GAMs induce EC activation
To study the relationships between GAMs and EC activation, we first examined the effect of GAMs on the gene expression profile of ECs. To accomplish this, we designed and developed an in vitro model examining the differential expression of angiogenic genes involved in GBM vascularity (Fig. 1a, left panel). To model GAMs, we first stimulated macrophages (RAW 264.7 cells) with conditioned medium (CM) collected from GBM cell lines, selecting from a panel of GBM cells (U87, U118, U251, or A172) (Fig. 1b, upper panel) and glioma stem cells (GSC28 and GSC267) derived from primary human GBM patients (Fig. 1b, lower panel). Subsequently, CM from stimulated macrophages was used to stimulate ECs for 24 h. ECs gene expression was analyzed using an RT-PCR angiogenesis panel comprised of 30 angiogenesis genes (Fig. 1a, right panel). Treatment of ECs with CM from GAMs resulted in a significant upregulation (> twofold) of four of the thirty angiogenic genes: VCAM1, ICAM1, CXCL5, and CXCL10 compared to control (Fig. 1b). Interestingly, VEGFA, a key angiogenic regulator was not significantly upregulated (> twofold) in Mφ-GBM cell lines when compared to Mφ-NHA, suggesting that GAMs induce EC activation independent of VEGFA (Additional file 1: Figure s1). Immunoblot analysis confirmed the increased expression of VCAM1 and ICAM1 proteins in ECs stimulated by GAM CM, validating the RT-PCR findings (Fig. 1c). Importantly, other tumor-stimulated (786-O, H1299, Hei193 and U2OS) macrophages, failed to induce expression of VCAM1, ICAM1, CXCL5 and CXCL10 as shown in Additional file 1: Figures S2. Furthermore, human cerebral ECs (hCMEC/D3) respond similar to HUVECs when stimulated with GAM-CM (Mφ-U87) (Additional file 1: Figure S3). These results demonstrate that GAMs specifically induce EC activation.
ECs activation associates with patient glioma grade and worse OS
ECs activation is characterized by increased expression of surface adhesion molecules such as VCAM1 and ICAM1 [13] and there is evidence that VCAM-1 is closely associated with tumor neoangiogenesis and progression [20]. To investigate the correlation of VCAM1 expression and glioma grades, we analyzed a cohort of IDH-wildtype(wt) human grade IV (GBM) (n = 14) and lower grade (LGG, grade II and III) glioma samples (n = 10) using dual immunohistochemical (IHC) staining for the endothelial marker CD31, and endothelial activation marker VCAM1. The classical glomerular-tufts seen in GBM, which are histological structures formed as a consequence of ECs proliferation and piling, showed a distinct VCAM1 positive staining restricted to the EC tufting and seen around the lumen of a vessel. Non-glomerular vasculature in the GBM by comparison were predominantly CD31 positive with low VCAM1 expression (Fig. 2a). We found significantly higher expression of VCAM1 in GBM cases (10/14) compared to LGG, where there was minimal to absent VCAM1 staining detected in all 10 cases examined (Fig. 2a, b).
To further validate these findings, we analyzed the TCGA glioma RNAseq database downloaded from the Genomic Data Commons (GDC) data portal (https://gdc.nci.nih.gov/) and analyzed VCAM1 expression in 137 IDH-wt GBM and 85 IDH-wt LGG samples. VCAM1 gene expression was statistically significantly higher in GBM vs. LGG (p = 0.01, Fig. 2c). Higher VCAM1 expression was associated with worse outcome as determined by median survival at 347 days vs. 427 days with lower expression of VCAM1 (log-rank test, p = 0.009, n = 222, Fig. 2d, upper panel). Multivariate analysis was performed in the IDH-wt glioma cohort (n = 222) to define the prognostic factors that contribute to OS. We found VCAM1 expression as a significant prognostic factor independent of tumor grade with HR > 1, indicating that it is significantly associated with shorter survival with 95% CI (p = 0.03, Fig. 2d, lower panel). We expanded our cohort to include all IDH-wt and IDH-mut gliomas in the TCGA RNAseq database (n = 593). Strikingly, VCAM1 expression remained a predictor of OS with median survival at 1152 days in VCAM1 high expression group vs. 2379 days in low expression group, which was based on median VCAM1 expression cutoff (p < 0.0001, Fig. 2e, upper panel). VCAM1 expression was significant in multivariate analysis, independent of tumor grade or IDH status with HR > 1, indicating that it is significantly associated with shorter survival with 95% CI (p = 0.007, Fig. 2e. lower panel). These results collectively support VCAM1 to be an important marker of activated ECs and that ECs activation as measured by high expression of VCAM1 can serve as a predictor of poor outcome in gliomas.
TNFα secreted by GAMs induces ECs activation and associates with OS in human GBM
We next wanted to uncover the specific cytokines secreted by GAMs that induce EC activation. We performed a multi-analytic inflammatory ELISA on conditioned medium from macrophages that were stimulated with the same panel of GBM cell lines (Mφ-GBM) as outlined above compared to macrophages conditioned with normal human astrocyte media (Mφ-NHA), which served as a control. Of the twelve inflammatory cytokines analyzed, only TNFα was significantly upregulated more than two-fold in all media obtained from Mφ-GBM (p < 0.05, Fig. 3a). This was further supported by Western immunoblot analysis showing that Mφ-GBM contains higher levels of TNFα in comparison to Mφ-NHA (Fig. 3b). Additionally, we show that TNFα expression is higher in the media from two primary glioma stem cell lines Mφ-GSC28 and Mφ-GSC267 compared to neural stem cell (NSC) control (Mφ-NSC, Fig. 3c).
Next to validate the effects of TNFα on EC activation, we treated human cerebral endothelial cells (hCMEC/D3) with recombinant human TNFα and found that it induced a statistically significant upregulation of VCAM1, ICAM1, CXCL5, and CXCL10 and no upregulation of VEGFA, similar to the effects seen with Mφ-GBM (Figs. 3d, 1b, Additional file 1: Figure s1). Taking together, our data strongly support that TNFα is the cytokine secreted by GAMs that induces EC activation.
To examine clinical significance of our findings, we investigated if human GAMs express TNFα and whether this associates with EC activation and OS. We used one of the cell surface macrophage markers, CD68 to identify GAMs [39]. CD68 is one of the most useful and descriptive markers for microglial function since both M1 polarized and M2 polarized microglia/macrophages can express CD68 [22]. We have also used F4/80 and MAC3 to confirm GAMs (data not shown). In a cohort of 39 human IDH-wt GBM patients, the serial sections from each patient were co-immunostained with CD68 and TNFα or TNFα andVCAM1. Dual IHC showed that GAMs (CD68+ cells) express TNFα at a range from 0 to 98.44% with average of 47.53% (Fig. 3E, Table 1). Co-staining for TNFα and VCAM1 in human GBM tissues also revealed a regional association of high VCAM1 and TNFα staining (Fig. 3F). Furthermore, we discovered that there is significantly positive correlation between extent of VCAM1 staining (Low, Medium or High) and percentage of TNF-alpha expressing GAMs cells per mm2 tumor area (Spearman, R = 0.8, p = 2.738e-009, Fig. 3g, Table 1).
Using publicly available RNAseq data from TCGA, correlation of expression between CD68 and TNFα was analyzed in 137 IDH-wt GBM. The expression of TNFα was significantly correlated with CD68 expression (Spearman, R = 0.46, p = 1.93E-08, Fig. 3h lower panel) and co-expression of TNFα and CD68 was significantly associated with worse OS (Spearman, p = 0.027, Fig. 3h upper panel). Additionally, this analysis showed that CD68 expression was significantly associated with another characteristic macrophage marker CD163 (Spearman, R = 0.74, p = 2.32E-25) and also with VCAM1 expression (Spearman, R = 0.25, p = 0.003) (Fig. 3h lower panel). Collectively, these results support that TNFα is positively correlated with the macrophage marker CD68 and worse OS in patients with GBMs.
GBM cells secrete interleukin 8 (IL8) and chemokine (C–C motif) ligand 2 (CCL2) to induce TNFα secretion in GAMs
To determine the factors secreted by GBM cells that in turn regulate GAMs to produce TNFα, we conducted a cytokine array analyzing 24 cytokines. Specifically, CM from different human GBM cell lines and NHA controls were collected and hybridized on a human cytokine array (Fig. 4a). We found that IL-8/CXCL8 secretion was increased in both U87 and U251 CM by 5.2- and 9.9-fold respectively compared to NHA CM. The secretions of CCL2/MCP-1 in U87 and U251 CM were 0.4 and 24-fold compared to NHA CM. Serpin E1/PAI-1 was also upregulated however since secreted by both GBM and NHA cell lines to similar levels (0.70–onefold, Fig. 4b) for subsequent studies, we focused on IL-8 and CCL2. To determine the key factor(s) and their optimal dosage responsible for GAM-induced TNFα secretion, we exposed RAW 264.7 macrophage cells to IL-8 or CCL2 alone or in combination at concentrations of 1 nM, 10 nM or 50 nM and compared the levels of TNFα secretion to that induced by Mφ-NHA CM (negative control), Mφ-U87 CM and Mφ-U251 CM (two positive controls). Consistent with our previous findings, we found that both Mφ-U87 CM and Mφ-U251 CM significantly induced TNFα secretion compared to Mφ-NHA CM (p < 0.05, Fig. 4c). Secondly, Mφ-U251 CM induced significantly less secretion of TNFα than Mφ-U87 CM even though U251 CM contains relatively higher amounts of IL-8 and CCL2 than U87 CM (9.9- vs. 5.2-fold and 24- vs. 0.4-fold, respectively, Fig. 4b, c), which suggests a non-linear relationship between either IL-8 or CCL2 and TNFα. Indeed, we establish this to be true since exposure of RAW264.7 cells to IL-8 or CCL2 alone or in combination at concentrations of 1 nM shows higher stimulation of TNFα secretion compared to those at concentrations of 10 or 50 nM (Fig. 4c). Thirdly, we found that while IL-8 or CCL2 alone significantly stimulates GAMs to secrete TNFα compared to NHA CM, only the combination of IL-8 and CCL2 stimulates the secretion of TNFα to the level comparable to U87 CM, which is the highest degree of response (Fig. 4c). Taken together, we found that both IL-8 and CCL2 are required to stimulate TNFα secretion in GAMs.
Inhibition of TNFα prevents EC activation and prolongs survival of mouse glioma model
To examine whether inhibition of GAM-secreted TNFα reduces EC activation, we inhibited TNFα with a neutralizing antibody or the selective inhibitor, Thalidomide, and evaluated the effects on EC activation genes signatures [32]. We showed that inhibition of TNFα with a neutralizing antibody was sufficient to block GAM-induced upregulation of VCAM1, ICAM1, CXCL5, and CXCL10 (Fig. 5a, upper panel), but had no effect on the expression of coagulation factor III (F3) that was not altered by GAM-CM (Fig. 5b). Inhibition of TNFα with Thalidomide also significantly downregulated the expression of VCAM1 and ICAM1, the two main genes characterized for EC activation [13] (Fig. 5a, lower panel).
We next examined the therapeutic potential of targeting TNFα in the established syngeneic glioma mouse model, GL261. GL261 mouse model is an immunocompetent model that allows for the engraftment of tumors in the brain without immediate immune rejection. This model is typically used to investigate the role of the tumour microenvironment [18] and it is the most widely used syngeneic model of glioblastoma [27]. This model recapitulates many of the genetic and phenotypic characteristics of GBM, including mutations in p53 and K-Ras that drive high expression of c-myc, cellular pleomorphism, angiogenesis, and pseudopalisading necrosis [35, 40]. Although IL-8 and its major cognate receptor, CXCR1, are not present in mouse cells, MIP-2 and KC are suggested to be functional homologues [17, 23, 24, 43]. Both MIP-2 and CCL2 are expressed by GL261 tumor cells [5, 9, 42]. Consistent with human GBM cell lines, CM from GL261 stimulated macrophages (Mφ-GL261) which in turn upregulated expression of VCAM1 and ICAM1 but not VEGF compared to macrophages conditioned with normal mouse astrocytes (Mφ-NMA) (Fig. 5c, Additional file 1: Figure s4). Importantly we show that baseline expression of TNFα in GL261-injected hemisphere (T) is elevated relative to non-GL261 injected hemisphere (N) as determined by ELISA and Western blotting (Fig. 5d).
To investigate the impact of the increased levels of the TNF-α expression on EC activation in GL261 syngeneic glioma mouse model, we performed double IHC staining with CD31 and VCAM1 antibodies. We assess the co-expression of the CD31 and VCAM1 on endothelial cells in the tumor and adjacent morphologically normal cortical brain tissue in this mouse model. There is significantly increased co-expression of the CD31 and VCAM1 in tumor areas compared to that in adjacent normal brain tissue or non-injected contralateral normal brain (scores 3.2 vs 1, p = 0.0004, Left panels in Fig. 5e, f). These findings support that the GL261 glioma model recapitulates the increased TNFα expression that induces EC activation, as seen in human GBMs detailed in the results above.
To further investigate the significance of the TNFα inhibition on the GL261 syngeneic glioma mouse model tumorigenesis, we used MP6-XT22, the TNFα monoclonal antibody derived from rat, which is an analog of Infliximab, a chimeric monoclonal antibody directed at TNFα [25, 25]. We confirmed that MP6-XT22 inhibits EC activation induced in vitro by macrophages stimulated with GL261glioma cells, as shown in Fig. 5c. Subsequently, we examined the therapeutic potential of MP-XTT22 in vivo using the mouse GL261 glioma model. First, we show that MP6-XT22 can cross the blood–brain barrier (BBB) as detected by the presence of rat IgG in capillary-depleted brain rich fractions isolated from GL261 tumors in treated mice and was also found at significantly higher levels in tumor versus normal contralateral brain (non- GL261 injected hemisphere) by ELISA (Fig. 5g). Next, we treated GL261 tumor-bearing mice with vehicle control or MP6-XT22 to determine the therapeutic effect of MP6-XT22 on EC activation and OS. MP6-XT22 was shown to decrease co-expression of CD31 and VCAM1 in the tumor areas compared to those in untreated vehicle control (score 3.2 vs 1.8, p = 0.03, right panels in Fig. 5e, f) and increase survival (log-rank test, p = 0.0008, Fig. 5h). The treatment reduced the level of co-expression of VCAM and CD31 in tumor areas to that in normal brain tissues (score 1.8 vs 1, p = 0.06, right panels in Fig. 5e, f) which suggests normalized ECs.
Taken together, these data demonstrate that inhibition of TNFα reduces EC activation and prolongs survival of mouse glioma models and provides support for TNFα serving as a novel therapeutic target in GBM.
Increased macrophage recruitment and local concentration of TNFα drive resistance to AATx
It has been reported that in response to Bevacizumab, an inhibitor targeting VEGF, GBMs recur in a more aggressive manner and are highly invasive and resistant to additional treatment [10, 15]. The other study has also shown that anti-VEGF treatment increases hypoxia which promotes macrophage infiltration [14, 16, 26]. We posit that increased macrophage recruitment will increase the local concentration of TNFα which in turn promote EC activation. To understand the role of TNFα in glioma resistance to bevacizumab, a bone marrow (BM) chimera-human GBM xenograft mouse model was developed. The BM of NOD/SCID mice were reconstituted with red fluorescent protein (RFP)-BM cells to create a chimeric mouse and then GFP-U87 cells were injected intracranially to create an orthotopic GBM xenograft. B20.4.1.1, the mouse analog of Bevacizumab or vehicle control was administrated one-week post intracranial injection of cells and the tumors were harvested two-weeks after treatment. The tumors were processed into single cell suspension prior to fluorescence-activated cell sorting (FACS). Both U87 tumor cells (GFP+) and BM-derived macrophages (RFP+/F4/80+ cells) were sorted and pooled for RNA extraction and microarray analysis (Fig. 6a). We detected a significant increase in TNFα expression in the B20.4.1.1 treated GBM xenografts compared to control as shown by ELISA analysis (n = 5, p < 0.05, Fig. 6b, left panel). The percentage of BM-derived macrophages (RFP+/F4/80+ cells) in tumor cells (GFP cells) were analysed by flow cytometry. B20.4.1.1 treatment increased the amount of BM-derived macrophages in tumors (GAMs) by 1.9 fold compared to control (Fig. 6b, middle panel). Additionally, B20.4.1.1 upregulated TNFα gene expression in these GAMs as determined by quantitative PCR in RNA sample extracted from pooled cells sorted from 5 different tumors (Fig. 6b, right panel). Most strikingly, all fourteen genes in the TNFα signaling pathway were upregulated by greater than twofold in GAMs treated with B20.4.1.1 compared to control (Fig. 6c). The upregulation of TNFα in GAMs was associated with an increase in activated ECs (as measured by aMVD) in B20.4.1.1 treated tumors (Fig. 6d). These experimental data show that B20.4.1.1 treatment stimulates macrophage recruitment to GBMs and increases the local concentration of TNFα, which in turn promotes EC activation and may contribute to resistance to AATx.
Upregulation of TNFα in GAMs predicted failed response to AATx
We sought to determine whether the above experimental findings were relevant to human GBM samples treated with Bevacizumab. To date, two phase III clinical trials with Bevacizumab have failed in human GBMs. Our study herein demonstrates that elevated TNFα expression in GAMs induced EC activation, providing a possible mechanism for resistance to AATx. To examine whether failure of AATx is due to elevated TNFα expression in GAMs, we examined expression of TNFα in CD68+ GAMs in GBM samples that were collected and processed prior to treatment with Bevacizumab and correlated the results with clinical outcome, using OS in Bevacizumab treatment as the end-point. We performed co-immunostaining of TNFα and CD68 on 13 GBM samples from patients to detect TNFα expression in GAMs per mm2 of tumor section. TNFα high and low groups were separated by median value (Fig. 6e) and correlated to survival following treatment with Bevacizumab. Patients in the low TNFα group had a median survival of 352 days following start of therapy, while patients in the high TNFα group had a median survival of 116 days after treatment with Bevacizumab, demonstrating that TNFα levels predict a statistically significant difference in response to Bevacizumab (log-rank test, p < 0.05, Fig. 6f). These results suggest that TNFα expression in GAMs can predict response to AATx with Bevacizumab in GBM patients and provides support for combining TNFα inhibition and Bevacizumab in clinical trials.