Harmonized single-cell landscape, intercellular crosstalk and tumor architecture of glioblastoma

Abstract

Glioblastoma, isocitrate dehydrogenase (IDH)-wildtype (hereafter, GB), is an aggressive brain malignancy associated with a dismal prognosis and poor quality of life. Single-cell RNA sequencing has helped to grasp the complexity of the cell states and dynamic changes in GB. Large-scale data integration can help to uncover unexplored tumor pathobiology. Here, we resolved the composition of the tumor milieu and created a cellular map of GB ("GBmap"), a curated resource that harmonizes 26 datasets gathering 240 patients and spanning over 1.1 million cells. We showcase the applications of our resource for reference mapping, transfer learning, and biological discoveries. Our results uncover the sources of pro-angiogenic signaling and the multifaceted role of mesenchymal-like cancer cells. Reconstructing the tumor architecture using spatially resolved transcriptomics unveiled a high level of well-structured neoplastic niches. The GBmap represents a framework that allows the streamlined integration and interpretation of new data and provides a platform for exploratory analysis, hypothesis generation and testing.

Optimizing Xenium In Situ data utility by quality assessment and best practice analysis workflows

“…Identifying the most reliable tools to define these domains is of general interest, although no independent comparison is yet available. Therefore, we benchmarked 4 domain finder algorithms (neighborhood-based 43 , Banksy 44 , DeepST 45 , SpaGCN) against the regions identified by expert manual annotation using the coronal P56 section from Allen Brain Atlas 46 ( Figure 5D-E ). A total of 36 domains were manually annotated, and thus each method was adjusted to predict a similar number of domains (35-37).…”

GBMdeconvoluteR accurately infers proportions of neoplastic and immune cell populations from bulk glioblastoma transcriptomics data

“…In contrast, the mode using the GBM-tissue specific immune and neoplastic cell markers listed in Tables S6 and S7 is denoted MCP GBM . In addition, at the time of preparing this manuscript a larger GBM-specific single cell resource, GBMap, was made available that amalgamated 26 single cell brain and GBM datasets [27]. We, thus, also ran MCPcounter using the GBMap marker genes, denoting this as MCP GBMap .…”

“…To make MCP GBM available to the neuro-oncology community, we have packaged it into an online application called GBMdeconvoluteR. We also give the user the option to use the marker genes from GBMap [27] because, although these did not quantify cell types as accurately as MCP GBM , the GBMap reference set extends the range of GBM non-neoplastic cell types that can be quantified from bulk expression data. GBMdeconvoluteR is, thus, a web-based application that enables users to upload bulk GBM expression profiles and output the relative proportion of immune and neoplastic GBM cells, or using GBMap markers genes as input, to also include normal brain and blood-vessel cells, across multiple samples.…”

Localization of protoporphyrin IX in glioma patients with paired stimulated Raman histology and two-photon excitation fluorescence microscopy

“…To verify the accumulation of PPIX in macrophages on a cellular level, we performed single cell deconvolution using the state-of-the-art Robust Cell Type Decomposition (RCTD) 46 and the GBMap reference dataset containing approximately ~1 million cells 47 . The inherent heterogeneity of PpIX accumulation in our specimens enabled us to compare cellular composition across areas of low and high PpIX accumulation (Fig.…”

“…The data contain samples from the tumor core area (Core), the contrast-enhancing rim (CE-Rim) and the in ltrative regions, which were de ned as weak PpIX positive areas without de ned histopathological classi cation which regions are samples according to the Ivy-GAP criteria. All samples (n=42) were deconvoluted to infer the cellular distribution using the pan-GBM single-cell data GBMap 47 . The full-scRNA-seq dataset was down-sampled by maintaining the quantitative distributions across all cellular subtypes, de ned as "annotation level 4".…”