CXCR4 EXPRESSION IS ASSOCIATED WITH PRONEURAL-TO-MESENCHYMAL TRANSITION IN GLIOBLASTOMA

Abstract

Glioblastoma (GBM) is the most common primary intracranial malignant tumor and consists of three molecular subtypes: proneural (PN), mesenchymal (MES), and classical (CL). Transition between PN to MES subtypes (PMT) is the glioma analogue of the epithelial-mesenchymal transition (EMT) in carcinomas and is associated with resistance to therapy. CXCR4 signaling increases expression of MES genes in glioma cell lines and promotes EMT in other cancers. RNA sequencing (RNAseq) data of PN GBMs in The Cancer Genome Atlas (TCGA) and secondary high-grade gliomas (HGGs) from an internal cohort were examined for correlation between CXCR4 expression and survival as well as expression of MES markers. Publicly available single cell RNA sequencing (scRNAseq) data was analyzed for cell type specific CXCR4 expression. These results were validated in a genetic mouse model of PN GBM. Higher CXCR4 expression was associated with significantly reduced survival and increased expression of MES markers in TCGA and internal cohorts. CXCR4 was expressed in immune and tumor cells based on scRNAseq analysis. Higher CXCR4 expression within tumor cells on scRNAseq was associated with increased MES phenotype, suggesting a cell autonomous effect. In a genetically engineered mouse model, tumors induced with CXCR4 exhibited a mesenchymal phenotype and shortened survival. These results suggest that CXCR4 signaling promotes PMT and shortens survival in GBM and highlights its inhibition as a potential therapeutic strategy.

INTRODUCTION

Glioblastoma (GBM) is the most common primary intracranial malignant tumor in adults^1,2. The prognosis of patients with GBM remains poor, with a median overall survival (OS) of 14.6 months despite aggressive surgical resection followed by chemotherapy and radiation^3,4. GBM is subdivided into three transcriptional subtypes—proneural (PN), mesenchymal (MES), and classical (CL) – each of which has characteristic molecular alterations, sensitivity to therapy, and prognosis⁵. For instance, PN GBM is associated with PDGFRA amplification and OLIG2 expression; whereas MES GBM is associated with NF1 loss and CD44 expression⁵. Furthermore, multiple subtypes may exist within the same tumor, reflecting the cellular and regional heterogeneity of GBM⁶. Although tumor recurrences are typically of the same class, chemoradiation may induce PN-to-MES transition (PMT), which contributes to therapeutic resistance^7,8. Known regulators of PMT in glioma include the transcription factors the signal transducer and activator of transcription 3 (STAT3) and CCAAT enhancer binding protein beta (CEBPβ)⁹. The transcription factor tafazzin (TAZ) additionally regulates PMT by a separate mechanism¹⁰.

CXCR4 is a G-protein coupled chemokine receptor whose primary ligand is CXCL12 (also known as SDF-1)¹¹. CXCL12-CXCR4 signaling is broadly implicated in tumor survival, proliferation, and migration¹². Inhibition of CXCR4 is associated with decreased expression of PMT genes in glioma cell lines¹³. Additionally, increased CXCR4 expression is associated with a more invasive phenotype in orthotopic GBM models¹⁴. In other systemic cancers, CXCR4 signaling is implicated in the epithelial-mesenchymal transition (EMT), which is considered a carcinoma analogue of PMT^11,12. Therefore, the CXCR4 signaling pathway may provide opportunities for therapeutic intervention by specifically impacting PMT.

Here, we investigated the role of CXCR4 in GBM outcomes with an emphasis on its effects on PMT. While CXCR4 has been previously linked to the MES signature in GBM, its specific effect on PN GBM outcomes have not been evaluated. Additionally, the relation of CXCR4 to known PMT master regulators has not been previously described. While CXCR4 expression is linked to MES signature, we sought to assess whether CXCR4 expression alone was sufficient to induce PMT. We found that increased CXCR4 expression was associated with reduced OS in patients with PN GBM, and CXCR4 expression directly correlated with an overall MES signature in secondary HGGs. CXCR4 expression is correlated with increased expression of the regulons of PMT master regulators STAT3/CEBPβ and TAZ. This suggests CXCR4 drives PMT through an interaction with STAT3/CEBPβ and TAZ. Publicly available single cell RNA sequencing (scRNAseq) data was leveraged to analyze effects of cell type specific CXCR4 expression. Finally, we demonstrate that CXCR4 is sufficient for promoting PMT in a mouse model of PN GBM.

METHODS

RNA expression analysis

The open access portal of the cBIO Cancer Genomics Portal (http://www.cbioportal.org) was used to obtain CXCR4 expression for a publicly available cohort of 56 patients with PN GBM classified by RNAseq¹⁵. Elevated mRNA expression levels as determined by Agilent microarray relative to normalized mean expression were obtained using the Onco Query language syntax. Correlations between CXCR4 expression and the expression of other genes of interest (e.g., CD44, OLIG2, RRBP1, SERPINH1, ICAM1, CD151, OSMR, ACTN1, COL4A2, ITGA7, CHI3L1, TIMP1, SOCS3, PTRF, SHC1, MVP, SERPINE1, SERPINA1, ACTG2, ACTN1, ADAMTS1, ANGPTL4, COL1A1, CXCL2, IL4R, IQGAP1, LUM, MYH9, TAGLN, TIMP1, TUBB, TRADD, CASP1, RELB, NKX-2, SOX-2, ERBB-3) were determined using the built-in comparison tools.

RNA sequencing

Resected secondary HGG were banked with IRB approval (PA140709). Total RNA was isolated from formalin-fixed, paraffin-embedded (FFPE) secondary GBM tissue using a MasterPure kit (Epicentre). Paired-end RNA sequencing (RNA-seq) was performed. Raw reads from RNA-seq data were processed with an in-house pipeline using STAR for read alignment, FastQC and RSeQC for read and alignment quality assessment, and FeatureCount for expression count. Samples with degraded RNA or unique mapping rates <30% were excluded. Reads were aligned to the GRCh38 human reference genome and mapped to the human transcriptome according to University of California Santa Cruz gene annotations. We then normalized the RNA-seq read counts for genes, applied a variance stabilizing transformation, and identified differentially expressed genes using the DESeq2 package. Differential gene expression was determined by a false discovery rate cutoff of 0.05. Single sample gene set enrichment analysis (ssGSEA) were performed using GSVA R package. We used the mesenchymal and proneural gene sets reported in a previous study⁸. The sequencing coverage and quality statistics for each sample are summarized in Supplementary Table 1.

Murine Models

Replication-competent avian sarcoma-leukosis virus long terminal repeat with a splice (RCAS)-platelet derived growth factor B (PDGFB) mice were generated as previously described¹⁶. Expression of PDGFB in Nestin-positive glioneuronal precursors induces endogenous gliomas with PN characteristics in immunocompetent Ntv-a mice¹⁶. This ectopic expression of PDGFB from Nestin-positive glioneuronal precursors results in gliomas that are predominantly histologically consistent with low-grade gliomas. RCAS-CXCR4 was created by cloning human CXCR4 cDNA into a Gateway-compatible RCAS vector using LR recombination (Invitrogen) and verified by sequencing. Immortalized DF-1 chick fibroblast cells were grown in DMEM containing 10% fetal bovine serum at 37°C. Live virus was produced by transfecting RCAS-PDGFB and/or RCAS-CXCR4 into DF-1 cells using FuGENE-6 (Roche).Transgenic mice expressing the avian tumor virus receptor A under the Nestin promoter (Ntv-a) were generated as previously described¹⁶. DF-1 cells transfected with RCAS-PDGFB (1 × 10⁴ cells in 1–2 μL PBS) or RCAS-PDGFB-CXCR4 (1 × 10⁵ cells) were injected bilaterally into the frontal lobes of Ntv-a mice within 24-72 h of birth when Nestin-positive cells are most proliferative. PDGFB (n = 31) and PDGFB-CXCR4 (n = 21) groups were generated. Mice were humanely euthanized by carbon dioxide asphyxiation 90 days after injection or sooner if symptoms related to tumor burden were present. Brains were removed, fixed in formalin, embedded in paraffin, and sectioned for immunohistochemical analysis. Tumors were designated as high-grade if they demonstrated microvascular proliferation or necrosis as described previously¹⁶. Animal experiments were approved by the Institutional Animal Care and Use Committee (protocol #An-8379).

Immunohistochemistry

Sections (4 μm) were cut from FFPE brains. Antigen retrieval was performed with heated citrate buffer. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide. Slides were incubated overnight at 4°C in 10% fetal bovine serum in PBS containing CD44 (1:500, ab41478, Abcam), CD31 (1:50 AF3628, RD systems), or phosphorylated histone H3 (pHH3) (1:100 06-570, Millipore) primary antibodies. On the following day, slides were washed in PBS and incubated with appropriate biotinylated secondary antibodies (1:200, Vector) for 1 h at room temperature. After washing in PBS, slides were incubated with Vectastain ABC reagent (Vector) for 1 h at room temperature and developed with diaminobenzidine as the chromogenic substrate and hematoxylin as the counterstain. Images were obtained using a Zeiss upright light microscope. The total number of CD44-positive cells in four non-overlapping high-power fields from randomly selected tumors was divided by the total number of cells and expressed as a percentage. The total number of pHH3-positive cells in five non-overlapping high-power fields from randomly selected tumors was divided by the total number of cells and expressed as a percentage. The total number of CD31-positive distinct blood vessels in five non-overlapping high-power fields from randomly selected tumors was counted and expressed as a number of distinct vessels per field. Higher magnification images demonstrating positive cells and vessels respectively are depicted in Supplemental Figure 1.

Immunofluorescence

Sections (4 μm) were cut from FFPE brains. Antigen retrieval was performed with heated citrate buffer. Slides were incubated overnight at 4°C in Opal antibody diluent (ARD1001EA, Akoya) containing ionized calcium binding adaptor molecule 1 (Iba1) primary antibody (1:500 178846, abcam) glial fibrillary acidic protein (GFAP) (1:500 7260, abcam). On the following day, slides were washed in PBS and incubated with appropriate fluorescent secondary antibodies (1:500 A1105 and A1101, ThermoFisher). The slides were mounted with Prolong gold with DAPI (P36931, Invitrogen). Images were obtained using a LSM900 confocal fluorescent microscope (Zeiss). The total number of Iba1-positive cells in five non-overlapping high-power fields from randomly selected tumors was divided by the total number of cells and expressed as a percentage.

Single Cell Sequencing

Single cell sequencing data was utilized from publicly available datasets. We used Seurat pipeline for standard procedures for single cell sequencing analysis ¹⁷. We first removed the cells with low quality and performed normalization and variance stabilization. Clustering is performed with the graph-based approach and data is visualized and explored using dimensionality reduction with Uniform Manifold Approximation and Projection (UMAP). Seurat R package, FindAllMarkers function is used to identify all differentially expressed genes between clusters. Clusters are annotated with cell types using the cluster markers identified using Seurat pipeline. We have visualized the CXCR4 gene expression using violin plots.

Statistics

Survival statistics were reported using Kaplan–Meier curves, which were compared using the log-rank test. TCGA gene expression association statistics were calculated using the built-in Spearman’s rank correlation. Gene expression levels between patient groups were compared using unpaired t-tests. Percent of CD44 positive cells between mouse samples was compared using 4 high power fields at 20X magnification dividing the number positive cells by total cells per field; statistical significance was determined with unpaired student’s t-test. Percent of pHH3 and Iba1 positive cells between mouse samples was compared using 5 fields at 20X magnification dividing the number positive cells by total cells per field; statistical significance was determined with one tailed unpaired student’s t-test. The total number of CD31-positive distinct blood vessels in five non-overlapping fields at 20X magnification was counted and expressed as a number of distinct vessels per field; statistical significance was determined with one tailed unpaired t-test. Patient characteristics were compared using the chi-squared test, student’s t-test, or Fisher exact test as appropriate. Chi-squared test was used to compare mouse tumor grade. Statistical analyses were performed using GraphPad Prism software (GraphPad-Prism Software Inc., San Diego, CA).

RESULTS

CXCR4 expression is associated with survival and expression of MES markers in PN GBM

Publicly available TCGA data on RNA expression levels in a set of molecularly subtyped PN GBM (N=56) was used to assess association of CXCR4 expression with survival and MES phenotype^15,18,19. CXCR4 expression was inversely correlated with survival in patients with PN GBM from the TCGA dataset (Spearman r = −0.39, p = 2.6e-3, Figure 1A). Patients with CXCR4 expression above the normalized mean (N=14) had a median OS of 6.84 months (95% CI 5.85-23.18 months), whereas the patients with CXCR4 expression below the normalized mean (N=42) had a significantly longer median OS of 14.7 months (95% CI, 10.62-35.11; log-rank test p = 0.026; Fig. 1B). Higher CXCR4 expression was correlated with higher expression of MES marker CD44 (Spearman r = 0.60, p = 8.9e-7) and lower expression of PN marker Olig2 (Spearman r = −0.39, p = 2.814e-3) (Fig. 1C, D), suggesting that CXCR4 expression is related to PMT. Consistent with this, CXCR4 expression was positively correlated with other previously identified markers of the MES phenotype and negatively correlated with markers of the PN phenotype (Fig. S2).

Figure 1.

(A) Correlation between CXCR4 mRNA expression and OS in 56 PN GBM patients registered in the TCGA database (Spearman r = −0.39, p = 2.6e-3). (B) Kaplan-Meier curves showing OS for PN GBM patients based on high (n = 14) or low (n = 42) CXCR4 expression (*p = 0.025, log-rank test). Correlation of CXCR4 expression with CD44 expression (C; Spearman r = 0.60, p = 8.9e-7) and OLIG2 expression (D; Spearman r = −0.39, p = 2.8e-3). (E) Correlation of CXCR4 expression with C/EBP, STAT3, and TAZ target gene expression.

CXCR4 expression correlates with STAT3, C/EBP, and TAZ targets in PN GBM patients

Previous studies suggest that PMT has multiple driver mechanisms including STAT3 and C/EBP and, separately, TAZ. Therefore, we assessed the co-expression of CXCR4 and target genes of these PMT regulators in patients with PN GBM using the same TCGA dataset. We found that CXCR4 overexpression was associated with higher expression of 10 out of 11 C/EBP targets, 5 out of 6 STAT3 targets, and 9 out of 13 TAZ targets (Fig. 1E). The magnitude of this correlation was greater for C/EBP and STAT3 targets relative to TAZ targets (Fig. S3). There was a negative correlation between CXCR4 and TUBB, a target of TAZ (Fig. S3). These findings indicate that CXCR4 expression positively correlates with the expression of many targets of C/EBP, STAT3, and TAZ and that CXCR4 overexpression is a common mechanism for multiple regulators of PMT.

CXCR4 expression in secondary HGG patients

To validate the findings from the TCGA cohort, we assessed the role of CXCR4 expression in an internal cohort of 64 secondary HGG patients (i.e., patients who had progressed from previously diagnosed WHO II to WHO III or IV tumors; Table 1). These tumors, sequenced after malignant transformation, are surrogates for PN GBM as most secondary GBMs are PN⁸. These patients had a median age of 35 years (range 18-65 years) (Table 1). Seven (11%) patients had gliomas that were isocitrate dehydrogenase (IDH) wild-type based on RNA-seq signatures (Table 1). Patients with the highest CXCR4 expression had lower OS relative to those patients with the lowest CXCR4 expression from the time of initial diagnosis (Fig. 2A). The patients with a CXCR4 expression z-score of >1 (i.e., highest expression, N=9) had a median OS of 30.9 months (95% CI, 27.3-NA), whereas the patients with a CXCR4 expression z-score of <−1 (i.e., lowest expression, N=9) had a median OS of 182.5 months (95% CI, 111.9-NA; p = 0.002, log-rank test; Fig. 2A). This was not significant when assessed based on the date of progression (p = 0.36; Fig. S4). The CXCR4 high patients had a higher proportion of IDH WT (N=4, 44%) versus the CXCR low group (IDH WT N=0, 0%) but this difference did not reach statistical significance (Chi-square, p=0.08). There was no significant difference in survival when excluding all IDH WT patients. The CXCR4 high group additionally were older at diagnosis (p=0.04, student’s t-test) and more likely to progress to grade IV glioma as opposed to grade III on recurrence (p=0.02, chi-square) (Table 1).

Table 1.

Clinical characteristics of in-house secondary HGG cohort

	Total	CXCR4 high (n = 9)	CXCR4 low (n = 9)
IDH Status ^
Mutant	57	5	9
WT	7	4	0
			p = .08
Age Average (years) ~	37.63	44.6 (29-59)	35.33 (22-50)
			P=.04
Death Status ^
Deceased	39	7	6
Alive	25	2	3
			P=1
Initial Grade ^
II	53	7	9
III	11	2	0
			P=.47
Initial Path #
A	30	6	3
OA	7	0	1
AA	9	2	0
O	17	1	5
AO	1	0	0
			P=.06
Progression Grade ^
III	38	2	7
IV	23	7	2
			P = .02
Progression Path #
AA	17	1	2
AO	15	1	4
AOA	6	0	1
GBM	23	7	2
No progression data	3	0	0
			P=.08

^*N=62

^**N=63

^{^}Chi-square

^#Fischer exact

^~Student T-test

Figure 2.

(A) Kaplan-Meier curves showing OS for secondary HGG patients based on high (n = 10) or low (n = 10) CXCR4 mRNA expression (*p = 5.0e-3, log-rank test). (B) Heatmap representing unsupervised clustering of all samples on the basis of MES versus PN enrichment. Correlation of CXCR4 expression with degrees of MES phenotype (R=0.85, p<f2.2e-16) (C) and PN phenotype (R=0.54, p=3.3e-6) (D). (E) Correlation of CXCR4 expression with C/EBP, STAT3, and TAZ target gene expression.

We assessed the correlation of CXCR4 level to transcriptional subtype within this cohort by single sample gene set enrichment analysis (ssGSEA). Using an unsupervised clustering approach, high CXCR4 expressors segregated from low expressors on the basis of enrichment of MES versus PN signatures respectively (Fig. 2B). Similar to the TCGA cohort, higher CXCR4 expression was associated with higher CD44 expression (p = 0.0002 ;Fig. S4) and lower OLIG2 expression (p = 0.01;Fig. S4) in HGG patients. Moreover, CXCR4 expression linearly correlated with enrichment of the MES signature (R=0.85, p<2.2e-16) and linearly anticorrelated with the PN signature (R=0.54, p=3.3e-6) suggesting a gene dosage effect (Fig. 2C, D).

We also assessed correlations between CXCR4 expression and the expression of targets of C/EBP, STAT3, and TAZ in our secondary HGG cohort. CXCR4 overexpression was associated with higher expression of 11 out of 11 C/EBP targets, 6 out of 6 STAT3 targets, and 12 out of 13 TAZ targets (Fig. 2E, Fig. S5). Notably, the lack of positive correlation between CXCR4 and TUBB remained consistent in both the HGG and TCGA cohorts (Fig. S5). These results indicate that increased CXCR4 expression portends a worse prognosis in secondary HGG, is associated with increased MES signature, and correlates with the expression of targets of known PMT regulators.

CXCR4 promotes MES features in a murine model of PN GBM

Expression of PDGFB in Nestin-positive glioneuronal precursors triggers gliomas that are predominantly histologically consistent with low-grade gliomas. To assess the role of CXCR4 in PMT, we leveraged this murine model by injecting mice with RCAS-PDGFB or RCAS-PDGFB+RCAS-CXCR4 constructs to induce gliomas. Mice co-expressing RCAS-PDGFB+RCAS-CXCR4 (n = 17) showed a lower median OS (60 days) relative to mice expressing RCAS-PDGFB (n = 31, 90 days; log-rank test, p = 0.04; Fig. 3A). Moreover, mice co-expressing RCAS-PDGFB+RCAS-CXCR4 demonstrated a higher incidence of HGG (HGG n = 9, LGG n = 8) compared with mice expressing RCAS-PDGFB (HGG n = 9, LGG n = 22; chi-square, p < 0.0001; Fig. 3B). Representative images of LGG and HGG are shown (Fig. 3B, S6).

Figure 3.

(A) Kaplan-Meier curves showing OS for mice injected with RCAS-PDGFB (n = 31) or RCAS-PDGFB+RCAS-CXCR4 (n = 17; * p = 0.041, log-rank test). (B) A higher proportion of HGGs was observed in PDGFB-CXCR4 mice (9/17) than in PDGFB mice on the basis of hematoxylin and eosin staining (9/31; *** p < 0.0001, chi-square test). In addition, histological sections of HGGs from PDGFB and PDGFB-CXCR4 mice were immunostained with antibodies against the MES marker CD44 (C, D), mitotic marker pHH3 (F, G), vascular endothelial marker CD31 (I, J), and combination of microglia/macrophage marker Iba1 and astrocyte marker GFAP (L, M). Tumors from PDGFB-CXCR4 mice showed a higher proportion of CD44-positive cells (** p = 0.0011, unpaired t-test; E), elevated rate of mitoses (* p = 0.033, unpaired t-test; H), increased microvessel density (** p = 0.0018, unpaired t-test; K), and greater accumulation of Iba1-positive microglia macrophages (* p = 0.040, unpaired t-test; N) than those from PDGFB mice. Scale bar, 50 μm.

In order to determine the impact of CXCR4 signaling on the tumor microenvironment, histologic sections of HGGs from RCAS-PDGFB and RCAS-PDGFB+RCAS-CXCR4 groups were stained with antibodies against CD44, pHH3 (mitotic marker), CD31 (endothelial marker), Iba1 (microglia/macrophage marker), and GFAP (astrocyte marker). Compared to tumors from the RCAS-PDGFB group (n = 5), tumors from the RCAS-PDGFB+RCAS-CXCR4 group (n = 5) exhibited increased proportion of CD44-positive cells (3.1% versus 9.7%, unpaired t-test, p = 0.0011; Fig. 3C, D, E). Mitotic activity was significantly elevated in the RCAS-PDGFB+RCAS-CXCR4 HGGs (2.04%, n = 5) versus RCAS-PDGFB tumors (0.85%, n = 4; unpaired t-test, p = 0.033; Fig. 3F, G, H). Mean density of CD31-positive microvessels in RCAS-PDGFB+RCAS-CXCR4 HGGs (n = 5) was 141 /mm², which was significantly higher than that in RCAS-PDGFB HGGs (n = 5; 75 /mm²; unpaired t-test, p = 0.0018; Fig. 3I, J, K). Finally, RCAS-PDGFB+RCAS-CXCR4 HGGs showed more pronounced accumulation of Iba1-positive microglia/macrophages (n = 4, mean density 31%) compared to RCAS-PDGFB HGGs (n = 4, mean density 19%; unpaired t-test, p = 0.040; Fig. 3 L, M, N). Taken together, these results indicate that expression of CXCR4 confers a MES phenotype of glioma and that CXCR4 expression correlates with a more aggressive phenotype of glioma with poorer OS. This further suggests that isolated CXCR4 expression is sufficient to induce PMT in a mouse model of PN glioma.

scRNAseq confirms CXCR4 expression in tumor cells

We concluded that increasing CXCR4 expression in PN GBM is associated with MES phenotype and reduced survival based on the bulk RNAseq experiments presented above. Furthermore, co-expression of RCAS-CXCR4 with RCAS-PDGFB was sufficient to induce PMT. We further analyzed publicly available scRNAseq data to clarify cell type specific effects of CXCR4 expression²¹. scRNAseq data segregates into various cell types with characteristic CXCR4 expression (Fig. 4A). CXCR4 expression was significantly highly expressed in cell clusters corresponding to microglia/monocytes, and T cells (Fig. 4B; microglia/monocytes adjusted p-value<0.05, log2 fold change=4.68, T-cell adjusted p-value=2.57e-100, log2 foldchange=4.52). However, CXCR4 expression was also notable in tumor cells (Fig. 4A), albeit at lower levels. This is possibly reflective of high CXCR4 expression levels in immune cells as CXCL12-CXCR4 signaling is known to be a key regulator of immune cell chemotaxis²². CXCR4 expression was particularly high in certain subsets of tumor cells, the significance of which remains to be explored (Fig. 4C; cluster 10 adjusted p-value= 4.17E-09, cluster 13, adjusted p-value 1.72E-33, cluster 19 adjusted p-value 1.01E-39). Consistent with our earlier findings, level of CXCR4 expression in tumor cells correlates positively with mesenchymal enrichment score (Fig. 4D; Pearson R=0.17, p<2.2e-16) and negatively with proneural enrichment score (Fig. 4E; Pearson R=−0.13, p<2.2e-16). Together, this data suggests a cell-autonomous effect, and supports the mouse studies where CXCR4 is expressed specifically within tumor cells. Interestingly, increased CXCR4 expression in microglia and monocyte cells was associated with increased MES phenotype within immune cells. This suggests that PMT may be promoted by the function of CXCR4 in the tumor microenvironment as well the direct action of CXCR4 within tumor cells.

Figure 4.

CXCR4 expression in different cell types are depicted in UMAP coordinates via (A) feature plot and (B) violin plot. Immune cells have significant expression of CXCR4 (microglia/monocytes adjusted p-value=0, log2 fold change=4.68, T-cell adjusted p-value=2.57e-100, log2 foldchange=4.52). (C) CXCR4 is significantly expressed only in some subclusters of tumor cells (cluster 10 adjusted p-value= 4.17E-09, cluster 13, adjusted p-value 1.72E-33, cluster 19 adjusted p-value 1.01E-39). CXCR4 expression within tumor cells correlates with enrichment of MES phenotype (D) (Spearman R=0.17, p<2.2e-16) and is inversely correlated with PN phenotype (E) (Spearman R=−0.13, p<2.2e-16).

DISCUSSION

GBM is molecularly classified into PN, MES, and CL subtypes⁵. MES GBM has worse outcomes than other subtypes, with PMT described as an analog of EMT²³. Here, we show that higher CXCR4 expression is associated with poorer OS in a publicly available PN GBM dataset as well as our own cohort of secondary HGG patients. Additionally, CXCR4 overexpression was linearly correlated with higher expression of genes associated with the MES phenotype, suggesting a gene dose phenomenon. CXCR4 expression was correlated with the activity of two sets of master regulators of PMT in GBM: STAT3/CEBP and TAZ. Taken together, these findings suggest that CXCR4 expression drives PMT in a dose dependent fashion through interactions with STAT3/CEBP and TAZ master regulators and is a marker of poor outcomes in PN GBM. Furthermore, we demonstrate that CXCR4 is sufficient to promote PMT by adding CXCR4 to an RCAS-PDGFB murine model of PN GBM. Additionally, we demonstrate increased MES phenotype and reduced PN phenotype when CXCR4 is increased on a cell type specific basis within tumor cells using publicly available scRNAseq data.

PMT is described as the glioma analogue of EMT in carcinomas^12,24. EMT is the loss of differentiated, polarized, and contact-inhibited properties of epithelial cells and the acquisition of mesenchymal invasive and stem-like properties²⁴. PN gliomas frequently present as lower-grade tumors, but PMT does not underlie malignant transformation⁸. Nonetheless, PMT is reported to occur because of radiation⁷, and recurrent tumors manifest a more MES phenotype^25,26. EMT and PMT may share molecular mechanisms^11,23 and chemokines have been shown to facilitate EMT in tumor cells²⁷. Specifically, increased expression of CXCR4 has been linked to EMT in many carcinomas¹¹. Increased CXCR4 expression is a marker of drug resistance in NSCLC cell lines and demonstrates similarity to cells undergoing EMT²⁸. Moreover, EMT linked to CXCR4 overexpression in NSCLC cell lines was related to signaling by MIF-1 as opposed to the canonical ligand CXCL12 in a partially IL-6 mediated manner²⁹. Interestingly, it is not yet established the activity of which CXCR4 ligand effects PMT in GBM. While there are many similarities between EMT and PMT, the specific CXCR4 signaling paradigm in PMT remains to be established.

Numerous mechanisms promote advancement along a PN-MES axis²³, and CXCR4 has been previously studied in the context of PMT. CXCR4 silencing in U87 GBM cell lines leads to decreased mesenchymal phenotype by loss of expression of N-cadherin, vimentin, b-catenin, TGF-b1, p-Smad2, and p-Akt, and the reduced activity of transcription factors NF-kB, AP-1, Snail, and twist¹³. Guanine Nucleotide binding-protein Gamma subunit 4 (GNG4) is hypermethylated in aggressive GBM, and exogenous over expression of GNG4 in tumor cell lines inhibited SDF1α/CXCR4-dependent chemokine signaling and reduced GBM cell migration³⁰. Prior query of the TCGA database has similarly shown CXCR4 expression to be a marker of the MES phenotype, with possible interaction with the MAPK pathway³¹. We show specifically that increased expression of CXCR4 in human GBM is associated with decreased overall survival in PN GBM possibly through induction of PMT.

Our murine model demonstrated ectopic expression of CXCR4 from tumor cells was sufficient to induce PMT thereby reducing survival. Whereas we selectively express CXCR4 in our murine model, scRNAseq data demonstrate highest CXCR4 expression in immune cells. One explanation is that this difference reflects a physiologically higher baseline CXCR4 expression in immune cells as CXCL12-CXCR4 signaling is a key mechanism of immune cell chemotaxis²². Nonetheless, increased CXCR4 expression in human tumor cells is associated with increased MES phenotype on scRNAseq, suggesting that increased CXCR4 expression within tumor cells drives PMT. Thus, taken together, the scRNAseq and murine experiments demonstrate CXCR4 overexpression in tumor cells is sufficient to induce PMT in PN GBM. However, CXCR4 may exert some effect on PMT via alterations of the tumor microenvironment. Increased CXCR4 expression in microglia/monocytes was associated with increased MES phenotype (Fig. S7) and increased CXCR4 expression in oligodendrocytes was associated with decreased PN phenotype (Fig. S7) within those same cell types. Additionally, tumors from mice co-expressing RCAS-PDGFB+RCAS-CXCR4 had a higher percentage of Iba1 positive macrophages/microglia than those of RCAS-PDGFB mice (Fig. 3N). It remains to be elucidated if the effect of CXCR4 in human GBM is driven by its expression in tumor cells or cells in the tumor microenvironment. Ultimately, the differential effects of CXCR4 expression in GBM on the basis of the cell type in which it is expressed is a future direction for our work.

Computational approaches identify two sets of non-interacting transcriptional master regulators of PMT: C/EBP, STAT3, and TAZ. Carro et al. identified a set of six transcription factors (STAT3, bHLH–B2, C/EBP, FOSL2, ZNF238, and RUNX1) responsible for PMT⁹. Of these, C/EBP and STAT3 were considered master regulators of the regulatory network. The authors looked at the combined effect of C/EBPβ and C/EBPδ as C/EBP due to their in vivo heteromeric binding. C/EBP is a basic leucine zipper containing transcription factor that plays critical roles in immune and inflammatory responses³². CXCR4 may be linked to C/EBPβ as liver-enriched inhibitory protein (LIP), an isoform of C/EBPβ, is a direct transcription factor for CXCR4 in breast cancer³³. STAT3 is a transcription factor that is activated by receptor-associated kinases typically in response to cytokine or inflammatory signaling³⁴. STAT3 is a key regulator of PMT in response to radiation, as blockade of its upstream regulator JAK2 prevents the development of MES features after radiation³⁵. Interestingly, there are also links between CXCR4 and JAK-STAT signaling. For example, SDF-1-activated CXCR4 directly activates JAK2, likely by transphosphorylation, in T cells³⁶. Additionally, CXCR4 upregulation is linked to the maintenance of stemness in a STAT3-mediated manner in cultured non-small cell lung cancer cells³⁷. CXCL12 mediated activation of CXCR4 in bladder cancer models drives induction of phosphorylated Stat3; this activation mediates invasiveness and is abrogated by CXCR4 knockdown³⁸.

Separately, Bhat et al. identified TAZ, a transcriptional co-activator that is a downstream target of the Hippo tumor suppressor, as a distinct master transcriptional regulator of PMT in glioma¹⁰. Specifically, using an RCAS/Ntv-a model, the addition of TAZ to PDGFB resulted in tumors with increased expression of MES features. Recently, it was shown that EGFR-driven TAZ activation lead to increased CXCR4 expression in patient-derived GBM oncospheres³⁹. In the present study, we found that tumors with higher CXCR4 expression had higher expression of both the CEBP/STAT3 and TAZ regulons. These findings indicate that increased CXCR4 expression is a common mechanism of distinct upstream regulators of PMT and suggests that CXCR4 signaling could be a therapeutic target that neutralizes redundant signaling mechanisms in PMT. Ultimately, the mechanism of interaction between CXCR4 and these PMT master regulators remains to be defined.

Many of the mechanisms of PMT are similar to those of the generation and maintenance of glioma stem cells (GSCs)²³. Cancer stem cells (CSCs) are self-regenerating pluripotent cells capable of generating tumors upon transplantation¹¹. CXCR4 has been studied as marker of CSCs in human GBM, with CXCR4-positive cells fulfilling the criteria for consideration as GSCs⁴⁰. These findings suggest that CXCR4 expression may be a proxy for the extent of GSCs within a tumor, which is important as GSCs are thought to survive conventional chemotherapy and radiation and give rise to recurrent GBM⁴¹. Additionally, as CXCR4 signaling is linked to both GSC maintenance and PMT⁴², CXCR4 activity may be a mechanism by which post-treatment glioma recurrence manifests a more MES phenotype through the maintenance of treatment-resistant GSC populations.

There is one commercially available CXCR4 antagonist, plerixafor, and it is currently used as an adjunct to mobilize hematopoietic stem cells prior to stem cell transplantation. Early studies have assessed the utility of CXCR4 antagonism in GBM. A phase I clinical trial demonstrated the safe use of plerixafor in conjunction with adjuvant temozolomide⁴³. In a follow-up phase I/II trial by the same group, plerixafor was continuously infused for 4 weeks beginning at day 35 of concurrent chemoradiation for primary GBM, resulting in a median OS of 21.3 months and PFS of 14.5 months⁴⁴. Another group demonstrated the safety of plerixafor in combination with bevacizumab in a phase I trial⁴⁵. Bevacizumab is a recombinant monoclonal antibody against vascular endothelial growth factor (VEGF) that is used to reduce VEGF function in endothelial cells and reduce tumor angiogenesis⁴⁶. The authors of this study hypothesize that CXCR4 inhibition with concurrent bevacizumab administration may target redundant mechanisms used by the tumor generate new blood vessels⁴⁵. Our data support their hypothesis that CXCR4 activity leads to angiogenesis as our murine model demonstrated increased vessel density by CD31 staining in tumors from RCAS-PDGFB+RCAS-CXCR4 mice. Beyond this, there have been case reports of CXCR4 antagonists used in combination with other novel therapeutics leading to prolonged remisson⁴⁷. Additionally, animal studies suggest the benefit of combined anti-PD-1 and anti-CXCR4 therapy by reduced infiltration of immunosuppressed leukocytes ⁴⁸. The continuing growth of this evidence base warrants larger clinical studies that assess the effect of CXCR4 blockade in patients with GBM.

This study has several limitations. Within our cohort of secondary HGG, the decreased survival of the CXCR4 high group is confounded by a higher proportion of IDH WT patients. These data were collected prior to the WHO 2016 guidelines distinguishing IDH wildtype and mutant disease as separate entities, and further study is needed to assess the role of CXCR4 in IDH mutant disease. As such, the internal cohort encompasses a variety of pathologic diagnoses which are no longer in use. Regardless, we dichotomized low- versus high grade based on histological criteria. Additionally, our results are in line with TCGA survival data as well as prior data from other groups showing reduced survival in patients with increased CXCR4 expression³¹. Additionally, we do not consider the type of treatment received at initial diagnosis of LGG and any potential effect this may have on the transcriptional subtype of secondary HGG. Many patients in this cohort received their initial surgical and oncologic care at other institutions before presenting to our hospital with secondary HGG, rendering this analysis unfeasible at present but a target of future study. Although it would have been beneficial to assess the level of CXCR4 protein in human tumor samples, this was not feasible in the current study. Additionally, a sequencing-based approach in the murine model would have helped confirm the signaling correlations observed in human patients. Although we found that CXCR4 expression correlated with the expression of other genes highlighted by previous studies, we do not demonstrate direct causative linkages or mechanistically define pathways. Thus, additional research is needed to assess the function of CXCR4 in glioma. Nonetheless, we demonstrate that increased CXCR4 expression in PN GBM is associated with reduced survival and an increased MES phenotype.

NOVELTY AND IMPACT.

Higher CXCR4 expression is associated with shorter survival and increased expression of MES markers in PN GBM in both bulk and single cell RNAseq. CXCR4 expression in RCAS-PDGFB murine model of PN GBM is sufficient to induce PMT.

ABBREVIATIONS

CD31

cluster of differentiation 31

CEBPβ

CCAAT enhancer binding protein beta

CL

classical

EMT

epithelial-mesenchymal transition

FFPE

formalin-fixed, paraffin-embedded

GBM

glioblastoma

GFAP

glial fibrillary acidic protein

Iba1

ionized calcium binding adaptor molecule 1

HGG

high-grade glioma

OS

overall survival

PDGFB

platelet derived growth factor B

pHH3

phosphorylated histone H3

PMT

PN to MES subtype transition

PN

proneural

RCAS

replication-competent avian sarcoma-leukosis virus long terminal repeat with a splice

ssGSEA

single sample gene set enrichment analysis

STAT3

signal transducer and activator of transcription 3

scRNAseq

single cell RNA sequencing

TCGA

The Cancer Genome Atlas

Data Availability Statement

The bulk RNA-seq data generated in this study is available in GEO under accession number GSE184941. These RNA data were generated in conjunction with the NIH and deposited in GEO at collection in 2021. These RNA data were used also in other studies. The bulk RNAseq data that support the findings of this study are available in TCGA at [doi: 10.1038/nature07385]. The single cell RNAseq data are derived from previously published reports [doi: 10.1038/nbt.3192]. Other data that support the findings of this study are available from the corresponding author upon request