EGFR wild-type amplification and activation promote invasion and development of glioblastoma independent of angiogenesis

We rigorously characterized the histological features of intracerebral xenografts established from 12 different patients with primary GBMs. Two distinct tumor subtypes were identified based on their ability to induce angiogenesis. As described previously, tumors in the first subtype were highly invasive and showed no signs of angiogenesis (Fig. 1a, Fig. S1) [53]. In the second subtype, tumors displayed angiogenic growth characterized by dilated macrovessels and endothelial hyperplasia. In addition, the majority of tumors also displayed typical microvascular proliferations and/or pseudopalisading necrotic areas, which are angiogenic hallmarks of GBM growth (Fig. 1a, Fig. S1, Table S1). Quantification of tumor vessels using the vascular marker vWF revealed that the area fraction of vascular elements was significantly higher in the angiogenic phenotype compared to its non-angiogenic counterpart (Fig. 1b). The angiogenic and invasive phenotypes were also verified by MRI. MRI is routinely used in the clinical setting to distinguish angiogenic (glioblastoma) from non-angiogenic tumors (low grade gliomas). Contrast enhancement represents vascular permeability and is only seen in highly angiogenic tumors [3]. In our model, the angiogenic phenotype showed more demarcated tumors and contrast enhancement on MRI, while the invasive tumors showed ill-defined borders and no contrast enhancement (Fig. 1c). To further compare the angiogenic and invasive phenotypes, we assessed expression of the pro-angiogenic factors VEGFA, ANGPT1, ANGPT2 and bFGF. As shown in Fig. 1d, the pro-angiogenic factors were mainly associated with the angiogenic phenotypes.

When comparing the invasive capabilities of both tumor subtypes, we found that the non-angiogenic phenotype showed extensive single cell infiltration into cortical brain areas in addition to migration into the contralateral hemisphere while tumor growth in the angiogenic phenotype was largely confined to the white matter with a more defined border and significantly fewer cells invading into the cortex (Fig. 1e, f).

Since various stem cell markers have been associated with a tumorigenic phenotype in human gliomas [30], we analyzed invasive and angiogenic xenografts for the expression of the stem cell markers nestin, sox2 and CD133. All tumors of either phenotype showed expression of both nestin and sox2. However, CD133, a controversial marker for stem-like cancer cells, was expressed at low to intermediate levels in two out of three xenografts of either phenotype (Fig. S2). Thus, the expression of stem cell markers was not linked to a specific phenotype.

To determine whether angiogenic and non-angiogenic tumors differed at the molecular level, we performed array comparative genomic hybridization (aCGH). While this analysis revealed several common genomic changes in both groups, such as losses of chromosomes 9p and 10q, amplification of EGFR occurred exclusively in the non-angiogenic, invasive phenotype (Fig. 2a; Table 1). FISH verified these results (Fig. 2a; Table 1), but also revealed some interesting patterns regarding the presence of EGFR amplification. High EGFR amplification was detected in the tumor cells from the invasive xenografts while corresponding patient tumors showed variable levels of EGFR amplification in the tissue sections (Fig. 2 b). In our model, the wtEGFR expression was strongly associated with the amplification status of the gene. Importantly, activation (phosphorylation) of EGFR was observed in tumors with genomic EGFR amplification, while tumors without amplification lacked EGFR expression and activation (Fig. 2 a). To investigate whether there was a difference in major downstream signaling pathways of both phenotypes, we performed western blotting of pAkt, pMAPK and pStat3. These pathways were activated in both phenotypes (Fig. S3), suggesting that common genomic changes independent of EGFR are sufficient drivers of major downstream signaling events.

We have previously shown that non-angiogenic tumors can spontaneously switch to an angiogenic phenotype upon in vivo passaging [53]. To determine whether wtEGFR amplification, expression, and activation were changed during the angiogenic switch, we analyzed four EGFR-amplified, non-angiogenic tumors by serial in vivo passaging. We observed that two tumors switched to an angiogenic phenotype, while the other two maintained a stable invasive phenotype (Fig. 2c). Both phenotypes harbored EGFR amplification; however, while the stably invasive tumors showed high levels of wtEGFR expression and phosphorylation (pEGFR), both proteins were downregulated in the core of tumors that switched to an angiogenic phenotype (Fig. 2c). In addition to the wtEGFR protein, various mutants exist, such as EGFRvIII [42]. However, EGFRvIII is expressed in fewer tumor cells than wtEGFR in most patient samples and amplification of EGFRvIII occurs to a much lesser extent compared to wtEGFR [4]. By evaluating EGFRvIII in our xenografts, we found its expression only in tumors with wtEGFR amplification (Fig. S4; Table 1), suggesting a strong association between wtEGFR amplification and EGFRvIII expression which has also been described in patient biopsies [57]. Interestingly, EGFRvIII was lost in stably invasive tumors after serial in vivo passages while it was expressed in xenografts that switched to an angiogenic phenotype (Fig. 2d, e). This shows opposite regulations of wtEGFR and EGFRvIII in our xenograft system and indicates that wtEGFR expression is necessary to drive the invasive phenotype, while EGFRvIII is dispensable and might be involved in stimulating angiogenic tumor growth when wtEGFR expression is lost. The relevance of EGFRvIII for tumor angiogenesis has been previously described [6, 11, 36].

As our animal model might mimic the development and progression of EGFR-amplified tumors in patients, we analyzed a tissue microarray from 206 GBM patients for EGFR amplification, protein expression and activation (pEGFR). EGFR amplification was found in 87 (41.4 %) patients which is in close agreement with previous results [12]. Moderate-to-high wtEGFR expression in a significant fraction (>10 %) of tumor cells (score ≥ 4) was detected in 77 (88.5 %) out of 87 EGFR-amplified tumors and in 26 (21.9 %) out of 119 non-amplified tumors (Table S2). Notably, a significant fraction (>10 %) of moderate-to-high pEGFR positive tumor cells (score ≥ 4) was exclusively found in amplified tumors with high EGFR expression (score ≥ 4). However, only 22 (25.3 %) of 87 amplified tumors had this moderate-to-high pEGFR (score ≥ 4), suggesting that activated EGFR is confined to a subpopulation within EGFR-amplified tumors (Table S2). Cells with activated EGFR were found in non-angiogenic or invasive areas where angiogenic vessels were rarely detected (Fig. 3a, upper panel). In contrast, angiogenic biopsies from EGFR-amplified tumors with abnormal vessels and high angiopoietin-2 levels showed only a few single or no EGFR-activated cells (Fig. 3a, lower panel). The expression of wtEGFR did not show the same association to invasiveness as pEGFR (Fig. 3a) suggesting that the activation of EGFR, rather than absolute levels of EGFR protein, is the most important factor determining non-angiogenic tumor growth. In addition, we analyzed in detail whole biopsy tissues from seven patients that had moderate-to-high pEGFR expression (score ≥ 4). To verify that pEGFR positive areas were non-angiogenic/invasive, we compared these areas to angiogenic areas from the same biopsies. Indeed, pEGFR positive areas had smaller vessels that covered significantly smaller area fractions of the tumor tissue, compared to pEGFR negative, angiogenic areas (Fig. 3b, c, Fig. S5). Thus, the activation of EGFR correlated with non-angiogenic, invasive growth in both the animal model and in patient biopsies.

To functionally verify that activation of EGFR is driving invasion in human GBM, we used a clinically relevant anti-EGFR antibody, cetuximab, to block wtEGFR activation in the stably invasive, EGFR-amplified xenografts. To ensure drug delivery to the invasive cells across the BBB, we used a CED technique [22], which leads to a broad distribution of compounds within the brain as described previously [19]. We delivered cetuximab to tumors 6 weeks after tumor implantation using osmotic minipumps. After 4 weeks of continuous cetuximab administration, we observed a dramatic effect on tumor growth and invasion as evidenced by MRI and histology (Fig. 4). The treated tumors were significantly smaller (125.75 mm³) compared to control tumors (810 mm³) (Fig. 4a, b), which caused neurological symptoms at this stage. Reflecting a slower growth rate, the proliferative activity of treated tumors was significantly lower compared to the controls (Fig. 4c). Importantly, there was a significant block of invasion into the brain in the treatment group, which was most pronounced proximal to the injection site, where tumors showed a demarcated border (Fig. 4d, e). In contrast, the same area in control tumors showed a prominent invasion of tumor cells (Fig. 4d, e). Distal to the injection site, tumor cells in the treatment group showed gradually increased invasiveness, yet infiltration was substantially less compared to control tumors (Fig. 4d, e). This effect might be explained by a lower concentration of cetuximab distal to the injection site. Upon terminating cetuximab administration, the tumors were followed up for two additional weeks. This revealed a reversion to the invasive phenotype similar to what was seen in control tumors (Fig. 4d, e). Thus, these data show that a continuous inhibition of wtEGFR activation is necessary to inhibit GBM invasion.

As cetuximab treatment was transient and performed on established tumors, we chose an additional method to stably block wtEGFR activation in the majority of tumor cells through the whole period of tumor development and progression. Using lentiviral vectors, we genetically modified EGFR function by introducing an inactive mutant of the receptor in EGFR-amplified tumors. This construct downregulates EGFR signaling in a dominant-negative manner (EGFR-CD533) [13]. Prior to implantation, tumor cells were mock infected, infected with a lentiviral GFP control vector or with a lentiviral EGFR-CD533 construct. Expression of dominant-negative EGFR protein was verified by western blot of tumor tissue (Fig. 5a). Notably, the expression of wtEGFR was strongly downregulated in the dominant-negative group compared to the control tumors (Fig. 5a, Fig. S6).

Kaplan–Meier analysis showed a significantly prolonged survival of animals in the EGFR-CD533 group compared to the control group (Fig. 5b; log-rank: P < 0.05), indicating a slower development of the modified tumors. To assess the tumor growth characteristics, we performed MRI. The images revealed circumscribed, contrast-enhancing tumors in the EGFR-CD533 group, while controls showed highly invasive, non-enhancing lesions (Fig. 5c, Fig. S6). The MRI findings were confirmed both at the macroscopic and microscopic level. Angiogenic growth was clearly evident in the EGFR-CD533 tumors, which were circumscribed and harbored microvascular proliferations and necroses. Angiogenesis was absent in the diffusely invasive growing tumors of the control group (Fig. 5d, Fig. S6). Importantly, EGFR activation (pEGFR) was significantly inhibited in the EGFR-CD533 tumors compared to the control tumors (Fig. 5d, e, Fig. S6).

To determine whether the tumors were hypoxic, pimonidazole staining was performed. While hypoxic areas were identified around necroses in the EGFR-CD533 group, no hypoxic regions were observed in control tumors (Fig. 5d). We further determined the expression of HIF1A and VEGFA to verify induction of an angiogenic switch at the molecular level. Upregulation of HIF1A and VEGFA occurred in the EGFR-CD533 transduced, angiogenic tumors whereas both proteins were absent/less expressed in the control tumors (Fig. 5f, Fig. S6). Functional analysis of gene expression data using Ingenuity Pathway Analysis (IPA) showed that the biological function “angiogenesis” was highly enriched (p 2.22 × 10⁸) in the dataset. Z score analysis revealed that angiogenesis was significantly increased (z score 3,305) in EGFR-CD533 tumors (Table S3). Subsequently, we used IPA to query for significantly altered upstream regulators in the dataset. HIF1A was the top enriched upstream regulator (P 3.45 × 10¹¹; Fig. S7).

To analyze whether inhibition of EGFR activation affected major downstream signaling pathways, we performed western blots for phosphorylated Akt, MAPK and Stat3. While Akt and MAPK activation were not changed, phosphorylation of Stat3 was enhanced in the EGFR-CD533 group (Fig. 5g). Thus, major downstream signaling was not blocked by inhibiting EGFR activation which is in line with previous clinical studies using EGFR inhibitors [24, 33, 41] and might in part explain the resistance to anti-EGFR therapy.

To verify that inhibition of EGFR activity affected the invasiveness of tumor cells, we performed detailed cell counts, which revealed less invasive cells in cortical areas within the EGFR-CD533 group compared to the control group (Fig. 6a, Fig. S6). Loss of invasiveness and cell motility is often accompanied by changes in morphology, also referred to as mesenchymal to epithelial transition (MET) [26]. In the EGFR-CD533 group, histology and nestin staining showed tumor cells with an epithelial-like phenotype compared to mesenchymal control cells (Fig. 6 b). MET was confirmed at the molecular level by analyzing vimentin, snail and twist expression, which are established markers for mesenchymal tumor cells and are upregulated upon epithelial to mesenchymal transition as well as during the metastatic process of epithelial cancers [8, 61]. Vimentin, snail and twist were all strongly downregulated in EGFR-CD533 tumors as compared to controls (Fig. 6c).