Phenotypic Mapping of Pathologic Cross-Talk between Glioblastoma and Innate Immune Cells by Synthetic Genetic Tracing

Abstract

Glioblastoma is a lethal brain tumor that exhibits heterogeneity and resistance to therapy. Our understanding of tumor homeostasis is limited by a lack of genetic tools to selectively identify tumor states and fate transitions. Here, we use glioblastoma subtype signatures to construct synthetic genetic tracing cassettes and investigate tumor heterogeneity at cellular and molecular levels, in vitro and in vivo. Through synthetic locus control regions, we demonstrate that proneural glioblastoma is a hardwired identity, whereas mesenchymal glioblastoma is an adaptive and metastable cell state driven by proinflammatory and differentiation cues and DNA damage, but not hypoxia. Importantly, we discovered that innate immune cells divert glioblastoma cells to a proneural-to-mesenchymal transition that confers therapeutic resistance. Our synthetic genetic tracing methodology is simple, scalable, and widely applicable to study homeostasis in development and diseases. In glioblastoma, the method causally links distinct (micro)environmental, genetic, and pharmacologic perturbations and mesenchymal commitment. SIGNIFICANCE: Glioblastoma is heterogeneous and incurable. Here, we designed synthetic reporters to reflect the transcriptional output of tumor cell states and signaling pathways' activity. This method is generally applicable to study homeostasis in normal tissues and diseases. In glioblastoma, synthetic genetic tracing causally connects cellular and molecular heterogeneity to therapeutic responses.This article is highlighted in the In This Issue feature, p. 521.

Figure 1. Glioblastoma-subtype synthetic Locus Control Regions (sLCR).

(A) Schematic representation of glioblastoma sLCRs generation from gene expression data. (B) Pairwise correlation heatmaps of significant TFBS motifs at glioblastoma subtype-specific loci. The number of transcription factors and signature genes used in the analysis are indicated above each panel. MES=Mesenchymal; PN=Proneural; CL=Classical. (C) Above, ssGSEA normalized scores for input genes for the indicated sLCRs (methods). The cell states identified by (12) are indicated in each quadrant, and the original single-cell position is maintained in the two-dimensional representation (methods). Below, TCGA subtypes (3) are shown for a head-to-head comparison. (D) Above, schematic representation of a sLCR and of the experimental steps to generate reporter cells. Below, heatmap of MGT#1 and PNGT#2 gene expression normalized by GAPDH and number of integrations relative to hGICs=human glioma-initiating-cells; GSCs=glioma-stem-cells. Selected non-brain tumor cell lines are also shown. (E) FACS profile of IDH-wt-hGICs and IDH-mut-hGICs transduced with the indicated reporters and FACS sorted for the reporter-independent marker H2B-CFP. (F) Above, schematic representation of bulk, MGT#1- and PNGT#2-expressing hGICs’ transcriptional profiling. Below, heatmap of GSEA adjusted p-values (see methods) for the indicated glioblastoma subtypes/state-signatures and comparisons in the indicated hGIC lines.

Figure 2. In vivo genetic tracing of mesenchymal trans-differentiation.

(A) Above, schematic representation of the experiment. Below, representative coronal forebrain images of IDH-wt-hGICs-MGT#1 xenografts in NSG mice at humane end point (n=10). Lower left, HE staining; lower right, insets showing magnification of mVenus, Tubulin and DAPI counterstained tissue with invasive glioma front being homogeneously MGT#1-high. (B) Representative lesion with mixed high and low mesenchymal reporter expression. (C-D) Representative H2B-CFP expression (arrowhead) in MGT#1 positive and negative lesions, respectively. (E) Above, schematic representation of the experiment. Below, representative t-SNE map of in vitro and in vivo reporter expression for IDH-wt-hGICs with the indicated dual-reporter combination. Gating strategy is shown in (Supplementary Figure S2). (F) Relative quantification of t-SNE data in (E). (G) Representation of the dual reporters’ expression in vitro and in vivo for the indicated pairs (n=3/group). Unpaired t-test reports significance for each in vivo reporter group compared to its relative in vitro control (****P<0.0001, ns=not significant). (H) Bubble plot of GSEA adjusted p-values for the indicated glioblastoma subtypes/states and comparisons. (I) Volcano plot of the differential expression analysis between in vitro PNGT#2-high and in vivo MGT#1-high. Selected genes are highlighted. (J) Ingenuity pathway top upstream regulator analysis of differential expression analysis in (I).

Figure 3. The mesenchymal glioblastoma identity is adaptive and reversible.

(A) Schematic description of phenotypic screening using sLCRs. (B) Bubble plot visualization of the screening of the indicated factors regulating MGT#1 in IDH-mut- and IDH-wt-hGICs (left) or MGT#1, MGT#2, PNGT#1 or PNGT#2 in IDH-wt-hGICs (right). Bubble size and color indicate the magnitude and the direction of the change. (C-D) Bar plot showing the individual response to the indicated factors/sLCRs after 48h of induction. (E-F) Representation of longitudinal expression of the MGT#1/2-mVenus in response to the indicated factors starting from day 0 (stimulation). The arrows indicate the time-point for cytokines withdrawal. (G) Bubble plot of GSEA adjusted p-values for the indicated glioblastoma subtypes/states and comparisons between the identified MES-inducing stimuli. (H) Upset plot of all intersections for the indicated MGT#1 activation cues sorted by intersection size. Interconnected circles in the matrix indicate common genes.

Figure 4. Mesenchymal glioblastoma genetic tracing reveals a swift cell state change driven by ionizing radiation but not hypoxia.

(A) MGT#1 activation in response to increasing doses of ionizing radiation (IR), at the 72h time-point. The insert shows the immunoblotting of the indicated antibodies and condition at 1h post ionizing radiation delivery. (B) Heatmap of ionizing radiation-induced significantly differentially regulated genes (padj<0.05 and log2FC±1.5) in MGT#1-high IDH-wt-hGICs fraction (pink, n=3) against non-irradiated control cells (blue, n=3). (C) GSEA plots for the indicated gene sets. (D) Representative FACS quantification of indicated sLCRs under the hypoxic (blue), low oxygen (green) or normoxic (pink) conditions. (E) Above, a schematic overview of the RNA-seq experimental design. Below, heatmap with differentially regulated genes of comparison between the hypoxic (blue, n=3) and normoxic (pink, n=3) conditions (padj<0.05 and log2FC±1.5). Heatmap color-coding is based on relative rlog-normalized gene expression values across samples. (F) Bubble plot of top gene sets enriched in response to hypoxia. Color codes and size indicate significance and gene ratio, respectively.

Figure 5. Genetic and pharmacological modulation of the mesenchymal state.

(A) Experimental design for the functional dissection of MGT#1 activation. (B) Volcano plot of sgRNA targets regulating MGT#1 expression in the screen (A). Fold-changes were calculated for all MGT#1-high fractions (naïve, TMZ+IR, TNFα+FBS, n=3 each) relative to all MGT#1-low fractions and unsorted controls (n=6 each). SGF29/CCDC101 highlighted in red as one of the top significantly upregulated hits within the comparison. sgRNAs associated with RAR/RXR agonist Tretinoin are labeled in yellow. (C) Representative FACS profiles of MGT#1 activation by the indicated conditions in SGF29 KO or control IDH-wt-hGICs-MGT#1 cells. (D) Ingenuity pathway analysis top upstream regulators of differential expression analysis in (B). Categories associated with acute inflammation pathways are in bold. (E) Representative FACS quantification of MGT#1 activation by the indicated treatments at 48h. (F) Top Ingenuity Pathway Analysis toxicity categories of differential expression analysis in (B). Categories associated with retinoic receptors signaling are in bold. (G) Bar plot representation of IDH-wt-hGICs-MGT#1 cells relative viability upon the indicated treatments. Error bars=±SEM, 2-way ANOVA followed by Welch’s correction (n=5, *p<0.05, ***p<0.001 and ****p<0.0001).

Figure 6. Innate immune cells drive non-cell autonomous mesenchymal commitment in tumor cells.

(A) Schematic representation of contact-free hGICs co-culture with immune cells (see methods). (B) Representative FACS profiles of IDH-wt- or IDH-mut-hGICsMGT#1 alone or co-cultured with human microglia (hMG#cl.20) or human CD34+ in vitro-derived myeloid-derived suppressor-like cells (MDSCs). (C) Representative FACS profiles of IDH-wt-hGICs-MGT#1 alone or co-cultured for the indicated time with human THP1-derived M1 or M2 macrophage-like cells. (D) Representative FACS profiles and gating strategy of IDH-wt-hGICs-MGT#1 alone or stimulated with TNFα or hMG co-culture. Below, Venn diagram of NFκB-related genes by Ingenuity Pathway Analysis of DRGs for the indicated conditions. DRGs are computed relative to control hGICs (log2FC>1, padj<0.05). (E) Heatmap of DRGs for the indicated conditions. RNA-seq reads were normalized as transcript per million, log2 transformed and z-scored. Statistical significance was assessed by using the limma R-package (control, n=3, hMG, n=3; TNFα n=2; padj<0.05). (F) Ingenuity Upstream Regulator Analysis of up-regulated genes by hMG co-culture compared to TNFα in IDH-wt-hGICs-MGT#1-high. (G) UMAP dimensional reduction of MGT#1 activation cues expression profiles combining all up-regulated genes (see methods). (H) Upset plot of all intersections for the indicated MGT#1 activation cue comparison sorted by intersection size. Interconnected circles in the matrix indicate common genes.

Figure 7. Therapeutic implications of phenotypic changes in glioma initiating cells driven by innate immune cells.

(A) Heatmap of the relative ssGSEA normalized score for the indicated gene sets in glioblastoma patients from the TCGA dataset. Including gene sets representing specific GBM subtype/state and up-regulated MGT#1 activation cues (Fig. 6 G–H). The status of IDH1 and NF1 mutations and the corresponding GBM subtypes are also indicated. (B) ssGSEA normalized scores for up-regulated MGT#1-high genes indicated in Figure 6D (see methods). Cell states identified by (12) are indicated in each quadrant, and the original dots position is maintained in the two-dimensional representation of GBM cell states (or meta-modules; methods). (C-D) Differential GSEA for the indicated comparisons. Significance is independently calculated by t-test and Kolmogorov-Smirnov. (E) Left, schematic depiction of chemosensitivity profiling of sLCR high and low states. Right, dot denotes log[IC₅₀] value in response to increasing concentrations of the indicated drugs for FACS-sorted MGT#1-high and MGT#1-low fractions of the indicated genotypes. Dotted line indicates threshold of 10μM concentration, unattainable in the brain tissue. (F) Dose-response curves of FACS sorted MGT 1-high, -low or na ve IDH-wt-hGICs subjected to increasing concentrations of selected compounds as summarized in (E). (G) A model for modulation of mesenchymal fate in glioblastoma.