Involvement of the FGF8 and Fibroblast Growth Factor Receptor signaling pathway in the maintenance and progression of fusion-positive rhabdomyosarcoma

Abstract

We previously used a myoblast model of fusion-positive rhabdomyosarcoma (FP-RMS) to show that FGF8, a PAX3-FOXO1 (P3F) transcriptional target, is required for P3F-driven tumorigenicity and, when aberrantly expressed, can maintain tumorigenicity in P3F-independent recurrent tumors. We report in this study that FGF8, FGFR1 and FGFR4 are often highly expressed in FP-RMS tumors. High FGF8 expression in FP-RMS cells is associated with high sensitivity to an FGFR4 inhibitor and a pan-FGFR inhibitor. While downregulating FGF8 resulted in loss of sensitivity to these inhibitors, FGF8 upregulation in myoblasts decreased FGFR4 expression and sensitized the cells to an FGFR1 inhibitor and a pan-FGFR inhibitor. FGF8 downregulation of FGFR4 expression was reverted by inhibitors of FGFR1, MEK or ERK, thus defining a signaling pathway by which FGF8 mediates this regulatory effect. Finally, high FGF8 expression in P3F-independent recurrent tumors was attributable to a rearrangement of viral LTR sequences into the FGF8 3’ UTR region, resulting in increased FGF8 mRNA stability. These findings indicate that FGF8 exerts oncogenic effects in FP-RMS via FGFR4 and may exert oncogenic effects in P3F-independent relapses via FGFR1. Our study reveals the functional significance of FGF8 in FP-RMS and provides a rationale for preclinical studies of FGFR inhibitors in FP-RMS.

Introduction

Fusion-positive (FP) rhabdomyosarcoma (RMS) is a highly aggressive pediatric soft tissue cancer for which the conventional treatment based on chemotherapy, radiotherapy and surgery is often ineffective, thereby resulting in frequent relapses. FP-RMS is distinguished by a recurrent 2;13 chromosome translocation that joins the PAX3 and FOXO1 genes and less frequently by a 1;13 translocation that joins the PAX7 and FOXO1 genes (1,2). These translocations generate PAX3::FOXO1 (P3F) or PAX7::FOXO1 fusion genes that encode novel transcription factors with potent oncogenic activity. Although the P3F fusion protein is responsible for much of the tumorigenic activity of FP-RMS, there is currently no treatment that directly targets this fusion protein. Using a human myoblast system that models targeted therapy directed against P3F, we demonstrated that expression of the fibroblast growth factor 8 (FGF8) gene, a direct transcriptional target of P3F (3,4), is required for the tumorigenic activity of primary tumors expressing P3F. In addition, we found that maintenance of FGF8 expression is sufficient for the development of recurrent tumors following P3F depletion in this model system and that FGF8 is required for proliferation, transformation, migration, and invasion of the recurrent tumor-derived (TD) cells (5).

FGF8 contributes to the tumorigenicity of several human cancers and is associated with poor prognosis and resistance to chemotherapy and radiotherapy (6–9). FGF8 signals by binding to cell surface receptors (FGFR1, FGFR2, FGFR3 and FGFR4) (10) and thereby modulates downstream pathways controlling cell proliferation, differentiation, survival migration and invasion. These FGF receptors (FGFRs) are known to be involved in the progression of many cancer types including RMS tumors (11–13). Stimulation of the FGF/FGFR pathway leads to the activation of numerous signaling molecules involved in tumorigenesis including RAS–RAF–MAPK, PI3K–AKT, and phospholipase Cγ. These pathways can be modulated by pharmacologically targeting the FGFRs with several small molecules, which have been reported to be effective in inhibiting growth and/or survival of several cancers (9,14,15).

In this paper, we conducted a comprehensive expression study of the FGF and FGFR families in RMS tumor samples and cell lines. We then used potent FGFR inhibitors to dissect the mechanisms by which FGF8, FGFR1 and FGFR4 function in the growth and tumorigenicity of FP-RMS tumors. By utilizing our cell culture models of primary FP-RMS tumors and P3F-independent recurrent tumors, we investigated novel downstream pathways induced by FGF8 in RMS tumors. Finally, we investigated the molecular mechanism responsible for high FGF8 expression in our model of P3F-independent recurrent tumor cells. Our data provide evidence that the expression status of FGF8 could be an indicator of tumor sensitivity to FGFR inhibitors in primary tumors and thus these inhibitors may be helpful in future clinical trials to treat FP-RMS tumors and prevent the development of recurrent tumors.

Materials and Methods

Cell culture

Human RMS and myoblast cell lines were cultured as described previously (16,17). The sources of the RMS cell lines are as follows: RH10 (RRID: CVCL_8750) - P. Houghton, 2015; RH30 (RRID: CVCL_0041) - ATCC (Manassa, VA), 1998; RH28 (RRID: CVCL_8752) - B. Emanuel, 1992; RH5 (RRID: CVCL_5917) - J. Khan, 2013; CW9019 (RRID: CVCL_N820) - J. Biegel, 1995; RH41 (RRID: CVCL_2176) - C. Linardic, 2014; MP4 - T. Cripe, 2015; NCI097 - L. Helman, 2015. A vial of RH30 cells provided in the remote past (1992) by E. Douglass was found to express low levels of P3F (1,2) but have an identical short tandem repeat genotype as the RH30 cells obtained from ATCC; these cells have been designated as “RH30 low P3F” [5]. The source of the human myoblast cell lines is as follows: Dbt - D. Trono, 2004; HuN2 and Human#2 - G. Pavlath (HuN2, 2007; Human#2, 2014); 18TL269121, 629287_D and 639629_D - Lonza (Basel, CH), 2019 ; B6M_new - B. Schaefer, 2018; HSMM3 and HSMM_b - C. Linardic, 2019. Primary and recurrent tumor-derived (TD) cells, Dbt/MYCN-EV and Dbt/MYCN-FGF8 cell were generated as previously described (5,18). Cell lines were authenticated using the short tandem repeat genotyping analysis with the AmpFLSTR profiler plus PCR amplification kit from Applied Biosystems (RRID:SCR_005039). Cell lines were periodically checked with a PCR-based Mycoplasma detection kit (ATCC, # 30–1012K) (RRID:SCR_001672) to rule out Mycoplasma contamination.

RNA extraction, sequencing, and gene expression profiling

Total RNA was extracted using the QIAGEN AllPrep Kit (RRID:SCR_008539). RNA-Seq samples were pooled and sequenced on HiSeq4000 (RRID:SCR_016386) using Illumina TruSeq Stranded Total RNA Library Prep and paired-end sequencing. The samples have 55 to 81 million pass filter reads with more than 94% of bases above the quality score of Q30. The RNA-Seq dataset was processed using the NIH CCBR Pipeliner utility (https://github.com/CCBR/Pipeliner) (RRID:SCR_003171). The quality of the sequencing reads was checked with FastQC version 0.11.5 (RRID:SCR_014583). After trimming the adapter sequence and removing low-quality reads with Cutadapt version 1.18 (RRID:SCR_011841), the sequences were aligned to the human reference genome (hg19/GRCh37) using STAR version 2.7.0f (RRID:SCR_004463) in two-pass mode (19,20). The genes and isoforms of the aligned sequence fragments were quantified with RSEM version 1.3.0 (RRID:SCR_000262) (21). The counts were normalized to the library size, and the voom algorithm from the limma R package was used for differential gene expression analysis (22). Those genes with log2 expression difference of greater than 1.5 and adjusted p-value of less than or equal to 0.05 were considered significantly different for the presented results.

The RMS RNA-Seq data were obtained from the OncoGenomics database (https://pob.abcc.ncifcrf.gov/cgibin/JK). RNA-seq was performed on the Illumina HiSeq2000 Sequencing System (RRID:SCR_020130) and the sequenced transcripts were mapped and analyzed with the TopHat2 pipeline (23). The relative abundance of mapped transcripts was further estimated with Cufflinks (RRID:SCR_014597) (24). The subsequent counts in fragment per kilobase of transcript per million mapped reads (FPKM) values were normalized, log2 transformed and used for the gene expression analysis.

Protein extraction and Western Blotting

Protein extraction and western blotting were performed as described previously (25). Membranes were incubated overnight with antibodies against FOXO1 (1:1000, Cell Signaling Technology Cat# 2880) (RRID:AB_2106495), FGFR1 (1:1000, Cell Signaling Technology Cat# 9740) (RRID:AB_11178519), FGFR2 (1:1000, Cell Signaling Technology Cat# 23328) (RRID:AB_2798862), FGFR3 (1:1000, Cell Signaling Technology Cat# 4574) (RRID:AB_2246903), FGFR4 (1:1000, Cell Signaling Technology Cat# 8562) (RRID:AB_10891199), FGF8 (1:1000, #16124, Sino Biological, Wayne, PA), pan-phosphotyrosine (Thermo Fisher Scientific Cat# MA5–38240) (RRID:AB_2898156) and GAPDH (1:2000, Santa Cruz Biotechnology Cat# sc-47724) (RRID:AB_627678), which was used as a loading control.

Immunoprecipitation analysis

For immunoprecipitation, whole cell protein extracts were obtained using NP-40 lysis buffer- 150 mM NaCl, 1% NP-40, 50 mM Tris-HCl pH 8.0, 0.15% (w/v) BSA, 10% (v/v) glycerol, protease and phosphatase inhibitor cocktail by vortexing cells every 5 min on ice for 30 min. Protein A/G Magnetic beads (Thermo Fisher Scientific, # 88802) (RRID:SCR_008452) were equilibrated with lysis buffer and lysates (0.30 mg) were pre-cleaned by incubation with protein A/G beads for 1h on a rotator at 4 °C. Next, 2 μg of antibody was added to the pre-cleared lysates and incubated on a rotator at 4 °C overnight. The following day, Protein A/G Magnetic beads were equilibrated with IP wash buffer (NP-40 lysis buffer containing 300 mM NaCl) added to the lysate-antibody mix for 1 h. Beads were washed three times, 5 min each, in IP wash buffer, resuspended in 50 μL of 2X loading buffer, boiled at 100 °C for 5 min, and subjected to SDS–PAGE and immunoblotting.

Lentiviral transduction

Lentiviruses were produced as previously described (5). All-in-one lentiCRISPR-v2-puro CRISPR/Cas9 with guide RNA (gRNA) targeting FGF8 and the non-targeting control (CR-control) vectors were purchased from GenScript (RRID:SCR_002891). For inducible knockdown experiments, we started with a plasmid containing a RB1 CRISPR gRNA (gift from Adam Karpf) (RRID:Addgene_87836) (26). The RB1 gRNA was removed and replaced with two different gRNAs targeting PAX3. Each gRNA was under a separate U6 promoter while the Cas9 and GFP were regulated by the same inducible promoter. gRNA sequences are listed in Supplementary Table 1.

Growth, transformation and clonogenic assays

For cell growth, a real-time imaging system (Sartorius IncuCyte S3 Live Cell Analysis System) (RRID:SCR_023147) was used to monitor cell density in culture up to 120 hours. Depending on cell line type, 500 to 4000 cells were seeded per well in 96-well clear bottom plates. The following day, growth medium was replaced with medium containing inhibitor (at the indicated concentration) or DMSO (control). Plates were imaged in the IncuCyte S3 system, with automatic recording every 6 hours under phase contrast and using 4x magnification. Cell confluency was assessed and analyzed using the IncuCyte software (Essen BioScience). Colony and focus assays were performed as previously described (5). The anti-FGFR1 inhibitor PD173074 (27), and the anti-FGFR4 inhibitor H3B6527 (28) were purchased form Selleck Chemicals LLC (RRID:SCR_003823). The pan-FGFR inhibitor Erdafitinib, ERK inhibitor SCH772984 and MEK inhibitor Trametinib were purchased form MedChemExpress (RRID:SCR_025062).

RNA decay assays

FGF8 mRNA decay was analyzed in Dbt primary and recurrent TD cells. Cells seeded in 6 well plates were treated with Actinomycin D (29) at 5 μg/mL in triplicates and then collected at 2, 8 and 24 hours for RNA isolation and cDNA synthesis. Untreated primary or recurrent cells were used as the time zero-point control. Real-time qPCR was performed using primers to: FGF8 and MYC (Supplementary Table 2). mRNA decay rate was determined by non-linear regression curve fitting using GraphPad Prism (Domatics, Boston, MA).

Mapping of the FGF8 3’ UTR end

The FGF8 3’ UTR ends were mapped in primary and recurrent TD cells using Rapid Amplification of cDNA Ends (RACE), as described previously (30). Briefly, cells cultured under normal conditions were scraped and RNA was isolated using the phenol/chloroform method. After DNase treatment, cDNA was synthesized with SuperScript IV Reverse Transcriptase (Thermo Fisher Scientific) (RRID:SCR_008452) primed with an oligo dT25 primer. PCR was performed using the forward primer (FGF8 Forward-1) and T7 reverse primer, and then a second (nested) PCR was run using the forward primer (FGF8 Forward-2) and T7 reverse primer. The PCR product was fractionated on an agarose gel and the imaged bands were excised, purified, cloned into the pCR™4-TOPO™ TA vector following the manufacturer’s instructions (Thermo Fisher Scientific) (RRID:SCR_008452) and sequenced. Primer sequences are listed in Supplementary Table 1.

Plasmids and cell transfections

The dual luciferase reporter plasmid psiCHECK2-miR-34 WT was a gift from Joanne Weidhaas (RRID:Addgene_78258) (31). The miR-34 WT region was excised and replaced with one of the following inserts: FGF8 3’ UTR-Full, FGF8 3’ UTR-Short-1, FGF8 3’ UTR-Short-2, FGF8 3’ UTR-LTR, and LTR only. All gene reporter assays were performed using parental Dbt/MYCN/iP3F cells (clone 2) (18). Cells were plated in 12-well plates and transfected in triplicate 24 hours later using Lipofectamine 3000 reagent following the manufacturer’s protocol (Invitrogen, Waltham, MA). Cells were incubated an additional 48 hours before assaying for luciferase activity, using a dual luciferase assay kit (Promega) (RRID:SCR_006724).

Data availability

The RNA-Seq data described in this article were deposited in and are available from the dbGaP database under accession phs004009.v1.

Results

1. Expression of FGF8 and FGFRs in RMS tumors and myoblast model system

We recently demonstrated that FGF8 is involved in the proliferation and transformation of FP-RMS tumor cells as well as in the tumorigenicity of primary and recurrent tumors in our myoblast model system (5). To better understand the role of FGF8 signaling in RMS tumors, we sought to analyze the expression of the genes encoding FGF8 and its known receptors (FGFR1, FGFR2, FGFR3 and FGFR4) in a large cohort of human primary RMS tumors and cell lines.

Overall, our analysis of the RNA-Seq data indicates that the expression of FGF8 is generally high in FP-RMS tumors and cell lines compared to very low expression in fusion-negative (FN)-RMS tumors and cell lines and myoblasts (Figure 1A-D, Supplementary Figure 1A). For the FGFRs, FGFR1 and FGFR4 are expressed at high levels in FP-RMS, FN-RMS and myoblasts. FGFR2 and FGFR3 are expressed at higher levels in FP-RMS than FN-RMS, but lower overall compared to FGFR1 and FGFR4 (Supplementary Figure 1B-C). There is a moderate to strong correlation (R² values from 0.39 to 0.81) between RNA and protein expression of FGF8, FGFR1 and FGFR4 in the FP-RMS cell lines (Figure 1B and 1E, Supplementary Figure 1D-G). Comparison of the FGF8, FGFR1 and FGFR4 RNA levels in the FP-RMS tumors and cell lines reveals greater variability of FGF8 expression compared to FGFR1 and FGFR4 expression (Figure 1C-D); this variability in FGF8 expression is also noted at the protein expression level (Figure 1E).

Figure 1. FGF8 and FGFR expression in RMS tumors and cell lines.

A-B. Heatmaps showing mRNA expression of FGF8 and FGFR family members in fusion-positive (n=38) and fusion-negative (n=67) tumors (A) and cell lines (B). Each column represents an individual cell line or tumor sample. In the top row, fusion status is shown as fusion-positive (FP) (black) or fusion-negative (FN) (grey). For all genes, transcript expression values were determined from RNA-Seq experiments and expression levels are represented by colors according to the scale shown at the right side of each heatmap. C, D. Box and whisker plots of FGF8, FGFR1 and FGFR4 RNA expression in RMS tumors (C) and RMS cell lines and myoblasts (D). ACTB expression was used as a comparative control. The two-sided unpaired Student’s t-test was applied to determine significant differences between the groups (* P<0.05; ** P<0.01; *** P<0.001; Ns, not significant). E. Western blot analysis of P3F, FGF8, FGFR1 and FGFR4 protein expression in FP-RMS cell lines. GAPDH expression was used as a control for protein loading normalization. F. FGFR4-phosphotyrosine content in FP-RMS cell lines. FGFR4 was immunoprecipitated from the indicated cell lines, and the eluted proteins or the input samples were analyzed by Western blot with antibodies to the indicated proteins. G-I. Violin plot showing expression of FGF8 (G), FGFR1 (H) and FGFR4 (I) genes across the tumor cell states previously identified in a single cell RNA-Seq study of FP-RMS (32). A total of 22,526 single cells derived from P3F FP-RMS samples were included in this analysis, and the tumor cell states were identified by UMAP unsupervised clustering projection and annotated according to the expression of specific gene markers as described in the study. The box limits indicate the first and third quartiles. Abbreviations: P, progenitor; Pr, proliferative; Gr, ground; Dif, differentiated; Neu, neuronal; Apo, apoptosis.

Tyrosine phosphorylation is a hallmark of receptor tyrosine kinase activation. Comparison of FP-RMS lines RH30 and RH41 with high FGFR4 and FGF8 expression to lines CW9019 and NCI097 with low FGFR4 and FGF8 expression revealed higher FGFR4-phosphotyrosine levels in the former lines (Figure 1F). These FGFR4-phosphotyrosine differences could be due to intrinsic differences in FGFR4 and/or FGF8 expression and may predict differential vulnerability to FGFR inhibitors.

Recent single cell RNA sequencing studies of FP-RMS identified subpopulations within these tumors (32). Reanalysis of published single cell RNA-Seq data showed that FGF8 and FGFR4 are most highly expressed in “Proliferative” and “Ground” subpopulations of FP-RMS tumors (Figure 1G-I). In contrast, FGFR1 expression is highest in the “Neuronal” subpopulation, in which previous studies demonstrated that P3F target signature was also most highly expressed.

2. FGF8 is a potential marker of FP-RMS sensitivity to FGFR inhibitors

We next investigated the role of FGF8, FGFR1 and FGFR4 in the proliferation of FP-RMS cells. As FGF8 binds to all FGFRs (33), we selected small molecule inhibitors specific for FGFR1 (PD173074) and FGFR4 (H3B6527) in addition to a potent pan-FGFR inhibitor (Erdafitinib) for these studies. We conducted a real-time growth assay on multiple FP-RMS cell lines and determined the IC₅₀ for each inhibitor in each cell line (Figure 2A-C). The most striking finding was a strong association between FGFR4 inhibitor sensitivity and FGF8 expression level such that the high FGF8-expressing lines were much more sensitive to the FGFR4 inhibitor than the low FGF8-expressing lines (p<0.05). The finding of decreased sensitivity to the FGFR4 inhibitor following FGFR4 depletion in RH41 cells confirms the specificity of this drug for FGFR4 (Figure 3A-B). While the high FGF8 tumor cells are sensitive to the pan-FGFR inhibitor, FGF8 expression was not significantly associated with a difference in sensitivity to the pan-FGFR and FGFR1 inhibitors in the FP-RMS cell lines.

Figure 2. Effect of FGFR inhibitors on FP-RMS cell growth.

A-C. Potency of FGFR1 (A), FGFR4 (B) and pan-FGFR (C) inhibitors in FP-RMS cell lines. Depending on the cell line, 500–4000 cells were seeded in clear-bottomed 96-well plates in 100 μl growth media. The following day cells were treated in assay medium with control (DMSO) or with a serial dilution of one of the three inhibitors (from 5 nM to 4 μM). DMSO was adjusted to 0.04% as the final concentration in all wells. Each time point was represented by 3–6 replicates. Cell growth was monitored by the IncuCyte live imaging system and IC₅₀ was calculated at 120 h using GraphPad software. The one-way ANOVA test was applied to determine significant differences between the high and low FGF8 groups, and the statistical significance is described in the top left beneath the inhibitor name. D-F. Effect of FGFR1 (D), FGFR4 (E), and Pan FGFR (F) inhibitors on colony formation in the RH30 (red) and CW9019 (black) FP-RMS cell lines. Colony formation assays are shown in Supplementary Figure 2. A linear regression [colony count versus log10 (inhibitor concentration)] was fitted, and the regression equations (in the form of y=b+mx) and correlation (R2) are shown. The m values represent the slope of the line and describe the impact of the inhibitor on colony formation. The statistical significance of the correlation coefficients was assessed by the F-test.

Figure 3. Impact of P3F, FGFR4 and FGF8 expression on the sensitivity of FP-RMS cells to FGFR inhibitors.

A. Western blot analysis of RH41 cells after transduction of lentivirus containing CRISPR/Cas9 constructs targeting FGFR4 or a non-targeting control construct. B. Relative IC₅₀ of FGFR4 inhibitor on RH41 cells with FGFR4 depletion as described in part A. C. P3F, FGF8, FGFR1 and FGFR4 expression in RH30 cells with high or low P3F expression levels. D. Relative IC₅₀ of FGFR1, FGFR4 and pan-FGFR inhibitors on RH30 cells with high or low P3F expression levels. E. Western blot analysis of P3F, FGF8, FGFR1 and FGFR4 expression after transducing RH30 and RH41 cells with doxycycline (dox)-inducible CRISPR/Cas9 constructs targeting P3F expression. F-G. Relative IC₅₀ of FGFR1, FGFR4, and pan-FGFR inhibitors on RH30 (F) and RH41 (G) with doxycycline-induced P3F depletion (+) compared to uninduced cells (–). H. Western blot analysis of FGF8 and FGFR family expression in RH30 high P3F cell line after transduction of lentiviruses containing CRISPR/Cas9 constructs targeting FGF8 or a non-targeting control construct. Three FGF8-targeted (CR-FGF8) subclones and one control (CR-control) are displayed. I. Relative IC₅₀ of FGFR1, FGFR4 and pan-FGFR inhibitors in RH30 FGF8-targeted and control subclones. For all Western blots (A, C, E, H), GAPDH expression was used as a control for protein loading normalization. For treatment with inhibitors (B, D, F, G, I), 500–1000 cells were seeded in 3 to 6 replicates. The growth was monitored by the IncuCyte live imaging system and IC₅₀ was calculated at 120 h using GraphPad software. Relative IC₅₀ is displayed as a ratio of the test condition compared to each control. The data displayed are representative of separate experiments that were performed three or four times. The two-sided unpaired Student’s t-test was applied to determine significant differences between the control and test groups (* P<0.05; ** P<0.01; *** P<0.001; ⁺ 0.06≥p≥0.05; Ns, not significant). Since different lots of inhibitors were used, a batch effect was noticed but did not affect the overall results of the experiments.

To further understand the involvement of the FGF pathway in RMS cell growth and survival, we performed colony formation assays using the three inhibitors at different concentrations. The assays showed that the RH30 cell line, which expressed high FGF8 levels, is more sensitive than the low FGF8-expressing CW9019 cell line to the FGFR4 and pan-FGFR inhibitors but comparably sensitive to the FGFR1 inhibitor (Figure 2D-F, Supplementary Figure 2A-B). The colony assay results for these two cell lines corroborate the growth assay results and show that high FGF8 expression is often associated with high sensitivity to the FGFR4 and pan-FGFR inhibitors in FP-RMS tumor cells.

3. FP-RMS sensitivity to FGFR inhibitors is dependent on FGF8 expression in the tumor cells.

As the FGF8 gene is a direct transcriptional P3F target, we asked whether the sensitivity to these FGFR inhibitors is linked to FGF8 and P3F expression in FP-RMS. First, we compared drug sensitivity between RH30 cells with high P3F expression to a spontaneous non-transformed variant of this cell line that stably expresses low levels of P3F along with its direct transcriptional targets FGF8 and FGFR4 (Figure 3C) (5). In the real-time growth assay, this low P3F variant showed a much lower sensitivity to the FGFR4 and pan-FGFR inhibitors compared to the RH30 high P3F cells (Figure 3D). Second, to model a P3F targeted therapy, we depleted P3F expression using a doxycycline-inducible CRISPR/Cas9 system in two FP-RMS cell lines (Figure 3E, Supplementary Figure 3A). In both RH30 and RH41, P3F depletion resulted in FGF8 and FGFR4 downregulation and rendered the cells less sensitive to the FGFR1, FGFR4 and pan-FGFR inhibitors (Figure 3F-G).

To investigate whether FGF8 loss is sufficient to recapitulate the change in FGFR inhibitor sensitivity seen with P3F expression loss, we used the CRISPR/Cas9 system to target FGF8 in the RH30 cell line and performed drug sensitivity assays using selected clones with very low FGF8 expression (Figure 3H). The results showed that depleting FGF8 in RH30 cells resulted in a lower sensitivity to the FGFR4 and pan-FGFR inhibitor but not to the FGFR1 inhibitor (Figure 3I). After depleting FGF8, there was no significant change in the expression of FGFR1, FGFR2 or FGFR3. However, the clones lost sensitivity to FGFR4 and pan-FGFR inhibitors, despite an increase in FGFR4 expression in all three FGF8-depleted clones compared to the control clones. The relationship between FGF8 and FGFR4 expression will be further explored in the next section. This finding confirms that FGF8 expression in FP-RMS cells is necessary for the high sensitivity of FP-RMS cells to the FGFR4 and pan-FGFR inhibitors.

To recapitulate our finding with the FP-RMS cells, we then verified whether FGF8 is important in the FGF/FGFR pathway inhibition using the same FGFR inhibitors in our myoblast model system. We first induced P3F expression in Dbt/MYCN/indP3F parental cells, which stimulated the expression of FGF8 and FGFR4 but not FGFR1 (5) (Figure 4A, Supplementary Figure 3B). Compared to unstimulated cells, this change in expression significantly augmented cell sensitivity to the FGFR4 and pan-FGFR inhibitor but not to the FGFR1 inhibitor (Figure 4B).

Figure 4. Impact of P3F and FGF8 expression on the sensitivity of transduced myoblasts to FGFR inhibitors.

A. P3F, FGFR1, FGFR4 and FGF8 protein expression in parental Dbt/MYCN/iP3F cells after doxycycline (dox)-induced P3F expression (+) compared to unstimulated (–) cells. In the FGF8 panel, there is a non-specific band (marked by “*”) that occurs with some batches of the FGF8 antibody. B. Comparison of the IC₅₀ for FGFR1, FGFR4 and pan-FGFR inhibitors between Dbt/MYCN/indP3F cells with (+) and without (–) doxycycline induction of P3F expression. C. FGF8, FGFR1 and FGFR4 expression in Dbt/MYCN-FGF8 cells compared to the Dbt/MYCN-EV cells. Dbt/MYCN-FGF8 and Dbt/MYCN-EV cells were generated after stably transfecting FGF8 cDNA or an empty control construct (EV) into the Dbt/MYCN cells as previously described (5). D. IC₅₀ of FGFR1, FGFR4 and pan-FGFR inhibitors in Dbt/MYCN-FGF8 cells compared to Dbt/MYCN-EV cells. In B and D, the two-sided unpaired Student’s t-test was applied to determine significant differences between the control and test groups (* P<0.05; ** P<0.01; ⁺ 0.06≥p≥0.05; Ns, not significant).

We next asked whether FGF8 is sufficient to sensitize the cells to the FGFR inhibitors in the absence of P3F expression. To address this question, we transfected the FGF8 cDNA into the Dbt/MYCN cells, which resulted in decreased expression of FGFR4 but no substantial change in FGFR1 expression compared to the control Dbt/MYCN-empty vector (Dbt/MYCN-EV) cells (Figure 4C). In this case, the Dbt/MYCN-FGF8 cells were more sensitive to FGFR1 and pan-FGFR inhibitors (Figure 4D). However, despite the decrease in FGFR4 expression, there was no change in sensitivity to the FGFR4 inhibitor. These results support the premise that FGF8 is sufficient to activate the FGF/FGFR pathway and thus sensitize the cells to the FGFR inhibitors. Furthermore, these results suggest that in the presence of P3F, FGF8 primarily signals through FGFR4 but in the absence of P3F (as would occur when FGF8 is upregulated by a P3F-independent mechanism), FGF8 primarily signals through FGFR1.

4. FGF8 downregulates FGFR4 through the FGF1/MEK/ERK pathway.

The finding of decreased FGFR4 expression in Dbt/MYCN-FGF8 cells (Figure 4C), and increased FGFR4 expression in RH30 cells following FGF8 loss (Figure 3H) suggests that FGF8 is involved in the regulation of FGFR4 expression. To verify this hypothesis, we used RNA-Seq to measure the expression of the four FGFRs in multiple Dbt/MYCN-FGF8 clones. We found that FGFR4 is downregulated in all the clones with high FGF8 compared to their respective EV controls, whereas FGFR1, FGFR2 and FGFR3 expression remains unchanged (Figure 5A). To also consider the effect of P3F on FGFR4 expression, we used a Dbt clone with inducible P3F (Dbt/MYCN/iP3F) and then stably transfected this clone with an FGF8 expression construct or empty vector control. Protein and RNA expression studies revealed that transfected FGF8 overexpression is capable of reducing FGFR4 expression at lower P3F levels, but the level of inhibition is reduced at higher P3F levels (Figure 5B, Supplementary Figure 4A-D), suggesting that FGFR4 stimulation by P3F counters FGFR4 inhibition by FGF8. The effect of FGF8 on FGFR4 expression will thus vary depending on the P3F level in the tumor cells and will be most prominent in tumors that have lost or have low residual P3F expression. Finally, stable transfection of FGF8 into the RH30-low P3F cells also resulted in a loss of FGFR4 expression (Figure 5C). The results indicate that FGF8 negatively affects FGFR4 RNA and protein expression, both in the presence or absence of P3F expression.

Figure 5. FGF8 controls FGFR4 expression through the FGFR1/MEK/ERK pathway.

A. Differential expression of FGF8 and FGFR family transcripts in Dbt/MYCN-FGF8 cells (Blue) compared to Dbt/MYCN-EV cells (Grey). B. P3F, FGF8, FGFR1 and FGFR4 protein expression in Dbt/MYCN/iP3F cells stably transfected with an FGF8 expression construct or EV control construct and treated with varying levels of doxycycline (Dox). C. FGF8, FGFR1 and FGFR4 protein expression in RH30 low P3F cells transfected with FGF8 expressing vector compared to EV control cells. D-F. Expression of FGFR1 and FGFR4 in Dbt/MYCN-FGF8 cells at the mRNA (D,E) and protein (F) levels after treatment with FGFR1, FGFR4, pan-FGFR, MEK and ERK inhibitors compared to the DMSO control. All drugs were used at 100 nM final concentration for a period of 72 hours. In D and E, the two-sided unpaired Student’s t-test was applied to determine significant differences between the control and test groups (* P<0.05; ** P<0.01; NS, not significant).

To elucidate the mechanism by which FGF8 regulates FGFR4 expression, we treated the Dbt/MYCN-FGF8 cells with each of the three FGFR inhibitors (at a concentration of 100 nM), and then measured FGFR1 and FGFR4 expression at both the RNA (Figure 5D-E) and protein levels (Figure 5F, Supplementary Figure 5A). Compared to treatment with vehicle alone, there is no significant increase in FGFR1 expression. In contrast, FGFR4 expression is enhanced by treatment with the FGFR1 and pan FGFR inhibitors but much less by treatment with the FGFR4 inhibitor. Comparable results were obtained when treating Dbt/MYCN-FGF8 with multiple concentrations of the above inhibitors (Supplementary Figure 5B-C). This result indicates that the FGF8-induced decrease of FGFR4 expression is efficiently reversed by inhibiting FGFR1 (or inhibiting all four FGFRs) but is not efficiently reversed by inhibiting FGFR4. These results suggest that FGF8 primarily signals through FGFR1 to downregulate FGFR4 expression.

Previous studies have shown that FGF8 binding to the FGFRs activates numerous downstream pathways, including the MAPK signaling cascade (34). To determine if this downstream signaling process is involved in FGF8-induced downregulation of FGFR4, we treated the Dbt/MYCN-FGF8 cells with MEK/ERK inhibitors and then similarly measured expression of FGFR1, FGFR4 and FGF8. Our results show that MEK or ERK inhibition leads to FGFR4 upregulation without increasing FGF8 (or FGFR1) expression (Figure 5D-F). These results indicate that FGF8 controls FGFR4 through the FGFR1/MEK/ERK pathway in this model system.

5. Mechanism of FGF8 upregulation in recurrent tumor-derived (TD) cells

Using our myoblast RMS system that models a P3F-targeted therapy, we observed the formation of recurrent P3F-independent tumors following the loss of P3F expression and regression in primary tumors. We identified high-level FGF8 expression in one subset of these recurrent tumors, and subsequently found that FGF8 was necessary and sufficient to promote tumorigenesis of these recurrent P3F-independent tumors (5). To elucidate the mechanism of P3F-independent FGF8 upregulation in these recurrent tumors, we first compared RNA-Seq data from recurrent TD cells with parental cells and primary TD cells (5). When aligned to the reference genome, we found that the FGF8 3’ UTR is shorter in the recurrent TD cells than in the parental and primary TD cells. As the 3’ UTR is known to be involved in the regulation of RNA translation and stability (35), we investigated whether FGF8 overexpression in the recurrent TD cells is related to changes in mRNA stability by measuring RNA decay following treatment with Actinomycin D, which inhibits RNA polymerase activity (36,37). Our results show that FGF8 expression following Actinomycin D remains stable up to 24 hours in the recurrent TD cells while FGF8 expression decreases rapidly in primary TD cells (Figure 6A). As a control, we assayed MYC expression, which showed rapid downregulation in both primary and recurrent TD cells (Figure 6B). These results suggest that the increased FGF8 abundance in the P3F-independent recurrent TD cells (5) is related to increased FGF8 RNA stability.

Figure 6. Mechanism of FGF8 overexpression in recurrent tumors.

FGF8 (A) and MYC (B) RNA expression following treatment of primary tumor-derived (TD) (Black) and recurrent TD (red) cells with Actinomycin D. Cells were seeded in 6 well plates and treated after 24 hours with Actinomycin D at 5 μg/mL and then collected at 2, 8 and 24 hours for RNA isolation and cDNA synthesis. Untreated primary or recurrent cells were used as time zero-point control and FGF8 and MYC mRNA abundance at each time point was normalized to mRNA level in the time zero-point control. The mRNA decay rate was determined by non-linear regression curve fitting using GraphPad software. Experiments were performed in biological triplicates and repeated 3 times. C. 3’ RACE analysis of FGF8 transcripts in parental (Par 1), primary TD (Prim 1, Prim 2) and recurrent TD (Rec 1a, Rec 2a) cells. D. Schematic illustration of the 3’ UTR region of the FGF8 transcripts and FGF8 gene in the primary and recurrent TD cells. The red segment represents the viral LTR sequence inserted into the FGF8 3’ UTR. The black segment represents the normal FGF8 3 ‘UTR and the gray segment represents the region downstream of the wild-type FGF8 3’ UTR. The orange segment represents an unknown sequence. The arrows show the position of PCR primers used in part E. E. PCR results from genomic DNA of parental, primary TD and recurrent TD cells using a forward primer from the FGF8 3’ UTR and a reverse primer from the viral LTR. F. PCR results from genomic DNA of parental, primary TD and recurrent TD cells using forward and reverse primers from the FGF8 3’ UTR surrounding the FGF8/viral LTR fusion junction. G. Gene reporter assay using Dbt/MYCN/iP3F cells transfected with dual luciferase reporter plasmid containing one of the indicated sequences cloned 3’ of the luciferase open reading frame. All results were normalized relative to the activity in the control empty vector (EV). The data are displayed as the mean ± SE of 3 replicates and are representative of four independent experiments. The two-sided unpaired Student t-test was applied to determine significant differences between EV and other samples (** P<0.01; *** P<0.001, ⁺ 0.06≥p≥0.05).

To further characterize the FGF8 3’ UTR in primary and recurrent TD cells, we used 3’ Rapid Amplification of cDNA Ends (RACE) methodology to isolate the 3’ UTR region. For parental (par 1) Dbt/MYCN/iP3F and two primary TD cells (Prim 1 and Prim 2), we amplified a single band that corresponds to the expected size of the FGF8 3’ UTR sequence. For the recurrent TD cells (Rec 1a and Rec 2a), we obtained a faint band with the same size as that of the primary TD cells in addition to an unexpected higher band, which is not present in the 3’ RACE PCR product of the parental and primary TD cells (Figure 6C). Sequencing of these PCR products revealed that the lower band corresponded to the wild-type FGF8 3’ UTR sequence. Surprisingly, the higher band consisted of the shortened FGF8 3’ UTR followed by an RNA segment of 447 bp, which corresponded to part of the viral 5’ LTR sequence of the MSCV2.2 plasmid vector. This 5’ LTR fragment was inserted at 132 bp following the FGF8 stop codon in one recurrent TD clone and at 137 bp following the FGF8 stop codon in the second recurrent TD clone (Figure 6D; compare mRNA diagram between the primary TD cells, and the two clones of the recurrent TD cells).

To identify this fusion between FGF8 and viral LTR sequences in the genomic DNA, we performed a PCR experiment with a forward FGF8 and reverse viral LTR primer. As predicted by the RNA results shown above, a junctional fragment consisting of the initial 132 bp or 137 bp of FGF8 3’ UTR joined to viral LTR was detected in genomic DNA from the recurrent TD clone 1 and clone 2, respectively (Figure 6E). To investigate whether the viral LTR was inserted within the FGF8 3’ UTR, we next designed forward and reverse PCR primers from the FGF8 3’ UTR sequence (Figure 6D, arrows). A single band of the expected size for the wild-type 3’ UTR was observed in the parental and primary TD cells (Figure 6F). In contrast, a second higher band was observed in one of the recurrent TD clones whereas only the lower band was observed in the other recurrent TD clone. The sequence of the lower band corresponded to the predicted wild-type FGF8 3’ UTR whereas the sequence of the larger PCR product from recurrent TD clone 1 revealed a 517 bp insertion of 5’ LTR viral sequence within the FGF8 3’ UTR region (Figure 6D). This 517 bp sequence aligned with the 5’ LTR of the MSCV2.2 plasmid vector and includes the 447 bp segment found at the mRNA level.

We then utilized a gene reporter assay to determine whether the 5’ LTR viral insertion contributes to increased FGF8 expression in the recurrent TD cells. We cloned the sequence corresponding to the FGF8 3’UTR-5’ viral LTR fusion found in each recurrent clone into the psiCHECK2 plasmid (with dual Luciferase/Renilla reporters) downstream of the luciferase reporter. We also cloned the full-length FGF8 3’ UTR alone, the truncated FGF8 3’ UTR alone, and the inserted viral 5’ LTR alone. Following transfection of these constructs into uninduced parental Dbt/MYCN/iP3F cells, we assayed the luciferase activity of the test and control constructs as a measure of the effect of these 3’ UTR sequences on expression. The full FGF8 3’ UTR and truncated 3’ UTR segments were associated with a small 1.4–1.9-fold increase in luciferase activity compared to that of the empty vector (EV) (Figure 6G). In contrast, the presence of the 5’ LTR sequence alone or fused to the FGF8 3’ UTR, as found in the transcripts of the two recurrent clones, was associated with a significantly higher increase (2.3- to 4.5-fold) in luciferase activity compared to the EV construct. The finding of no comparable increase in luciferase activity when the 5’ LTR segment was inserted upstream of the luciferase coding region suggests that this element is not acting as an enhancer (Supplementary Figure 6). These results thus support the premise that the fusion of the viral 5’ LTR to the FGF8 3’ UTR in the recurrent TD cells augments FGF8 expression by enhancing FGF8 mRNA stability.

Discussion

The development of approaches that directly antagonize P3F function or expression is challenging since fusion transcription factors are difficult to target with small molecule inhibitors (38). Efforts to develop alternative and effective therapies to indirectly target P3F are ongoing at different levels (38,39). Recent studies have pointed to the possibility of targeting P3F modulators like PLK1, coregulators like BRD4, and downstream effectors like MET, IGF1R, and FGFR4 (13,40–42). Identification of downstream P3F targets that are essential to P3F-driven oncogenicity is of great interest since targeting these molecules will potentially impair a major part of the P3F-induced program of tumorigenicity in FP-RMS. We previously demonstrated that FGF8 stimulates proliferation and transformation in a myoblast model of primary FP-RMS tumorigenesis, and discovered that a subset of P3F-independent recurrent tumors in this model system depends on sustained expression of FGF8 (5). Based on these findings, studying the mechanisms of FGF/FGFR involvement in the tumorigenicity of FP-RMS using small molecules that target FGFRs is important since these drugs may constitute a potential effective treatment of FP-RMS.

In this study, we first conducted a comprehensive expression analysis of the FGF and FGFR families in both human RMS tumors and cell lines. For the FGF family, the expression of FGF8 at higher level in FP-RMS than in FN-RMS is consistent with the fact that FGF8 is a direct transcriptional target of P3F and with a recent study that showed high FGF8 in FP-RMS (43). However, the variation of FGF8 expression at the RNA and protein levels in FP-RMS cells does not appear to correlate with P3F expression levels and may be related to chromatin accessibility or other cofactors. A recent study using a tandem of ATAC-seq and RNA-Seq measurements of chromatin accessibility showed that the expression of some but not all genes is concordant with the accessibility (44). In contrast to FGF8, FGF7 is a P3F target that is present at comparable levels in both FP- and FN-RMS tumors and is expressed at a lower level than FGF8 in FP-RMS. This finding suggests that FGF8 may be more responsive to P3F induction than FGF7 or that other cofactors contribute to higher FGF8 expression levels. Other than FGF13, all the other FGFs are expressed at the RNA level at low or very low levels in both FP- and FN-RMS. It should be noted that FGF13, FGF11, FGF12, and FGF14 do not bind to the FGFRs and represent intracellular FGFs with no well-known functions (45,46). Our results thus strongly indicate that FGF8 is the major FGF in many FP-RMS tumors.

As FGF8 was the most highly expressed FGF in FP-RMS, we studied the relationship between FGFR inhibitor sensitivity and FGF8 expression status in FP-RMS cells in addition to the status of FGFRs in these cells. We found that FP-RMS cells with high FGF8 expression tend to be more sensitive to the FGFR4 inhibitor than FP-RMS cells expressing low FGF8. In addition, targeted inactivation of FGF8 in RH30 cells lowers the sensitivity to the FGFR4 and pan-FGFR inhibitors. These findings suggest that FGF8 signaling is mainly driven through FGFR4 in FP-RMS cell lines with high FGF8. Our findings also indicate that the relative sensitivity of these RMS cells to FGFR inhibitors is not directly related to the level of FGFR expression or P3F expression in FP-RMS tumors.

Our myoblast model system provides substantial flexibility in dissecting oncogenic events in RMS pathogenesis and uncovering new regulatory pathways. In the formation of high FGF8-expressing P3F-independent recurrent tumors in the myoblast system, we found that the loss of FGFR4 expression was not only related to the loss of P3F expression but also was due to the high expression level of FGF8. We have shown that the ability of FGF8 to downregulate FGFR4 expression can be simply recapitulated by introducing an FGF8 expression construct into Dbt-MYCN myoblasts. In subsequent experiments, the impairment of FGF8-induced downregulation of FGFR4 expression in the presence of FGFR1, MEK or ERK inhibitors suggests that FGF8 signals through an FGFR1/MEK/ERK pathway to exert regulatory influence on FGFR4 mRNA expression. We postulate that this pathway ultimately impinges on specific transcription factors or RNA processing factors to exert the regulatory effect at the level of the FGFR4 gene or mRNA.

Based on these findings, we propose that FGF8 signals primarily through FGFR4 in FP-RMS tumors whereas FGF8 primarily signaling through FGFR1 in recurrent RMS tumors that have lost P3F expression. In FP-RMS, P3F induces expression of both the ligand FGF8 and the receptor FGFR4, and then FGF8 is involved in a negative feedback loop that lowers and fine-tunes FGFR4 expression. In P3F-independent recurrent tumors, increased FGF8 signals through FGFR1 to downregulate FGFR4 and thereby reinforce its primary signaling through FGFR1.

In a series of final experiments, we defined a molecular mechanism responsible for the high-level FGF8 expression in two P3F-independent recurrent RMS tumors from our myoblast model system. In these two tumors, a rearrangement event occurred in which a viral 5’ LTR sequence was juxtaposed with the FGF8 3’ UTR sequence. The result of this rearrangement was truncation of the wild-type FGF8 3’ UTR sequence and addition of the viral LTR sequence to the end of the FGF8 transcript. These structural changes at the 3’ end of the FGF8 transcript were consistent with the enhanced stability of the FGF8 transcript found in Actinomycin D treatment studies. In one of the two tumors, we were able to define the rearrangement event as an insertion of the 5’ LTR (557 bp) into the FGF8 3’ UTR genomic DNA segment. Of interest, this region matched the sequence of the LTR region from the MSCV2.2 plasmid, which is part of the pK1 retroviral vector used to overexpress MYCN in our myoblast system (16,18). As individual clones of Dbt/MYCN/iP3F cells were used in the previous mouse experiments, it is unlikely that a pre-existing insertion was selected during tumorigenesis. The presence of this insertion only in the genomic DNA of the recurrent tumor cells suggests that a recombination event occurred during tumorigenesis and that the resulting deregulated FGF8 expression drove expansion of this subclone in response to the selective pressure caused by the loss of P3F expression. Although similar fusion transcripts juxtaposing the viral LTR into the FGF8 3’ UTR formed in two recurrent tumors, the specific recombination events differ between the two recurrent tumors that we analyzed. As discussed above, we defined a 517 bp insertion of the 5’LTR sequence within the FGF8 3’UTR region in one recurrent tumor. Using a similar PCR strategy, we could not identify a simple small insertion in the second recurrent tumor. This difference could be due to an insertion that was too large to be efficiently amplified or a more complicated rearrangement forming the junction between FGF8 and viral sequences.

Our studies of these recurrent tumors indicate that the insertion of the viral LTR into the FGF8 3’ UTR most likely increases FGF8 expression by increasing FGF8 mRNA stability. A previous report, in a mouse mammary tumor model, showed that the integration of the mouse mammary tumor virus upstream of the murine fgf8 gene can act as an enhancer and induce fgf8 overexpression without a change in the fgf8 transcript sequence (47). Despite the difference in the mechanism of overexpression of FGF8 between our finding and this report, the data suggest that the FGF8 locus is susceptible to recombination events, and when these events lead to upregulated expression, a strong oncogenic stimulus is created. FGF8 has been reported to be highly expressed within several cancer categories and high expression is associated with differences in outcome or therapeutic response (43,48–50). In cases in which there is unusually high FGF8 expression in a tumor, the possibility that FGF8 may be aberrantly upregulated by genetic alterations, such as rearrangements, should thus be considered. However, it should be emphasized that recurrent clones with FGF8 rearrangements arose in these studies in response to loss of P3F expression in the primary tumor. Since current RMS therapy does not affect P3F expression or function, FGF8 overexpression would not be expected in recurrent tumors arising from current RMS therapy.