Abstract Cancer stem cells that display tumor-initiating properties have recently been identified in several distinct types of malignancies, holding promise for more effective therapeutic strategies. However, evidence of such cells in sarcomas, which include some of the most aggressive and therapy-resistant tumors, has not been shown to date. Here, we identify and characterize cancer stem cells in Ewing's sarcoma family tumors (ESFT), a highly aggressive pediatric malignancy believed to be of mesenchymal stem cell (MSC) origin. Using magnetic bead cell separation of primary ESFT, we have isolated a subpopulation of CD133+ tumor cells that display the capacity to initiate and sustain tumor growth through serial transplantation in nonobese diabetic/severe combined immunodeficiency mice, re-establishing at each in vivo passage the parental tumor phenotype and hierarchical cell organization. Consistent with the plasticity of MSCs, in vitro differentiation assays showed that the CD133+ cell population retained the ability to differentiate along adipogenic, osteogenic, and chondrogenic lineages. Quantitative real-time PCR analysis of genes implicated in stem cell maintenance revealed that CD133+ ESFT cells express significantly higher levels of OCT4 and NANOG than their CD133− counterparts. Taken together, our observations provide the first identification of ESFT cancer stem cells and demonstration of their MSC properties, a critical step towards a better biological understanding and rational therapeutic targeting of these tumors. [Cancer Res 2009;69(5):1776–81]

Supplementary Data from Identification of Cancer Stem Cells in Ewing's Sarcoma

PURPOSE. Loss of cyclin-dependent kinase inhibitors p16INK4a, p27,and p21 has been linked to melanoma progression. We evaluated their expression in conjunctival melanocytic proliferations and explored correlations among genomic, RNA, and protein levels of p16INK4a. METHODS. Expression of p16INK4a, p27,and p21 was analyzed by immunohistochemistry in 51 conjunctival nevi, 38 conjunctival melanocytic intraepithelial neoplasia (C-MIN), and 49 conjunctival melanomas (CJMs). Whole-genome/exome sequencing and RNA sequencing were available for 14 and 11 CJMs, respectively. Three CJM cell lines were treated with a DNA methyltransferase inhibitor. RESULTS. p16INK4a was expressed in 92% of nevi, 23.7% of C-MIN, and 42.3% of CJMs; p27 in 98%, 13.5%, and 26.5%, respectively; and p21 in 33.3%, 2.8%, and 28.6%, respectively. Downregulation of p16INK4A and p27 in CJMs versus nevi was significant. No CDKN2A mutations were identified. One CJM showed homozygous CDKN2A loss and two had heterozygous loss, all lacking p16INK4a expression. Among 11 cases with normal ploidy, p16 mRNA was detected in eight (72.3%), but protein in only four (36.4%), suggesting post-transcriptional and post-translational regulation. Treatment with 5-aza-2'-deoxycytidine upregulated p16INK4a and reduced proliferation in two CJM cell lines. CONCLUSIONS. p16INK4a and p27 are significantly downregulated in CJM and may help in the histopathological differential diagnosis between melanoma and conjunctival nevus. p16INK4a loss involves genomic, epigenetic, and post-translational mechanisms. Restoring p16 function may reduce CJM aggressiveness.

KDM6A KO decreases Ewing sarcoma tumor growth. A, Colony formation assay of KDM6A KO cells (sgRNA#1 and sgRNA#2) compared with parental cells and sgCTRL in A-673. B, Tumor growth curves of the average volume of xenografts derived from subcutaneous injection of KDM6A KO cells compared with parental and sgCTRL A-673 cells in nude athymic mice (left, n = 11 for parental, n = 11 for sgKDM6A#1, and n = 9 for sgCTRL; right, n = 9 for parental, n = 7 for sgKDM6A#2, and n = 12 for sgCTRL; n, number of tumors). C, RT-qPCR determination of KDM6A targets CDH11 and IRS2 in RNA extracts from four tumors excised from parental, sgCTRL, and sgKDM6A (sgRNA#1) xenografts of A-673 cells. EWSR1::FLI1 and KDM6A expression are also shown. GAPDH was used as a housekeeping gene. D, Kaplan–Meier curve representing the percentage of xenograft tumors that reach a tumor volume of 1,000 mm3 within days after injection in parental, sgCTRL, and KDM6A KO (sgRNA#1 and sgRNA#2) in A-673 cells. E, As in D, representing the percentage of xenograft tumors that reach a tumor volume of 1,500 mm3 in A-673 sgCTRL+ empty vector (sgCTRL; n = 11), sgKDM6A#1 + empty vector (sgKDM6A#1; n = 9), and upon overexpression of KDM6A WT (sgKDM6A#1 + KDM6A WT; n = 8) and the dead mutant (H1146A/E1148A; sgKDM6A#1 + KDM6A-mut; n = 11) in sgKDM6A#1. F, IHC staining of CD99, KDM6A, Ki-67, and NF200 on sections of tumors excised from sgCTRL and sgKDM6A (sgRNA#1) xenografts of A-673 cells. White scale bar, 50 µm. H&E, hematoxylin and eosin. Statistical significance was determined by repeated measures two-way ANOVA (B) and the Kruskal–Wallis test with Dunn multiple comparison correction (C) relative to sgCTRL. Survival analysis was performed using the log-rank (Mantel–Cox) test (D and E) relative to sgCTRL (D) and to sgCTRL and sgKDM6A#1 (E). Error bars, SEM (B and C). *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not significant.

KDM6A recruits BRG1 to EWSR1::FLI1-activated enhancers in a demethylase-independent manner. A, Western blot of KDM6A, KDM6B, and EWSR1::FLI1 in whole cell extracts upon KDM6A KO with two sgRNA sequences (#1 and #2) in A-673 cells. Tubulin was used as a loading control. B, Heatmap showing expression levels of significantly downregulated genes in the vicinity (100 kb) of EWSR1::FLI1-KDM6A ChIP-seq peaks upon sgKDM6A#1 and sgKDM6A#2 in A-673 cells. C, RT-qPCR of EWSR1::FLI1-KDM6B targets with or without KDM6A ChIP-seq peaks in A-673 sgKDM6A#1 and sgKDM6A#2. D, Box plot depicting the average ChIP-seq signal of H3K27me3 at 4,369 peaks exclusive to sgCTRL or 1,616 peaks for sgKDM6A#1 (sgCTRL+/sgKDM6A− and sgCTRL−/sgKDM6A+, respectively) and 1,610 peaks in common (sgCTRL+/sgKDM6A+). E, Venn diagram depicting genes associated with H3K27me3 peaks in sgCTRL and sgKDM6A#1. F, UCSC Genome Browser signal tracks for H3K27me3 at the SYT1 intronic enhancer in sgCTRL and sgKDM6A#1. G, RT-qPCR of EWSR1::FLI1-KDM6A targets with (left) or without KDM6B (middle) in A-673 sgCTRL+ empty vector (sgCTRL), sgKDM6A#1 + empty vector (sgKDM6A#1), and upon overexpression of KDM6A WT (sgKDM6A#1 + KDM6A WT) and the dead mutant (H1146A/E1148A; sgKDM6A#1 + KDM6A-mut) forms in sgKDM6A#1. H, Venn diagram depicting the overlap between EWSR1::FLI1-KDM6A and BRG1 peaks in A-673 cells. I, Box plot of BRG1 ChIP-seq signal on the common set of 1,808 EWSR1::FLI1-KDM6A and BRG1 peaks upon sgKDM6A#1 in A-673. J, ChIP-qPCR of BRG1, KDM6A, and H3K27me3 enrichment at the enhancer and promoter regions of IRS2 and SYT1 KDM6A-activated targets upon sgKDM6A#1. DICER was used as a negative control region. Kruskal–Wallis test with Dunn correction relative to sgCTRL (C and I), Wilcoxon signed-rank test (D), two-tailed Mann–Whitney test relative to sgKDM6A#1 (G), and ordinary two-way ANOVA with Holm–Šídák test relative to control (J) were applied. GAPDH was used as a housekeeping gene (C and G). Error bars indicate SEM (C, G, and J) of three independent biological experiments and SD (D and I). *, P < 0.05; **, P < 0.01; ***, P < 0.001;****, P < 0.0001; ns, not significant.

Knockdown of KDM6A and KDM6B downregulates EWSR1::FLI1-activated targets. A, Box plot representing the RNA levels of KDM6A, KDM6B, and EWSR1::FLI1 in control and EWSR1::FLI1 knockdown in A-673 Ewing sarcoma cells from Orth and colleagues (45). B, Metagene plot showing KDM6A (left) and KDM6B (right) ChIP-seq signals in 3,634 and 1,000 peaks, respectively, at TSS (0) and within a 1,000 kb window in control (shCTRL) and upon doxycycline induction of EWSR1::FLI1 knockdown in A-673 (shEWSR1::FLI1). C, ChIP-qPCR of EWSR1::FLI1 (EF), KDM6A (left) and KDM6B (right) enrichment at the enhancer region of IRS2 and NKX2-2 genes, respectively, upon doxycycline-induced knockdown (+dox) of the fusion in A-673 cells. D, Box plot representing the RNA levels in A-673 cells of KDM6A, KDM6B, and EWSR1::FLI1 target genes from Riggi and colleagues (16). RPKM, reads per kilobase of transcript per million mapped reads. E, Western blot showing levels of KDM6A and KDM6B in whole cell extracts (top) and H3K27me3 in histone extracts (bottom) upon KDM6A (left) or KDM6B (right) knockdown with two shRNA sequences (sh#1 and sh#2) at 72 hours of doxycycline induction in A-673 cells. Tubulin and H3 were used as loading controls. Levels were quantified relative to tubulin/H3 and control. F, Gene set enrichment analysis curves and normalized enrichment scores (NES) for the Hallmark collection of epithelial-to-mesenchymal transition (EMT) upon shKDM6A#2 (top) and shKDM6B#2 (bottom) targets in A-673 cells. FDR is also shown. G, Heatmap and dendrogram showing expression levels of genes in the vicinity (100 kb) of EWSR1::FLI1-KDM6A (left) and EWSR1::FLI1-KDM6B (right) ChIP-seq peaks significantly downregulated upon knockdown of each demethylase in A-673 cells. n indicates the number of deregulated direct targets upon knockdown. H, UCSC Genome Browser signal tracks for KDM6A, KDM6B, EWSR1::FLI1, and H3K27ac at the SYT1 gene. RNA-seq tracks for shKDM6A (shRNA#2) and shKDM6B (shRNA#2) and the corresponding shCTRL in A-673 are shown. I, As in H, at the IGF1 gene. Statistical differences between groups were assessed by the Student t test (A), Mann–Whitney t test (C), and Wilcoxon signed-rank test (D). Error bars in A, C, and D indicate SD. *, P < 0.05; ***, P < 0.001; ns, not significant.

Impact of KDM6B KO on EWSR1::FLI1 targets. A, Western blot showing levels of KDM6A, KDM6B, and EWSR1::FLI1 in whole cell extracts upon KDM6B KO with two sgRNA sequences (#1 and #2) in A-673 cells. Tubulin was used as a loading control. B, Heatmap showing expression levels of significantly downregulated genes in the vicinity (100 kb) of EWSR1::FLI1-KDM6B ChIP-seq peaks upon sgKDM6B#1 and sgKDM6B#2 in A-673 cells. C, RT-qPCR of EWSR1::FLI1-KDM6B targets with or without KDM6A ChIP-seq peaks in A-673 sgKDM6B#1 and sgKDM6B#2. GAPDH was used as a housekeeping gene. D, Box plot depicting the average ChIP-seq signal of H3K27me3 at 3,248 peaks exclusive to sgCTRL or 2,366 peaks for sgKDM6B#2 (sgCTRL+/sgKDM6B− and sgCTRL−/sgKDM6B+, respectively) and 3,473 peaks in common (sgCTRL+/sgKDM6B+). E, Venn diagram depicting the overlap between genes associated with H3K27me3 peaks in sgCTRL and sgKDM6B#2. F, UCSC Genome Browser signal tracks for H3K27me3 at the intronic enhancer of JARID2 in sgCTRL and sgKDM6B#2. G, ChIP-qPCR of H3K27me3 enrichment at the enhancer region of NKX2-2 and CCND1 genes upon KDM6B KO with sgRNA#2. ENC1 was used as a negative control region. H, Venn diagram depicting the overlap of EWSR1::FLI1-KDM6B and BRG1 peaks in A-673 cells. Statistical significance was determined by the Kruskal–Wallis test with Dunn multiple comparison correction relative to sgCTRL (C), Wilcoxon signed-rank test (D), and ordinary two-way ANOVA with Dunnett multiple comparisons test relative to IgG (G). Error bars indicate SEM (C and G) of three independent biological experiments or SD (D). *, P < 0.05; **, P < 0.01; ****, P < 0.0001.

KDM6A and KDM6B colocalize genome-wide with EWSR1::FLI1 at primed and active enhancers. A, Pie chart showing the genomic distribution of KDM6A, KDM6B, and EWSR1::FLI1 peaks relative to functional categories, including promoters, genic regions, and intergenic regions in A-673 cells. B, Bar plot depicting the percentage of total regulatory elements in the genome overlapping with KDM6A, KDM6B, and EWSR1::FLI1 peaks. C, Top, Venn diagram depicting the overlap of KDM6A, KDM6B, and EWSR1::FLI1 peaks in A-673 cells. Bottom, table shows the top MEME DNA motifs and the corresponding E value for every group of peaks. D, Box plots depicting the average ChIP-seq signal of H3K27ac (top) and H3K4me1 (bottom) on each subset of peaks. E, Heatmap of KDM6A, KDM6B, EWSR1::FLI1, H3K27ac, H3K4me3, and H3K4me1 ChIP-seq signals for each group of peaks. F, UCSC Genome Browser signal tracks for KDM6A, KDM6B, EWSR1::FLI1, H3K27ac, and H3K4me1 in A-673 cells at the CDH11 enhancer (top) and the NKX2-2 enhancer (bottom). EWSR1::FLI1-KDM6B peaks, with or without KDM6A (A-B-EF or B-EF, respectively), are represented as black bars below the tracks. Statistical significance was determined by the Wilcoxon signed-rank test relative to KDM6B-EF peaks (D). Error bars in D, SD. ****, P < 0.0001.

H3K27me3 genome-wide redistribution upon EWSR1::FLI1 overexpression in hpMSCs. A, Western blot showing levels of EWSR1::FLI1 and H3K27me3 in whole cell and histone extracts, respectively, in control (CTRL) and upon EWSR1::FLI1 overexpression in hpMSC (EF). Tubulin and histone H4 were used as loading controls. H3K27me3 was quantified relative to H4 and control. B, Heatmap depicting H3K27me3 and EWSR1::FLI1 (EF) ChIP-seq in 4,080 H3K27me3-EWSR1::FLI1 coinciding bins for control and EWSR1::FLI1 hpMSC. C, Ranking of genes associated with H3K27me3 signal strength in bins (red, genes gaining; blue, genes losing H3K27me3) upon EWSR1::FLI1 overexpression in hpMSC. D, UCSC Genome Browser signal tracks for H3K27me3, H3K27ac, and EWSR1::FLI1 in control and EWSR1::FLI1 hpMSC at the NKX2-2 gene. Up or down bins correspond to regions that gain or lose H3K27me3 upon EWSR1::FLI1 introduction, respectively, and are represented as black rectangles below tracks. Clusters of bins are highlighted in light blue. E, As in D at the TGFBI gene. F, Violin plots representing mRNA levels of KDM6A (top) and KDM6B (bottom) in a panel of Ewing sarcoma (EwS), osteosarcoma (OS), rhabdomyosarcoma (RMS), and synovial sarcoma (SS) cell lines extracted from Barretina and colleagues (50). G, As in F, in primary sarcoma tumors from the NCBI GEO public repository. MSCs derived from healthy bone marrow were used as control tissue. H, IHC staining of KDM6A (top) and KDM6B (bottom) on sections of representative primary Ewing sarcoma tumors (PT) from the cohort of 43 tumors from our institution, counterstained with hematoxylin–eosin. Tumor samples are numbered with “#.” White scale bar, 100 µm. Statistical significance was determined by one-way ANOVA test with Holm–Šídák multiple comparison test (F and G) and relative to Ewing sarcoma. *, P < 0.05; **, P < 0.01; ****, P < 0.0001.

KDM6A knockout decreases EwS tumor growth.

Knockdown of KDM6A and KDM6B downregulates EWSR1::FLI1-activated targets.

Excel file showing genes with gain or loss of H3K27me3 upon EWSR1::FLI1 introduction in hpMSC.

H3K27me3 genome-wide redistribution upon EWSR1::FLI1 overexpression in hpMSCs

KDM6B KO decreases tumor growth. A, Colony formation assay of KDM6B KO cells (sgRNA#1 and sgRNA#2) compared with parental and sgCTRL A-673 cells. B, Bar charts showing the number of colonies from A, in parental, sgCTRL, and sgKDM6B (sgRNA#1 and sgRNA#2) A-673 cells. C, Tumor growth curve of the average volume of xenografts derived from the subcutaneous injection of KDM6B KO cells (sgRNA#1 and sgRNA#2, n = 11 each) compared with sgCTRL A-673 cells (n = 8, where n indicates the number of tumors) in nude athymic mice. D, Kaplan–Meier curve representing the percentage of xenograft tumors that reach a tumor volume of 1,000 mm3 within days after injection in sgCTRL and KDM6B KO (sgRNA#1 and sgRNA#2) A-673 cells. E, IHC staining of CD99, KDM6B, and Ki-67 on sections of tumors excised from sgCTRL and sgKDM6B (sgRNA#1 and sgRNA#2) xenografts of A-673 cells. Black scale bar, 50 µm. H&E, hematoxylin and eosin. F, Left,depletion of KDM6A, enriched at single GGAA regions bound by EWSR1::FLI1, perturbs BRG1 recruitment causing repression of target genes in a demethylase-independent manner. KDM6B colocalizes with EWSR1::FLI1 and BRG1 at GGAA repeats, and its depletion results in repression and a mild accumulation of H3K27me3. Statistical significance was determined by the Kruskal–Wallis test with Dunn multiple comparison correction (B) and repeated measures two-way ANOVA (C). Survival analysis was performed using the log-rank (Mantel–Cox) test (D) relative to sgCTRL. Error bars, SEM (B). *, P < 0.05; **, P < 0.01; ***, P < 0.001; F, Created in BioRender. Figuerola, E. (2025) https://BioRender.com/ikv5h9i.

Excel file showing associated genes for KDM6A, KDM6B and EWSR1::FLI1 ChIP-seq data.

KDM6A and KDM6B co-localize genome-wide with EWSR1::FLI1 at primed and active enhancers.

Ewing sarcoma is a highly aggressive tumor arising in bones and soft tissues, driven by the fusion oncoprotein EWSR1::FLI1. This aberrant transcription factor binds to GGAA microsatellites, causing epigenetic reprogramming through the formation of active neoenhancers in a permissive cellular context. Inhibition of the oncogene remains challenging, and current efforts seek to exploit emergent epigenetic treatments targeting EWSR1::FLI1 cofactors. In Ewing sarcoma, neoenhancers are characterized by strong enrichment of the active H3K27ac mark and a concomitant lack of the repressive H3K27me3 mark. These regions are typically decorated with high levels of H3K27me3 prior to EWSR1::FLI1 activation. In this study, upon expression of EWSR1::FLI1 in human pediatric mesenchymal stem cells, considered the putative cell of origin of Ewing sarcoma, we unraveled the genome-wide redistribution of H3K27me3. Stemming from these results, we elucidated the contribution of the H3K27me3 demethylases KDM6A/UTX and KDM6B/JMJD3 in the transcriptional activity of EWSR1::FLI1-induced enhancers. KDM6A had a demethylase-independent role in recruiting BRG1 at EWSR1::FLI1-primed enhancers containing single GGAA motifs, which was critical for Ewing sarcoma tumor growth. Conversely, KDM6B demethylated H3K27me3 at specific EWSR1::FLI1-active enhancers, colocalizing with BRG1 at GGAA repeats. Loss of KDM6B impaired the growth of Ewing sarcoma tumor xenografts. These results highlight KDM6 demethylases as EWSR1::FLI1 functional partners with potential as targets for treating Ewing sarcoma.Significance: KDM6A and KDM6B mediate critical mechanisms behind EWSR1::FLI1 transcriptional activation program involving demethylase-independent and dependent functions, respectively, supporting the development of therapeutic strategies targeting these demethylases in Ewing sarcoma.

Excel file with reagents information.

Impact of KDM6B knockout in EWSR1::FLI1 targets.