Upregulation of FGF9 in Lung Adenocarcinoma Transdifferentiation to Small Cell Lung Cancer

Lung adenocarcinoma and small cell lung cancer (SCLC) constitute distinct subgroups of human lung cancer (1). Cell of origins of lung adenocarcinoma and SCLC have been reported to be type II alveolar and neuroendocrine cells of the lung epithelium, respectively (2–5). However, the transdifferentiation of adenocarcinoma to SCLC has been frequently reported in clinical cases (6–12). Transdifferentiation between histological subtypes of cancer is usually a rare phenomenon, but has been reported in prostate cancer and lung cancer that have acquired resistance to molecular targeted therapeutics (6–9, 13–15), which suggests lineage plasticity of cancer cells. For example, the frequency of transdifferentiation of prostate adenocarcinoma to a more aggressive neuroendocrine prostate cancer has been reported to be 17% in 202 prostate cancer cases (16). Similarly, 5–15% of lung adenocarcinomas harboring activating epidermal growth factor receptor (EGFR) mutations transdifferentiate to SCLC after treatment with EGFR tyrosine kinase inhibitors (EGFR-TKI; refs. 7–9). Small cell transdifferentiation can also occur in anaplastic lymphoma kinase (ALK)-translocated non-SCLC (NSCLC) after ALK-TKI treatment (10), in NSCLC after immune-checkpoint inhibitor treatment (11) or in NSCLC without any treatment (12). These transdifferentiated cancers are no longer sensitive to either targeted therapy or immune-checkpoint inhibitor and show poor prognosis, which is in sharp contrast with the good prognosis of patients with NSCLC with effective targeted therapy (15, 17, 18). Several studies have reported that patients with transdifferentiated SCLC showed poorer prognosis than patients with de novo SCLC (12, 19). Therefore, to improve the prognosis of patients with lung cancer with transdifferentiated SCLC, understanding the mechanisms underlying transdifferentiation is essential.

Primary SCLC accounts for around 13% of lung cancer cases and is considered a deadly disease, with a 5.6% 5-year survival rate (20). SCLC is pathologically characterized by neuroendocrine features, and the expression levels of synaptophysin (SYP), chromogranin A (CHGA), calcitonin gene-related peptide (CGRP), neural cell adhesion molecule 1 (NCAM1, also known as CD56), and achaete-scute homolog 1 (ASCL1; refs. 21, 22) are used clinically or preclinically as markers of neuroendocrine differentiation. Recent genomic profiling of primary SCLC identified alterations in the retinoblastoma (RB1) and tumor protein 53 (TP53) genes (4, 5, 23), leading to simultaneous inactivation of these canonical tumor suppressors in almost all SCLC, indicating that these are prerequisites for SCLC development. Additional genetic alterations such as MYC amplification and inactivating mutations in epigenetic modifiers such as histone acetyl transferases, CREBBP and EP300, and NOTCH family members have also been reported (23–25). In addition, functional roles of several genes including PTEN, RBL2, MYC, and NOTCH pathway signaling genes (26, 27) in SCLC development have been reported in genetically engineered mouse models, indicating that these genetic and epigenetic alterations can be the “third hit” for SCLC development.

Several studies have determined some of the molecular characteristics of transdifferentiated SCLC (28–30). The genetic alterations of TP53 and RB1 genes have been repeatedly reported. A study has reported the inactivation of RB1 in transdifferentiated SCLC, but it was not sufficient for the transdifferentiation to SCLC (28). Another report that evaluated the evolutional trajectories of SCLC transdifferentiation identified both TP53 and RB1 inactivation at an early stage of clonal evolution in transdifferentiation (30). The essential roles of the pathognomonic inactivation of TP53 and RB1 in transdifferentiation are consistent with primary SCLC, indicating that common pathway alterations exist between primary and transdifferentiated SCLC. However, inactivation of these canonical tumor suppressors is a prerequisite for transdifferentiation. Definitive additional events, the third hit, for transdifferentiation have not been identified. In addition, no prospective experiments have provided evidence of transdifferentiation. To understand the key biological processes in transdifferentiation, the identification of the third hit event and a prospective and functional evaluation are needed.

In this study, using paired samples from patients with lung cancer with confirmed transdifferentiated SCLC, we identified FGF9 upregulation as a key molecular event for SCLC transdifferentiation. Furthermore, we prospectively confirmed FGF9-mediated neuroendocrine differentiation in established mouse and human lung adenocarcinoma cells in vitro and in vivo, providing direct evidence for FGF9-mediated transdifferentiation.

MLE12, H1975, H1650, H2228, H2087, H2110, A549, H441, H358, H69, H209, COR-L51, COR-L88, H82, and H146 were purchased from the ATCC and cultured in RPMI-1640 (Life Technologies) growth medium supplemented with 10% FBS and 1% penicillin/streptomycin. The BID007 cell line was established as described previously (31). To make stable cell lines that constitutively express FGF9, human FGF9 and mouse Fgf9 cDNA was amplified and cloned into the MigR1 retrovirus vector (#27490; Addgene). The MigR1 empty vector and MigR1-FGF9 vectors were transfected into Phoenix AMPHO or Bosc23 cells to make retrovirus particles. The medium containing the retroviruses was collected 48 and 72 hours after transfection. Infection was performed with polybrene. Of note, the MigR1 retrovirus vector contains an internal ribosome entry site (IRES)-eGFP sequence; highly positive GFP cells were, thus, sorted using flow cytometry (MoFlo XDP, ML99030; Beckman Coulter). The expression of FGF9 in these cells was confirmed using quantitative RT-PCR and Western blotting. The stable cell lines that express EGFR that harbors activating mutations (L858R and an exon 19 deletion) were prepared as previously described (53). To create RB1-knockout cells, we used the RB1 CRISPR system (#KN206933; OriGene). gRNA vectors (scrambled or RB1) and donor DNA were transfected into target cell lines. Puromycin-resistant cells were selected as RB1-knockout cells, and a single clone was picked for each gRNA. The protein expression of RB1 in these cells was confirmed using Western blotting.

AZD4547 (#S2801) was purchased from Selleck Chemicals. Recombinant FGF9 protein was purchased from Peprotech. Anti-phospho-p44/42 MAPK (Tyr202/Tyr204; #3126), anti-total p44/42 MAPK (#3127), anti-RB1 (#9309) anti-phospho-AKT (Ser473; D9E; #4060), and anti-total AKT (#9272) antibodies were purchased from Cell Signaling Technologies. An anti-actin antibody produced in mice (#A5441) and rabbit polyclonal antibodies against CGRP (#C8198) were purchased from Sigma-Aldrich. An anti-chromogranin antibody (#412751) and anti-synaptophysin antibody (#413831) were purchased from Nichirei Biosciences Inc. An anti-FGF9 antibody was purchased from R&D Systems (#AF-273NA) and Abcam (#ab71395). A goat anti-GFP polyclonal antibody (#600–101–215) was purchased from Rockland. A rabbit polyclonal antibody against FGFR1 (#sc-121) was purchased from Santa Cruz Biotechnology. Anti-EGFR (clone 3C6) and anti-Ki67 (clone 30–9) antibodies (CONFIRM kits) were obtained from Ventana-Roche.

This assay was performed as described previously (32); we used 1 × 10⁴ MLE12-FGF9 and MLE12-empty cells per well. After 4 weeks of culture, colonies were counted using a colony counter (Lumi vision analyzer; AISIN).

Total cellular RNA was isolated from cells using the RNeasy Mini Kit (#74106; Qiagen), and 2 μg of RNA was reverse-transcribed into cDNA using the High Capacity RNA-to-cDNA Kit (#4387406; Applied Biosystems). For qRT-PCR, we used an ABI Prism 7000 Sequence Detection System (Life Technologies). Relative quantification was performed using the ΔC_t method (cycle threshold), in which the ΔC_t value was calculated by subtracting the C_t value of each gene from that of human and mouse GAPDH. The data represent the average of three replicates from at least two independent experiments.

The MTS assay was performed as described previously (31). Briefly, H1975 cells (2 × 10³ cells/well) were seeded into 96-well plates on day 1 and cultured for 72 hours. Subsequently, the relative cell numbers were measured. All conditions were tested in triplicate. The data represent the average of three technical replicates; a representative experiment from at least two independent experiments is shown.

MLE12 cells were transfected with a final concentration of 20 nmol/L of siRNA targeting FGFR1, FGFR2, or FGFR3 or of negative control siRNA (FGFR1: #s66023, #66024, FGFR2: #s201347, #s201348, FGFR3: #s66030, #s66031, Life Technologies). For transfection, SilentFect (#1703361; Bio-Rad) was used according to the manufacturer's protocol. Knockdown of the expression of FGFR1, FGFR2, and FGFR3 was confirmed using qRT-PCR. SCLC cells were also transfected with siRNA targeting FGF9 (#s230578, #s5146, #s5147) or negative control siRNA (#4390844; Thermo Fisher Scientific) at a final concentration of 20 nmol/L. Seventy-two hours after transfection, the relative number of viable cells was measured using the MTS cell proliferation assay.

Total proteins were extracted using cell lysis buffer (#9803; Cell Signaling Technologies). Protein concentrations were measured using the BCA protein assay (#23225; Thermo Fischer Scientific), and equal amounts of proteins were denatured and reduced using the sample buffer. After boiling the samples, their aliquots were subjected to SDS-PAGE. The fractionated proteins were transferred onto polyvinylidene difluoride membranes, which were then incubated first with primary antibodies, and then with secondary antibodies. For protein detection, the membranes were incubated with agitation in the LumiGLO reagent and peroxide (#7003S; Cell Signaling Technologies) and exposed to X-ray films.

All animal experiments were approved by the Laboratory Animal Center, Keio University School of Medicine. Female NOD/SCID and BALB/C nu/nu nude mice were purchased from Charles River. Mice were anesthetized with ketamine. H2228, H69, H209, COR-L88, and MLE12 cells transduced with or without FGF9 were suspended in Matrigel (#356237; Corning) and were subcutaneously injected into the mice. Tumor volume was monitored using calipers. To evaluate drug efficiency, when the average tumor volume reached 100 to 200 mm³, mice were randomized to receive either placebo (1% Tween 80, #T0546; Tokyo Chemical Industry Co.) or an FGFR inhibitor (AZD4547, 12.5 mg/kg) once a day via intragastric administration. Formalin-fixed and paraffin-embedded (FFPE) tissues from sacrificed animals were then subjected to IHC and immunofluorescence analysis. On the other hand, to examine tumor formation in the mouse lungs, MLE12-FGF9 cells were intravenously injected into the tail vein of NOD/SCID mice. Ten weeks after injection, tumor formation in the mouse lungs was confirmed using micro-CT (μCT). Subsequently, mice were sacrificed, and their lungs were harvested.

Immunofluorescence was performed on cultured cells and paraffin-embedded sections from tumor-bearing mice as described previously (31).

IHC was performed, and slides were evaluated by expert pathologists from Keio University Hospital, Tokyo Saiseikai Central Hospital, and Yeouido St. Mary Hospital. FGF9 was stained as described previously (33). Of note, cell blocks of MLE12-empty and H1975 cell lines were used as negative controls, whereas those of MLE12-FGF9 and COR-L51 cell lines were used as positive-controls for FGF9 staining (Supplementary Fig. S1). The staining intensity in the cytoplasm was quantified based on an empirical 3-step scoring method (0, no staining; 1+, weak staining; 2+, strong staining); the samples with 1+ and 2+ staining intensity were defined as positive. Anti-EGFR and anti-Ki67 antibodies (prediluted) were also used on a Benchmark automated IHC platform (Ventana-Roche) using the CC1 m protocol and ultraView-HRP-based detection (34–36).

Relative gene expression data for FGF9, other FGFs, and neuroendocrine markers were individually obtained from the Cancer Cell Line Encyclopedia (CCLE) database (https://portals.broadinstitute.org/ccle/; ref. 37). Fifty-three SCLC and 134 NSCLC cell lines were evaluated.

Gene expression data from 81 human SCLC samples (23) were used to obtain an expression heatmap and correlation matrix with pair-wise Pearson correlation coefficients, comparing the expression levels of FGF9, FGF2, SFTPC, and neuroendocrine marker genes within the dataset.

Exome sequencing libraries using FFPE samples from case #1 were constructed following the manufacturer's protocol (SureSelectXT Human All Exome V6/V6 + UTRs Kit; Agilent). Using Trim Galore, we discarded short reads and reads with insufficient base qualities. The trimmed reads were aligned to the reference genome (GRCm38) using the Burrows–Wheeler Aligner (BWA); the index files required by BWA were generated separately. Several postprocessing steps were required to prepare the files for the detection of SNVs, loss of heterozygosity, and copy-number variations. CleanSam was used to obtain information on soft-clipped reads, which are only partially aligned to the reference genome. Next, these files were sorted using samtools. In addition, using Picard Readgroups, we marked reads that have been sequenced together. Then, duplicated reads (which were possibly PCR duplicates) were marked. Base recalibration was conducted in the final step of post-processing. Somatic point mutations and indels were called simultaneously, and the results were stored as a VCF file by running Mutect2 (tumor-only mode). Using FilterMutectCalls provided in the GATK package, probable technical or germline artifacts were removed. Using SelectVariants, all indels >10 bp were filtered out. Additional filters were used to decrease the false-positive rate of reported mutations; we applied filters for mutant allele frequency (≥10%) and coverage at particular positions in tumor samples (≥10×). To further reduce false-positive callings, the SNVs and indels were compared with different databases: CLINVAR (Benign/Likely_benign), Exome Aggregation Consortium and gnomAD (over 10% common SNP in East Asia), PolyPhen2 (score under 0.5), and SIFT (score over 0.05). All mutation information was visualized using mafrools (ver. 1.6.15). Of note, we tested whether two bam files were from the same patient using BamixChecker and calculated the genotype-concordance score (38). Mutation signature analysis was performed using SigProfiler (39).

RNA was extracted from the FFPE samples (case #1). Using Trimomatic (ver. 0.38), we discarded short reads and reads with insufficient base quality. The trimmed reads were then aligned with the reference genome (GRCm38) using STAR (ver. 2.7.0.f). TMM normalization and gene set enrichment analysis (GSEA) were performed using edgeR (ver. 3.22.3) and gprofiler2 (version 0.1.6), respectively.

Statistical analysis was performed using GraphPad Prism 5.0 (Graph Pad Prism Software Inc.). The Student t test was used for comparisons between two groups. For correlation studies, the Pearson correlation test was performed after the confirmation of parametric distribution. All P values were two-sided; P values <0.05 were regarded as statistically significant.

The datasets generated during and/or analyzed in this study are available from the corresponding author upon reasonable request.

This study was approved by the ethical review boards of Keio University School of Medicine (approval no. 20110171), Tachikawa Kyosai Hospital (approval no. 2016–06), Tokyo Saiseikai Central Hospital (approval no. 30–12), and St. Mary's Hospital (approval no. XC16SIMI0048S), and was performed in accordance with the Declaration of Helsinki. All human specimens were acquired from patients under the guidance of each institution. Written informed consent was obtained in all cases and all ethical regulations were followed. In addition, all animal experiments were approved by the Laboratory Animal Center, Keio University School of Medicine (approval no. 12115).

First, we focused on a case (case #1) with confirmed adenocarcinoma to SCLC transdifferentiation. The patient was a 69-year old male with 40 pack-years of smoking history. This patient with lung adenocarcinoma harboring the EGFR L858R mutation was treated with gefitinib, a first-generation EGFR-TKI. After 24 months, when the patient experienced disease progression, a second biopsy revealed the L858R and T790M mutations. Therefore, osimertinib, a third-generation EGFR-TKI was administered. Five months after osimertinib treatment, the patient experienced disease progression and underwent a third biopsy (Fig. 1A). The third biopsy revealed transdifferentiation of the adenocarcinoma into SCLC with the EGFR L858R mutation. Fortunately, we were able to obtain paired samples, before and after EGFR-TKI treatment, and performed integrative molecular profiling, including whole-exome sequencing (WES) and RNA sequencing (RNA-seq). First, WES confirmed the identical patient origin of the paired samples (genotype-concordance score = 0.889). The original EGFR mutation L858R was confirmed in both pre- and postsamples, indicating that the EGFR mutation is a clonal event. The list of mutant genes is shown in Supplementary Fig. S2. A nonsense mutation in RB1 was detected in the post-SCLC sample but not in the presample, and a missense mutation in TP53 DNA binding domain was detected in the both samples (Fig. 1B). In addition, PIK3CA mutations and MYC amplification were not observed. Mutation signature analysis identified smoking-related signatures in both samples (Supplementary Table S1), reflecting the heavy smoking history of the patient. RNA-seq analysis revealed distinct expression profiles of the samples. The expression of neuroendocrine markers such as NCAM1, CHGA, SYP, CHGB, and ASCL1 or small cell neuroendocrine signatures (SCN signatures; ref. 6) was upregulated in the transdifferentiated SCLC (Fig. 1C). Gene Ontology (GO) term analyses with upregulated/downregulated genes showed that 12 of the 40 pathways significantly altered by SCLC transformation were “neuron”-related terms (Supplementary Figs. S3A and S3B). Recently, primary SCLC has been classified into four groups depending on key transcriptomic patterns: ASCL1, NEUROD1, POU2F3, and YAP1 (40). In this case, the SCLC belonged to the ASCL1 type (Fig. 1D). The expression level of the EGFR gene was slightly lower in the post- than in the presample, which may reflect the loss of sensitivity to EGFR-TKIs. The expression levels of TP53 and RB1 were comparable (Fig. 1E). Interestingly, upregulation of the FGF9 was observed in the post-SCLC sample (Fig. 1E). Of the upregulated genes, we focused on FGF9 because its overexpression has been associated with malignant phenotypes in multiple cancer cells (41–44), and FGF9-mediated neuroendocrine differentiation was observed in a mouse prostate cancer model (45). To confirm FGF9 upregulation at the protein level, we performed immunohistochemistry (Fig. 1F). Again, a robust upregulation of FGF9 was confirmed in the SCLC sample. In addition, to evaluate the specificity of FGF9 upregulation in the context of transdifferentiation, the expression levels of other FGFs were also evaluated. A clear upregulation was observed only for FGF9, and not for other FGFs (Fig. 1G).

To confirm FGF9 upregulation in multiple cases of transdifferentiation to SCLC, we collected samples from multiple institutions in Japan and Korea. We obtained paired samples from five additional patients. The clinical characteristics or CT images of the patients are shown in Supplementary Table S2 and Supplementary Fig. S4, respectively. Although the EGFR-mutated NSCLC is associated with no or light smoking history (46), five of six cases (83.3%), including case #1, had dense smoking exposures. The mean smoking exposure was 44.1 pack-years in our cases, in clear contrast with the 3.1 pack-year history in a previous report (30). Notably, robust FGF9 upregulation was observed in four of six (66.7%) cases (Fig. 2A and B). Decreased expression of EGFR and increased expression of Ki67 were confirmed in cases #5 and #6; samples from the other cases were not available due to clinical reasons (Supplementary Fig. S5; Supplementary Table S3). Increased expression of FGFR1 was observed in case #6 (Supplementary Fig. S5). Altogether, these data indicate that the upregulation of FGF9 is a common event in adenocarcinoma to SCLC transdifferentiation, although the functional roles of FGF9 in transdifferentiation still remain elusive.

To seek the clinical relevance of FGF9 expression in the transdifferentiated SCLC, we evaluated FGF9 expression in primary SCLC because primary and transdifferentiated SCLC share common molecular alterations in the oncogenic process. Interestingly, the FGF9 expression level was positively correlated with the expression of neuroendocrine markers such as ASCL1, CHGA, and NCAM1 in previously reported human primary SCLC dataset (Fig. 3A and B; ref. 23). In a cancer cell line dataset (37), FGF9 mRNA expression level was significantly higher in SCLC than in NSCLC cells (Fig. 3C; Supplementary Fig. S6), and it was positively correlated with neuroendocrine marker genes (Fig. 3D; Supplementary Fig. S7A and B). Interestingly, the knockdown of FGF9 or the inactivation of the FGFR pathway inhibited the growth of an FGF9-expressing SCLC cell line, H69 cells, in vitro and in vivo (Supplementary Figs. S8A–S8C), although no functional roles of FGF9 were observed in other FGF9-expressing SCLC cell lines. Therefore, these data suggest that FGF9 plays a functional role in a subgroup of primary SCLC.

In fact, the expression of FGF9 was confirmed in 45 of 69 (65.2%) of our original SCLC samples (Fig. 3E and F; Supplementary Table S4). FGF9 is reported to be expressed in epithelial and mesothelial cells of embryonic lungs and downregulated in normal adult lungs (47). Consistent with this report, FGF9 expression was not observed in adjacent noncancer tissues. These data support the clinical relevance of FGF9 upregulation in transdifferentiation.

Our observation that FGF9 is upregulated in multiple cases of transdifferentiation and that FGF9 expression is positively associated with neuroendocrine marker gene expression prompted us to prospectively evaluate the functional roles of FGF9 in transdifferentiation.

First, we introduced FGF9 into MLE12 cells (Supplementary Fig. S8). MLE12 cells, which express the large T antigen that inactivates Tp53 and Rb1, are mouse lung epithelial cells derived from mouse lung adenocarcinoma (32, 48). FGF9 induction significantly increased 3D colony formation in vitro (Fig. 4A), highlighting the capacity of FGF9 to promote transformation. Interestingly, we found that some MLE12-FGF9 cells detached from the surface of the culture dish and continued to grow as floating aggregates after 3 weeks of culture, which is a characteristic of SCLC cell lines (Fig. 4B). These floating cells were collected and examined for expression of Fgf9 and of several neuroendocrine markers, including Cgrp, compared with that in cells that remained attached to the plate. Cgrp expression in floating MLE12-FGF9 cells was significantly higher than that in the attached MLE12-FGF9 cells (Fig. 4C). These data suggest the possibility that FGF9 functionally contributes to neuroendocrine differentiation.

To confirm the transdifferentiation capacity of FGF9 in vivo, mouse allograft model experiments were performed. MLE12-FGF9 or MLE12-empty vector control cells (1.0 × 10⁶) were subcutaneously injected into NOD/SCID mice (n = 6/group). Four weeks after injection, all six mice injected with MLE12-FGF9 cells harbored subcutaneous tumors, whereas none of the mice injected with MLE12-empty vector cells had tumors (Fig. 4D). Interestingly, the histology of the subcutaneous MLE12-FGF9-derived tumors resembled that of human SCLC, which consists of a pleomorphic population of small, round cells with scanty cytoplasm and hyperchromatic nuclei with homogeneous chromatin dispersion (Fig. 4E). Quantitative reverse transcription PCR (qRT-PCR) confirmed increased mRNA levels of the neuroendocrine markers Ascl1 and Cgrp in the formed tumors (Fig. 4F). CGRP expression was confirmed at the protein level (Fig. 4G). These data are suggestive of FGF9-mediated neuroendocrine differentiation at the molecular level. To rule out potential effects of the subcutaneous tissue microenvironment on epithelial cell differentiation and to confirm that transformation to SCLC can also occur in the lung microenvironment, we injected MLE12-FGF9 cells intravenously via the tail vein of NOD/SCID mice. Ten weeks after injection, we confirmed tumor formation in the mouse lungs using μCT (Fig. 4H). Again, pathologic analysis revealed histopathologic similarities with human SCLC (Fig. 4I). We also transduced a vector encoding the EGFR gene harboring the activating L858R mutation or the exon 19 deletion (49), a common lung adenocarcinoma oncogene, into MLE12 cells in order to rule out the possibility of nonspecific SCLC transdifferentiation of MLE12 cells. Transplanting these MLE12-EGFR cells into NOD/SCID mice (n = 3, respectively) subcutaneously did not result in tumor formation (Supplementary Fig. S9). These data provide preclinical evidence that FGF9 functionally contributes to SCLC transdifferentiation in vitro and in vivo.

To elucidate the mechanism underlying FGF9-mediated SCLC transdifferentiation, immunoblotting was performed for MLE12-empty and -FGF9 cells. Increased phosphorylation of ERK but not AKT was observed (Fig. 5A), indicating MAPK pathway activation by Fgf9. Recombinant human FGF9 (rhFGF9) also activated the MAPK pathway (Fig. 5B). To clarify the corresponding receptors of FGF9-mediated signals, gene knockdown experiments using siRNA was performed. The phosphorylation of ERK was attenuated by Fgfr1, Fgfr2, and Fgfr3 knockdown, indicating that all of these receptors transduce downstream signals; of note, variability was detected partly due to the knockdown efficiency (Fig. 5C). Notably, qRT-PCR experiments identified significantly increased mRNA levels of FGF receptors Fgfr1 and Fgfr2 (Fig. 5D) in MLE12-FGF9 mouse tumors compared with those in MLE12-empty and -FGF9 cells, which suggests FGF9-mediated upregulation of the FGFR pathway in tumors in vivo. Immunofluorescence showed the colocalization of FGFR1 and FGF9 or GFP signals in MLE12 tumors (Fig. 5E). In addition, to compare the expression levels of Fgfr genes between MLE12-FGF9 cells in vitro and MLE12-FGF9 cells in the tumors in vivo, GFP-positive cells were sorted from the developed tumors. Interestingly, qRT-PCR analysis revealed the clear upregulation of Fgfr1 and Fgfr2 in MLE12-FGF9 cells from tumors (Supplementary Fig. S10), suggesting the cell-autonomous expression of FGF receptors. To further confirm that in vivo FGF9-mediated transformation occurred through the FGFR pathway and to exclude the contribution of other signaling pathways, we used AZD4547 (50, 51), a pan-FGFR inhibitor, to treat MLE12-FGF9 tumors. Nine NOD/SCID mice were injected subcutaneously with MLE12-FGF9 cells and observed until tumor nodules reached 200 mm³, at which point, they were treated with either AZD4547 (12.5 mg/kg/day; n = 5) or the vehicle (n = 4) for 24 days. AZD4547 treatment significantly reduced tumor growth (Fig. 5F) compared with vehicle treatment. Collectively, these results indicate that FGF9 induces the transdifferentiation of mouse lung adenocarcinoma-derived cells to SCLC through FGF9-mediated upregulation of the FGFR pathway.

Next, to confirm the FGF9-mediated neuroendocrine differentiation in established human lung adenocarcinoma-derived cells, we performed further in vitro and in vivo experiments. To obtain the preclinical evidence for neuroendocrine differentiation in EGFR-mutated lung adenocarcinoma cells, we used H1975 cells, which harbor TP53 and EGFR L858R + T790M mutations. We performed a gene knockout of RB1 using the CRISPR/Cas9 technology in H1975 cells, with or without FGF9 induction (Fig. 6A), because inactivation of RB1 is frequently reported in transdifferentiation (28–30). However, no upregulation of the neuroendocrine genes was observed, indicating that TP53 and RB1 inactivation and FGF9 overexpression were not sufficient to induce neuroendocrine differentiation in H1975 cells. Further, to mimic the clinical situation wherein EGFR-TKI-resistant lung NSCLC transdifferentiates to SCLC, H1975 cells with RB1 knockout and FGF9 overexpression were subjected to chronic exposure to osimertinib, which subsequently established osimertinib-resistant (osiR) H1975 cells (Fig. 6B). The expression of ASCL1, a master regulator of neuroendocrine differentiation, was minimal in the H1975 parental, in the H1975 RB1-knockout, and in the H1975 RB1-knockout with FGF9 transduction cells (Fig. 6C). However, ASCL1 expression was significantly increased in the H1975 RB1-knockout osiR and in the H1975 RB1-knockout osiR with FGF9 transduction cells. These data indicate that inactivation of TP53 and RB1, FGF9 overexpression, and EGFR pathway inhibition are essential to induce neuroendocrine differentiation in H1975 cells. We evaluated two additional EGFR mutant lung cancer cell lines, H1650 and BID007 (Supplementary Table S5; Supplementary Figs. S11A and S11B). The inactivation of RB1, overexpression of FGF9, and the inhibition of the EGFR pathway did not induce the upregulation of the neuroendocrine markers in these cells. Thus, it can be inferred that FGF9 induces neuroendocrine differentiation in only a subset of EGFR mutant NSCLC cell lines.

Finally, to examine whether FGF9 expression induces neuroendocrine differentiation in other lung adenocarcinoma cells, we transduced FGF9 into H2087, H2110, and H2228 lung adenocarcinoma cell lines, which harbor both TP53 and RB1 inactivating mutations but not EGFR-activating mutations (Supplementary Table S5). We transplanted these FGF9-transduced lung adenocarcinoma cells into immunodeficient mice. Although the growth rate of H2087 and H2110 cells was comparable irrespective of FGF9 expression (Supplementary Figs. S12A and S12B), it was significantly higher in FGF9-transduced H2228 cells (H2228-FGF9) than in the H2228 cells transduced with an empty vector (H2228-empty; Fig. 6D). Histologic analysis revealed that H2087 and H2110 cells formed moderately differentiated adenocarcinoma tumors regardless of FGF9 expression (Supplementary Fig. S12C). In contrast, H2228 cells exhibited histological differences depending on FGF9 expression. Compared with tumors generated from H2228-empty cells, which formed well-differentiated adenocarcinomas, H2228-FGF9 cells formed poorly differentiated adenocarcinomas. The palisading pattern, a characteristic of neuroendocrine differentiation, was observed in H2228-FGF9 tumors but not in H2228-empty tumors (Fig. 6E). IHC revealed that the neuroendocrine markers SYP and CHGA were expressed in H2228-FGF9 tumors but not in H2228-empty tumors (Fig. 6F), indicating neuroendocrine differentiation in established adenocarcinoma cells. Collectively, the findings provide direct evidence that prospective FGF9 induction functionally contributes to the transdifferentiation of adenocarcinoma to SCLC in established human lung cancer cells.

In this study, we identified FGF9 upregulation in lung cancer cases of adenocarcinoma to SCLC transdifferentiation. Although the number of cases evaluated in this study was small (n = 6), upregulation of FGF9 was found in 66.7% of patients examined. In addition, aberrant FGF9 expression was observed in a significant proportion of primary SCLC, and it was positively correlated with neuroendocrine marker genes. FGF9 is a developmentally important FGF and its expression is repressed in normal adult lungs. In the pseudoglandular stage of lung development, FGF9 activates signaling pathways to direct epithelial specification and lung branching (52). Recently, the oncogenic roles of FGF9 in multiple cancer types, including lung cancer, have been reported by our research group and other groups (33, 44). However, the role of FGF9 upregulation in primary and transdifferentiated SCLC remain elusive. Although robust upregulation of FGF9 mRNA was observed, no genetic alterations such as FGF9 mutations or amplification was identified by WES, indicating that the FGF9 expression level was regulated through epigenetic mechanisms. Recently, long noncoding RNA H19- or microRNA-mediated regulation of the expression of FGF9 has been reported in multiple types of cancers, including SCLC (53–55); in addition, the high expression of H19 was reported in ASCL1 positive cell lines (56). To understand the mechanism of FGF9 upregulation in primary and transdifferentiated SCLC, further epigenetic evaluation such as noncoding RNA, DNA methylation, and chromatin modification studies should be performed.

Interestingly, our cases with transdifferentiation had heavy smoking history, in contrast to the light smoking history of other cohorts (29, 30), and several case reports (14, 15). Tobacco smoking may affect the susceptibility to transdifferentiation. However, to verify the effects of tobacco smoking on transdifferentiation, further clinical evaluations are required.

Transdifferentiation was prospectively demonstrated in multiple mouse and human lung adenocarcinoma-derived cells. To our knowledge, this is the first report that provides direct evidence for transdifferentiation. However, the level of transdifferentiation was variable among cells. In addition, transdifferentiation was not induced in the H2087 and H2110 cells even with TP53 and the RB1 inactivation and FGF9 overexpression. EML4-ALK positive H2228 has transdifferentiated to SCLC, which may explain that transdifferentiation can also occur in ALK-translocated NSCLC after ALK-TKI treatment (10). These findings indicate that specific genetic or epigenetic contexts may affect the level of transdifferentiation. Interestingly, SCLC transdifferentiation was most frequently observed in lung adenocarcinoma with EGFR-mutations after long-term EGFR-TKI treatment, when the cancer cells become resistant to EGFR pathway inhibition. Thus, EGFR pathway inhibition may be required for transdifferentiation. In this study, this was supported by the downregulation of EGFR mRNA levels in the RNA-seq data of case #1 and the upregulation of the ASCL1 gene in H1975 osiR cells. The effect of EGFR pathway inhibition in transdifferentiation needs further evaluation.

Clinically, the transdifferentiated patients with SCLC are treated using systemic chemotherapy, including platinum-etoposide or platinum-irinotecan, which is administered in patients with primary SCLC (57). The recent approval of combination therapy with immune-checkpoint inhibitors as first-line treatment for extensive disease SCLC provides another choice of treatment for transdifferentiated patients with SCLC (58, 59). However, the efficacy of immune-checkpoint inhibitors for transdifferentiated SCLC should be further evaluated, because none of 17 patients who received immune-checkpoint inhibitors experienced a response in a retrospective cohort (29). To improve the prognosis of transdifferentiated patients with SCLC, further identification of potential drug targets and the development of a novel therapeutic strategy are needed. The paucity of druggable genetic or epigenetic alterations has hampered the development of molecular targeted therapy in SCLC. Interestingly, we have observed cell-autonomous upregulation of Fgfr1 in trans-differentiated MLE12-FGF9 tumors. Similar cell-autonomous upregulation of Fgfr1 was observed in a genetically engineered mouse lung adenocarcinoma model (60). The upregulation of FGFR1 may be relevant with the finding that amplification of FGFR1 was observed in approximately 6% of primary SCLC (24). In addition, several studies have suggested the oncogenic roles of FGFR1 in SCLC (23, 61, 62). The upregulation of the FGF9–FGFRs axis may open a new window for the use of FGFR inhibitors in the treatment of transdifferentiated SCLC, as supported by the in vivo experiment using AZD4547. To evaluate the efficacy of FGFR inhibitors for the treatment of transdifferentiated SCLC, clinical prospective studies are needed.

In conclusion, we provide preclinical and clinical evidence for FGF9-mediated SCLC transdifferentiation and propose a potential avenue for treatment of transdifferentiated SCLC.

No disclosures were reported.

Kota Ishioka: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Hiroyuki Yasuda: Conceptualization, data curation, supervision, funding acquisition, writing–original draft, writing–review and editing. Junko Hamamoto: Conceptualization, data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Hideki Terai: Conceptualization, data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Katsura Emoto: Data curation, validation, investigation, visualization, methodology. Tae-Jung Kim: Data curation, validation, investigation, visualization, methodology. Shigemichi Hirose: Data curation, validation, investigation, visualization, methodology. Takashi Kamatani: Investigation. Sachiyo Mimaki: Investigation. Daisuke Arai: Investigation. Keiko Ohgino: Investigation. Tetsuo Tani: Investigation. Keita Masuzawa: Investigation. Tadashi Manabe: Investigation. Taro Shinozaki: Investigation. Akifumi Mitsuishi: Investigation. Toshiki Ebisudani: Investigation. Takahiro Fukushima: Investigation. Mari Ozaki: Investigation, writing–review and editing. Shinnosuke Ikemura: Investigation. Ichiro Kawada: Investigation. Katsuhiko Naoki: Investigation. Morio Nakamura: Investigation. Takashi Ohtsuka: Investigation. Hisao Asamura: Validation, investigation. Katsuya Tsuchihara: Writing–review and editing. Yuichiro Hayashi: Validation, Investigation, visualization, methodology. Ahmed E. Hegab: Investigation, writing–review and editing. Susumu S. Kobayashi: Investigation, writing–review and editing. Takashi Kohno: Investigation, writing–review and editing. Hideo Watanabe: Formal analysis, investigation, visualization, methodology. David M. Ornitz: Validation, methodology, writing–review and editing. Tomoko Betsuyaku: Conceptualization, supervision, funding acquisition, project administration, Writing–review and editing. Kenzo Soejima: Conceptualization, supervision, funding acquisition, investigation, project administration, writing–review and editing. Koichi Fukunaga: Supervision, validation, project administration, writing–review and editing.

This work was supported in part by the Japan Society for the Promotion of Science to K. Soejima (19H03671), T. Betsuyaku (15H04833), H. Terai (18K08184), J. Hamamoto (19K08610), and H. Yasuda (17K09667). This work was also supported in part by the Japan Agency for Medical Research and Development (AMED), grants 20cm0106576 and 20ck0106471 to H. Yasuda, by the Takeda Science Foundation (to H. Terai and H. Yasuda), and by NIH to D.M. Ornitz (HL11119008). We thank Ms. Mikiko Shibuya for her excellent technical assistance. We also thank the Collaborative Research Resources at the Keio University School of Medicine for assistance with cell sorting.

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.