Zebrafish rhabdomyosarcoma reflects the developmental stage of oncogene expression during myogenesis

Rhabdomyosarcoma (RMS) is a pediatric malignancy thought to arise from the aberrant regulation of developmental programs during myogenesis. It is the most common soft-tissue sarcoma of childhood (Wexler and Helman, 1994; Arndt and Crist, 1999). Treatment approaches are often aggressive and multimodal, involving a combination of chemotherapeutic agents and local surgical resection and/or irradiation. Patients with locally unresectable or metastatic disease are particularly difficult to treat, and current treatment approaches have a considerable number of severe sequelae, including secondary malignancies (Wexler and Helman, 1994; Arndt and Crist, 1999). There is a need for novel therapies that specifically target the unique molecular underpinnings of this disease.

There are two major pediatric subtypes of RMS, embryonal (ERMS) and alveolar (ARMS), distinguished by histological appearance (Wexler and Helman, 1994; Arndt and Crist, 1999). ARMS is characterized in most cases by chromosomal translocations between the PAX3 or PAX7 and the FKHR (FOXO1 - Human Gene Nomenclature Database) loci (Barr, 2001). By contrast, ERMS is often associated with loss of heterozygosity at chromosome 11p15 and a number of whole chromosomal gains (Wexler and Helman, 1994; Barr, 1997; Arndt and Crist, 1999). Activating mutations in RAS family members have been described in a minority of ERMS cases (Stratton et al., 1989; Chen et al., 2006; Martinelli et al., 2009), and gene expression analysis on a number of tumors suggests that RAS pathway activation might be a common event in ERMS formation, even in the absence of RAS mutations (Langenau et al., 2007).

The vast majority of skeletal muscle progenitors during vertebrate embryogenesis are derived from somites, epithelial condensations of the paraxial mesoderm that transiently form during anterior-posterior patterning of the embryo (Pownall et al., 2002; Parker et al., 2003). The myogenic regulatory factors (MRFs) MyoD (Myod1), Myf5, myogenin (also known as Myf4) and Myf6 (also known as Mrf4) are basic helix-loop-helix (bHLH) transcription factors that are important regulators of muscle development and differentiation. MyoD and Myf5 are expressed in early muscle progenitors within somites and together are required for and regulate muscle progenitor specification (Pownall et al., 2002). Myf6 and myogenin are expressed in more differentiated muscle cells and are the key regulators of muscle differentiation (Pownall et al., 2002). Postnatal muscle progenitors are thought to arise from satellite cells, the resident stem cells of adult skeletal muscle (Parker et al., 2003). The MRFs function similarly to their embryonic roles in directing the differentiation of satellite cell-derived progenitors into terminally differentiated muscle (Parker et al., 2003).

The myogenic program is disrupted in human RMS. RMS tumors consistently overexpress MYOD1, and variably express the other MRFs (Anderson et al., 1999; Blandford et al., 2006). They also express other myogenic factors that are characteristic of multiple stages of myogenesis and often express various structural proteins that are typically found in terminally differentiated muscle (Anderson et al., 1999; Blandford et al., 2006). Yet RMS tumors do not readily form myotubes and myofibers, suggesting an inability to activate myogenic programs appropriately (Anderson et al., 1999). Interestingly, expression of myogenin is correlated with disease progression and poor clinical outcome (Blandford et al., 2006; Heerema-McKenney et al., 2008), yet it is unclear whether it directly modulates disease phenotype or represents a marker of disease state in RMS.

A handful of recent experiments have yielded insights into the tumor-initiating potential of a number of cell types during muscle development. In murine models, ARMS can develop from the introduction of PAX-FKHR into mesenchymal stem cells or Myf6-expressing terminally differentiating myofibers, but not in Pax7-expressing satellite cells (Keller et al., 2004b; Keller et al., 2004a; Ren et al., 2008). In another line of experiments, the introduction of T/t-Ag (SV40 large-T/small-T antigen), hTERT and H-Ras^V12G expression into committed adult human skeletal muscle myoblasts resulted in ERMS formation, whereas the introduction of the same genetic alterations into low-passage human fetal skeletal muscle precursor cells resulted in sarcomas lacking histopathological features of RMS (Linardic et al., 2005). Rubin and colleagues investigated the tumorigenic potential of multiple myogenic lineages using Ptch1 and/or p53 (Trp53 - Mouse Genome Informatics) conditional mouse models (Rubin et al., 2011). Although a number of myogenic lineages were capable of giving rise to ERMS tumors, Myf6-expressing differentiating muscle cells had the highest propensity to form ERMS tumors, whereas Pax7-expressing satellite cells formed mostly undifferentiated sarcomas. In a study by Hettmer and colleagues, the introduction of KrasG12V and disruption of CDKN2A;p16p19 in prospectively isolated satellite cells resulted in pleomorphic RMS formation; in a heterogeneous population of adipogenic and differentiating myogenic cells, these genetic changes led to tumors with varying levels of myogenic commitment (Hettmer et al., 2011). Finally, Hatley and colleagues demonstrated that murine ERMS can form when a conditionally active Smo allele is introduced into the adipocyte lineage (Hatley et al., 2012). Taken together, these data suggest that a number of cell types from myogenic and non-myogenic lineages harbor differing potentials to generate RMS. However, the pathways activated in these cell types that contribute to differing cell biology and myogenic differentiation remain to be uncovered.

In this study, we generated novel models of RMS in the zebrafish by inducing KRAS^G12D expression at different stages of muscle development using zebrafish rag2, cdh15 and mylz2 (mylpfa - Zebrafish Information Network) promoters. The models differed in their ability to recapitulate normal myogenesis, with rag2 and cdh15 tumors consisting of a large proportion of myoblast-like cells and mylz2 tumors most closely resembling mature skeletal muscle. rag2 and cdh15 tumors were more aggressive than mylz2 tumors and were more similar to less differentiated human RMS in terms of gene expression. We found that tumor-propagating activity was enriched in the least differentiated, myf5-expressing cell populations within the tumors, suggesting that a cellular hierarchy reminiscent of untransformed developing muscle cells exists within RMS tumors. This study highlights the importance of the developmental timing of oncogene expression on subsequent tumor formation and suggests that pathways regulating normal development can also have profound effects on tumor behavior.

Zebrafish were maintained and developmentally staged as previously described (Westerfield, 1989) and according to Institutional Animal Care and Use Committee guidelines. The Animal Care and Use Committee, Boston Children’s Hospital, approved all animal protocols.

cdh15 (4.1 kb) and mylz2 (2.2 kb) promoter elements were PCR-amplified from wild-type AB strain genomic DNA (for all primer sequences, see supplementary material Table S3). Each forward primer contains a XhoI restriction digest site, and each reverse primer contains a BamHI site. PCR products were purified, digested using XhoI and BamHI, and cloned into the rag2-GFP vector (Langenau et al., 2003) to generate cdh15:GFP and mylz2:GFP vectors. These vectors were digested with BamHI and HindIII to release the GFP cassette, and KRAS^G12D from the rag2-KRAS^G12D vector (Langenau et al., 2007) was inserted to generate cdh15:KRAS^G12D and mylz2:KRAS^G12D vectors. The generation of the mylz2:mCherry and rag2-KRAS^G12D constructs were previously described (Langenau et al., 2007; Smith et al., 2010).

The cdh15:GFP, mylz2:mCherry, rag2:KRAS^G12D, cdh15:KRAS^G12D and mylz2:KRAS^G12D vectors were linearized with XhoI, purified and diluted to 100 ng/μl in 0.5×TE + 0.1 M KCl. One nl of the cdh15:GFP and mylz2:mCherry vector dilutions were microinjected into one-cell-stage AB strain zebrafish embryos, and resulting F0 adult fish were incrossed to identify fluorescence-positive stable F1 transgenic fish. The F1 progeny were outcrossed to AB strain fish to propagate the lines. mylz2:mCherry fish were outcrossed to myf5:GFP (Chen et al., 2007) zebrafish to generate myf5:GFP; mylz2:mCherry double transgenic animals. The rag2:KRAS^G12D, cdh15:KRAS^G12D and mylz2:KRAS^G12D vector dilutions were microinjected into one-cell stage AB strain or myf5:GFP; mylz2:mCherry transgenic embryos to generate mosaic transgenic tumor-bearing fish.

Segments (350-700 bp) of open reading frames for GFP, mCherry, mylz2, cdh15, rag2 and KRAS^G12D genes were PCR amplified from the rag2-GFP or cdh15:mCherry vectors or from zebrafish tumor cDNA. Digoxigenin-labeled anti-sense and sense RNA probes were generated from purified PCR products using SP6 and T7 RNA polymerase, respectively, with the exception of the KRAS^G12D probes, which were generated using T7 and SP6 RNA polymerase, respectively. Whole-mount in situ hybridization on zebrafish embryos was performed as described (Thisse et al., 1993).

Fish were euthanized and fixed in 4% paraformaldehyde overnight at 4°C and decalcified in 0.5 M EDTA, pH 8. Paraffin embedding, sectioning, Hematoxylin and Eosin (H&E) staining and RNA in situ hybridization were performed according to standard techniques by the Brigham & Women’s Pathology Core. RNA in situ probes used for staining of sections have been previously described (Langenau et al., 2007). For immunohistochemistry, paraffin-embedded sections were treated with xylene, rehydrated in a graded ethanol series, incubated in Citrate-based Antigen Unmasking Solution (Vector Laboratories) at 97°C for 20 minutes, and incubated in BLOXALL Blocking Solution (Vector Laboratories) for 10 minutes. Subsequent incubations were performed following the manufacturer’s instructions using the Vectastain Elite ABC Kit (Vector Laboratories). Phospho-S6 Ribosomal Protein (Ser235/236) antibody (Cell Signaling #2211) was used at 1:400, and slides were counterstained with Hematoxylin.

RNA was isolated from whole tumor or normal muscle samples using TRIzol reagent (Invitrogen), treated with DNase I and purified using the RNeasy Mini Kit (Qiagen). cDNA synthesis was performed from equal amounts of RNA using SuperScript III First-Strand Synthesis SuperMix for qRT-PCR (Invitrogen), and quantitative RT-PCR was performed using SYBR GreenER qPCR SuperMix for iCycler (Invitrogen). Primer sequences used for QPCR are included in supplementary material Table S3. For each gene tested for every sample, relative gene expression was calculated from experimental triplicates by using the 2^-ΔΔCT method (Livak and Schmittgen, 2001). For each experimental sample, ΔC_T was calculated by subtracting the C_T value for EF1a from the C_T value for the target gene. ΔΔC_T was then calculated by subtracting the ΔC_T for a single reference sample per gene from ΔC_T for the sample. Normalized relative gene expression was calculated using the 2^-ΔΔCT formula and then averaged across experimental replicates, and gene transcript expression levels between tumor types were compared using Student’s t-test.

RNA was isolated from seven tumors for each of the rag2, cdh15 and mylz2 tumor types and from seven normal zebrafish muscle samples using TRIzol reagent (Invitrogen). cDNA was prepared and hybridized to zebrafish Affymetrix arrays. Zebrafish data were analyzed using the Genepattern software suite (Broad Institute). Raw CEL files were converted into GCT format using the ExpressionFileCreator Module, with RMA and quantile normalization. The GCT file was then used in the PreprocessDataSet module, where threshold and ceiling values were set, and any value lower/higher than the threshold/ceiling were reset to the threshold/ceiling value, and data were processed to discard any invariant genes (genes with less than twofold change and mindelta 100). Comparisons between groups were performed using the ComparativeMarkerSelection modules, using two-sided t-test with Benjamini-Hochberg multiple hypothesis testing. Normalized, pre-processed data were used for unsupervised, hierarchical clustering via GenePattern. Row and column distance measurement was performed with the Pearson correlation, using pairwise complete-linkage, and both rows and columns were normalized. Microarray data have been deposited in Gene Expression Omnibus under accession number GSE39731.

The data from Davicioni et al. (Davicioni et al., 2009) were modified to stratify samples into survival of < or >2.5 years, or < or >5 years. Any sample without full survival data was censored, bringing the total number of usable samples from 186 to 156. Classifications (^*.cls files) were made based on the conventions of Davicioni et al., including histological subtype, molecular subclass, and survival of 2.5 years or 5 years. This yielded a gct file containing the expression data of the 156 relevant human arrays, as well as several cls files used as the input to GSEA. Because the annotation between zebrafish and human datasets is necessarily imperfect, we converted the zebrafish Affymetrix probe IDs into the human Affymetrix U133A array probe ID, if possible. This then allowed for derivation of the human gene symbols (HUGO identifiers) for each relevant zebrafish comparison. GSEA was run using up- and downregulated HUGO gene lists from each zebrafish RMS type, querying against the various human RMS classes. Permutation number=1000, collapse dataset to symbols=true, and permutation type=gene_set as several subclassifications had fewer than seven samples. A false discovery rate (FDR) of <0.25 was considered a significant enrichment.

Tumor-bearing fish were injected intraperitoneally with 10 μl of 2.5 mg/ml EdU per 0.25 g body weight. After 24 hours, fish were euthanized and frozen in Optimal Cutting Temperature medium at -80°C overnight. Cryostat sections (12 μm) were prepared, and EdU labeling was performed using the Click-iT EdU Alexa Fluor 594 Imaging Kit (Invitrogen). Labeled sections were imaged at 400× magnification using a compound fluorescent microscope, and the number of EdU-positive and DAPI-positive nuclei were counted in three separate 1.37×10⁴ μm² fields per tumor. The ratios of EdU-positive to DAPI-positive nuclei in the three fields were averaged to calculate an EdU/DAPI ratio for each tumor.

Blood cells from AB strain fish, muscle cells from mylz2:mCherry fish, and tumor cells from KRAS^G12D-injected fish were isolated. Blood cells were re-suspended in 10 ml 0.9× PBS. Muscle and tumor cells were mechanically disrupted in 10 ml 0.9× PBS and incubated with 125 μl 2.5 mg/ml (13 WU/ml) Liberase TM enzyme blend (Roche) at room temperature for 30 minutes. Heat-inactivated FBS (500 μl) was added to each cell preparation, and samples were then filtered (40-μm cell strainer) and centrifuged at 1000 g for 5 minutes. Muscle and tumor cell suspensions were re-suspended in 1 ml 0.9× PBS + 2 mM MgCl₂ and incubated with 10 μl of 10 U/μl benzonase (Sigma) for 15-30 minutes at room temperature. To this, 9 ml of 0.9× PBS + 5% fetal bovine serum (FBS) was added, and samples were centrifuged at 1000 g for 5 minutes, re-suspended in 0.9× PBS + 5% FBS + 200 ng/ml DAPI, and re-filtered (40-μm cell strainer). FACS was performed as described (Traver et al., 2003; Langenau et al., 2007) on a BD FACSAria II Cell Sorter. Doublets were excluded by size, and dead cells were excluded by DAPI incorporation. For all samples, test sorts were performed to measure the purity of sorted populations.

Transplantation of tumor cells was performed as described (Langenau et al., 2007). Sorted cell populations were transplanted into irradiated casper fish (White et al., 2008) (10 Gy for primary transplantation, 20 Gy for secondary transplantation, 2 days prior to transplantation). Red blood cells were used as carrier cells for a total of 2×10⁴ cells per transplantation. To assess for tumor engraftment, recipients were scored for GFP and mCherry fluorescence under a fluorescence dissecting microscope at 8 and 15 days post-transplantation. Fish with GFP- and/or mCherry-positive tumor masses were defined as having positive engraftment, and a subset of these fish was sectioned to verify engraftment by histology.

5′ upstream promoter elements of the cdh15 (cadherin 15, M-cadherin) and mylz2 (myosin light chain, phosphorylatable, fast skeletal muscle a) genes were cloned from wild-type zebrafish genomic DNA. cdh15 is a marker of satellite cells and is also expressed in differentiating myoblasts where it contributes to myoblast fusion (Irintchev et al., 1994; Zeschnigk et al., 1995), whereas mylz2 is a marker of more differentiated muscle in zebrafish (Xu et al., 2000). The cdh15 and mylz2 promoters were cloned into GFP and mCherry expression vectors, respectively, and injected into AB strain zebrafish embryos at the one-cell stage. The resulting transgenic animals displayed fluorophore expression throughout the musculature beginning at 1 day post-fertilization (dpf) for cdh15:GFP embryos, and 2 dpf for mylz2:mCherry embryos (data not shown). By RNA in situ hybridization, the cdh15 and mylz2 promoters were shown to drive gene expression similarly to their respective endogenous genes during development (Fig. 1E-T). myoD (myod1 - Zebrafish Information Network) is expressed in the most newly formed somites of developing embryos (Fig. 1A-D), reflecting its early role in the specification of muscle progenitors. GFP transcript can be detected in nascent somites of cdh15:GFP embryos (Fig. 1I-K; supplementary material Fig. S1). By contrast, mCherry transcript is consistently found only in older and more anterior somites of mylz2:mCherry embryos during somitogenesis (Fig. 1Q-S). In addition, whereas mCherry transcript can be seen at high levels throughout the developed musculature of 5 dpf larvae, GFP transcript is barely detectable in the trunk musculature at 5 dpf (Fig. 1H,L,P,T). Gene expression analyses from adult fish also suggest that mCherry-positive cells from mylz2:mCherry fish express markers of muscle differentiation at higher levels than do GFP-positive cells from cdh15:GFP fish (supplementary material Fig. S2). Although both promoters drive gene expression in the muscle of developing zebrafish, the cdh15 promoter drives expression earlier during myogenesis than the mylz2 promoter.

A 6.9-kb zebrafish rag2 promoter drives gene expression at low levels in the mononuclear compartment of the skeletal musculature (Langenau et al., 2007) that are undetectable by in situ hybridization of 5 dpf larvae (supplementary material Fig. S3) but can occasionally be seen in mononuclear cells of older larvae and adult fish (Langenau et al., 2007). Injection of a rag2:KRAS^G12D construct, containing human oncogenic KRAS^G12D, into zebrafish embryos results in the formation of RMS tumors (Langenau et al., 2007). To investigate further the effects of KRAS^G12D expression in the musculature, rag2:KRAS^G12D, cdh15:KRAS^G12D and mylz2:KRAS^G12D transgenic constructs were individually injected into AB strain zebrafish embryos at the one-cell stage. The resulting embryos express the KRAS^G12D transgene in a mosaic pattern consistent with endogenous expression patterns of the respective muscle promoters (supplementary material Fig. S4). In older larvae (5 dpf), KRAS^G12D expression can be detected in the musculature for all of the constructs. The mosaic transgenic zebrafish developed tumors that differed in histological appearance (Fig. 2A-G). Like rag2 tumors, cdh15 tumors comprise a heterogeneous population of spindle-shaped cells resembling myoblasts, multinucleated myofibers, and blood cells, consistent with classification as the spindle cell variant of ERMS. By contrast, mylz2 tumors are relatively hypocellular, more closely resemble mature skeletal muscle, and contain a number of large, vacuolated cells (Fig. 2D-G). Tumors from the different models were assigned a differentiation score in a blinded fashion based on histological characteristics, including degree of cellularity and how closely tumors resemble mature skeletal muscle. Overall, rag2 and cdh15 tumors appear less differentiated than mylz2 tumors by histology (Fig. 2X).

Like rag2 tumors, cdh15 and mylz2 tumors express myoD, myogenin and desmin, clinical diagnostic markers of RMS (Fig. 2L-W). Whereas a large proportion of cells in rag2 and cdh15 tumors display myoD and myogenin expression, only a small fraction express these markers in mylz2 tumors. By quantitative RT-PCR, the rag2 and cdh15 tumors express the activated satellite cell markers myf5 and cdh15, but not the quiescent satellite cell marker pax7, at significantly higher levels compared with the mylz2 tumors (Fig. 3A-C; P<0.05). rag2 tumors express higher levels of the myoblast differentiation gene myogenin compared with mylz2 tumors (Fig. 3D; P<0.05). rag2 tumors also express lower levels of desmin, which encodes an intermediate filament found in differentiated muscle (Fig. 3E; P<0.05). No difference was seen in levels of mylz2 expression across the tumors (Fig. 3F; P>0.05). There was a detectable level of KRAS^G12D expression in rag2 tumors. Although KRAS^G12D transcript levels appear higher in cdh15 and mylz2 tumors compared with rag2 tumors, this difference was not statistically significant (Fig. 3G; P>0.05). Phospho-S6 ribosomal protein (pS6RP) levels, indicative of active RAS signaling, were similar among tumor types by immunohistochemistry (supplementary material Fig. S5). rag2 and cdh15 tumors express early and intermediate markers of myogenesis at higher levels than do mylz2 tumors, reflecting a greater number of primitive myoblast-like cells in those tumors.

rag2 and cdh15 tumors behave more aggressively than mylz2 tumors. mylz2 tumors develop with lower incidence compared with the rag2 and cdh15 models (mylz2, n=19 of 117; rag2, n=50 of 144; cdh15, n=42 of 146) and with longer latency (Fig. 4A; rag2 versus mylz2, P=0.0004; cdh15 versus mylz2, P<0.0001; rag2 versus cdh15, P=0.73; log-rank test). mylz2 tumor-bearing fish display increased survival over time compared with those bearing rag2 or cdh15 tumors (Fig. 4B; rag2 versus mylz2, P=0.001; cdh15 versus mylz2, P=0.009; rag2 versus cdh15, P=0.72; log-rank test); median survival times were 63 days for rag2 tumors, 46 days for cdh15 tumors and 156 days for mylz2 tumors. By histology, mylz2 tumors generally appear more circumscribed and less invasive compared with the other tumor types (data not shown). The rate of EdU incorporation per total nuclei within the tumors over a 24-hour time period is not significantly different (Fig. 4C-L; P>0.05), and TUNEL staining within the tumors does not differ (data not shown). The differences in survival observed in the tumor types correspond to varying abilities to undergo muscle differentiation but do not correlate with measurable differences in the rate of proliferation or apoptosis.

Given the histological and phenotypic variation among the tumor types, we investigated whether this was reflected by differences in gene expression. We performed gene expression analysis using microarrays for each of the three subtypes of zebrafish RMS, along with normal muscle samples as controls. Unsupervised clustering revealed distinct clusters for the wild-type muscle and mylz2 tumors, whereas the rag2 and cdh15 tumors were closer to each other (Fig. 5A), similar to what was seen at the histological level. This indicates that each of these tumor types harbor distinct molecular signatures. To quantify this on a global level, the up- and downregulated genes for each tumor type were generated by comparing array data for each of the tumor types to normal muscle (greater than twofold change, FDR<0.05) as shown in Fig. 5B and supplementary material Table S1. Across the three tumor types, we found that 664 genes were dysregulated (compared with normal muscle), demonstrating considerable molecular overlap between these histologically distinct tumors. However, as suggested by clustering, the rag2 and cdh15 tumors appear much more similar to each other than to the mylz2 tumors: 208 genes (96 upregulated, 112 downregulated) were shared between the rag2 and cdh15 tumors, whereas the mylz2 tumors only shared 56 genes (30 upregulated, 26 downregulated) with the rag2 tumors. The mylz2 tumors exhibited upregulation of genes more reminiscent of normal muscle, such as troponin (tnnt3a) and myosin light peptide 7 (myl7), and had significantly higher levels of genes such as tropomyosin 1 (tpm1) than did the rag2 and cdh15 tumors. This suggests that at the molecular level, consistent with the histological analysis, the mylz2 tumors appear to be more differentiated than the rag2 or chd15 tumors.

To understand whether these zebrafish gene signatures correspond to the molecular changes found in human RMS, we compared the zebrafish tumor signatures with a large collection of annotated human RMS tumors using GSEA (Davicioni et al., 2009). We utilized the histological and molecular classification scheme in the Davicioni analysis, and considered the 156 human tumors that had complete annotation. This allowed for a classification of the human tumors into well-differentiated (WD), moderately differentiated (MD) and undifferentiated (UDS) subtypes. Genes upregulated in the mylz2 tumors (compared with the rag2 or cdh15 tumors) were enriched in the human WD tumors [mylz2 versus rag2: normalized enrichment score (NES)=2.06, family-wise error rate (FWER)=0.00, FDR=0.00; mylz2 versus cdh15: NES=1.88, FWER=0.001, FDR=0.001], consistent with the observation that the zebrafish mylz2 tumors are the most well-differentiated (Fig. 5C,D; supplementary material Table S1). Similarly, genes upregulated in the rag2 tumors were enriched in the human UDS tumors (rag2 versus cdh15: NES=1.50, FWER=0.112, FDR=0.089) compared with human MD tumors, indicating that the rag2 tumors are most similar to human undifferentiated rhabdomyosarcoma (data not shown).

To understand which classes of genes are most affected in each tumor subtype, the genes characterizing each subtype were analyzed using Ingenuity Pathway Analysis. The mylz2 and rag2 tumors were most easily differentiated by genes that play roles in glycolysis (e.g. pgm1, hk1, aldoa, upregulated in the mylz2 versus rag2 tumors) as well as the phosphatase and tensin homolog (PTEN) signaling pathway (e.g. fgfr3, fgfr4, ccnd1, rac3, magi), suggesting that alterations in metabolism might underlie differences in tumor subtypes. Taken together, these data indicate that the histological observation of differentiation status in the zebrafish tumors recapitulates the molecular distinctions of human RMS differentiation.

Previous experiments have shown that fluorescently labeled cell populations of rag2 tumors generated in myf5:GFP; mylz2:mCherry double transgenic zebrafish are molecularly distinct and represent cells at differing stages of muscle differentiation (Ignatius et al., 2012). Moreover, in this background, myf5:GFP-positive/mylz2:Cherry-negative cells in rag2 tumors, representing the least differentiated cell types, were highly enriched for tumor-propagating activity (Ignatius et al., 2012). In order to determine whether similar cell populations are enriched for tumor-propagating activity in the other RMS types, rag2:KRAS^G12D, cdh15:KRAS^G12D or mylz2:KRAS^G12D constructs were injected into one-cell-stage myf5:GFP; mylz2:mCherry zebrafish embryos (Fig. 6). When observed grossly, all tumors in this transgenic background display high levels of mCherry expression (Fig. 6C,F,I,L). Whereas rag2 and cdh15 tumors brightly express GFP, mylz2 tumors appear either extremely dim or grossly negative for GFP fluorescence (Fig. 6B,E,H,K). FACS analysis on the mononuclear component of the tumors confirmed that only a small fraction of cells in the mylz2 tumors are dimly GFP-positive compared with rag2 and cdh15 tumors (Fig. 7).

Tumors were then FACS-sorted based on GFP (G) and mCherry (R) positivity into G+R-, G+R+, G-R+ and G-R-populations (Fig. 7B-E,G-J,L-O). In the examples depicted in Fig. 7, the purity of cell populations were: for the rag2 tumor, G+R- 92.0%, G+R+ 91.4%, G-R+ 73.7%, G-R-99.8%; for the cdh15 tumor, G+R- 93.1%, G+R+ 80.9%, G-R+ 73.6%, G-R-99.8%; and for the mylz2 tumor, G+R- 92.1%, G+R+ 94.9%, G-R+ 89.0%, G-R-100.0%. Sorted cell populations were transplanted into irradiated casper recipients at limiting dilution (1×10⁴, 1×10³, 100 and ten cells). casper strain zebrafish are almost entirely transparent as adults, allowing for the rapid detection of transplanted tumor cells (White et al., 2008). Interestingly, the G+R- and G+R+ populations were enriched for engraftment potential in irradiated recipients compared with G-R+ and G-R-populations for each of the tumor models (Table 1; n=2 for each tumor type). For example, when 1000 G+R-, G+R+ or G-R+ cells were transplanted, engraftment was seen, respectively, in: 8/10, 7/10 and 0/10 fish for the rag2 tumor; 4/8, 2/8 and 0/8 fish for the cdh15 tumor; and 6/6, 8/10 and 1/9 fish for the mylz2 tumor. rag2 and cdh15 engrafted tumors were consistently larger than mylz2 tumors (data not shown). Whereas the G-R+ population of rag2 and cdh15 tumors exhibited no engraftment potential at 1×10⁴ cells, the G-R+ population of mylz2 tumors was able to engraft when as few as 100 cells were transplanted. When G+R- cells were transplanted into casper recipients, the engrafted tumors that developed closely resembled the parent tumor by fluorophore expression for each of the tumor models (Fig. 6M-U). G+R- cells from rag2 and cdh15 tumors yielded engrafted tumors that were positive for both GFP and mCherry expression, whereas dim G+R- cells from mylz2 tumors yielded tumors that were mCherry-positive, but only dimly GFP-positive or GFP-negative as observed grossly.

To confirm the engraftment potential of GFP-expressing cells, engrafted tumor cells from rag2 and cdh15 G+R- primary transplants were FACS sorted and found to contain G+R-, G+R+, G-R+ and G-R- cell populations. Sorted cells were transplanted into secondary irradiated recipients at limiting dilution (10³, 100 and ten cells). As in the primary recipients, the G+R- and G+R+ populations of rag2 and cdh15 engrafted tumors were enriched for tumor-propagating activity, with as few as ten GFP-positive cells required for engraftment (supplementary material Table S2). As previously reported (Ignatius et al., 2012), G+R- cells from the rag2 tumors exhibited a significant enrichment for long-term, serial engraftment potential compared with either the G+R+ or G-R+ cell subpopulations (P=0.034, limdil software). Secondary transplantation of mylz2 tumors could not be completed owing to the low number of engrafted cells within primary recipients. From these transplantation experiments, we conclude that the myf5:GFP-expressing subpopulation within rag2, cdh15 and mylz2 tumors is enriched for tumor-propagating activity.

Clinical RMS is characterized by the inability to undergo normal muscle differentiation, despite the overexpression of MRFs in tumors (Anderson et al., 1999; Blandford et al., 2006). The mechanisms underlying this block in myogenesis and the importance of dysregulation of myogenic programs in RMS are largely unknown. In this study, zebrafish RAS-induced RMS displayed varying capabilities to implement the myogenic program. rag2 and cdh15 tumors had a large number of primitive, myoblast-like cells with inefficient myotube and myofiber formation, reminiscent of the spindle cell variant of ERMS, whereas mylz2 tumors more closely resembled mature skeletal muscle. This difference was reflected at the gene expression level; rag2 and cdh15 tumors expressed higher levels of the satellite cell and early myoblast markers myf5 and cdh15, and mylz2 tumors preferentially expressed structural proteins associated with terminally differentiated myofibers. rag2 and cdh15 tumors expressed high levels of MRFs in the absence of terminal muscle differentiation, suggesting that overriding mechanisms in RMS cells block the capacity to undergo myogenesis.

Myogenin is an MRF that directs terminal differentiation of muscle progenitor cells (Pownall et al., 2002). Interestingly, a high proportion of myogenin-positive cells and overexpression of myogenin transcript are associated with poor prognosis in human (Blandford et al., 2006; Heerema-McKenney et al., 2008) and zebrafish RMS. Recent data have demonstrated a crucial role for myogenin-positive cells in the invasion and metastatic potential of RMS tumors (Ignatius et al., 2012). Our data further highlight the importance of myogenin as a prognostic marker and suggest that it is also a marker of aberrant regulation of the myogenic program in RMS. In rag2 and cdh15 tumors, myogenin-expressing cells, which make up the bulk of the tumor, fail to undergo terminal differentiation. In mylz2 tumors, these myogenin-expressing cells do not accumulate but probably undergo differentiation. We propose that high myogenin expression, which is associated with poor survival in human RMS, might represent an inability of tumor cells to differentiate. Our data suggest that aberrant regulation of myogenesis is an important contributor to clinical outcome in RMS.

Gene expression analyses of RAS-induced zebrafish tumors highlight the ability of zebrafish tumors to model differing subtypes of human RMS. The gene signatures of the zebrafish tumors, although sharing a great deal of overlap, demonstrate that tumors driven by mylz2 are more well differentiated than either rag2 or cdh15 tumors. Moreover, pathway analysis suggests that dysregulation of the PTEN tumor suppressor pathway may discern the mylz2 from rag2 tumors. This is consistent with studies showing that some human RMS might be sensitive to inhibition of downstream mTOR signaling with rapamycin (Hosoi et al., 1999). Subtypes of human RMS expressing PAX3 or PAX3-FKHR downregulate PTEN, allowing for continued tumor growth (Li et al., 2007). Several of the genes upregulated in the rag2 and cdh15 tumors, including cyclin D1 and fgfr4, are known to interact with the PTEN pathway, and are involved in muscle differentiation. Cyclin D1 overexpression suppresses muscle differentiation in myoblasts by inhibiting the ability of MyoD to transactivate muscle-specific genes (Skapek et al., 1995) and might function similarly in RMS to block differentiation programs. Activating mutations in FGFR4 have recently been described in RMS, and FGFR4 is a transcriptional target of PAX3 and the PAX3-FKHR fusion protein (Lagha et al., 2008a; Taylor et al., 2009; Cao et al., 2010). FGF signaling has also been implicated in the maintenance of stem cell compartments in embryonic and adult muscle (Lagha et al., 2008b). Our data suggest that dysregulated FGF signaling contributes to RMS progression, perhaps through the activation of self-renewal pathways at the expense of differentiation programs, and that pathways required for normal muscle development and stem cell maintenance are utilized in tumors to mediate important aspects of tumor behavior.

Prior studies have suggested that multiple cell types are capable of giving rise to RMS (Keller et al., 2004b; Keller et al., 2004a; Linardic et al., 2005; Langenau et al., 2007; Ren et al., 2008; Hettmer et al., 2011; Rubin et al., 2011; Hatley et al., 2012). In line with these studies, we have demonstrated that RMS formation can occur from the induction of oncogene expression at different times during myogenesis. The ability of tumors to recapitulate normal patterns of myogenesis differs considerably based on the timing of oncogene induction. Our data suggest that RAS activation in the context of differentiating muscle cells results in transformation but does not inhibit myogenic programs as it does in myoblasts. Similarly, ERMS tumors from the Myf6 differentiating myoblast lineage in p53^-/- mice show a higher degree of in vitro myogenic differentiation compared with tumors from the Pax7 satellite cell lineage (Rubin et al., 2011). These data suggest that the differentiation capability of tumor cells reflects that of their respective cells of origin during development. Our data also raise the intriguing possibility that differences in clinical outcome can reflect differences in the tumor-initiating cell in RMS.

The rag2 and cdh15 tumors are highly similar despite differences in developmental expression patterns of these promoters. Transgene expression in mosaic rag2:KRAS^G12D and cdh15:KRAS^G12D embryos is consistent with endogenous expression patterns, with significant KRAS^G12D expression seen in developing somites of cdh15 but not rag2 transgenics. In older larvae (5 dpf), rag2 and cdh15 promoter expression is undetectable within the trunk musculature of stable transgenics, consistent with endogenous expression (Fig. 1L; supplementary material Fig. S3H). In 5 dpf mosaic transgenics, KRAS^G12D transgene expression could be seen in a subset of muscle cells, possibly reflecting an expansion of muscle progenitor cells in larvae that go on to form tumors (supplementary material Fig. S4H,X,FF). In a rare (1 out of 71) 24 hpf rag2:KRAS^G12D embryo, an aggregate of transgene-positive cells could be seen (supplementary material Fig. S4O), suggesting an expansion of KRAS^G12D-expressing cells even at this early stage.

The tumor models described here do not show dramatic differences in global RAS expression or signaling within tumors. However, we cannot exclude the possibility that there are subpopulations of cells within tumors with very high or low levels of RAS expression that contribute to the differences in phenotypes that we have observed. It also remains unclear how RAS levels differ in the cells of origin at the time of transformation and how RAS expression may influence initial tumor formation and development. Because RAS levels are known to affect tumor phenotype in other models (Le et al., 2007), both cell of origin and RAS expression levels might contribute to the observed differences in tumor phenotype in our studies.

Recent experiments have demonstrated that myf5-expressing cells are highly enriched for tumor-propagating activity in rag2 ERMS tumors (Ignatius et al., 2012). Here, we have shown that myf5-expressing cells are enriched for tumor-propagating activity in all zebrafish RMS tumor types. myf5 expression, which is high in activated satellite cells and proliferating myoblasts, decreases during the course of muscle differentiation (Tomczak et al., 2004). Our data suggest that the most immature subpopulations within RMS tumors have the highest self-renewal capacity and also raise the interesting possibility that RMS cells behave according to a cellular hierarchy reminiscent of normal myogenesis, in which the most immature cells are capable of self-renewal and differentiation into cells that lack self-renewal and proliferative capacity.

Myf5 is a myogenic regulatory factor with important roles in muscle specification and regeneration (Rudnicki et al., 1993; Cooper et al., 1999; Ustanina et al., 2007). It is highly expressed in activated satellite cells and is upregulated in zebrafish, human and murine ERMS (Cornelison and Wold, 1997; Langenau et al., 2007; Zibat et al., 2010; Rubin et al., 2011). In rag2 and cdh15 tumors, myf5-expressing cells are abundant and have high self-renewal capability, reminiscent of activated satellite cells. In mylz2 tumors, these cells are rare, express myf5:GFP at low levels, and display a strong propensity to undergo terminal differentiation. These observations suggest that satellite cell programs regulated by myf5 may be reactivated in differentiating cell types that normally lack self-renewal and proliferative capacity, and that myf5 might be an important driver of self-renewal in ERMS. mylz2 tumors were the only tumors in which myf5-negative cells had discernible tumor-propagating activity. As these cells form the bulk of mylz2 tumors, they represent a significant contributor to tumor growth in this model and suggest that self-renewal capability might remain active in more differentiated cell types in mylz2 tumors. Taken together, our data suggest that the cellular context of the tumor-initiating cell during muscle development and pathways regulating normal muscle development may determine the self-renewal and differentiation programs used during RMS generation and growth.

In summary, we have demonstrated that induction of oncogenic RAS expression at different stages during muscle development has profound effects on the ability of tumor cells to recapitulate normal myogenesis, thus altering the tumorigenic capability of these cells. Our studies also suggest an important role for developmental pathways in mediating tumor behavior. Our zebrafish models will be useful reagents for elucidating the pathways that govern differentiation versus self-renewal and for understanding how cellular context can inform tumor behavior in RMS.

We thank Chuck Kaufman, Alison Taylor and Ginggi Storer for helpful discussions regarding this manuscript; Christian Lawrence, Isaac Adatto and Li-Kun Zhang for expert fish care; and Ellen Durand, Emily Huang, Ronald Mathieu and John Daley II for technical assistance.