Introduction

Glioblastomas (GBM) stand as the most prevalent and aggressive primary brain malignancy among adults.1 The current standard of care includes maximal surgical resection followed by radiotherapy and chemotherapy with temozolomide. Unfortunately, this aggressive management is rarely curative, cancer recurrence and death are almost inevitable. Individuals diagnosed with GBM exhibit a median survival of 15.4 months, with less than 5% surviving beyond 5 years.2 Currently, many novel GBM immunotherapy trials, including vaccines, chimeric antigen receptor T cells, and immune checkpoint blockade (such as programmed death 1 (PD-1)/ligand-1 (PD-L1) blockade therapy), have shown only modest benefits in patients with GBM.3 Lack of tumor-infiltrating T cells, as well as uniquely immunosuppressive macrophages, are significant barriers limiting the efficacy of immunotherapy in GBM and the relapsing and refractory treatment of patients with GBM.4 There is an urgent need to decipher the pathogenic mechanism of GBM and develop more effective treatments for the patients.

Cytomegalovirus (CMV) is a β-herpes virus which exhibits strict host specificity and establishes life-long persistence with alternating phases of latency and productive infection.5 CMV was reported present in many solid cancers, including breast,6 prostate7and medulloblastoma.8 Present mainly in tumor cells, the CMV infection may be tumorigenic, inducing the activation of mitotic signals transmitted by the products of proto-oncogenes such as c-fos c-jun and c-myc.5 9 Cobbs et al first reported the detection of human CMV antigens in GBM in 2002.10 Since then, many studies have attempted to address the controversial question of whether there is an association between CMV and GBM.11 CMV encodes for a plethora of gene products that serve to evade the immune system by intercepting, for example, interferon-regulatory factor, signal transducer and activator of transcription, and nuclear factor κB dependent signaling.12 However, it remains largely elusive how immune cells in the GBM tumor environment respond to the presence of CMV, especially macrophages and T cells. Meanwhile, it is unknown how CMV-infected GBM altered depending on the disease stage whether CMV-specific cell subsets are found in different uninfected CMV tumor microenvironments and/or exhibit distinct immune checkpoint, and how to design to use CMV as a novel, potential therapeutic target for the tumor.

In this work, we confirmed the presence of CMV infection in 60% of GBM cases. For the first time, we have constructed a single-cell transcriptomic atlas for GBM with CMV infection. Our findings indicate that CMV infection promotes the expansion of a subset of bipositive SOX2+CD68+ tumor-associated macrophages (TAMs). CMV manifests through mediating interactions between bipositive TAMs and novel identified immunosuppressive FXYD6+ T cells, leading to T-cell exhaustion and exacerbating the immune-suppressive microenvironment in GBM. Furthermore, we have developed CMV-T-cell receptor (TCR)-T cells, which demonstrated encouraging therapeutic effects when introduced into the GBM, providing a promising avenue for the treatment of GBM.

MethodCollection of tissue and blood samples

Relevant clinical information of the patients is succinctly presented in online supplemental table 1. Surgically resected GBM specimens were promptly placed on ice, ensuring a transfer time of no more than 30 min following lesion excision. Blood samples from patients were collected 1 day before the surgical procedure. The cohort of patients with GBM comprised 8 women and 12 men, spanning an age range from 22 to 73 years. A total of 20 samples were analyzed in this study, encompassing 20 tumor samples, and 20 patient blood samples.

Immune electron microscopy

The GBM tumor tissue, post-removal, is stored in immunoelectron microscopy fixative at 4°C, avoiding freezing during transportation. Immediate embedding and immunolabeling tests are vital to prevent antigen loss. After washing the tissue blocks thrice with pre-cooled 0.1 M phosphate buffer (PB) (pH 7.4), resin penetration occurs in stages at 4°C, followed by embedding in London Resin (LR) white resin and polymerization at −20°C for 48 hours. Ultrathin sections (70–80 nm) are cut from resin blocks on an ultra-microtome and then fished onto nickel grids. Immunolabeling involves multiple steps, including incubation with primary (Anti-Cytomegalovirus IE1/2, ab53495, Abcam; anti-HCMV-pp65, BIOSS, bs-0271R) and secondary antibodies (Goat-anti-mouse IgG (HL)−4 nm Gold, 115-185-146, Jackson; Goat-anti-rabbit IgG (HL)−10 nm Gold, G7402, Sigma), rinsing, staining, and drying the nickel grids in a controlled manner. Finally, observation under a transmission electron microscope (TEM) and image capture is conducted to identify positive signals, which appear as 4/10 nm black golden particles.

Droplet digital PCR for CMV copy number

Microdroplet preparation involves placing the generation chip in a holder, adding 40 µL microdroplet to the oil phase well, and introducing 20 µL PCR reaction water phase to the sample well. After dealing with the chip’s pad, it’s placed in the MicroDROP-100A instrument for microdroplet formation. Once formed, transfer droplets (45–55 µL) to a 96-well PCR plate. Seal the plate and puncture the film, then heat-seal it. Perform PCR within 30 min. After PCR, use a MicroDROP-100B biochip reader for detection. Preheat the detector for 30 min, insert the plate, and press the button for detection. Afterward, analyze results using Quant Drop software.

Immunohistochemistry

To assess the cell composition density and spatial positioning of bipositive SOX2+CD68+ TAMs in CMV-infected GBM, multiplex immunohistochemical (multi-IHC), and multispectral imaging were performed on formalin-fixed paraffin-embedded slides from 20 patients with GBM. The multi-IHC kit was used for staining specific cell markers, including CD68 (anti CD68 mAb,14-0688-82, Thermo Fisher), SOX2 (Anti-SOX2, #ab97959, Abcam), CMV-IE1/2 (Anti-Cytomegalovirus IE1/2, #ab53495, Abcam), TNF-β (Thermo fisher), FXYD6 (Santa Cruz), CD3 (CD3 Rabbit mAb, #49268, Signalway Antibody), KI67 (anti KI67, #P41816, ProMab Biotechnologies). After sequential application of primary antibodies, slides underwent incubation with secondary antibodies and tyramide signal amplification. Microwave heat-treated antigen retrieval was applied, and nuclei were stained with 4',6-diamidino-2-phenylindole (DAPI). Multispectral images were obtained by scanning stained slides using the Mantra System (CaseViewer) across the fluorescence excitation spectrum (420–720 nm). In addition, histological staining and immunofluorescence staining of mice brain tissue used CD3 (Anti-CD3 mAb, #ab86883, Abcam), Hematoxylin and Eosin Staining Kit (Reedbio) and TdT-mediated dUTP nick end labeling (TUNEL, #afihc030, AiFang biological) Kit. InForm image analysis software (CaseViewer) performed multispectral unmixing using a spectral library created from auto-fluorescence and single-stained sections, resulting in reconstructed images with auto-fluorescence removed.

Flow cytometry

Cryopreserved single-cell suspensions from GBM tumor tissue (and peripheral blood mononuclear cells (PBMCs) for use as controls) were thawed. In the bipositive SOX2+CD68+ TAMs identification experiment, samples were stained for viability with Zombie NIR Fixable Viability Dye Kit (1:500 dilution, BioLegend, #423101) at 25°C for 20 min in the dark. Samples were then washed in phosphate-buffered saline (PBS) supplemented with 2% fetal bovine serum (FBS), before staining with a master mix of surface antibodies (Alexa Fluor 647 anti-human CD68 Antibody, BioLegend, #333820), and in PBS supplemented with 2% FBS at 25°C for 60 min in the dark. Appropriate antibody concentrations were determined previously by titration. FIX & PERM Cell Permeabilization Kit (Thermo Fisher, #GAS003) are intended for the fixation (Reagent A), permeabilization (Reagent B), and intracellular antibodies (Alexa Fluor 488 anti-SOX2 Antibody, BioLegend, #656110) of the cells. Samples were washed with PBS supplemented with 2% FBS and resuspended in 1× stabilizing fixative for use in flow cytometry.

Tissue dissociation and preparation for single-cell RNA sequencing

The fresh tissues were preserved in the sCelLiVE Tissue Preservation Solution (Singleron) on ice within 30 min post-surgery. Following this, the specimens underwent a triple wash with Hanks Balanced Salt Solution, were minced into small pieces, and subsequently digested with 3 mL of sCelLiVE Tissue Dissociation Solution (Singleron) using the Singleron PythoN Tissue Dissociation System at 37°C for 15 min. The resulting cell suspension was collected and passed through a 40 micron sterile strainer. Next, GEXSCOPE red blood cell lysis buffer (RCLB, Singleron) was introduced to the cell suspension. This mixture, in a ratio of one part cell to two parts RCLB by volume, was incubated at room temperature for 5–8 min to eliminate red blood cells. After incubation, the mixture underwent centrifugation at 300×g at 4°C for 5 min to remove the supernatant. The remaining pellet was gently resuspended in PBS.

Single-cell RNA sequencing and data preprocessing

The single-cell RNA sequencing (scRNA-seq) libraries were prepared with the Chromium Single Cell 5′ Reagent Kit from the Chromium platform (10x Genomics), following the manufacturer’s guidelines. Sequencing was performed on the Illumina HiSeq X Ten platform. Preprocessing of PE150 Illumina sequencing reads was conducted using Cell Ranger software (V.7.0.0). Raw format reads were converted to FASTQ using “cellranger mkfastq”. Subsequently, reads in FASTQ format were aligned to the human genome reference (hg38, GRCh38) using STAR. The “cellranger count” tool was employed to generate gene expression matrices for each sample.

Quality control, dimension-reduction, and clustering

Cells were filtered by gene counts below 200 and the top 2% gene counts and the top 2% unique molecular identifier (UMI) counts. Cells with over 50% mitochondrial content were removed. Following the removal of low-quality and doublet cells, single cells were normalized, 121,637 cells were retained for the downstream dimension reduction and clustering analyses. We applied principal component analyses to reduce the dimensionality of the data using the top 2,000 most variable genes in the data set. Computed principal components were batch corrected for variations between patients using the Harmony R package V.1.0. We used batch-corrected principal componets (PCs) as input for Louvain-based graphing and chose resolution parameters between 0.1 and 1 depending on the single-cell data sets. Seurat was used to identify cluster-specific marker genes and visualization with dot and feature plots.

scRNA-seq-based CNA detection

The InferCNV package was used to detect the copy number alterations (CNAs) in malignant cells (cancer cells). non-malignant cells were used as baselines to estimate the CNAs of malignant cells. Genes expressed in more than 20 cells were sorted based on their loci on each chromosome. The relative expression values were centered to 1, using a 1.5 SD from the residual-normalized expression values as the ceiling. A slide window size of 101 genes was used to smoothen the relative expression on each chromosome, to remove the effect of gene-specific expression.

Pathway enrichment analysis

To explore the potential functions of the differentially expressed genes (DEGs), Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes analyses were conducted using the “clusterProfiler” R package V.4.10.0.13 Pathways with an adjusted p value (p_adj) below 0.05 were deemed significantly enriched. GO gene sets, encompassing molecular function, biological process, and cellular component categories, served as the reference. Gene Set Enrichment Analysis (GSEA) was applied to assess the DEGs between two groups. Additionally, for Gene Set Variation Analysis (GSVA) pathway enrichment analysis, the average gene expression of each cell type was employed as input data, using the GSVA package.

Cell–cell interaction analysis

The R package CellChat V.1.6.114 was employed to quantitatively infer and analyze intercellular communication networks using our scRNAseq data. CellChat uses network analysis and pattern recognition approaches to predict major signaling inputs and outputs for cells, elucidating how cells and signals coordinate for various functions. This tool classifies signaling pathways, outlining both conserved and context-specific pathways through manifold learning and quantitative contrasts. CellChat calculates the communication probability of a ligand-receptor pair between two cell types using a law of mass action model. This calculation takes into account ligand and receptor concentrations, any known cofactor concentrations, and the number of cells in each cell type. The significance of the communication probability is determined by assessing whether it is statistically higher between the known cell types than between randomly permuted groups of cells.

Trajectory analysis

For the classification of a cell as a lymphocyte in the context of trajectory inference, a T cell was required to exhibit at least one UMI for each of the genes protein tyrosine phosphatase receptor type C (PTPRC/ CD45). After excluding B cells and natural killer cells, uniform manifold approximation and projection (UMAP) was reapplied for dimensionality reduction. Post dimensionality reduction for both trajectories, hierarchical clustering was conducted on the UMAP coordinates. The Slingshot V.2.10.0 tool15 was used to fit a minimum spanning tree to these clusters, thereby determining the approximate trajectory structure. This piecewise linear trajectory underwent smoothing via simultaneous principal curves to derive the final trajectories and pseudotime values.

Bulk RNA-seq

The published bulk RNA-seq data of GBM matched survival information was downloaded from UCSC Xena (https://xenabrowser.net/datapages/). Raw reads were processed and quantification and identification of DEGs were performed using DEseq2.

The detection of HLA types

Human leukocyte antigen (HLA) was detected by single-molecule real-time (SMRT) sequencing. A reaction mastermixture was prepared before use, which contained the 10 µL of reaction mix contained 4 µL of PCR product, 5 µmol/L barcoded adaptor, 1×T4 DNA ligase buffer, 1 mmol/L ATP, 200 µmol/L dNTP, 2.5 units of T4 polynucleotide kinase, 0.75 units of T4 DNA polymerase and 180 units of T4 DNA ligase (HC). The purified PCR product (120–250 ng) was mixed with the enzyme mixture. Reaction mixes were then incubated at 37°C for 20 min, 25°C for 15 min, and 65°C for 10 min. Exonucleases I and III were then used to remove failed ligation products, and the pre-library was purified with Ampure PB beads. After pooling, the pre-libraries were purified twice more. The final library was bound with sequencing enzymes and primers using specific Pacific Biosciences kits, loaded with DNA-polymerase complexes, and sequenced on the Sequel II platform for 20 hours. The primary analysis of output data was carried out with SMRT Link V.10.1.0 software. All HLA alleles were named according to the International Society of Blood Transfusion working group that develops and maintains guidelines for blood group antigen and allele nomenclature.

TCR analysis

TCR analysis involved enriching full-length TCR V(D)J segments from 5′ library-amplified complementary DNA via PCR using the Chromium Single-Cell V(D)J Enrichment kit (10x Genomics). Cell Ranger vdj pipeline assembled TCR sequences, identifying CDR3 sequences and rearranged TCR genes. Loupe V(D)J Browser V.2.0.1 (10x Genomics) facilitated analysis. Some T cells exhibited the same TCR with two α and/or two β chains. Four TCR forms (1α1β, 1α2β, 2α1β, and 2α2β) were considered, treating the same TCR form with one class, resulting in a unique TCR name. The TCR diversity metric, encompassing clonotype frequency and barcode information, was derived for analysis.

Grouping of Lymphocyte Interactions with Paratope Hotspots 2 algorithm

The GLIPH2 (Grouping of Lymphocyte Interactions with Paratope Hotspots 2) algorithm functions by grouping lymphocyte interactions based on complementary paratope hotspots.16 It dissects highly diverse TCR sequences into shared specific groups that may recognize the same peptide-major histocompatibility complex ligand. These shared specific groups are established through strong homology within the same amino acid sequence motif or complementarity determining region 3 (CDR3) of the TCRβ chain. Prediction of shared CDR3 sequences using the GLIPH2 algorithm: Amplified TCR or B cell receptor (BCR) sequences are input into the system, and the GLIPH2 algorithm is used for analysis. TCR sequences are grouped and clustered, and shared CDR3 sequences are identified, resulting in the generation of visualized outcomes.

Generation of TCR-T cells

We designed a gRNA to target the first exon of the constant chain of the TCRα gene (TRAC). The sequence targeted is located upstream of the transmembrane domain of the TCRα. This domain is required for the TCRα and β assembly and addressing to the cell surface. Both, non-homologous end joining and integration of the TCR by homology directed repair (HDR) at this locus would then efficiently disrupt the TCR complex. In the process of gene targeting, primary T cells are isolated from peripheral blood by lymphocyte isolation fluid in patients. T-cell activation was initiated, and 48 hours later, the CD3/CD28 beads were magnetically removed. Subsequently, T cells underwent transfection through electrotransfer of Cas9 mRNA and gRNA using the NEPA21 system from the NEPA GENE apparatus. A mixture of 1×106 cells, 1.6 µL(25 pmol/µL) of Cas9, 1.2 µL(100 pmol/µL) of gRNA, and 4 µg of dsDNA (TCR) were electroporated in a Nepa Electroporation Cuvettes (NEPA GENE, #EC-002S). Following electroporation, cells were diluted into the culture medium and incubated at 37°C with 5% CO2. The culture is 2–4 hours post-electroporation (1×105 to 1×106). Transduced T cells were cultured with replaced fresh medium and IL-2 (Recombinant Human Interleukin-2, PrimeGene, GMP-101–02) every 3 days edited cells were then cultured under standard conditions at 37°C and expanded in T-cell growth medium, with regular replenishment to maintain a density of approximately 1×106 cells per ml every 2–3 days. For a more in-depth understanding of targeting constructs and strategies, TRAC gRNA and dsDNA(TCR) sequence, refer to figure 5F.

Co-culture U251 and TCR-T cell in vitro

A total of 6×105 CMV-TCR-T cells/untransduced (UTD) T cells were co-cultured with an equivalent number of mixed U251-GFP-Luci-Vector cells (derived from a human glioma cell line and engineered with a green fluorescent protein and luciferase vector, previously constructed by our laboratory), were infected with CMV (Strain AD169, sourced from the Microbial Virus Strain Bank of Central South University). This co-culture took place at 37°C for 16 hours in Roswell Park Memorial Institute (RPMI) 1640, supplemented with L-glutamine (2 mM), antibiotics (penicillin 100 U/mL, streptomycin 100 µg/mL), and 10% FBS. Brefeldin A (10 µg/mL #420601, BioLegend) was added 4 hours before harvest. Following incubation, cells were washed twice with PBS containing 1% FBS, then stained with CD3 (BV421 Mouse Anti-Human CD3), CD8 (PerCP/Cyanine5.5 anti-human CD8a, BioLegend, #301032) and Zombie NIR Fixable Viability Dye Kit at 4°C for 30 min. After washing and fixation using the FIX & PERM Cell Permeabilization Kit (Thermo Fisher, #GAS003), cells were further stained with interferon (IFN)-γ (APC Mouse Anti-Human IFN-γ Antibody, BioLegend, #506510) for 45 min at 4°C. After two additional washes, cell analysis was conducted using an FACS flow cytometer, and FlowJo software was employed for data analysis.

Intracranial in situ tumorigenic mice models

All animal experiments were approved by the Animal Care and Use Committee of Central South University. U251-GFP-Luci-Vector cells infected with CMV were harvested during the logarithmic growth phase, washed twice with PBS, and then loaded into a 50 µL syringe with a 25-gage needle attachment. Using a stereotactic frame for precision, the tumor cells were implanted 2 mm to the right of bregma, at a depth of 4 mm from the surface of the skull at the coronal suture. Resuspended at 5×105 cells in 5 µL of serum-free medium per mouse. In mouse models, effector cells were systematically infused via tail vein injection in a 100 µL volume. Each mouse in the corresponding groups received an intravenous transplant of 2×107 CMV TCR-T cells or UTD T cells. Simultaneously, interleukin (IL)-2 (1×106 U per mouse) was intraperitoneally injected. Tumor progression was longitudinally monitored by assessing bioluminescence emission using a BLT AniView100 optical imaging system after intraperitoneal substrate injection.

Statistical analysis

In scRNA-seq data, cell distribution comparisons between two groups were evaluated through unpaired two-tailed Wilcoxon rank-sum tests. For comparisons of gene expression or gene signature between two groups of cells, unpaired two-tailed Student’s t-tests were employed. Paired two-tailed Wilcoxon rank-sum tests were used for the analysis of cell distribution within paired cell types. All statistical analyses and data presentations were conducted using R. The specific statistical tests applied in the figures were delineated in the figure legends. Statistical significance was defined at p<0.05.

For statistical analysis of other experimental, data are presented as the mean±SD from at least three separate experiments, and the data were analyzed using GraphPad Prism V.9.5 Differences between the variables of the two groups were tested using the Student’s t-test, and one-way analysis of variance was used to evaluate the differences among variables of multiple groups. Survival analysis was calculated by the Kaplan-Meier method. Values of p<0.05 were considered statistically significant.

ResultsCMV infection induces bipositive CD68+SOX2+ TAMs to remodel GBM immune microenvironment through scRNA-seq

We combined immunogold electron microscopy with droplet digital PCR (ddPCR) to assess CMV viral particles and DNA in fresh postoperative tumor tissue for confirmed 20 patients with GBM (online supplemental figure S1A). The rare electron-dense particles labeled with gold were confirmed that be morphologically consistent with CMV virions (figure 1A); the low viral copy numbers were displayed in 60% of tissues of patients with GBM (12/20) by ddPCR detection (online supplemental figure S1B). The clinical details of these patients are summarized in online supplemental table 1.

Figure 1

CMV infection triggers bipositive TAMs-driven remodeling of the GBM immune microenvironment. (A) Transmission electron microscopy images of CMV-infected GBMs. Scale bars respectively, 5 μm 1 µm 500 nm. The red arrows are pp65 at 10 nm. green arrows are IE1/2 at 4 nm, and the white box shows electron-dense particles. (B) The UMAP above shows the composition of the annotated cells, with different colors representing different cell populations. The below is the CMV infected map, and the dark blue is the CMV infected. (C) The dot plot shows the expression levels of different classical cell marker genes in annotated cell populations. (D) UMAP feature plot representation of marker gene expression within individually identified macrophage and cancer cell populations and phenotypic states. (E) Representative images of multiplex immunofluorescence staining in formalin-fixed paraffin-embedded tissues, indicating CD68+SOX2+ TAMs, in paired CMV+ and CMV− samples. Scale bar, 200 µm and 50 µm. Boxplots illustrating the fraction of CD68+SOX2+ TAMs in CMV− (light blue) and CMV+(deep blue), respectively. Box center lines, bounds of the box, and whiskers indicate medians, first and third quartiles, and minimum and maximum values within 1.5×IQR of the box limits, respectively. Significance was determined using a two-sided, unpaired Wilcoxon rank-sum test relative to CMV+ (n = 12) for CMV− (n=8, p value<0.0001). (F) Representative flow cytometry plots and summary data showing the percentage of cellular staining for CD68+SOX2+ TAMs in CMV− and CMV+ GBM (n=20). CMV, cytomegalovirus; GBM, glioblastomas; TAM, tumor-associated macrophages; UMAP, uniform manifold approximation and projection.

Next, a comprehensive exploration of the functional and transcriptional diversity in CMV-infected GBM tissues was performed by scRNA-seq. After initial quality control and filtering, transcriptomic data were available for 121,637 cells from GBMs including CMV-infected GBMs (CMV+, n=6) and CMV-uninfected GBMs (CMV−, n=3). Using known lineage markers, we identified seven major cell populations, including cancer cells, macrophages, stromal cells, oligodendrocytes, endothelial cells, and T cells (figure 1B and online supplemental table 2). In the GBM landscape, we observed chromosome 7 amplification and chromosome 10 loss specifically in glioma cells, which was consistent with published whole-exome sequencing data for gliomas (online supplemental figure S1C). We found a novel cell type: bipositive TAMs (Bipositive TAMs) co-expressing markers of both macrophages and tumor cells, which were relatively specific in patients with CMV+ GBM (figure 1C, online supplemental figure S1D). We excluded the effect of common molecular features on cellular compartmentalization (eg, TP53 mutation and MGMT methylation status) (online supplemental figure S1E,F). We scored our single-cell data using the gene set (COATES_MACROPHAGE_M1_VS_M2_UP) from GSEA, and the results showed that bipositive TAMs were more biased towards M1-type macrophages (online supplemental figure S2A). However, the polarization of the characterized genetically analyzed macrophages was not obvious (online supplemental figure S2B). By analyzing the expression levels of classical markers of macrophages and tumor cells, we found that CD68 and SOX2 were more significantly expressed in bipositive TAMs (figure 1D). Intriguingly, immunofluorescence and flow cytometry detection indicated that GBM tissues infected with CMV are enriched with bipositive CD68+SOX2+ TAMs (figure 1E,F, online supplemental figure S2C), suggesting that CD68 and SOX2 serve as a reliable marker for bipositive TAMs. The functional analysis revealed that differentially expressed genes in bipositive TAMs and T cells were predominantly associated with human CMV infection and viral protein interaction with cytokine and cytokine receptor pathways (online supplemental figure S2D,E). These results suggested that CMV infection remodels the GBM immune microenvironment through bipositive CD68+SOX2+ TAMs.

Bipositive TAMs are major contributors to the clinical outcome of GBMs

To better understand the functional characteristics of bipositive TAMs, we conducted a meticulous re-annotation analysis of the published scRNA-seq data from glioma tumor tissue, which encompassed 44 segments obtained from 18 patients with glioma, comprising 2 LGG, 11 ndGBM, and 5 rGBM. This data set comprehensively covered eight major cell types, namely cancer cells macrophages, oligodendrocytes, endothelial, stromal cells, and so on. Bipositive TAMs were reproducibly found in these patients with glioma (figure 2A,B). Functional analysis revealed that the differential genes of bipositive TAMs were mainly enriched in the human CMV infection pathway (figure 2C, online supplemental figure S2F).

Figure 2

Bipositive TAMs significantly influence the clinical outcomes of GBMs. (A) UMAP plot of 236,575 cells from tumor tissue samples of patients with glioma from GSE182109 data. Each cluster is shown in a different color. (B) Violin plots showing the expression levels of different classical cell marker genes in the 8 cell clusters. (C) Radargram demonstrating Gene Ontology analysis associated with infection by pathogenic organisms. (D) The bar graph shows the percentage of each cell population in each sample in single-cell RNA sequencing (E) Box plot showing the percentage of different cell populations in low-grade gliomas (LGG), newly diagnosed glioblastomas (ndGBM), and recurrent glioblastomas (rGBM) (F) Comparison of the signatures of each cell type gene multivariable Cox regression was used to obtain the HRs (with Wald 95% CIs shown as horizontal bars, and p values given on the right) based on the cross-validated prognostic scores derived using the Cox model and applied to pairwise differences of expression of the genes. (G) Kaplan-Meier plot of cross-validated macrophage prognostic score from (F) at 50% cut-off, showing the genes with opposite effects selected by the model. GBM, glioblastomas; RNA-seq, RNA sequencing; TAMs, tumor-associated macrophages; t-SNE, t-distributed stochastic neighbor embedding; UMAP, uniform manifold approximation and projection.

To assess the correlation of bipositive TAMs with the clinical progression of patients with glioma, we analyzed the percentage of type of cells in the tumor tissue. Analysis of the percentage of each cell type in patient tumor tissue indicated that we observed more pronounced heterogeneity in cell composition in ndGBM (figure 2D). In comparison to ndGBM, rGBM displays lower spatial heterogeneity and a less complex array of cellular components within the tumor region. Notably, we found bipositive TAMs were prevalent at the disease stage of glioma progression and particularly prominent in patients with GBM (figure 2E, online supplemental figure S2G). Bipositive TAMs potentially originate from the fusion of macrophages engulfing tumor cells in the early stage of disease progression (LGG), and the number of cells is relatively more, but bipositive TAM cells may form tumor cells with the progression of the disease to late stage (GBM). We sought to determine which cells carried the most prognostic information, only bipositive TAMs signatures displayed to be an independent prognostic factor by multivariate analyses (figure 2F). Kaplan-Meier survival analyses showed that the more the bipositive TAMs infiltration, the shorter were overall survival of patients with GBM (figure 2G). These results suggested that bipositive TAMs might act as a pointer of GBM progression. Taken together, these data collectively suggest that bipositive TAMs, which are associated with CMV infection gene expression programs, played a significant role in shaping the progression of GBM.

CMV infection enhances a novel immunosuppressive FXYD6+ T cell and facilitates GBM progression

To decipher how bipositive TAMs drive the tumor progression, we used a public repository of ligand-receptor interactions, CellChat V.1.6.0, to infer interactions between bipositive TAMs, cancer cells, and all T populations in our scRNA-seq data. First, seven lymphocyte subpopulations were classified by differential gene expression and examination of known lineage markers (figure 3A, online supplemental figure S3A), in which five T-cell clusters across tumor samples were identified, each with unique signature genes (online supplemental figure S3B). Specifically, the central memory T cells were characterized by high expression of the IL-7R and CD40LG genes. The T regulatory cells highly expressed the genes encoding inhibitory FOXP3, CTLA-4, and IL-2R, while effector T cells demonstrated high expression of cytotoxic cytokines such as CD8A/B, GZMH, and NKG7 genes (figure 3B).

Figure 3

CMV infection amplifies a newly identified immunosuppressive FXYD6+ T cell, promoting the progression of GBM. (A) Distribution and expression of representative genes in lymphocyte clusters. (B) Violin plots showing the expression levels of different classical lymphocyte marker genes in the seven cell clusters. (C) Gene network diagram showing functional analysis of FXYD6+ T cells characterized genes. (D) Gene function network diagram demonstrating the analysis of the major functions of the FXYD6+ T cells signature gene cluster. (E) Over-represented GO terms of FXYD6+ T cells (pink) compared with other T cells (violet). (F) The inferred MDK signaling networks between cancer cells/ bipositive TAMs and T-cell clusters. (G) Expression score of the FXYD6+ T cells gene sets of non-responder and responder after neoadjuvant anti-PD-1 therapy. (H) Representative example of CMV− (G15) and CMV+ (G3) GBM tumors. Tumor staining by multiplexed immunofluorescence shows the spatial distributions of FXYD6+ T cells, KI67+ cancer cells (tumor cell proliferation), and TNF-β (T-cell cytotoxicity). (I) Circos plots showing well-known ligand-receptor pairs from CellChat databases under the requirement that either the ligand or the receptor (or both) both being expressed. Arrows are pointing from the ligand toward the receptors. Different colors represent different cells, and the width is proportional to the number of events. Only selected example pairs are labeled and highlighted for the network. P values are adjusted for the number of cell types. CMV, cytomegalovirus; GBM, glioblastomas; PD-1, programmed death 1; TAM, tumor-associated macrophages; UMAP, uniform manifold approximation and projection.

Thrillingly, we discovered a cluster of T cells that was characterized by high expression of FXYD6, LSAMP, MEG3, and NCAM1 genes. We define it as FXYD6+ T cells, detailed FXYD6+ T cells characteristic genes are shown in online supplemental table 3. Compared with other T cells, FXYD6+ T cells highly expressed the related genes of cell cycling, gliogenesis development cell growth, and differentiation, such as MDK, ASCL1, SOX10, and SOX11 (figure 3C, online supplemental figure S3C), GO terms enriched in FXYD6+ T cells were associated with regulation of microtubule polymerization or depolymerization and regulation of nervous system development (figure 3D). These results suggested that FXYD6+ T cells supported the gain-promoting gliogenesis, which was different from the reported proliferating T cells; meanwhile, FXYD6+ T cells the majority of genes defining the GO term “positive regulation of immune system process” and “T cell mediated cytotoxicity” was downregulated in FXYD6+ T cells compared with other T cells (figure 3E), which suggested FXYD6+ T cells induce immunosuppression. It is not characterized by high expression of corresponding immunosuppressive molecules, like PDCD1, CTLA-4, TIGIT, LAG3, and BTLA, etc (online supplemental figure S3D), so, FXYD6+ T cells are a group of T cells which is different from the classical exhausted T cells. In summary, FXYD6+ T cells are a group of novel immunosuppressive T cells.

Subsequently, we performed the analysis of the interactions between bipositive TAMs, cancer cells, and all T populations in our scRNA-seq data. Compared with the uninfected group, we found that CMV infection altered interactions between bipositive TAMs, cancer cells, and T cells, primarily mediated via the MDK (Midkine) pathway (online supplemental figure S3E), in which either the cell–cell interactions of cancer cells and T cells or bipositive TAMs and T cells mainly focus on FXYD6+ T cells (figure 3F). We next analyzed data from published spatial transcriptome.17 Analysis of the co-localization FXYD6+ T cells signature in spatial transcriptome data showed increased FXYD6+ T cells signatures in anti-PD-1/anti-PD-L1 treatment no responder patients with GBM (figure 3G). Multiplex IHC showed that FXYD6+CD3+ cells were numerically higher in patient with CMV+ GBM samples than in patient with CMV− GBM samples. Meanwhile, patients with concurrent CMV infection showed more tumor cell proliferation and less T-cell toxicity (figure 3H, online supplemental figure S3F). In addition, we found more clues about interactions between FXYD6+ T cells and cancer cells. We detected the neurotrophic factors MDK and pleiotrophin (PTN) in cancer cells. PTN and MDK are involved in inducing and stimulating neuronal differentiation. Receptor protein tyrosine phosphatase type Z 1 (PTPRZ1) and nucleolin (NCL), a shared receptor of MDK and PTN, were highly expressed in FXYD6+ T cells (figure 3I). The binding of MDK to PTPRZ1 is important for the MDK-dependent survival of embryonic neurons.18 The interaction of PTN with PTPRZ1 is known to play an important role in cell–cell adhesion, cell motility, migration, cell division, and promoted glioblastoma stem cells (GSC) proliferation and self-renewal.19 In short, in GBM with CMV infection, the interactions of cancer cells and FXYD6+ T cells via ligand-receptor binding MDK and PTN with PTPRZ1/NCL mediated the tumor immunosuppressive microenvironment and promoted GBM progression.

CMV infection increased T-cell exhaustion in GBM

To better understand the changes in T cells in patients with CMV+ GBM, each type of T cell was found in individual patient tissues (figure 4A,B). We found that most T cells in CMV-infected GBM tumors were IL-7R+ T cells (46.3%), while CMV-uninfected were GZMH+ T cells (57.7%) (figure 4C,D). We used Slingshot for trajectory inference with markers PTPRC/CD45, CD3D, and CD3E (online supplemental figure S4A). The analysis produced a two-lineage trajectory, with one terminal point to FOXP3+ T cells and another to FXYD6+ T cells. We first observed that the position of individual cells along the trajectory (ie, in pseudotime) varied largely according to their CMV infection. T cells from CMV− tissue localized almost entirely in the early portion of the trajectory, while CMV-infected tumors were heavily skewed toward the ends of the trajectory (online supplemental figure S4B). We have referred to the article published by Carrasco et al20 to analyze the hallmarks of T-cell aging in our data to validate the accuracy of the starting/ending point of the trajectory (see online supplemental figure S4C). Through differential gene expression analysis, numerous inhibitory checkpoints had increased expression in T cells from CMV+ GBM tumors compared with the CMV− GBM. Their expression, including PDCD1/PD-1, HAVCR2/TIM-3, and LAG3, was substantially increased late in pseudotime (figure 4E,F). Their expression levels which are associated with a more terminally exhausted phenotype were consistently low throughout pseudotime (figure 4G), potentially explaining the low efficacy of immune checkpoint inhibitors in GBM. The trajectory analysis indicates that T cells were predominantly enriched in the later stage of the trajectory in CMV+ GBM, suggesting that CMV infection drives T-cell exhaustion, and potentially interferes with the response to inhibitor treatment.

Figure 4

T cells profile in glioblastomas with CMV infection. (A) The UMAP shows the composition of the T cells, with different colors representing different T-cell populations. (B) UMAP representation from colored based on patient ID. (C) The bar shows the percentage of T-cell subsets comparing CMV-infected and CMV-uninfected groups. (D) Bar graph showing the percentage of T-cell subsets in patients. (E) Distribution and expression of exhaustion-related genes in T cells in the CMV-uninfected group. (F) Distribution and expression of exhaustion-related genes in T cells in the CMV-infected group. (G) Slingshot trajectory analysis demonstrates the average expression pattern of each gene and signature across pseudotime (scaled from minimum to maximum average expression for each gene). CMV, cytomegalovirus; UMAP, uniform manifold approximation and projection.

Identifying clonally expanded TCRs in patients with GBM with CMV infection

We further explored whether T cells could generate specific activated TCR in CMV-infected GBM tumor microenvironment. We investigated the TCR of T cells in CMV+ GBM and uninfected tissue by performing single-cell TCR/BCR sequencing (scTCR/BCR-seq) on the cells for which we had transcriptome data, enabling a linkage between clone type and phenotype (figure 5A). We sought to determine the antigens that drive the clonal expansion of T cells in the tissue of patients with CMV infection (figure 5B). Conspicuously, patients with CMV+ GBM have more TCR clones, especially in CD8+ and GZMH+ T cells (online supplemental figure S5A,B). We next compared CMV infected TCR clone type with the uninfected (defined as the percentage of total TCRαβ sequences that are identical to one or more TCRαβ sequences), which revealed that the levels of TRBV4-3 were higher in CMV-infected GBMs (figure 5C,