Cell-Autonomous and Non–Cell-Autonomous Mechanisms of Transformation by Amplified FGFR1 in Lung Cancer

The discovery of genomic alterations in kinase genes has changed the clinical care of patients with lung adenocarcinomas bearing such alterations (1, 2). However, the landscape of genome aberrations differs dramatically between the most common lung cancer subtypes lung adenocarcinoma, squamous cell lung cancer, and small-cell lung cancer. Only a few therapeutically tractable genome alterations have so far been found in squamous cell lung cancer (3, 4) and small-cell lung cancer (5). Of these, amplifications of 8p12 (FGFR1, WHSC1L1) and mutations of DDR2 have been associated with preclinical sensitivity to kinase inhibition (6–9). Furthermore, early signs of clinical activity of fibroblast growth factor receptor (FGFR) inhibitors in FGFR1-amplified small and squamous cell carcinomas underscore the validity of such approaches (10, 11). Unfortunately, though, it is presently unclear which of the tumors bearing FGFR1 amplifications are also sensitive to FGFR inhibition.

The FGFR family consists of four receptor tyrosine kinases, which are common targets of deregulation by translocation, point mutation, and amplification in cancer (12–14). FGFRs are activated via 22 different FGF ligands resulting in downstream FRS2 and extracellular signal-regulated kinase (ERK) phosphorylation (13); their signaling is modulated endogenously by multiple negative intrinsic feedback loops (15–17). Furthermore, alternative splicing of FGF receptors mediates different responses to FGF ligands in epithelial (expressing IIIb splice variants) and mesenchymal tissues (expressing IIIc splice variants) in development (18). In particular, alternative splicing of exon 8 determines the difference between epithelial and mesenchymal variants of FGFR1; the mesenchymal FGFR1-IIIc-β variant differs from full-length FGFR1-IIIc-α by skipping exon 2 (Ig1 loop; ref. 12, 19).

Several recent studies described the 8p12 amplicon (FGFR1, WHSC1L1) as recurrently amplified in lung cancer; the amplicon spans approximately 10 Mbps and contains about 50 genes in total (5, 6). Here, we sought to characterize in detail the structure of the 8p12 amplicon to identify mechanisms of oncogenic transformation induced by amplified FGFR1 and to determine modulators of FGFR dependency in lung cancer.

In lung cancer, amplification of 8p12 occurs most frequently in the squamous cell and small-cell subtypes (5–7). We analyzed single-nucleotide polymorphism (SNP) array data (20–23) of 306 surgically removed squamous cell carcinoma specimens that were generated as part of a large international lung cancer genomics consortium project (24). Most samples revealed multiple individual broad alterations (termed macro lesions, size ≥ 20 Mbp), including gains of entire chromosomal arms (36% of copy-number segments), so that focal copy-number alterations were masked (Fig. 1A left). We therefore developed a novel approach, termed Focal Amplification Peak Purification (FAPP), which smoothens broad alterations and thus extracts the primary focal amplification peak (Fig. 1). We subsequently performed a significance analysis (using GISTIC; ref. 25) on this purified dataset, resulting in 16 regions of recurrent focal amplifications affecting a total of 206 genes (Fig. 1A, right, 1B and C, and Supplementary Fig. S1). We identified highly focal and significant recurrent amplification peaks for 3p11 (containing EPHA3) and 12p15 (containing the FGFR adaptor gene FRS2), not previously described in squamous cell lung cancer (Fig. 1D). The 12p15 locus (FRS2) included 8 genes and was amplified at high amplitude (copy number ≥4) in 2% of all cases (Fig. 1A and D bottom). Because FRS2 encodes the central adaptor protein downstream of FGFRs (15), FRS2- amplified lung tumors may depend on signaling downstream of FGF receptors. Overall, FAPP dramatically reduced the number of genes (n_Gene) that occurred in broad amplification peaks (Fig. 1B and C), such that in particular the 3q26 and 8q24 amplicons gained focality (Fig. 1A).

As an independent validation of our findings, we applied the same computational approach to another cohort of squamous cell lung cancer specimens [n = 299, The Cancer Genome Atlas (TCGA); ref. 3] and obtained similar results (Supplementary Fig. S2A). In the TCGA cohort, we also observed a marked heterogeneity of the 8p12 amplicon (Supplementary Fig. S2B). Furthermore, we found similar amplification patterns of EPHA3 and FRS2 (Fig. 1D, right). Thus, the 8p12 amplicon in squamous cell lung cancer is profoundly heterogeneous on the genomic level; FGFR1 was not even included in the amplicon of several tumor specimens.

We next assessed co-occurrence of amplification events and performed hierarchical clustering of purified copy-number data (26). Using this approach, we were able to sort 210 of 306 squamous cell lung cancer primary tumor samples (69%) into clusters (Supplementary Fig. S3). We found that 8p12 (FGFR1) clustered together with the 11q13 (CCND1, FGF4, FGF19) amplicon in some cases, a finding that has been recently associated with FGFR inhibitor sensitivity in the presence of beta-klotho expression (27). Most of the amplified regions formed single clusters [e.g., 3q26 (SOX2), 4q12 (KIT, PDGFRA), 8q24 (MYC)] or were split into minor subclusters [7p11 (EGFR), 12q15 (FRS2), 19q12 (CCNE1)]. The 8p12 (FGFR1, WHSC1L1) amplicon was, however, identified as the most heterogeneous alteration, covering 5 segments and forming 4 subclusters (Supplementary Fig. S3).

We subsequently plotted the segmented copy-number data sorted by the genomic coordinate of their amplification center in a 15-Mbp region. This approach revealed the full extent of the heterogeneity of the 8p12 amplicon (Fig. 1C and Supplementary Figs. S2B, S3, and S4); the 8p12 amplicons of only 28% of all specimens centered on FGFR1 (Supplementary Fig. S4). In marked contrast, 90% of the amplifications of 7p11 were centered on EGFR and displayed a homogeneous amplification pattern (Fig. 1C, bottom, and Supplementary Figs. S3 and S4). Similarly, CCND1, SOX2, and MYC were each in the center of the respective amplicon in the vast majority of the cases (Supplementary Fig. S4).

Finally, we analyzed chromosomal regions bearing high copy-number fluctuations that indicate the presence of genomic breakpoints. Accordingly, 8p12-amplified samples were frequently found to harbor chromosomal breakpoints in a 10-Mbp region that contained the FGFR1 gene. In addition, breakpoints of the 8p12 locus occurred in non–8p12-amplified samples, too. In contrast, the homogeneous amplifications of 3q26 (SOX2), 7p11 (EGFR), and 11q13 (CCND1) revealed chromosomal breakpoints only within a 5-Mbp range around the respective oncogene (data not shown). Conclusively, this data emphasizes that FGFR1 lies in a genomic locus with a highly heterogeneous pattern of amplification.

In summary, the 8p12 amplification pattern exhibits striking differences from other recurrent amplicons in squamous cell lung cancer; only a fraction of the amplified samples harbor FGFR1 in the epicenter of the amplicon. Thus, beyond the absolute copy-number amplitude, the geographic amplicon extension may be biologically relevant for diagnostic purposes.

Because many of the FGFR ligands are present in the tumor microenvironment of malignant tumors (28), we next sought to determine whether FGFR1-amplified lung tumor cell lines are dependent on ligand binding. To this end, we analyzed the genomic features of 148 lung cancer lines to identify cell lines that capture the genomic pattern that we observed in the primary tumors (Supplementary Fig. S5). We tested in particular for the presence or absence of genome alterations that occur specifically in different lung cancer subtypes (24). Because only few FGFR1-amplified cell lines exist that are of squamous cell histology (HCC95 and H520 in this study), we additionally included the 8p12-amplified lines H1581 (large-cell) and DMS114 (small-cell; Supplementary Fig. S5). We note, however, that large-cell carcinomas can be mostly reclassified to the other lung tumor subtypes (24). Because FGFR1 amplifications are almost entirely restricted to squamous cell and small-cell carcinomas (3, 5, 6, 29, 30), we postulate that these cell lines are valid models of FGFR1-amplified lung cancers (Supplementary Fig. S5). Furthermore, the two FGFR-dependent cell lines, H1581 and DMS114, both include FGFR1 in the main peak of the amplicon.

We found that H1581 cells showed high basal activation of the FGFR1 pathway, which was completely antagonized by FGFR inhibition (Fig. 2A). Basal levels of phospho-FGFR (p-FGFR) were lower in the other FGFR inhibitor-sensitive line, DMS114 (Fig. 2A), and phospho-ERK (p-ERK) levels were enhanced by the addition of FGFs 1, 2, and 4. Both cell lines exhibited high levels of FGF secretion under steady-state conditions and increased ligand secretion under serum starvation (Fig. 2B). Thus, FGFR-dependent lung tumor cells may be sustained in their growth through autocrine and/or paracrine FGFR activation. High baseline levels of p-ERK were unchanged under FGFR inhibition and starvation in the NRAS-mutant cell line HCC15; in FGFR1-amplified but inhibitor-insensitive H520 cells, p-ERK could be activated by FGFs 1, 2, and 4, but p-FGFR levels were unaffected by the addition of inhibitor (Supplementary Fig. S6). Thus, FGFR1-amplified lung cancer cells display a variable degree of autoactivation of FGFR signaling, which might be due to autocrine secretion of FGFs in the most inhibitor-sensitive cell lines, H1581 and DMS114 (Fig. 2B).

We therefore sought to determine formally whether secretion of extracellular ligands was required for the viability of FGFR-dependent cells (Supplementary Fig. S7). To this end, we ectopically expressed FGFR1 mutants, which have reduced affinity to FGF ligands (Fig. 2C, extracellular domain; ref. 31) in cis with the V472M “gatekeeper” mutation that induces resistance to FGFR inhibition (6). As expected, ectopic expression of the V472M gatekeeper mutation alone resulted in resistance to FGFR inhibition in H1581 cells (Fig. 2D). We next cloned the gatekeeper mutant V472M together with the extracellular FGFR1β mutant R161Q (Fig. 2C and D), which is analogous to the R251Q mutation of FGFR2. This mutation strongly interferes with ligand binding to FGFR2 (31). We further predicted five other extracellular mutation sites in the Inter–IgG2–IgG3 domain (ref. 32; L76T, A78L, K83E, D157N, D193N) to potentially interfere with ligand binding (Fig. 2D). The combination of the gatekeeper mutant V472M with any of the extracellular mutants K83E, D157N, and D193N did not alter the resistance to inhibitor treatment (Fig. 2D). In contrast, expression of either L76T or R161Q abrogated the resistance induced by the V472M mutation expressed in cis. However, increasing doses of recombinant FGF-2 restored inhibitor resistance induced by V472M in these cells (Fig. 2D). Thus, exogenously added ligands overcome the reduction in ligand affinity induced by L76T or R161Q. As a consequence, ERK signaling of the double mutant was reduced by FGFR inhibition and restored by high doses of FGF-2 (25 ng/mL, Fig. 2E; ref. 31). As expected, under FGFR inhibition by PD173074, FGF-2 was insufficient to reactivate ERK signaling in the case of wild-type FGFR1β (Fig. 2E). These experiments support the notion that autocrine FGFR activation is required for survival of H1581 cells that exhibit focal amplification of FGFR1.

Finally, we tested whether ligand dependency of FGFR1-amplified cells was also relevant for tumor formation in vivo. To this end, we applied adenoviruses expressing a soluble version of the extracellular domain of FGFR fused to the immunoglobulin heavy chain (AdsFGFR; ref. 33) that competes with cellular FGFRs for soluble ligand (34). Subcutaneous injection of these viruses inhibited tumor formation in H1581 xenografts (Fig. 2F and G), but not the development of KRAS-mutant A549 tumors (Fig. 2F and G). Treatment with AdsFGFR, but not empty-vector virus, delayed a gain in weight in both the mice bearing H1581 (P = 0.04) and A549 (P = 0.012) tumors, thus suggesting that FGF signaling is required for body weight maintenance (35). Efficacy of viral infection was confirmed by adenovirus-specific PCR from fixed liver tissue of all treatment groups.

In summary, we provide evidence that FGFR1-amplified and inhibitor-sensitive H1581 cells depend on ligand binding in vitro and in vivo.

As a next step, we sought to test whether wild-type FGFR1 was oncogenic when overexpressed and analyzed the oncogenic phenotype of NIH3T3 cells ectopically expressing FGFR1 in soft agar assays. Whole-transcriptome sequencing (RNAseq) of six primary FGFR1-amplified squamous cell lung cancer tumors as well as four amplified cancer cell lines (Supplementary Fig. S8A) revealed that mesenchymal splice variants of FGFR1 were predominantly expressed in the FGFR inhibitor-sensitive cell lines. We therefore cloned these splice variants (FGFR1-IIIc-α, FGFR1-IIIc-β) from H1581 cells and transduced NIH3T3 cells with these variants of FGFR1 either alone or together with six additional genes (REL, SOX2, MYC, CCND1, DYRK1B, AKT2) with a possible role in squamous cell lung cancer biology (Fig. 3A and Supplementary Fig. S9). The latter genes are located in or close to recurrent amplicons in this lung tumor subtype. Both FGFR1 variants reproducibly induced mild transformation of NIH3T3 cells to anchorage-independent growth (q = 8 × 10⁻⁹; Fig. 3A and B). In our hands NIH3T3 cells did not survive transduction with MYC alone. However, transduction of NIH3T3 cells with MYC and FGFR1 (q = 2 × 10⁻⁵) was strongly oncogenic as determined by the number and size of colonies in soft agar (Fig. 3B).

Similar to FGFR-dependent H1581 cells (6), treatment with the FGFR inhibitor PD173074 induced apoptosis in these FGFR1–MYC cotransduced NIH3T3 cells, but not in cells expressing FGFR1 alone (Fig. 3C). Thus, FGFR1-amplified cells coexpressing MYC may be more susceptible to FGFR inhibition, which has been similarly reported for FGFR2-mutant breast cancer (36).

Injection of NIH3T3 FGFR1-IIIc-α and -β cells into nude mice led to palpable subcutaneous tumors after a median of 20 days (Fig. 3D, top, and Supplementary Fig. S10A). HEK293 cells, transduced with FGFR1, similarly induced subcutaneous tumors in vivo (Supplementary Fig. S10B), and intravenous injection of NIH3T3 FGFR1α cells led to tumor growth in the lungs (data not shown). Treatment with the FGFR inhibitor BGJ398 (15 mg/kg, q.d.) repressed tumor growth of NIH3T3 cells expressing either of the mesenchymal FGFR1 splice variants (Fig. 3D and Supplementary Fig. S10A). Thus, the catalytic activity of FGFR1 was required for tumor formation in vivo. However, FGFR inhibition by BGJ398 did not induce tumor shrinkage in tumors expressing FGFR1 alone. In contrast, this treatment led to regressions of tumors coexpressing FGFR1 and MYC (Fig. 3D; P < 0.001).

Of note, the tumors expressing FGFR1 alone also exhibited low nuclear expression levels of MYC (Fig. 3E). However, MYC was expressed at much higher nuclear levels in the double-transduced cells, which was subject to FGFR-dependent regulation (Fig. 3E and Supplementary Fig. S11). Thus, FGFR1-expressing tumors upregulate MYC in vivo, but only very high levels of MYC expression are likely to govern susceptibility to FGFR inhibition.

Supporting the notion that MYC may interplay with FGFR1 signaling, we found it to be strongly regulated by FGFR1 in the FGFR-dependent cell lines H1581 and DMS114. Accordingly, levels of MYC and of cyclin D1 decreased upon FGFR inhibition within 24 hours (Fig. 4A). In contrast, expression levels remained relatively stable in FGFR1-amplified HCC95 and H520 cells, which are resistant to FGFR inhibition (6), as well as in the NRAS-mutant HCC15 cells (Fig. 4A and Supplementary Fig. S10C). MYC was also highly regulated on the transcriptional level in H1581, but not in H520 cells (data not shown).

To formally test whether MYC expression levels dictate sensitivity to FGFR inhibition, we stably silenced MYC in H1581 cells. This manipulation led to FGFR inhibitor resistance (Fig. 4B). Unfortunately, we could not test this hypothesis in DMS114 cells because they did not tolerate MYC knockdown. We next examined the regulation of downstream effectors in MYC signaling and found that the mitochondrial apoptosis mediators were predominantly affected by FGFR inhibition (PD173074, 1 μmol/L; Supplementary Fig. S12A); loss of the mitochondrial membrane potential as well as cytochrome C release occurred robustly after 72 hours in FGFR-dependent cell lines (Supplementary Fig. S12B).

Further analysis of RNAseq data revealed that tumor samples, in which the amplicon centered on FGFR1, expressed higher levels of MYC (P = 0.002) compared with other 8p12-amplified samples (Supplementary Fig. S8B). However, we were not able to detect a statistically significant co-occurrence of amplified 8p12 and MYC (Supplementary Fig. S3). Therefore, we analyzed the transcription levels of MYC in our cell line panel (n = 14). Levels of MYC gene expression predicted FGFR inhibitor sensitivity in individual 8p12-amplified cell lines (P = 0.02; Fig. 4C) as well as in groups of sensitive versus insensitive cell lines (Fig. 4D).

Altogether, we used cell culture and xenograft experiments (Fig. 3B and D) to study the interplay of FGFR1 with MYC. In all independent approaches, we observed that MYC modulates oncogenic transformation, cell-autonomous signaling, and FGFR inhibitor response in FGFR1-amplified or overexpressing cells (Fig. 4).

To extrapolate our finding that MYC expression levels dictate FGFR inhibitor sensitivity of FGFR1-amplified lung cancer to a larger panel of primary tumors, we screened a cohort of 306 squamous cell lung cancer biopsies for the presence of FGFR1 amplification by FISH (37). In this cohort 8p12 amplification occurred at a frequency of approximately 20%. A subcohort (n = 86) enriched for FGFR1 amplification (78%) was further analyzed for p-FGFR1 and MYC expression by immunohistochemistry using a 4-tier scale by three independent observers (Fig. 5A and Supplementary Fig. S13). We found strong membranous p-FGFR1 staining in this cohort (Supplementary Fig. S13). Only 26% of the amplified samples exhibited low scores of FGFR1 phosphorylation (Fig. 5A and Supplementary Fig. S13A). In contrast, high levels of nuclear MYC staining did not segregate with amplification status of FGFR1 (frequency 40% in FGFR1^amp vs. 46% in FGFR1^non-amp; P = 0.76; Fig. 5A and Supplementary Fig. S13A). Thus, whereas most FGFR1-amplified squamous cell lung cancers exhibited FGFR1 phosphorylation, only a fraction of these cases also showed nuclear MYC expression. The finding that only a minority of FGFR1-amplified lung tumors are likely to respond to FGFR inhibition (10, 11) is consistent with the possibility that MYC expression predicts FGFR dependency in this cohort.

We identified a 65-year-old caucasian man with a 70-pack-per-year smoking history. The patient was diagnosed with stage IV squamous cell lung cancer and had been initially treated with two chemotherapy lines (a combination of carboplatinum and paclitaxel and docetaxel monotherapy). We observed amplification of FGFR1 (2.6 ratio-signals per cell on average, plus 88% of the cells harbored 5 or more gene copies) in the patient's tumor (Fig. 5B). Immunohistochemical assessment revealed elevated expression levels of MYC with a score of 3 (Fig. 5B and Supplementary Fig. S13B). The patient agreed to treatment with BGJ398, a highly specific FGFR inhibitor (38), which was being evaluated in a first-in-humans trial at our center. After cardiac assessment and baseline thoracic computed tomography (CT; ref. 10), treatment with 100 mg BGJ398 was started. We observed a regression without cavitation of the tumor [CT scans after 4 and 8 weeks, partial response (PR) according to RECIST 1.1 criteria] and the patient experienced improvement of symptoms (Fig. 5C). After 10 months of therapy, progressive disease (PD) was diagnosed in the kidney (PD as to RECIST1.1 criteria), so that BGJ398 treatment was stopped.

Another patient was diagnosed with metastatic squamous cell lung cancer and high-level amplification of FGFR1 (10.1 signals per cells on average) and high expression of MYC (Supplementary Fig. S14). The FGFR1 amplification was highly focal, as determined by hybrid-capture–based massively parallel sequencing of 302 genes, enriched for the chromosomal region covering the 8p12 amplicon (Supplementary Fig. S14A). The patient refused chemotherapy, but consented to off-label use of pazopanib, a multikinase inhibitor with weak activity against FGFR (39, 40). After cardiac assessment and baseline thoracic CT (Supplementary Fig. S14B), treatment with pazopanib 400 mg b.i.d. was started. Four weeks and eight weeks after the start of pazopanib, CT showed tumor regression with cavitation (Fig. S14b). Because of grade 2 fatigue, stomatitis, and gastrointestinal side effects, the patient decided to stop pazopanib after 6 months. At that time, no clinical or radiologic signs of tumor progression were present. We note that the inhibitory profile of pazopanib and the pseudocavernous response are also compatible with a predominant antiangiogenic effect. However, in light of our preclinical findings, we speculate that the patient's response might also be attributable to FGFR inhibition in the context of an MYC-expressing, FGFR1-amplified lung cancer.

Preliminary results from an early FGFR inhibitor trial as well as our conclusions from cell culture experiments suggest that not all patients with FGFR1 amplification will benefit from FGFR inhibition (6, 10, 11, 41). Here, we show marked heterogeneity of the 8p12 amplification event in squamous cell lung cancer. This heterogeneity was also evident when assessing mechanisms of receptor activation and signaling modulation. We report that the 8p12 amplification results from broad genomic rearrangements, such that FGFR1 is located in the epicenter of the amplicon in only 28% of all 8p12-amplified cases. Furthermore, although FGFR1 phosphorylation was present in 74% of FGFR1-amplified primary lung tumors and is therefore unlikely to be predictive of the much more infrequent sensitivity to inhibition alone, high expression levels of MYC associated with FGFR dependency across different cell line models. Supporting an oncogenic role for amplified FGFR1, we show that the mesenchymal splice variants that we found to be expressed by amplified tumors are transforming in vitro and in vivo. This effect was strongly enhanced by coexpression of MYC. The role of high-level expression of MYC in mediating FGFR dependency was further strengthened by clinical observations of two patients with amplified FGFR1 and high MYC-expressing squamous cell lung cancer who responded to FGFR inhibition. Even though an FGFR-independent mechanism of response could not be ruled out, these clinical observations were in line with our preclinical observations. Beyond these cell-autonomous mechanisms of activation and modulation of drug response, we provide support for a role of autocrine and/or paracrine ligand-dependent receptor activation in modulating sensitivity to FGFR inhibition, which may be of potential therapeutic relevance.

In conclusion, we reveal that high expression of MYC as well as FGF ligand concentrations modulate oncogenic transformation and response to FGFR inhibition in FGFR1-amplified lung cancer. We hope that our findings may help in refining the selection of patients who are most likely to benefit from treatment with FGFR inhibitors.

Cancer cell lines, HEK293T and NIH3T3 cells were purchased from American Type Culture Collection and German Resource Centre for Biological Material (DSMZ) and cultured using either RPMI or Dulbecco's Modified Eagle Medium (DMEM) high-glucose media, supplemented with 10% fetal calf serum (FCS). Adherent cells were routinely passaged by washing with PBS buffer and subsequent incubation in Trypsin/EDTA. Trypsin was inactivated by the addition of culture medium and cells were plated or diluted accordingly. Suspension cell lines were passaged by suitable dilution of the cell suspension. All cells were cultured at 37°C and 5% CO₂. The identity of all cell lines included in this study was authenticated by genotyping (SNP 6.0 arrays, Affymetrix) and all cell lines are tested for infection with mycoplasma (MycoAlert, Lonza). Furthermore, the identity of the H1581 cell line was ensured by short tandem repeat profiling (DNA fingerprinting).

Cell lines were starved from bovine serum for 24 hours and stimulated by a collection of 6 FGF ligands (1 ng/mL) and heparin (10 μg/mL) for 20 minutes. In addition, the FGFR inhibitor PD173074 (1 μmol/L) was added 40 minutes before stimulation by FGF-1 and FGF-2. Phosphorylation of FGFR, ERK, AKT, and the FGFR1 signaling adapter protein FRS2α as well as total expression of ERK and FGFR1 were assessed by immunoblotting.

Total RNA was extracted from fresh-frozen lung tumor tissue containing at least 60% tumor cells. Depending on the tissue size, 15–30 slides were cut using a cryostat (Leica) at −20°C. Material for RNA extraction was disrupted and homogenized for 2 minutes at 20 Hz by Tissue Lyser (Qiagen). RNA was extracted using the Qiagen RNeasy Mini Kit. RNA quality was assessed by a Bioanalyzer; samples showing an RNA integrity number (RIN) > 8 were retained for transcriptome sequencing. We cloned cDNA strands of 250 bp into a sequencing library, allowing us to sequence 95-bp paired-end reads without overlap. All RNAseq libraries were analyzed on the Illumina Genome Analyzer IIx.

Gene coverage was used to differentiate splice variants of FGFR1. Mesenchymal splice variants of FGFR1 were differentiated by coverage of exon 2, whereas coverage of tissue-specific exons 8 (IIIb/IIIc) distinguished epithelial (IIIb) from mesenchymal (IIIc) forms.

All animal procedures were approved by the local animal protection committee and the local authorities. Transduced NIH3T3 and tumor cells were resuspended in RPMI or DMEM medium and injected (5 × 10⁶ cells per tumor) subcutaneously into the flanks of 8- to 15-week-old male nude mice [Rj:NMRI-nu (nu/nu), Janvier Europe] under 2.5% isoflurane anesthesia (42).

To assess the effect of FGFR inhibitors in vivo, NVP-BGJ 398 (Novartis) was dissolved in a vehicle solution (33% PEG300, 5% glucose) for xenograft application (6, 38). Tumor size was monitored every second day by measurement of perpendicular diameters by an external caliper (42) and calculated by use of the modified ellipsoid formula [V = 1/2 (Length × Width²)]. Oral therapy was started when tumors reached a volume of 100 mm³. Mice received daily either BGJ398 (15 mg/kg) or vehicle solution. After 14 (NIH3T3 FGFR1β + MYC), 16 (NIH3T3 EML4–ALK, KRAS G12V), or 25 (NIH3T3 e.V., FGFR1α/β) days of therapy, respectively, mice were sacrificed by intraperitoneal injection of ketamine/xylazine (300/60 mg/kg).

To examine ligand dependency in vivo, AdCMV-null virus (Vector Biolabs) and AdsFGFR virus (titer: 1 × 10¹⁰, contributed as a kind gift by Gerhard Christofori, University of Basel) were mixed with tumor cells in DMEM for subcutaneous injection. Tumor formation was monitored twice a week by careful visual inspection and palpation of the skin. As soon as tumors became palpable, diameters were measured by an external caliper to determine tumor volumes. In addition, animal weights were documented weekly. Eight weeks after injection of H1581 and A549 tumor cells, animals were sacrificed.

Subcutaneous tumors as well as livers were resected and fixed in 4% formaldehyde for immunohistochemical staining and virus detection, respectively.

Cell culture supernatants were collected, centrifuged (200 rcf, 5 minutes), concentrated by ultracentrifugation units (Satorius AG) and analyzed for FGF concentrations by ELISA (Abcam). In addition, protein was extracted from cells, collected in equal amounts of lysis buffer (Cell Signaling Technology), and measured by Bradford assay (Pierce). Normalized FGF concentrations (c_Norm) were derived as ratios of FGF and lysate protein concentrations.

For further details, refer to Supplementary Methods.

F. Malchers is a consultant/advisory board member of Blackfield AG. L. Nogova has received honoraria from the speakers' bureaus of Novartis and Roche and is a consultant/advisory board member of Amgen, Lilly, and Bayer. J.M. Heuckmann has ownership interest (including patents) and is a co-founder and shareholder in Blackfield AG. J. Diebold is a consultant/advisory board member of Roche and Pfizer. M. Scheffler is a consultant/advisory board member of Boehringer Ingelheim. M. Peifer has ownership interest (including patents) and is a founder and shareholder of Blackfield AG, and is a consultant/advisory board member of the same. T. Zander has received honoraria from the speakers' bureaus of Amgen, Roche, Novartis, and Boehringer Ingelheim and is a consultant/advisory board member of Amgen. F. Ringeisen is employed as a Senior Medical Director, Clinical Research, in Novartis Pharma and has ownership interest (including patents) in the same. J. Wolf has received honoraria from the speakers' bureau of Novartis and is a consultant/advisory board member of the same. R.K. Thomas has received commercial research grants from AstraZeneca, Merck, and EOS, has ownership interest (including patents) in Blackfield AG, and is a consultant/advisory board member of JNJ, Blackfield, AstraZeneca, Roche, Lilly, Sanofi, and Merck. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Malchers, F. Dietlein, P.S. Mainkar, S. Chandrasekhar, D. Rauh, R. Buttner, J. Wolf, R.K. Thomas

Development of methodology: F. Malchers, F. Dietlein, J.M. Heuckmann, M. Peifer, A. Florin, N. Karre, D. Rauh, R. Buttner, L.C. Heukamp, R.K. Thomas

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Malchers, F. Dietlein, J. Schöttle, L. Nogova, K. Albus, J.M. Heuckmann, O. Gautschi, J. Diebold, D. Plenker, M. Gardizi, M. Scheffler, M. Bos, D. Seidel, S. Chandrasekhar, J. George, S. Silling, T. Zander, R. Ullrich, H.C. Reinhardt, R. Buttner, L.C. Heukamp, J. Wolf, R.K. Thomas

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Malchers, F. Dietlein, J. Schöttle, X. Lu, L. Nogova, J.M. Heuckmann, M. Bos, F. Leenders, M. Peifer, P.S. Mainkar, N. Karre, S. Chandrasekhar, D. Rauh, R. Ullrich, R. Buttner, J. Wolf, R.K. Thomas

Writing, review, and/or revision of the manuscript: F. Malchers, F. Dietlein, L. Nogova, O. Gautschi, J. Diebold, M. Scheffler, M. Bos, J. George, R. Ullrich, H.C. Reinhardt, F. Ringeisen, R. Buttner, L.C. Heukamp, J. Wolf, R.K. Thomas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Dietlein, L. Fernandez-Cuesta, D. Seidel, A. Richters, A. Florin, S. Chandrasekhar, S. Silling, H.C. Reinhardt, R. Buttner, R.K. Thomas

Study supervision: F. Malchers, F. Dietlein, S. Chandrasekhar, D. Rauh, J. Wolf, R.K. Thomas

The authors thank Will Morrel for invaluable technical help and Gerhard Christofori for providing the AdsFGFR virus. They also thank Dr. Klaus Strobel and Dr. Peter Hofman for CT images and photograpy.

This work was co-funded by the German federal state North Rhine Westphalia (NRW) and the European Union (European Regional Development Fund: Investing In Your Future) as part of the PerMed NRW initiative (grant 005-1111-0025; to R.K. Thomas and J. Wolf). This work was also supported by the EU-Framework Program CURELUNG (HEALTH-F2-2010-258677; to R.K. Thomas and J. Wolf), by the Deutsche Forschungsgemeinschaft through TH1386/3-1 (to R.K. Thomas) and through SFB832 (TP6 to R.K. Thomas; TP5 to L.C. Heukamp), by the German Ministry of Science and Education (BMBF) as part of the NGFNplus program (grant 01GS08100; to R.K. Thomas and J. Wolf), by the Deutsche Krebshilfe as part of the Oncology Centers of Excellence funding program (to R. Buttner and R.K. Thomas) and through the Mildred-Scheel-Doktorandenprogramm (grant 110770; to F. Dietlein and R.K. Thomas), by the Max Planck Society, by the Behrens-Weise Foundation (M.I.F.A.NEUR8061; to R.K. Thomas), and by an anonymous foundation (to R.K. Thomas). R.K. Thomas is supported by a Stand Up To Cancer Innovative Research Grant, a Program of the Entertainment Industry Foundation (SU2C-AACR-IR60109).