Colon cancer-derived oncogenic EGFR G724S mutant identified by whole genome sequence analysis is dependent on asymmetric dimerization and sensitive to cetuximab

Activation of the epidermal growth factor receptor (EGFR) oncoprotein, a member of the ErbB family of receptor tyrosine kinases, is among the most common oncogenic driving events in human cancer[1]. Genomic mechanisms for activating the EGFR gene include nucleotide substitutions and in-frame insertions/deletions of the kinase domain in lung adenocarcinoma and papillary thyroid carcinomas, and multi-exonic deletions (exons 2 through 7: EGFR variant III or vIII), nucleotide substitutions of the extracellular domain and carboxyl terminal deletions in glioblastoma[2‐6]. EGFR is also activated by high-copy amplifications in many epithelial cancer types, prominently in lung and upper gastrointestinal carcinomas as well as glioblastoma and head and neck cancer[7‐10]. Furthermore, EGFR protein is over-expressed in many cancers even without evidence of focused genomic alteration, as observed in many cases of colorectal carcinoma where EGFR kinase domain mutations were found in only 3 out of 224 cases, 1.3% subjected to whole exome sequencing[11, 12]. Given the elevated expression and genomic alterations present in EGFR, multiple cancer therapies have targeted EGFR, as both its kinase activity and its dependence on extracellular ligand signaling have rendered EGFR vulnerable to therapeutic intervention. FDA-approved EGFR targeted inhibitors include the low-molecular-weight ATP-competitive kinase inhibitors, such as gefitinib and erlotinib, and humanized monoclonal antibodies directed against the extracellular domain, notably cetuximab and panitumumab[13]. Although high-level expression of EGFR ligands and/or increased EGFR gene copy numbers may be predictive markers for antitumor response by cetuximab in colon cancer[14‐16], and patients with RAS driven cancers are known not to benefit from cetuximab treatment, a clear molecular explanation of cancer response to cetuximab has remained elusive.

Colorectal adenocarcinoma has been a classic model to study the progressive accumulation of genomic lesions leading to cellular transformation. Key genomic features of these tumors involve inactivation of tumor suppressors such as APC, TP53 and SMAD4 and mutational activation of oncogenes including KRAS, NRAS, BRAF and PIK3CA[17]. Given the role of cetuximab in therapy of these cancers, initial efforts to identify activating EGFR mutations identified few such events, though potentially activating events such as G719S were seen[18]. More recent reports have also identified potentially mutations of ERBB3[19] and amplifications and mutations of ERBB2 in CRC[12, 20]. We have previously reported whole genome sequence analysis of nine colorectal carcinoma/normal pairs,leading to the identification of activating translocations of TCF7L2 and of the association of Fusobacterium nucleatum with colorectal carcinomas[21, 22]. Here, we report genomic analysis of a tenth anonymized case of colorectal carcinoma. Whole genome sequencing was performed on the genomic DNA from colorectal carcinoma tissue and adjacent non-neoplastic colonic tissue to a median coverage of 32.5x and 34.2x coverage, respectively, with 86.8% of the genome sequenced to adequate depth for mutation calling.

An analysis of somatic genome structural alterations by comparison of tumor-derived and non-neoplastic derived sequences identified 63 somatic structural rearrangements, including a deletion of the APC tumor suppressor gene (Figure 1A, Additional file1: Figure S1A, and Additional file2: Table S1). Comparison of nucleotide sequences between the colorectal tumor and normal colon identified an overall mutation rate of 6.7 mutations/Mb including 18,401 somatic nucleotide substitutions, and 983 somatic insertions and deletions of < 37 bases (Figure 1B and Additional file3: Table S2). As observed in other colorectal cancers[21, 23, 24], mutation analysis identified a marked elevation in the rate of C to T transitions at CpG dinucleotides (82/Mb). Analysis of non-synonymous coding mutations revealed a total of 119 alterations in 116 genes (Additional file4: Table S3). Prominent mutations included a somatic R175H substitution in the TP53 tumor suppressor gene and a somatic G724S substitution in the EGFR oncogene (Figure 1B and Additional file1: Figure S1B). Somatic mutations of common colorectal adenocarcinoma oncogenes KRAS, BRAF, NRAS and PIK3CA[23] were not detected.

The absence of both KRAS and BRAF mutations are common features seen in colorectal cancers that are responsive to cetuximab[25, 26], thus making the EGFR mutation in this case of particular interest. The somatic G724S mutation in EGFR occurs at the final glycine of the GxGxxG nucleotide-binding motif that is essential for ATP binding and is conserved among all protein kinases (Figure 1C)[27, 28]. Substitution of EGFR G719, the first residue of this motif, to serine, cysteine, or alanine, has been observed in lung adenocarcinomas (~1%), and one G719S mutant and four G724S mutants have been reported in colorectal carcinomas that were sequenced for EGFR (Figure 1C)[18, 29] (COSMIC database). In addition, these EGFR mutations were found to be mutually exclusive with well known KRAS, BRAF and PIK3CA oncogenic driver mutations, demonstrating their potential role in tumorigenesis (COSMIC database).

To determine whether the G724S mutant is oncogenic and to evaluate its pharmacologic sensitivity, we generated this mutant in vitro and retrovirally transduced it into NIH-3T3 cells. While wild-type EGFR expressing NIH-3T3 cells form colonies in soft agar only in the presence of ligand, NIH-3T3 cells that express EGFR G724S form colonies in the absence of exogenous ligand, as do the lung and colon cancer-derived G719S mutants (Additional file1: Figure S2A)[30]. Furthermore, both G719S and G724S mutants undergo constitutive tyrosine-phosphorylation, which is further increased by EGF treatment, whereas phosphorylation of wild-type EGFR requires induction by EGF (Additional file1: Figure S2B). These data demonstrate that colon-cancer derived G719S and G724S mutants are oncogenically active in the absence of ligand stimulation.

To test whether G719S and G724S EGFR mutants are cetuximab-sensitive in vitro, we ectopically expressed these mutants in Ba/F3 cells, rendering these cells IL-3 independent but dependent on exogenous oncogenic EGFR signaling[31]. Ba/F3 cells expressing the colon-cancer derived G724S mutant and the lung/colon cancer-derived G719S mutant showed sensitivity to cetuximab with an IC₅₀ value of ~ 0.3 μg/ml (Figure 2A). In contrast to their high cetuximab sensitivity, Ba/F3 cells dependent on EGFR G719S and G724S mutants were only moderately sensitive to erlotinib with an IC₅₀ value of ~ 0.3 μM (Additional file1: Figure S3A), consistent with previous reports on G719X mutants in vitro and in lung cancer clinical trials[30‐33] and with the failure of a previously identified patient with EGFR G724S mutant colorectal cancer to respond to gefitinib[29].

In order to further examine the efficacy of cetuximab, we expanded our studies by generating xenograft mouse models with either SW48 or HCT8 colon cancer cells, which harbor either EGFR G719S or KRAS G13D mutation, respectively. Here, we found that consistent with in vitro response (Additional file1: Figure S3B), cetuximab treatment dramatically suppressed tumor formation driven by SW48 cells (Figure 2B), suggesting the anti-tumor effect of cetuximab against tumors harboring EGFR G719S mutant. In contrast, cetuximab treatment was ineffective for the tumors driven by HCT8, which is consistent with the previous findings that KRAS-mutant tumors are insensitive to cetuximab (Figure 2C). Taken together, we found that G719S and G724S mutants are oncogenic in the absence of ligand stimulation and effectively respond to cetuximab in vivo and in vitro.

Recently, we reported that a subset of lung cancer-derived oncogenic EGFR mutants such as L858R require asymmetric dimerization for biochemical activation and oncogenic transforming activity, meaning that their oncogenic ability depend on formation of an EGFR homodimer in which two distinct regions of the two molecule dimerize[34]. By contrast, we showed that other EGFR mutants are oncogenic without the requirement for dimerization, including the gefitinib-resistant exon 20 insertion mutant and the T790M mutant[34]. Furthermore, we found that dimerization-dependent L858R mutant shows a dramatic response to cetuximab, whereas tumors driven by dimerization-independent mutants such as T790M are resistant to the antibody, suggesting that there is a close correlation between dimerization dependency of lung cancer-derived oncogenic mutant EGFR and pharmacological effects of cetuximab[34]. Given that colon cancer-derived G719S and G724S mutants are sensitive to cetuximab, we sought to examine whether oncogenic potential of these mutants are dependent on the asymmetric dimerization like lung cancer-derived L858R mutant.

To test the hypothesis that cetuximab sensitivity of G719S and G724S mutants is a function of their dimerization dependence, we generated G719S and G724S mutants with compound substitution mutations at the dimerization interface in the N-lobe or C-lobe that disrupt the asymmetric dimerization of EGFR[35]. Specifically, we generated epitope-tagged EGFR expression constructs that combined a receiver-impairing mutation (L704N) and/or an activator-impairing mutation (I941R) with oncogenic G719S and G724S mutants. The single or compound EGFR mutants were expressed in NIH-3T3 cells by retroviral transduction, and the EGFR mutant-expressing cells were assayed for their ability to grow in soft agar. In this system, the transforming ability of dimerization-dependent mutants is predicted to be abolished by cis mutation of the L704 or I941 mutations. Furthermore, co-expression of the L704N and I941R mutant forms, in contrast, is predicted to restore transforming ability that is dimerization-dependent, because the two mutant forms can heterodimerize. Therefore, this experiment allows us to test whether specific EGFR kinase domain mutants can induce cellular transformation in a dimerization dependent or independent fashion.

Dimerization-impairing cis mutations in EGFR, L704N and I941R, significantly reduced the ability of the cetuximab-sensitive G719S and G724S mutants to promote colony formation upon retroviral transduction (Figure 3A and B, L704N or I941R); colony-forming activity was partially or completely restored by co-expression of the L704N and I941R mutants in trans (Figure 3A and B, L704N&I941R). Consistent with these results, we found that constitutive and EGF-inducible autophosphorylation of the G719S and G724S mutants (Figures 3C and D, lanes 1 and 2), is attenuated by the introduction of L704N and I941R dimerization-impairing mutations (Figures 3C and D, lanes 3, 4, 5, and 6). Furthermore, receptor phosphorylation is partially rescued by co-expression of either G719S/L704N or G719S/I941R mutants or the cognate combination in a G724S mutant background (Figure 3C and D, lanes 7 and 8). Taken together, these data suggest that colon cancer-derived cetuximab-sensitive G719S and G724S mutants acquire their oncogenic potentials following asymmetric dimerization, respectively, which are similar to lung-cancer derived L858R mutant[35]. Furthermore, these results are consistent with our model that EGFR variants that are dependent on dimerization can be inhibited by cetuximab.

In summary, our findings suggest that EGFR mutation may underlie at least some cases of cetuximab responsiveness in colorectal carcinoma. While EGFR mutation has historically been believed to be rare in colorectal carcinomas, whole exome sequencing published through The Cancer Genome Atlas identified somatic non-synonymous coding EGFR mutations in 10 of 224 colorectal carcinoma cases (or 4.5%)[12]. While many such mutations may be passenger alterations that do not activate EGFR signaling, these results do speak to the potential for mutational activation of EGFR to result in susceptibility to anti-EGFR antibodies in a small fraction of CRC cases. Indeed, the recent demonstration of secondary somatic mutation in the EGFR extracellular domain conferring acquired resistance to cetuximab[36] is consistent with attribution of responses to cetuximab to EGFR blockade. Anti-EGFR therapy in metastatic colorectal cancer has been reserved for second-line therapy after failure of initial empiric chemotherapy but is now increasingly also used as part of first-line therapy for RAS wild-type patients. As genomic diagnostics enter into routine clinical practice, patients whose CRCs harbor potentially oncogenic EGFR mutations will be identified, including those at codons 719 and 724. These results suggest that in such patients, therapeutic approaches utilizing EGFR-directed antibody as part of their initial therapy should be evaluated given the greater potential dependence on EGFR signaling in these patients.