The expression profile and tumorigenic mechanisms of CD97 (ADGRE5) in glioblastoma render it a targetable vulnerability

Abstract

Glioblastoma (GBM) is the most common and aggressive primary brain malignancy. Adhesion G protein-coupled receptors (aGPCRs) have attracted interest for their potential as treatment targets. Here, we show that CD97 (ADGRE5) is the most promising aGPCR target in GBM, by virtue of its de novo expression compared to healthy brain tissue. CD97 knockdown or knockout significantly reduces the tumor initiation capacity of patient-derived GBM cultures (PDGCs) in vitro and in vivo. We find that CD97 promotes glycolytic metabolism via the mitogen-activated protein kinase (MAPK) pathway, which depends on phosphorylation of its C terminus and recruitment of β-arrestin. We also demonstrate that THY1/CD90 is a likely CD97 ligand in GBM. Lastly, we show that an anti-CD97 antibody-drug conjugate selectively kills tumor cells in vitro. Our studies identify CD97 as a regulator of tumor metabolism, elucidate mechanisms of receptor activation and signaling, and provide strong scientific rationale for developing biologics to target it therapeutically in GBM.

Keywords: ADGRE5; CD97; CP: Cancer; Warburg metabolism; adhesion G protein-coupled receptor; antibody-drug conjugate; glioblastoma; receptor signaling.

Figure 1.. CD97 (ADGRE5) is de novo expressed in GBM.

(A) Heatmap showing log₂ expression of transcripts encoding aGPCRs and known markers in healthy brain tissue and GBM (log₂(RPKM)). CD97 (ADGRE5) is indicated by an arrow. (B) Heatmap showing protein expression of aGPCRs and known markers in healthy brain tissue and GBM. CD97 (ADGRE5) is indicated by an arrow. (C) Volcano plot showing upregulated genes in a PDGC compared to human NSCs. All aGPCRs are labeled. CD97 (circled) was significantly upregulated (adjusted p value [p_adj] < 0.00001, log₂(fold change) > 1.5). (D) Transcripts from bulk RNA-seq data collected from two PDGC replicates and two NSC replicates were aligned along the CD97 locus. The red arrowhead signifies the first CD97 exon. (E) Immunofluorescent and Hoechst 33342-stained specimen of human temporal lobe (healthy) and GBM. MAP2A is a neuronal marker. (F) Confocal microscopy images of a GBM specimen stained with an antibody against the CD97 extracellular domain (ECD) (red) and DAPI (blue). A merged color image is accompanied by black and white single-channel images. (G) CD97 surface staining via flow cytometry using an allophycocyanin (APC)-conjugated antibody in proneural (n = 6), classical (n = 6), and mesenchymal (n = 4) PDGCs, NSCs (n = 6), and NHAs (n = 4) (unpaired t test; ****p < 0.0001). (H) Single-cell RNA/ATAC-seq data from IDH-mutant astrocytoma and IDH-WT GBM surgical specimens. Single cells are color-coded based on their tumor of origin. (I) Cluster identities were characterized and named based on their corresponding gene expression profiles. Single cells are colored based on broad cluster identities. (J) CD97 expression in cell clusters based on integrated data. (K) ATAC peaks associated with the CD97 locus were extracted from single-cell ATAC-seq and integrated based on the clusters in (I). The red arrowhead signifies the CD97 promoter. Also shown are peaks corresponding to the DDX39A promoter, which is expressed in all cell clusters. (L) H3K27ac ChIP-seq from the GEO shows peaks associated with the CD97 locus for IDH-WT GBM and IDH-mutant anaplastic astrocytoma. The red arrowhead signifies the CD97 promoter. (M) Schematic representation of the structure of three CD97 splice isoforms. EGF, epidermal growth factor-like repeat; RGD, arginylglycylaspartic acid domain; GAIN, GPCR autoproteolysis-inducing domain; GPS, GPCR proteolysis site; NTF, N-terminal fragment; CTF, C-terminal fragment. Note that for isoforms 2 and 3, only the NTF is shown. (N) Bulk RNA-seq data from the GBM dataset of TCGA Splice Variant Database (TSVdb) were used to quantify CD97 isoform expression. Orange boxes are centered around the median and encompass the two middle quartiles (n = 166 per isoform; ANOVA F_2,330 = 283.3, p < 0.0001; Tukey’s multiple comparisons test: isoform 1 vs. isoform 2: ****p < 0.0001; isoform 1 vs. isoform 3: ****p < 0.0001; isoform 2 vs. isoform 3: ****p < 0.0001). (O) Schematic representation of the tet-inducible vector used for CD97 isoform overexpression. (P) Immunoblots for the CD97 intracellular domain (ICD, equivalent to the CTF) and the ECD (equivalent to NTF). CD97 S531A is an uncleavable point mutant of isoform 1. Non-specific bands are indicated by a hash symbol (#). The same membrane was stained for CD97 ICD and then stripped for CD97 ECD staining. Error bars indicate SEM.

Figure 2.. CD97 is essential for tumor initiation in vitro and in vivo.

(A) Schematic representations of the shRNA and CRISPR constructs used to knock down and knock out CD97, respectively. (B) Histograms of CD97 surface staining following knockdown or overexpression of isoform 3 of CD97 in a PDGC. (C) Bar graph visualizing tumorsphere formation following knockdown of CD97 and followed by rescue with a dox-inducible and a shRNA-resistant form of CD97 isoform 3 (n = 5 for each PDGC; two-way ANOVA F_3,36 = 9.795, p < 0.0001; Tukey’s multiple comparisons test: empty vector (EV) SCRsh vs. EV CD97sh: ***p < 0.001; EV CD97sh vs. overexpressed (OE) CD97sh: *p < 0.01; OE SCRsh vs. OE CD97sh: ns, p > 0.05). (D) Representative wells from the tumorsphere formation assay quantified in (C) before and after dox-induced rescue. (E) ELDA plots for three PDGCs are displayed. (F) Summary plot showing calculated clonogenic frequencies for all tested PDGCs after CD97 knockdown based on ELDA assays. Paired samples are connected with a line (n = 3 per PDGC; two-way ANOVA F_1,28 = 75.24, p < 0.0001). (G) Graph showing compromised viability of PDGCs infected with three separate gRNAs against CD97 (orange) compared to a control gRNA against the human homolog of ROSA26 (black) in a cell competition assay (n = 3 per gRNA; two-way ANOVA F_4,40 = 10.37, p < 0.001; Tukey’s multiple comparisons test: ROSA26 vs. CD97: days 4, 8, and 12: ns, p > 0.05; days 16 and 20: ***p < 0.001). (H) Bioluminescent images taken with IVIS of intracranial GBM xenografts 90 days after injection. (I) Immunofluorescent images of mouse brains injected with PDGCs lentivirally infected with the indicated shRNA construct, the TagRFP fluorophore, and luciferase. White arrows indicate the resulting tumor at the site of injection. (J) Graph depicting the average bioluminescent radiance captured by IVIS for mice harboring PDGC xenografts (n = 9 mice per group; unpaired t test; *p < 0.05). (K) Kaplan-Meier survival curve showing significantly increased survival of mice xenografted with a PDGC harboring CD97 shRNA (n = 5 mice per group; log-rank test; **p < 0.01). Error bars indicate SEM.

Figure 3.. CD97 knockdown impairs glycolytic metabolism in GBM.

(A) Volcano plot derived from bulk RNA-seq data collected from PDGCs lentivirally infected with the SCR or CD97 shRNA (n = 3 biological replicates for each). Genes involved in canonical glycolysis and glucose transport are represented as large yellow points. Genes involved in the TCA cycle and OXPHOS (cytochrome oxidase subunits) are represented as large green points. CD97 is represented by an orange point. (B) The top ten enriched and depleted pathways determined by the largest fold enrichment among downregulated genes using GO PANTHER pathway enrichment analysis. Stars indicate metabolic pathways. (C) Correlation matrix from bulk RNA-seq data collected from PDGCs following knockdown or overexpression of CD97 shows high correlation between CD97 and glycolysis-related genes and anti-correlation with TCA cycle-related genes. HK2 (hexokinase 2) and SLC2A3 (glucose transporter 3 [GLUT3]) transcripts are both included because they have been implicated in Warburg metabolism. (D) Bar graph showing decreased lactate production after knockdown of CD97 (n = 6 per PDGC; two-way ANOVA F_1,15 = 27.24, p < 0.001) and increased lactate production after CD97 overexpression in PDGCs (n = 3 per PDGC; two-way ANOVA F_1,6 = 26.07, p < 0.01). (E) Steady-state metabolomic data reveal depletion of glycolytic metabolites after knockdown of CD97 in PDGCs (PN [proneural], n = 2–3 [one replicate was removed for technical reasons]; CL [classical], n = 3). (F) Depleted (red) and enriched (green) heavy-labeled glycolytic and TCA cycle metabolites after a heavy-labeled glucose tracing experiment. (G) Representative Seahorse Cell Energy Phenotype graph showing OCR and ECAR changes before (baseline) and after (maximal) addition of mitochondrial stressors. (H and I) Bar graphs quantifying baseline and maximal ECAR (n = 6 per PDGC; baseline: two-way ANOVA F_1,10 = 14.06; **p < 0.01; maximal: two-way ANOVA F_1,10 = 17.87, p < 0.01). (J and K) Bar graphs quantifying baseline and maximal OCR (n = 6 per PDGC; baseline: two-way ANOVA F_1,10 = 2.820, p > 0.05; maximal: two-way ANOVA F_1,10 = 5.474; *p < 0.05). Error bars indicate SEM.

Figure 4.. CD97 activates the MAPK signaling pathway to regulate glycolysis and tumorsphere formation.

(A) Immunoblot for phosphorylated ERK1/ERK2, total ERK1/ERK2, and GAPDH in PDGCs following CD97 knockdown. Red pixels indicate saturation. Quantification of densitometry ratios is also shown (phosphorylated [p]-ERK/ERK: n = 5–6 per PDGC; two-way ANOVA F_1,17 = 8.407, p < 0.01; ERK/GAPDH: n = 5–6 per PDGC; two-way ANOVA F_1,17 = 2.796, p > 0.05). (B) Immunoblot for p-ERK1/ERK2, total ERK1/ERK2, and GAPDH in a PDGC following dox-induced overexpression of CD97. Quantification of densitometry ratios is also shown (all isoforms compiled for each PDGC) (p-ERK/ERK: n = 4–8 per PDGC; two-way ANOVA F_1,21 = 10.98, p < 0.01; ERK/GAPDH: n = 4–8; two-way ANOVA F_1,21 = 1.554, p > 0.05). (C) Bar graph showing that CD97 overexpression rescues lactate levels after CD97 knockdown (n = 3–6; ANOVA F_3,15 = 25.75, p < 0.0001; Tukey’s multiple comparisons test: EV SCRsh vs. EV CD97sh: ***p < 0.001; CD97 rescue SCRsh vs. CD97 rescue CD97sh: ns p > 0.05). (D) Bar graph showing restored lactate levels upon rescue with the MEK^DD mutant (n = 3–8; ANOVA F_3,19 = 8.216, p < 0.01; Tukey’s multiple comparisons test: EV SCRsh vs. EV CD97sh: *p < 0.05; MEK^DD SCRsh vs. MEK^DD CD97sh: ns p > 0.05). (E) Seahorse Cell Energy Phenotype graphs showing OCR and ECAR changes before (baseline) and after (maximal) addition of mitochondrial stressors in knockdown cells after introduction of the MEK^DD mutant or an EV control. (F) Quantification of the difference in baseline OCR and ECAR between SCR and CD97 knockdown PDGCs after introduction of the MEK^DD mutant or an EV control (n = 13 [all PDGCs compiled]; unpaired t test; ns p > 0.05; **p < 0.01). (G) Bar graph visualizing tumorsphere formation following CD97 knockdown and introduction of the MEK^DD mutant (n = 8–9 per PDGC; two-way ANOVA F_3,66 = 8.707, p < 0.0001; Tukey’s multiple comparisons test: EV SCRsh vs. EV CD97sh: **p < 0.01; MEK^DD SCRsh vs. MEK^DD CD97sh: ns p > 0.05). (H) Representative wells from the assay in (G). Error bars indicate SEM.

Figure 5.. CD97 activates the MAPK signaling cascade through recruitment of β-arrestin to its phosphorylated C terminus and by association with THY1/CD90.

(A) Schematic of possible CD97 signaling mechanisms tested. Listed in red are the methods for testing. (B) Phosphoproteomic data collected from GBM samples identify five phosphorylation sites on the CD97 cytosolic C terminus. The heatmap depicts the percentage of GBM samples with the detected phosphorylation site. (C) Homogenous time-resolved fluorescence (HTRF) ratios from a β-arrestin recruitment assay performed after overexpression of WT CD97 or the ΔPS mutant. Two PDGCs (n = 2 for each) are compiled (n = 4 per condition; unpaired t test; ns p > 0.05; *p < 0.05; **p < 0.01). (D) Immunoblot for p-ERK1/ERK2 after overexpression of WT CD97 or the ΔPS mutant. (E) Bar graphs displaying densitometry ratios from (D) (n = 7; paired t test; ns p > 0.05; *p < 0.05). Two PDGCs (n = 3–4 for each) are compiled. (F) Bar graphs depicting the number of tumorspheres in a tumorsphere formation assay after CD97 knockdown followed by overexpression of shRNA-resistant forms of WT CD97 or the ΔPS mutant. The SCR shRNA groups used for normalization are not shown (n = 4 per PDGC; ANOVA F_5,12 = 80.79, p < 0.0001; Tukey’s multiple comparisons test: EV vs. WT: ***p < 0.001; EV vs. ΔPS: ns p > 0.05; WT vs. ΔPS: *p < 0.05). (G‒I) Single-cell RNA-/ATAC-seq data from Figure 1I showing expression of CD97, CD55, and THY1/CD90. (J) A bar graph depicting the percentage of positive cells measured by flow cytometry after surface staining of PDGCs with fluorescein isothiocyanate (FITC)-conjugated antibodies against CD55, THY1/CD90, and an IgG control (n = 3 for each PDGC; two-way ANOVA F_2,20 = 95.11, p < 0.0001; Tukey’s multiple comparisons test: IgG vs. CD55: **p < 0.01; IgG vs. CD90: ****p < 0.00001; CD55 vs. CD90: ****p < 0.01). (K) Immunoblot for p-ERK1/ERK2 from CD97-overexpressing PDGCs plated on laminin or recombinant forms of putative ligands CD55 and THY1/CD90. (L) Quantification of densitometry ratios from immunoblots in (K) (n = 2–4 per PDGC; EV p-ERK/ERK: two-way ANOVA F_2,6 = 0.9450, p > 0.05; EV ERK/GAPDH: two-way ANOVA F_2,6 = 1.047, p > 0.05; CD97 overexpression p-ERK/ERK: two-way ANOVA F_2,16 = 12.48, p < 0.001; Tukey’s multiple comparisons test: laminin vs. CD55: ns, p < 0.05; laminin vs. CD90: ***p < 0.001; CD97 overexpression ERK/GAPDH: two-way ANOVA F_2,16 = 2.257, ns p > 0.05). Error bars indicate SEM.

Figure 6.. A CD97-targeting antibody-drug conjugate demonstrates enhanced killing of PDGCs compared to NSCs and NHAs.

(A) Schematic of the MMAF-conjugated ADC binding to the CD97 receptor. (B) Dose-response curves generated from WST8 cell viability assays (n = 3 biological replicates based on technical triplicates) after treatment of PDGCs, U87 GBM cells, NHAs, NSCs, and HEK293 cells with the CD97 ADC. (C) Bar graph displaying the LD₅₀ values (in nM) for the CD97 ADC on all six cell lines as calculated from dose-response curves in (B). Curves were based on a nonlinear regression fit. (D) Representative images of Hoechst 33342-stained cells after 6 days of treatment with different concentrations of the CD97 ADC. (E) Quantification of Hoechst intensity in Figure 5D. Values are normalized to the 0 nM condition. NHAs, NSCs, and HEK293 cells are compiled (non-GBM; black) and the three GBM lines are compiled (GBM; red) to make differences more evident (n = 1 for each cell line; two-way ANOVA F_2,8 = 12.13, p < 0.01; Tukey’s multiple comparisons test: non-GBM: 0 vs. 1.2 nM: ns p > 0.05; 0 vs. 6 nM: *p < 0.05; GBM: 0 vs. 1.2 nM: *p < 0.05; 0 vs. 6 nM: *p < 0.05). (F) Diagram of the CD97 ADC killing CD97-expressing GBM cells and sparing healthy brain cells. Error bars indicate SEM.