Mouse PDX Trial Suggests Synergy of Concurrent Inhibition of RAF and EGFR in Colorectal Cancer with BRAF or KRAS Mutations

Colorectal cancer is one of the leading causes of cancer-related mortality worldwide. In the United States, it was estimated that nearly 137,000 people were diagnosed, and more than 50,000 died from the disease in 2014 (1). Although early detection, surgery, and other localized therapeutic intervention are successful in the management of early-stage disease, metastatic colorectal cancers are typically associated with a poor prognosis with the majority of patients dying within 2 years upon diagnosis, resulting in a 5-year survival rate of just 11% (2, 3). The key reasons for the limited success of colorectal cancer–directed therapies include complex genetic alterations and intrinsic heterogeneity of colon tumors, which are more dominant in the metastatic setting (4, 5).

MAPK pathway activation through KRAS or BRAF mutation and EGFR overexpression are common among colorectal cancer patients (4–6). The somatic mutation rates of KRAS and BRAF in colorectal cancer are approximately 50% and 10%, respectively (5, 7, 8). Due to its epithelial origin, colon tumors generally express high levels of EGFR, a member of the ERBB family of RTKs (6). The anti-EGFR antibodies, including cetuximab and panitumumab, both can inhibit ligand binding and EGFR dimerization, which is necessary for full activation of EGFR (9). Cetuximab is a chimeric human-murine IgG₁ antibody, and panitumumab is a humanized IgG₂ antibody (10, 11). Both have become part of the first-, second-, and third-line therapy for colorectal cancer (11, 12). Data from clinical trials have shown the therapeutic benefit of anti-EGFR antibody therapy either as a standalone agent or in combination with chemotherapy regimens (13–15).

It is well documented that KRAS mutation is a key mechanism of resistance to anti-EGFR therapy in colorectal cancer. An initial study published in 2006 showed the lack of response to cetuximab in patients with mutations in KRAS occurring at codons 12 and 13 (16), followed by another study confirming the causal role of these KRAS mutations (17). A host of subsequent studies, including further clinical trials and retrospective analyses of tumor samples, verified the notion that the presence of KRAS mutations underlies the lack of response to anti-EGFR therapies (18–21). These findings led to the issuance of a provisional clinical opinion by the American Society of Clinical Oncology in 2009, stating that patients with metastatic colorectal cancer eligible to receive an anti-EGFR antibody therapy should have their KRAS mutational status profiled, and those with a codon 12 or 13 mutation in KRAS should be excluded from this treatment (22). Beyond intrinsic resistance, KRAS mutations also play an important role in acquired resistance through clonal selection due to tumor heterogeneity (23, 24). In addition to KRAS mutations at codons 12 and 13, MAPK pathway activation through other genetic alterations was demonstrated to be associated with resistance (25, 26). These include KRAS mutations in other codons (19, 27, 28), KRAS amplification (29), and NRAS and BRAF mutations (19, 27, 30, 31). Collectively, these studies have established that MAPK pathway activation is a key factor in treatment stratification of patients with metastatic colorectal cancer, as well as a key biomarker predictive of resistance to anti-EGFR antibodies in this tumor setting.

Although MAPK activation confers resistance to EGFR therapies in colorectal cancer, EGFR activation is associated with resistance of BRAF-mutant colorectal cancer to MAPK inhibitors. For example, BRAF inhibitor vemurafenib was demonstrated to be ineffective for treatment of BRAF-mutant colorectal cancer (32), although it showed significant clinical benefit for melanoma patients with BRAF mutation (33, 34). Recent studies revealed that the unresponsiveness of BRAF-mutant colorectal cancer was due to feedback activation of EGFR that leads to MAPK reactivation (35–37). We have recently developed a class of RAF inhibitors including LSN3074753 and LY3009120, and the latter is being currently evaluated in clinical trials (38, 39). These inhibitors are capable of inhibiting the kinase activities of all RAF isoforms, as well as RAF dimers, and they demonstrated activity against tumor cells with MAPK pathway activation driven by BRAF monomer or RAF dimers including BRAF- or KRAS-mutant colorectal cancer (38–40). We hypothesize that the combinational inhibition of EGFR and MAPK may provide synergistic benefit for patients with colorectal cancer, particularly those with KRAS or BRAF mutation. In this study, we have comprehensively characterized a panel of colorectal cancer patient-derived xenograft (PDX) models at molecular level and conducted a single-mouse (n = 1) PDX trial aimed at evaluating antitumor activities of cetuximab, LSN3074753, or their combination in this setting. We have demonstrated that the n = 1 PDX mouse clinical trial paradigm could reliably recapitulate the clinical results of cetuximab, and that combination of cetuximab and LSN3074753 has synergistic efficacy for colorectal cancer PDX models with KRAS or BRAF mutation.

HCT116 cells with a KRAS mutation, and HT-29 and A375 cells with a BRAF V600E mutation were obtained from the ATCC and stored within a central cell bank that performs cell line characterizations. The cells were passaged for less than 2 months after which time new cultures were initiated from vials of frozen cells. Characterization of these cells is done by a third-party vendor (RADIL), which included profiling (by PCR) for contamination of various microorganisms of bacterial and viral origin. As a result, no contamination was detected. The samples were also verified to be of human origin without mammalian interspecies contamination. The alleles for nine different genetic markers were used to determine that the banked cells matched the genetic profile that has been previously reported. All cells were maintained in DMEM (Thermo Scientific) supplemented with 10% FBS (Invitrogen). RAF inhibitors vemurafenib, dabrafenib, sorafenib, and LSN3074753 were synthesized by Eli Lilly and Company. Antibodies against phospho-ERK1/2, phospho-MEK, and actin were purchased from Cell Signaling Technologies Inc. Antibodies against BRAF and CRAF used for Western blot analysis were from Bethyl Laboratories. Antibodies against BRAF and CRAF used for immunoprecipitation (IP) were from BD Biosciences.

Cells were lysed using MSD lysis buffer (Meso Scale Discovery), containing 1x phosphatase and proteinase inhibitor cocktail (Thermo Scientific). The protein concentration of individual sample was determined with the DC Protein Assay Kit (Bio-Rad). SDS-PAGE was performed on cell lysates containing 20 μg of total protein using 4%–20% Novex tri-glycine gradient gels (Invitrogen). Protein was transferred onto 0.2 μm nitrocellulose membranes using Trans-Blot Turbo Transfer system (Bio-Rad) as per the manufacturer's instructions. Western blot analysis was conducted as described (39, 41, 42). Cell proliferation assay was conducted by CellTiter-Glo as described (39).

HCT116 cells were plated in 10-cm culture dishes and cultured in growth media at 37°C for 24 hours before treatment. Cells were treated with individual compound for 2 hours and lysed in lysis buffer (Thermo Scientific) with 1x proteinase/phosphatase inhibitor cocktail. The cells lysates were then centrifuged at 14,000 rpm for 10 minutes at 4°C. Supernatants were collected for protein quantification and IP. Anti-BRAF or anti-CRAF antibody was coated onto Dynabeads at 1 μg antibody/0.75 mg beads and incubated at room temperature for 20 minutes. Beads were washed with Wash Buffer (PBS + 0.02% Tween-20) three times. Protein lysates were incubated with coated beads at 250 μg lysate/1 μg antibody with rotation at 4°C overnight. Beads were washed with Wash Buffer three times. SDS sample buffer (Novex, NP0007) with reducing agent was added to the beads and heated at 95°C for 5 minutes. Proteins were separated on SDS-PAGE for Western blot analysis.

All animal studies were performed in accordance with American Association for Laboratory Animal Care institutional guidelines. Colorectal cancer PDX models were developed with nude mice in either Oncotest GmbH or Champions Oncology, Inc. as described (43). The mice were individually enrolled for treatment when tumor sizes reached 200 mm³. Mice bearing established tumors were treated with cetuximab at 20 mg/kg twice per week by intraperitoneal injection for 4 weeks, LSN3074753 at 40 mg/kg twice daily orally for 4 weeks, or their combination. Control group with three mice was administered with vehicles.

Whole-exome sequencing was performed on an Illumina HiSeq 2000 using the SureSelectXT Human All Exon v4 (51 Mb) protocol (Agilent). Paired-end sequencing with read length of 100 base pairs and 100X average on-target coverage was achieved. As PDX specimens include both human (tumor) and mouse (stromal) origins, the sequencing reads were mapped to a hybrid human and mouse genome (hg19 and mm10) using the Burrows-Wheeler Aligner MEM to separate human and mouse reads. Human variants were called with SAMtools, Genome Analysis Toolkit (GATK-Unified Genotyper), and FreeBayes (Garrison and Marth). Appropriate filters were used to discard low-quality variants as well as enriching for likely functional somatic mutations by excluding silent variants and germline SNPs.

RNA was prepared with the Illumina TruSeq RNA Sample Preparation Kit v2 to conduct paired-end sequencing on an Illumina HiSeq 2000 at read length of 100 bp and targeted read depth of 100 mol/L. Reads were quality trimmed by sickle v1.33 with default parameters. Sequencing reads were mapped in parallel to both mouse (GRCm38.p1) and human (GRCh37.p13) reference genomes using GSNAP (v.2013-11-27) followed by sorting of reads into “best-match mouse,” “best-match human,” or “equivalent match.” Read counts were generated against exons annotated in NCBI gene models for these genome builds downloaded in December 2013 and then summed at the gene level to provide a single number per gene per sample using a custom Perl script for each species using the respective best-match mapped reads.

Affymetrix SNP6.0 data generation was conducted as per Affymetrix's protocol (http://www.affymetrix.com/support/technical/byproduct.affx?product=genomewidesnp_6). The CEL files were processed using CNIP pipeline described elsewhere (http://software.broadinstitute.org/cancer/software/genepattern/affymetrix-snp6-copy-number-inference-pipeline).

Commercial TaqMan Gene Expression assays were purchased from ThermoFisher Scientific for quantification of AREG (AREG, Assay ID: Hs00950669_m1), EREG (EREG, Assay ID: Hs00914313_m1), and EGFR (EGFR, Assay ID: Hs01076078_m1). Total RNA was isolated from snap-frozen tumor tissue using the TissueLyser with stainless steel beads (5 mm; QIAGEN) and the MagMAX 96 Total RNA isolation Kit (ThermoFisher Scientific). RNA was reverse-transcribed with a high-capacity cDNA reverse transcription kit (ThermoFisher Scientific) and random primers according to the manufacturer's instructions. cDNA was then diluted 1:25 (4 ng/mL) for qPCR use. qPCR was performed in a 384-well format on 20 ng of cDNA (4 ng/mL) with 1 μL of TaqMan Gene Expression Assay (20×) and 10 μL of TaqMan Gene Expression Master Mix (2x) in a final volume of 20 μL. Each sample was tested in triplicate. All assays were run on the ViiA-7 Real-Time PCR System using standard setting (10-minute incubation at 95°C, followed by 40 cycles at 95°C for 15 seconds and 60°C for 1 minute). Absolute values (mRNA copies/ng of cDNA) were derived from a standard curve generated with a serial dilution of plasmid containing the cloned sequence of the target gene. Each target standard curve demonstrated linearity with regression coefficients (r² values) above 0.997. All assays had efficiency between 90% and 100% with a standard curve slope between −3.3 and −3.6. Data were extracted with the ViiA-7 Software (ThermoFisher Scientific).

For each PDX model, the three vehicle animals were summarized by fitting a line to Log volume versus time using robust linear regression. From there, the best median percent change in tumor volume compared with baseline or % regression (%ΔT/T₀) was calculated for values above or below baseline, respectively, for each time point. For each treated animal, the reported result used here was the best median percent response calculated from moving windows of 3 consecutive days. Disease control was defined as having <20% median tumor growth or regression compared with baseline. Complete response (CR) was defined as no measurable tumor volume with nearly −100% regression, and partial response (PR) was defined as −50% to >−100% regression, a criteria used by others in preclinical studies (44). Stable disease (SD) was defined as <20% median tumor growth to >−50% regression compared with baseline. The Kaplan–Meier analysis was performed across PDX models using time to one tumor size doubling.

LSN3074753, N-[2-fluoro-4-methyl-5-[2-methyl-7-(methylamino)-1,6-naphthyridin-3-yl]phenyl]-N′-[(2S)-2-hydroxy-3,3-dimethylbutyl]- urea (Fig. 1A), is a close analog of LY3009120 (Fig. 1B), and both are pan-RAF and RAF dimer inhibitors (38, 39). As described previously (38) and further demonstrated in Supplementary Table S1, LSN3074753 and LY3009120 share a very similar pharmacology profile with similar in vitro potencies in inhibiting the kinase activities of BRAF, BRAF V600E, and CRAF. The cellular activities of LSN3074753 as demonstrated by phospho-ERK inhibition in BRAF V600E–mutant A375 and KRAS G13D–mutant HCT116 cells are 22 and 160 nmol/L, respectively, which are equivalent to the activities of LY3009120 of 37 and 150 nmol/L in these two cell lines. Both compounds are RAF-selective inhibitors sharing similar off target activities against a few other kinases, such as p38, PDGFRα, PDGFRβ, and Kit. We have previously demonstrated that LY3009120 is able to induce RAF dimer and inhibit the kinase activity of RAF dimers with minimal paradoxical pathway activation in KRAS-mutant HCT116 cells (39). As demonstrated in Fig. 1C in HCT116 cells, LSN3074753 is also able to inhibit the phospho-MEK and ERK activities in a concentration-dependent manner with minimal paradoxical pathway activation. Similar to LY3009120, LSN3074753 induces BRAF and CRAF heterodimer as demonstrated by IP studies (Fig. 1D), and inhibits the kinase activity of the BRAF–CRAF dimers (Fig. 1E). Overall, these results confirm that LSN3074753 is a pan-RAF and RAF dimer inhibitor and share a similar in vitro pharmacology profile to LY3009120.

Seventy-nine colorectal cancer PDX models were established at either Oncotest GmbH or Champions Oncology, Inc. as described (43). In collaboration with Oncotest and Champions, we have comprehensively characterized these colorectal cancer models by exome and RNA sequencing, as well as SNP6.0 analysis. The key genetic mutations and status of EGFR expression of these colorectal cancer models were summarized in Supplementary Table S2. Overall, the mutation rates for the most relevant genes are APC 75.9% (60/79), p53 70.9% (56/79), KRAS 55.7% (44/79), NRAS 5.1% (4/79), BRAF 15.2% (12/79), PIK3CA 26.6% (21/79), PTEN 12.7% (10/79), PIK3R1 6.3% (5/79), and CTNNBI 3.8% (3/79). Note that 32.9% (26/79) models are wild-type for KRAS and BRAF, 13.9% (11/79) models have EGFR amplification based on gene copy numbers (gene copy numbers 3.6 or more), and 13.9% (11/79) models have EGFR overexpression based on RNA sequencing data. Six models showed both EGFR copy-number gain and EGFR overexpression. Western blot analysis of tumor lysates from these PDX tumors revealed almost all colorectal cancer PDX models expressed EGFR, and many of them exhibited EGFR activation exemplified by receptor phosphorylation, whereas HER2 overexpression and activation were rare in these models (Supplementary Fig. S1). For 12 models with BRAF mutations, seven are BRAF V600E and five are non-V600 atypical BRAF mutations. The BRAF V600E mutation is generally mutually exclusive with KRAS and NRAS mutation with the exception of one model CTG0360, in which the BRAF V600E mutation co-occurs with an uncommon KRAS Y157H mutation. For five atypical BRAF-mutant models, they all co-occur with either KRAS (3) or NRAS (2) mutations (Supplementary Table S2).

The mice were individually enrolled for treatment when PDX tumor sizes reached 200 mm³. Mice bearing established tumors were treated with cetuximab at 20 mg/kg twice per week for 4 weeks, LSN3074753 at 40 mg/kg twice daily for 4 weeks, or their combination with a single mouse per treatment group (n = 1). Control group with three mice was administered with vehicle. The tumor growth or regression was recorded as described under the “Materials and Methods.” The population response to different treatments was summarized in Table 1. Based on RECIST criteria, disease control was defined as <20% median tumor growth versus baseline or regression which was indicated by red font in Table 1. Treatment that led to −50% to >−100% tumor size regression was defined as PR, and treatment that led to −100% regression was defined as CR. The mice with PR or CR by treatment were indicated by bold red font. Cetuximab monotherapy was efficacious in 16 of 79 PDX models exemplified by antitumor activities with tumor size <20% median tumor volume change or regression compared with the baseline with a disease control rate (DCR) of 20.3%. Among them, five showed −50% or greater regression with an overall response rate (ORR) of 6.3%. Single-agent LSN3074753 treatment demonstrated antitumor activity in 6 of 79 PDX models with a DCR of 7.6%. Among these models, only one showed greater than −50% regressions with an ORR of 1.3%. The combination treatment was efficacious in 28 of 79 PDX models with a DCR of 35.4%. Seven of these models displayed −50% or greater regression with an ORR of 8.9% (Table 2).

Cetuximab monotherapy demonstrated a DCR of 20.3% and ORR of 6.3% as summarized in Tables 1 and 2. We analyzed the association of single-agent activity of cetuximab with specific genetic alterations and found that cetuximab was primarily active in colorectal cancer PDX models that harbored a wild-type of KRAS and BRAF. As shown in Fig. 2A, nine PDX models that regressed in response to cetuximab treatment (including five models with −50% or greater regression) were enriched for wild-type KRAS and BRAF. Among 11 models that displayed SD, five had KRAS mutation and 1 had a KRAS and a BRAF mutation. All other 11 BRAF-mutant PDX mice have progressive disease in the cetuximab single-agent arm. Figure 2B and C provides examples of single-agent cetuximab activity in two models with wild-type KRAS and BRAF. In CTG0075 model, treatment of cetuximab for 4 weeks exhibited tumor regression, and this efficacy was sustained after drug withdrawal, whereas all three PDX mice in the vehicle arm had progressive disease. Similarly, in CXF2039 model, treatment of cetuximab resulted in CR that was maintained after discontinuation of drug treatment.

To further evaluate single-agent cetuximab activity, we analyzed copy-number variation and expression of EGFR and its ligands based on data from SNP6.0, RNA sequencing, and qPCR analysis. As shown in Fig. 2D, the cetuximab single-agent activity is strongly correlated with EGFR copy numbers with P value of 0.0002 in 26 colorectal cancer models with KRAS and BRAF wild-type. Models that responded to cetuximab have significantly higher EGFR copy numbers. These results were further confirmed by RNA sequencing data in the same 26 models (Fig. 2E). In addition, we also conducted qPCR analysis for expression of EGFR and its ligands in all models evaluated in this study. These data indicate that expression of EGFR as well as low-affinity ligands (amphiregulin or AREG, and epiregulin or EREG) differentiated colorectal cancer models that displayed sensitivity to cetuximab monotherapy from those that did not (Fig. 2F–H).

Single-agent treatment with pan-RAF inhibitor LSN3074753 showed little activity with DCR of 7.6% and ORR of 1.3% (Table 2). Further analysis suggested that the activity appeared to be associated with KRAS or BRAF mutation status. Among five modes that displayed tumor regression in response to LSN3074753 treatment, three are KRAS mutation and one harbors both KRAS and BRAF mutations, and one (CTG0125) is wild-type for both KRAS and BRAF (Fig. 3A). CTG0125 model has a TP53 mutation but no clear oncogenic driver mutation (Supplementary Table S2). Among 7 PDX models with BRAF V600E mutation, LSN3074753 demonstrated modest tumor growth inhibitory activities ranging from 62.3% tumor growth inhibition (based on tumor growth inhibition compared with vehicle controls) to −31.75 tumor regression (Fig. 3A; Table 1). The limited single-agent activity of LSN3074753 in BRAF V600E mutation–positive colorectal cancer models was consistent with clinical study of vemurafenib in patients with colorectal cancer (32). Figure 3B illustrated the best case of LSN3074753 treatment in a BRAF V600E–positive model. LSN3074753 treatment leads to tumor stasis, followed by tumor relapse after discontinuation of treatment. Similarly, in KRAS G12D mutation–positive CTG0079 model, LSN3074753 treatment caused a PR with −56.8% tumor growth regression, and the tumor eventually relapsed once the drug was withdrawn (Fig. 3C).

Cetuximab and LSN3074753 monotherapy arms showed a DCR of 20.3% and 7.6%, respectively. Additive and synergistic effects were observed when cetuximab and LSN3074753 were given in combination to the 79 PDX mice, and the overall DCR in the combination arm was 35.4% (28/79). As shown in Fig. 4A, three mice had CR and four had PR with a total ORR of 8.9%. There were a total of 19 mice with different degrees of tumor volume regression, including models with KRAS or BRAF mutation, as well as KRAS and BRAF wild-type. Further analysis revealed that 26 models treated in the combination arm showed an improved antitumor efficacy, and three models displayed an inferior efficacy compared with the single-agent treatment (Supplementary Fig. S2). Subgroup analysis suggested that BRAF mutations including V600E or other atypical mutations are the best predictor of this combination activity. In 12 models with BRAF mutations, the combination arm has a DCR of 41.7% (5/12), whereas cetuximab or LSN3074753 alone showed similar DCR of 8.3% each (Table 2). The presence of KRAS mutations is also associated with combination efficacy. Among 44 KRAS mutation models, cetuximab or LSN3074753 alone has a DCR of 13.6% or 11.4%, respectively, and the combination arm has a DCR of 34.1%. However, in models with wild-type KRAS and BRAF, cetuximab or LSN3074753 alone has a DCR of 38.5% and 3.8%, respectively, and the combination arm had no benefit over cetuximab alone for this group with a DCR of 34.6%, similar to cetuximab single-agent arm (Table 2).

Figure 4B–E illustrates the synergistic effect of cetuximab and LSN3074753 given in combination in different colorectal cancer PDX models. In CXF1103 model with a KRAS G12D mutation, cetuximab was inactive, LSN3074753 had moderate antitumor growth activity, and the combination of the two led to tumor growth regression (Fig. 4B). In CXF2129 model with a KRAS G12D mutation, neither cetuximab nor LY3074753 was active, and their combination caused complete tumor regression. Despite drug withdrawal, the antitumor effect was maintained (Fig. 4C). In BRAF V600E–mutant CTG0978 model, cetuximab or LSN3074753 single agent showed a moderate antitumor activity during treatment, whereas the combination treatment caused significant tumor regression followed by tumor relapse after discontinuation of therapy (Fig. 4D). In CTG0358 model with atypical BRAF G596V and NRAS mutations, both cetuximab and LSN3074753 showed minimal antitumor activity, and the combination of these two agents caused tumor stasis while the animal was on treatment (Fig. 4E). Overall, these models exemplified that cetuximab and LSN3074753 combination led to synergistic antitumor activity in a number of colorectal cancer PDX models, particularly in those harboring KRAS or BRAF mutations.

To further assess combination synergy, a Kaplan–Meier analysis was performed across PDX models using progression-free survival (PFS) based on time to one tumor doubling. In this analysis, the proportion not doubled and the tumor size doubling time were plotted. As demonstrated in Fig. 5A in all 79 mice, all three treatment groups showed better PFS than vehicle control group with P value < 0.001. Combination of cetuximab and LSN3074753 showed the best PFS and is statistically better than pan-RAF inhibitor LSN3074753 monotherapy (P = 0.001), but not statistically better than cetuximab single agent (P = 0.147). However, in 53 PDX mice with KRAS or BRAF mutation, the combination did show statistically significant benefit over either monotherapy or vehicle-treated group (Fig. 5B). This analysis suggests that the efficacy of cetuximab and LSN3074753 combination may be restricted to colorectal cancer with KRAS or BRAF mutations, whereby representing a potentially viable treatment option for patients with colorectal cancer whose tumors harbor BRAF or KRAS mutations.

To further understand the molecular mechanism of synergy of cetuximab and LSN3074753 combination in KRAS- or BRAF-mutant colorectal cancer, we utilized HT-29 cells with a BRAF V600E mutation for cell signaling analysis. As demonstrated in Fig. 5C, pan-RAF inhibitor LSN3074753 was active against HT-29 cells with IC₅₀ of 35 nmol/L. Addition of EGF in the growth medium of HT-29 cells caused resistance to LSN3074753 with IC₅₀ of 1,297 nmol/L. However, cetuximab was able to restore the sensitivity of HT-29 cells to LSN3074753 in the presence of EGF (Fig. 5C). Cell signaling analysis by Western blot revealed that treatment of HT-29 cells by LSN3074753 for 48 hours appeared to lead to activation of EGFR as indicated by multiple phospho-EGFR forms and partial feedback of phospho-ERK (Fig. 5D). Combination of cetuximab and LSN3074753 was able to reduce the phospho-ERK feedback (Fig. 5E). In addition, we previously developed a pan-RAF inhibitor LY3009120-resistant cell line with KRAS-mutant colorectal cancer HCT-116 cells (45). LY3009120 was active for HCT116 cells with IC₅₀ of 147 nmol/L and inactive for resistant cells with IC₅₀ of greater than 10,000 nmol/L. Molecular analysis revealed that phospho-EGFR, phospho-CRAF, phospho-MEK, phospho-ERK, and phospho-AKT were significantly enhanced in these resistant cells (45). All together, these data suggest that EGFR activation is a key resistant mechanism to pan-RAF inhibition in colorectal cancer.

The EGFR-directed antibody therapies, cetuximab and panitumumab, are approved for the treatment of patients with chemorefractory metastatic colorectal cancer. Both drugs demonstrated clinical benefit and achieved similar objective response rates when used as monotherapy for chemorefractory colorectal cancer (13, 14, 46, 47). Investigations into the molecular basis of response to EGFR therapies, particularly retrospective analyses with archived tumor tissues from subsets of patients participating in clinical trials, revealed that resistance to EGFR blockade in metastatic colorectal cancer is related to KRAS, NRAS, or BRAF mutation (16, 17). Therefore, upon progression of chemotherapy, patients with KRAS, NRAS, or BRAF mutation–positive colorectal cancer have very limited treatment options. We are interested in exploring potential combinational treatment options for patients with colorectal cancer, particularly those with KRAS, NRAS, or BRAF mutation. We utilized a 1 PDX mouse trial to mimic intertumor heterogeneity encountered in clinical trials. Our study results revealed that combination of cetuximab and pan-RAF inhibitor LSN3074573 showed a synergistic efficacy in CXC PDX models with KRAS or BRAF mutations, suggesting that combined EGFR and pan-RAF inhibition could potentially provide clinical benefit to this subset of patients with coloretal cancer.

It is believed that PDX models better recapitulate the biology of human cancer and more accurately predict response to experimental treatments (48). These models are superior to traditional tumor cell line xenograft models because they maintain more similarities to the parental tumors. Detailed examination of PDX mice tumors indicates that histology and gene expression profiles are retained, along with single-nucleotide polymorphism and copy-number variants (48). In this study, we have molecularly characterized 79 colorectal cancer PDX models by whole-exome and RNA sequencing, as well as SNP6.0 analysis. The common genetic mutations, such as TP53, APC, KRAS, BRAF, and PIP3CA, and their frequencies are consistent with the data reported for human tumors (5, 7, 8). More importantly, in this PDX trial, the results appear to recapitulate the clinical observations. In the cetuximab treatment arm, the ORR is 6.3% and all models displaying tumor shrinkage had wild-type KRAS and BRAF. In addition, there is also an association between expression of EGFR and low-affinity EGFR ligands (amphiregulin and epiregulin) and cetuximab response. These results are similar to findings observed in the clinic (13, 14, 46). In an early phase II study for patients with colorectal cancer refractory to chemotherapy, five of 57 patients achieved PR with an ORR of 9% in the cetuximab treatment arm (46). Based on the response rate and the responders' genetic background in the cetuximab arm, we can conclude that the one colorectal cancer PDX mouse trial is able to reliably recapitulate the clinical results. Therefore, the one PDX mouse trial may be reliably predictive of clinical response in colorectal cancer.

In this study, we found that 39 of 79 (49.4%) models have a KRAS mutation at codon 12 or 13, and none of these 39 models responded to cetuximab. This is consistent with clinical findings that KRAS mutations at codon 12 or 13 confer resistance to EGFR antibody therapies. In addition to codon 12 or 13 mutations, there are 4 models with KRAS mutations in other codons (codons 61, 146, and 157), four models with NRAS mutation, and 12 models with BRAF mutations including BRAF V600E and other atypical BRAF mutations. Interestingly, none of these models showed a response to cetuximab either, suggesting that MAPK pathway activation through KRAS, NRAS, or BRAF mutation is a common resistance mechanism to EGFR therapy. These results are consistent with recent findings described elsewhere. In addition to KRAS mutations at codons 12 and 13, MAPK activation through other genetic alterations was demonstrated to be associated with resistance to anti-EGFR therapies (19, 25–31). These include KRAS mutations in other codons (26–28), KRAS amplification (29), and NRAS and BRAF mutations (19, 27, 30, 31). Collectively, our results and that of others have established that MAPK activation is a key determinant of treatment stratification in patients with metastatic colon cancer, as well as a major biomarker of resistance to anti-EGFR antibody therapy. Therefore, for patients with metastatic colon cancer with MAPK activation through a KRAS, NRAS, or BRAF mutation, a combinational treatment with anti-EGFR therapy and MAPK pathway inhibitor might be a viable option.

Indeed, in this PDX mouse trial, we have found that the combination of cetuximab and pan-RAF inhibitor LSN3074753 showed additive and synergistic effects for colorectal cancer PDX models, particularly those with KRAS or BRAF mutation. In 43 models with KRAS mutation and 12 models with BRAF mutation, a significant improvement of overall DCR was observed in the combination arm when compared with single-agent treatment of cetuximab or LSN3075753. Because KRAS and BRAF are downstream effectors of EGFR, activating mutations of KRAS or BRAF rendering resistance to EGFR antibody therapies are mechanistically understandable. Both KRAS and BRAF mutations activate MAPK signaling, and we have demonstrated that pan-RAF inhibitors including LY3009120 and LSN3074753 are able to effectively inhibit MAPK activity and cell proliferation of tumor cells with KRAS or BRAF mutation (38, 39). In this study, pan-RAF inhibitor LSN3074753 exhibited a moderate efficacy with DCR of 11.4% or 8.3% for KRAS or BRAF mutant models, respectively. Combination treatment of cetuximab and LSN3074753 significantly improved DCR in these models. Our molecular characterization with RNA sequencing, SNP6.0, and Western blot analysis suggests that EGFR expression and activation are common among colorectal cancer PDX models including those with KRAS or BRAF mutation. Therefore, EGFR and MAPK pathway activation represents two molecular hallmarks of colorectal cancer biology, consistent with previously reported analysis with patient samples (5, 7, 8). In addition, as demonstrated in this study, treatment of BRAF V600E–mutant HT-29 cells with pan-RAF inhibitor appeared to cause EGFR activation and phospho-ERK feedback. In our previous study, we have developed a pan-RAF inhibitor–resistant HCT-116 (KRAS mutation) cell line. This resistant cell line showed significant increase of phospho-EGFR, phospho-ERK, and phospho-AKT (45). These results are consistent with the phosphoproteomics results reported from a recent study (49). In this study, phosphoproteomics revealed that MAPK inhibitors, including LY3009120, enhanced EGFR- and MET-driven ERK and AKT signaling in KRAS-mutant lung cancer cells (49). In our study in colorectal cancer HCT-116 cells, MET activation was not significant (45). Altogether, these results explain why combination treatment of EGFR and pan-Raf inhibitor displays synergy in colorectal cancer with KRAS or BRAF mutation.

EGFR activation also appears to be a key mechanism to cause resistance of BRAF-mutant colorectal cancer to BRAF-selective inhibitors. For example, vemurafenib was demonstrated to be ineffective for treatment of BRAF V600E–mutant colorectal cancer (32), although vemurafenib showed significant clinical benefit for melanoma patients with the same BRAF mutation (34). Recent studies revealed that the refractoriness of BRAF-mutant colorectal cancer to BRAF inhibitors was due in part to feedback activation of EGFR that leads to MAPK reactivation (35–37). In melanomas, tumor cells express low or undetectable levels of EGFR and are therefore not subject to EGFR-mediated feedback. However, colon tumor cells generally express EGFR, and BRAF inhibition by a selective BRAF inhibitor causes a rapid feedback activation of EGFR, which supports continued proliferation in the presence of a BRAF inhibitor in colorectal cancer. Combination of BRAF and EGFR inhibition blocked MAPK reactivation and markedly improved efficacy in vitro and in vivo (35, 37). These findings support the evaluation of combined BRAF and EGFR inhibition in patients with BRAF-mutant colorectal cancer.

In conclusion, our molecular characterization of 79 colorectal cancer PDX models revealed that EGFR activation through EGFR overexpression and MAPK activation via KRAS, NRAS, or BRAF mutations are two major molecular hallmarks of colorectal cancer. The single-mouse colorectal cancer PDX trial was able to recapitulate clinical results of cetuximab. Combined inhibition of EGFR and MAPK pathway demonstrated significant synergy in colorectal cancer PDX models with KRAS or BRAF mutation. Therefore, cetuximab in combination with an MAPK inhibitor, such as pan-RAF inhibitor LY3009120, is a viable treatment option that can be explored in a clinical trial setting in patients with colorectal cancer with KRAS or BRAF mutation.

G.P. Donoho and A. Aggarwal hold ownership interest (including patents) in Eli Lilly & Company. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y.-m.M. Yao, G.P. Donoho, P.W. Iversen, A. Aggarwal, J.R. Henry, G.D. Plowman, S.-B. Peng

Development of methodology: Y.-m.M. Yao, G.P. Donoho, Y. Zhang, A. Forest, A. Aggarwal, S.-B. Peng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Zhang, R.D. Van Horn, A. Forest, R.D. Novosiadly

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.-m.M. Yao, G.P. Donoho, P.W. Iversen, Y. Zhang, R.D. Van Horn, Y.W. Webster, P. Ebert, S. Bray, J.C. Ting, A. Aggarwal, R.V. Tiu, G.D. Plowman, S.-B. Peng

Writing, review, and/or revision of the manuscript: Y.-m.M. Yao, G.P. Donoho, P.W. Iversen, R.D. Van Horn, A. Forest, R.D. Novosiadly, P. Ebert, S. Bray, A. Aggarwal, J.R. Henry, R.V. Tiu, G.D. Plowman, S.-B. Peng

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.D. Van Horn, A. Forest, S. Bray

Study supervision: Y.-m.M. Yao, G.P. Donoho, S.-B. Peng

This work was supported by Eli Lilly and Company.

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.