene expression and cell hashing, enabling integrated demultiplexing and analysis of multimodal libraries (https://www.10xgenomics.com/support/software/cell-ranger). Data analysis was performed in R (version 4.2.0) with Seurat (version 4.3.0). All illustrations were generated with the SCpubr package (version 2.0.2) and ggplot2 (version 3.5.1). The following criteria were set for a cell to pass QC: 300 < nCount < 20000, 200 < nFeature < 5000, mitochondrial gene content < 10%. Cells with a high content of ribosomal genes were also filtered out of the analysis. Samples from each GEM were then demultiplexed based on the presence of the barcode hashtag sequence using Seurat::HTODemux. Cells featuring two barcode sequences were excluded as doublets. Cells featuring no barcode sequences were excluded as negatives. The SCTransform (version 0.4.2, glmGamPoi) was used for data normalization. Afterwards, Seurat::FindVariableFeatures() was employed to identify N = 2000 variable features which were then used for data scaling through Seurat::ScaleData(), accounting for mitochondrial and ribosomal content. Linear dimensionality reduction was performed on the variable features with Seurat::RunPCA() to generate principal components (PC) and identify the main variation sources. Based on the generated elbow plot featuring all PCs (Seurat::ElbowPlot), and by assessing the variance gain with each additional vector, we manually selected the first 30 PCs for downstream clustering and batch correction. Batch effect correction across samples was performed by applying the Harmony algorithm (version 1.1.0) to eliminate bias from different samples and perform PC embedding-based integration. The RunHarmony() function was executed with the group.by.vars parameter assigned to sample information-featuring variable. To visualize biologically meaningful clusters of cells, an array of cluster resolutions ranging from 0.1 to 2.0 was used to generate Uniform Manifold Projection (UMAP) plots (Seurat::FindNeighbors(), Seurat::FindClusters(), and Seurat::RunUMAP() functions). For clustering analysis, resolution 1.1 was selected for downstream processing. Cluster annotation was performed manually based on known cell markers and by using the Seurat::FindAllMarkers function. Literature research was conducted, and the panglaoDB database was screened to validate the connection of identified markers to specific cell types. Differentially expressed genes between treatment groups were identified with Seurat::FindMarkers() function. Only genes which were expressed in at least 30% of the cells within an annotated cluster in either treatment arm (min.pct = 0.3) and passed multiple testing correction (p-value < 0.05) were considered for downstream analysis. Gene ontology (GO) analysis was performed with the ClusterProfiler package (version 4.4.4). The AnnotationDbi::select() command was executed to obtain ENSEMBL IDs for each gene and subsequently, the enrichGO() function mapped the IDs to mouse GO terms (OrgDb = “org.Mm.eg.db”), incorporating only those related to biological processes (ont = “BP”). Results were visualized with enrichplot (version 1.16.2). Cell-cell communication was inferred from the scRNAseq dataset using CellChat (version 1.6.1). 13 Separate CellChat objects were generated for the pMSCV and mCAR27 treatment groups using the createCellChat() function, with the annotated broad immune-cell clusters used as grouping identities. To enable direct comparison of the predicted communication networks between treatment conditions, the individual CellChat objects were combined using the mergeCellChat() function. Differences in pathway-level information flow between pMSCV- and mCAR27-treated mice were visualized using the rankNet() function, whereas differential ligand-receptor interactions between selected sender and receiver cell populations were visualized using netVisual_bubble(). Predicted pathway-level communication networks between immune-cell populations were additionally visualized as chord diagrams using netVisual_chord_gene(). RT-qPCR RNA from 106 GL261/MCS or GL261/CD70 murine GB cells was extracted with the RNeasy Mini Kit (#74104, Qiagen). Subsequently, the MultiScribe™ Reverse Transcriptase kit (#4311235, Invitrogen) was used to reverse transcribe 1 μg RNA. Gene expression levels were assessed on the QuantStudio-3 real time PCR system (Applied Biosystems) with the TaqMan™ Gene Expression Master Mix (#4369016, Applied Biosystems) and TaqMan probes targeting mouse Gapdh (assay ID: Mm99999915_g1, Applied Biosystems), mouse Sh3bgrl2 (assay ID: Mm00552919_m1, Applied Biosystems), mouse Cd200 (assay ID: Mm00487740_m1, Applied Biosystems), and mouse Prkg2 (assay ID: Mm00435938_m1, Applied Biosystems). Bulk RNA sequencing Library preparation and sequencing A total of 500 ng RNA from each sample (GL261/MCS or GL261/CD70, N = 4 biological replicates per condition) was processed with the Illumina Stranded mRNA Prep Ligation Kit (#20040532, Illumina) to generate libraries for sequencing. Indexing was achieved with the IDT for Illumina RNA UD Indexes Set A Ligation Kit (#20040553, Illumina). The size and quality of generated libraries was assessed with the D1000 ScreenTape Assay (#5067-5582, Agilent) loaded on a 4150 TapeStation System (Agilent) and the Qubit™ 1× dsDNA High Sensitivity Assay Kit (#Q33230, Thermo Fisher Scientific). Generated libraries were sequenced on a NextSeq550 instrument (Illumina) with the NextSeq 500/550 High-Output v2.5 75-cycle Kit (#20024906, Illumina). Data analysis The RNA-Seq data was aligned using STAR (version 2.5.3a) against the murine genome (GRCm39). Reads duplication was marked using Sambamba (version 0.6.5). A gene-count matrix was generated using Featurecounts (Subread version 1.6.5). The processing pipeline and parameter settings were based on the DKFZ ODCF workflow (https://github.com/DKFZ-ODCF/RNAseqWorkflow). Genes with total counts < 10 among all samples were removed. Differential expression analysis was performed between the CD70 and MCS control groups using DESeq() function with default settings in the DESeq2 (version 1.32.0). The results were shrunk by the “apeglm” method via the lfcShrink function. The volcano plot of results was generated using EnhancedVolcano (version 1.10.0). Genes with adjusted p-values < 0.05 and absolute fold-change ≥ 2 were retained as differentially expressed. The scaled vst-normalized expression levels of the DEGs were visualized in a heatmap using ComplexHeatmap (version 2.8.0). Flow cytometry For assessment of CD70 expression in brain tumor-infiltrating immune cells, single cell suspensions from tumor-bearing hemispheres were generated as described above. Cells underwent a PBS wash and were blocked as above. Subsequently, they were stained with antibodies in different combinations at 4 °C for 30 min in the dark. All flow cytometry antibodies used in this study are provided in Supplementary Table S3. After incubation, cells were washed twice with PBS and data acquisition was performed on a FACSymphony A5 SE (BD Biosciences) instrument. Data analysis was conducted with FlowJo (version 10.10.0, Treestar). Statistical analysis Data were analyzed/visualized with GraphPad Prism (version 10.1.2; GraphPad Software). Results are expressed as mean (SD) unless otherwise stated. All applied statistical tests are mentioned in the corresponding figure captions. Data normality was investigated. Illustrations were made using BioRender (BioRender Inc). Figures were assembled using Inkscape (version 1.3.2, Inkscape Developers). Animal housing and welfare All animals used in this study were maintained in a facility free of pathogens, at a constantly monitored temperature of 22 °C ± 2 °C, a humidity level of 55% ± 10%, and a 12-h light/dark cycle. Housing took place in individually ventilated cages, with the maximum number of animals in each cage being 5. An ad libitum standard diet according to the protocols of the German Cancer Research Center (DKFZ) was provided to the animals. To ensure ethical compliance, only the minimum number of animals necessary to achieve statistical significance was used, and the 3R principles were applied in every experiment to reduce animal distress. Meticulous animal monitoring occurred daily throughout the whole duration of the experiments, and animals were sacrificed upon reaching pre-determined neurological signs, or a body weight loss of 20%, criteria not met by any animal in the present study. All animal procedures took place at the preclinical research facility of DKFZ. To prevent corneal injury, protective eye cream was applied to the animals' eyes during the orthotopic implantation procedure. To maintain physiological body temperature, all animals undergoing anesthesia were placed on a warming mat with a temperature set at 37 °C. Post-surgery pain relief was provided via subcutaneous (s.c.) injection of carprofen (#15, Cp Pharma) at a dose of 10 mg/kg body weight, diluted in a 0.9% (w/v) NaCl solution (#630138, Braun) with a second dose given 24 h later. Carprofen was then administered to the animals via drinking water for 48 h following any surgical intervention. To minimize technical variability, all animals underwent orthotopic GB cell implantation and local CAR-T cell administration according to a harmonized experimental schedule, with only minimal temporal differences between experimental units. Tumor signals were subsequently acquired for all experimental units within the same measurement session using the same instrumentation, reagents from identical production batches, and standardized data acquisition and analysis procedures with identical processing parameters. Results CD70 has a tumor-promoting role in murine GB Driven by the sustained anti-tumor effects of murine CD70 CAR-T cells in our previous study, 11 and to better define the tumor-intrinsic impact of CD70 expression in the utilized syngeneic model, we performed bulk RNA sequencing on GL261/CD70 cells and their non-expressing control counterpart (GL261/MCS). Differential gene expression (DGE) analysis revealed significant upregulation of Lef1, a transcription factor implicated in GB aggressiveness and self-renewal, 14 Nts, which promotes GB proliferation and invasion via NTSR1 activation and is associated with poor prognosis in GB patients, 15 , 16 and Wnt6, a regulator of oncogenic signaling linked to enhanced tumorigenicity and reduced overall survival in in vivo models, 17 in GL261/CD70 cells compared to the control. Notably, we also observed upregulation of genes with immunosuppressive functions. We report upregulation of Fgl2 in GL261/CD70 cells, a gene that establishes an immunosuppressive niche characterized by Treg recruitment, expansion of anti-inflammatory M2-polarized macrophages, and impaired CD103+ DC differentiation. 18 , 19 Additionally, Cd200, a checkpoint molecule with proposed inhibitory properties was also upregulated in GL261/CD70 cells. 20 In contrast, several genes with reported tumor-suppressive functions in GB were downregulated, including Sh3bgrl2, which has been shown to inhibit GB growth, migration, and self-renewal, 21 and Prkg2, implicated in restraining GB cell proliferation (Figure 1A and B). 22 Expression of selected differentially expressed genes was validated by RT-qPCR (Figure 1C, upper panel) and Cd70-OE was further verified by RNAseq quantification (Figure 1C, bottom panel). Gene ontology (GO) analysis of upregulated genes revealed enrichment of pathways linked to GB tumorigenicity and stemness, including TGF-beta and Wnt signaling (Figure 1D). To further evaluate the GB-promoting role of murine CD70 in vivo, immunocompetent C57BL/6J mice were orthotopically implanted with GL261/MCS or GL261/CD70 cells (N = 12 animals per group). MRI assessment 14 d later revealed significantly larger tumors in mice receiving GL261/CD70 cells compared to control GL261/MCS cells (Figure 1E). Interestingly, flow cytometric analysis of tumor-infiltrating leukocytes highlighted a trend towards increased immunosuppressive Treg frequency in GL261/CD70 tumors (p = 0.06, Figure 1F). Collectively, these findings provide evidence for the biological relevance of CD70 expression in syngeneic GB. Transcriptomic profiling and evaluation of the oncogenic potential of murine Cd70 overexpression. (A) Volcano plot of differentially expressed genes between GL261/CD70 and GL261/MCS cells. Genes fulfilling the Log2FC criterion are marked in green. Genes fulfilling the p-value criterion are marked in blue. Genes fulfilling both criteria are marked in red. Genes fulfilling no criteria are marked with gray. The Log2FC cutoff was set at Log2FC < −1 or Log2FC > 1 and p-value cutoff was p < 0.05. (B) Expression heatmap of genes fulfilling the criteria from (A). (C) Upper panel: Validation of expression levels of selected genes from (A) with RT-qPCR. Bottom panel: Cd70 FPKM expression values from (A). (D) Gene ontology pathways and biological processes significantly enriched in GL261/CD70 cells from (A). (E) Measurement of established tumor volumes 14 d after orthotopic implantation by MRI. (F) Quantification of GB-infiltrating Tregs in the formed tumors from (E) by flow cytometry. Data gated on live and single CD45+/CD3+/CD4+/CD8a− cells. Gating was established using unstained and isotype control antibodies. For (A) and (C), N = 4 biological replicates. For (E) and (F), each dot represents an animal. For (C), (E), and (F), data presented as mean (SD). Statistical significance was assessed with an unpaired two-tailed t-test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s., not significant. scRNAseq identifies all major immune cell populations in locally treated mice The substantial and sustained tumor remissions observed following treatment with murine anti-CD70 CAR-T cells in our previous study 11 prompted us to profile the brain tumor immune landscape after interception by CD70-targeted CAR-T cells by employing scRNAseq on enriched immune cells from the brains of treated mice (Figure 2A). Specifically, we orthotopically implanted immunocompetent C57BL/6J mice (N = 16 animals) with GL261/CD70 GB cells and verified the presence of tumors 14 d later by MRI. One day later, mice were intratumorally treated with a single dose of 1.5 * 106 mock-transduced non-targeting T-cells (pMSCV) or CD70-targeting ligand-based mCAR27 CAR-T cells (Supplementary Figure S1A and B, N = 8 animals per group). 11 Contrary to the control, mCAR27 CAR-T cells led to marked tumor remission, determined six days later (Day-21) by MRI (Figure 2B, Supplementary Figure S1C). On Day-22, animals were sacrificed and single-cell suspensions from the tumor-bearing hemispheres were generated. Subsequently, CD45+ immune cells were sorted (Supplementary Figure S1D and E). No significant difference in the number of sorted CD45+ immune cells between treatment groups was discovered (Supplementary Figure S1F). Incorporation of tdTomato in the constructs allowed us to identify transduced T-cells in the sorted immune cell pool (Supplementary Figure S1G). Enriched CD45+ immune cells were subjected to scRNAseq by utilizing the Chromium Next GEM Single Cell 5′ technology (10× Genomics). A total of 38923 cells passed QC and served as input to perform cluster annotation based on the expression of known marker genes (Figure 2C and D, Supplementary Figure S2). All clusters expressed global immune cell marker Ptprc (Cd45). We identified three P2ry12 +/Tmem119 + microglia (MG) clusters: one characterized by elevated Kcnma1 expression (MG_Kcnma1+; P2ry12 + /Tmem119 + /Kcnma1 + ), one by Spp1 expression (MG_Spp1+; P2ry12 + /Tmem119 + /Spp1 + ), and one by increased Crybb1 expression (MG_Crybb1+; P2ry12 + /Tmem119 + /Crybb1 + ). Macrophages (MP) segregated into anti-inflammatory M2-like cells (M2-MP; Cd68 + /Arg1 + /Ms4a7 + ) and pro-inflammatory M1-like cells (M1-MP; Cd68 + /Clec4e + /Trem1 + ). Analysis of dendritic cells (DC) resulted in subdivision into conventional DCs (DC3; Flt3 + /Xcr1 + /Clec9a + , widely known as cDC1 in the literature), plasmacytoid DCs (pDC; Flt3 + /Ccr9 + ), a cluster positive for Cacnb3 (DC2; Flt3 + /Cacnb3 + ), and a cluster of monocyte-derived DCs, expressing Cd209 (DC1; Flt3 + /Cd209 + ). Regarding natural-killer (NK) cells, we identified two clusters, both positive for universal markers Ncr1 and Itga2, but expressing different levels of Gzma (NK1; Ncr1 + /Itga2 + /Gzma high , and NK2; Ncr1 + /Itga2 + /Gzma low ). Two B-cell clusters were discovered (BC1; Cd19 + /Ebf1 + /Vpreb3 − /Cd79a low , and BC2; Cd19 + /Ebf1 + /Vpreb3 + /Cd79a high ). Concerning T-cells (TC), nine clusters were identified. We annotated Cd4-positive helper T-cells (CD4 TC; Cd3e + /Cd4 + ) and two clusters of Cd8-expressing cytotoxic T-cells, with different degrees of Cd8 and Gzmb expression (CD8 TC1; Cd3e + /Cd8 high /Gzmb high , and CD8 TC2; CD3e + /Cd8 low /Gzmb low ). A cluster of proliferative T-cells was identified based on high expression of Mki67 and Stmn1 (Proliferative TC; CD3e + /Mki67 + /Stmn1 + ), and in contrast, two exhausted T-cell clusters were also observed (Exhausted CD4 TC; Cd3e + /Cd4 + /Tox + /Pdcd1 + , and Exhausted CD8 TC; Cd3e + /Cd8 + /Tox + /Pdcd1 + ). Exogenously administered CD70-targeting CAR-T cells or mock-transduced T-cells were discriminated based on the tdTomato fluorescent reporter (CD8+/tdTom+; CD3e + /Cd8 + /tdTomato + ) and γδ T-cells were also identified based on the expression of TrcgC1 (γδ TC; Cd3e + /TrcgC1 + ). Expression of Foxp3 transcription factor was a determinant for identification of Tregs (TREG; Cd3e + /Cd4 + /Foxp3 + ). Finally, additional annotated subtypes included basophils (BAS; Il6 + /Cd200r3 + ), mast cells (MC; Fcer1a + ), and neutrophils (NEU; S100a8 + /S100a9 + ). In conclusion, all main immune cell subsets were identified by our clustering approach in mouse brains post-treatment. Identification of immune cell subsets in treated mice. (A) Schematic representation of the in vivo single-cell transcriptomic study. (B) Assessment of tumor volume after treatment by MRI. Each dot represents an animal (N = 8 animals per group). A paired two-tailed t-test was performed to assess statistical significance. (C) UMAP embedding of GB-infiltrating immune populations at a resolution of 1.1 (D) Dot plot of key marker gene expression in identified immune cell clusters from (C). The size of each dot represents the expression percentage and the color represents the expression levels. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s., not significant. CD70 CAR-T cell treatment alters the immune landscape in the brain TME Our analysis revealed that CAR-T cell treatment resulted in a shift of immune cell absolute numbers and frequencies (Figure 3A and B, Supplementary Figure S3). Concerning adoptively transferred cells, we observed a significantly higher incidence of Cd8 +/tdTom + cells in the mCAR27 group, suggesting successful CAR-T cell in vivo expansion upon target interception. This observation also applied to both identified Cd8-expressing T-cell clusters (CD8 TC1 and CD8 TC2), suggesting that CAR-T cell therapy facilitates elevated brain tumor infiltration by endogenous T-cell populations with cytotoxic functions. Coherently, this was accompanied by an increase in Cd8 T-cells with an exhaustion-like phenotype (Exhausted CD8 TC cluster), reflecting previous target antigen exposure in the GB TME. The opposite was observed for Cd4 + exhausted T-cells, a subset whose prevalence decreased upon CAR-T cell therapy. Critically, mCAR27 treatment also led to a decrease in immunosuppressive populations. Specifically, we discovered a reduction in the frequency of anti-inflammatory tumor-promoting Cd68 + /Arg1 + /Ms4a7 + M2-like macrophages (M2-MP cluster) and Tox +/Pdcd1 +/Foxp3 + Tregs (TREG cluster), subsets that suppress T-cell activity in GB by secretion of IL-10, TGF-β and expression of arginase, as well as by dampening antigen presentation and dendritic cell maturation. 23 Although minor differences were reported in the case of B-cells and microglia, we observed a reduction in the incidence of monocytic DCs (cluster DC1; Flt3 + /Cd209 + ) and Cacnb3 + DCs (cluster DC2), as well as one NK cell subset (cluster NK1) after treatment with anti-CD70 CAR-T cells (Figure 3C). Taken together, we showcase that treatment with CD70-targeting CAR-T cells led to a remodeling of the TME, manifested by effector cell expansion, reduction of immunosuppressive populations, but also downregulation of certain DC and NK cell populations, raising the possibility of a potential on-target or indirect off-tumor effect. Comparison of immune cell frequencies between treatment arms. (A) UMAP distribution of identified immune cell populations across treatment groups. (B) Average frequency of immune cells per identified cluster across treatment arms. N = 8 animals per group. (C) Direct comparisons of immune cell frequencies per identified subset. Each dot represents an animal. The median is displayed. A Mann-Whitney test was performed to assess statistical significance in the following clusters: M1-MP, NK1, and all microglial clusters. An unpaired two-tailed t-test was used for all other clusters. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s., not significant. CD70 is expressed in GB-infiltrating immune cells The observed alterations in immune cell frequencies after mCAR27 treatment prompted us to examine whether they were attributable to surface CD70 expression. Although absent from normal brain tissue, Cd70 is expressed in immune cell subsets, primarily DCs, but also NK cells and a fraction of activated T-cells. 24–26 Using in-house scRNAseq data of primary GB patients, we previously highlighted Cd70 expression on Tregs and TAMs, providing a rationale for employing CAR-T cells not only to intercept CD70+ tumors but also to remodel the TME by depleting immunosuppressive populations. 11 To interrogate CD70 expression on the surface of GB-infiltrating immune cells, we orthotopically implanted C57BL/6J mice with GL261/CD70 cells (Figure 4A). Single-cell suspensions from the tumor-bearing hemispheres were generated and analyzed by flow cytometry. Concerning the lymphoid lineage, CD70 expression was marginal on B-cells, and was additionally observed in a subset of NK cells, potentially justifying the lower NK cell frequency upon mCAR27 treatment (Figure 4B). CD70 was also expressed in a significant portion of T-cells. This may have resulted from tumor-driven T-cell signals, as Cd70 is induced upon T-cell stimulation. 27 Further investigation unveiled the presence of CD70 in the vast majority of immunosuppressive Treg cells (CD45+/CD3+/CD4+/CD25+/Foxp3+, Figure 4C), offering a mechanistic explanation for their reduced frequency following CD70-targeting. Regarding the myeloid compartment, we detected CD70 on the membrane of a low fraction of DCs, a phenomenon well-documented in the literature, 25 but also on a higher percentage of macrophages (Figure 4D). Accounting for their M1/M2-like polarization, we discovered CD70 expression in macrophages expressing both CD86 (CD45+/CD11b+/F4.80+/CD86+), a co-stimulatory marker linked to the M1-phenotype, and CD206 (CD45+/CD11b+/F4.80+/CD206+), a mannose receptor expressed predominantly on M2-polarized macrophages (Figure 4E). Taken together, CD70 was detected across multiple GB-infiltrating immune subsets. This distribution potentially provides a mechanistic explanation for the induced immune remodeling following CD70-targeting, particularly through depletion of Treg cells and M2-like MPs. Our data also indicate that our CD70-directed approach could extend to NK cells and DCs, warranting consideration of potential effects on innate immunity and antigen presentation. Evaluation of CD70 expression on the surface of GB-infiltrating immune cells. (A) Schematic overview of the pipeline. (B) Gating strategy and quantification of lymphoid cells expressing CD70 by flow cytometry. (C) Gating strategy and quantification of Treg cells expressing CD70 by flow cytometry. (D) Gating strategy and quantification of myeloid cells expressing CD70 by flow cytometry. (E) Gating strategy and quantification of M1- and M2-like MPs expressing CD70 by flow cytometry. For (B)–(E), the mean is displayed. N = 3 animals were used for this experiment (lymphoid and myeloid). For (B)–(E), representative gating strategy dot plots from a single animal. Isotype control antibodies and CD70- an FOXP3-FMO control samples were used to determine gating. Tumor clearance by mCAR27 orchestrates inflammatory resolution and alleviates immunosuppression To elucidate whether our immunotherapy approach leads to transcriptional remodeling in the identified populations, we performed DGE analysis. To facilitate robust results and allow for a clearer interpretation of mCAR27-induced transcriptional remodeling, very closely related subclusters were consolidated into broader lineage-defined immune cell compartments (Supplementary Figure S4A). Among all clusters, M2-like macrophages exhibited the most pronounced transcriptional remodeling upon CD70-targeted CAR-T cell therapy (Figure 5A and B). Specifically, we observed a downregulation of multiple type-I interferon-stimulated genes (ISG, Figure 5C, Supplementary Figure S4B), accompanied by decreased expression of Ccl5, Cxcl2, and Cxcl9, cytokines that mediate immune cell infiltration (Supplementary Figure S4C). Notably, anti-inflammatory M2-like macrophages in mCAR27-treated brains also downregulated key immunosuppressive markers such as Cd274 (PD-L1), Tgfbi and Pilra, but also Slc7a2 and Arg1 (most downregulated gene), two genes encoding products that deplete arginine from the TME, establishing a suppressive niche (Figure 5D and E). 23 An arginine-depleted TME in pMSCV-treated brains is also suggested by the upregulation of the arginine synthase gene Ass1 as a potential stress-coping mechanism in pMSCV M2-like MPs. In contrast, mCAR27 M2-MPs upregulated Tmem176a, Tmem176b, Cd83, and Gpnmb, genes that promote resolution of tumor-related inflammation and enhance tissue protection and homeostasis programs (Figure 5F). 28–30 A similar pattern was observed with inflammatory M1-like MPs. Namely, we again observed downregulation of type-I ISGs (Supplementary Figure S4D–F); and Tmem176a and Tmem176b were the only significantly upregulated genes in this cluster (Supplementary Figure S4G). GO analysis of downregulated genes in both MP clusters confirmed enrichment of processes linked to interferon response as well as cytokine-mediated pathways (Figure 5G, Supplementary Figure S4H). The DC hypercluster underwent similar transcriptional remodeling upon mCAR27 treatment. DCs from mCAR27-treated mice also downregulated ISGs (Figure 6A–C, Supplementary Figure S5), showing overlap with MPs (Figure 6D), and exhibited a less metabolically stressed phenotype. We also discovered Ass1 downregulation, an observation potentially leading to sufficient arginine supply within the TME. Furthermore, key suppressors Socs1 and Pdcd1lg2 (PD-L2) were also repressed, implying that the removal of tumor-driven inflammatory cues not only alleviates stress responses but also attenuates immune suppression (Figure 6E). Importantly, the convergence in ISG downregulation between MPs and DCs was further recapitulated at the signaling pathway level by performing GO analysis (Figure 6F). Analogous to its effects on macrophages and DCs, mCAR27 treatment also reprogrammed Tregs toward a less immunosuppressive state. Notably, mCAR27 Tregs showed decreased expression of Lag3, Zc3h12d (promotes degradation of pro-inflammatory mRNAs), 31 and Dusp16 (inhibits secretion of pro-inflammatory cytokines). 32 These findings were accompanied by downregulation of proliferation-enhancing Cdk6, as well as Pcyt1a, a gene that supports membrane integrity. 33 Similarly, Myo10, a gene associated with tumor growth and migration, 34 , 35 was also downregulated in mCAR27 Tregs (Figure 7A and B). These transcriptional alterations may act in concert with potential CD70-targeting on the surface of Tregs, contributing to their lower abundance in mCAR27-treated brains (Figures 3B, 7C). Collectively, these findings suggest that CAR-T cell treatment not only diminishes Treg incidence but also attenuates their suppressive programs, thereby contributing to a TME more permissive for effective long-term tumor surveillance. Finally, regarding the microglial hypercluster, although no extensive transcriptional alterations were identified, our analysis showed upregulation of antigen presentation-related genes H2-Oa, H2-Ab1, H2-Aa, Cd74 and H2-Eb1 in control-treated mice, likely responding to the ongoing tumor-derived inflammatory signals (Supplementary Figure S6A and B). In contrast, these genes were significantly downregulated in mCAR27-treated mice, plausibly a result of mCAR27-mediated elimination of GB antigenic stimuli and the marked global downregulation of ISGs, which are dominant upstream drivers of MHC-II genes. Taken together, our findings indicate that following tumor clearance, mCAR27 CAR-T cells reprogram the immune landscape towards a protective state that restrains overactivation, resolves inflammation, and preserves homeostasis, while simultaneously dampening immunosuppressive activity to support robust immune surveillance. DGE analysis in M2-MPs post-CAR-T cell therapy. (A) Heatmap of significantly altered genes in M2-MPs between control- and mCAR27-treated mice (N = 2506 cells total). (B) Volcano plot of differentially expressed genes from (A). Genes with log2FC < −1 and log2FC > 1 are marked with blue and red respectively. (C) Expression plot of selected genes from (A). (D) Expression levels of genes linked to L-arginine metabolism (top) and immunosuppression (bottom) from (A). (E) Feature plots depicting expression of L-arginine metabolism genes from (A). (F) Expression levels of genes associated with resolution of inflammation and tissue repair from (A). (G) Enriched GO terms and biological processes of significantly downregulated genes in mCAR27 M2-MPs. For (D) and (F), each dot represents an animal (N = 8 per group), and data are presented as mean (SD). Significance was investigated with an unpaired two-tailed t-test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s., not significant. DGE analysis in the DC hypercluster post-CAR-T cell therapy. (A) Heatmap of significantly altered genes in DCs between control- and mCAR27-treated mice (N = 3548 cells total). (B) Volcano plot of differentially expressed genes from (A). Genes with log2FC < −1 and log2FC > 1 are marked with blue and red respectively. (C) Expression of the ISG signature from (A) across treatment arms. (D) Venn diagram depicting overlap of significantly downregulated ISGs in M1-MPs, M2-MPs, and DCs. (E) Expression levels of selected genes from (A). Each dot represents an animal (N = 8 per group), and data are presented as mean (SD). Significance was assessed with an unpaired two-tailed t-test. (F) Enriched GO terms and biological processes of significantly downregulated genes in mCAR27 DCs. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s., not significant. DGE analysis in Tregs post-CAR-T cell therapy. (A) Heatmap of significantly altered genes in Tregs between control- and mCAR27-treated mice (N = 1169 cells total). (B) Expression levels of selected genes from (A). Each dot represents an animal (N = 8 per group), and data are presented as mean (SD). An unpaired two-tailed t-test was used to assess significance for all genes except from Lag3, for which a Mann–Whitney test was utilized. (C) Absolute number of GB-infiltrating Tregs identified in each treatment arm. Each dot represents an animal. Significance was assessed with an unpaired two-tailed t-test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s., not significant. mCAR27 treatment remodels predicted intercellular communication involving endogenous CD8 T cells Given the increased representation of endogenous CD8 T-cells and concomitant reduction of TREG and M2-like macrophage populations following mCAR27 treatment, we finally investigated whether these compositional changes were accompanied by alterations in predicted intercellular communication. CellChat analysis was performed to compare ligand–receptor-mediated communication between the broader lineage-defined immune cell populations (Supplementary Figure S4A) in pMSCV- and mCAR27-treated tumors, with particular emphasis on signaling involving endogenous CD8 T cells. At a pathway level, mCAR27 treatment was associated with extensive reorganization of the predicted TGF-β communication network (Figure 8A). Specifically, whereas pMSCV-treated tumors displayed a highly interconnected TGF-β network involving multiple immune cell populations, the network observed following mCAR27 treatment exhibited a markedly reduced interconnectivity and architecture. Nevertheless, TGF-β signaling remained detectable after mCAR27 treatment, indicating redistribution rather than complete loss of predicted TGF-β-mediated intercellular communication. Predicted intercellular communication following mCD70-targeted therapy. (A) CellChat-inferred TGF-β signaling networks in pMSCV- and mCAR27-treated tumors visualized as chord diagrams, showing predicted communication between the indicated major immune-cell populations. (B) Predicted TREG-to-CD8 T-cell communication. (C) Predicted M2-like macrophage (M2-MP)-to-endogenous CD8 T-cell communication. For (B) and (C), left, relative information flow of signaling pathways compared between pMSCV and mCAR27 treatment conditions; right, ligand-receptor interactions increased in the pMSCV condition, with dot color representing the predicted communication probability and dot size indicating statistical significance. We next examined communication from immunoregulatory populations toward endogenous CD8 T-cells. Predicted TREG-to-CD8 T-cell communication differed substantially between treatment conditions (Figure 8B, left panel). TGF-β signaling was selectively associated with the pMSCV condition in the pathway-level comparison, while several other signaling pathways remained detectable or were preferentially represented following mCAR27 treatment. Analysis of ligand-receptor interactions increased in pMSCV further identified Tgfb1 (Tgfbr1 + Tgfbr2) as a predicted TREG-to-CD8 T-cell interaction in control-treated tumors that was not detected in the corresponding mCAR27 interaction network (Figure 8B, right panel). Additional predicted interactions selectively detected in the pMSCV condition included several ICAM1- and CD80-associated ligand:receptor pairs (e.g. CD80-CTLA4), indicating that the treatment-associated alteration in TREG-to-CD8 T-cell communication extended beyond the TGFβ pathway. A similar treatment-associated alteration was observed for predicted M2-like macrophage-to-CD8 T-cell communication. TGF-β signaling was again selectively associated with the pMSCV condition (Figure 8C, left panel), with Tgfb1 (Tgfbr1 + Tgfbr2) identified among ligand-receptor interactions increased in pMSCV (Figure 8C, right panel). In addition, predicted PD-L1 signaling was relatively stronger in pMSCV-treated tumors, and Cd274-Pdcd1 communication between M2-like macrophages and endogenous CD8 T-cells showed lower predicted communication probability following treatment with mCD70-targeting mCAR27 cells (Figure 8C, right panel). These changes occurred alongside other treatment-associated alterations in communication, indicating remodeling rather than a complete loss of interplay between these populations upon CD70 targeting. Finally, we investigated predicted communication between the adoptively transferred tdTomato+ population and endogenous CD8 T-cells. The pathway composition of this communication network differed markedly between the two treatment conditions. Communication associated with the pMSCV condition included TGF-β, SPP1, CCL and MIF signaling, whereas the mCAR27 condition was associated with a distinct set of predicted pathways, including CXCL, ICAM, ALCAM, CD80, CD86 and MHC-II signaling (Supplementary Figure S7), suggesting that the predicted communication landscape between adoptively transferred and endogenous CD8 T cells was substantially remodeled following mCAR27 treatment. Collectively, these analyses demonstrate that the previously observed compositional remodeling of the tumor-infiltrating immune compartment following mCAR27 treatment is accompanied by extensive alterations in predicted intercellular communication involving endogenous CD8 T-cells. These findings complement the observed changes in immune-cell composition and identify treatment-associated alterations in immunoregulatory signaling networks that may contribute to the remodeled immune landscape following mCAR27 therapy. Discussion In recent years, scRNAseq has become a powerful tool for profiling the complex cellular and molecular effects of CAR-T cell therapy on the TME. 3 , 4 Capturing immune populations and dynamic transcriptional programs at high resolution may reveal resistance mechanisms, therapy-induced remodeling, and potential vulnerabilities that can be exploited to potentiate treatment outcomes. Driven by the remarkable survival prolongation of our established syngeneic model conferred by mCD70-targeting CAR-T cells, 11 this follow-up study is the first to employ scRNAseq to comprehensively characterize the brain immune landscape following anti-CD70 CAR-T cell treatment. Our data indicate that post tumor interception, mCAR27 CAR-T cells triggered a coordinated immune recalibration consistent with resolution of inflammation, while simultaneously reducing immunosuppressive signals, facilitating long-term surveillance providing insights into the immune landscape after tumor elimination. An important consideration of the present model is the high and homogeneous CD70 expression resulting from lentiviral Cd70 overexpression. While this approach was necessary to establish a CD70-positive syngeneic GB model in an immunocompetent setting, GL261/CD70 should therefore be considered a CD70-high experimental model rather than one representative of average CD70 expression in human GB. Nevertheless, Cd70 expression in GL261/CD70 cells (mean 52.6 FPKM) was within the upper range of CD70 expression observed in our previously published human GB RNAseq cohort, in which several human tumors exhibited expression exceeding 50 FPKM. 11 This cross-species comparison should, however, be interpreted cautiously, given the differences between cultured murine tumor cells and heterogeneous human bulk tumor specimens, as well as other differences including genome annotation, library preparation, sequencing procedures, and methodological data analysis pipelines. A principal barrier limiting the success of immunotherapy is proliferative exhaustion. Despite the marked tumor reduction observed, our analysis revealed an increase in the fraction of exhausted cytotoxic T-cells post-treatment (Cd45 + /Cd3e + /Cd8 + /Tox + /Pdcd1 + ). This finding paves the way for future studies combining CAR-T cells with PD1-blockade to further enhance anti-tumor efficacy. In a relevant clinical study, Bagley et al. demonstrated that although safe and well-tolerated, peripheral infusions of EGFRvIII-directed CAR-T cells in combination with pembrolizumab showed no detectable clinical efficacy for GB treatment (NCT03726515). 36 However, patients in this study were treated systemically. In contrast, our approach focuses on local delivery, a strategy that has been associated with improved anti-tumor activity. 37 , 38 Preclinical studies have underscored the added benefit of incorporating checkpoint interference with CAR-T cell therapy for GB treatment. 39–42 Therefore, future studies could potentially investigate whether the sustained tumor remissions obtained after CD70-directed CAR-T cell therapy can be further augmented by checkpoint blockade. A relevant finding of our study was the downregulation of arginine depletion genes Arg1 and Slc7a2 in immunosuppressive macrophages after mCAR27 treatment, presumably leading to a more L-arginine-enriched TME compared to the control arm. Previous research has highlighted the importance of L-arginine for T-cell metabolism, proliferation, and survival in vitro and in vivo. 43 , 44 Furthermore, this amino-acid is central for T-cell expression of the Cd3ζ chain, entailing the immunoreceptor tyrosine-based activation motifs that trigger signaling upon target engagement. 45 Regarding immunotherapy, studies have demonstrated that L-arginine production potentiates the in vitro and in vivo anti-tumor effects of transgenic as well as CAR-T cell products for solid and hematological malignancies. 44 , 46 Taking these studies into consideration, the downregulation of Arg1 and Slc7a2 in mCAR27-treated mice might result in an L-arginine-rich TME, potentially explaining the enhanced incidence of cytotoxic T-cells and exogenous therapeutic cells in mCAR27-treated mice compared to the control. Another contributing factor could also be the elimination of GL261/CD70 malignant cells, as GB CD70 is shown to induce T-cell apoptosis. 7 , 8 , 47 Therefore, elimination of GB cells might also remove the chronic CD70-driven T-cell overstimulation that leads to cell death, thereby supporting increased survival and accumulation after treatment. In line with this, our previous study demonstrated that mCAR27 CAR-T cells not only significantly extended survival compared to control-treated mice but also persisted within the brain for months after a single intratumoral administration, 11 linking sustained anti-tumor surveillance with durable in vivo CAR-T cell persistence. The profoundly immunosuppressive GB TME represents a major obstacle to achieving long-lasting clinical responses after CAR-T cell therapy. Populations such as Tregs and M2-polarized TAMs are key mediators of immune evasion, functioning by secreting suppressive cytokines, depleting metabolites, and impairing antigen presentation. 1 Previous research suggests that targeting such populations may enhance anti-tumor responses. Tilsed et al. showed that CCR4-directed CAR-T cells can selectively deplete Treg cells ex vivo and in vivo. 48 Similarly, Watanabe et al. demonstrated that in T-cell lymphoma, CCR4-guided CAR-T cells not only eliminated malignant cells but also depleted Treg cells, while sparing Th1 and CD8+ cytotoxic T-cell subsets that are critical for effective anti-tumor immunity. 49 This approach may also improve the functional quality of the CAR-T cell product itself, as Treg depletion during manufacturing has been shown to augment the CAR-T cell cytotoxic capacity. 50 In agreement with published research, 51 our experiments revealed CD70 expression on the surface of CD3+/CD4+/CD25+/FOXP3+ Tregs, the abundance of which decreased after CD70-directed CAR-T cell therapy. Similarly, we detected CD70 on M2-polarized macrophages, corroborating previous findings showcasing Cd70 gene expression on TAMs co-expressing Cd163 and Tgfbi. 11 Several studies have underscored the potential of CAR-T cell-mediated TAM depletion to augment adoptive immunotherapy. 52–54 Future research will investigate the functional consequences of CD70 incidence on these populations. Co-culture of mCAR27 CAR-T cells with GB-infiltrating Tregs could clarify whether the observed reduction in Treg abundance results from direct CAR-T cell-mediated killing, complementing the observed reduced Cdk6 Treg expression levels as well as Ccl5 downregulation, a cytokine that recruits Tregs via interaction with CCR5. 55–57 Conversely, experiments will also aim to discover the effect of M2-MP CD70 expression on the elicited anti-tumor response and to determine whether the less immunosuppressive M2-MP transcriptome stems from direct mCAR27 engagement. Our CellChat analysis revealed extensive treatment-associated changes in predicted intercellular communication. Particularly notable was the altered predicted communication from immunoregulatory populations toward endogenous CD8 T-cells, with TGF-β signaling from both TREGs and M2-like macrophages toward endogenous CD8 T-cells preferentially associated with the pMSCV condition. While these observations provide an additional mechanistic layer linking the compositional and transcriptional changes identified in this study, CellChat infers communication from ligand and receptor expression and therefore cannot establish whether these predicted interactions are functionally active in vivo or whether they contribute directly to tumor control. Direct functional experiments will be required to address this question. Depletion of endogenous CD8 T-cells could determine whether this population contributes to therapeutic efficacy, while rechallenge of successfully treated animals with CD70-expressing tumor cells could test whether treatment results in functional endogenous antitumor protection. Such experiments would provide an important functional extension of the present single-cell analyses and determine whether the treatment-associated remodeling contributes to durable tumor control. An additional consideration when interpreting the treatment-associated immune remodeling observed in this study is the marked difference in tumor burden between the two treatment groups at the time of scRNAseq analysis. Namely, mCAR27-treated mice exhibited pronounced tumor regression compared with pMSCV-treated mice, and the present experimental design therefore does not allow us to fully disentangle direct effects of mCAR27 treatment on the immune compartment from secondary changes resulting from shrinking tumor sizes. Indeed, the reduced inflammatory and immunosuppressive features observed following mCAR27 treatment may, at least in part, reflect resolution of tumor-associated inflammatory stress following tumor regression. Thus, the immune landscape described here should be interpreted as the state associated with effective mCAR27-mediated tumor control, rather than as evidence that all observed immune alterations are directly induced by CAR-T cells independently of changes in tumor burden. Future studies incorporating longitudinal immune profiling at multiple time points after CAR-T cell administration, including early time points preceding substantial differences in tumor burden, could help distinguish early treatment-associated immune changes from those emerging secondary to tumor regression. One limitation of CAR-T cell therapy is the potential for on-target/off-tumor effects, reported in the clinic. 58–60 Although our local administration strategy is expected to minimize the risk of systemic toxicities, and even though CD70 is not expressed in normal brain tissue, its demonstrated presence on a fraction of GB-infiltrating DCs and NK cells may represent a potential side-effect. We observed a reduction in the frequency of specific DC subsets following mCAR27 treatment. Importantly however, we found no difference in the fraction of conventional Flt3 + /Xcr1 + /Clec9a + DCs (cluster DC3 in our analysis, widely referred to as cDC1 in the literature), a subset which is critical for eliciting anti-tumor responses by cross-presenting antigens to MHC-I to prime CD8+ T-cells. 61 , 62 This suggests that the principal antigen presentation machinery to cytotoxic T-cells is not directly compromised in our model. On the other hand, we observed a decrease in the frequency of CD45 +/Flt3 +/Cd209 + cells (cluster DC1), resembling monocytic DCs. This could be a result of either direct CD70 targeting on their surface or the general TME shift towards a less inflammatory state after mCAR27 treatment. Although monocytic DCs are considerably less efficient in stimulating T-cells compared to conventional DCs, 63 and some reports even suggest a tumor-promoting role in GB, 64 these cells can still support Th1 and Th17 polarization, present tumor antigens to T-cells, and enhance cytotoxicity of CD8+ T-cells. 65 Therefore, the consequences of their reduction following mCAR27 treatment will be investigated further. Like our interpretation for DCs, the reduced NK abundance after treatment may also reflect a combination of diminished inflammatory recruitment following GB eradication and selective depletion of CD70+ NK cells. Interestingly, NK cells also secrete CCL5 and XCL1 which function by recruiting conventional DCs (cluster DC3 in our study), stimulating their differentiation and expansion. 66 Based on their elicited anti-tumor characteristics, an in-depth investigation of the consequences of NK cell reduction as a CAR-T cell side effect will be the focus on future research. In summary, our study provides the first single-cell resolution map of the brain immune landscape following CD70-directed CAR-T cell therapy, revealing remodeling of the TME characterized by resolution of inflammation and reduced immunosuppression. Our data establish a mechanistic framework for understanding post-CAR-T cell immune dynamics in GB, underscore the emergence of potential on-target/off-tumor effects, and lay the groundwork for investigation of potential combination strategies aimed at further enhancing the therapeutic efficacy of utilized mCAR27 cells. Supplementary Material Supplementary Material Decoding_immune_landscape_with_scRNAseq_SUPPLEMENTARY_REVISED_CLEAN.docx