Search

Glioblastomas are malignant tumors of the central nervous system hallmarked by subclonal diversity and dynamic adaptation amid developmental hierarchies. The source of the dynamic reorganization within the spatial context of these tumors remains elusive. Here, we characterized glioblastomas in-depth by spatially resolved transcriptomics, metabolomics and proteomics. By deciphering regional shared transcriptional programs across patients, we infer that glioblastomas are organized by spatial segregation of lineage states and adapt to inflammatory or metabolic stimuli reminiscent of reactive transformation in mature astrocytes. Integration of metabolic imaging and image mass cytometry uncovered locoregional tumor-host interdependence resulting in spatially exclusive adaptive transcriptional programs. Inferring copy-number alterations emphasizes a spatially cohesive organization of subclones associated with reactive transcriptional programs, confirming that environmental stress gives rise to selection pressure. A model of glioblastoma stem cells implanted into human and rodent neocortical tissue mimicking various environments confirmed that transcriptional states originate from dynamic adaptation to various environments. Funding: DHH is funded by the Else Kroner-Fresenius Foundation. The work is part of the MEPHISTO project (PI: DHH and DD), funded by BMBF (iGerman Ministry of Education and Research) (project number: 031L0260B). The Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) supports the work of AW (project number: SFB 824 C04). VR, KJ and UGH funded by BMBF (Bundes Ministerium fur Bildung und Forschung) (project number: FMT 13GW0230A), US was supported by the Fordergemeinschaft Kinderkrebszentrum Hamburg. We thank Dietmar Pfeifer for here helpful advice. We thank Biorender.com. We thank Stella Maria Carro for her support and the provision of her laboratory facilities and equipment. Declaration of Interest: No potential conflicts of interest were disclosed by the authors. Ethical Approval: The local ethics committee of the University of Freiburg approved the data evaluation, imaging procedures and experimental design (protocol 100020/09 and 472/15_160880).

High-dimensional technologies have provided insights into transcriptional heterogeneity and dynamic plasticity which are hallmarks of brain tumors. Although scRNA-seq recovers the diversity of transcriptional states, their spatial context within the neuronal environment has remained unexplored. Here, we integrated spatially resolved transcriptomics and metabolomics to characterize the glioma landscape at multiple molecular levels. We integrated spatial transcriptomics (10X Visium, n= 28) and metabolomics (MALDI, n= 6) from primary and recurrent glioblastoma patients. Unsupervised cluster analysis and pattern recognition uncovered 5 spatially distinct transcriptional programs, shared across patients. These included three cell-specific developmental stages largely reflecting those that are part of recently suggested models. By integrating metabolome data, we identified an additional program encompassing reactive responses to hypoxia. Areas of hypoxic response were negatively correlated with proliferation (R2= -0.34, p< 0.001) and significantly enriched for gene expression signatures from the S-phase (p< 0.001). Modeling of transient spatial gradients using vector field predictions showed opposing vector directions of hypoxia response and migratory capacity, underpinning the “go-or-growth” theory, where cells either proliferate or migrate. Inferred copy-number alterations (CNA) revealed a significant increase in genomic instability, highly correlated to hypoxia response (R2= 0.78, p< 0.001). Near necrotic areas, we observed a significant accumulation of CNAs while proliferation was inhibited, and cells remained in the S-phase. We validated this hypothesis of hypoxia-driven accumulation of CNAs by chronic hypoxia cultures of primary patient-derived cell lines. A gain of chromosomal instability after long-term hypoxia was observed, suggesting that hypoxic areas in glioblastoma function as bioreactors for genomic instability. Our findings elucidate the evolution of resistant subclones in glioblastoma. They provide novel insights into the dynamic regulation and interaction between host and tumor and cast a new light on hypoxic and necrotic areas, which may represent the source of the heterogeneous and resistant nature of glioblastomas.

Regional diversity of anti-tumor immunity in glioblastoma (GBM) has been poorly investigated due to technical limitations. To overcome these hurdles, we developed SPTCR-seq (SPatial TCell Receptor sequencing) to spatially profile T cell immune response. Here, we performed 10X-Visium spatial transcriptomics and SPTCR-seq on 9 de-novo IDH wt glioblastoma. For SPTCR-seq, we captured enriched T-cell receptor (TCR) sequences by hybridization followed by Oxford nanopore (ONT) long-read sequencing. We established a computational pipeline for spatial demultiplexing, error correction and TCR reconstruction using a Seq2Vector model, followed by a long-term memory autoencoder to deconvolute and compare CDR3 sequence similarities across patients. The on-target rate for detected TCR genes reached over 85%, of which our long-TCR algorithm could completely reconstruct TCRs in over 70%. To decipher the regional diversity of T-cell clonality, we performed spatially weighted regression to estimate the overlap of T-cell clonal expansion to the recently described regional transcriptional niches. We found that myeloid cell infiltration was highly correlated with increased T cell diversity. We integrated single cell RNA-seq data and annotated T cell subtypes through the “natural barcode” of the CDR3 sequence, revealing that CD4+ T cells interact with myeloid cells via MHCII in segregated areas, also described as the Reactive Immune niche. In contrast, CD8+ T cells exhibit a dysfunctional/exhausted phenotype at the same location. In comparison to the myeloid-infiltrated regions, the hypoxia-associated niches showed increased CD8+ T cells with greater clonal expansion, suggesting an immune response anti-cancer immunity. Global spatial vector field analysis revealed a spatial co-occurrence of metabolic vulnerability and a mutational burden with CD8+ T effector cells negatively associated with myeloid cell infiltration. Integration of CDR3 sequences in different patients revealed higher similarity of motifs of CD4+ compared with CD8+ T cells in hypoxia areas. Our data provide new insights into the regional heterogeneity of cancer immunity.

Glioblastomas are highly malignant tumors of the central nervous system. Evidence suggests that these tumors display large intra- and inter-patient heterogeneity hallmarked by subclonal diversity and dynamic adaptation amid developmental hierarchies. However, the source for dynamic reorganization of cellular states within their spatial context remains elusive. Here, we in-depth characterized glioblastomas by spatially resolved transcriptomics, metabolomics and proteomics. By deciphering exclusive and shared transcriptional programs across patients, we inferred that glioblastomas develop along defined neural lineages and adapt to inflammatory or metabolic stimuli reminiscent of reactive transformation in mature astrocytes. Metabolic profiling and imaging mass cytometry supported the assumption that tumor heterogeneity is dictated by microenvironmental alterations. Analysis of copy number variation (CNV) revealed a spatially cohesive organization of subclones associated with reactive transcriptional programs, confirming that environmental stress gives rise to selection pressure. Deconvolution of age-dependent transcriptional programs in malignant and non-malignant specimens identified the aging environment as the major driver of inflammatory transformation in GBM, suggesting that tumor cells adopt transcriptional programs similar to inflammatory transformation in astrocytes. Glioblastoma stem cells implanted into human neocortical slices of varying age levels, independently confirmed that the ageing environment dynamically shapes the intratumoral heterogeneity towards reactive transcriptional programs. Our findings provide insights into the spatial architecture of glioblastoma, suggesting that both locally inherent tumor as well as global alterations of the tumor microenvironment shape its transcriptional heterogeneity. Global age-related inflammation in the human brain is driving distinct transcriptional transformation in glioblastomas, which requires an adjustment of the currently prevailing glioma models.

The diversity to T cell responses and clonality in spatially heterogeneous glioblastoma is of paramount importance to explore underlying mechanisms of anti-tumor immunity. Spatial transcriptomics, a novel technology to map the transcriptional architecture, is technically limited to discover T cell receptor (TCR) sequences as the 3' approach lacks sufficient coverage. Here, we established SPTCR-seq, a method to capture TCR sequences followed by long-read sequencing to enable full-length TCR reconstruction. We performed 10X Visium spatial transcriptomics on 9 primary and recurrent glioblastoma with both 3’-sequencing and SPTCR-seq. For SPTCR-seq, we target enriched T cell receptor sequences by capturing by hybridization followed by Oxford-Nanopore long-read sequencing. The on-target rate was above 80% for captured TCR genes and spatial barcode was successfully aligned in more than 60%. IgBlast and MixCR were used to reconstruct the TCR and map T cell clonality. Within our recent developed spatial transcriptomic analysis framework (SPATA2), we build a novel toolbox, SPATA-Immunology, which enables integration of stRNA-sequencing data and spatially resolved TCR sequencing. Our data showed that clonal evolution of T cells is limited to regional areas underpinned by significant spatial autocorrelation coefficient (0.6-0.95, padj< 0.001). In the surrounding tumor cell spots, the recently described transcriptional program “reactive immune” (RI), was significantly enriched. Using spotlight, a computational approach to project scRNA-sequencing into the spatial space, we found a local enrichment of CD163 positive macrophages exclusively in areas of large T cell clonality. Imaging mass cytometry of a consecutive section confirmed the spatial confluence of T-cell infiltration and CD163-positive macrophages. Through DeepTCR we uncovered potential epitopes which correlate with T cell clonality and might help to discover novel targets for CART therapy. Spatial profiling of TCR sequences through SPTCR-seq is a powerful tool to investigate anti-tumor immunity in glioblastoma and allows to discover general and personalized targets for immunotherapy.

Glioblastomas are embedded into an immunosuppressive microenvironment resulting in limited success of immunotherapies. Although, we and others observed defined myeloid-tumor crosslinks reducing T cell homing and activation, the spatial context of these interactions remained unexplored. Here, we provide evidence, that local T cell infiltration results in a defined activation of myeloid cells causing transcriptional reprogramming of tumor cells reminiscent of reactive transformation in mature astrocytes. Through integration of spatially resolved transcriptomics and imaging mass cytometry (n=18, 39 protein glioma panel) we mapped defined transcriptional responses in areas of high or low T cell infiltration respectively. Functional analysis revealed that areas of large T cell infiltration are enriched for glial (CHI3L1, GFAP and VIM) and inflammatory genes (HLA-DRA, C3, CCL4, CCL3). We found that marker genes of common reactive states in mature astrocytes significantly overlap with the reactive immune program of glioblastoma cells (f-score 0.76, p=2.2e-10). Increased numbers of CD163+ cells were found in surrounding areas of T cell infiltration and spatially linked to defined immunosuppressive release of IL10, recently reported as a major driver of T cell exhaustion. To support our findings, we injected a primary glioblastoma cell line into cortex slices of three different human and rodent donors. After 7 days of tumor growth, we performed scRNA-sequencing of FACS-sorted tumor cells and baseline cell culture cells. Compared to baseline, we found cells of all reported transcriptional states, confirming the dynamic adaptation of cells within a neural environment. In elderly donors, a significant accumulation of reactive immune programs (ANOVA p< 0.001) was observed. Immunostainings confirmed an increased myeloid cell activation in these donors. Our data suggest that inflammatory stimuli in the glioblastoma microenvironment cause transcriptional reprogramming in glioblastoma similar to inflammatory transformation of reactive astrocytes. The spatial exclusivity of these programs highlights the value of a spatial perspective on heterogeneity.