Epithelial-to-Mesenchymal Transition Defines Feedback Activation of Receptor Tyrosine Kinase Signaling Induced by MEK Inhibition in KRAS-Mutant Lung Cancer

Lung cancer is the leading cause of cancer-related mortality, with a 5-year survival rate of less than 15% across all stages of disease. KRAS is the most frequently mutated oncogene in non–small cell lung cancer (NSCLC), which is detected in 20% to 25% of Caucasian patients with lung adenocarcinoma (1, 2). In the past decade, the discovery of driver oncogenes such as EGFR mutations and ALK rearrangements in lung adenocarcinomas has led to the development and implementation of targeted therapies that block the function of these oncogenes. In contrast, there are no approved targeted therapies for the treatment of KRAS-mutant cancers because direct inhibition of mutant KRAS has proven challenging, despite the development of small-molecule inhibitors that interfere with the localization of KRAS or inhibit the activity of mutant KRAS (3).

Recently, targeting KRAS effector pathways has gained traction as a therapeutic alternative (1, 2, 4). The MAPK pathway is a well-known effector pathway of KRAS; however, targeting this pathway by MEK inhibition results in limited activity in patients with KRAS-mutant lung cancer (5, 6). The lack of effectiveness may be associated with activation of multiple other effectors by KRAS, including the PI3K–AKT and NF-kB pathways (3, 4). In addition, MEK inhibition leads to relief of physiologic negative feedback loops and results in activation of several upstream receptor tyrosine kinases (RTK; refs. 7–9). As such, the combination of MEK inhibitors with IGF1R inhibitors was shown to enhance cell death in KRAS-mutant colorectal cancer (10) and lung cancer (11). Additionally, cotargeting MEK and ERBB3 was shown to be effective in KRAS-mutant lung and colorectal cancers (12). The different RTKs that are activated in KRAS-mutant cancers reflect significant heterogeneity of these tumors, even within the same tissue type. Furthermore, recent integrative analysis of genomic, transcriptomic, and proteomic data identified three subclasses of KRAS-mutant lung adenocarcinoma, supporting the idea of heterogeneity among tumors with the same tissue type (13).

The epithelial-to-mesenchymal transition (EMT) is an essential mechanism in the developmental process and repair of tissue (14). EMT also contributes to the progression of cancer by promoting loss of cell–cell adhesion, leading to a shift in cytoskeletal dynamics. Hallmarks of EMT include the loss of E-cadherin expression and concomitant increase of mesenchymal markers, such as vimentin (15, 16). Emerging evidence suggests that EMT has been shown to associate with primary and acquired drug resistance. For example, EMT predicts sensitivity to EGFR inhibitors in NSCLC (17). In addition, EMT is a cause of resistance to EGFR inhibitors in EGFR-mutant lung cancers (18). Tumors with a mesenchymal phenotype are also associated with aggressive disease and poor prognosis (19, 20).

Herein, we report that MEK inhibition leads to distinct activation of RTKs in KRAS-mutant lung cancers depending on the epithelial or mesenchymal state of the cancer. In epithelial-like KRAS-mutant cancer cells, MEK inhibition upregulates ERBB3, which in turn activates PI3K–AKT and MAPK signaling. In contrast, mesenchymal-like KRAS-mutant lung cancer cells lack ERBB3 expression. Instead, these cells show higher basal expression of FGFR1 protein, and MEK inhibition relieves inhibition of FGFR1 signaling by suppressing Sprouty expression. Importantly, combinatorial inhibition of FGFR1 and MEK in mesenchymal-like KRAS-mutant lung cancers resulted in robust apoptosis in vitro and tumor shrinkage in vivo. These data suggest biomarker-driven combinations of FGFR and MEK inhibitors may be effective for mesenchymal-like KRAS-mutant NSCLC.

We first evaluated whether MEK inhibitor treatment leads to feedback activation of PI3K and MEK–ERK signaling. Two allosteric MEK inhibitors, trametinib and selumetinib, were used to determine the lowest drug concentration that maximally inhibited ERK phosphorylation (Supplementary Fig. S1A and S1B). Following 48-hour treatment with MEK inhibitors, rebound activation of ERK signaling occurred with both drugs, as evidenced by phosphorylation of ERK, whereas the rebound was less significant following treatment with the newer MEK inhibitor trametinib, which also disrupts RAF–MEK complexes (refs. 21, 22; Fig. 1A). Treatment with trametinib in a panel of KRAS-mutant lung cancer cell lines further demonstrated that long-term MEK inhibition induces rebound activation of ERK signaling (Supplementary Fig. S1C). Furthermore, upregulation of AKT phosphorylation was observed in many of the KRAS-mutant lung cancer cell lines. Inhibition of PI3K/AKT or MEK is known to relieve feedback suppression of RTKs (8, 12, 23–25). To determine whether activation of RTK signaling accounts for the rebound activation of ERK and upregulation of AKT signaling following MEK inhibitor treatment, we interrogated global RTK phosphorylation before and after trametinib treatment. The results showed that phosphorylation of ERBB3 and its binding partner ERBB2 were upregulated following trametinib treatment (Fig. 1B); consistently, ERBB3-expressing epithelial-like KRAS-mutant NCI-H358 and NCI-H1573 cells showed clear upregulation of ERBB3 phosphorylation following trametinib treatment (Fig. 1C). Furthermore, immunoprecipitation of p85, the regulatory subunit of PI3K, identified that the interaction of p85 with ERBB3 was upregulated following trametinib treatment (Fig. 1D). SiRNA knockdown or pharmacologic inhibition of ERBB3 by a pan-ERBB inhibitor, afatinib, negated rebound activation of ERK and upregulation of AKT phosphorylation induced by trametinib and resulted in increased apoptosis (Fig. 1E and Supplementary Fig. S1D and S1E). We next determined the effects of the drug combination in vivo. Whereas trametinib monotherapy slowed tumor growth, tumor regression was observed with the combination of afatinib with trametinib treatment (Fig. 1F). Collectively, these results showed that the feedback ERK and AKT activation induced by ERBB3 mitigates the effect of MEK inhibitor in a subset of KRAS-mutant lung cancer cell lines in vitro and in vivo.

Because we found that the MAPK pathway was also reactivated in KRAS-mutant cells that lacked expression of ERBB3, and its reactivation was not negated by afatinib treatment (Fig. 1G), we sought to uncover other distinct signaling pathways leading to reactivation of the MAPK pathway in order to develop effective MEK inhibitor–based targeted therapy for these cancers. Previous studies demonstrated that ERBB3 loss is associated with the mesenchymal phenotype in cancers, including lung cancer (17, 20, 26). We therefore hypothesized that it was mesenchymal KRAS-mutant lung cancers that reactivated the MAPK pathway following MEK inhibition independent of ERBB3.

We first performed a Western blot analysis of our KRAS-mutant lung cancer cell line panel and determined that our cells that expressed ERBB3 were epithelial, whereas those null for ERBB3 were mesenchymal (Fig. 2A). To further study this correlation, we induced EMT via chronic TGFβ1 treatment in the epithelial-like NCI-H358 cells, and upregulated and downregulated genes were analyzed (Fig. 2B; Supplementary Table S1). As expected, ERBB3 and CDH1 (E-cadherin) were strongly suppressed following induction of EMT. Interestingly, we found the receptor tyrosine kinase FGFR1 to be one of the highest upregulated genes in NCI-H358 TGFβ1-treated cells (Fig. 2B). Western blot analysis confirmed that FGFR1 was prominently upregulated following EMT induction, whereas ERBB family proteins EGFR, ERBB2, ERBB3, and ERBB4 had reduced expression (Fig. 2C). We next performed unsupervised hierarchical clustering of 39 KRAS-mutant NSCLC cell lines from the Cancer Cell Line Encyclopedia (CCLE; ref. 27). A total of 2,635 genes, which demonstrated more than a 5-fold expression change among cell lines with median value of 6 or more, were extracted and used for further analyses. These analyses identified two distinct subsets of KRAS-mutant cancer cell lines (Fig. 2D). One subtype, including the NCI-H358 and NCI-H1573 cells, typically expressed ERBB3 and CDH1, consistent with the epithelial-like phenotype. The second group had a mesenchymal phenotype with low CDH1, high VIM (vimentin), and low ERBB3. Furthermore, these cells had significantly higher FGFR1 expression compared to the epithelial-like cell lines. Linear regression analyses demonstrated positive correlations between ERBB3 and CDH1 as well as FGFR1 and VIM. Conversely, inverse relationships between FGFR1 and CDH1 as well as ERBB3 and VIM were found (Fig. 2E and F; Supplementary Fig. S2). Collectively, these data demonstrate an ERBB3 high/FGFR1 low expression pattern in epithelial-like KRAS-mutant lung cancers, contrasting with an FGFR1 high/ERBB3 low expression pattern in mesenchymal-like KRAS-mutant cancers.

Consistent with the gene expression data from CCLE, Western blot analysis of a panel of KRAS-mutant cancer cells identified mesenchymal cells as having higher FGFR1 expression (Fig. 3A). To further determine the relationship between mesenchymal phenotype and FGFR1 expression, we tried to revert EMT to determine how that affected the expression of FGFR1. Knockdown of the EMT transcriptional repressor ZEB1 partially reversed EMT with upregulation of E-cadherin and downregulation of vimentin demonstrated in the TGFβ1-treated (and thus EMT-induced) NCI-H358 cells (Fig. 3B). Indeed, ZEB1 knockdown resulted in suppression of FGFR1 and induction of ERBB3 expression (Fig. 3B), consistent with our other findings. We next investigated whether FGFR1 expression is related to the feedback activation of ERK signaling following MEK inhibitor treatment in mesenchymal-like KRAS-mutant cancers. In NCI-H1792 and LU99 mesenchymal-like KRAS-mutant lung cancer cell lines, while MEK inhibitor treatment strongly suppressed ERK phosphorylation, long-term MEK inhibition resulted in activation of FRS2, an adaptor protein of FGFR (Fig. 3C). Furthermore, immunoblot analyses identified that trametinib did not upregulate phosphorylation of several other RTKs, such as EGFR family proteins, AXL, MET, and PDGFRα and IGF1R adaptor protein IRS1, suggesting that FRS2 may be the primary signal induced by MEK inhibition (Supplementary Fig. S3). To determine if the upregulation of FRS2 phosphorylation is mediated by FGFR1, we treated the NCI-H1792 and LU99 cells with trametinib after FGFR1 knockdown. Indeed, although either treatment with an FGFR inhibitor, NVP-BGJ398 (28), or FGFR1 knockdown modestly downregulated ERK signaling, in the presence of trametinib, FGFR1 knockdown negated trametinib-induced FRS2 phosphorylation and feedback activation of ERK signaling (Fig. 3D; Supplementary Fig. S4). In addition, AKT phosphorylation was upregulated following trametinib treatment, especially in the LU99 cells, which was also negated by FGFR1 knockdown (Fig. 3C and D).

Next, in order to further investigate how MEK inhibition leads to the induction of the FGFR pathway in KRAS-mutant mesenchymal lung cancers, we compared gene expression profiles of the mesenchymal-like KRAS-mutant lung cancer NCI-H1792 cells before and after treatment with trametinib. Interestingly, we found no significant autocrine upregulation of FGF ligands or receptors following trametinib treatment (Supplementary Table S2). The addition of an FGFR1 ligand modestly rescued mesenchymal-like KRAS-mutant lung cancer cells from trametinib-induced growth inhibition, suggesting some, but minor, paracrine contribution from tumor stroma or systemic production (Supplementary Fig. S5). The MEK/ERK pathway can also be regulated by ligand-independent feedback regulation in which downstream proteins transcriptionally regulated by ERK in turn negatively regulate ERK signaling (29). Indeed, trametinib significantly suppressed the expression of inhibitory regulators of the MAPK pathway, including Sprouty proteins (SPRY1, SPRY2, and SPRY4) and MAPK phosphatases (MPKs and DUSPs; Supplementary Table S3). Sprouty proteins antagonize FGFR-mediated MAPK activation by competing with FRS2 for binding to GRB2 and the SOS complex (30, 31). Therefore, we hypothesized that suppression of SPRY or DUSP proteins led to FRS2 activation. We chose to investigate SPRY4 and DUSP6 because these proteins showed the sharpest downregulation following trametinib treatment among proteins in each family (Supplementary Fig. S6). Whereas both SPRY4 and DUSP6 expression were downregulated following trametinib treatment (Fig. 4A), FRS2 phosphorylation was strongly induced by SPRY4 suppression (Fig. 4B). Furthermore, overexpression of SPRY4 protein negated feedback activation of FRS2 following MEK inhibitor treatment (Fig. 4C), resulting in further suppression of ERK phosphorylation and reduction of AKT phosphorylation induced by trametinib and greater induction of apoptosis in LU99 cells (Fig. 4D). Moreover, in low FGFR1-expressing epithelial-like KRAS-mutant lung cancer cells, FRS2 phosphorylation was not induced by trametinib despite SPRY4 downregulation (Fig. 4E). In line with this, ZEB1 knockdown abrogated FRS2 activation following trametinib treatment (Fig. 4F). Collectively, MEK inhibition leads to downregulation of expression of SPRY proteins in KRAS-mutant lung cancers, which relieves suppression of basal FGFR–FRS2 function, leading to reactivation of MAPK signaling in the presence of FGFR1.

To further interrogate the relationship between EMT and FGFR1-mediated ERK reactivation and upregulation of AKT signaling, epithelial and mesenchymal KRAS-mutant lung cancer cell lines were treated with the combination of trametinib and NVP-BGJ398. In NCI-H1792 and LU99 mesenchymal-like KRAS-mutant lung cancer cell lines, the combination of NVP-BGJ398 and trametinib showed a more complete suppression of ERK phosphorylation and improved suppression of AKT phosphorylation compared to trametinib monotherapy, consistent with a key role of FGFR1 in MEK inhibitor–induced feedback (Fig. 5A). The combination of FGFR and MEK inhibition led to profound inhibition of cell survival compared with single-agent pan-ERBB, FGFR, and MEK inhibitors as well as combined pan-EGFR and MEK inhibition (Fig. 5B and C). This was accompanied by greater induction of apoptosis (Fig. 5D and E). On the contrary, the combination of FGFR and MEK inhibition had no effect on downstream signaling in epithelial-like KRAS-mutant cells (Fig. 5F). Accordingly, the combination of FGFR inhibitor with trametinib induced similar levels of apoptosis compared to trametinib monotherapy in the epithelial-like NCI-H358 and NCI-H1573 cells (Fig. 5F). In contrast, in the EMT-induced NCI-H358 cells, the addition of NVP-BGJ398 led to further suppression of ERK phosphorylation compared to trametinib monotherapy and negated AKT activation induced by trametinib (Supplementary Fig. S7A). The combination of trametinib and NVP-BGJ398 or FGFR1 knockdown induced robust apoptosis in these cells (Supplementary Fig. S7B and S7C). Collectively, these data demonstrate MEK inhibition induces distinct RTK activation depending on EMT status, and the activation of the FGFR1–FRS2 pathway is dominantly induced in mesenchymal-like KRAS-mutant lung cancer cells.

Encouraged by the activity of the combination of trametinib and NVP-BGJ398, we expanded the number of mesenchymal-like KRAS-mutant lung cancer cell lines to further investigate the effects of cotreatment with trametinib and NVP-BGJ398. Immunoblot analysis confirmed that an FGFR inhibitor suppressed feedback activation of ERK following trametinib treatment for 48 or 72 hours (Fig. 5A and G; Supplementary Fig. S8A and S8B). In addition, AKT phosphorylation was upregulated following trametinib treatment in many of the cell lines and the combination showed a trend to induce better suppression of AKT compared to trametinib monotherapy (P = 0.08; Fig. 5A; Supplementary Fig. S8A and S8C). Among eleven mesenchymal-like KRAS-mutant lung cancer cell lines, an FGFR inhibitor in combination with a MEK inhibitor induced robust apoptosis in four, modest apoptosis in three, and minimal apoptosis in four of the cell lines, suggesting heterogeneity of response to the therapy (Fig. 5H; Supplementary Fig. S8D). Moreover, apoptosis induced by NVP-BGJ398 with trametinib was significantly correlated with that induced by a PI3K inhibitor with trametinib (Supplementary Fig. S9A and S9B), suggesting that the FGFR1–FRS2 pathway activated by trametinib plays roles in both PI3K and MEK signaling. To exclude the possibility that residual AKT activity was causing resistance to FGFR and MEK inhibition, SW1573 and NCI-H460 insensitive cell lines were treated with NVP-BGJ398, trametinib, and a PI3K inhibitor, GDC-0941 (pictilisib). Although the combination induced complete suppression of both AKT and ERK signaling (Supplementary Fig. S10A), modest apoptosis was observed only in these cell lines (Supplementary Fig. S10B and S10C). Recent studies have demonstrated that an abnormality in the apoptotic machinery, including BCL-XL overexpression by YAP1, and decreased ratios of BIM/BCL-XL and PUMA/BCL-XL underlie intrinsic resistance to a MEK inhibitor with or without a PI3K inhibitor (32, 33). In line with this, the addition of the BCL2/BCL-XL inhibitor ABT-263 (navitoclax) to an FGFR inhibitor and a MEK inhibitor induced apoptosis in the SW1573 cells, suggesting that deregulation of apoptotic proteins may be a cause of the differential sensitivity to FGFR and MEK inhibition (Supplementary Fig. S10D and S10E).

These findings led us to test the efficacy of the combination of an FGFR inhibitor with a MEK inhibitor in vivo. Whereas either NVP-BGJ398 or trametinib monotherapy had minimal effect on tumor growth in the LU99 tumor xenograft model, tumors regressed when treated with the combination of FGFR and MEK inhibitors (Fig. 6A and B). The drug combination was well tolerated over a 4-week treatment period (Supplementary Fig. S11). Pharmacodynamic studies of the drug-treated tumors recapitulated the in vitro results; trametinib partially suppressed ERK phosphorylation; however, it induced FRS2 and AKT phosphorylation as well. The addition of the FGFR inhibitor suppressed feedback activation of FRS2, leading to greater suppression of ERK and AKT, which resulted in downregulation of S6 phosphorylation (Fig. 6C). Moreover, the efficacy of the combination of an FGFR inhibitor with a MEK inhibitor was validated in a second NCI-H23 xenograft model (Fig. 6D).

Lastly, to validate our findings in a clinically relevant setting, we used a patient-derived xenograft (PDX) of KRAS-mutant lung cancer with a representative mesenchymal phenotype identified by the expression of E-cadherin and vimentin (Fig. 6E). Consistent with the in vitro findings and the cell line xenograft mouse models (Fig. 6A and D), the addition of the FGFR inhibitor sensitizes these tumors to trametinib (Fig. 6F), supporting the notion that EMT may serve as an informative biomarker to predict the responsiveness of KRAS-mutant lung cancers to the combination of FGFR inhibition and MEK inhibition.

To determine whether FGFR1 expression was also associated with the mesenchymal phenotype in primary KRAS-mutant lung cancers as it was in our cell line studies, we analyzed RNA sequence expression data of 75 KRAS-mutant lung adenocarcinomas from The Cancer Genome Atlas (TCGA; ref. 34). Using a set of 28 genes that is highly correlated with the EMT process as reported by Kalluri and Weinberg (35), hierarchical clustering analysis revealed that KRAS-mutant cancer could be classified into two groups based on epithelial and mesenchymal markers (Fig. 7A). Importantly, mesenchymal-like KRAS-mutant tumors demonstrated significantly higher FGFR1 expression and significantly lower ERBB3 expression compared to epithelial-like KRAS-mutant cancer (P < 0.001 for FGFR1 and P = 0.03 for ERBB3, respectively). These data confirm that FGFR1 expression is high, specifically in the mesenchymal-like KRAS-mutant cancer subgroup.

MAPK is the best-characterized downstream pathway of KRAS; however, MEK inhibitor monotherapy demonstrates only modest efficacy in vitro and in vivo (5, 6, 9). These data are reminiscent of BRAF inhibitors in BRAF-mutant colorectal cancer: The subsequent identification and drugging of the EGFR pathway as a feedback activator of ERK signaling has led to clinical trials of combination BRAF and EGFR inhibitors (36, 37). In this study, we have shown that resistance to MEK inhibition in KRAS-mutant lung cancers is caused by the feedback activation of RTK signaling. Although feedback activation is mediated by ERBB3 in epithelial-like KRAS-mutant cancer, the FGFR1–FRS2 pathway plays a critical role in the feedback activation of MAPK signaling in mesenchymal-like KRAS-mutant cancers. Importantly, the combination of trametinib with an FGFR inhibitor induced robust apoptosis in vitro and tumor regressions in vivo in mesenchymal-like KRAS-mutant lung cancers, suggesting that the biomarker-directed combination of these two drugs would benefit this subset of patients with KRAS-mutant lung cancer.

Reactivation of ERK signaling has played critical roles in primary and acquired resistance to targeted therapies, even outside of KRAS-driven cancers (36–41). It has been demonstrated that activation of MAPK signaling induces transcription of negative feedback genes, including SPRYs and DUSPs (9, 42). Interestingly, this pathway is active in benign tumors, presumably acting as a fail-safe negative feedback signaling program enabling the cell to protect itself from malignancy (42). In our study, we found that this pathway was also active in the malignant KRAS-mutant lung cancers, and relieving this feedback through MEK inhibitor treatment leads to the inhibition of the SPRY proteins. In mesenchymal KRAS-mutant cancers, this is sufficient to activate the FGFR1 pathway, as FGFR1 appears “primed” in these cancers via high levels of expression. Therefore, in mesenchymal-like KRAS-mutant cancer cells, FGFR1 activity is suppressed by the active KRAS–MEK signal, perhaps part of an initial effort by the cell to prevent malignancy; the FGFR1 signal is transduced once negative feedback is relieved by trametinib, and only blocking both the MEK/ERK signal and the FGFR1 signal allows for sustained MEK/ERK inhibition and subsequent apoptosis and tumor regressions.

In mesenchymal-like KRAS-mutant lung cancer cells, SPRY4 plays a primary role in the induction of FRS2 phosphorylation following trametinib treatment. Although SPRY proteins have been shown to be involved in ERK-dependent feedback inhibition of EGFR family proteins in BRAF-mutant melanomas and colon cancers, and FGFR family proteins in gastrointestinal stromal tumors (GIST; refs. 37, 38, 41), the role of SPRYs and DUSPs is cell type– and context-dependent (43). SPRY4 has been shown to suppress FGF pathways, but it has not been shown to inhibit EGF signaling, in contrast to the involvement of other SPRY family proteins (30). Intriguingly, FRS2 phosphorylation was not induced in epithelial-like KRAS-mutant cancer cells expressing low FGFR1 protein, despite downregulation of SPRY4 expression by trametinib. These results support the notion that Sprouty proteins are downregulated following trametinib treatment in KRAS-mutant lung cancers; however, only in the high FGFR1-expressing mesenchymal lung cancers does this lead to FGFR1-dependent FRS2 phosphorylation and subsequent reactivation of the MAPK pathway, resulting in resistance to MEK inhibitors (Fig. 7B). SPRY4 expression has been shown to be positively regulated by the ERK pathway previously (43). In line with this, a RAS-responsive element was located between nucleotide positions −69 and −31 in the SPRY4 promoter region, and transfection of dominant-negative RAS N17 significantly inhibited promoter activity of the gene by up to 70% in A549 cells (44). However, the precise mechanism of SPRY4 regulation remains to be determined.

We identified that MEK inhibition led to upregulation of PI3K–AKT signaling in many of the KRAS-mutant lung cancer cell lines despite their epithelial or mesenchymal status. In epithelial-like KRAS-mutant lung cancer cells, we showed that upregulation of PI3K–AKT signaling was mediated through ERBB3 by upregulation of protein expression and phosphorylation, which led to bound ERBB3:p85, the regulatory subunit of PI3K. Selumetinib was shown to induce degradation of c-MYC, which relieves transcriptional repression of ERBB3 in breast cancer and KRAS-mutant lung and colon cancers (8, 12). Furthermore, in BRAF-mutant cancers, a BRAF inhibitor or trametinib upregulates ERBB3 by inducing the expression of the transcription factor FOXD3, or by decreasing the expression of the transcription repressors C-terminal binding protein 1 and 2 (45, 46). We also demonstrated that FGFR1 mediated the upregulation of PI3K–AKT signaling in mesenchymal-like cells; PI3K could be activated by FGFR1–FRS2, in which several tyrosine residues of FRS2 serve as docking sites for proteins, such as GRB2 and GAB1, allowing recruitment of PI3K. Alternatively, PI3K activation may be mediated by RAS, whose activation is upregulated following SPRY4 suppression.

We found ERK suppression–mediated relief of negative feedback of FGFR1 signaling occurred independently of FGF ligands and receptors. Recently, FGFR was also shown to mediate reactivation of the MAPK pathway and attenuate the antiproliferative effects of imatinib in gastrointestinal stromal tumors (41). In that study, whereas treatment with FGF ligands reduced the antitumor activity of imatinib in GIST cell lines, no significant alterations in the expression of FGF ligands and receptors were observed after long-term imatinib treatment. Our results demonstrated only modest induction of trametinib resistance by the FGF ligand. These results indicate that relief of ERK-dependent feedback may play important roles in the reactivation of ERK in KRAS-mutant cancers independently of ligand-mediated activation.

Although our study identified subgroups of KRAS-mutant lung adenocarcinoma based on EMT status and identified a distinct therapeutic approach corresponding to each subgroup, other factors affecting heterogeneity of KRAS-mutant lung cancer could contribute to intrinsic resistance. In fact, we observed heterogeneity within each subgroup. For example, the level of ERBB3 and FGFR1 expression was variable among tumors (Fig. 7A). This heterogeneity may be at least partly attributed to coexisting genomic alterations. Recent bioinformatic analysis by Skoulidis and colleagues identified that KRAS-mutant lung cancer can be classified into three subgroups by co-occurring genetic alterations in STK11, TP53, and CDKN2A/B (13). In line with this finding, concurrent loss of LKB1 is a cause of resistance to the combination of selumetinib with docetaxel or a PI3K inhibitor in a KRAS-mutant lung tumor model, likely through activated AKT and/or SRC pathways (47, 48). Although we noted that TP53 or LKB1 status did not affect the level of apoptosis induced by NVP-BGJ398 and trametinib (Supplementary Fig. S12) and that there was no relationship between EMT and subgroups classified by Skoulidis and colleagues (Supplementary Fig. S13; ref. 13), it remains likely that other tumor-specific mutations are affecting drug response.

Another important consideration is the differential activation of effector pathways by mutant KRAS other than PI3K and MEK in the tumors: For instance, KRAS G12C and G12V mutants harbor increased signaling through the RAL effector pathway and decreased signaling through AKT, whereas KRAS G12D activates both PI3K and MEK signals (49). However, in our study, apoptosis induced by FGFR and MEK inhibitors was not correlated with the site of KRAS amino acid substitutions (Supplementary Fig. S14). Lastly, we noted a marked difference in the amount of apoptosis induced by PI3K/MEK inhibitors or FGFR/MEK inhibitors in the different models (Fig 5H; Supplementary Fig. S9B). This suggests that the apoptotic machinery plays a role in the resistance of KRAS-mutant NSCLCs to FGFR/MEK inhibitors (33). In line with this, differences in the balance of proapoptotic and antiapoptotic BCL2 family proteins could lead to the insensitivity of the combination in SW1573 cells, and, more generally, may underlie heterogeneity in the apoptotic response to FGFR/MEK inhibitor combination therapy in KRAS-mutant NSCLCs. These heterogeneities among KRAS-mutant lung cancer indicate the need to identify other targets to combine with MEK inhibitors.

Other studies have identified factors influencing the response of KRAS-mutant cancers to targeted inhibition. For instance, YAP1 was recently identified as being able to counteract loss of mutant KRAS in cell lines and mouse models, including lung cancer, by activating the FOS transcription factor and inducing expression of mesenchymal genes (50, 51). Interestingly, in the same study that identified YAP1 as a survival factor following mutant KRAS suppression (50), members of the FGF family were also identified as genes that rescue loss of viability induced by KRAS knockdown. A second screen also identified FGFR1 as a kinase that was able to rescue KRAS suppression in KRAS-dependent cells (50). In addition, FGFR1 has been demonstrated to induce EMT in embryonic development and prostate cancer models (52, 53). These results invoke a link between FGFR1 expression, EMT, and KRAS dependency, and are consistent with some of our findings in this study.

In conclusion, we report that the FGFR1–FRS2 pathway mediates feedback activation of MAPK following MEK inhibition in mesenchymal-like KRAS-mutant lung cancer. Our findings support clinical trials with stratification of patients with KRAS-mutant lung cancer based on EMT status in order to target the mesenchymal tumors with this novel therapy.

The lung cancer cell lines A549, NCI-H358, NCI-H1792, NCI-H23, SW900, and SW1573 were purchased from the ATCC. SK-LU-1, Calu-1, Calu-6, and NCI-H460 were provided by Takashi Takahashi (Nagoya University, Japan). RERF-LC-AD2, LU-65, and LU-99 cells were obtained from the Japanese Cell Research Bank (Osaka, Japan). HCC2108 cells were obtained from the Korean Cell Line Bank (Seoul, South Korea). The NCI-H1573 and NCI-H2030 cells were provided by Massachusetts General Hospital Cancer Center. Cells were cultured in RPMI-1640 (Invitrogen) with 10% FBS. Characteristics of cell lines used in this study were summarized in Supplementary Table S4. Cells were obtained between 2012 and 2016. Experiments using A549, SW900, LU-65, and HCC2108 cells were done within 6 months from the acquisition of these cells from cell banks. All other cell lines were tested and authenticated by short tandem repeat (STR) analysis with GenePrint 10 System (Promega) by the Japanese Cell Research Bank at the time of submission. Cells were regularly screened for Mycoplasma using a MycoAlert Mycoplasma Detection Kit (Lonza). NVP-BGJ398, GDC-0941, ABT-263, selumetinib, afatinib, and trametinib were obtained from Active Biochem. Compounds were dissolved in DMSO to a final concentration of 10 mmol/L and stored at −20°C when not in use.

Lysates were prepared using Cell Lysis Buffer (Cell Signaling Technology). The procedure for Western blotting was as previously described (10). Antibodies used in this study are given in Supplementary Table S5.

Cells were seeded into 6-well plates at a density of 1 × 10⁵ to 2 × 10⁵ cells/well. Twenty-four hours later, cells were transfected with 20 nmol/L siRNA against FGFR1 (Dharmacon), DUSP6 (Dharmacon), ZEB1 (Dharmacon and Santa Cruz Biotechnology), SPRY4 (Ambion and Dharmacon), or ERBB3 (Invitrogen), or Stealth RNAi-negative control low GC Duplex #3 (Invitrogen) using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instructions.

The SPRY4 expression construct was obtained from GE Dharmacon. Preparation of lentivirus and infections was performed as previously described (10).

Cells were seeded at approximately 30% to 40% confluence in 6-well plates. After overnight incubation, the media were changed and DMSO or the indicated drugs were added. After 72 hours, floating cells in media and adherent, trypsinized cells were collected in a single tube. Cells were pelleted and washed once with PBS. Apoptotic cells were stained with Annexin V using Annexin V: PE Apoptosis Detection Kit I and assayed on an BD Accuri C6 flowcytometer (BD). All experiments were performed in triplicate.

Suspension of 5 × 10⁶ cells was injected subcutaneously into the flanks of 6-to-8-week-old male nude mice (Clea). The care and treatment of experimental animals were in accordance with institutional guidelines. Mice were randomized (n = 8 in LU99 xenograft and 5 in NCI-H23 and NCI-H358 xenograft) once the mean tumor volume reached approximately 230 to 300 mm³. Drugs were administered once daily by oral gavage. NVP-BGJ398 was dissolved in acetic acid/acetate buffer pH 4.6/PEG300 1:1; trametinib and afatinib were dissolved in 7% DMSO, 13% Tween 80, 4% glucose, and HCl equivalent molar concentration to each drug. Mice were monitored daily for body weight and general condition. Tumors were measured twice weekly using calipers, and volume was calculated using the following formula: length × width² × 0.52. According to institutional guidelines, mice were sacrificed when their tumor volume reached 1,000 mm³.

For the PDX study, a xenograft model from a 53-year-old female with KRAS G12D–mutant lung cancer was purchased from the Jackson Laboratory. Female NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice at 6 to 8 weeks of age were engrafted with tumor fragments at passage P4 at The Jackson Laboratory. Then, mice were transferred to the animal facility at Virginia Commonwealth University and randomized once the mean tumor volume reached approximately 300 mm³.

RNA was isolated using the RNeasy kit (QIAGEN) following the manufacturer's recommendations. The microarray analysis was performed using the SurePrint G3 Human Gene Expression 8 × 60K v2 Microarray Kit (Agilent Technology). The microarray accession number in GEO is GSE79235.

Gene expression data of lung cancer cell lines were obtained from the CCLE (27). Out of 187 lung cancer cell lines, 39 NSCLC cell lines with KRAS mutation were extracted for further analysis based on the information obtained from the CCLE (CCLE_sample_info_file). Unsupervised hierarchical clustering was performed on 39 KRAS-mutant NSCLC cell lines using Cluster 3.0 program (54) to perform clustering and JAVA TreeView program (55) to visualize the results. A total of 2,635 genes were used for the analysis, which demonstrated more than a 5-fold change among cell lines with a median expression value of 6 or more. Three distinct subclusters were extracted, which included the marker genes ERBB3, CDH1, FGFR1, and VIM.

Messenger RNA profiling for 230 resected lung adenocarcinomas was obtained from the TCGA (34). Out of 230 adenocarcinoma samples, 75 samples with KRAS mutation were extracted for further analysis based on the clinical information. We performed unsupervised hierarchical clustering of 75 adenocarcinoma samples with KRAS mutation based on the 28 genes, which were listed as EMT-related genes by Kalluri and Weinberg (35). Expression data for two additional genes, FGFR1 and ERBB3, were extracted and ordered based on the order of hierarchical clustering.

Linear regression analyses and Student t tests were performed where indicated. Linear regression analyses were performed using r software and the slope was considered significantly non-zero whenP < 0.05. For Student t tests, populations were considered significantly different at P < 0.05. All statistical tests were two-sided.

Mouse xenograft experiments were approved by the ethical committee on the Institute for Experimental Animals, Kanazawa University Advanced Science Research Center (Kanazawa, Japan). The PDX experiment was approved by the Virginia Commonwealth University Institutional Animal Care and Use Committee (IACUC protocol #AD10001048).

H. Ebi reports receiving a commercial research grant from AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Ebi

Development of methodology: H. Kitai, H. Ebi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Kitai, H. Ebi, K.V. Floros, Y. Adachi, A.C. Faber

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Kitai, H. Ebi, S. Tomida, A.C. Faber

Writing, review, and/or revision of the manuscript: H. Ebi, Y. Adachi, M. Nishimura, A.C. Faber

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Kitai, H. Ebi, H. Kotani, S. Yano

Study supervision: H. Ebi, M. Nishimura, S. Yano

Other (provision of key reagents and expertise): S. Oizumi

This work is supported by Grants-in-Aid for Cancer Research 26830105 (to H. Ebi) and 21390256 (to S. Yano); the Scientific Research on Innovative Areas “Integrative Research on Cancer Microenvironment Network” (22112010A01; to S. Yano and H. Ebi); a Grant-in-Aid for Project for Development of Innovative Research on Cancer Therapeutics (P-Direct) from the Ministry of Education, Culture, Sports, Science, and Technology (MEXT) of Japan (to S. Yano); an AstraZeneca R&D Grant Award (to H. Ebi); and a research grant for developing innovative cancer chemotherapy from the Kobayashi Foundation for Cancer Research (to H. Ebi). A.C. Faber is supported by an NCI Transition Career Development Award (K22CA175276), an American Lung Association Lung Cancer Discovery Award, and the George and Lavinia Blick Research Fund.

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.