MAPK-driven glioma progression and reprogramming of the tumor-associated immune response

Abstract

Background

In brain tumors, the immunosuppressive microenvironment leads to tumor aggressiveness and immunotherapy resistance. To identify factors that regulate the glioma-immune landscape, we took advantage of the disease trajectory of the MAPK-driven glioma, pleomorphic xanthoastrocytoma (PXA), which evolves from an immune-rich low-grade tumor to an aggressive glioma.

Methods

Using a multiomics approach that included single-nucleus RNA sequencing, spatial transcriptomics, and spatial proteomics, a cohort of 38 tumors, including 10 longitudinal pairs, was profiled.

Results

Tumor progression was associated with transition to a more proliferative progenitor state and increased hypoxia. Hypoxic reprogramming of the immune landscape included increased TGF-β signaling and altered chemokine signaling, including decreased CXC-chemokine ligand 14 (CXCL14) and CXCL16-CXCR6 signaling important in immune cell recruitment and activation. The resulting immune landscape was spatially reorganized with reduced HLA class II expression, reduced CD8+ and CD4+ T cells, and dominated by immunosuppressive myeloid cells.

Conclusions

Hypoxia and chemokine signaling emerged as key components of the glioma-immune landscape that evolve with tumor progression. These findings identify potential therapeutic opportunities in PXA, and they argue that optimal immunomodulatory strategies in glioma will differ along the disease trajectory.

Graphical Abstract

Graphical Abstract.

Key Points.

• MAPK-altered pleomorphic xanthoastrocytoma and its immune microenvironment coevolve.

• Hypoxia is accompanied by altered immune composition, spatial organization, and cell signaling.

• Transition to an immunosuppressive myeloid phenotype at progression.

Importance of the Study

Integrating multiomic data from an immune-rich MAPK-driven glioma, we uncover coordinated evolution of the tumor and immune microenvironment. With glioma progression, increased tumor hypoxia and decreased chemokine signaling were associated with altered immune cell composition and spatial organization, including decreased antigen-presenting cells, decreased CD8 T cells, and increased immunosuppressive myeloid cells. These data highlight the dynamic interplay between tumor, vasculature, and immune elements in glioma and identify factors associated with immunosuppressive reprogramming of the immune landscape. In pleomorphic xanthoastrocytoma, they argue that optimal immunomodulatory strategies may differ along the disease trajectory.

Glioma progression is driven by the coevolution of the tumor and its microenvironment, including the tumor-associated immune response.^1-3 Within glioma molecular subtypes, characteristic patterns of tumor cell evolution have been identified with clonal selection of mutations, copy number alterations, and epigenetic states that are advantageous to the tumor.³ How the glioma-immune landscape evolves, and the factors that drive these changes are less clear.

Molecularly distinct glioma, defined by specific genetic and epigenetic alterations, demonstrates significant differences in the recruitment, composition, and function of tumor-associated immune cells.^4-7 For instance, isocitrate dehydrogenase (IDH)-mutant gliomas are associated with an immunosuppressive microenvironment characterized by few infiltrating lymphocytes. Several mechanisms may contribute to immune evasion, including reduced tumor cell expression of interferon-gamma-inducible chemokines, including CXC-chemokine ligand 10 (CXCL10).⁸ In contrast, pleomorphic xanthoastrocytoma (PXA), an MAPK-driven glioma with oncogenic RAF kinase signaling and CDKN2A biallelic inactivation, exhibits a CD8 T cell-rich phenotype.^9,10 This immune-enriched phenotype was associated with the secretion of CXCL14 by tumor cells, and, in a murine glioma model, CXCL14 was sufficient to drive an antitumor CD8 T cell response and prolong survival.

Therapeutic targeting of the glioma-associated immune response has emerged as a promising strategy to slow disease progression and improve outcomes.^11,12 Eliciting ­antitumor immunity, however, depends on multiple coordinated processes, including efficient antigen presentation by myeloid cells, T cell recruitment, and sustained effector cell activation within the tumor microenvironment (TME). Dysregulation in any of these processes can profoundly influence whether an immune response leads to tumor elimination or immune evasion. Just as tumor molecular characteristics help shape the immune microenvironment, they also influence responsiveness to immunomodulation such as checkpoint blockade.¹³ Alterations in cytokine expression and antigen presentation can shape both baseline immune cell infiltration and dynamic responses to therapy.¹³ For example, in GBM, responsiveness to PD-1 blockade was associated with mutations that activate the MAPK pathway, suggesting a link between oncogenic signaling and T cell-mediated antitumor activity.^13-15

Successful immunomodulation of the immune microenvironment in glioma requires an understanding of the factors driving immunosuppression and their temporal and spatial regulation. To investigate this, we took advantage of the disease trajectory of MAPK-driven PXA, which evolves over years from a low-grade, immune-enriched tumor to an aggressive, high-grade disease.^16,17 A large cohort of 38 PXA tumors, including 10 longitudinal pairs, was profiled using a multiomics approach that included single-nucleus RNA sequencing (snRNA-seq), targeted gene expression profiling, spatial transcriptomics, and multiplexed single-cell imaging. Through this integrated approach, we demonstrate a coevolution of both tumor and its microenvironment, characterized by dynamic changes in cellular composition, phenotype, and cellular crosstalk, with implications for the optimal timing of immunotherapy.

Methods

Patient Samples and Inclusion Criteria

A total of 38 PXA cases (mean age 27.9 years), diagnosed per the 2021 World Health Organization (WHO) Classification of Tumors of the Central Nervous System (CNS) , were identified from records in the UCSF Brain Tumor Center Biorepository at UCSF, see Supplementary Material S1. The samples, patient demographics, clinical data, clinical molecular profiling, assays performed, and inclusion in prior publications^10,17,18 are specified in Table S1 and  Supplementary Material S1. The ethics approval number for the use of deidentified human biospecimens is #10-01318. These studies were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments.

Image Acquisition and Analysis

Tumor blocks for analyses were selected based on tumor content ≥70%. Images of immunostained slides were acquired at 200× magnification using a Zeiss Axioscan 7 (Zeiss). Digital quantification of whole slide images or 3-5 representative regions per tissue block was performed with FIJI software (RRID: SCR_002285). Identical exposure times and threshold settings were used for each channel on all sections of similar experiments, see Supplementary Material S1.

snRNA-seq Sample Preparation, Data Acquisition, Preprocessing, and Data Analysis for Cell Type Clustering, Annotation, and Pathway Analysis

Flash-frozen tumor tissue for snRNA-seq analysis was selected based on the demonstration of >70% viable tumor nuclei in adjacent H&E-stained sections. Approximately 15 000 nuclei were loaded per capture, and snRNA-seq was performed using the Chromium Next GEM Single Cell 3ʹ Reagent Kits v3.1 (Dual Index; 10× Genomics). Data preprocessing and downstream analyses were performed in Seurat (v5.1.0), see Supplementary Material S1.

Gene set enrichment analysis of Gene Ontology (GO) terms was performed using the gseGO function in the clusterProfiler package (v3.0.4). Pseudobulk analysis used the AggregateExpression() function from the Seurat package based on grade status. Cell-cell communication between tumor and other cell types was analyzed using the CellChat package (v2.1.2).

Multiplex Immunophenotyping Via Multiplexed Ion Beam Imaging-TOF

The patient PXA tumors, data acquisition and analyses, quality control, cell clustering, and annotation have been described previously¹⁹ (see also Supplementary Material S1). Statistical significance was defined as P-value <.05 by paired or unpaired Student’s t-test.

Quantitative InterCellular Niche Enrichment Analysis

Quantitative InterCellular Niche Enrichment (QUICHE) analysis was performed to identify differentially enriched niche neighborhoods across conditions (PXA G2 vs G3)^19,20 using the Python package at https://github.com/angelolab/publications/tree/main/2024-Ranek_etal_QUICHE. In brief, we identified cellular niches according to spatial proximity within each image (radius = 200 pixels), performed distribution-focused downsampling to select a subset of niches per patient sample, constructed a k-nearest neighbor (=50) graph for differential neighborhood enrichment testing, and labeled niche neighborhoods according to the top 3 most abundant cell types. The downsampling sketch size was determined to be within 1 SD below the mean and adjusted depending on the spread of cells per field of view. Niche neighborhoods were considered significant if they had a spatial false discovery rate <0.20.^19,20

Results

Evolution of MAPK-Driven Glioma and Its Cellular Microenvironment With Progression

To investigate the temporal evolution of tumor-immune interactions in MAPK-driven glioma, we analyzed the transcriptional and proteomic profiles of tumor and immune cells using snRNA-seq and high-resolution spatial mapping using targeted protein probe-based profiling, single-gene RNA in situ hybridization, and multiplex immunophenotyping via multiplexed ion beam imaging by time of flight (MIBI-TOF) and multiplex immunostaining (Figure 1A). Additionally, spatial transcriptomic analysis was performed to assess the spatial relationships between tumor cells and diverse immune cell populations. A total of 38 MAPK-driven glioma diagnosed per the fifth edition of the WHO Classification of Tumors of the CNS as PXA were profiled, including 14 WHO CNS5 grade 2 (G2) and 24 WHO CNS5 grade 3 (G3) tumors (Table S1 and  Supplementary Material S1). Among these, 10 patients had longitudinal samples collected from 2 or more surgical timepoints.

Figure 1.

Coevolution of the MAPK-driven glioma PXA and its microenvironment. (A) Schema depicting the experimental workflow to evaluate the temporal evolution of the transcriptomic and proteomic landscape of PXA WHO CNS grade 2 and grade 3 (hereafter referred to as G2 and G3). (B) UMAP visualization of unbiased clustering analysis of 36 781 cells in PXA (n = 10). The clusters are color-coded, and each dot represents a cell. (C) Dot plot illustrating the expression of top marker genes across different cell types. Dot size represents the percentage of cells expressing the gene, while color intensity denotes the expression level. (D) UMAP plot depicting the overlay clusters for neoplastic cells in G2 (blue) and G3 (red) PXA (n = 10). (E) Bar plot showing enriched pathways from pseudobulk DE genes in neoplastic cells from G2 (blue) and G3 (red) PXA (n = 10). Selected significant pathways are depicted with a positive normalized enrichment score (NES) for G2 and a negative NES for G3. (F) Dot plot showing expression of select genes in tumor cells as a function of grade. Dot size represents the percentage of cells expressing the gene, while color intensity denotes the expression level. (G) Bar plot showing the proportion of cell cycle phase in G2 and G3. The S and G2M phases were calculated using the “Cell Cycle Scoring” function, with segregated list markers for G2/M phase and the S phase. (H) Dot plot showing representative cell cycle genes related to the replication phase. Dot size represents the percentage of cells expressing the gene, while color intensity denotes the expression level. (I) (left) Bar plot showing Ki67 protein expression in a PXA cohort (n = 18) and (right) representative image. Scale bar in the image denotes 200 µm. Abbreviation: PXA, pleomorphic xanthoastrocytoma. (A) Created in BioRender. Hashemi, E. (2026) https://BioRender.com/crk3ymp.

For snRNA-seq, nuclei were isolated from 10 PXA (n = 4 G2 and n = 6 G3), including 3 patients with longitudinally paired specimens over their disease trajectory, and profiled using the 10× Genomics platform. High-quality nuclei were selected based on mitochondrial gene content and read count thresholds. A total of 36 781 cells were captured. Unsupervised clustering identified 7 distinct cell populations present in both G2 and G3 tumors (Figure 1B). Neoplastic cells were distinguished from nonmalignant neuroglial, endothelial, and immune populations using previously described methods.¹ These approaches incorporate both gene-signature profiling and inferred copy number alterations from snRNA-seq data (Figure 1C, Figure S1A). For example, T cell clusters expressed CD3E, CD3G, and granzymes (GZMs); monocyte clusters expressed CSF1R and major histocompatibility complex (MHC) class II genes; oligodendrocyte clusters expressed MOG, MAG, and PLP1; astrocytes expressed GFAP and AQP4; and putative neoplastic clusters expressed EGR1, SOX2, SOX6, and CDK6. The immune clusters comprised distinct populations and were the focus of our study. As expected, putative neoplastic clusters shared expression of some genes with nonneoplastic neuroglial populations; however, the putative neoplastic clusters exhibited frequent CNAs (Figure S1A). While cell-type frequencies, including immune cells, differed across samples and between tumor grades, the differences were not significant (Figure S1B-D).

With disease progression to G3, the transcriptomic profile of putative neoplastic cell clusters changed (Figure 1). Pseudobulk analysis of differentially expressed genes (DEGs) demonstrated that G2 tumors were enriched for pathways associated with glial cell and oligodendrocyte differentiation (Figure 1E). Conversely, G3 tumors showed enrichment for pathways related to angiogenesis, response to hypoxia, DNA replication, and telomere maintenance. Expression of markers associated with a mesenchymal phenotype (CD44) and a progenitor cell state (nestin and Sox2) increased in G3 tumors (Figure 1F). In GBM, transitions among 4 coexisting neoplastic cellular states have been described.²¹ Assigning each putative PXA tumor cell to a dominant cell state based on signature enrichment identified 4 cellular states in each tumor, with proliferation predicted to be predominantly in astrocyte-like and mesenchymal-like cells (Figure S1E-G). However, the distribution of these states did not consistently differ between G2 and G3. In silico cell cycle phase analysis across all tumor cells suggested an expanded proportion of tumor cells in G2/M and S phases in G3 vs G2 tumors (Figure 1G). Consistent with these findings, G3 tumors had higher expression of cell cycle-associated genes, including PCNA, TOP2A, MCM6, and MKI67 (Figure 1H) and increased proliferation (n = 18 tumors) (Figure 1I). These data suggest that with disease progression, tumor cells transition to a more proliferative progenitor state with an altered transcriptional profile.

Decreased Antigen Presentation and Increased Immunosuppressive Myeloid Phenotype With Progression

Antigen presentation by tumor-associated myeloid cells (subsequently referred to as TAMs), including macrophages and microglia, is critical for an effective antiglioma-immune response.²² Using protein multiplexed single-cell imaging in a cohort of 18 PXA, progression to G3 was associated with a striking reduction in antigen-presenting cells, as denoted by decreased MHC class II protein expression on myeloid cells (P = .008) (Figure 2A-C). To investigate gene expression changes associated with this reduction, we used snRNA-seq. Pseudobulk analysis of TAM DEGs demonstrated reduced MHC class II complex assembly and antigen processing and presentation pathways in G3 tumors (Figure 2D). While expression was greatest in TAMs, a subset of tumor cells expressed MHC class II, and this expression was also decreased with G3 progression (Figure S2A and B). In contrast, MHC class I was expressed widely across cell types and was more variable with progression (Figure S2C and D). Enriched pathways in G3 TAMs included those involved in cellular response to hypoxia, glucose starvation, and transforming growth factor beta (TGFB) stimulus (Figure 2D). In addition, CD47 expression on tumor cells was increased in G3 tumors (Figure S2E), (P = 1.03 × 10⁻¹³). CD47 is known to engage with its receptor SIRPα on macrophages, delivering inhibitory signals that suppress tumor cell phagocytosis.²³

Figure 2.

Reprogramming of the TAM population with disease progression. (A) Representative images of HLADR⁺ TAMs in G2 and G3 tumors. Scale bar in image denotes 100 µm. (B) Bar plot showing quantification for HLADR protein intensity on TAMs in G2 (n = 5) and G3 (n = 13) PXA. (C) Dot plot showing HLA class II gene expression in TAMs in G2 and G3 tumors. Dot size represents the percentage of cells expressing the gene, while color intensity denotes expression level. (D) Bar plot showing enriched pathways from pseudobulk DE genes in TAMs from G2 and G3 PXA (n = 10). Selected significant pathways are depicted with a positive normalized enrichment score (NES) for G2 and a negative NES for G3. (E) UMAP plot showing unbiased clustering of the TAM population in PXA (n = 10), where each dot represents a single cell and colors denote TAM populations (blue = C1, red = C2, purple = C3, green = C4). (F) Bar plots showing the percentage of each TAM population over total; color code for each cluster, and (G) for cluster#2 in G2 and G3. Dots represent the samples. Statistical significance was assessed using Student’s t-test. (H) Heatmap showing the top 10 differentially expressed genes for the 4 different TAM clusters. Gene expression is denoted with red higher and blue lower, and the top color bar denotes cluster#. (I) UMAP plot showing the overlayed macrophage and microglia signature genes onto TAMs (macrophage = yellow; microglia = purple). (J) The relative expression of microglial signature genes on TAM clusters. (K) Identity signature gene programs expressed in TAM clusters in PXA. Dot size denotes the percentage of cells expressing the signature, and color intensity indicates the level of expression. Abbreviation: PXA, pleomorphic xanthoastrocytoma.

To better understand TAM population shifts, and further characterize their heterogeneity, TAM cells from the initial clustering were reclustered at a higher resolution and 4 unique clusters were identified (Figure 2E). With G3 progression, there was a significant increase in immunosuppressive TAMs (TAM cluster #2) (Figure 2F and G), characterized by increased immunosuppressive markers, including MRC1 (mannose receptor C-type 1; CD206), C5aR1 (complement C5a receptor 1), NRP1 (Neuropilin-1), and HIF1A, denoted by the significant differentially expressed genes within each cluster (Figure 2H, Table S2). In contrast, myeloid clusters associated with high antigen presentation and processing (TAM cluster #1) and inflammatory microglia (TAM cluster #3) trended down with progression. TAM cluster #1 was enriched in HLA class II genes, including HLA-DRB1 and HLA-DPA1, FCGR3A (CD16), and CD14. TAM cluster #3 had high expression of microglia and microglia-associated genes (P2RY12, CSF1R, CX3CR1, TREM2, and A2M). TAM cluster #4 represented a small subpopulation in our dataset and was characterized by ribosomal and mitochondrial transcribed genes.

TAM evolution may reflect alterations in cellular ontogeny, cellular activation programs, or both. Using exemplary gene signatures (Tables S3 and S4), the total number of TAMs, total number of microglia (predominantly within TAM cluster #3), and total number of macrophages (predominantly within TAM clusters #1, 2, and 4) were not significantly different across grades (Figure 2I and J, Figure S2F and G). However, applying consensus cell identity programs²⁴ to further define cell subsets, the signatures for macrophages, monocytes, and neutrophil programs were enriched, while those for microglia and classical dendritic cells (cDCs) trended down with G3 progression (Figure 2K). MHC class II was most highly expressed in cells expressing canonical macrophage markers and in cells within cluster #1 (Figure S2H-J). Using canonical DC markers from Azimuth²⁵ (Table S5), predicted DC-like cells, partially overlapping the HLA class II-high TAM cluster #1, decreased with tumor progression (Figure S3A and B). Moreover, immunoprofiling revealed a significant reduction in CD11c protein expression in G3 tumors (Figure S3C). Using canonical gene signatures (Table S6) all 3 major DC populations (cDC1, cDC2, and pDC) were enriched in MHC class II-high TAM cluster #1 and decreased with G3 progression (Figure S3D and E). Overall, local TAM proliferation was low (Figure S3F).

Myeloid cell activity programs may help identify functionally relevant myeloid subsets.²⁴ With G3 progression, the scavenger- and complement-immunosuppressive programs, observed primarily in the immunosuppressive TAM cluster #2, were enriched (Figure 3A, Figure S4A). Immunoprofiling of TAMs for CD163, a canonical protein marker of immunosuppressive myeloid cells, demonstrated an increase in G3 tumors, including in 4 patients with paired longitudinal samples (Figure 3B-D), despite similar Iba1+ TAM cells (Figure S4B). Taken together, these data suggest G3 progression is accompanied by significant alterations in TAM activity and reduced HLA class II expression with more subtle alterations in cellular ontogeny. In support of this, pseudotime analysis predicted a transition over time from an HLA class II-high TAM population dominated by cluster #1 to an immunosuppressive TAM population dominated by cluster #2 (Figure 3E).

Figure 3.

Enhanced hypoxia and immunosuppressive TAM remodeling in high-grade tumors. (A) Activity signature programs² expressed in G2 and G3 tumors in PXA. (B) Bar plot showing quantification for dual Iba+CD163+ cells over all cells in G2 (n = 6) and G3 (n = 8), (C), and for PXA paired samples (n = 4 patients). (D) Representative images of multiplex immunostaining showing Iba+ (green) and CD163+ (violet) cells in G2 and G3 tumors. Arrowhead in the inset denotes dual-positive cells. Scale bar within inset denotes 50 µm. (E) Pseudotime analysis of the TAM clusters using the Monocle3 package. The line indicates the direction of developmental progression. Color represents the temporal trajectory of the cells. (F) Dot plot showing selected significant GO terms from pathway analysis on pseudobulk DE genes in G2 and G3 tumors across all cells (n = 10 tumors). Dot size represents gene-signature set size, and NES = Normalized Enrichment Scores. (G) Violin plot showing Hallmark hypoxia gene-signature score in G2 and G3 tumors. (H) Feature plots indicating hypoxia gene-signature differences between G2 and G3 tumors. The more intense color denotes higher expression of signature genes. (I) Dot plot illustrating the expression of hypoxia signature genes across different cell types in PXA. (J) Violin plot depicting HIF1A target gene signatures in G2 and G3 PXA. (K) Expression of the Hallmark hypoxia signature score on TAM clusters. (In panels A, I, and K, the size of the dots shows the percentage of cells expressing signatures, and the intensity of color indicates the level of expression.) Abbreviations: NES, normalized enrichment scores; PXA, pleomorphic xanthoastrocytoma.

Increase in Tumor Hypoxia and Role for Tumor Vasculature in TME Crosstalk

Given the importance of cues from the TME in shaping glioma-associated myeloid cell identity and activity,²⁴ we investigated global shifts in gene expression across all cells using pseudobulk analysis of DEGs. Cellular response to hypoxia, including angiogenesis and response to oxygen levels, emerged as the most enriched pathways in G3 tumors (Figure 3F). A hypoxia signature score, as determined by the calculated Hallmark hypoxia module score for each cell, was significantly higher in G3 compared to G2 tumors (P<2.2e-16) (Figure 3G and H, Table S7). Notably, this hypoxia signature was prominent in nearly all cell types within the tumor ecosystem (Figure 3I). G3 tumors also had elevated expression of HIF1A, a central regulator of the hypoxic response (Figure S4C). Moreover, a canonical HIF1A target gene signature,²⁶ containing 100 HIF1A target genes, was present in multiple cell types and was increased in G3 tumors (Figure 3J, Figure S4D, Table S8). Notably, an HIF1A target gene signature excluding the overlapping hypoxia signature genes was increased in G3 tumors (Figure S4E and F). Among the TAM clusters, the immunosuppressive TAM cluster #2 was most enriched for the hypoxia signature gene score (Figure 3K).

Given the potential impact of hypoxia on cell-cell communication in the TME, we used CellChat to infer how intercellular communication networks may change with tumor progression. Computing the aggregated cell-cell interactions, neoplastic cells and vascular cells had the highest number of interactions (Figure S4G-I) with predicted VEGF signaling, an HIF1A target gene elevated in G3 tumors, from neoplastic cells to vascular cells (Figure S4J). Neoplastic cells represented the major source of the angiogenic factor VEGFA, an HIF1A target gene elevated in G3 tumors (Figure S4K). By comparing the differential number and strength of interactions between tumor grades, vascular cells emerged as dominant communicators, including with neoplastic cells, monocytic cells, T cells, and vascular cells (Figure 4A).

Figure 4.

Spatial reorganization of immunosuppressive niches under hypoxia with vascular involvement. (A) Heatmap showing the predicted differential number of interactions or interaction strength among different cell populations in G2 and G3 PXA. The bar height indicates the degree of change between the 2 conditions. In the color-bar, red (or blue) represents increased (or decreased) signaling in the G3 compared to the G2. (B) Pearson correlation between normalized hypoxia score and the proportion of TAM-2 cells relative to total TAM across FOVs (r = 0.50, P = .005). (C) Spatial map depicting cell-type niches within the same FOV. TAM-1 (HLA class II-high), TAM-2 (immunosuppressive), vascular, and other cell types are color-coded. (D) Spatial density maps of hypoxia signature score, HIF1A, VEGFA, and CD163 expression within a representative field of view (FOV). White circles highlight regions of interest. (E) Pearson correlation between normalized hypoxia score and the proportion of HLA class II-high TAM relative to total TAM across FOVs (r = −0.50, P = .005). (F) Spatial map depicting cell-type niches in the corresponding FOV, including TAM-1 (HLA class II-high), TAM-2 (immunosuppressive), CD8⁺ T cells, and other cell types. (G) Spatial density maps of hypoxia signature score and HLA class II gene expression (HLA-DRA, HLA-DRB1) in a second representative FOV. Regions of interest are highlighted by white circles.

In GBM, hypoxia is implicated as a critical driver of spatial organization²⁷ and malignant cell state.²⁸ Investigating patient-level tumor heterogeneity in PXA using Numbat, we identified recurrent subclonal CNA patterns, including both gains and losses (Figure S5). Subclones from a single patient showed differences in copy-number burden and in expression-derived hypoxia and proliferation signature scores. However, due to limited inferred subclonal architecture in the 10 samples, a relationship between hypoxia and malignant cell state could not be established.

Investigating intratumoral heterogeneity, hypoxia emerged as a potential driver of spatial organization in PXA. Spatial transcriptomics was performed on a total of 28 fields of view (FOVs) from 4 PXA and annotated based on snRNA-seq data and canonical gene markers for each cell type (Figure S6 and  Supplementary Material S1). Two TAM populations were identified, spatial transcriptomic TAM cluster #1 (stTAM-1) characterized by high expression of antigen presentation genes (HLA class II) and spatial transcriptomic TAM cluster #2 (stTAM-2) characterized by expression of immunosuppressive marker genes (SPP1). Across FOVs, immunosuppressive-like stTAM-2 cells were positively associated with the normalized hypoxia score (Pearson correlation coefficient, r = 0.50, P = .005) (Figure 4B). A representative spatial map and associated spatial density maps illustrate this relationship within a single FOV (Figure 4C and D). Conversely, the proportion of HLA class II-high stTAM-1 cells were inversely associated with the normalized hypoxia score (Pearson correlation coefficient, r = −0.50, P = .005) as illustrated in spatial maps and spatial density maps from a single FOV (Figure 4E-G). Consistent with hypoxia being an important driver of the immunosuppressive TAM response, VEGF-high tumors had increased CD163+ cells relative to VEGF-low tumors (CD163+/total cells, 6.8% [n = 10] vs 1.3% [n = 10], respectively, P = .02; see Supplementary Material S1). VEGF-high tumors also had a trend toward increased TGFB1 expression relative to VEGF-low tumors (TGFB1, 741.0 [n = 10] vs 550.3 [n = 10], respectively, P = .06).

Reduced T Cell Number at Progression

Given increased tumor hypoxia, a more immunosuppressive TAM phenotype, and decreased antigen presentation, we investigated the impact on the T cell response. Multiplexed single-cell imaging on a cohort of G2 and G3 tumors demonstrated a significant decrease in CD8+ and CD4+ T cells relative to all immune cells with progression to G3 (Figure 5A). Examining longitudinal tumor pairs from the same patient over time also identified a significant decrease in CD8+ and CD4+ T cells over time (Figure 5B-D).

Figure 5.

Altered T cell numbers and phenotype. (A) Violin plot showing T cell subsets, CD8+ T cells (left) and CD4+ T cells (right), as a percentage of all immune cells enumerated in G2 and G3 PXA. The middle line denotes the median (G2 n = 14, G3 n = 36). Statistical significance was assessed using an unpaired Student’s t-test. P = .05. (B) Percentage of CD8+ or CD4+ T cells over all immune cells in individuals with longitudinally paired samples over time, denoted initial and recurrent (n = 7). Statistical significance was assessed using a paired Student’s t-test. P = .05. (C) Representative images from multiplex single-cell imaging showing immune, endothelial, and tumor cell distribution in initial and recurrent tumor pairs from the same patient, highlighting CD8+ T cells and (D) CD4+ T cells. (E) UMAP plot depicting clustering of the lymphoid populations using the Azimuth data reference in G2 (left) and G3 (right) PXA (n = 10). Dots represent cells, and color represents each cell type. (F) Heatmap comparing select differentially expressed genes in G2 and G3. Genes are grouped into differentiation, tissue-resident, and cytotoxicity markers, which are color-coded. Color scale denotes relative expression level in each condition. (G) Dot plot showing expression level and percent T cells expressing a select group of anti-inflammatory cytokines. Color indicates expression, and the size of the dots represents the percentage of the cells. (H) Dot plots showing expression of known markers for Tregs, including the gene marker and transcription factor, (I) and for exhaustion markers on CD4+ T cells. The size of the dots indicates the percentage of cells expressing the genes, and the color intensity denotes the level of expression. Abbreviations: HSPC, hematopoietic stem/progenitor cells; NK, natural killer cells; PXA, pleomorphic xanthoastrocytoma.

To investigate T cell phenotype with progression, we used the Azimuth reference dataset²⁵ to use exemplary genes to annotate lymphoid populations from snRNA-seq. The identified populations were similar across grades, including CD8+ T cells, CD4+ T cells, NK cells, mucosal-associated invariant T cells, hematopoietic stem/progenitor cells, and B cells (Figure 5E). With progression to G3, we observed a shift from a Teff/Teffm phenotype toward a central memory (Tcm) phenotype with reduced cytotoxicity, as evidenced by upregulation of IL-7R, CCR7, and SELL, along with downregulation of cytotoxic effector genes, including granzymes and perforin (Figure 5F, Figure S7A and B). This shift was accompanied by reduced expression of T-bet (TBX21), a key transcription factor that promotes Th1 and CD8+ cytotoxic differentiation (Figure 5F). Despite an increase in a Tcm phenotype, both CD69 and CD103 (ITGAE), markers of tissue-resident memory T cells,²⁹ were diminished in G3 (Figure 5F, Figure S7A and B), suggesting T cell differentiation is not directed toward a tissue-resident memory phenotype. In parallel, lymphoid cells in G3 tumors had increased expression of anti-inflammatory cytokines (IL-10 and TGF-β1) (Figure 5G, Figure S7A and B).

While regulatory T cells (Tregs) did not form a distinct cluster, approximately 7% of CD4 T cells expressed FOXP3 and with G3 progression, canonical markers for Tregs,³⁰ such as FOXP3, IL2RA (CD25), and STAT5, were increased (Figure 5H). Furthermore, CD4 T cells had increased expression of exhaustion markers, including CTLA-4, TIGIT, and HAVCR2 (TIM-3) (Figure 5I), indicating a transition to a more exhausted and dysfunctional T cell state. In parallel with these changes in T cell phenotype, immunnoprofiling demonstrated an upward trend in the CD4/CD8 T cell ratio with G3 progression in 4 patients with paired longitudinal samples (Figure S7C). The expression of HLA-DR, a marker associated with T cell activation and cytotoxic potential,³¹ was also reduced at both the pathway and individual gene transcript levels in G3 T cells (Figure S7D and E).

Altered Cellular Neighborhoods and Chemokine Signaling at Progression

To investigate the evolution of cellular neighborhoods, we compared G2 and G3 tumors profiled by multiplexed single-cell imaging using QUICHE,^19,32 which permits the identification of differentially enriched cellular niches. Tumor progression to G3 was associated with systematic shifts in the spatial organization of tumor and immune cells (Figure 6A-D, Figure S8A and B). G2 enriched niches suggested complex interactions between antigen-presenting cells (APC or CD11b-HLA-DR+ cells), tumor cells, other myeloid cells, and CD8+ T cells. As illustrated in Figure 6A and B, enriched G2 niches included those containing CD11b+ and CD11b+ HLA-DR+ myeloid cells and CD8+ T cells. In contrast, G3 niches were dominated by tumor cell interactions with immunosuppressive myeloid cells, including CD163+ microglia and CD14+ myeloid cells (Figure 6C and D). Thus, tumor progression was associated with altered spatial organization of tumor-immune neighborhoods.

Figure 6.

Altered cellular neighborhoods and chemokine signaling and their potential impact on the immune landscape. (A) Spatial interaction graph derived from differentially abundant niches neighborhoods in G2, nodes represent cell types, node size is proportional to connectedness as measured by eigenvector centrality, and edge weights correspond to the number of unique patients with the corresponding interaction; Niche neighborhoods were considered significant if they had a spatial false discovery rate (FDR) < 0.20. (B) Representative images for a cellular niche differentially enriched in G2 and (C) in G3. (D) Niche network for cellular niche neighborhoods differentially enriched in G3 PXA. (E) A dot plot showing differentially expressed chemokines expressed by TAMs and (F) chemokine receptors expressed by T cells implicated in altered cellular communication with progression. (G) Dot plot showing CXCL16 expression in TAMs. (H) Pearson correlation for CXCR6 and GZMA in T cells that express both genes from snRNA-seq data. (P = .0017 and r = 0.47). (I) Neoplastic cell expression of CXCL14 from snRNA-seq data (n = 10 tumors) in G2 and G3 (P = 4.42e-06). Expression values in plots are “LogNormalized.” PXA tumor grade is color-coded: blue (G2) and red (G3). (J) Comparison of CXCL14 expression (normalized count) from Nanostring targeted gene panel in G2 (n = 8) and G3 (n = 14) PXA. Each dot represents a value from a single patient, horizontal lines represent the mean, and error bars indicate SD. (K) Representative RNA scope images showing CXCL14-expressing cells (red) in PXA from the same patient over time: G2 (top) and G3 (bottom). Scale bar denotes 200 µm, and the boxed region denotes the magnified region. (L) Pearson correlation between relative expression of CXCL14 and ITGAX within tumor regions in PXA (r = 0.39, P = .03). (M) Representative image illustrating the spatial relationship between tumor cells (yellow hexagon) expressing CXCL14 (blue dot) and TAMs (green hexagon) expressing the DC-marker, ITGAX (red dot), within a single field. All other cell types are shown in gray (hexagon). Two distinct regions of the image are magnified to highlight local spatial interactions. (E-G) Dot size represents the percent expressed, and dot color denotes average expression. Abbreviation: PXA, pleomorphic xanthoastrocytoma.

Investigating inferred cellular crosstalk, with tumor ­progression predicted immunostimulatory signals from TAMs to T cells decreased, including CXCL16-CXCR6, PECAM1-CD38 and HLA-CD8, suggesting impaired T cell recruitment and activation (Figure S9A). Pseudobulk analysis of DEGs identified several chemokines and chemokine receptors with differential expression between G2 and G3 tumors in TAMs and T cells (Figure 6E and F), including decreased expression of CXCL16 and its receptor CXCR6. CXCL16, expressed by TAMs (Figure S9B) and enriched in TAM cluster #1, characterized by a high antigen presentation signature (Figure 6G), is a key scavenger receptor involved in both antigen presentation by dendritic cells and recruitment of T cells into the brain.^33-37  CXCR6, the receptor for CXCL16, was highly expressed by T cells and its expression was positively correlated with GZMA (Figure 6H, Figure S9C). Moreover, proximity analysis of spatial transcriptomic data suggested spatial colocalization of CXCL16-expressing TAMs with CXCR6-positive lymphoid populations (Figure S9D). In MAPK-altered tumors, we and others have demonstrated that the chemokine CXCL14 can promote an antitumor immune response.^10,38 With disease progression, CXCL14 expression was significantly decreased in tumor cells by snRNA-seq (Figure 6I) and globally in tumors using a targeted gene expression panel (n = 22) (Figure 6J). The decrease in CXCL14 could be visualized in a single patient over time using chromogenic in situ hybridization (Figure 6K). Investigating spatial relationships, CXCL14-expressing tumor cells were correlated with CD11c (ITGAX)-expressing cells (Figure 6L and M). Taken together, these data suggest PXA progression is associated with altered cellular communication, including alterations in spatial niches and chemokine signaling, that act to further impair the antitumor immune response.

Discussion

Unlike the lymphocyte-poor immune response dominated by immunosuppressive TAMs in IDH-wildtype GBM and IDH-mutant diffuse glioma, we and others have identified MAPK-driven tumors such as PXA as unique, enriched for CD8 T cells and antigen processing and presentation.^9,10,19,39 With PXA progression, however, there is a reduction in T cells and an increase in immunosuppressive myeloid cells.¹⁹ To investigate factors shaping the tumor-immune microenvironment in PXA over the disease trajectory, we used a multiomics approach to study a large cohort of immune-enriched MAPK-driven glioma, including 10 patients with multiple tumor samples over time. Multiomic analyses, including snRNA-seq, targeted gene expression profiling, spatial transcriptomics, and multiplexed single-cell imaging, suggest disease progression is associated with a reprogramming of the tumor ecosystem. With increased tumor growth and hypoxia, altered communication networks and cellular niches promote an immunosuppressive TME characterized by altered chemokine signaling, increased TGF-β signaling, reduced antigen presentation, and reduced ­effector CD8+ T cells. We believe our study is the first to demonstrate coevolution of the tumor and its immune microenvironment in an MAPK-driven glioma. Highlighting the dynamic interplay between the tumor, vasculature, and immune elements, our study identifies factors associated with disease progression that are common to diffuse glioma, such as hypoxia, and unique to PXA, such as altered CXCL14 signaling.

Glioma malignant progression is characterized by genomic evolution with the emergence and expansion of genetically distinct subclones.^3,10 In PXA, intratumoral heterogeneity,¹⁷ contributes to disease progression to G3 defined by an increase in proliferative index^16,40 and an association with increased genomic complexity, including copy number ­alterations.^17,40 With G3 progression, snRNA-seq demonstrated that tumor cells transition to a more proliferative progenitor state with an altered transcriptional profile and increased hypoxia. In our cohort, telomere maintenance regulation also emerged as enriched in G3 tumors. While TERT promoter mutations in PXA have been associated with progression and unfavorable prognosis,^17,41 a recent study failed to identify an association.⁴² In PXA, the mechanisms of telomere maintenance, if any, and their association with disease outcome require further study.

Tumor-associated myeloid cells, including microglia and macrophages, are abundant in glioma and orchestrate the glioma-immune landscape.^24,43 While the total number of TAMs in PXA was not changed at progression, the myeloid cell phenotype and cellular architecture were markedly altered, as demonstrated by transcriptional profiling, multiplexed single-cell imaging, and analysis of cellular neighborhoods. Antigen-presenting TAMs were significantly decreased. Gene and protein expression of HLA class II and ITGAX (CD11C) was decreased and TAMs with a DC-like signature were decreased. In contrast, immunosuppressive TAMs were increased, including expression of markers such as CD163, CD204, and CD206. In parallel, predicted cellular neighborhoods underwent significant restructuring with decreased HLA-DR+ myeloid-CD8 T cell interactions, decreased myeloid-myeloid interactions, and increased interactions between tumor cells and CD163+ TAMs in G3 tumors.

With tumor progression, the myeloid scavenger immunosuppressive program was enriched in PXA. In IDH-WT GBM, Miller et al.²⁴ recently demonstrated that TAMs with this immunomodulatory program were enriched in hypoxic and vascular niches and were an important determinant of the T cell response, the clinical response to checkpoint therapy, and overall survival. The retrospective nature of our study and limited cohort of flash-frozen tumor samples limited our analyses of important TAM subsets, including border-associated macropahges.^44-47 Future studies will be required to investigate how these important subsets change with disease progression and whether TAM reprogramming in PXA is associated with an altered response to immunomodulatory therapy.

TAM phenotype and activity are critical determinants of the T cell response. With G3 progression effector CD8 T cells and CD4 T cells were reduced. Moreover, the remaining T cells were phenotypically altered with reduced cytotoxicity and cytokine production and increased expression of exhaustion markers. This phenotypic switch may be attributable, in part, to decreased TAM MHC class II antigen presentation in higher grade tumors, as shown by Kilian et al.²² Our data suggests that the T cell phenotype shifts from effector to more Tcm states. While Tcm cells may provide a reservoir for long-term immune surveillance, their limited cytotoxic activity in combination with the immunosuppressive myeloid phenotype may reduce the immediate immune pressure on tumor cells, potentially facilitating tumor progression. Consistent with these data, prior reports have suggested a phenotypic shift in tumor-infiltrating T cells from effector toward memory states in recurrent GBM.⁴⁸

Hypoxia emerged as a dominant feature of the TME in G3 tumors with hypoxic niches associated with a more immunosuppresssive TAM population. Hypoxic tumor cells also upregulated angiogenic factors, driving angiogenesis and the vasculature emerged as a central communication hub. Prior studies in glioma have demonstrated that ­activation of HIF1A promotes TGF-β signaling, which can facilitate macrophage recruitment and promote M2-like macrophage polarization.⁴⁹ Hypoxia-induced upregulation of chemokines such as CCL5 and CXCL8 (IL8), as seen in our cohort, can further promote an immunosuppressive microenvironment.^50,51 Hypoxic stress can drive stem cell survival and tumor invasiveness.⁵² Similar to GBM,²⁷ our data suggest increased tumor hypoxia may be an important driver of spatial organization in PXA.

Chemokines, critical mediators of cellular crosstalk, orchestrate the tumor immune landscape. In G3 PXA, several chemokines implicated in CD8 T cell recruitment were decreased, including CXCL16,³³ CXCL14,¹⁰ and CCL2. Unbiased ligand-receptor network analysis in PXA predicted a decrease in CXCL16-CXCR6 interaction in G3 tumors. The predicted decrease was associated with a grade-dependent reduction in CXCL16 expressing myeloid cells, particularly the HLA-DR-high myeloid cluster (#1), and a reduction in CXCR6 expressing T cells. CXCL16-CXCR6 signaling has recently been identified as important in regulating T cell recruitment, activation, and retention in the brain in diverse pathologic conditions, including viral infection, neurodegenerative disease, and brain tumors.^35-37 CXCR6 has also been shown to be important for CD8+ T cell-dendritic cell interactions within the tumor.⁵³ While the function of CXCL16-CXCR6 in chronic neurologic disease is complex, the observed association between CXCR6 and GZMA expression in PXA supports its role in promoting a cytotoxic T cell response in G2 PXA. Subsequent studies will be required to dissect the mechanistic role of CXCL16-CXCR6 function in the antitumor immune response.

CXCL14 is highly expressed in immune-enriched PXA and CXCL14 can promote a CD8⁺ T cell antitumor immune response in a murine model for glioma.¹⁰ In addition to direct effects on CD8 T cells,³⁸ CXCL14 has also been implicated in the recruitment and activation of antigen-presenting cells to tumors, particularly dendritic cells (DCs).^10,54,55 In spatial analyses, we identified a correlation of CD11c-expressing DCs with CXCL14-expressing tumor cells. Thus, decreased CXCL14 in G3 PXA may promote the decrease in CD8+ T cells and decrease in antigen-presenting myeloid cells. Given the potential role for CXCL14 in the glioma-associated immune response, including TLR9-mediated antitumor responses,⁵⁶ ongoing studies are investigating whether CXCL14 can be used to promote a therapeutic antiglioma-immune response.

Integrating data from snRNA-seq, spatial transcriptomics, and multiplexed single-cell imaging, we uncover a coordinated evolution of the tumor and its microenvironment. Highlighting the dynamic interplay between tumor, vasculature, and immune elements, hypoxia and chemokine signaling emerged as key components of the glioma-immune landscape that evolve with tumor progression. These findings identify new therapeutic opportunities in PXA, and they argue that optimal immunomodulatory strategies may differ along the disease trajectory.

Supplementary Material

Supplementary material is available online at Neuro-Oncology (https://academic.oup.com/neuro-oncology).

Author Contributions

E.H. and J.J.P. designed and directed the study, interpreted data, and wrote the manuscript. E.H., R.L.A., L.W., H.P., B.O., A.K., A.S., and A.K. conducted experiments, acquired data, analyzed data, and commented on the manuscript. S.C.B., J.S.R., and V.V. analyzed data and commented on the manuscript. M.S.B., M.C.O., E.F.C., R.M.P., S.C.B., M.A., S.L.H.-J., H.O., and A.D. provided resources and commented on the manuscript.

Conflict of Interest Statement

The authors declare that no conflicts of interest exist.

Funding

This work was supported by the National Institutes of Health (R01 NS131474 to J.J.P., T32 CA151022 to E.H., R35 1R35NS142982-01 to H.O., R01LM013897 to A.D., and P50CA097257 to A.D.), the Department of Defense (W81XWH-19-1-0348 to A.D.), the T.J. Martell Foundation (J.J.P.), the American Brain Tumor Association Basic Research Fellowship, fully supported by the Karl Schmidt Oligodendroglioma Research Fund (BRF2500040 to R.L.A.), the Cancer Research Society Postdoctoral Fellowship (H.P.), the Parker Institute for Cancer Immunotherapy (M.A.), and the Alliance for Cancer Gene Therapy (M.A.). Resources were provided by the UCSF Brain Tumor SPORE Biorepository (National Institutes of Health P50 CA097257) and by a gift from the Panattoni family.

Ethics Statement

All tumor samples were deidentified prior to analysis and the ethics approval number for the use of deidentified human biospecimens is 10-01318. Written informed consent was obtained for tissue banking from each of the participants. These studies were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments.

Data Availability

SnRNA-seq data that support the findings of this study have been deposited in NCBI GEO with the # GSE312124 accession codes.