Research ArticleCancer

MET signaling in keratinocytes activates EGFR and initiates squamous carcinogenesis

See allHide authors and affiliations

Sci. Signal.  21 Jun 2016:
Vol. 9, Issue 433, pp. ra62
DOI: 10.1126/scisignal.aaf5106

MET and RAS converge on EGFR in skin cancer

Whereas a subset of cutaneous melanomas harbor oncogenic mutations in N-RAS that lead to out-of-control signaling through this proliferative pathway, RAS mutations are infrequent in cutaneous squamous cell carcinomas (SCC). Cataisson et al. found that cutaneous SCC was associated with high amounts of hepatocyte growth factor (HGF), which activates the receptor MET. In transgenic mouse models, HGF and oncogenic RAS each activated a common downstream transcriptional program in skin cells, resulting in stimulation of the epidermal growth factor receptor (EGFR) pathway through the production and secretion of its ligands. Furthermore, established squamous tumors in mice regressed when this activation of or signaling by this EGFR pathway was blocked. In some human SCCs, high HGF and activated MET correlated with an EGFR-activating gene signature like the one identified in the mice, indicating that this pathway may offer therapeutic targets for patients.


The receptor tyrosine kinase MET is abundant in many human squamous cell carcinomas (SCCs), but its functional significance in tumorigenesis is not clear. We found that the incidence of carcinogen-induced skin squamous tumors was substantially increased in transgenic MT-HGF (mouse metallothionein–hepatocyte growth factor) mice, which have increased abundance of the MET ligand HGF. Squamous tumors also erupted spontaneously on the skin of MT-HGF mice that were promoted by wounding or the application of 12-O-tetradecanoylphorbol 13-acetate, an activator of protein kinase C. Carcinogen-initiated tumors had Ras mutations, but spontaneous tumors did not. Cultured keratinocytes from MT-HGF mice and oncogenic RAS-transduced keratinocytes shared phenotypic and biochemical features of initiation that were dependent on autocrine activation of epidermal growth factor receptor (EGFR) through increased synthesis and release of EGFR ligands, which was mediated by the kinase SRC, the pseudoproteases iRhom1 and iRhom2, and the metallopeptidase ADAM17. Pharmacological inhibition of EGFR caused the regression of MT-HGF squamous tumors that developed spontaneously in orthografts of MT-HGF keratinocytes combined with dermal fibroblasts and implanted onto syngeneic mice. The global gene expression profile in MET-transformed keratinocytes was highly concordant with that in RAS-transformed keratinocytes, and a core RAS/MET coexpression network was activated in precancerous and cancerous human skin lesions. Tissue arrays revealed that many human skin SCCs have abundant HGF at both the transcript and protein levels. Thus, through the activation of EGFR, MET activation parallels a RAS pathway to contribute to human and mouse cutaneous cancers.


Studies using the multistage induction of squamous cancers on mouse skin have revealed much about the biology of tumor formation and first defined the operational and functional distinctions of initiation, promotion, premalignant progression, and malignant conversion. Furthermore, the standard protocol of 7,12-dimethylbenz[a]anthracene (DMBA) followed by 12-O-tetradecanoylphorbol 13-acetate (TPA) application confirmed the primacy of Hras mutations as an initiating event for squamous tumors in vivo and displayed the importance of regenerative hyperplasia and inflammation as the selective forces in tumor promotion, leading to the emergence of HRAS-initiated tumors. Through genetic modification of mice, the identification of potential initiating events for skin tumors has expanded to include Kras and Nras as well as known RAS targets in the epidermal growth factor receptor (EGFR) pathway, such as EGFR, ErbB2, or SOS, and more downstream factors such as v-FOS, c-MYC, IGF1, and components of the nuclear factor κB (NF-κB) pathway. In most of these cases, a proinflammatory tumor promoter is also required for tumor formation (1). This requirement stimulated research to elucidate how promoter-induced inflammation provides the selection stimulus for tumor outgrowth and led to a broader understanding of the role of inflammatory cells, chemokines, and molecular pathways in carcinogenesis in that they have both protective and negative consequences (24).

It is now recognized that initiating events themselves can have consequences on the inflammatory milieu, and this can be essential to their oncogenic potential. For example, transduction of keratinocytes with oncogenic Hras activates EGFR signaling, leading to release of interleukin-1α (IL-1α), activation of NF-κB, and elaboration of CXC motif chemokine receptor 2 (CXCR2) ligands that are essential components for HRAS-mediated keratinocyte neoplastic transformation (5, 6). The magnitude of induction of these intermediate pathways is greatly enhanced by activation of protein kinase Cα (PKCα), yet the overexpression and activation of PKCα in mouse skin are not sufficient to initiate tumors in the absence of Ras mutations (5, 7, 8). Nevertheless, transgenic mice that overexpress PKCα in the epidermis (K5-PKCα mice) are exquisitely sensitive to tumor promotion after DMBA initiation (5). This exquisite sensitivity to tumor promotion provides a model that could help identify initiating events of less potency than Ras mutations but of great relevance to human cancer. Signaling by hepatocyte growth factor (HGF) through its receptor tyrosine kinase MET has been studied in multiple epithelial carcinomas (9, 10). Several studies suggest that MET protein is abundant in a subset of human skin cancers (11, 12) and various other epithelial cancers (10, 13). In human cancers where MET appears to contribute, the abundance of MET protein is most frequently increased as a result of transcriptional up-regulation or gene amplification. Constitutively active MET mutations have been identified in hereditary papillary renal cell carcinoma patients (14, 15), but in general, activating MET mutations are infrequent (10). The downstream targets of activated MET include the adapter protein GAB1, the phospholipase PLC-γ, the guanosine triphosphatase RAS, and the kinases PI3K, RAF, ERK, and MAPK, leading to mitogenic, motogenic, and morphogenic responses in many cell types (9). Aside from intrinsic changes in MET, mutations in the regulatory region of the gene encoding HGF that induce its overexpression and activation of MET contribute to both breast and bladder cancer (16, 17). Copy number amplification of HGF has been identified in head and neck squamous cell carcinoma (HNSCC) (14), and HGF is described as a cancer driver gene across squamous cancers independent of tissue origin (18). HGF mutations have also been detected in about 20% of human cutaneous SCCs ( (19, 20), and HGF expression increases as a primary response to ultraviolet (UV) light exposure in the skin (21, 22). In most normal tissues, HGF is produced predominantly by stromal cells in the microenvironment and elicits both paracrine and autocrine MET signaling (16, 23). Thus, an experimental model that links HGF and MET to squamous cancer has relevance for various human cancers. In transgenic mice overexpressing HGF driven by the metallothionein (MT) 1 promoter (MT-HGF mice), melanocytes relocate to the epidermal-dermal junction, closely approximating the cellular distribution of human skin (24, 25). A single dose of UV irradiation or DMBA followed by TPA in MT-HGF neonates induces melanomas and SCCs through the activation of MET (26, 27). Here, we used the MT-HGF mouse in combination with the tumor promotion–sensitive K5-PKCα mouse to explore the stage-specific contribution of increased MET activity to the development of squamous cell tumors in the skin. In this setting, we found that HGF-activated MET was a fully functional tumor initiator for skin tumor formation, and the downstream signaling from MET activation co-opted many of the properties of oncogenic RAS, signaling in keratinocytes indicating that together HGF and MET comprise a relevant oncogenic pair for the propagation of human skin cancers.


Double-transgenic MT-HGF/K5-PKCα mice develop greater numbers of squamous tumors than do wild-type, K5-PKCα, and MT-HGF littermates

To generate the age-matched experimental and control genotypes to conduct this study, we crossed FVB/N K5-PKCα female mice with FVB/N MT-HGF transgenic male mice to breed MT-HGF-K5-PKCα mice, hereinafter referred to as double-transgenic (DT) mice. In vitro fertilization of wild-type FVB/N females with sperm from DT males enabled the generation of the four experimental genotypes of the same age. Wild-type, K5-PKCα, MT-HGF, and DT littermates were subjected to a DMBA-TPA protocol that induces melanomas and squamous tumors in MT-HGF mice (27). In this protocol, DMBA was applied topically once to 4-day-old pups, and then TPA was applied twice weekly for 5 weeks starting at 6 weeks of age, when hair follicles are in a resting state called telogen, and tumor growth was assessed up to 20 weeks of age (Fig. 1A). A maximum squamous tumor incidence of 40 to 60% was evident at 10 weeks of age in the DT and MT-HGF mice, in contrast to just 10% incidence in K5-PKCα at that same time point (Fig. 1B). By week 20, tumor-bearing DT mice averaged six tumors per mouse, whereas K5-PKCα and MT-HGF mice averaged just one tumor per mouse (Fig. 1C). Wild-type mice did not develop tumors in this protocol, as expected from the suboptimal (5 week; low dose) TPA regimen (28). The malignant conversion rate (that is, the % squamous carcinoma among all squamous lesions) was higher in MT-HGF than DT mice, at 73 and 23%, respectively (fig. S1), suggesting that many of the excess lesions in the DT group were low risk (29). All tumors in the DT group and 70% of those in the K5-PKCα group contained the expected Hras codon 61 A→T transversion mutation characteristic of the DMBA initiation–TPA promotion protocol. In contrast, 6 of 18 tumors in the MT-HGF group had Kras mutations, suggesting a distinct selection of initiated cells in the presence of excess HGF and absence of PKCα to facilitate promotion (fig. S2A). Wild-type, K5-PKCα, MT-HGF, and DT littermates were also compared for squamous tumor formation in the absence of DMBA-driven initiation by application of TPA (1 μg or 1.6 nmol) twice weekly for up to 10 weeks starting at 6 weeks of age (Fig. 1D). At the termination of the experiment (at 16 weeks old), 100% of DT mice had developed multiple squamous papillomas (2.7 average number of tumors per mouse), whereas none of their wild-type, K5-PKCα, or MT-HGF littermates developed squamous lesions (Fig. 1, E and F). We further screened for Kras, Hras, and Nras mutations at codons 12, 13, and 61 in the squamous lesions observed in DT animals without DMBA initiation and found none (fig. S2B). This result suggests that activated MET is sufficient to drive skin carcinogenesis in a tumor promotion–sensitive environment. To further confirm whether the initiated phenotype is conferred by activated MET on keratinocytes, we performed orthotopic grafting of MT-HGF or wild-type keratinocytes and primary dermal fibroblasts on syngeneic MT-HGF or wild-type recipients without further treatment (6, 30). Four of five MT-HGF mice developed squamous papillomas in the MT-HGF keratinocyte grafts (fig. S3, A and B), whereas none were observed in the wild-type grafts on wild-type mice. Mutations in Ras alleles were not detected in papillomas from orthografts. When wild-type keratinocytes were grafted to MT-HGF recipients, 40% of the mice formed tumors, suggesting that paracrine HGF and a wound environment are sufficient stimulus to initiate and promote tumor formation (fig. S3A). Collectively, these results indicated that aberrant HGF signaling and MET activation can substitute for Ras mutation to initiate skin carcinogenesis.

Fig. 1 Responsiveness of wild-type, K5-PKCα, MT-HGF, and DT mice to chemically induced skin carcinogenesis.

(A) Breeding scheme and DMBA-TPA protocol timeline, which initiated with topical application of DMBA (20 μg per 0.1 ml of acetone) at day 4 after birth and promoted with topical application TPA (1 μg or 1.6 nmol per 0.2 ml of acetone) twice a week from 6 to 11 weeks of age. Tumors were counted every week. (B and C) Percentage of mice with tumors (B) and the mean number of skin tumors per tumor-bearing animal (C). Wild-type (WT), n = 7; K5-PKCα, n = 18; MT-HGF, n = 14; and DT, n = 9 mice. In (B), *P < 0.05 and **P < 0.001 compared with K5-PKCα were analyzed by a two-sample t test. Data in (C) are means ± SEM; *P < 0.05 DT versus MT-HGF or K5-PKCα was analyzed by Mann-Whitney U test. (D) Promotion timeline in the absence of DMBA initiation. TPA (1 μg or 1.6 nmol per 0.2 ml of acetone) was applied to mice twice a week from 6 to 16 weeks of age. (E and F) Tumor incidence in mice treated as in (D) at 16 weeks of age (E), with a photograph of representative DT mice (F). WT, n = 6; K5-PKCα, n = 16; MT-HGF, n = 6; and DT, n = 11.

HGF/MET does not sensitize mice to TPA responses

The foregoing results suggested that MET activation was sufficient to initiate skin tumor formation but have not addressed a possible effect on tumor promotion. When TPA was applied to the skin of mice with any of the four genotypes, only the skin of those expressing the K5-PKCα allele demonstrated enhanced promotion responses, notably hyperplasia and inflammation (as assessed by the induction of various markers), whereas the skin of MT-HGF and wild-type mice responded identically (fig. S4). Thus, the TPA-mediated promotion component responsible for the high tumor burden in DT mice was not enhanced by HGF-activated MET signaling but was primarily a response to PKCα activation.

MT-HGF and DT primary keratinocytes exhibit a RAS-like phenotype in vitro

To examine cell-autonomous effects of HGF-MET signaling in the skin tumor model, isolated keratinocytes from the various genotypes were studied in culture and their phenotypes were compared to those produced by oncogenic RAS transduction (5, 6), hereinafter called RAS keratinocytes. RAS keratinocytes with an otherwise wild-type background exhibited refractile and elongated spindle-like morphology and proliferated to a higher density at confluence than did nontransduced keratinocytes (Fig. 2A). DT or MT-HGF but not K5-PKCα keratinocytes had similar morphology to RAS-transduced cells with wild-type or genotype-matched (DT or MT-HGF) backgrounds (Fig. 2A). Treatment of MT-HGF keratinocytes with a MET inhibitor (PHA665752) reversed this phenotype in vitro (fig. S5A). These observations suggested that MT-HGF and DT keratinocytes were cell-autonomously activated in vitro to resemble an initiated phenotype. MET was constitutively activated in cultured keratinocytes from MT-HGF or DT mice, and this activation was not further increased by oncogenic RAS transduction (Fig. 2B). As expected, RAS transduction activated mitogen-activated protein kinase (MAPK) signaling in keratinocytes from all four genotypes, a change reproduced by constitutive MET signaling in MT-HGF and DT keratinocytes without oncogenic RAS (Fig. 2B). Similar results were seen for the activation of EGFR signaling, a main effector of RAS transformation in primary keratinocytes (Fig. 2B) (6, 31). Although the MET inhibitor PHA665752 reduced the activation of EGFR in MT-HGF keratinocytes, the EGFR inhibitor AG1478 did not inhibit the activation of MET in DT keratinocytes, indicating a unidirectional interaction of these two receptor kinases (fig. S5, C and D). Induction of four EGFR ligands by oncogenic RAS was essential for the autocrine activation of EGFR in RAS keratinocytes (31). The expression of transcripts for those ligands was also increased in MT-HGF and DT keratinocytes in the absence of RAS transduction (Fig. 2C), suggesting that autocrine MET signaling could mediate the activation of EGFR through the up-regulation of its cognate ligands.

Fig. 2 MT-HGF and DT keratinocytes exhibit activated MET and EGFR signaling.

Primary keratinocytes from WT, K5-PKCα, MT-HGF, and DT newborn mice were cultured in 0.05 mM Ca2+ medium and transduced with v-rasHa (RAS) for 3 days. (A) Morphology of control and RAS-transduced WT, MT-HGF, and DT keratinocytes. Scale bar, 50 μm. (B) Immunoblotting of total cell extracts from primary control or RAS-transduced keratinocytes. p-, phosphorylated. Blots are representative of six experiments and are quantified in fig. S13A. HSP90, heat shock protein 90. (C) Real-time polymerase chain reaction (PCR) quantification of mRNA encoding Areg, betacellulin (Btc), heparin-binding EGF-like growth factor (Hbegf), and Tgfa in control and RAS keratinocytes 3 days after transduction. Blots are representative of three independent experiments. Data are means ± SEM of three biological replicates. ****P < 0.0001 versus WT, one-way ANOVA with Dunnett’s posttest.

The morphological and biochemical similarities between transduction by RAS and activation of MET in keratinocytes prompted us to examine the connection further. Both RAS transduction and MET activation in the absence of RAS transduction increased keratinocyte proliferation as measured by [3H]thymidine incorporation (Fig. 3A, left). Furthermore, both RAS-transduced MT-HGF and DT keratinocytes resisted the growth inhibition produced by inducing keratinocyte differentiation with increased calcium, a property of more advanced neoplastic keratinocytes (Fig. 3A, right) (32). Biochemically, resistance to differentiation-induced growth inhibition was documented by the persistence of cyclin D1 abundance in differentiating MT-HGF keratinocytes regardless of RAS transduction (Fig. 3B). Additional phenocopies of RAS and MET activity in keratinocytes were the suppression of suprabasal keratin production (K1 and K10) in response to increased calcium (Fig. 3, C and D), de novo induction of keratin 8 (Fig. 3C), and increases in the abundance of components of oncogenic RAS-induced inflammation, which are key parts of the gene signature (Fig. 3D) that is associated with initiation of keratinocyte neoplasia by oncogenic RAS (6). We determined the EGFR dependence of these signature changes associated with MET activation by treating EGFR-null keratinocytes (33, 34) with HGF or by treating MT-HGF keratinocytes with small-molecule inhibitors of EGFR (AG1478 and PD168393) and monitoring signature responses (fig. S6). In keratinocytes genetically depleted of EGFR, the basal levels of transcripts for EGFR ligands, cytokines, and keratins were essentially unchanged, but the induced signature alterations associated with HGF exposure were substantially reduced (fig. S6A). An almost identical pattern in these signature markers was seen for Hras-transduced EGFR-null keratinocytes (fig. S6B). Furthermore, blocking EGFR activity using the small-molecule inhibitors (AG1478 and PD168393) in MT-HGF keratinocytes reduced Cxcl1 and Il1a expression and restored the expression of mRNAs encoding K1 and K10 (figs. S6C and S7), again reproducing studies on RAS keratinocytes (5, 6, 31). No consistent association of these changes with MET directly activating an endogenous Ras allele could be discerned, because knocking down each Ras allele independently with a specific small interfering RNA (siRNA) in MT-HGF keratinocytes did not reverse the Cxcl1, Krt1, and Il1a signature over several days of this study (fig. S8, A and B).

Fig. 3 MT-HGF and DT keratinocytes exhibit a RAS-like phenotype.

(A) Tritiated thymidine incorporation for 24 hours was measured in control and RAS-transduced WT, K5-PKCα, MT-HGF, and DT keratinocyte cultures grown under proliferative (0.05 mM Ca2+; left) or differentiating conditions (0.12 mM Ca2+; right). Right: Percent inhibition of thymidine incorporation in differentiating conditions relative to respective culture maintained under proliferative conditions. Data are means ± SEM of four biological replicates. ****P < 0.0001 versus WT; #P < 0.0001 versus WT RAS, one-way ANOVA with Dunnett’s posttest. (B and C) Immunoblotting in total denatured cell extracts from control and RAS-transduced keratinocytes for (B) cyclin D1 or (C) early markers of differentiation (K1 and K10) and simple epithelial marker (K8). Actin served as a loading control. In (B) and (C), lysates were analyzed from keratinocytes 3 days after RAS transduction in cultures that were switched to 0.12 mM Ca2+ medium for an extra 24 hours. Blots are representative of three experiments and are quantified in fig. S13 (B and C). (D) Real-time PCR analysis of mRNA encoding CXCL1, CXCL2, K1 (Krt1), K10 (Krt10), TNFα, MMP9, GM-CSF (Csf2), SLPI, and IL-1α in control and RAS keratinocytes 3 days after RAS transduction. **P < 0.01, ****P < 0.001 versus WT, one-way ANOVA with Dunnett’s posttest. Data are means ± SEM of three biological replicates.

Activation of ADAM17 by HGF/MET and RAS is central to the initiated phenotype

The central role of EGFR activation in the MET signature indicated that the up-regulation of EGFR ligands must also be associated with release from the cell surface to engage in receptor binding, a process catalyzed by the disintegrin and metalloproteinase ADAM17 (35, 36). Consistent with this expectation, amphiregulin (AREG) was present in culture supernatants from MT-HGF keratinocytes at substantially higher amounts than supernatants from the wild-type keratinocytes, and this was reduced by blocking MET with PHA665752 (fig. S5B). Treatment of MT-HGF keratinocytes with increasing concentrations of the ADAM17 inhibitor GM6001 also decreased free AREG in culture supernatant (Fig. 4A) and reduced EGFR activation (Fig. 4B). Targeting ADAM17 with several distinct siRNAs in MT-HGF keratinocytes also reduced the activation of EGFR (Fig. 4C). Furthermore, deleting ADAM17 with adenoviral Cre infection in ADAM17fl/fl keratinocytes (37) reduced the HGF-induced release of AREG in culture supernatants and EGFR activation (Fig. 4, D and E). Along with the other parallels among active MET and oncogenic RAS in keratinocytes, EGFR activation was also reduced in oncogenic RAS keratinocytes by knocking down ADAM17 with an siRNA (Fig. 4F). Knocking down ADAM17 also effectively reversed the transcriptional signatures that characterize MT-HGF keratinocytes by 72 hours after transduction (Fig. 4G). Consistent with ADAM17 serving to release the EGFR ligands, neutralizing antibodies against transforming growth factor–α (TGFα) and AREG in culture supernatants of MT-HGF keratinocytes also reduced EGFR activation (fig. S9). Two predominant pathways controlling ADAM17 maturation and activation are SRC and iRhom1 and iRhom2 (3841), and treating keratinocytes with HGF activated SRC (Fig. 5A). Using the release of AREG into culture supernatants of HGF-treated keratinocytes as a readout for ADAM17 activity, genetic (siRNA) inhibition of SRC reduced AREG release (Fig. 5B). HGF treatment of wild-type keratinocytes substantially elevated iRhom2 mRNA expression (Fig. 5C), and selective siRNA was used to reduce either iRhom1 or iRhom2 (fig. S10). As seen for reduction of SRC, siRNA knockdown of iRhom1 and iRhom2 in wild-type keratinocytes reduced AREG release in response to HGF (Fig. 5D). The combination of siRNA to both SRC and iRhom essentially eliminated AREG release (Fig. 5D). Thus, it appears that activation of MET sets in motion an SRC-iRhom–mediated pathway to activate ADAM17, release EGFR ligands to stimulate EGFR activity, and contribute to the initiated keratinocyte phenotype.

Fig. 4 MET transactivates EGFR through ADAM17-mediated release of EGFR ligands.

(A) AREG concentration in the supernatant of confluent cultures of WT or MT-HGF keratinocytes treated for 24 hours with ADAM17 inhibitor (GM6001) or vehicle [dimethyl sulfoxide (DMSO); 0]. Data are means ± SEM of three biological replicates. ***P < 0.001 versus MT-HGF DMSO (0 μM), one-way ANOVA with Dunnett’s posttest. (B and C) Immunoblotting of lysates from confluent MT-HGF keratinocyte cultures treated with vehicle or GM6001 (B) or with control (Cont. si) or ADAM17-targeting siRNA [from Qiagen (Q5 and Q6) or Dharmacon (Dhar.)]. Blots are representative of three experiments. (D and E) Supernatant AREG concentration (D) (inset, ADAM17 abundance) and immunoblotting (E) from confluent cultures of keratinocytes from ADAM17fl/fl mice were transduced for 48 hours with control (Cont.) or Cre adenovirus and challenged with HGF for either 6 hours (D) or 10 min (E). Data are means ± SEM of three biological replicates. ***P < 0.001 versus control + HGF, by two-sided Student’s t test. (F) Immunoblotting in lysates from WT or RAS keratinocytes transduced with ADAM17-targeting siRNA for 48 hours. Blots are representative of three experiments. All blots are quantified in fig. S14 (A to E). (G) PCR-assessed expression of signature transcripts in WT and MT-HGF keratinocytes transduced with control or ADAM17 siRNA. Data are means ± SEM of four biological replicates. **P < 0.01, ****P < 0.0001 versus MT-HGF control, two-sided Student’s t test.

Fig. 5 MET activates SRC and iRhom to mediate the release of AREG.

(A) Immunoblotting in total cell extracts from primary keratinocytes treated with HGF as indicated. Blots are representative of three experiments and are quantified in fig. S14F. (B) AREG concentration in the supernatant of primary keratinocytes cultured to confluence in 0.05 mM Ca2+ medium and treated for 6 hours with HGF in the presence of one of four siRNAs targeting SRC. Inset, SRC abundance in lysates quantified in fig. S14G. Data are means ± SEM of three biological replicates. *P < 0.05, ***P < 0.001 versus control + HGF, two-sided Student’s t test. (C and D) Real-time PCR analysis of expression of Rhbdf1 and Rhbdf2 (encoding iRhom1 and iRhom2, respectively) (C) and enzyme-linked immunosorbent assay (ELISA) analysis of the supernatant concentration of AREG (D) in cultures of primary keratinocytes treated as in (B) in the presence of siRNAs targeting iRhom1, iRhom2, SRC, or combinations thereof. The amount of AREG released in response to HGF above that in response to phosphate-buffered saline (PBS) in cells transfected with control siRNA was set as the baseline. Data are means ± SEM of four biological replicates. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 versus control siRNA, two-sided Student’s t test.

The cytokine signature of MT-HGF keratinocytes is IL-1– and NF-κB–dependent, and active EGFR is required for tumor formation

We have previously reported that the RAS keratinocyte signature involves autocrine EGFR signaling and IL-1α–mediated activation of NF-κB (6). Treatment of MT-HGF keratinocyte cultures with the IL-1 receptor antagonist (IL-1ra) or NF-κB signaling blockade by transduction with inhibitor of NF-κB super-repressor (IκBsr) adenovirus reduced the expression of Il1a and Cxcl1 (Fig. 6A). Blockade of IL-1 did not alter the expression of transgenic Hgf [48.9 ± 6.7 arbitrary units (AU) by densitometry and 41.5 ± 8.68 AU in control- and IL-1ra–treated cultures, respectively; P = 0.537]. Collectively, these results suggested that MET signaling in keratinocytes leads to the activation of an EGFR–IL-1–NF-κB axis that contributes to the requirements for a cancer-initiated phenotype previously shown for oncogenic RAS in keratinocytes (6). To test whether EGFR activation was required for MET-driven skin carcinogenesis in vivo, we induced squamous papillomas by grafting MT-HGF keratinocytes to syngeneic hosts. Both EGFR and ERK are activated in tumors from such grafts as detected by increased phosphorylation of EGFR and ERK (Fig. 6B). Once tumors were established, we treated the mice daily with the orally competitive EGFR tyrosine kinase inhibitor gefitinib at a dose that is nontoxic to skin (100 mg/kg) (42). Over 14 days of treatment, gefitinib nearly eliminated established squamous papillomas, whereas tumors on the vehicle control–treated mice continued to grow (Fig. 6, C and D). The regressing lesions displayed a reduced percentage of Ki67-positive basal keratinocytes and reduced microvessel density based on CD31+ cells compared to the vehicle group (fig. S11).

Fig. 6 RAS-like phenotype and tumor development by HGF-overexpressing keratinocytes is mediated by NF-κB and EGFR.

(A) Real-time PCR analysis of the expression of mRNA encoding IL-1α and CXCL1 in WT and MT-HGF keratinocytes 3 days after transduction with the NF-κB dominant-negative (IκBsr) adenovirus (Ad) or treatment with the IL-1ra. Data are means ± SEM of three biological replicates. ****P < 0.001, two-sided Student’s t test. (B) Immunoblotting on lysates from MT-HGF tumors from orthotopic grafts on MT-HGF host (n = 7 mice) and MT-HGF normal skin (n = 4 mice). Blots are quantified in fig. S15A. (C and D) MT-HGF keratinocytes (6 × 106) were mixed respectively with 6 × 106 MT-HGF primary dermal fibroblasts before grafting in the interscapular region of syngeneic hosts. Once squamous papillomas were clearly established, mice were treated daily by oral gavage with vehicle control or gefitinib (100 mg/kg) for 2 weeks. (C) Waterfall plot of tumor response (% change of tumor volume) to 2-week treatment with gefitinib. Each bar represents tumor volume on an individual mouse. (D) Representative photographs of orthotopic squamous papillomas at the start of the treatment (day 0) and at termination (day 14).

MT-HGF– and RAS-transformed keratinocytes share strongly concordant global gene expression profiles

Because both MT-HGF keratinocytes and RAS keratinocytes were able to initiate carcinogenesis in a manner that appeared to involve similar pathways, we considered that a more global view through a microarray-based gene expression analysis might reveal a common signature essential for tumor formation. Using analysis of variance (ANOVA) contrasts, we generated lists of differentially expressed genes in MT-HGF keratinocytes and keratinocytes transduced with oncogenic RAS compared to normal controls all in 0.05 mM Ca2+ proliferation medium (adjusted P < 0.001). Both activated RAS and activated MET induced abundant and substantially overlapping transcriptional responses in keratinocytes (Fig. 7, A and B, and table S1). A highly statistically significant overlap between the activated RAS and activated MET signatures (P < 0.0001) comprised 6247 genes, where 93% of the mRNAs were concordantly up- or down-regulated relative to wild-type keratinocytes. Correlation analysis of fold changes in the shared signature (Fig. 7B) confirmed that the gene regulation was highly concordant between the two (Pearson’s correlation coefficient, r = 0.88; linear coefficient of determination, R2 = 0.77). For further in-depth analyses, we selected from the concordant signature the top 372 genes with at least twofold changes in both experiments (287 up-regulated and 85 down-regulated mRNAs) (Fig. 7C). Notably, most of these genes (88%) were significantly affected by the EGFR inhibitor and negatively correlated (r = −0.63) with the changes in gene expression between MT-HGF and wild-type control (fig. S7 and table S1).

Fig. 7 A model RAS/MET signature can be identified.

(A) Venn diagram depicting the number of differentially expressed genes in the MT-HGF and WT-RAS experiments [false discovery rate (FDR) <1%]. (B) Scatterplot of the genes from the MT-HGF and WT-RAS overlapping signature (6247 genes); linear regression between the two experiments is shown by the solid black line (R2 = 0.77); 93% of 6247 genes in the overlap are concordantly up- or down-regulated between the two experiments. (C) Heatmap visualization of the estimated expression of 372 genes selected for the model RAS/MET signature (concordant meant at least twofold change in both MT-HGF and WT-RAS). Genes (columns) and samples (rows) are ordered by hierarchical clustering using Euclidean distance and complete linkage. (D) GSEA enrichment plots are shown for top-enriched gene sets in the model RAS/MET signature. The green line is the running enrichment score calculated along the ranked gene list represented by the red blue horizontal bar (average t statistic from MT-HGF and WT-RAS comparisons ranked from the highest positive to the highest negative value); the vertical black bars in the plot indicate the position of the genes from the respective GO terms, which are mostly situated within up-regulated genes. (E) GSEA enrichment map for nonredundant and overlapping GO gene sets generated with REVIGO algorithm (SimRel <0.9); highly similar GO terms (3% of the strongest pairwise semantic similarities) are linked to each other with edges weighted by the semantic similarity; GO nodes are color-coded by the significance of enrichment (GSEA FDR <5%) and sized proportionally to the percentage of genes annotated to the term. Unweighted edges connect genes from the model RAS/MET signature, which are also among the GSEA core genes (“leading edge”), to the GO terms to which they are annotated. The gene nodes are color-coded by the average difference in expression of MT-HGF and WT-RAS versus WT-control. Table S2 contains detailed statistical results of the GSEA.

Gene set enrichment analysis [GSEA (43)] identified a total of 37 ontologies with 33 nonredundant gene functions significantly enriched in the 372-gene shared signature (see details in table S2). Most significantly enriched terms, in particular Gene Ontology (GO) domains (Fig. 7D), were keratinization [nominal enrichment score (NES) = 2.1, Q < 2 × 10−4; biological process], serine-type endopeptidase activity (NES = 2.2, Q < 2 × 10−4; molecular function), and cornified envelope (NES = 2.0, Q = 4 × 10−4; cellular component), indicating up-regulation of these functions. Semantic similarity networks were built from the enriched GO terms using REVIGO algorithm (Fig. 7E) (44). The semantic similarities over the GOs provided quantitative information on gene functional relationships, helping to gain a more holistic view on the processes affected. Using this approach, we identified eight GO networks with highly overlapping gene functions, which indicated that the top concordant RAS/MET signature captures the very essential genes related to keratinocyte biology; these include biological processes implicated in tissue development (epidermis and ectoderm development, and epithelial differentiation) and extracellular matrix components. Peptidase endopeptidase was revealed as a major category of concordant gene expression changes not previously studied in initiated keratinocytes. Included in this group are members of the TMPRSS11 family of membrane-bound serine proteases not previously specifically linked to RAS or MET transformation. Several of these proteases have the capacity to release membrane-bound growth factors to stimulate growth as well as migration of tumor cells. Lipid and fatty acid metabolism pathways emerged from this analysis as prominently associated biological and molecular functions, consistent with recent associations of these pathways with EGFR activity (45). The IPA (Ingenuity Pathway Analysis) Upstream Regulator tool enabled us to infer the activity state of upstream factors responsible for the observed expression changes in a set of genes. IPA and its Ingenuity Pathway Knowledge Base predicted inhibition (bias-corrected z score < −1.96) of 17 and activation (bias-corrected z score > + 1.96) of 20 upstream regulators whose targets were enriched in the 372-gene signature of RAS and activated MET (P ≤ 0.01) (table S3). HRAS, KRAS, and NUPR1 appeared on this list of upstream regulators.

MET signaling contributes to human cutaneous SCC

In a limited study of acute HGF treatment of normal human keratinocytes, both EGFR ligands and cytokine markers followed a similar pattern of change as seen in MT-HGF mouse keratinocytes (fig. S12). To address the frequency that MET is detected in human cutaneous SCC, we used human skin SCC tissue arrays (Biomax) to probe for MET by immunohistochemical (IHC) staining and histomorphometric analysis using the Aperio software. In most of the 76 cases examined, intense MET staining was readily detected (Fig. 8A) in many grade 1 lesions, and intensity was increased in grade 2 lesions when compared to grade 1 (Fig. 8B). Although the number of grade 3 lesions was too small to obtain statistical significance, the mean scores of grade 2 and 3 lesions were the same. These data suggest that MET is commonly abundant in human cutaneous SCC development and could contribute to disease progression. We were unsuccessful in obtaining a signal for phosphorylated MET in adjacent cutaneous SCC tissue arrays using multiple phosphorylated MET antibodies and putative positive control samples. Instead, we probed for expression of HGF transcripts by in situ hybridization (Fig. 8C). In this analysis, 72% of 66 evaluable human cutaneous SCC tumor punches were positive for HGF expression with much of the signal detected in the tumor epithelium and individual stromal cells (Fig. 8C). Parallel IHC staining for HGF protein produced the same pattern. We also examined a series of six human established SCC cell lines without RAS mutations to detect whether MET was constitutively activated when probed for phosphorylated MET by immunoblotting. Although only SCC9 in this panel showed MET inhibitor–responsive phosphorylation of MET (activation), all lines, including SCC9, responded to exogenous HGF through an increase in the phosphorylation of MET and the consequent reduction of MET, indicating an intact MET-responsive pathway (Fig. 8D).

Fig. 8 MET protein and HGF transcripts are abundant in many human cutaneous SCC.

(A) Immunostaining for MET in human skin SCC tissue array, visualized by bright-field microscopy. (B) Quantification of MET staining intensity in the epithelial compartment using the Aperio software ImageScope according to tumor grade. Grade 1, n = 49; grade 2, n = 20; grade 3, n = 5. *P < 0.05 versus grade 1, two-sided Student’s t test. (C) HGF RNA in situ hybridization and protein IHC in human SCC. RNA in situ signal (red) was deconvoluted with the Aperio ScanScope algorithm and pseudocolored with Adobe imaging software. Scale bars, 200 μm. Representative positive and negative tumors are shown for in situ hybridization and IHC. (D) Immunoblotting in human SCC cell lines grown to confluence and treated for 24 hours with DMSO, capmatinib (Cap.), PHA665752 (PHA), or recombinant HGF. Blots are quantified in fig. S15B.

From the 372 most concordant genes in the GSEA, we applied a systems approach, Denoising Algorithm based on Relevance network Topology (DART) (46), which enabled us to derive a cross-species, overall measure of the RAS/MET signature activity in an individual patient sample and compare the activity scores across the different tissues in the training and validation patient data sets. Activity scores in patient-matched samples from the normal epidermis (NE), actinic keratosis (AK), and SCC displayed 10 of 13 patients having higher activity scores in the AK sample than in the NE (P = 0.003) (Fig. 9A). In all but one patient, the estimated activity of the RAS/MET signature was higher in the SCC than that in NE tissue (P = 0.0005). Increased activity scores are also observed in the SCC compared to the AK samples (P = 0.048). For further verification, we computed the DART scores using the same network in two independent SCC test data sets (47, 48). Higher activation of the RAS/MET signature in the AK and SCC samples than in the NE was again observed (P = 0.001 and P = 0.0001, respectively), although the difference between AK and SCC activity scores was not confirmed (P = 0.68) (Fig. 9B). Overall, the results support the conclusion that MET signaling could occur early in many squamous skin cancers and persist through the advanced stage.

Fig. 9 The model RAS/MET signature is active in skin cancers from patients.

(A) Predicted DART scores of the gene signature activity in patient-matched samples (GSE32979; n = 13) from NE, AK, and SCC. (B) Predicted DART scores of the gene signature activity in patients and healthy controls, combining data from GSE2505 and GSE42677: NE, 16 samples; AK, 9 samples; SCC, 15 samples.


Our work addresses the early changes required for the conversion of a normal keratinocyte into a premalignant tumor on the road to forming a cancer. The discoveries in the mid-1980s that an activated Hras allele is present in nearly all benign tumors induced on mouse skin by DMBA-TPA exposure together with experimental evidence that activated HRAS is sufficient to initiate normal keratinocytes to produce benign tumors were milestones in carcinogenesis research (49). In the interim decades, numerous laboratories have filled in the intricate cascading signaling molecules downstream from oncogenic Ras alleles in multiple model systems (50). In mouse keratinocytes, crucial biochemistry for oncogenic HRAS initiation appears to involve the up-regulation of ligands for and autocrine activation of EGFR. Because RAS is downstream from EGFR signaling, this requirement for EGFR activation may indicate that the signal strength of a single mutated Ras allele is not sufficient to drive early neoplasia. After EGFR activation, IL-1α is released and activates IL-1R on keratinocytes, creating a second autocrine loop, leading to the activation of NF-κB signaling that modifies expression of specific keratinocyte genes involved in tumor formation including increasing expression and release of CXC ligands such as CXCL1 (6). The consequences of cytokine release are both autocrine, stimulating tumor cell migration through activation of keratinocyte CXCR2, and paracrine, attracting immune cells into the tumor stroma (oncogene-induced inflammation) (51). Inhibition of any one of these autocrine loops impairs tumor formation. We now show that activation of keratinocyte MET through elevated autocrine or paracrine HGF in the skin microenvironment uses these same pathways to be oncogenic for keratinocytes and is sufficient for tumor formation in the absence of Ras mutations but in the presence of a strong promoting stimulus such as wounding or enhanced cutaneous PKCα. The focal nature of tumor formation in the HGF-MET mice suggests that a subpopulation of keratinocytes in the skin epithelium is particularly sensitive to MET activation and the promoting stimulus. In normal mouse skin, HGF is detected in the hair follicle dermal papilla and MET in the hair follicle including the hair follicle bulge where stem cells reside (52). In the MT-HGF mouse, the expression of HGF is more widespread. Therefore, it is reasonable to speculate that bulge stem cells give rise to MET-induced squamous tumors, but further studies are required to support this possibility. The combination of DMBA and MT-HGF favors the selection of Kras mutant tumors in addition to the expected Hras mutant tumors, potentially arising from the same cell compartment. The skin carcinogenesis literature is peppered with examples of Kras mutant tumors emerging after carcinogen initiation when the promoting environment is modified from the standard TPA protocol (5355). In our model, it is notable that only mutant Hras tumors and no mutant Kras tumors were detected after DMBA in the DT group, emphasizing the importance of context in the selection of incipient mutant tumor cells. Furthermore, many of the tumors in this group were at lower risk for malignant conversion (fig. S1), whereas the increased frequency of Kras mutations in the DMBA-MT-HGF group might have contributed to malignant conversion (54).

There is limited information regarding the precise contribution of the HGF-MET axis in normal skin homeostasis. Deletion of Met from the skin epithelium in mice does not produce a skin phenotype (52, 56) but does impair wound healing, whereas HGF administration to chronic wounds accelerates healing (56, 57). The MT-HGF mouse develops spontaneous internal tumors and cutaneous melanomas (25), but the epithelial skin compartment is not a particular target for spontaneous tumor development (58). Nevertheless, MET is abundant in a substantial fraction of human HNSCCs, and activating mutations of MET, although rare, may contribute to radioresistance in these tissues (11, 12, 59). An intact HGF-MET pathway was essential for skin tumor formation in a transgenic mouse model of skin-targeted overexpression of matriptase, a serine protease that is capable of converting pro-HGF to the mature MET ligand (12). Furthermore, Met copy number increases are frequent in DMBA-induced mouse skin papillomas and increase further during tumor progression to SCC (60). The high frequency of human skin SCCs expressing MET and HGF (Fig. 8), the higher intensity of MET protein in progressing human skin cancers (Fig. 8), and the increasing activity level of the RAS/MET activation signature derived from human skin precancer and cancer databases (Fig. 9) support a contribution of MET signaling to human skin cancer. Limited results suggest that both mouse and human keratinocytes respond similarly to HGF stimulation.

The mechanism through which cell-autonomous or paracrine activation of MET in keratinocytes produced tumors became clearer when we determined that, phenotypically and biochemically, these keratinocytes reproduced the biology of RAS-transformed keratinocytes. Like RAS, MET activates EGFR through enhancing expression of the EGFR cognate ligands and controlling their maturation through the membrane-bound protease ADAM17, thus establishing the autocrine loops necessary for tumor formation. Although previous studies have shown that EGFR can lead to MET activation (6163), our data indicate that EGFR is an obligatory effector of MET-driven mouse skin carcinogenesis. Blocking EGFR reverses the MET biochemical signature and causes MET-driven tumors to regress. Pharmacological inhibition of EGFR causes major changes in the gene expression profile of MT-HGF keratinocytes (fig. S7). This reliance on EGFR activity for tumor growth in vivo is also true for oncogenic RAS (31), further highlighting the commonalities among those two initiators of skin carcinogenesis. The crosstalk between MET and EGFR in our model is unidirectional because oncogenic RAS (and subsequent EGFR activation) does not cause MET activation (Fig. 2B), and treatment of DT keratinocytes with an EGFR inhibitor does not decrease phosphorylated MET levels (fig. S5).

The activation of ADAM17 by oncogenic RAS has been recently reported in both pancreatic and colorectal cancers (64, 65), releasing EGFR ligands that are essential for tumor growth in both models. A previous report suggested HGF-MET–induced ADAM17 activity in invading trophoblast cells (66), but this relationship in tumors has not been reported before as far as we can tell. We now show that ADAM17 is a central player in producing the EGFR activation critical for the oncogenic signature in both RAS- and HGF-MET–transformed keratinocytes, further linking these two tumor initiators. Our data have unexpectedly revealed that MET activation enhances SRC activity and iRhom expression, particularly iRhom2, leading to ADAM17 activation likely through translocation and maturation of ADAM17 proteolytic activity (40). Accumulating evidence suggests that iRhoms have important functions in cutaneous biology (6769), but the connection to MET activation has not been made. When stabilized by mutation, the short-lived iRhom2 protein enhances cutaneous wound healing through activation of ADAM17 and release of EGFR ligands (69, 70). Of particular interest regarding cancer, stabilizing mutations in iRhom 2 cause tylosis esophageal cancer characterized by cutaneous dyskeratosis and inherited susceptibility to squamous esophageal cancer attributed to enhanced EGFR signaling through ADAM17 (71). This new link to MET suggests that exploring iRhom2 and ADAM17 in cancers associated with MET activation is worthy of further study.

Beyond the establishment of the autocrine loops emanating from EGFR activation, the pathways involved in both MET and RAS initiation of keratinocyte neoplasia converged on the expression of many genes and common pathways. The vast majority of the aberrantly expressed genes from keratinocytes initiated by activated RAS or MET overlapped and were concordant. Selecting the 372 most modulated and concordant RAS and MET genes to produce a highly enriched tumor-associated gene expression data set coupled with GO analyses derived from GSEA allowed us to identify biologically meaningful and coherent sets of gene functions involved. The concordant gene list (table S1) confirmed the changes in cytokines, growth factors, and differentiation markers that we have come to know as the signature of initiation. Not surprisingly, the functional gene set analysis confirmed the importance of pathways associated with epidermal development and keratinocyte differentiation but revealed two unexpected highly relevant pathways. A strong functional association with endopeptidase-peptidase activity was revealed (Fig. 7E and table S2). Matrix metalloproteinases (MMPs) have been associated with transformation of keratinocytes and a number of other cancers, but the profiling revealed an association with a number of type II transmembrane serine proteases. The functions of members of this family are not well known (72), but TMPRSS13 is reported to activate pro-HGF to the mature ligand (73) and TMPRSS11E is reported to decrease in HNSCC (74). This is an area worthy of further study. Pathways controlling lipid biosynthesis, lipid transport, and fatty acid synthesis were also unexpectedly revealed in the functional enrichment analysis. Recent studies indicate a strong association among EGFR signaling, lipid metabolism, and cancer growth. Much of the data are derived from glioblastomas, but our results suggest that the concept may be more widespread (45, 75, 76).

It is premature to speculate on what regulates the expression of downstream effectors in the RAS/MET signature profiles, but the IPA Upstream Regulator tool identified several transcriptional regulators previously associated with skin carcinogenesis or RAS transformation (table S3). Among these, reduction in p53-regulated genes has been associated with skin tumor progression (77) and NUPR1 is required for RAS transformation of pancreatic cells (78). It is notable that HRAS, KRAS, and RAF1 all appear on the upstream regulator algorithm as does tumor necrosis factor (TNF) representing the key role of NF-κB in the transformation process.

Although our study focused on the role of HGF-MET in cutaneous cancer, MET signaling has been linked to a broader variety of human cancers, prompting the development of MET inhibitors as cancer therapeutics (10). Both preclinical and clinical experiences have revealed crosstalk of MET and EGFR in the therapeutic setting. In particular, MET amplification is identified as a resistance mechanism for tumor cell lines and lung cancer patients treated with EGFR inhibitors (79). Conversely, EGFR inhibitors enhance the antitumor activity of MET inhibitors in cell lines, xenograft models, and lung cancer patients (8082). Thus, clinical trials of combined anti-MET and anti-EGFR treatment in advanced internal cancers show promise. Our data suggest that cutaneous cancers, perhaps in the setting of organ transplant patients where cutaneous SCC can be life-threatening, should be considered as a model for combined MET-EGFR kinase inhibitor therapy where the dynamics of tumor response can be followed visually and sampled temporally.


Mice and treatments

Mouse studies were performed under a protocol approved by the National Cancer Institute (NCI) and the National Institutes of Health (NIH) Animal Care and Use Committee. The construction and characterization of K5-PKCα and MT-HGF mice on an FVB/N background were previously described (58, 83). K5-PKCα and MT-HGF mice were crossed to produce F1 mice MT-HGF/K5-PKCα DT mice. These DT mice were then crossed to wild-type FVB/N background mice to produce F2 mice of all four genotypes. At various time points after TPA application, skin biopsy samples were fixed in formalin (ED Biosciences) and embedded in paraffin for hematoxylin and eosin (H&E) analysis and IHC. Epidermal hyperplasia was quantified by measuring the epidermal height at five randomly chosen sites per skin biopsy (×400 magnification; Nikon Eclipse E400). ADAM17fl/fl (37) mice were purchased from The Jackson Laboratory.

Tumor induction experiments

Initiation with DMBA (Sigma-Aldrich) was done by a single topical application of 20 μg of DMBA in 0.1-ml acetone on the backs of 4-day-old mice. Twice a week, TPA (LC Laboratories) treatments [1 μg (1.6 nmol) in 0.2 ml of acetone] were started when mice reached 6 weeks of age and continued for 5 weeks. Squamous tumors induced by the initiation-promotion protocol were counted weekly until 20 weeks after DMBA treatment. For tumor studies where exogenous initiation was eliminated by excluding DMBA treatment, biweekly TPA treatment [1 μg (1.6 nmol) in 0.2 ml of acetone] was started when mice reached 6 weeks of age and continued for 10 weeks. Portions of skin tumors were either frozen or formalin-fixed at times, and tumor type (squamous papilloma or SCC) was verified by visualizing H&E-stained sections.

Syngeneic mouse grafting

Confluent cultures of wild-type and MT-HGF primary keratinocytes were used for grafting as described previously (30). Six million keratinocytes (wild-type or MT-HGF) were respectively mixed with 6 million wild-type or MT-HGF mouse primary dermal fibroblasts (cultured for 1 week) and grafted onto the back of syngeneic recipient (wild-type or MT-HGF) mice on a prepared skin graft site located in the interscapular region. Three to four weeks after grafting, when tumors had developed on the MT-HGF grafts, daily gavage treatment with gefitinib (100 mg/kg) or vehicle control (10% DMSO in water) was conducted for 2 weeks. These dosage and regimen were chosen on the basis of efficacy and lack of side effect as reported previously (42). Tumor dimensions were measured weekly using calipers, and approximate tumor volumes were determined by multiplying tumor height × length × width. Tumors were collected in 10% formalin and processed for H&E staining (Histoserv Inc.), Ki67, and/or platelet endothelial cell adhesion molecule 1 (PECAM-1) (CD31) (Santa Cruz Biotechnology; sc-1506). Immunostaining was performed at the Pathology-Histotechnology Laboratory (NCI-Frederick). The percentage of Ki67-positive cells was quantified by counting Ki67-labeled basal cells from at least three randomly chosen areas. The microvascular density was determined by counting the number of CD31-stained cells in at least five fields (×400 magnification; Nikon Eclipse E400).

Cell culture

Primary mouse keratinocytes were isolated from newborn transgenic and wild-type littermate epidermis as described and cultured in modified Eagle’s medium (MEM), 7% Chelex-treated fetal calf serum (Gemini Bio Products), and 0.05 mM calcium unless otherwise indicated (30). EGFR inhibitors AG1478, PD168393 (Calbiochem), gefitinib (LC Laboratories), MET inhibitor PHA 665752 (Tocris), MET inhibitor capmatinib (Selleckchem), and ADAM17 inhibitor GM6001 (Calbiochem) were diluted in DMSO. Antibodies against TGFα (Abcam), AREG (R&D Systems), and IL-1ra (5 μg/ml, IL-1ra or anakinra; Division of Veterinary Resources, NIH) were diluted in culture medium and added to cell culture medium as indicated before cell harvesting. Nonsilencing (control), Hras, Kras, Nras, Src, iRhom1, iRhom2 siRNAs (Qiagen), and ADAM17-silencing siRNA (Qiagen and Thermo Scientific) were transfected using RNAiMAX (Life Technologies) at a final concentration of 20 nM. At 24 or 48 hours after transfection with siRNA, cells were treated with HGF (PeproTech; 40 ng/ml) or harvested for analysis. Primary mouse keratinocytes were isolated from newborn EGFR-deficient and wild-type mice, and confluent cultures were treated for 3 hours with HGF. Human epidermal carcinoma cell line A431 [American Type Culture Collection (ATCC) CRL-1555] was cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum. SCC-4 (tongue; ATCC CRL-1624), SCC-9 (tongue; ATCC CRL-1629), SCC-15 (tongue; ATCC CRL-1623), and SCC-25 (tongue; ATCC CRL-1628) were cultured in DMEM:F12 (Lonza) supplemented with 0.5 mM sodium pyruvate, hydrocortisone (400 ng/ml), and 10% fetal bovine serum. Cutaneous SCC-13 (84) was cultured in MEM (Lonza). Confluent cultures were treated with MET inhibitors capmatinib INCB28060 (Selleckchem) and PHA 665752 (Tocris) for 24 hours.

Retroviral and adenoviral constructs

The v-rasHa replication-defective ecotropic retrovirus was prepared using Psi 2 producer cells (85). Retrovirus titers were routinely 1 × 107 virus/ml. Cultured primary keratinocytes were infected with v-rasHa retrovirus (here referred to as RAS or oncogenic RAS) on day 3 at a multiplicity of infection (MOI) of 1 in medium containing Polybrene (4 μg/ml; Sigma-Aldrich). The IκBsr (86) adenovirus was introduced into primary keratinocytes using an adenoviral construct driven by a cytomegalovirus (CMV) promoter, and empty adenovirus was used as control (A-CMV). Keratinocytes were adenovirus-infected for 30 min in serum-free medium with an MOI of 10 viral particles per cell and Polybrene (4 μg/ml; Sigma-Aldrich) to enhance uptake. Serum-containing medium was added to the cells for the next 48 hours after the infection. Primary keratinocytes from ADAM17fl/fl (The Jackson Laboratory) mice were cultured in 0.05 mM Ca2+ medium to confluence and transduced for 48 hours with Cre adenovirus (gift of F. Gonzalez, NCI) and then challenged with HGF (PeproTech; 40 ng/ml).


Cultured keratinocytes with or without v-rasHa transduction were lysed in MPER lysis buffer (Pierce) supplemented with 200 μM NaVO3, 10 mM NaF, and cOmplete Mini tablets (Roche). Lysates prepared for ADAM17 analysis were supplemented with the metalloproteinase inhibitor 1,10-phenanthroline (Sigma-Aldrich) at 10 mM. Proteins were quantified by the Bradford method (Bio-Rad) and separated by 10.5 to 14%, 4 to 20%, or 10% tris-HCl gels (Bio-Rad). To prepare lysates for immunoblotting from skin biopsy samples, flash-frozen skin biopsies in liquid nitrogen were processed using a ball mill pulverizer (Mikro-Dismembrator S, Sartorius) for 1 min at 2000 rpm. These samples were subsequently resuspended in radioimmunoprecipitation assay buffer supplemented with a protease inhibitor cocktail (Halt, Thermo Scientific). For analysis of keratinocyte differentiation in vitro, cultures were washed once with PBS (Ca2+- and Mg2+-free), and total cell lysates were prepared in situ on the dish using lysis buffer [10 μl/cm2; 5% SDS and 20% 5-mercaptoethanol in 0.25 M tris (pH 6.8)]. Total EGFR (Santa Cruz Biotechnology), phosphorylated EGFR (Invitrogen), total MET, and phosphorylated MET (Tyr1234/1235; Cell Signaling) were detected after overnight incubation with 1:500 dilution of each antibody. Antibodies against TACE/ADAM17 (QED) and SRC (Cell Signaling) were incubated overnight at 1:1000 dilution. COX-2 antibodies (Cayman Chemical) were incubated overnight at 1:750 dilution. Antibodies against HSP90 (BD Transduction Laboratories) were incubated for 2 hours at room temperature at 1:5000 dilution. Antibody for K8 (University of Iowa) was used at 1:100, and antibodies against K1 and K10 (Covance) were used 1:10,000, both incubated overnight. Antibody for actin (Chemicon) was used 1:10,000 for 1 hour at room temperature. ECL SuperSignal (Pierce) system was used for detection. The intensities of immunoblots were quantified using ImageJ (NIH), and the relative expression of targeted proteins was normalized.

[3H]Thymidine incorporation assay

Keratinocytes were plated in 24-well plates, and 3 days after plating, [3H]thymidine (1 μCi per well) was added for 4 hours. Cultures were trypsinized, well content was transferred to glass fiber filters using a Brandel cell harvester, and incorporated counts were read using a Wallac TriLux 1450 MicroBeta scintillation counter (PerkinElmer).

Isolation of tumor DNA

Sections (30 μm) were cut from formalin-fixed, paraffin-embedded papillomas arising from DT mice treated with TPA only as described above. Genomic DNA was isolated using QIAamp DNA FFPE Tissue Kit (Qiagen) or DNeasy Tissue Kit (Qiagen) for tumors that were snap-frozen (DMBA-TPA studies) in liquid nitrogen at collection time.

PCR amplification and sequencing of K-Ras, H-Ras, and N-Ras genes

Primers specific for mouse Kras, Hras1, and Nras were purchased from Invitrogen and are listed in table S4. Nested primers were used for Kras and Hras1 analysis. All PCRs used Taq Master Mix (Qiagen) and a PCR protocol previously described (54). After PCR amplification, product was purified using QIAquick PCR Purification Kit (Qiagen). Sequencing was performed using the ABI Prism Dye Terminator Kit and was performed by the NCI sequencing core.

Reverse transcription PCR analysis

RNA was isolated from cultured cells and posthomogenized tissue biopsies with TRIzol using the manufacturer’s protocol (Invitrogen). Complementary DNA synthesis and real-time PCR analysis were conducted as previously described (87). Predesigned QuantiTect primers (Qiagen) were used for all genes except for Gapdh, where the following primers were designed: 5′-CATGGCCTTCCGTGTTCCTA-3′ (forward) and 5′-GCGGCACGTCAGATCCA-3′ (reverse).

Myeloperoxidase assay

Skin samples were homogenized in potassium phosphate buffer (pH 6.0) containing 0.5% hexadecyltrimethylammonium bromide, sonicated, and freeze-thawed three times, after which sonication was repeated. The suspension was centrifuged at 40,000g for 15 min, and 10 μl of supernatant was added to 290 μl of potassium phosphate buffer (pH 6.0) containing o-dianisidine dihydrochloride (0.167 mg/ml; Sigma-Aldrich) and 0.0005% hydrogen peroxide. Changes in optical density were monitored at 460 nm at 25°C over a 4-min period.

IHC analysis

Human skin cancer tissue arrays were obtained from U.S. Biomax (SK802a) and were used for MET and HGF expression and localization studies. IHC staining used the MET antibody (Cell Signaling D1C2; 1:100 dilution) and the HGF-specific antibody (5 μg/ml; R&D Systems AF-294) and procedures described by the array manufacturers. Target retrieval (Dako) was performed in a microwave for 10 min. Secondary antibody (anti-rabbit, 1:300; Vector) and avidin-biotinylated conjugation were performed with ABC Kit (Vector) according to procedures described by the manufacturer. Slides were scanned and analyzed using a ScanScope XT scanner and ImageScope viewing software version (Aperio Technologies Inc.).

HGF in situ hybridization

This procedure was performed by a commercial vendor (Phylogeny Inc.) according to the published method (88). Human cutaneous SCC tissue arrays from U.S. Biomax were deparaffinized for 5 min in xylene, immersed in 100% ethanol for 5 min, and then air-dried. Treatment was with Bond Epitope Retrieval Solution 2 from Leica (AR9640) for 30 min. LNA probes HGF-1-1, HGF-1-2, and Scramble-miR were prepared according to the manufacturer’s recommended conditions (Exiqon), and each was labeled at the 5′ end with digoxigenin. Probes were diluted to 25 fmol/μl in Exiqon hybridization buffer (208022). Probe solutions were placed on tissue sections, covered with polypropylene coverslips, and heated to 60°C for 5 min, followed by hybridization at 37°C overnight. Sections were washed in intermediate stringency solution (0.2× SSC with 2% bovine serum albumin) at 55°C for 10 min. Sections were treated with anti–digoxigenin–alkaline phosphatase conjugate [1:150 dilution in tris buffer (pH 7); Roche] at 37°C for 30 min. Development was carried out with nitro blue tetrazolium/bromochloroindolyl phosphate from Thermo Fisher (34042). Development was closely monitored and stopped when the control sections appeared light blue. Development time with the chromogen was between 15 and 30 min. Sections were counterstained with nuclear fast red for 3 to 5 min, rinsed, and mounted with coverslips. Slides were scanned using Aperio ScanScope XT System and analyzed using ImageScope (version; Leica). To distinguish signal from counterstain, a color deconvolution algorithm (Aperio) was used to separate chromogen-generated signal from counterstain and pseudocolored using Adobe Photoshop.

Enzyme-linked immunosorbent assay

Quantikine ELISA kits for CXCL1 and AREG were used according to the manufacturer’s protocol (R&D Systems).


Total RNA was extracted from keratinocyte cultures with TRIzol (Invitrogen) using the manufacturer’s protocol. Three independent biological replicates were evaluated. Gene expression profiling was performed using Affymetrix Mouse Gene 1.0 ST Array platform. RNA quality testing, microarray hybridization, and processing were performed at the Laboratory of Molecular Technology in Frederick, MD. Raw gene expression data (.CEL files) were processed with the Robust Multichip Average algorithm and quantile normalization (89) implemented in the BRB-ArrayTools version 4.3.2 (software developed by R. Simon and BRB-ArrayTools Development Team; Gene annotation and mapping of the mouse-human orthologs were imported from mAdb ( For genes with multiple probesets, average expression was calculated from the two most correlated probesets if the Pearson correlation reached at least 0.8; otherwise, the probeset with the highest median was selected. A total of 17,114 mouse-human orthologs were input into the statistical analysis. The data presented here have been deposited in the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) database and are accessible through GEO series accession number GSE58671.

Genomic analyses

To analyze for variance, one-way RVM-ANOVA and contrast comparisons (90) were used to identify differential expression between wild-type and MT-HGF, MT-HGF-AG1478, and RAS keratinocytes. FDR (91) less than 1% was chosen as the significance cutoff. In addition, genes showing at least twofold change and the same direction of change in both RAS keratinocytes and MT-HGF keratinocytes were selected to the model RAS/MET initiation signature. The calculations were performed in R programming environment (R version 3.0.2) and R/Bioconductor qvalue 1.36.0 library (92).

To analyze for GO, GSEA was conducted using GO gene set collections with the broadest terms eliminated (GO_FAT) from DAVID database (93, 94). In GSEA, genes were ranked by the average ANOVA t statistic, and gene sets were considered enriched at 5% FDR. Furthermore, subsets of enriched gene sets were selected if the core genes (GSEA leading edge) included at least five genes from the top RAS/MET signature and were summarized with semantically nonredundant terms using REVIGO algorithm (44). REVIGO interaction maps were visualized using R/CRAN igraph 0.7.0 library (95).

For pathway discovery, genes from the RAS/MET signature were also used as input to the Upstream Regulator tool in the IPA (Qiagen; The enrichment P value (Fisher’s exact test) less than 0.01 and bias-corrected z score (predicted activation state) exceeding ±1.96 were used to select candidate upstream regulators.

An unsupervised strategy of Jiao et al. (46) for inferring molecular activation of a model perturbation signature (DART) was used to evaluate expression of the model RAS/MET signature in SCC patients. Microarray data sets available in GEO were used in the analysis, namely, patient data from Hameetman et al. (96), Mitsui et al. (47), and Nindl et al. (48). Gene expression in each data set was standardized by means of z score (the mean of zero and unit SD). Predicted activity levels of the RAS/MET signature were compared between NE, AK, and SCC samples using the nonparametric Wilcoxon signed-rank test (paired data) or Wilcoxon rank sum test (unpaired data). The computations were performed using R/Bioconductor DART version 1.8.0 (46).


Unless otherwise specified, biochemical data were analyzed by Prism software, and significance values were assigned through Mann-Whitney U test, Student’s t test, or one-way ANOVA with Dunnett’s posttest. P < 0.05 was considered to be significant.


Fig. S1. Malignant conversion rate is increased in MT-HGF compared to DT animals.

Fig. S2. Ras mutation analysis in DMBA-TPA and MET-generated skin lesions.

Fig. S3. MT-HGF keratinocytes can form squamous papillomas when orthotopically grafted.

Fig. S4. HGF-MET does not enhance responses to TPA.

Fig. S5. Treatment of MT-HGF keratinocytes with a MET inhibitor (PHA665752) reverses their phenotype in vitro.

Fig. S6. EGFR dependence of the MET and RAS signatures in keratinocytes.

Fig. S7. Effects of EGFR inhibition on the transcriptional profile of MT-HGF keratinocytes.

Fig. S8. Activated MET gene signature is not dependent on individual Ras allele expression.

Fig. S9. TGFα and AREG neutralizing antibody activities reduce the activation of EGFR in MT-HGF keratinocytes.

Fig. S10. Both iRhom1 and iRhom2 contribute to the release of AREG upon MET activation.

Fig. S11. Gefitinib treatment reduces proliferation and microvessel density in MT-HGF squamous papillomas.

Fig. S12. MET activation causes EGFR ligand, cytokine, and chemokine mRNA up-regulation in human keratinocytes.

Fig. S13. Quantification for immunoblots represented in Figs. 2 and 3.

Fig. S14. Quantification for immunoblots represented in Figs. 4 and 5.

Fig. S15. Quantification for immunoblots represented in Figs. 6 and 8.

Table S1. The list of 5812 significant genes concordantly up-regulated or down-regulated in the wild-type RAS and MT-HGF keratinocytes.

Table S2. GO functions enriched in the RAS/MET 372-gene signature.

Table S3. Upstream regulators predicted by IPA to be responsible for expression changes in the RAS/MET 372-gene signature.

Table S4. PCR primers.


Acknowledgments: We thank S. Walters for care of the mouse colonies; D. Butcher and M. Anver from the Pathology-Histotechnology Laboratory (SAIC-Frederick) for excellent technical assistance; C. Ostermeier from the Cryopreservation and Assisted Reproduction Laboratory (SAIC-Frederick); and M. Malik and D. Goldstein for the use of the CCR Research Exchange (CREx). G. Vande Woude provided valuable advice. Funding: This work was funded by the intramural program of the Center for Cancer Research of the NCI. Author contributions: C.C., K.S., A.R., M.K., L.W., W.D., F.L., A.Z., and K.B.R. performed experiments; A.M.M. performed computational work; S.H., J.D., M.R.S., and A.R. contributed IHC analysis; G.M. provided the MT-HGF mouse model and wrote the paper; and S.H.Y. designed experiments, analyzed data, and wrote the paper. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The microarray data are deposited to NCBI (GEO accession number GSE58671).
View Abstract

Navigate This Article