Single-cell RNA sequencing suggests different progenitor lineages between IDH mutant and IDHwt glioma

Abstract

Introduction

Single-cell RNA sequencing has elucidated the heterogeneity in cancer. Single-cell glioblastoma (GBM) analyses have also proposed the resemblance of GBM cells to radial glia and outer radial glia (oRG) supporting the hypothesis that remnants of developmental tissue get reactivated in cancer. A recent study isolated neural progenitor cells (NPCs) from developing fetal human brain (gestational week 17-19) and classified NPCs based on their expression of THY1 (CD90), CD24 and EGFR. Ventricular radial glia are THY1⁻CD24⁻EGFR⁺ whereas oRG are THY1⁻CD24⁻EGFR⁻. Early neuron precursors are CD24⁺THY1⁻EGFR⁺ and glial progenitor cells (GPCs) are THY1⁺EGFR⁺. GPCs give rise to THY1⁺EGFR⁺PDGFRA⁺ pre-oligodendrocyte progenitor cells.

Methods

We aimed to apply the classification above to IDH mutant astrocytoma and oligodendroglioma as well as IDHwt GBM samples. We used 3 publicly available datasets: Wang (paired 74 IDHwt primary and recurrent samples), Tirosh (6 primary oligodendroglioma samples) and Venteicher (10 primary IDH mutant astrocytoma).

Results

In IDH mutant astrocytoma, 82.67% of cells are THY1⁺. In oligodendroglioma, 80.47% of cells are THY1⁺ (mostly EGFR⁺PDGFRA⁺ in both disease entities). In IDHwt EGFR amplified primary GBM samples, 87.5% of cells are THY1⁻CD24⁻EGFR⁺. This percentage drops to 70.4% in the recurrent setting. THY1⁻CD24⁻EGFR⁻ cells increase from 9.7% to 23.1% at recurrence. In IDHwt EGFRwt primary GBM samples, 48.6% of cells are THY1⁻CD24⁻EGFR⁺ and 44.15% are THY1⁻CD24⁻EGFR⁻. In the recurrent setting, 43.26% of cells are THY1⁻CD24⁻EGFR⁺ and 49.58% are THY1⁻CD24⁻EGFR⁻.

Conclusion

IDH mutant gliomas and IDHwt GBM express different progenitor cell markers. THY1 is highly expressed in IDH mutant gliomas.

Key Points.

• Most IDH mutant astrocytoma/oligodendroglioma cells express THY1 based on single-cell RNA expression.

• Most IDHwt GBM cells are THY1⁻CD24⁻, on the other hand, based on single-cell RNA expression.

• A different cell population than the dominant one in the primary setting becomes more prevalent in the recurrent setting.

Importance of the Study

IDH mutant astrocytoma and oligodendroglioma and IDH wild-type astrocytoma have vastly different pathophysiology and clinical behavior. It is reasonable to assume that they may differ in their cell of origin as well. In this study, we aimed to apply a newly described neural progenitor cells (NPCs) classification to IDH mutant and IDHwt gliomas. We find that most IDH mutant astrocytoma/oligodendroglioma cells are THY1⁺ resembling glial progenitor cells (GPCs), whereas most IDHwt GBM cells are THY1⁻CD24⁻ resembling ventricular and outer radial glia.

Astrocytomas include IDH1/2 mutant and IDH wild-type (IDHwt) subtypes. Glioblastoma (GBM) is the most common malignant brain tumor; the term is now reserved for IDHwt astrocytoma grade 4.¹

Little advances have been made in the treatment of GBM despite enormous research efforts. In fact, GBM was among the first cancers to be profiled by The Cancer Genome Atlas Project.² Since then, we have well understood the epigenetics and genetics of GBM at the DNA and transcriptomic levels. Receptor tyrosine kinase (RTK/RAS) pathway alterations (eg, via amplification of epidermal growth factor receptor (EGFR) are among the most common altered pathways in GBM. Chromosome 7 gain (containing EGFR) and chromosome 10 loss (containing PTEN among other tumor suppressor genes) are thought to be early initiating events in the gliomagenesis of IDHwt GBM.³

Taking this information into account, and given the prognostic relevance, certain molecular markers have been added as elements for CNS tumor grading based on the 2021 WHO classification,⁴ whereas grading was traditionally solely based on histological features. For IDHwt astrocytoma specifically, the presence of TERT promoter mutation, EGFR amplification and/or chromosome 7 gain/10 loss upgrade the tumors to molecular grade 4.

Advanced technologies such as single-cell RNA (scRNA) sequencing/single-nucleus RNA sequencing (snRNA-seq) and spatial transcriptomics have helped to further elucidate the cell-type composition as well as the complexity of this deadly cancer and its microenvironment. Recent scRNA/single-nucleus RNA sequencing (snRNA-seq) studies have demonstrated that GBM cells exhibit a high degree of heterogeneity and plasticity and seamless transitions between cellular states.^5,6 Moreover, interestingly, single-cell GBM analyses have proposed resemblance of GBM cells to radial glia and supporting the ‘‘embryonic rest’’ hypothesis and reactivation of remnants of developmental tissue in cancer.^7,8 A newly published study isolated neural stem and progenitor cells from developing fetal human brain (gestational week 17-19) and classified radial glia and outer radial glia (oRG) based on their expression of EGFR, CD24 and THY1.⁹ Ventricular radial glia (vRG) are CD24⁻THY1⁻EGFR⁺, whereas oRG are CD24⁻THY1⁻EGFR⁻. Early neuron precursors are CD24⁺THY1⁻EGFR⁺ and glial progenitor cells (GPCs) are THY1⁺EGFR⁺. GPCs are lineage-restricted to astrocytes and oligodendrocytes but not to neurons. THY1⁺ cells are further classified based on EGFR and PDGFRA expression to pre-OPCs (THY1⁺EGFR⁺PDGFRA⁺) and OPCs (THY1⁺EGFR⁻PDGFRA⁺) that finally give rise to mature oligodendrocytes (THY1⁺EGFR⁻PDGFRA⁻). The importance of EGFR and PDGFRA as markers for these progenitor cells again highlights the resemblance to markers of gliomagenesis.

Most of the GBM single-cell publications were limited by small sample size, given how expensive these technologies are. Some studies combined IDH mutant and IDHwt samples in the analysis, and some combined primary and recurrent samples. Not all studies reported on the mutations/copy number change data relevant to the new IDHwt GBM classification. The purpose of this study is to describe the single-cell make-up of glioma based on the WHO 2021 classification. To accomplish this, we use publicly available data from 3 recent scRNA/snRNA glioma studies: Wang et al.,⁶ Venteicher et al.¹⁰ and Tirosh et al.¹¹ We aimed to apply the CD24/THY1/EGFR classification above to IDHwt GBM, IDH mutant astrocytoma and IDH mutant oligodendroglioma, respectively. We find that most IDH mutant astrocytoma/oligodendroglioma cells are THY1⁺ resembling GPCs, whereas most IDHwt GBM cells are THY1⁻CD24⁻ resembling ventricular and oRG.

Methods

Data from the Wang, et al. study⁶ was downloaded using GEO accession GSE174554. The study profiled 86 primary and recurrent GBM specimens with snRNA sequencing. 76 of these samples represent IDHwt GBM. 52 of these samples had patient-matched pair identifiers. The Supplementary Tables were downloaded from the original paper. Information about EGFR amplification/mutation status, TERTp mutation as well as chromosome 7 gain/10 loss were extracted manually from Figure 1 from the original manuscript into ‘sample_df.csv’ (Table S1). EGFR and TERTp data were based on the UCSF500 clinical assay panel. Chr7/10 data were derived from snRNA analysis per Wang et al.

Figure 1.

This figure demonstrates a bar graph of the cellular composition of IDH mutant astrocytoma, 1p19q co-deleted oligodendrogliomas and IDHwt GBM based on single-cell RNA sequencing. The asterisks denote statistical significance.

For the Venteicher et al. 2017¹⁰ and Tirosh et al.¹¹ studies, the pre-processed matrix file was downloaded from the 3CA database.¹² The 3CA database houses 77 scRNA datasets where the quality control, filtering and cell-type annotation were all consistently applied to all the datasets and made available to download. The Venteicher et al. study included 10 IDH mutant astrocytoma samples of various grades: 1 primary grade 2, 6 primary grade 3, 1 recurrent grade 3, 1 primary grade 4 and 1 recurrent grade 4. The Tirosh et al. study included 6 IDH mutant/1p19q co-deleted oligodendroglioma samples: 5 were grade 2 and 1 was “grade 2/3”. The 3 datasets were not ­integrated, and batch correction was not applied. Bioinformatic analyses were applied to each dataset separately.

We used R 4.3.1 to analyze the scRNA/snRNA datasets. Seurat objects were created for the above studies per the Seurat V5 workflow.¹³ For the Wang et al. study, quality control for was completed per the Seurat workflow. Cells were selected for further analysis after excluding potential empty droplets (cells with less than 200 genes per cells) and doublets or multiplets (cells with more than 2500 genes per cell). Low quality or dying cells were excluded by selecting cells with less than 5% mitochondrial genes. The data were then normalized, highly variable features were selected, the data were scaled and dimensionality reductions were applied. We excluded samples that were comprised of <10 tumor cells.

For the Wang samples, cell-type annotation (tumor cells versus microenvironment) was extracted from “GSE174554_Tumor_normal_metadata.txt” from GEO. For the Venteicher and Tirosh samples, cell-type annotation was downloaded from 3CA.

Positive expression of markers of interest (EGFR⁺, CD24⁺, THY1⁺ and TERT⁺) was defined as feature expression values >0 using the FetchData command from Seurat. We used Seurat’s FindMarkers function to find the differentially expressed genes between the cellular groups of interest. As a default, Seurat uses the non-parameteric Wilcoxon rank sum test to perform this analysis. Differentially expressed genes were selected if the adjusted p-value was <.01 and the absolute log₂ fold-change was greater than 1 (log₂FC > 1 or log₂FC < −1). Gene set enrichment analysis (GSEA) was performed using clusterProfiler.¹⁴ The C2 curated gene collection set was used from the Molecular Signatures Database (MSigDB) using the msigdbr package in R.

IBM SPSS Statistics 27 was used for statistical analyses. The Kruskal-Wallis test was used to compare the percentage of cellular compositions across groups. p <.05 were considered significant.

Results

72 of the Wang samples met the quality control criteria above (32 primary samples and 40 recurrent samples). After quality control, the number of tumor cells ranged from 22-4224 (median 590 cells). Of the 32 primary samples, 9 were EGFR amplified and 23 were EGFR wild-type (EGFRwt), 27 had Chr7 gain/Chr10 loss while 5 did not, and 16 were positive for the TERTp mutation and 16 were negative. For the Venteicher samples, the number of tumor cells ranged from 112-1273 (median 441 cells). And for the Tirosh samples, the number of tumor cells ranged from 428-1174 (median 638 cells).

Applying the CD24/THY1/EGFR Classification to Astrocytoma

In primary IDHwt EGFR amplified samples, 87.51% of cells are CD24⁻THY1⁻EGFR⁺. The percentage drops to 70.39% in the recurrent setting. THY1⁻CD24⁻EGFR⁻ cells increase from 9.76% to 23.11% at recurrence. THY1⁺ and CD24⁺ cells comprise <2% of cells each. While the proportion differences are intriguing, they did not reach statistical significance.

In primary IDHwt EGFRwt samples, 48.63% of cells are CD24⁻THY1⁻EGFR⁺. The percentage drops to 43.27% in the recurrent setting. THY1⁻CD24⁻EGFR⁻ cells increase from 44.15% to 49.58% at recurrence. THY1⁺ cells comprise 4.3% and 3.8% and CD24⁺ cells comprise 2.92% and 3.35% in the primary and recurrent settings, respectively. The proportion differences did not reach statistical significance.

In IDH mutant astrocytoma, on the other hand, 82.67% of cells overall are THY1⁺, while 14.75% are THY1⁻CD24⁺. Unlike IDHwt GBM, only 2.57% of cells are THY1⁻CD24⁻. More specifically, and regardless of the tumor grade, most cells were THY⁺EGFR⁺PDGFRA⁺ and comprised 86.89%-88.3% of primary IDH mutant samples. In the one grade 4 recurrent sample, THY1⁻CD24⁺ cells increased to 26.18%. Of the THY1⁺ cells, THY1⁺EGFR⁺PDGFRA⁻ cells increased from an average of 8.11% in primary grade 3 samples to 12.07% in the recurrent grade 3 sample, and from 11.98% in the primary grade 4 sample to 73.87% in the recurrent grade 4 sample. THY1⁺EGFR⁺PDGFRA⁺ cells decreased from an average of 86.89% in the primary grade 3 samples to 66.59% in the recurrent grade 3 sample, and from 87.19% in the primary grade 4 sample to 18.32% in the recurrent grade 4 sample.

Similarly, of 1p19q co-deleted oligodendrogliomas, 80.47% of cells were THY1⁺, 14.36% of cells were THY1⁻CD24⁺ and 4.94% of cells were THY1⁻CD24⁻.

As expected by the percentages above, THY1⁺ cellular abundance in IDH mutant astrocytoma and oligodendroglioma was statistically significantly higher than IDHwt GBM (Kruskal-Wallis p-value <.0001). Similarly, THY1⁻CD24⁺ cells were significantly higher proportionally in IDH mutant astrocytoma and oligodendroglioma than in IDHwt GBM (Kruskal-Wallis p-value .001). CD24⁻THY1⁻EGFR⁺ cell proportion was statistically higher in IDHwt EGFR amplified GBM than IDH mutant astrocytoma and oligodendroglioma (Kruskal-Wallis p-value .013 and .002 respectively), and while the proportion is numerically higher than IDHwt EGFRwt, Kruskal-Wallis p-value was .173. Similarly, THY1⁻CD24⁻EGFR⁻ cell proportion was significantly higher in IDHwt EGFRwt GBM than IDH mutant astrocytoma and oligodendroglioma (Kruskal-Wallis p-value .003 and <.001 respectively), but not than IDHwt EGFR amplified GBM (Kruskal-Wallis p-value .097).

Figures 1 and 2 show the cellular composition of all primary IDH mutant astrocytoma, 1p19q co-deleted oligodendroglioma and IDHwt GBM samples. In Figure 1, the asterisks denote statistical significance as per above.

Figure 2.

The figure shows a schematic diagram of the cellular composition IDH mutant astrocytoma, 1p19q co-deleted oligodendrogliomas and IDHwt GBM based on single-cell RNA sequencing (Created in BioRender. Alnahhas, I. (2025) https://BioRender.com/s25e327).

The Effect of Chromosome 7 Gain/10 Loss on the CD24/THY1/EGFR Classification in IDHwt GBM

Of the 9 primary EGFR amplified IDHwt GBM samples in the Wang cohort, 8 had Chr 7 gain/10 loss. On the other hand, of the 23 primary EGFRwt GBM samples, 19 had Chr 7 gain/10 loss. Therefore, and in order to assess the pure effect of Chr7 gain/10 loss on the cellular composition in IDHwt GBM—isolated from the EGFR amplification effect—we compared the cellular composition of the 19 EGFRwt samples with Chr 7 gain/10 loss to the 4 EGFRwt samples that were negative for Chr 7 gain/10 loss.

Primary Chr7 gain/10 loss negative samples are composed of 85.69% of CD24⁻THY1⁻EGFR⁻ cells and 10.84% of CD24⁻THY1⁻EGFR⁺ cells. The percentage becomes 61.54% and 33.18% respectively in the recurrent setting. On the other hand, primary Chr7 gain/10 loss positive samples include on average 46.5% of CD24⁻THY1⁻EGFR⁻ cells and 45.87% of CD24-THY1-EGFR⁺ cells. The percentage becomes 43.12% and 49.56% respectively in the recurrent setting.

TERT Expression per TERTp Mutation Status

Of the 26 pairs of samples, TERTp mutation status only changed once from absent to present in the recurrent setting. Overall, TERT expression was very low regardless of the TERTp mutation status. Only 0.99% of cells express TERT when TERTp mutation is negative and 1.2% of cells express TERT when TERTp mutation is positive.

Find Differentially Expressed Markers between EGFR⁺ and EGFR⁻ Cells

We then used Seurat’s FindMarkers function to find differentially expressed genes between cells expressing EGFR and cells not expressing EGFR in IDHwt GBM samples. We performed these analyses in EGFR amplified and EGFRwt as well as primary and recurrent samples separately. Tables S2-S5 include data on the differentially expressed genes that met a significant adjusted p-value of <.01 and log fold-change of the average expression > 1 or < −1 in EGFR amplified primary and recurrent samples, and EGFRwt primary and recurrent samples. Figures S2 and S3 show volcano plots of differentially expressed genes between EGFR⁺ and EGFR⁻ cells that are shared between primary and recurrent samples.

Functional Enrichment Analysis of Differentially Expressed Genes

GSEA of the differentially expressed genes between EGFR⁺ and EGFR⁻ cells was applied. Within the IDHwt EGFR amplified samples in the primary setting (Table S6), and as expected, the VERHAAK_GBM_CLASSICAL set is highly enriched in EGFR⁺ cells (normalized enrichment score (NES) -2.3, adjusted p-value .00016). On the other hand, the VERHAAK_GBM_ MESENCHYMAL set is enriched in EGFR⁻ cells (NES 1.97, adjusted p-value .0008). Seven hypoxia gene sets are enriched in the EGFR⁻ cells but none in EGFR⁺ cells.

Within the IDHwt EGFR amplified samples in the recurrent setting (Table S7), the VERHAAK_GBM_CLASSICAL set remains enriched in EGFR⁺ cells (NES −3.38, adjusted p-value .0000000001). However, VERHAAK_GBM_PRONEURAL becomes enriched in EGFR⁻ cells (NES 3, adjusted p-value 7.741667e-08). Interestingly, 11 hypoxia gene sets are enriched in EGFR⁺ cells and one hypoxia gene set is enriched in EGFR⁻ cells.

Within the IDHwt EGFRwt samples in the primary setting (Table S8), the VERHAAK_GBM_CLASSICAL set is also enriched in EGFR⁺ cells (NES—2.9, adjusted p-value 1.499300e-08). VERHAAK_GBM_MESENCHYMAL and VERHAAK_GBM_PRONEURAL are enriched in EGFR⁻ cells (NES 2.35, adjusted p-value 3.466847e-07) and (NES 1.52, adjusted p-value 4.842140e-02), respectively. Twenty hypoxia gene sets are enriched in the EGFR⁻ cells but none in EGFR⁺ cells.

Finally, within the IDHwt EGFRwt samples in the recurrent setting (Table S9), the VERHAAK_GBM_CLASSICAL and VERHAAK_GBM_PRONEURAL sets are enriched in EGFR⁺ cells (NES −2.78, adjusted p-value 1.238314e-08) and (NES −2.07, adjusted p-value 2.092531e-02), respectively. On the other hand, the VERHAAK_GBM_ MESENCHYMAL set is enriched in EGFR⁻ cells (NES 3.03, adjusted p-value 1.215676e-08). Twenty hypoxia gene sets are enriched in the EGFR⁻ cells but none in EGFR⁺ cells.

Discussion

Advanced technologies such as next-generation sequencing and scRNA sequencing have helped to elucidate the cell-type composition as well as the complexity of cancer and its microenvironment. Moreover, single-cell glioma analyses have proposed resemblance of GBM cells to radial glia and oRG. The purpose of this study is to describe the single-cell make-up of glioma based on the WHO 2021 classification. To this purpose, we used a newly described classification of neural stem and progenitor cells from developing fetal human brain using the markers CD24, THY1 and EGFR. vRG are CD24-THY1-EGFR⁺, whereas oRG are CD24⁻THY1⁻EGFR⁻. Early neuron precursors are CD24⁺THY1-EGFR⁺ and GPCs are THY1⁺EGFR⁺. GPCs are lineage-restricted to astrocytes and oligodendrocytes but not to neurons.

We find that primary IDHwt GBM samples are mainly composed of cells that express markers similar to ventricular and oRG. Chromosome 7 gain and EGFR amplification drive a higher proportion of CD24⁻THY1⁻EGFR⁺ cells. In primary EGFR amplified samples, 87.51% of cells are CD24⁻THY1⁻EGFR⁺. Whereas THY1⁺ and CD24⁺ cells comprise <2% of cells. EGFRwt GBM that lacks chromosome 7 gain is composed of 85.69% of CD24⁻THY1⁻EGFR⁻ cells, on the other hand. In EGFRwt GBM with chromosome 7 gain, cells are roughly evenly split between CD24⁻THY1⁻EGFR⁺ and CD24⁻THY1⁻EGFR⁻ cells. The proportion of THY1⁺ and CD24⁺ cells are higher than EGFR amplified samples. The difference is striking in IDH mutant astrocytoma and oligodendroglioma. In IDH mutant astrocytoma 82.67% of cells overall are THY1⁺ and 14.75% are THY1⁻CD24⁺. Similarly, of 1p19q co-deleted oligodendrogliomas, 80.47% of cells were THY1⁺ and14.36% of cells were THY1⁻CD24⁺. Most of the THY1⁺ cells in IDH mutant glioma are also EGFR⁺PDGFRA⁺, an expression profile similar to pre-OPCs. THY1 is transcriptionally highly expressed in IDH mutant gliomas. THY1 expression has been previously reported in gliomas but not linked to IDH status.¹⁵ These data question whether by expressing different progenitor markers, IDH mutant gliomas and IDHwt GBM may have different cells of origin. Surely, this work remains purely observational and hypothesis generating and validation with bigger datasets and in-vivo functional experiments is essential before making final conclusions. Better understanding of the cell-of-origin of these tumors will help to better understand the biology of these diseases and therapy response differences.

Another interesting phenomenon is that, consistently, a different cell population than the dominant one in the primary setting becomes more prevalent in the recurrent setting. CD24⁻THY1⁻EGFR⁻ cells increase from 9.76% to 23.11% at recurrence in EGFR amplified GBM, and from 44.15% to 49.58% at recurrence in EGFRwt GBM. CD24⁺ and THY1⁺EGFR⁺PDGFRA⁻ increase in proportion in recurrent IDH mutant astrocytoma, while the percentage of the initiating population of THY1⁺EGFR⁺PDGFRA⁺ cells drops.

GSEA shows that -as expected- the VERHAAK_GBM_CLASSICAL set is consistently enriched in EGFR⁺ cells regardless of EGFR amplification status. The VERHAAK_GBM_ MESENCHYMAL set is enriched in EGFR- cells in EGFRwt primary and recurrent samples. More hypoxia gene sets are enriched in EGFR⁻ cells in EGFRwt primary and recurrent samples compared to EGFR amplified samples.

Finally, TERT expression was very low, even in the presence of TERTp mutation (1.2% of cells). This is consistent with recent literature that suggests that unlike the TERTp mutation, which was seen throughout the entire tumor, TERT expression was detected only in a subset of cells.¹⁶

We highlight the limitations of this study. First, the Wang et al. dataset utilized snRNA-seq, whereas the Ventricular et al. and Tirosh et al. datasets utilized scRNA-seq. Glioma whole and intact cells are challenging to isolate, explaining the higher number of nuclei than cells in the studies used. We used expression values >0 in designating positive and negative marker expression to address this technical difference. Moreover, in single cell sequencing, a large fraction of observed “zeros” (ie, no reads for a gene in a cell) may result from technical limitations and not represent true biological lack of expression, especially when looking at singular markers. In addition, whether the number of cells captured in each sample is sufficient to be biologically representative of the cellular composition is an intrinsic challenge to single-cell studies. Factors that come into play include the depth of sequencing and the expected frequency and fractions of the rare cell types. The Satija lab “How many cells” guidance gives an example (eg, to be 95% confident of seeing ≥5 cells from cell types at 2% fraction of the cell population, ∼619 cells total is needed). We included all the cells from the selected samples from each cohort. A few samples had limited number of cells which may have resulted in false negative results. We also note that we did not perform batch-correction across datasets. Batch effects come from technical variation across samples and datasets. While computational methods can improve data integration by reducing batch effects, they have the potential of overcorrecting the biological variability. Given that the datasets included different tumor pathologies (IDHwt GBM, IDH mutant astrocytoma and oligodendroglioma), there is true biological variation across the cohorts hindering complex and assumptious integration. We therefore analyzed each dataset independently.

We then re-iterate that this work remains purely observational and hypothesis generating and validation with bigger datasets and in-vivo functional experiments is essential.

Supplementary Material

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

Lay Summary

Gliomas are brain tumors made up of many different kinds of cells that resemble those seen in the developing brain. The authors of this study wanted to compare different types of gliomas to understand how their cell origins differ. To do this, they examined gene activity in individual tumor cells from several published datasets of glioblastoma, astrocytoma, and oligodendroglioma and grouped the cells based on patterns of genes linked to early brain development. They found that the different glioma types showed distinct gene activity patterns, suggesting they arise from different kinds of brain cells.

Author Contributions

Conception and design of the study: I.A., A.K., M.K., W.S. Acquisition and coding: I.A. Interpretation of data: I.A., A.K., M.K., W.S. Drafting and revision of the written manuscript: I.A., A.K., M.K., W.S. All authors discussed and reviewed the manuscript and approved the manuscript for publication.

Conflict of Interest Statement

The authors report no conflicts of interest.

Funding

The study received no funding.

Ethics Statement

We only used de-identified data. Under the federal regulations for human subjects (45 CFR Part 46), research involving publicly available datasets would not require institutional review board review.

Code Availability

All code used is available on GitHub (https://github.com/iyadalnahhas/scRNA_astrocytoma).

Data Availability

Publicly available data were used as per the Methods section.