Combinations of Bevacizumab and Erlotinib Show Activity in Colorectal Cancer Independent of RAS Status

This article is featured in Highlights of This Issue, p. 2465

The epidermal growth factor receptor (EGFR) and the vascular endothelial growth factor (VEGF) pathways are validated targets for treatment of patients with metastatic colorectal cancer (mCRC). There is extensive cross-talk between these two signaling pathways (1) and combinations of VEGF(R) and EGFR-directed compounds have consistently shown at least additive activity in preclinical models (2). However, the results of clinical trials were disappointing because the addition of EGFR-targeted monoclonal antibodies (mAb) to bevacizumab plus chemotherapy was no better, in terms of overall survival, than bevacizumab plus chemotherapy alone, even for patients with wild-type RAS tumors, and was detrimental, in terms of progression-free survival, for most patients with KRAS-mutant tumors (3, 4).

To determine if these disappointing results were pathway- or agent-related, we compared the antitumor activity of two mAbs with two small molecule tyrosine kinase inhibitors (TKI) in the same colorectal cancer xenograft models (5). The results showed that the combination of the two antibodies was no more active than either agent alone while the combination of the TKIs showed synergistic antitumor activity. Unexpectedly, the TKI combination was also active in colorectal cancer models with mutant RAS status (5), indicating that mutant RAS may not be a limiting factor for EGFR-directed TKIs like it is for the mAbs (6–8). These findings provided the rationale for the current study, which shows that erlotinib enhances the in vivo activity of bevacizumab independent of RAS status. We further demonstrate a differential activity of cetuximab and erlotinib on EGFR-mediated functions in colorectal cancer cells revealing fundamental differences between mAbs and TKIs targeting the same pathway.

Bevacizumab (Avastin) was provided by Roche, cetuximab was from Merck, and erlotinib (Tarceva) was purchased from LC Laboratories.

Colorectal cancer cells were maintained in cell culture as described previously (9). SW48, LS174T, and DLD-1 colorectal cancer cells were a kind gift from Richard Hamelin (Saint-Antoine Research Center, Paris, France), while SW480 and LIM1215 cells were purchased from American Type Culture Collection. HT-29, CoLo205, and SW620 cells were kindly provided by Richard Camalier (NCI, Bethesda, MD) while HCT-116 cells were a generous gift from Bert Vogelstein (John Hopkins, Baltimore, MD). Upon defreezing and prior to each animal experiment, cells were monitored for mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza). Cells were characterized by next-generation sequencing of 50 cancer-related genes using the Ion AmpliSeq Cancer Hotspot Panel v2 (Thermo Fisher Scientific) and were replaced after 3 months in culture.

The antitumor effects of bevacizumab, erlotinib, and their combination were evaluated in athymic mice (female NMRI-Foxn1 nu/nu, 6 weeks old) from Taconic bearing SW48, HT-29, or SW620 xenografts. Two to 5 million cells were injected into the right flank, and the treatments were started when the tumors were palpable. Animals were weighed daily and the tumor size was determined three times per week (10). Tumor volumes (mm³) were calculated according to the formula: [(length × width²)/2]. Boxplot analysis was carried out using the GraphPad prism 6.04 software (GraphPad). Treated/control (T/C) values were calculated as follows: average tumor volume of treated animals/average tumor volume of control animals × 100. Animals were treated according to institutional guidelines and the protocol was approved by the institutional ethics committee for animal experimentation (Institutional Animal Care and Use Committee).

The following antibodies were used for immunohistochemistry (IHC) analysis: anti-CD31 (#550274, BD Bioscience) to determine the vascular density, anti-VEGF-A (sc-152, Santa Cruz Biotechnology, CliniSciences), anti-phospho-VEGFR1 (07-758, Millipore) that recognizes phospho-Tyr 1213, anti-phospho-VEGFR2 (# 2478, Cell Signaling Technology, Ozyme) that recognizes phospho-Tyr 1175, anti-phospho-EGFR (sc-12351, Santa Cruz Biotechnology) that recognizes phospho-Tyr 1173, anti-EGFR (sc-03, Santa Cruz Biotechnology), and anti-amphiregulin (AF262, R&D System). The relevant Cy3-conjugated secondary antibodies were obtained from Jackson ImmunoResearch.

The following antibodies were used for Western blot analysis: anti-phospho-EGFR antibody (#3777, Cell Signaling Technology) that recognizes Tyr1068 EGFR, anti-EGFR antibody (sc-03, Santa Cruz Biotechnology) and anti-actin-HRP antibody (sc-1615, Santa Cruz Biotechnology) followed by incubation with the appropriate horseradish peroxidase-conjugated secondary antibodies (Cell Signaling Technology).

Biomarker analysis was carried out with tumors collected after 4 weeks of treatment and processed for immunohistochemistry (IHC) analysis. All images were captured by a fluorescence microscope, and the fluorescence intensities were determined by the MetaMorph software (Universal Imaging Corporation) for quantitative image analysis. When indicated, tumors were counterstained by 4′,6-diamidino-2-phenylindole to reveal the nuclei of individual cells. For the quantitative analysis of the signal intensity, the data represent the average fluorescence intensity of treated tumors compared with the treatment intensity of control tumors and is the average of 6 fields/tumor for at least 3 different tumors. Blood vessel density is expressed as the CD31-positive area, in percentage of total and represent the averages of at least 6 fields/tumor for at least 3 different tumors.

For the determination of pEGFR expression by the TECs, double labeling was carried out for pEGFR and CD31 and the degree of colocalization was determined by semiquantitative analysis as illustrated in Fig. 4. Only blood vessels where the entire rim could be assessed were included in the analysis with at least 12 blood vessels per treatment.

Tumor tissues were collected from untreated, frozen tumors (3 tumors per treatment) and protein extracts were prepared in RIPA buffer according to the manufacturer's instructions. VEGF levels were determined by Quantikine ELISA (# DVE00 and MMV00, R&D Systems) for human, tumor-derived VEGF and for murine, stroma-derived VEGF, respectively. Amphiregulin levels were determined by Quantikine ELISA (R&D Systems, # DAR00, human AREG). VEGFR1 and VEGFR2 levels were determined by DuoSet ELISA (DY321B and DYC1766, R&D Systems). The values represent the average of 3 independent experiments, each done in duplicate.

For Western blot, equal amounts of proteins (100 μg/lane) were loaded into SDS-PAGE gels, transferred onto nitrocellulose membranes and blotted with antibodies directed against phosphorylated and total EGFR. Protein expression was quantified by densitometric analysis of the immunoblots using Image Lab software (Bio-Rad) after normalization with β-actin.

Cells were seeded and allowed to attach for 24 hours followed by 72- hours incubation in 5% FCS in the absence or presence of cetuximab (1 μg/mL) or erlotinib (3 μmol/L) as described (11). Conditioned media were collected and concentrated using Amicon ultracentrifugal filter units. The levels of secreted TGFα and amphiregulin were determined by Quantikine ELISA (R&D Systems, #DTGA00 for TGFα and #DAR00 for amphiregulin) according to the manufacturer's instructions. The values represent the average of 3 independent experiments, each done in duplicate.

Cell migration was assessed by the Transwell assay (Boyden chamber, Dutscher) according to the manufacturer's instructions. Briefly, 50,000 to 100,000 cells in serum-free media with or without erlotinib or cetuximab were plated in the upper chamber on membranes with a pore size of 8 μm while the lower chamber contained culture media with EGF (20 ng/mL) as chemoattractant. After 6 hours, remaining cells were removed from the top side of the inserts whereas migrating cells on the bottom of the inserts were stained with Diff-Quik (Thermo Fischer Scientific), and all migrating cells were counted. Results are expressed as means ± SEM and represent triplicate samples from at least two independent experiments.

ANOVA was performed to determine the significance of observed differences between groups using the tool pack from Excel (Microsoft). Post hoc comparisons were made using Student paired t test using GraphPad Prism (GraphPad Software). Differences between two groups are presented as the mean ± SEM or mean ± SD as noted in the figure legends. All tests were two-sided and P < 0.05 were considered statistically significant.

Three different colorectal cancer xenograft models were used including SW48 [RAS wild-type (WT), bevacizumab sensitive], HT-29 (RAS WT, bevacizumab resistant), and SW620 (RAS mutant and bevacizumab sensitive; ref. 10). The antitumor activities of bevacizumab (1 mg/kg) and erlotinib were comparable for SW48 and SW620 xenografts with T/C values (average tumor volume of treated animals/average tumor volume of control animals × 100) of 55 and 54 for bevacizumab and 65 and 74 for erlotinib, respectively (Fig. 1). Combined treatment with bevacizumab and erlotinib increased the antitumor activity to T/C values of 33 and 39, respectively, which is significantly different from bevacizumab alone (P < 0.01). To confirm the activity of the bevacizumab + erlotinib combination toward the RAS-mutant SW620 tumors, the experiment was repeated with a standard dose of bevacizumab (5 mg/kg). Under these conditions, the T/C value was 27 for bevacizumab alone and 16 for the combination (P < 0.001; Supplementary Fig. S1, top).

Bevacizumab exposure (5 mg/kg) was accompanied by modest tumor growth inhibition for HT-29 xenografts (T/C of 68), whereas erlotinib treatment was associated with unexpected strong antitumor activity (T/C of 36). The combination of the two drugs gave a T/C value of 26, which is significantly (P < 0.001) more than what was observed for bevacizumab alone (Fig. 1), whereas the difference between erlotinib and erlotinib + bevacizumab did not reach significance (P = 0.07). It should be noted that in order to detect a meaningful difference between the agents when given alone or in combination, most studies were carried out with suboptimal doses of the two drugs.

The potent activity of erlotinib toward HT-29 xenografts was unexpected because HT-29 tumors show limited sensitivity to the EGFR-targeted antibody cetuximab (5). HT-29 cells express the BRAF V600E mutation (http://cancer.sanger.ac.uk/cosmic). To explore if the expression of mutant BRAF was associated with erlotinib sensitivity, we selected a different BRAF V600E-mutant cell line, CoLo205. However, the response of CoLo205 tumors to bevacizumab, erlotinib, and their combination (T/C values of 53, 64, and 32, respectively) was comparable with SW48 and SW620 rather than with HT-29 (Supplementary Fig. S1, bottom). Therefore, BRAF status as such cannot explain the unexpected erlotinib sensitivity of HT-29 xenografts.

Bevacizumab has preferential activity for human VEGF. To estimate the relative contribution of human tumor–derived and murine stromal-derived VEGF within the tumor microenvironment, tissue extracts were prepared from untreated SW48, HT-29, and SW620 tumors, and the concentration of murine and human VEGF was determined by species-specific ELISA analysis (data not shown). The results show that human VEGF represents more than 95% of the total VEGF in the tumor environment in agreement with previous findings (10, 12).

The microvascular density of SW48, HT-29, and SW620 xenografts was compared by immunohistochemistry with a CD31-directed antibody followed by quantitative image analysis. The results (Fig. 2, top left) indicate that bevacizumab reduced the microvascular density by ∼65% in the bevacizumab-sensitive SW48 and SW620 xenografts compared with only ∼30% in the bevacizumab-resistant HT-29 xenografts. In comparison, erlotinib treatment was accompanied by ∼50% reduction of the vascular density in all 3 xenograft models. Interestingly, the addition of erlotinib to bevacizumab strongly diminished the microvascular density of HT-29 xenografts (from 30% to 64% of the vehicle control), suggesting that EGFR signaling contributes to the bevacizumab resistance of this tumor model. The combination of bevacizumab + erlotinib was also significantly more effective than bevacizumab alone for the SW48 and SW620 tumor models, although the effect was less pronounced than for HT-29.

Bevacizumab exposure was accompanied by ∼40% increased expression of tissue-associated VEGF that diminished significantly (P < 0.01) when erlotinib was added to bevacizumab (Fig. 2, top right). Most colorectal cancer cells and tumors express functional VEGFR1/Flt-1 that promotes cellular survival under environmental stress (13). Bevacizumab exposure was accompanied by increased levels of active phospho-VEGFR1 for all xenograft models which was attenuated (P < 0.001) by the addition of erlotinib (Fig. 2, middle right and bottom). This was most marked for the HT-29 xenografts where the levels of phospho-VEGFR1 were 171% in the presence of bevacizumab alone and 71% in the presence of bevacizumab + erlotinib, compared with untreated vehicle controls. The differences in VEGFR1 phosphorylation were not linked to altered protein levels, as determined by ELISA of total VEGFR1 in tumor extracts (Supplementary Fig. S2). In bevacizumab-treated tumors, the active, autophosphorylated form of pVEGFR1 was mostly localized inside the cells as indicated by a prominent cytoplasmic signal (Fig. 2, bottom, indicated with a white arrow).

In addition to VEGFR1, colorectal cancer xenografts may express low levels of VEGFR2/Flk-1. Bevacizumab exposure was accompanied by up to 2-fold increase in the levels of phospho-VEGFR2 that was significantly attenuated in the presence of erlotinib, with the exception of the SW620 model, where the modest decrease did not reach significance (Fig. 2, middle right). It is noticeable that the levels of VEGFR1 and VEGFR2 protein are very different, with about 20-fold less VEGFR2, compared with VEGFR1, in SW48 and SW620 tumors and almost 80-fold less VEGFR2, compared with VEGFR1, in HT-29 xenografts (Supplementary Fig. S2). These differences may explain the current controversy concerning the expression of VEGFR2 by colorectal cancer cells (13). Taken together, our findings show that bevacizumab activates autocrine VEGF signaling in all three xenograft models which can be attenuated by erlotinib.

Next, the levels of total and phosphorylated EGFR were determined (Fig. 3, top). The results show that erlotinib exposure was accompanied by decreased levels of the active, autophosphorylated form of EGFR which was most pronounced for SW48 (66% inhibition, compared with the vehicle control) followed by SW620 (50% inhibition) and HT-29 xenografts (35% inhibition). Unexpectedly, bevacizumab treatment was accompanied by strong EGFR activation ranging from ∼140% for the bevacizumab-sensitive SW48 and SW620 xenografts to more than 200% for the bevacizumab-resistant HT-29 xenografts. The addition of erlotinib to bevacizumab was accompanied by a highly significant (P < 0.001) decrease in phospho-EGFR levels in all models shown by quantitative immunohistochemistry as well as by Western blot analysis (Fig. 3, top and middle). In comparison, bevacizumab had no detectable influence on total EGFR levels whereas erlotinib exposure, alone or in combination, was accompanied by a modest increase of total EGFR that never reached significance (Fig. 3 top and middle).

As observed for pVEGFR1, bevacizumab treatment was accompanied by a prominent cytoplasmic signal of pEGFR (Fig. 3, bottom, indicated with a white arrow).

Tumor-associated endothelial cells (TEC) frequently express functional EGFR (14, 15). To characterize the expression of active EGFR on the TECs, a semiquantitative approach was developed as illustrated (Fig. 4, top left). Tumors were double-labeled for phospho-EGFR and CD31, and the degree of colocalization was determined as detailed in the figure legend. The results show that bevacizumab treatment was accompanied by a strong increase in TEC-associated phospho-EGFR while phospho-EGFR was attenuated following treatment with erlotinib. Importantly, the addition of erlotinib to bevacizumab was associated with a highly significant (P < 0.001) decrease in TEC-associated phospho-EGFR, compared with bevacizumab alone, for all three xenograft models.

The EGFR ligands amphiregulin (AREG) and TGFα are known to form positive feedback loops with activated EGFR (16–18). Amphiregulin is of particular interest, because it is expressed at high levels in colorectal cancer and has potential predictive value for the response to EGFR-targeted agents (19, 20). Tumor extracts were prepared, and the ligand expression was determined by ELISA analysis. A modest increase in amphiregulin expression was observed in the bevacizumab-treated tumors while erlotinib treatment resulted in a ∼50% decrease. Importantly, amphiregulin expression was significantly decreased in all bevacizumab + erlotinib groups, compared with bevacizumab alone. For TGFα, the levels were too close to the detection limit to give reproducible results.

It has been reported that amphiregulin promotes endothelial tube formation in vitro, suggesting a direct role for amphiregulin in tumor angiogenesis (21). To determine if tumor-derived amphiregulin colocalize with the TECs, tumors were double-labeled for human amphiregulin and CD31. Unexpectedly, the results (Fig. 4, bottom) show prominent colocalization of amphiregulin and the TECs (indicated in yellow) compared with the more faint red labeling of amphiregulin in the tumor cells, coherent with a close in vivo association between the TECs and tumor-derived amphiregulin.

EGFR activation is accompanied by a positive feedback loop with some of its ligands, including TGFα and amphiregulin (16–18). Intriguingly, a recent study reports that EGFR-directed antibodies are unable to downregulate amphiregulin and TGFα in RAS-mutant colorectal cancer cells, which could, at least in part, explain the lack of activity of the mAbs toward such cells (11). The activity of erlotinib and cetuximab toward different RAS-mutant colorectal cancer cell lines was compared (Fig. 5). The results show that cetuximab exposure increased TGFα expression up to 4-fold in all cellular models, but had no significant influence on amphiregulin. In clear contrast, erlotinib decreased both TGFα and amphiregulin secretion, with highly significant (P < 0.001) differences between cetuximab and erlotinib in all cases (Fig. 5, top).

EGFR activation promotes tumor cell migration. Four colorectal cancer cell lines with good migratory capacity were selected, including LIM1215 (RAS WT), HCT-116, LS174T, and SW480 (all RAS mutant). EGF strongly stimulated the migration of all cell lines, which could be attenuated by erlotinib (Fig. 5, middle and bottom). In clear contrast, cetuximab was only able to attenuate migration of the RAS WT LIM1215 cells, without any detectable influence on the migration of the RAS-mutant cells. Importantly, the viability at the end of the incubation period was comparable for all groups (Supplementary Fig. S3).

We here report that erlotinib increases the activity of bevacizumab in colorectal cancer models independent of RAS status and bevacizumab resistance. We further document differential activity of cetuximab and erlotinib on EGFR-mediated functions in colorectal cancer cells as a function of RAS status revealing fundamental differences between mAbs and TKIs targeting the same pathway. These findings inspired the DREAM phase III clinical trial (22) and provide the mechanistic framework to better understand why the bevacizumab + erlotinib combination would be more active than bevacizumab alone.

Besides RAS status, there is increasing interest in the impact of primary tumor location. Current evidence suggests that metastatic colorectal cancer (mCRC) arising from different sides of the colon (left vs. right) have different clinical outcomes, which may, at least in part, be a reflection of the different embryonic origin of the normal tissue forming the left and right sides of the colon. In mCRC, patients with right-sided tumors have generally worse prognosis than those with left-sided tumors (23, 24). Furthermore, retrospective analysis of pooled clinical trials suggests that EGFR-targeted antibodies may be more active toward left-sided RAS WT tumors compared with right-sided RAS WT tumors (24–26). Interestingly, it has been reported that VEGF targeting may also differ between right- and left-sided tumors. In particular, bevacizumab may only provide a survival advantage for left-sided and rectal tumors (27, 28), suggesting that right- and left-sided tumors may respond differently to both EGFR- and VEGF-targeted anticancer agents. However, a large prospective trial is still needed to confirm these data.

Unfortunately, little is known about the origin of the vast majority of colorectal cancer cell lines including the ones used here. However, based on putative site-specific markers, it has been suggested that HT-29 is likely derived from a right-sided tumor while SW620 is likely derived from a left-sided tumor (29). Although the origin of these cell lines can never be determined with full certainly, it is tempting to speculate that erlotinib, alone or in combination with bevacizumab, may prove particularly interesting for treatment of patients with right-sided mCRC, a patient subgroup in need of novel, active agents.

Usually, EGFR inhibition is believed to principally influence tumor cells while VEGF targeting is acting on endothelial cells. The results presented here provide evidence for a more complex model. Pruning of the tumor-associated microvasculature is accompanied by tumor cell hypoxia, activation of tumor-associated VEGF signaling as well as EGFR activation. In addition, vascular pruning interfered with tumor cell receptor recycling as evidenced by a prominent cytoplasmic signal for both pVEGFR1 and pEGFR in agreement with an influence of the oxygen-sensing pathway on endocytosis (30). Interestingly, bevacizumab exposure was also accompanied by EGFR signaling activation and accumulation of tumor-derived amphiregulin in the TECs. Taken together, these results suggest that both tumor cells and TECs are targeted by EGFR blockage.

The finding that an EGFR-targeted TKI may have different activity compared with EGFR-targeted antibodies in vitro as in vivo is initially surprising. However, it should be noted that the mAbs and the TKIs owe their activity to different mechanisms. Binding of the mAbs to EGFR is followed by endocytosis and lysosomal degradation, thereby decreasing receptor density on the cell surface (31). In the case of the TKIs, it is the catalytic activity of EGFR that is inhibited without an immediate effect on EGFR protein levels. Importantly, the activity of the TKIs is not restricted to surface-associated EGFR but includes additional pools of EGFR, like the active EGFR contained in endosomes following receptor internalization. Furthermore, recent results suggest that the interaction between EGFR and Ras proteins ceases within minutes after EGFR activation, with EGFR accumulating in the endosomes while Ras remain associated with the plasma membrane, thereby providing a physical separation between the two molecules and rapid signal downregulation (32).

Traditionally, EGFR signaling has been associated with canonical cellular signaling pathways like Ras/MAPK, PI3K/Akt, phospholipase C/PKC, STAT, and Src pathways as depicted in Fig. 6. However, several novel EGFR signaling pathways have been described recently including the binding of EGFR to and phosphorylation of the tyrosine kinase Lyn, which then phosphorylate MCM7, a licensing factor critical for DNA synthesis (33), beclin1 (BCN-1), a crucial mediator of autophagy (34), and argonaute 2 (AGO2) a modulator of microRNA processing (35). Importantly, the phosphorylation of AGO2 and Beclin1 is mediated by catalytically active EGFR present on intracellular vesicles after internalization, but before lysosomal degradation (34, 35), thereby representing a subfraction of EGFR that can only be inhibited by EGFR-directed small molecules, but not by EGFR-directed mAbs.

In conclusion, our study provides strong evidence that mAbs and TKIs target different elements of the EGFR signaling pathway, both as single agents and in combination with VEGF-targeted therapies. These findings may provide a rationale for novel therapeutic strategies for treatment of patient subgroups that are currently unlikely to benefit from EGFR-targeted antibodies. We further suggest that differential activity of mAbs and TKIs targeting the same signaling pathway is likely applicable for other tumor types.

A. Savina is an employee of Institut Roche. B. Chibaudel is a consultant/advisory board member for Bayer, Lilly, Roche, and Sanofi. T. Andre reports receiving speakers bureau honoraria from and is a consultant/advisory board member for Roche. A.K. Larsen reports receiving commercial research support from Roche. No potential conflicts of interest were disclosed by the other authors.

The sponsors had no role in the study design, data collection and analysis, interpretation of the results, the preparation of the manuscript, or the decision to submit the manuscript for publication.

Conception and design: A. Savina, C. Tournigand, A. de Gramont, A.K. Larsen

Development of methodology: P. Mésange, A. Bouygues, A.K. Larsen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Mésange, N. Ferrand, A. Bouygues

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Mésange, A. Bouygues, N. Ferrand, A. Savina, B. Chibaudel, C. Tournigand, T. André, A. de Gramont, A.K. Larsen

Writing, review, and/or revision of the manuscript: P. Mésange, A. Bouygues, N. Ferrand, M. Sabbah, A.E. Escargueil, A. Savina, B. Chibaudel, C. Tournigand, T. André, A. de Gramont, A.K. Larsen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Sabbah, A.E. Escargueil, A.K. Larsen

Study supervision: A. Savina, A.K. Larsen

Other (suggestions and technical advice): T. André

We thank Mériam Ayadi, Olivier Bernadini, Karima El Ouadrani, Virginie Megalophonos, Delphine Muller, Amélie Petitprez, and Virginie Poindessous for help with the animal experiments. We appreciate the contribution of Tatiana Ledent from the animal facilities at Saint-Antoine Research Center and Sylvie Dumont and Fatiha Merabtene from the pathology platform at Hôpital Saint-Antoine.

This work was financed in part by research funding to A.K. Larsen from Roche. P. Mésange and A. Bouygues are generously supported by a Mouna Nasrallah fellowship.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.