Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2017 Mar 31;355(6332):eaai8478.
doi: 10.1126/science.aai8478.

Decoupling Genetics, Lineages, and Microenvironment in IDH-mutant Gliomas by Single-Cell RNA-seq

Free PMC article

Decoupling Genetics, Lineages, and Microenvironment in IDH-mutant Gliomas by Single-Cell RNA-seq

Andrew S Venteicher et al. Science. .
Free PMC article


Tumor subclasses differ according to the genotypes and phenotypes of malignant cells as well as the composition of the tumor microenvironment (TME). We dissected these influences in isocitrate dehydrogenase (IDH)-mutant gliomas by combining 14,226 single-cell RNA sequencing (RNA-seq) profiles from 16 patient samples with bulk RNA-seq profiles from 165 patient samples. Differences in bulk profiles between IDH-mutant astrocytoma and oligodendroglioma can be primarily explained by distinct TME and signature genetic events, whereas both tumor types share similar developmental hierarchies and lineages of glial differentiation. As tumor grade increases, we find enhanced proliferation of malignant cells, larger pools of undifferentiated glioma cells, and an increase in macrophage over microglia expression programs in TME. Our work provides a unifying model for IDH-mutant gliomas and a general framework for dissecting the differences among human tumor subclasses.


Fig. 1
Fig. 1. Expression differences between IDH-A and IDH-O are governed by the tumor microenvironment and genetics
(A) Workflow. Freshly resected tumors were dissociated to single-cell suspension, sorted by FACS, and profiled by Smart-seq2 in 96-well plates. (B) Differential expression between IDH-A and IDH-O across bulk TCGA tumors (left) and across single cells (center), and the averages from each of these two analyses (right). (C) Differentially expressed genes by bulk analysis include microglia/macrophage-specific genes (left column) and neuron-specific genes (right column). (D) Distribution of expression differences between bulk IDH-A and IDH-O samples for microglia/macrophage-specific genes (black) and neuron-specific genes (gray). (E) Microglia/macrophage scores versus neuron scores (11) for bulk IDH-O (blue) and IDH-A (purple) tumors. (F) Left: Differentially expressed genes that are neither microglia/macrophage-specific nor neuron-specific, assigned to four categories of genetic influences (11), from top to bottom: genes residing in chromosome arms 1p or 19q, genes activated by CIC, genes repressed by CIC, and P53 target genes. Right: Observed and expected percentages of IDH-A–specific genes assigned to the first two categories and IDH-O–specific genes assigned to the last two categories. Expected percentages were defined by analysis of all genes rather than only the IDH-A– and IDH-O–specific genes.
Fig. 2
Fig. 2. Glial lineages are shared among IDH-A and IDH-O
(A) Average expression levels of oligodendrocyte-specific and astrocyte-specific genes across all IDH-A (y axis) and IDH-O (x axis) malignant cells. (B) Correlations of oligodendrocyte-specific and astrocyte-specific genes with PC1 (x axis) and PC2 (y axis) from a PCA of all IDH-A malignant cells. (C) Classification of malignant cells (columns) from IDH-A and IDH-O according to the differential expression of 50 oligodendrocytic and 50 astrocytic genes. Bottom: Relative expression of the 100 genes (rows). Top: Significance of differential expression [−log10(P value of a t test)] between oligodendrocytic and astrocytic genes. Cells were sorted by significance from the most oligodendrocytic-like to the most astrocytic-like cells; dashed lines indicate a significance threshold of P <0.01. (D) For each malignant cell in IDH-A and IDH-O, we show its differentiation scores (x axis, maximum of oligodendrocytic and astrocytic scores) versus the average expression of IDH-A–specific or IDH-O–specific genes (left and right y axes, excluding those genes that exhibit differential expression due to genetic alterations). Lines indicate the corresponding local weighted smoothing regression (LOWESS), demonstrating the decreased differences between IDH-A and IDH-O programs in cells with low glial differentiation scores.
Fig. 3
Fig. 3. Undifferentiated cells in IDH-A and IDH-O are associated with cycling cells and a putative stemness program
(A) Percentage of cycling cells (y axis) in sliding windows of 200 cells ranked by differentiation scores (x axis) for either IDH-A or IDH-O malignant cells. (B) Pearson correlations (color scale) between the expression profiles of 90 genes preferentially expressed in undifferentiated cells, across IDH-A (top) and IDH-O (bottom) undifferentiated cells. Genes are ordered by their correlation with the highest-scoring cluster in each analysis (11). (C) Pearson correlations of the 90 genes in (B) with the highest-scoring clusters in (B) in IDH-A (x axis) and IDH-O (y axis). The most consistent genes are labeled. (D) In situ RNA hybridization shows mutually exclusive expression of astrocytic (APOE) and oligodendrocytic (APOD) lineage markers; mutually exclusive expression of astrocytic and proliferation (Ki-67, arrow) markers; and coexpression of proliferation and stem/progenitor (SOX4, arrow) markers.
Fig. 4
Fig. 4. Analysis of tumor architecture by tumor grade and in genetic subclones
(A) The percentage of cycling cells (top), percentage of undifferentiated cells (middle), and negative correlation between the two lineage scores (bottom) are all associated with tumor grade (P < 0.05, one-way analysis of variance). For each feature, bars show the average value across groups of tumors defined by tumor type and grade. Error bars denote SE. (B and C) CNV inference in MGH103 (B) and MGH57 (C) reveals large-scale CNVs that vary between cells of the same tumor. Cells were clustered on the basis of their CNV patterns at specific chromosomal regions (black bars at top) to define putative subclones. (D and E) Comparison of the two lineage scores (left) and percentage of cycling cells (right) between the two subclones indicated for MGH103 (D) and for MGH57 (E). Significant differences are indicated (*P <0.05, **P < 0.001; Kolmogorov-Smirnov test for lineages and hypergeometric test for cell cycle).
Fig. 5
Fig. 5. Microglia and macrophages across IDH-mutant gliomas
(A) Microglia (y axis) and macrophage (x axis) expression levels (32) of genes with high and low PC1 scores from PCA of tumor microglia/macrophages. (B) Top: Distribution of scores by average expression of microglia (PC1-high) versus macrophage (PC1-low) genes (11). Bottom: Differential expression of selected microglia- and macrophage-specific genes among all cells ranked by the scores at top. (C)Fraction (color code) of cells in bins of scores, as defined in (B), top, for each glioma; macrophages from melanoma (5) are included for reference (top row). Tumor grades are indicated at the right. (D) Average endothelial scores (x axis) versus macrophage or microglia (y axis) across IDH-A and IDH-O tumors of grades II to IV. Arrows indicate grade-specific changes associated with increased expression of endothelial program. (E) Correlation between endothelial scores and macrophage/microglia scores across all IDH-A or IDH-O bulk TCGA tumors. (F)In situ RNA hybridization for microglia (CX3CR1) and macrophage (CD163) markers. Left to right: MGH56 contains a few CX3CR1-positive cells and is negative for CD163. MGH43 contains microglia-like cells and macrophage-like cells (two blood vessels are highlighted by arrows). MGH43 contains cells expressing both CD163 and CX3CR1 (three cells highlighted by arrows). MGH42 stains exclusively for CD163.

Similar articles

See all similar articles

Cited by 141 articles

See all "Cited by" articles

Publication types