Comprehensive Molecular Profiling of Desmoplastic Small Round Cell Tumor

This article is featured in Highlights of This Issue, p. 1097

Desmoplastic small round cell tumor (DSRCT) is a rare and aggressive small round blue cell malignancy, which occurs predominantly in adolescent and young adult males, and is characterized by abdominopelvic sarcomatosis exhibiting multilineage cellular nests of epithelial, muscular, mesenchymal, and neural differentiation admixed with desmoplastic stroma (1). The exact incidence of DSRCT is poorly defined, although because its histopathologic description at least 850 cases have been reported in the literature (2). Overall survival at 5 years from retrospective series ranges from 11% (3, 4) to 28% (5, 6). Presenting signs and symptoms are often gastrointestinal in nature, and can include abdominal pain and bloating, change in stool pattern, weight loss or weight gain secondary to ascites, and nausea. The cancer is thought to originate along the peritoneal surface and is almost universally metastatic at the time of presentation with common sites including liver, spleen, and nodal involvement above the diaphragm (6), although disease can arise in disparate sites including the testes (7–9) and the central nervous system (10–12). Several staging systems have been proposed (13), with the most recent using imaging characteristics to define intermediate (no liver involvement or ascites), high-risk (either liver involvement or ascites), and very high-risk disease (both liver involvement and ascites) (6).

Given the rarity of this disease, no randomized trials addressing treatment have been performed, and nearly all the available literature describes anecdotal or retrospective experiences. As such, no standard of care therapy has emerged, although treatment almost always includes high-dose alkylator-based chemotherapy (14), and an attempt at complete cytoreductive surgery. The inclusion of whole abdomen/pelvis radiation therapy using an intensity-modulated approach is advocated by some centers, (3, 15, 16) but not by others (5). Additional consolidative local control approaches include investigational treatment with radiolabeled antibody (17), as well as hyperthermic intraperitoneal chemotherapy (HIPEC; refs. 18–21). Recurrence despite high-intensity, multi-modality therapy is common (22–24).

DSRCT is molecularly characterized by the pathognomonic EWSR1–WT1 t(11;22) (p13:q12) translocation (25–27). Beyond this disease defining translocation, the understanding of recurrent oncogenic alterations or genomically defined subsets of DSRCT has remained limited (28). Moreover, there is a marked paucity of established and characterized preclinical DSRCT models. An early study based on hotspot mutation genotyping of 29 common oncogenes (not including FGFR4) in 24 DSRCT samples did not find any mutations (29). An additional study described that nearly one third of 135 mutated genes detected from whole exome sequencing of six DSRCT samples were related to the DNA damage response (DDR) or epithelial–mesenchymal transition (EMT) pathways, suggesting their biologic importance for this disease (30). Another series compared 35 DSRCT samples with 88 Ewing sarcoma samples using a combination of next-generation sequencing (NGS), IHC, chromogenic in situ hybridization or fluorescence in situ hybridization and detected higher expression levels of androgen receptor (AR), TUBB3, EGFR, and TOPO2A (31). The most recently published series describes 83 DSRCT samples analyzed by the FoundationOne Heme assay, and identified recurrent alterations in FGFR4, ARID1A, TP53, MSH3, and MLL3 (32).

Herein, we describe a multimodal approach to define the molecular landscape of DSRCT using hybridization capture-based NGS assay MSK-IMPACT (33) of 68 matched tumor-normal samples from 53 patients, a targeted RNA sequencing assay MSK-Solid Fusion (34) from 17 patients, and whole-genome and RNA sequencing of 10 samples from 9 patients. Furthermore, we describe MSK-IMPACT data from 24 established DSRCT patient-derived xenograft (PDX) models from 10 patients.

This project was approved by the Institutional Review Board of Memorial Sloan Kettering Cancer Center (MSKCC) and was conducted in accordance with the U.S. Common Rule. Formalin-fixed paraffin-embedded DSRCT samples from a subset of patients treated at MSKCC between 2014 and 2019 were submitted for clinical sequencing using the MSK-IMPACT panel (35).

The MSK-IMPACT assay interrogates all exons and select introns of 468 genes to identify mutations, copy number changes, microsatellite instability status, and select structural variants in its current iteration (previous versions interrogated 341 and 410 genes), and was conducted as described previously (36). DNA from both formalin-fixed paraffin-embedded or fresh tissue biopsies as well as a matched normal blood sample was extracted and sheared. Barcoded libraries from tumor and normal samples were captured, sequenced, and subjected to a custom pipeline to identify somatic mutations. Analysis consisted of deep sequencing of all exons and selected introns of a cancer-associated gene panel. Allele-specific copy number analysis was performed with FACETS, utilizing the additional information from a tumor/matched normal design. A segment mean ≥0.2 or ≤0.2 was used to define low-level gains and losses, respectively. In an unmatched analysis of our DSRCT cohort, the matched normals were removed and the pipeline was run using a pooled normal control to find suspected somatic mutations.

Tumor was sequenced to approximately 80X with a matched normal sample at 60X. Single-base substitutions were called using CaVEMan (Cancer Variants through Expectation Maximization; http://cancerit.github.io/CaVEMan/; ref. 37), Strelka2 (https://github.com/Illumina/strelka), and MuTect2 (https://github.com/broadinstitute/gatk). Copy number and cellularity information for CaVEMan were predicted with the Battenberg algorithm (37) using 1,000 Genomes (38) loci within the NGS data. Small somatic insertions and indels were identified using a modified version of Pindel (https://github.com/cancerit/cgpPindel; ref. 39), Strekla2 (https://github.com/Illumina/strelka), and MuTect2 (https://github.com/broadinstitute/gatk). Structural rearrangements were detected by BRASS (Breakpoints via Assembly; https://github.com/cancerit/BRASS), GRIDSS (https://github.com/PapenfussLab/gridss), and SvABA (https://github.com/walaj/svaba). To improve specificity, a number of custom postprocessing filters were applied. Mutational signature analysis of the substitutions was performed using the R package MutationalPatterns (https://bioconductor.org/packages/release/bioc/html/MutationalPatterns.html). Small insertion/deletions were interrogated for the presence of either short tandem repeat or microhomology at the breakpoints, as described previously (40) Data relating to this patient will be deposited in the European Genome-phenome Archive (www.ebi.ac.uk/).

Tumor specimens were analyzed using the MSK-Solid Fusion assay, a targeted RNA-based panel that utilizes the Archer Anchored Multiplex PCR (AMPTM) technology and NGS to detect genes fusions in solid tumor samples (41). Unidirectional gene-specific primers (GSP) were used to target specific exons in 62 genes known to be involved in chromosomal rearrangements. Final targeted amplicons were sequenced on an Illumina MiSeq instrument and data were analyzed using the ArcherTM Software (V4.0.10).

Raw expression data from Filion and colleagues on 137 sarcomas samples was obtained (42). Probe set data were normalized and expression estimates were derived using robust multi-array average (RMA) as previously described (43). Specifically, FGFR4 expression in DSRCT was compared with four other fusion-positive sarcomas.

JN-DSRCT1 cells were plated in 384-well plates with a density of 500 cells/well, and were maintained in DMEM/F12 media with 10% FBS, 1% penicillin/streptomycin, and fungizone at 37°C. The cells were tested for mycoplasma as described previously (44) approximately once every 4 months (last performed October 2020). The alamar blue assay was utilized as described previously (45) with reagent added at hour 72, and results recorded at hour 96. A negative control with 1% DMSO, and a positive control with 1% killer mix in 1% DMSO were utilized. IC₅₀ values were calculated using Prism GraphPad.

The clinical characteristics of the 53 patients with DSRCT are summarized in Table 1. The median age at diagnosis was 21, with a range of 7 to 47 years. There was a strong male predominance as has been noted in prior studies with 48 (91%) of patients being male, and 5 (9%) being female.

A total of 68 matched tumor/normal samples from 53 patients were analyzed by MSK-IMPACT, a hybridization capture-based NGS assay interrogating all exons and select introns of 468 genes. Almost all samples (63, 92%) represented intra-abdominal disease, whereas five samples (8%) were obtained from extra-abdominal sites. Fifty-four (79%) samples were obtained following treatment of any type, and 14 (21%) prior to any intervention. Somatic mutations as detected by MSK-IMPACT in the 68 samples from 53 patients are shown in Fig. 1A and Supplementary Table S1. Recurrent events were uncommon but included alterations in the TERT promoter (3%), ARID1A (6%), HRAS (4%), FGFR4 (7%; further described below), and TP53 (3%). Alterations in HRAS included one amplification, one deletion, and one G12V mutation known to be a hotspot. The average tumor mutational burden (TMB) was 0.87 mutations per megabase (range 0–3.9), and none of the samples showed microsatellite instability (MSI) based on analysis of the MSK-IMPACT sequencing reads using the MSIsensor algorithm (46).

Although the MSK-IMPACT hybrid capture-based NGS assay routinely detects the EWSR1–WT1 fusion of DSRCT by tiling the EWSR1 introns involved in the genomic rearrangement, a subset of these specimens (n = 17) was also analyzed via an anchored multiplex RT-PCR assay for the detection of gene fusions, the MSK-Solid Fusion assay (47). The EWSR1–WT1 fusion was orthogonally confirmed using the MSK-Solid Fusion assay in all 17 samples tested, and detected breakpoints between exons 7 of EWSR1 and exon 8 of WT1 in 12 samples, exons 9 of EWSR1 and exon 8 of WT1 in 3 samples, and exons 10 of EWSR1 and exon 8 of WT1 in 2 samples. This variability in the exon structure of the EWSR1–WT1 fusion transcript is well described (48).

A recent study reported frequent mutations in multiple genes including MSH3 (14%) and MLL3 (16%) in a cohort of tumors profiled using the FoundationOne Medicine Heme assay (32). Our results using MSK-IMPACT identify a much lower overall mutation rate, with a single sample harboring an MLL3 somatic mutation (R199fs) and none with MSH3 somatic alterations. A major difference between these two studies is the use of a matched germline normal control for each sample profiled in our analysis. Therefore, we compared our MSK-IMPACT profiling results using a matched germline control versus an unmatched analysis where presumptive germline mutations were identified and removed on the basis of population databases (Fig. 1B). The mean TMB was significantly higher in the unmatched analysis (14.34 vs. 0.85; P value <0.01). Furthermore, 12 (18%) KMT2C/MLL3 and 6 (8%) MSH3 variants were identified in the unmatched analysis (whereas only one MLL3 mutation and no MSH3 mutations were found by our standard matched tumor:normal analysis). These data suggest that previous unmatched analyses may have overestimated the somatic mutation rate in DSRCT.

Copy number alterations of 65 matched tumor/normal samples were analyzed using the Fraction and Allele-Specific Copy Number Estimates and Tumor Sequencing (FACETS) algorithm as described previously (49). Three samples were excluded due to a lack of matched normal material. Recurrent loss of heterozygosity (LOH) in chromosome arms 11p, 11q, and 16q were identified in 18%, 22%, and 34% of 50 patients, respectively (Supplementary Fig. S1A). Hierarchical clustering of observed gains and losses reveals that these alterations separate into non-overlapping clusters (Supplementary Fig. S1B).

Paired DNA/RNA whole genome and transcriptome sequencing was performed on 10 DSRCT samples from 9 patients, all but one of whom were previously treated. Mutational burden in these samples was found to range from 0.32 to 3.05 mutations/Mb, with a median of 0.80 mutations/Mb. A circos plot depicting whole-genome sequencing of a representative DSRCT sample is shown in Fig. 2A–D, and the remainder are shown in Supplementary Fig. S2A. A majority of the analyzed DSRCT genomes are relatively simple, with low mutational burden, minor involvement of arm or chromosome level aneuploidies, and few structural rearrangements as exemplified in Fig. 2A–D. However, this simplicity was not universal as demonstrated in additional circos plots in Supplementary Fig. S2A.

Given recent evidence that some canonical gene fusions in Ewing sarcoma are generated from chromoplexy (50, 51), structural variant analysis was performed to determine if similar events were occurring in these DSRCT samples. This analysis revealed that the EWSR1–WT1 fusions in this subset were predominantly simple, balanced translocations as shown in the representative plot in Fig. 2E, and the remainder as shown in Supplementary Fig. S2B.

Small deletions were also observed scattered across the genome, but none were recurrent or thought to involve known tumor drivers (Supplementary Fig. S3). Although the samples overall had relatively few structural events aside from the canonical EWSR1–WT1 translocation, most samples had additional nonclustered, nonrecurrent translocations (Supplementary Fig. S3).

Clonality analysis from WGS data revealed the presence of subclones in all samples analyzed. Signature analyses comparing intratumoral subclones did not reveal significant differences between primary clones and subclones. Furthermore, no suspected secondary tumorigenic drivers of disease were noted in these subclones (data not shown).

In the 10 DSRCT samples analyzed by whole-genome sequencing, mutational signatures 1, five (associated with aging) and eight (associated with double stranded break repair) were found (52), but could not be related to presentation, treatment, or outcome.

Alterations in FGFR4 were identified in 5 of 67 (7%) of tumor samples analyzed by MSK-IMPACT. Amplification of FGFR4 was found in two samples, and a deep deletion in one sample. Notably, the two samples with amplification (from different patients) exhibited remarkably similar copy number profiles as shown in Fig. 3A. In addition, an exon 13 p.V550 L point mutation, previously identified as oncogenic (53), was identified in two samples (Fig. 1A).

To investigate expression levels of FGFR4 in DSRCT, whole transcriptome RNA sequencing results from a cohort of 154 pediatric malignancies, including 8 DSRCT samples, were analyzed (Fig. 3B). A clear trend toward increased transcripts per million (TPM) was noted amongst the DSRCT samples, as the 4 samples with the highest FGFR4 TPMs in the entire cohort were DSRCT. None of 8 DSRCT samples in this cohort had FGFR point mutations, amplification, or deletion. Additionally, microarray expression data from 137 tumor samples across 5 different fusion-associated sarcoma subtypes were analyzed and found to demonstrate very strong differential expression of FGFR4, with a mean of 4.9X higher expression in DSRCT when compared with all other sarcoma types in the analysis (Fig. 3C; Supplementary Table S1).

To explore the potential susceptibility of FGFR4 overexpression in DSRCT to targeted inhibition, the antiproliferative effect of a panel of FGFR4 inhibitors was assessed against the only widely available DSRCT cell line, JN-DSRCT (54). RNA sequencing revealed high expression of FGFR4 in the JN-DSRCT cell line which was similar to DSRCT patient samples, and was higher than that found in Ewing sarcoma patient samples (Supplementary Fig. S4A). Notably JN-DSRCT does not harbor FGFR4 mutation or amplification. In addition, protein expression of FGFR4 was confirmed by Western immunoblotting (Supplementary Fig. S4B). Hepatocellular carcinoma cell lines with established sensitivity and resistance (Hep 3B and H2122, respectively) and the Ewing sarcoma cell line A673 which was expected to be resistant were used as controls. The drug panel consisted of four tyrosine kinase inhibitors with varying activity against FGFR4 including ponatinib, rogaritinib, fisogatinib, and BLU9931. Only ponatinib demonstrated modest activity against the JN-DSRCT cell line with an IC₅₀ of 0.46 μmol/L. None of the other inhibitors tested demonstrated activity against JN-DSRCT with all IC₅₀ values >25 μmol/L (Supplementary Figs. S4C and S4D). These data suggest that FGFR4 overexpression in the absence of amplification or activating mutation is not sufficient to render DSRCT cells responsive to FGFR4 inhibition.

To further characterize the molecular landscape of DSRCT, as well as to produce a set of validated preclinical, in vivo models, a total of 120 freshly extracted DSRCT surgical samples from 39 patients were implanted beginning in September 2017. At the time of this publication, 44 of these samples had engrafted, 10 were pending engraftment or failure, and 66 had failed engraftment, for an overall engraftment rate of 36.6%. Tumor material from a subset of the engrafted models in their first or second passage consisting of 24 patient-derived DSRCT xenografts from 10 separate patients was analyzed using the same 468 gene MSK-IMPACT assay as described above. These data were then compared with MSK-IMPACT data from matched patient tumor samples (Fig. 4).

The pathognomonic EWSR1–WT1 fusion was observed by MSK-IMPACT in all PDX and patient samples. Two patient tumor samples harbored additional oncogenic mutations in HRAS and ARID1A, both of which were recapitulated in the corresponding PDX models. In addition, copy number alterations noted in tumor were nearly all represented in the PDX models. Several mutations were identified in PDX samples (TERT, TP53, BRAF), which were not present in matched patient tumor. MSK-IMPACT sequencing reads were reviewed and the lack of these aberrations in the original patient sample was confirmed.

The primary tumorigenic driver of DSRCT, the EWSR1–WT1 fusion, has been well described (25, 26, 48, 55), but few recent series have described any additional recurrent molecular findings (28, 30–32). As is the case for most malignancies driven by oncogenic transcription factors, knowledge of the tumorigenic fusion has yet to inform rational therapeutic strategies. Mutations in FGFR4 (7%), ARID1A (6%), and TP53 (3%) were identified in the MSK-IMPACT cohort described herein. The somatic alteration rate in this study is much lower than that described in a recent study by Chow and colleagues evaluating a cohort of DSRCT samples profiled using the FoundationOne Heme targeted NGS assay (32). Recurrent mutations in MSH3 and MLL3 were not detected in our cohort when the analysis was performed with a matched germline control but were seen at a significantly higher frequency when an unmatched analysis was performed. The use of public SNP databases for filtering of variant calls from unmatched, tumor-only sequencing is imperfect, as it does not eliminate so-called “private” SNPs that are not present in those databases (36). Therefore, we suspect that some genomic alterations reported in previous studies may represent germline SNPs, rather than tumor-specific somatic mutations.

Despite the quiescent genome suggested by this paucity of recurrent mutations, copy number alterations were noted in a striking number of DSRCT samples. This finding aligns with similar observations in other pediatric solid tumor malignancies including Ewing sarcoma (56), rhabdomyosarcoma (57), and osteosarcoma (58, 59). The clinical significance of these copy number-related subgroups is unknown, and may be elucidated through prospective studies with universal molecular profiling and standardized treatment regimens.

Whole-genome sequencing of DSRCT samples did not reveal additional information regarding secondary driver oncogenic events beyond the EWSR1–WT1 fusion. However, it did identify these driver fusions as simple and balanced translocations (Fig. 2B), distinguishing them from subsets of those found in Ewing sarcoma which have been shown to be generated from chromoplexy (50). This finding is somewhat unsurprising given the same orientation of EWSR1 and WT1 on their respective chromosomes, unlike EWSR1 and ERG, for example, which have opposite orientations on their respective chromosomes necessitating events like chromoplexy for fusion formation (51).

In this report, activating alterations in FGFR4 were noted in a subset of tumors (6%; Fig. 1A). Differential expression of FGFR4 was demonstrated by transcriptome analysis and Affymetrix array-based profiling (Fig. 3B and C; Supplementary Table S2), and did not correlate with FGFR point mutations, amplifications, or deletions. These data are aligned with a recently published series which also identified FGFR4 as a potential therapeutic target in DSRCT (60), as well as the experience in rhabdomyosarcoma in which high FGFR4 expression has not been found to be genetically encoded by FGFR4 amplification (53). The FGFRs consist of four highly conserved transmembrane receptor tyrosine kinases (FGFR1–4), and have received significant attention as anticancer targets (53, 61, 62). Specifically, FGFR4 has been of particular interest in pediatric sarcomas, including both embryonal and alveolar subtypes of rhabdomyosarcoma (53, 63). Furthermore, the EWSR1–WT1 fusion oncoprotein has been shown to contribute to FGFR4 transcriptional upregulation (64), suggesting a mechanism for the differential expression described above. However, in vitro testing of multiple FGFR inhibitors demonstrated limited activity in cell proliferation assays in the established JN-DSRCT cell line. Our data suggest that FGFR4 overexpression alone is insufficient to render sensitivity to biologic inhibition of FGFR4. Rare cases of DSRCT with FGFR4 mutations which have been established as activating in other malignancies (e.g., V550 L point mutation) may however demonstrate susceptibility to FGFR inhibition. Indeed, our finding of two patients with DSRCT with FGFR4 V550L, together with a previous report of FGFR4 V510 L in 3 patients with DSRCT (65), establish FGFR4 as the only receptor tyrosine kinase recurrently activated by somatic mutations in DSRCT, and present in 3% to 4% of cases. Additional investigation of the utility of FGFR4 inhibition in DSRCT in the setting of activating FGFR4 point mutations is warranted and would be facilitated by the generation of isogenic pair-matched cell lines. Although small molecule inhibition of the FGFR4 kinase is unlikely to provide significant antitumor effect in the majority of patients with DSRCT, we demonstrate high expression of FGFR4 across the majority of DSRCT samples in our cohort. This finding provides a rationale for investigation of FGFR4 as a therapeutic target for antibody–drug conjugates or novel immunotherapeutics including chimeric antigen receptor (CAR) T-cell or bi-specific T-cell engaging products.

Some preclinical vulnerabilities have been identified in DSRCT cells without discrete genomic data to explain them. For example, DSRCT PDX models have been shown to be exquisitely sensitive to CHK1 kinase inhibition without clear genomic rationale (66). Although one publication analyzing seven DSRCT samples suggested genomic instability due to functional pathway analysis which converged on the DDR pathway (30), a subsequent analysis in a larger cohort did not confirm this (32), nor was this suggested in the current analysis. However, the strength of the preclinical data has supported the initiation of an ongoing clinical trial evaluating CHK1 kinase inhibition in combination with irinotecan in patients with DSRCT (NCT04095221). These events highlight the gaps in knowledge regarding biologic activity, which are critically needed to dovetail with ongoing molecular discoveries.

Although therapies directed against the underlying fusion oncogene in DSRCT have proved elusive thus far, drug development against this and other similar central drivers may still be possible. Direct shRNA approaches (NCT02736565) and direct inhibition of the E26 transformation-specific (ETS) family of oncoproteins (NCT02657005) are being evaluated in Ewing sarcoma and provide guideposts around which to try to develop additional innovative clinical approaches for DSRCT. Any such effort will significantly benefit from the rich PDX bank described and molecularly characterized here, in which these approaches can be rigorously evaluated.

W.D. Tap reports personal fees from Eli Lilly, EMD Serono, Eisai, Munipharma, C4 Therapeutics, Daiichi Sankyo, Blueprint, GlaxoSmithKline, Agios, NanoCarrier, and Deciphera during the conduct of the study; and has a patent for Companion Diagnostic for CDK4 inhibitors - 14/854,329 pending to MSKCC/SKI and Companion Diagnostic for CDK4 inhibitors - 14/854,329 pending to MSKCC/SKI; scientific advisory board and stock ownership at Certis Oncology Solutions, co-founder and stock ownership at Atropos Therapeutics, and scientific advisory board at Innova Therapeutics. G. Gundem reports other support from ISABL Technologies outside the submitted work. L.H. Wexler reports personal fees from EUSA Pharma outside the submitted work. P.A. Meyers reports other support from Amgen, personal fees from Astellas, Lilly, and Salarius outside the submitted work. E. Pappaemanuil reports other support from Isabl Inc. during the conduct of the study; and is a founder and equity holder of Isabl Inc. No disclosures were reported by the other authors.

E.K. Slotkin: Conceptualization, resources, data curation, supervision, writing–original draft, writing–review and editing. A.S. Bowman: Data curation, investigation, writing–review and editing. M.F. Levine: Data curation, investigation, writing–review and editing. F. Dela Cruz: Resources. D.F. Coutinho: Data curation, investigation. G.I. Sanchez: Investigation. N. Rosales: Investigation. S. Modak: Resources, supervision. W.D. Tap: Resources. M.M. Gounder: Resources. K.A. Thornton: Resources. N. Bouvier: Project administration. D. You: Project administration. G. Gundem: Supervision. J.T. Gerstle: Resources. T.E. Heaton: Resources. M.P. LaQuaglia: Resources. L.H. Wexler: Resources. P.A. Meyers: Resources. A.L. Kung: Conceptualization, resources, data curation, formal analysis, funding acquisition, investigation, writing–review and editing. E. Papaemmanuil: Formal analysis, supervision, writing–review and editing. A. Zehir: Data curation, formal analysis, supervision, investigation, writing–review and editing. M. Ladanyi: Resources, data curation, supervision, investigation, writing–review and editing. N. Shukla: Resources, supervision, investigation, writing–review and editing.

Funding for this work was provided by the Scarlett Fund, the Paulie Strong Foundation, the Montag Family Initiative for Pediatric Cancer Research, the Grayson Fund, the Willens Family Fund, and the MSK Cancer Center Support Grant (No. P30 CA008748).

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.