FGFR Inhibition Overcomes Resistance to EGFR-targeted Therapy in Epithelial-like Cutaneous Carcinoma

Cutaneous squamous cell carcinoma (cSCC) is the second most prevalent form of nonmelanoma skin cancer, and its incidence has increased in recent years (1). Most cSCCs conserve epithelial differentiation features and are considered well or moderately differentiated SCCs (WD/MD- or G2-SCC). A subset of cSCCs exhibits poorly differentiated traits (PD-SCC or G3-SCC) and is eventually undifferentiated and spindle shaped (PD/S- or G4-SCC). The latter histopathologic traits, along with tumor size, invasion level, and perineural invasion, are associated with enhanced recurrence and metastasis (2). Around 90% of cSCCs are successfully treated by surgical excision. Radiotherapy and/or cisplatin-based chemotherapy have been used until recently to treat recurrent, metastatic, and advanced/unresectable tumors (1, 3). However, these therapies are highly toxic to elderly patients with cSCC and produce limited clinical outcomes (4–6). Recently, the FDA has approved a treatment for advanced and metastatic cSCCs based on cemiplimab, an anti–PD-1 antibody that inhibits the PD-1–mediated immune checkpoint. Although a response was reported in almost half of the treated patients, 12%–20% of the cases showed primary resistance, and more than 40% of initially responding patients showed disease progression after 6 months of treatment (short-term response; ref. 7). Identifying the mechanisms that control cSCC growth and metastasis is critical for designing effective targeted therapies that enhance patient progression-free survival (PFS), including in those patients who are not suitable candidates for anti–PD-1–based immunotherapy.

EGFR expression is frequently upregulated in solid tumors, such as lung and colorectal carcinomas, glioblastomas, and head and neck SCCs (HNSCCs; refs. 8, 9). More than 50% of cSCCs overexpress EGFR, but the relevance of the signaling pathway in this tumor type is unclear (10, 11). EGFR mutations and gene amplifications have been identified in a small percentage of cSCCs (11–14) and HNSCCs (15). Cetuximab, an antibody that blocks EGFR activity, was approved for the treatment of HNSCCs and colorectal carcinomas without RAS-activating mutations by the FDA (16, 17). There have been several clinical trials of EGFR inhibitor–based therapies in cSCCs. In one such trial, cetuximab showed clinical activity, based on the evaluation of 31 patients with unresectable or metastatic cSCC that expressed EGFR and had not previously received chemotherapy, and yielded an overall disease control rate of 69% and a response rate (RR) of 28% at 6 weeks, although the disease progressed in 17%–19% of patients. The mean PFS and overall survival (OS) times were shorter than 6 and 9 months, respectively (18).

Gefitinib produced a moderate response in a single-arm phase II study (40 patients with metastatic and locoregional recurrent cSCC were treated with gefitinib, until the disease progressed or toxicities became intolerable; the responses of 37 patients were evaluated), with an overall RR of 16% and a disease control rate of 51% (19). The median duration of response was 31.4 months, median OS was 12.9 months, and median PFS was 3.8 months. As neoadjuvant therapy before standard surgery or radiotherapy, gefitinib produced a 45.5% RR in patients with aggressive or recurrent cSCC (20). In a single-arm phase II clinical trial, erlotinib exhibited an RR of 10% and PFS of 4.7 months in patients with recurrent or metastatic cSCC (21).

Predictive biological markers of response or even molecular traits of tumor cells were lacking from most of the studies, and no correlation between EGFR expression and response to EGFR-targeted therapy was observed in advanced cSCCs (18).

Here, we investigate the relevance of EGFR signaling to promote patient-derived cSCC growth, and the effect of EGFR-targeted therapy in these tumors, as well as the mechanisms involved in resistance to this treatment.

Fresh samples of human skin SCCs were supplied by the Plastic Surgery and Pathology Units of the Hospital Universitari de Bellvitge (IDIBELL, Barcelona, Spain). Samples were recovered in RPMI1460 (Life Technologies) medium containing 10% FBS (Life Technologies), 20 mmol/L HEPES (Sigma), and 1% of antibiotic/antimycotic (Sigma) shortly after surgery. Paraffin-embedded cSCC samples from cetuximab-treated patients were recovered from the Pathology Unit of the Hospital Universitario de Bellvitge/IDIBELL Biobank. These studies were carried out in compliance with the principles of the Declaration of Helsinki (Fortaleza, 2013). The treatment of the personal data was adjusted to the provisions of the European data protection regulation. This study was approved by the Research Ethics Committee of the Hospital Universitari de Bellvitge (Barcelona, Spain, IRB00005523). All patients were fully informed about the study before giving their signed consent to being included in it.

Fresh patient cSCC samples (2–4 mm³) were engrafted in the back skin of 6-week-old male NOD-scid IL2Rγ^null (NSG) mice. Tumor growth was monitored and the volume was calculated (V = π/6 × L × W²) every 2–3 days. Tumors of a critical size were surgically excised and small pieces were serially engrafted in a new NSG mouse. Resected mice with symptoms of poor health were sacrificed and examined for metastatic lesions. Animal housing, handling, and all procedures (reference No. 10402) involving mice were approved by the Institutional Animal Care and Use Committee IDIBELL (Barcelona, Spain) and Spanish authorities to be developed at IDIBELL Animal Facility (Barcelona, Spain, AAALAC Uni 1155). Protocols were performed in accordance with the guidelines of European Directive 2010/63/UE, Federation of European Laboratory Animal Science Associations, Animal Welfare Act, and “The guide for the use and care of Laboratory Animals.”

Tumor cells isolated from patient samples or PD/S-SCC patient-derived xenografts (PDXs; SCC11 cells) were grown in basic medium, comprising DMEM-F12 Medium (Life Technologies), 1× B27 (Life Technologies), and Penicillin/Streptomycin (PAA Laboratories). Tumor cells isolated from patient samples or MD/PD-SCC PDXs were grown in basic medium, alone (SCC10 and SCC34) or supplemented with EGF (20 ng/mL; Sigma; SCC16 and SCC24). Cells were cultured and grown at 37°C in a humidified 5% CO₂ incubator, as reported previously (22).

To in vitro–generated gefitinib-resistant (iGefR) cell cultures, SCC10, SCC16, and SCC24 cells were initially grown in basic medium with 2 μmol/L gefitinib and FGF2 (10 ng/mL; PeproTech) until they reached confluence. Surviving/tolerant cells were then split and cultured with increasing doses of gefitinib in the presence of FGF2, until cells were able to grow exponentially in 10 μmol/L gefitinib.

To determine the effect of pharmacologic inhibition of EGFR, patient cSCC–derived xenograft (cSCC-PDX) samples were engrafted in NSG mice. When their tumors were detectable (upon reaching a volume of ∼380 mm³), mice were randomly assigned to a control or inhibitor treatment group and orally treated with gefitinib (LC Laboratories; 75 mg/kg; diluted in sterile water with 0.05% Tween 80) or vehicle every 48 hours. Tumor volume was monitored every 2 days. Treatment continued until mice had received 21 doses of the drug, and the gefitinib-treated (GefT) tumors exhibited a stable and significant reduction in their growth. Upon halting the treatment, tumor relapse was allowed to occur in the absence of gefitinib. Samples of relapsed tumors were engrafted in new NSG mice and, when they became detectable, gefitinib treatment was resumed for one or two cycles to measure resistance acquisition.

To determine the effect of pharmacologic inhibition of FGFRs on gefitinib-resistant (GefR) tumors, mice carrying detectable GefR tumors were randomly assigned to one of two groups: a control group, which was treated every 2 days with gefitinib, following the regimen described above; and the NVP-BGJ398 + gefitinib group, which was treated every 2 days with gefitinib and daily with NVP-BGJ398 [MedChem; 25 mg/kg; diluted in acetate buffer and PEG300 1:1 volume/volume (v/v)]. To determine the effect of pharmacologic inhibition of FGFR as monotherapy, and to test the effect of this treatment in advance of the subsequent gefitinib treatment, mice carrying GefR tumors were randomly assigned to the control (gefitinib), NVP-BGJ398 monotherapy, and combined gefitinib + NVP-BGJ398 therapy groups, which were treated as described above. Mice from the NVP-BGJ398 monotherapy group were treated daily with the drug. After 25–30 days of treatment, mice were assigned to an NVP-BGJ398 treatment group, in which the same treatment was continued, or to an NVP-BGJ398 post-gefitinib group, at which point NVP-BGJ398 treatment was halted and animals were treated every 2 days only with gefitinib (75 mg/kg).

To analyze the effect of FGFR2 knockdown on GefR tumor growth, 3 × 10⁶ sh-control and sh-FGFR2-1 iGefR SCC10 cells, or sh-control and sh-FGFR2-1 GefR1 SCC10 cells, previously treated with doxycycline for 72 hours, were resuspended in 100 μL of DMEM and mixed with 100 μL of DMEM containing 1 × 10⁶ new born dermal fibroblasts (1:1 v/v). Mixed tumor cells and fibroblasts were subcutaneously injected in NSG mice. When tumors became detectable, mice were orally treated with gefitinib every 48 hours, and doxycycline (2 mg/mL) was continuously administered in their water bottle. The doxycycline solution was renewed every 48–72 hours. Tumor volume was monitored every 2 days. Tumors were surgically excised when they reached a critical size.

To determine the relevance of EGFR signaling in patient cSCC growth, we generated cSCC-PDXs by engrafting surgical samples with different grades and histopathologic features in immunodeficient mice. This subset of patient samples included two MD-SCCs (G2 grade), three MD/PD-SCCs (G2/G3 grade), and one undifferentiated PD/S-SCC (G4 grade; Supplementary Table S1). Most PDXs conserved the histopathologic features of the parental patient samples (Supplementary Fig. S1A), although some had a higher (SCC23-PDX) or even a lower (SCC34-PDX) histologic grade than their parental samples (Supplementary Table S1). This suggests that patient tumors have high intratumoral heterogeneity, and specific tumor cell populations may be selected during PDX establishment. In accordance with the findings of our previous studies (22, 23), patient G2/G3 cSCCs, matched primary culture cells, and cSCC-PDXs conserved the expression of E-cadherin and EpCAM epithelial markers. Conversely, advanced SCC11 lost the expression of these markers and strongly induced epithelial-to-mesenchymal transition (EMT) and mesenchymal markers, such as vimentin, ZEB1, and ZEB2 (Supplementary Fig. S1B–S1D).

cSCC-PDXs and cultured cells recapitulated the expression profiles of various tyrosine kinase receptors and ligands of their respective parental samples, following a specific pattern in accordance with their epithelial or mesenchymal features. Indeed, epithelial-like cSCC-PDXs (G2/G3 tumors) and derived tumor cells strongly expressed EGFR, ERBB2, ERBB3, as well as FGFR2B, FGFR2C, and ligands, whereas these receptors were strongly downregulated in mesenchymal-like SCC11-PDXs and cells (Fig. 1A and B; Supplementary Fig. S2A, S2C, and S2D). In contrast, the expression of FGFR1c, which is mostly associated with fibroblast and mesenchymal cells (24), was upregulated along with PDGFRA and PDGFRB in SCC11-PDXs and parental tumors, in accordance with the mesenchymal features of these tumors (refs. 22, 23; Supplementary Fig. S2B–S2D).

To determine the mutational and transcriptional landscape of the epithelial-like cSCC-PDXs, we performed RNA sequencing (RNA-seq) analysis of SCC10-PDXs, SCC16-PDXs, SCC24-PDXs, and SCC34-PDXs. The number of genetic alterations (single-nucleotide variations), including missense variants, frameshifts, in-frame insertions, deletions, and stop gains (Supplementary Fig. S3A; Supplementary Table S2), was within the range of those described in cSCCs (12–14). Indeed, mutations in TP53, NOTCH (NOTCH1, NOTCH2, or NOTCH3), FAT1, KMT2A, KMT2C, and IGF2R affected 75% of tumors, and alterations in a large subset of genes (Supplementary Fig. S3B) were detected at a similar frequency to those reported previously (12–14). No mutations were detected in EGF or FGF receptor family members, in either RAS or BRAF genes (Supplementary Fig. S3B; Supplementary Table S2), in accordance with the low frequency of these mutations in cSCCs (12–14). SCC34-PDXs showed E545K PIK3CA, G577E MET, and G1169R IGF1R mutations (Supplementary Table S2). PIK3CA mutations were previously detected in 10%–13% of cSCCs, whereas MET and IGF1R mutations were infrequent in this tumor type (12, 14). Analysis of a subset of mutations confirmed that they were also present in the respective parental patient tumors (Supplementary Fig. S3C). Therefore, cSCC-PDXs recapitulated most of the histopathologic features and expression profile of regulatory factors of primary patient tumors, making them exceptional preclinical models for evaluating the response of EGFR-targeted therapies.

Most cSCCs are diagnosed as G2 MD-SCCs and G3 PD-SCCs (25). As these epithelial-like tumors strongly express EGFR and ligands, we evaluated the relevance of this signaling pathway for cancer cell proliferation and cSCC growth. We detected EGFR phosphorylation in epithelial-like cSCC cells growing under basal culture conditions, which was inhibited, along with the phosphorylation of AKT and ERK1/2 downstream effectors, in response to gefitinib treatment (Fig. 1C). In addition, gefitinib significantly blocked tumor cell proliferation and survival, under basal culture conditions and after EGF stimulation (Fig. 1D). Similarly, these cells were highly sensitive to other EGFR inhibitors, such as afatinib and cetuximab (Fig. 1E and F). These results indicate that autocrine activation of EGFR signaling promotes the in vitro proliferation and survival of epithelial-like cSCC cells. In contrast, gefitinib treatment did not affect the proliferation and survival of mesenchymal-like cells from PD/S-SCCs with reduced EGFR expression (Fig. 1G). The E545K PIK3CA-activating mutation is known to be related to resistance to EGFR inhibitors in lung and colorectal tumors (26, 27), whereas the significance of the identified MET and IGF1R mutations to the activity of these receptors, or to the response to EGFR inhibitors, has not been reported. In this regard, we observed that proliferation of SCC34 cells was less sensitive to lower doses of gefitinib (0.1 μmol/L) than SCC10, SCC16, and SCC24 cells, which express nonmutated forms of PIK3CA (Fig. 1D).

To determine the effect of gefitinib on in vivo epithelial-like cSCC growth, samples of SCC10-, SCC16-, SCC24-, and SCC34-PDXs were engrafted in immunodeficient mice, which were treated with vehicle or gefitinib (Fig. 2A). Gefitinib treatment strongly blocked the growth of SCC10-, SCC16-, and SCC24-PDXs (Fig. 2B–D). This effect was dose dependent, and the strongest effect on suppressing tumor growth was observed with the highest doses of gefitinib (75 mg/kg; Supplementary Fig. S6E). This treatment induced a significant regression of tumors with a size close to the ethical endpoint (Supplementary Fig. S4A), coinciding with diminished tumor cell proliferation, as determined by phosphohistone H3 (Ser10) labeling (Supplementary Fig. S4E and S4F). In contrast, SCC34-PDXs responded poorly to gefitinib, although treated tumors grew more slowly than their respective control tumors (Fig. 2E). This primary resistance to gefitinib was maintained after several treatment cycles (GefT SCC34; Fig. 2I; Supplementary Fig. S4B), suggesting that intrinsic resistance to EGFR may be associated with the constitutive activation of PI3K signaling, or with other genetic alterations present in control SCC34-PDXs. Altogether, these results indicate that gefitinib treatment blocks the growth, and even promotes strong remission of patient-derived MD/PD-SCCs, which conserves epithelial differentiation traits in the absence of PIK3CA-activating mutations.

Intrinsic or acquired resistance to EGFR-targeted monotherapy has been described in a substantial percentage of patients with non–small cell lung cancer (NSCLC; 20%–40% of cases) and HNSCC (70%–80% of cases; ref. 28). To determine whether gefitinib responder cSCC-PDXs might eventually acquire resistance, we halted the treatment after tumor growth had become stably blocked (Fig. 2A–D). Despite the dramatic initial response to EGFR inhibition, GefT SCC10-, SCC16-, and SCC24-PDXs regrew after treatment withdrawal, indicating that a subset of surviving/tolerant tumor cells was able to drive tumor relapse. Relapsed tumors were then engrafted in immunodeficient mice, which were treated continuously with gefitinib (Fig. 2A; Supplementary Fig. S4B). A total of 68.4% of relapsed SCC10-PDXs were resistant to EGFR inhibitor after this second treatment cycle (Fig. 2F; Supplementary Fig. S4B). In contrast, only 23.5% and 17.4% of relapsed SCC16-PDXs and SCC24-PDXs, respectively, grew in the presence of gefitinib (Fig. 2G and H; Supplementary Fig. S4B). Further characterization of GefR tumors showed that these exhibited a tumor cell viability and proliferation ratio similar to their respective untreated control tumors (Supplementary Fig. S4C–S4F). Therefore, these results demonstrate that epithelial-like cSCCs initially responding to gefitinib may become resistant to this drug after long-term treatment at a frequency that is dependent on the intrinsic features of each tumor.

Amplification or acquisition of EGFR-activating mutations, downstream effectors, or other tyrosine kinase receptor genes has been associated with resistance to EGFR inhibitors in other tumor types (29). Analysis of RNA-seq data did not reveal the acquisition of EGFR, RAS, or BRAF mutations in GefR SCC10-, SCC16-, or SCC24-PDXs (Supplementary Table S2). A subset of specific mutations appeared in each GefR SCC-PDX model (Supplementary Table S2; Supplementary Fig. S5B), but no common mutation was identified in these three resistant PDXs (Supplementary Fig. S5A). Interestingly, E542K PIK3CA, an activating mutation previously associated with resistance to EGFR inhibitors (30, 31), was identified in two of the three GefR cSCC16 replicates. This mutation was undetectable in parental and nontreated cSCC16 (Supplementary Fig. S5C), suggesting that it is acquired, or that a small subset of SCC16 cells carrying the E542K PIK3CA mutation may be selected by therapy pressure. As a similar PIK3CA-activating mutation was present in intrinsic resistant SCC34-PDXs (Supplementary Fig. S5D), we suggest that activation of the PI3K pathway, among others, may promote survival of epithelial-like cSCC cells upon inhibition of EGFR signaling.

To further characterize the molecular mechanisms involved in EGFR inhibitor resistance, we compared the transcriptomic profiles between tumor cells isolated from control and GefR SCC10-, SCC16-, and SCC24-PDXs (Supplementary Table S1). The expression of a large subset of genes was deregulated in resistant tumors relative to their respective controls (Fig. 3A; Supplementary Table S1). In addition, multiple genes were differentially expressed in control and GefT SCC34-PDXs (Fig. 3A; Supplementary Table S1). Given that control SCC34-PDXs were already resistant to gefitinib, the gene signature of GefT SCC34-PDXs may be related to the tumor cell response to EGFR inhibition, rather than resistance to EGFR inhibitors. The comparison of gene signatures of the cSCC-PDXs with acquired resistance identified only one gene each that was commonly upregulated (ANKEF1) and downregulated (PDZK1; Supplementary Fig. S5E and S5F) in GefR SCC10-, GefR SCC16-, and GefR SCC24-PDXs. Although we cannot discount the possibility that these proteins may play a role in gefitinib resistance; because no specific inhibitors for them are available, we focused on other altered factors or signaling pathways in GefR tumors to identify a possible targeted therapy to overcome gefitinib resistance.

Gene ontology analysis showed that the expression of genes related to cell-to-cell contact, migration, extracellular matrix remodeling, stemness, and EMT was downregulated not only in the GefR cSCC models, but also in GefT SCC34, which may be a consequence of EGFR inhibition (Fig. 3B; Supplementary Fig. S5G). We observed that epithelial-like traits and histopathologic features were similar in control and GefR tumors (keratin production, etc.), and we did no detect any mesenchymal features in GefR tumor cells. These results indicate that resistance was not associated with progression to the mesenchymal state. RNA-seq data identified several signaling pathways that may be deregulated in GefR tumors. Indeed, the expression of FGFR2 and FGF2 was significantly upregulated in GefR SCC10 and GefR SCC16, respectively (Fig. 3B). A more thorough characterization of the expression of FGFRs and ligand by quantitative PCR revealed that the level of FGF2 expression was higher in control SCC34-PDX and SCC10-PDX cells, which showed the strongest trend toward resistance, than in the other gefitinib-sensitive SCC-PDXs (Fig. 3C and F). Furthermore, the expression of different isoforms of FGFR was significantly induced in tumor cells from GefR SCC10 and SCC16 relative to their respective untreated tumors (Fig. 3C and D). Although FGFR and FGF2 expression was not significantly altered in GefR SCC24, these tumor cells strongly expressed FGFR2B and FGFR3 (Fig. 3E). Furthermore, ITGB3, which was associated with EGFR inhibitor resistance in breast cancer cells (32), was one of the 10 most strongly upregulated genes in GefR SCC24 (Fig. 3B). Indeed, ITGB3 could enhance FGFR signaling in GefR SCC24, as reported previously in acute myeloid leukemia cells (33). Similarly, SCC34 cells strongly expressed all FGFR isoforms, which were not significantly altered in response to long-term gefitinib treatment (Fig. 3F).

Our results suggest that autocrine or paracrine activation of FGFR signaling could promote tumor cell proliferation/survival upon EGFR inhibition. In this regard, FGF2 and FGF7 rescued the proliferation of cSCC cells upon acute gefitinib treatment (Supplementary Fig. S6A), and long-term in vitro treatment with increasing doses of inhibitor and FGF2 gave rise to GefR cells (called here iGefR SCC10, iGefR SCC16, and iGefR SCC24 cells; Fig. 4A). EGFR signaling was attenuated in iGefR cells, as phosphorylation of this receptor and ERK1/2 was less strongly induced in response to EGF in iGefR cells than in control tumor cells (Fig. 4B; in the absence of gefitinib and treated with EGF). Gefitinib treatment greatly reduced the levels of EGF-induced pERK1/2 in iGefR SCC16 and SCC24 cells, the effect being more marked than that in iGefR SCC10 cells. The higher level of FGF2 expression detected in SCC10 cells than in SCC16 and SCC24 cells (Supplementary Fig. S2C) suggests that iGefR SCC10 cells could activate the FGFR signaling by an autocrine pathway. In turn, this could account for the higher levels of pERK1/2 observed in iGefR SCC10 cells upon gefitinib treatment than in iGefR SCC16 and SCC24 cells. We suggest that FGFR signaling could be activated by an autocrine pathway to a greater extent in iGefR SCC10 cells than in iGefR SCC16 and SCC24 cells. Accordingly, established iGefR SCC10 cells can grow with gefitinib, but without adding FGF2 to the medium, in contrast to iGefR SCCC16 and SCC24 cells, which are more dependent on added FGF2 to grow in the presence of gefitinib.

Furthermore, tumors generated from iGefR SCC10 cells were resistant to in vivo gefitinib treatment in contrast to those generated from control cells (Fig. 4D).

The expression of FGFR2B and FGFR2C was significantly induced in the three iGefR cell types, whereas FGFR1B and FGFR1C were upregulated in iGefR SCC10 and SCC16. Furthermore, iGefR SCC24 strongly induced the expression of FGFR3 relative to levels in their corresponding control cells (Fig. 4C). Accordingly, iGefR cells responded more strongly to FGF treatment than did control cells, as characterized by the increased phosphorylation of FRS2 downstream effector, which was significantly blocked in response to the FGFR pan-inhibitor, NVP-BGJ398 (Fig. 4E). Taken together, these results indicate that FGFR signaling is induced in GefR SCC cells.

To determine the relevance of FGFR signaling in the regulation of GefR SCC cell proliferation, we analyzed the effect of two FGFR inhibitors, PD173074 (which inhibits FGFR3 and FGFR1) and NVP-BGJ398 (infigratinib; a pan-inhibitor of FGFR; refs. 34, 35) in iGefR cells. We observed that iGefR cells were highly sensitive to FGFR inhibitors, as proliferation was significantly reduced in response to gefitinib and low doses of FGFR inhibitors, whereas proliferation of control cells was inhibited at higher doses of FGFR inhibitors (Supplementary Fig. S6B–S6D). In accordance with the detectable expression of distinct FGFRs in cSCC control cells, the proliferation of control cells was also inhibited in response to the highest doses of NVP-BGJ398, suggesting that FGFR signaling could also contribute at some degree to promote the proliferation of control cells. However, we observed a limited effect of NVP-BGJ398 on growth of control SCC10-PDX compared with that in response to gefitinib (75 mg/kg; Supplementary Fig. S6E). Interestingly, the hypersensitivity of iGefR cells to FGFR inhibitors indicates that activation of FGFR signaling enhances iGefR cell proliferation, which could promote the resistance to EGFR inhibitor.

As FGFR2B was consistently induced in iGefR cells, we tested the effect of FGFR2B knockdown on the iGefR cell proliferation, and on tumor cells isolated from in vivo–generated GefR1 SCC10-PDXs (Supplementary Fig. S6F–S6I). FGFR2 knockdown (sh-FGFR2-1) inhibited the proliferation of GefR cells in the presence of gefitinib, the effect being more marked in iGefR and GefR1 SCC10 cells than in iGefR SCC16 and SCC24 cells (Supplementary Fig. S6G and S6I). These differences may be related to the lower level of induction of FGFR2 expression in iGefR SCC16 and SCC24 cells (Fig. 4C), and/or the involvement of other FGFRs in overcoming the effect of the EGFR inhibition. Furthermore, FGFR2 abrogation significantly blocked the growth of iGefR SCC10 and GefR1 SCC10 tumors in the presence of gefitinib (Fig. 5A and B), concomitantly with a significant reduction in tumor cell proliferation (Fig. 5C). These findings indicate that upregulation of FGFR2-mediated FGFR signaling promotes the proliferation of cSCC10 cells upon gefitinib treatment. To determine the involvement of FGFR1 or FGFR3 in the activation of FGFR signaling and in promoting GefR cell proliferation, we knocked down the expression of each of these receptors in iGefR SCC10, SCC16, and SCC24 cells. A moderate downregulation of FGFRs was obtained using different short hairpin RNA constructs (Supplementary Fig. S6H), which showed some different power to interfere the expression of their respective receptors in each iGefR cells. We found that the knockdown of each FGFR gave rise a similar inhibition rate of iGefR cell proliferation, although some differences were observed between the three analyzed iGefR cells, probably due to the intrinsic differences in these cells (Supplementary Fig. S6I). These results suggest that FGFR1, FGFR2, and FGFR3 could contribute at some degree to the activation of FGFR signaling to promote the proliferation of iGefR cells upon EGFR inhibition, and the use of an FGFR pan-inhibitor is a good strategy to block this signaling pathway (by acting simultaneously on different receptors) and the proliferation of GefR cells.

In addition, the growth of not only GefR SCC10-PDXs, GefR SCC16-PDXs, and GefR SCC24-PDXs, but also of intrinsically resistant SCC34-PDXs was significantly reduced in response to combined gefitinib and NVP-BGJ398 therapy (Fig. 5D), in accordance with the significant reduction of tumor cell proliferation observed in these tumors (Fig. 5E). To determine whether FGFR inhibitor treatment alone was sufficient to block the growth of GefR tumors, GefR SCC10-PDXs and GefR SCC16-PDXs were treated with gefitinib, NVP-BGJ398 pan-inhibitor, or a combination of gefitinib + NVP-BGJ398 (Fig. 5F). Our results demonstrate that NVP-BGJ398 treatment alone had a more limited response than combined gefitinib + NVP-BGJ398 treatment. Indeed, inhibition solely of FGFR signaling was not enough to significantly reduce the growth rate of GefR tumors, indicating that, in the absence of gefitinib, EGFR could be reactivated, thereby promoting GefR tumor growth. Furthermore, the sequential treatment with NVP-BGJ398 for almost 3–4 weeks followed by gefitinib, did not reduce the growth ratio of these GefR tumors (Fig. 5F). Indeed, the best response was obtained when both EGFR and FGFR signaling were inhibited simultaneously (Fig. 5D and F). Overall, our results demonstrate that pharmacologic inhibition of FGFR signaling blocks the growth of GefR cSCCs, even in the presence of PIK3CA-activating mutations, and suggest that a bypass from EGFR to FGFR signaling may be responsible for overcoming the effect of the EGFR inhibitor.

To determine the relevance of FGFR signaling activation in the response of patients with cSCC to EGFR-targeted therapy, we studied the phosphorylation status of FRS2 in a subset of recurrent cSCCs from patients who received paclitaxel plus cetuximab as palliative treatment. Patients were classified as short- or long-term responders when a response of up to or more than 6 months, respectively, was achieved (Fig. 6A). We found that tumor cells from short-term responders exhibited stronger, but more heterogeneous pFRS2 labeling than those from long-term responders (Fig. 6B), with the exception of cSCC-Ctx5, which expressed a strong pFRS2 signal. Although this patient responded for more than 6 months, the disease progressed and the patient died 8 months after treatment had begun. Despite the small number of patients analyzed here, our results indicate that induced FGFR signaling may be associated with a poor response to cetuximab and with reduced patient outcome.

Furthermore, using public databases (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE1029959), we compared the expression of FGFR signaling genes in a cohort of patients with recurrent and metastatic HNSCCs (R/M HNSCC) that were treated with platinum-based chemotherapy plus cetuximab. This cohort had previously been classified as short PFS patients (PFS ≤ 5 months) or long PFS patients (PFS > 5 months; refs. 36, 37). The levels of FGF2 and FGFR1 expression were higher in tumors from short PFS patients than in those from long PFS patients after treatment (Fig. 6C and D). Taken together, these findings indicate that activated FGFR signaling, induced by the overexpression of FGFR and/or ligands, can limit the response of patients with cSCC and R/M HNSCC to EGFR inhibitors.

Clinical trials with patients with recurrent, metastatic, and/or nonresectable cSCC had a variable response ratio (up to 45% of cases) to EGFR-targeted therapies (18, 21). However, the mechanisms responsible for EGFR inhibitor resistance were not known at that time. Here, we generated a subset of cSCC-PDXs to determine the role of EGFR signaling in patient cSCC growth and the response to EGFR-targeted therapy. Characterization of parental samples and derived cSCC-PDXs and primary cells demonstrated that cSCCs could be stratified into epithelial-like cSCCs (WD/MD-SCCs and MD/PD-SCCs), whose tumor cells conserved epithelial differentiation markers, and mesenchymal-like cSCCs (PD/S-SCCs), which contained tumor cells that strongly induced the expression of EMT markers. These findings are consistent with our previous results from an analysis of the expression of epithelial markers and mesenchymal markers in a larger cohort of WD/MD-SCCs and PD/S-SCCs (23). Analysis of the RNA sequences of G2/G3 cSCC-PDXs revealed no mutations in EGF or FGF receptors, KRAS, HRAS, or BRAF in these tumors. In contrast, SCC34-PDXs bore the E545K PIK3CA-activating mutation, which has been linked to EGFR inhibitor resistance in other tumor types (26, 27), as well as G577E MET and G1169R IGF1R mutations, whose relevance to the activity of the respective receptors is unknown.

Levels of EGFR expression and activity in tumor cells were correlated with the extent of expression of epithelial markers. In contrast to mesenchymal-like cSCC cells, epithelial-like cSCC cells exhibited autocrine activation of this signaling, which promoted in vitro tumor cell proliferation and survival. Furthermore, growth was strongly blocked in the epithelial-like cSCC-PDXs without PIK3CA mutations in response to a first cycle of gefitinib, whereas SCC34-PDXs containing the PIK3CA-activating mutation showed primary resistance to treatment.

Despite the strong response to gefitinib in epithelial-like cSCCs, all gefitinib-sensitive tumors progressed after cessation of treatment, as expected from the reversible activity of this EGFR inhibitor (38). However, residual tumors contained gefitinib-resistant/tolerant cells, as indicated by the acquired resistance exhibited by relapsed tumors after subsequent cycles of treatment. The percentage of resistant tumors was different in each cSCC-PDX model, possibly due to the variable frequency of resistant/tolerant cells in initial untreated tumors that were selected upon treatment pressure, or due to the acquisition of mutations in specific genes and/or the activation of signaling pathways that may overcome the EGFR inhibition. A subset of mutations was identified in each resistant cSCC-PDX model, but no common genetic alteration was detected. E542K PIK3CA, which was previously associated with resistance to EGFR inhibitors (30, 31), was identified in some GefR cSCC16 replicates, but not in untreated or in the parental cSCC16 samples of the patient. The fact that it was undetectable in some GefR cSCC16 replicates indicates that other molecular events help promote growth of these patient's tumors upon EGFR inhibition.

Resistance to EGFR inhibitors has been described in patients with lung SCC, HNSCC, and colorectal carcinoma (28). The resistance has been related to mutations in RAS and EGFR genes, loss of PTEN function, activation of the cMET/HGF pathway, and EMT induction (39–42). However, no mutations in EGFR, RAS, RAF, or MET were identified in cSCCs that acquired resistance after long-term gefitinib treatment. In addition, genes associated with stemness and EMT were significantly downregulated in most GefR cSCCs, which conserved their epithelial differentiation traits, as indicated by pathologic analysis, in accordance with the most strongly differentiated phenotype induced by gefitinib in HNSCC cell lines (43). This scenario differs from that reported in EGFR-mutated NSCLCs that are resistant to EGFR inhibitors. In this tumor type, the induction of the EMT program was associated with initially surviving and drug-tolerant cells. This transition occurs in a subset of patients with EGFR-mutant NSCLC who acquire resistance to EGFR inhibitors, either independently, or jointly with genetic resistance mechanisms, such as EGFRT790M (44–46).

Although the mechanisms responsible for EMT induction in tumors with resistance to EGFR-targeted therapy are yet to be identified, differences between the induction of EMT in NSCLC and the maintenance, or even the enhancement of epithelial traits in cSCC and HNSCC tumor cells, could be associated with the plasticity that NSCLC cells exhibit upon therapy pressure. Indeed, several studies demonstrated that a subset of NSCLCs experienced a fundamental histology transformation from NSCLC to small-cell lung cancer (SCLC) at the time of tyrosine kinase inhibitor resistance, whereas the original EGFR mutations were maintained in these cases (44, 46). It is important to highlight that tumor stroma, and consequently the signals provided by the tumor microenvironment, originate from mice in GefR cSCC-PDXs, in contrast to NSCLC patient biopsies that are resistant to EGFR therapy, in which there is a switch to the mesenchymal phenotype. Therefore, one possible scenario may be that NSCLC cells present an epigenetic landscape that predisposes them to plasticity, allowing the switch from an epithelial-like to a mesenchymal-like phenotype in response to tumor microenvironment–derived factors, or even from NSCLC to SCLC histology.

Our results indicate that FGFR signaling pathway is commonly upregulated in GefR cells. cSCC cells strongly induced FGFR expression and activity upon continuous in vitro gefitinib treatment, and these iGefR cells were highly sensitive to FGFR inhibitors. Induction of FGFR signaling may be mediated by different FGFRs that are expressed at different levels in GefR tumor cells and which may contribute to the overall induction of FGFR signaling. Furthermore, pharmacologic inhibition of FGFR signaling, by using a pan-inhibitor that simultaneously inhibits different FGFRs, significantly reduced the growth of not only tumors generated from iGefR cells, but also of in vivo–generated GefR SCC-PDXs, including those carrying PIK3CA-activating mutations. Overall, these findings demonstrate that induction of FGFR signaling overcomes the effect of long-term EGFR inhibition. In accordance with this, recurrent and metastatic cSCCs, which had a short-term response to combined paclitaxel and cetuximab therapy, showed an increased level of FGFR activity. However, further analysis of a large patient cohort is needed to confirm our findings. The meta-analysis of patients with HNSCC treated with chemotherapy and cetuximab showed the short-term response to be associated with an increased expression of FGF2 or FGFR, thereby supporting this finding. Similarly, stronger FGFR signaling was described in lung and breast tumor cells with acquired resistance to EGFR inhibitors (47–49).

Our studies indicate that the best response to block the growth of GefR SCCs was provided by a combined therapy consisting of EGFR and FGFR inhibitors, in contrast to the monotherapy based on NVP-BGJ398, or an alternative treatment with FGFR inhibitors after resistance acquisition and followed by gefitinib treatment. Even after EGFR and FGFR inhibition, tumor growth was not completely blocked, suggesting that other currently unidentified signaling pathways or mechanisms may also contribute to the proliferation of GefR SCC cells. An important question is how the bypass from EGFR to FGFR signaling is induced during long-term gefitinib treatment. The distinct incidence of resistance observed in cSCC10 compared with in cSCC16 and cSCC24 indicates that not all the tumor cells of these tumors were equally capable of switching to FGFR signaling upon therapy pressure, and that their frequency may dictate the overall response to EGFR inhibitors.

The main limitation of this study was the small cohort of patients with cSCC treated with cetuximab and paclitaxel, as well as the limited number of GefR SCC-PDXs analyzed. However, it is important to highlight that very few patients with cSCC were treated with cetuximab and paclitaxel as palliative treatment, after previous authorization by the Pharmacy and Therapeutics Commission of ICO/HUB. On the other hand, a few cSCC-PDXs were available, and we choose those with epithelial-like features to test the in vivo response to gefitinib and to generate GefR cSCCs, after long-term in vivo treatments with gefitinib.

Considered as a whole, our results demonstrate that epithelial-like cSCCs respond optimally to EGFR-targeted therapies, although long-term treatment and PIK3CA-activating mutations give rise to resistance to these treatments, which is overcome by a combined treatment with EGFR and FGFR inhibitors. Further studies are needed to identify markers of cSCC cells that have a strong ability to switch from EGFR to FGFR signaling, to predict the acquisition of resistance to EGFR-targeted therapy.

M. Taberna reports grants, personal fees, and nonfinancial support from Merck and MSD, personal fees from BMS, and personal fees and nonfinancial support from AstraZeneca during the conduct of the study. N. Vilariño reports personal fees from Roche, Boehringer Ingelheim, and from Lilly outside the submitted work. R. Mesia reports personal fees from BMS, MSD, Roche, Merck KGaA, and AstraZeneca outside the submitted work, and travel/accommodations/expenses from Merck KGaA and Bristol Myers Squibb. No disclosures were reported by the other authors.

A. Bernat-Peguera: Conceptualization, formal analysis, investigation, writing-review and editing. J. Navarro-Ventura: Conceptualization, formal analysis, investigation, writing-review and editing. L. Lorenzo-Sanz: Conceptualization, formal analysis, investigation, writing-review and editing. V. da Silva-Diz: Conceptualization, formal analysis, investigation, writing-review and editing. M. Bosio: Data curation, formal analysis. L. Palomero: Data curation, formal analysis, writing-review and editing. R.M. Penin: Conceptualization, resources, formal analysis. D. Pérez Sidelnikova: Resources, formal analysis. J.O. Bermejo: Resources, formal analysis. M. Taberna: Conceptualization, resources, formal analysis, writing-review and editing. N. Vilariño: Resources, formal analysis. J.M. Piulats: Resources, formal analysis. R. Mesia: Resources, formal analysis, writing-review and editing. J.M. Viñals: Resources. E. González-Suárez: Conceptualization, writing-review and editing. S. Capella-Gutierrez: Data curation, formal analysis. A. Villanueva: Resources, formal analysis, methodology. F. Viñals: Conceptualization, resources, formal analysis, writing-review and editing. P. Muñoz: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, writing-original draft, writing-review and editing.

A. Bernat-Peguera, J. Navarro-Ventura, and L. Lorenzo-Sanz each received an IDIBELL Fellowship. V. da Silva-Diz was funded by the Spanish Ministry of Science and Innovation Fellowships. The research of P. Muñoz's group was supported by the Spanish Ministry of Economy and Competitiveness MINECO (SAF2014-55944R and SAF2017-84976R), cofunded by FEDER funds/European Regional Development Fund (ERDF- a way to build Europe) and by the Catalan Department of Health (CERCA, Generalitat de Catalunya; 2017SGR595). The Coordination Node at the Barcelona Supercomputing Center is a member of the Spanish National Bioinformatics Institute (INB), ISCIII-Bioinformatics platform and was supported by grant No. PT17/0009/0001, funded by the Instituto de Salud Carlos III (ISCIII) and ERDF. We thank J. Comas (Universitat de Barcelona-SCT) for technical support with flow cytometry; the CRG Genomics and Bioinformatic Units; the patients enrolled in this study for their participation; the Hospital Universitario Ramón y Cajal, Hospital Virgen de la Salud, and Biobanco del Principado de Asturias and Fundación Instituto Valenciano de Oncología, which are members of the Spanish Hospital Platform Biobanks Network, and the Tumor Bank of the Hospital de Bellvitge for help with human tumor sample collection; and the IDIBELL animal facility service for mouse care.

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.