Kinome-wide shRNA screen identifies the receptor tyrosine kinase AXL as a key regulator for mesenchymal glioblastoma stem-like cells

Abstract

Glioblastoma is a highly lethal cancer for which novel therapeutics are urgently needed. Two distinct subtypes of glioblastoma stem-like cells (GSCs) were recently identified: mesenchymal (MES) and proneural (PN). To identify mechanisms to target the more aggressive MES GSCs, we combined transcriptomic expression analysis and kinome-wide short hairpin RNA screening of MES and PN GSCs. In comparison to PN GSCs, we found significant upregulation and phosphorylation of the receptor tyrosine kinase AXL in MES GSCs. Knockdown of AXL significantly decreased MES GSC self-renewal capacity in vitro and inhibited the growth of glioblastoma patient-derived xenografts. Moreover, inhibition of AXL with shRNA or pharmacologic inhibitors also increased cell death significantly more in MES GSCs. Clinically, AXL expression was elevated in the MES GBM subtype and significantly correlated with poor prognosis in multiple cancers. In conclusion, we identified AXL as a potential molecular target for novel approaches to treat glioblastoma and other solid cancers.

Figure 1

Patient-Derived PN and MES GSCs Used for In Vitro Screening (A) MRI captions taken pre-operation and post-operation of GBM patients (upper) and H&E staining of the xenotransplanted cells in NOD/SCID mice (lower). Magnification is 20× (left) and 100× (right). (B) Representative FACS plots demonstrating expression of CD133 and CD44 PN GSCs (PN_528) and in MES GSCs (MES_83). (C) Neurosphere formation assay of CD133+ versus CD133- PN_528 GBM cells. CD133 mRNA expression level in sorted cells was verified by qRT-PCR (RT-PCR). Error bars represent the SD from three technical replicates. Stem cell frequency was calculated by ELDA analysis. (D) Flowchart depicting the experimental procedure used to investigate kinases whose silencing induces cell death or impairs proliferation. See also Figure S1.

Figure 2

Knockdown of a Subset of Kinases Differentially Induces Cell Death in MES or PN GSCs (A) List of kinases that induce cell death significantly (^∗, adjusted p < 0.05) in MES GSCs (red) or in PN GSCs (blue). mRNA expression FC between PN and MES GSCs is indicated on the right side of the heatmap. (B) Stacked bar charts of the top two genes inducing cell death in MES (above) or in PN (below) GSCs. Average of the shNT used for normalization and the shRNAs targeting the indicated gene is shown. Error bars represent the SD. (C) Venn diagram grouping kinases whose silencing induces a significant change of the cell-cycle phases specific to MES or to PN GSCs (see also Figures S2 and S3). Asterisks highlight 54 kinases whose silencing altered the cell cycle specifically in MES GSCs. (D) Ranking of AXL according to the cell number FC in SubG1 in MES and PN GSCs. Data are all initially normalized to the non-targeting shRNA.

Figure 3

AXL Is More Highly Expressed in MES Than PN GSCs (A) Microarray analysis of AXL mRNA expression in PN GSCs from individual patients (n = 18), MES GSCs (n = 12), astrocytes (n = 5), and neural progenitors (n = 3) (Mao et al., 2013) (^∗p < 0.05; ^∗∗p < 0.01; ^∗∗∗∗p < 0.0001; Mann Whitney t test in comparison to MES GSCs). (B) AXL mRNA expression as determined by qRT-PCR in PN (n = 3) and MES (n = 3) (^∗∗∗p < 0.0001; t test). Error bars represent the SD from three technical replicates. (C) Representative western blot of total AXL and phospho-AXL (pY779) expression in PN (528, 19, and 84) and MES GSCs (83, 326, and 1123). (D) Representative immunofluorescent images of total AXL and phospho-AXL staining in MES and PN GSCs (scale bar represents 50 μm). (E and F) mRNA expression of AXL as determined by gene expression profiling from previously published datasets: (E) De Bacco et al. (2012) (^∗∗p < 0.01, unpaired t test) and (F) Bhat et al. (2013) (^∗p = 0.0124, unpaired t test). (G) AXL expression is higher in the MES subtype (n = 18) compared with the IDH (n = 5), RTKI (n = 5), and RTKII (n = 28) subtypes, as determined by RNaseq data from individual patients (TCGA dataset, classification by Sturm et al., 2012). Error bars represent the minimum and maximum data points (^∗p < 0.05; unpaired t test). (H) AXL is higher in the MES subtype (n = 55) compared with the classical (n = 42), PN (n = 31), and G-CIMP (n = 8) subtypes but not to the neural (n = 28) subtype, as determined by RNaseq data from individual patients (TCGA dataset) (^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗∗p < 0.0001; unpaired t test). Error bars represent the minimum and maximum data points.

Figure 4

AXL Is Required for the Growth and Clonogenicity of MES GSCs (A) Average FC of cell number in different phases of the cell cycle upon AXL knockdown normalized to shNT. Error bars represent the SD from 5 independent shRNAs targeting AXL. (B) Representative western blot results of shNT and shAXL samples from 83 and 1123 GSCs with GAPDH as a loading control (upper). AXL expression was quantified and normalized to GAPDH (lower). (C) In vitro growth assay in 83 and 1123 MES GSCs transduced with shNT or two different shRNA targeting AXL (shAXL#1 and #2) measured at days 2, 4, and 6. Error bars represent the SD of three technical replicates. (D) Representative pictures of GSCs cultivated as neurospheres transduced with shNT or AXL directed shAXL#1 or shAXL#2. Pictures were taken 5 days after transduction. Scale bar represents 60 μm. (E) Limiting dilution neurosphere formation assay in MES (83 and 1123) and PN (528) GSCs. Cells were transduced with shNT or shAXL#1 or shAXL#2 in ten technical replicates. Readout was performed 12 days after transduction (^∗p < 0.5; ^∗∗p < 0.01; ^∗∗∗p < 0.001; linear regression analysis). Stem cell frequency was calculated by ELDA analysis. (F) Caspase activity assay of PN 528 GSCs and MES GSCs (83 and 1123) 4 days after transduction with shNT and shAXL#1 and shAXL#2. Results are normalized to shNT. Error bars represent the SD of three biological replicates. Significance was assessed by comparing the rate of caspase activity for each shRNA in the PN GSC line with the MES GSC lines (N.S. = not significant; ^∗∗∗p < 0.005; ^∗∗∗∗p < 0.001; paired t test). (G) AXL inhibitor treatment effects on MES_83 and PN_528 GSC growth. Results were obtained 72 hr after BGB324 treatment at the indicated concentrations (IC₅₀ of 1.027 μM [MES_83] and 2.035 μM [PN_528]). Error bars represent the SD of three technical replicates. Western blot phospho-AXL expression in MES 83 GSCs with or without BGB324 treatment for 24 hr. β-Actin served as loading control.

Figure 5

AXL Contributes to Tumor Propagation of MES GSCs (A) Representative photographs of mouse brains injected with 83 and 1123 GSCs transduced with shNT or shAXL#2 and representative H&E staining of shNT and shAXL mouse xenografts at the indicated time points. (B) Kaplan-Meier survival curves for mice injected with 83 and 1123 GSCs transduced with shNT (n = 6 and n = 5, respectively) and shAXL#2 (n = 6 and n = 5, respectively) (see also Figure S4).

Figure 6

AXL Expression Correlates with CD44 (A) AXL mRNA expression significantly correlates with CD44 mRNA expression. RNaseq data from n = 169 GBM patients are represented. (B) Relative mRNA expression of AXL and CD44 upon AXL knockdown. Data are normalized to shNT and are represented as mean of technical replicates ± SD. (C) Immunofluorescent staining of AXL and CD44 on MES_83 GSCs in the mouse brain with (shAXL#2) or without (shNT) knockdown of AXL. Tumors were removed 24 days after injection (see also Figure S5). Scale bar represents 50 μm.

Figure 7

AXL Is Highly Expressed in Clinical High-Grade Glioma Tumors (A) Survival analysis based on the impact of the multi-gene prognostic index. Affymetrix data generated by TCGA Research Network that are available at the TCGA data portal were used. Visualization of the data was performed using the GBM-BioDP (Celiku et al., 2014). (B) AXL is more highly expressed in astrocytoma and GBM than in non-tumor tissue (Sun et al., 2006) (n = 23 non-tumor; n = 7 grade II; n = 19 grade III; n = 81 GBM). ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗∗p < 0.0001 (Mann-Whitney test). (C) Representative immunohistochemistry (ICH) staining of phospho-AXL in WHO grade IV (GBM), grade III and non-tumor samples. Scale bar represents 50 μm. (D–H) Patient survival data based on AXL expression. (D) GBM dataset. The cutoff was based on the median, n = 156, p = 0.060 with log-rank test. (E) Ovarian cancer dataset through Oncomine. The cutoff was based on the median, n = 288, p < 0.0001 with log-rank test. (F) Poorly and moderately differentiated colon cancer samples. The cutoff was based on the median, n = 164, p < 0.05 with log-rank test. (G) Top and bottom 25% samples from the pancreatic ductal carcinoma dataset through Oncomine, n = 13, p < 0.01 with log-rank test. (H) Top and bottom 25% samples from the Burkitt’s lymphoma dataset through Oncomine, n = 60, p < 0.05 with log-rank test.