Cooperation of cancer drivers with regulatory germline variants shapes clinical outcomes

Pediatric malignancies including Ewing sarcoma (EwS) feature a paucity of somatic alterations except for pathognomonic driver-mutations that cannot explain overt variations in clinical outcome. Here, we demonstrate in EwS how cooperation of dominant oncogenes and regulatory germline variants determine tumor growth, patient survival and drug response. Binding of the oncogenic EWSR1-FLI1 fusion transcription factor to a polymorphic enhancer-like DNA element controls expression of the transcription factor MYBL2 mediating these phenotypes. Whole-genome and RNA sequencing reveals that variability at this locus is inherited via the germline and is associated with variable inter-tumoral MYBL2 expression. High MYBL2 levels sensitize EwS cells for inhibition of its upstream activating kinase CDK2 in vitro and in vivo, suggesting MYBL2 as a putative biomarker for anti-CDK2-therapy. Collectively, we establish cooperation of somatic mutations and regulatory germline variants as a major determinant of tumor progression and highlight the importance of integrating the regulatory genome in precision medicine.

T he advent of high-throughput "omics" technologies in oncology enabled assignment of patients to targeted therapies based on somatic mutations in the protein coding genome 1 . However, many childhood cancers including Ewing sarcoma (EwS)-a highly aggressive bone-associated cancerhardly exhibit any recurrent genetic alteration other than pathognomonic and uniformly expressed driver mutations 2,3 . Yet, these tumors show substantial inter-individual heterogeneity concerning clinical behavior and treatment response, which cannot be solely explained by their few additional (epi-)genetic alterations [3][4][5][6] .
Recent studies in humans and model organisms suggested that the effects of a dominant oncogene may depend on variations in the regulatory genome [7][8][9][10][11] . Thus, we hypothesized that oncogenic cooperation of driver-mutations with specific regulatory germline variants may explain inter-individual diversity of clinical outcomes in cancer.
We explore this possibility in EwS, which constitutes a genuine model to study such cooperation for several reasons: First, it is characterized by a simple, nearly diploid genome with a single driver-mutation resulting from chromosomal rearrangements fusing the EWSR1 gene to various members of the ETS family of transcription factors (in 85% FLI1) 2,3,12-14 . Second, EWSR1-FLI1 steers~40% of its target genes by binding DNA at GGAAmicrosatellites, which are thereby converted into potent enhancers [15][16][17][18] . Third, the enhancer activity of EWSR1-FLI1-bound GGAA-microsatellites strongly depends on the inter-individually variable number of consecutive GGAA-repeats 16,17,19 . Together, these characteristics provide an ideal framework to analyze how cooperation of a dominant oncogene (here EWSR1-FLI1) with polymorphic germline regulatory elements (here GGAA-microsatellites) influences the expression of disease-promoting genes that could explain clinical diversity in cancer. In this study, we show, in the EwS model, how such cooperation steers the expression of the functionally and clinically relevant EWSR1-FLI1 target gene MYBL2, thereby determining tumor growth, patient survival, and drug response.

Results
EWSR1-FLI1 regulates MYBL2 via a polymorphic GGAAmicrosatellite. To identify candidate genes with high clinical relevance, we crossed two datasets. The first comprised expression microarrays of A673 EwS cells harboring a doxycycline (DOX)inducible shRNA against EWSR1-FLI1 (A673/TR/shEF1) profiled with/without DOX-treatment (Supplementary Data 1). The second comprised 166 transcriptomes of primary EwS with clinical annotation (Supplementary Data 2). We calculated for each gene represented in both datasets the fold change (FC) of its expression after DOX-induced EWSR1-FLI1 knockdown in A673/TR/ shEF1 cells and the significance for association with overall survival (OS) stratifying patients by expression quintiles of the corresponding gene. Specifically, the latter analysis was carried out by a custom software (GenEx) that automatically calculates the P values for each gene in a given overall survival dataset with matched gene expression data by a Mantel-Haenszel test for patients grouped in the highest versus the lowest expression quintile of the given gene (adjusted for multiple comparisons by the Bonferroni method) (see Methods). This analysis identified MYBL2 (alias B-MYB), encoding a central transcription factor regulating cell proliferation, cell survival, and differentiation 20 , as the top EWSR1-FLI1 upregulated gene, whose high expression was significantly associated with poor OS (nominal P = 9.6×10 −7 , Bonferroni-adjusted P = 0.018) (Fig. 1a, b; Supplementary Data 3).
The EWSR1-FLI1-dependency of MYBL2 expression was validated in time-course experiments in A673/TR/shEF1 on the mRNA and protein level in vitro and in vivo (Fig. 1c,   Supplementary Fig. 1a-c), and in nine additional cultured EwS cell lines (Supplementary Fig. 1d).
Despite this tight regulation of MYBL2 by EWSR1-FLI1, we noted a marked inter-tumor heterogeneity of MYBL2 mRNA expression in 166 primary EwS ( Supplementary Fig. 1e) and in an independent cohort of 208 EwS on protein level stained for p-MYBL2 ( Supplementary Fig. 1f). Interestingly, MYBL2 expression did not correlate with minor variations of EWSR1-FLI1 expression ( Supplementary Fig. 1g), suggesting that interindividual diversity of MYBL2 transcription may be caused differently.
In accordance, re-analysis of published chromatin immunoprecipitation followed by next-generation sequencing (ChIP-seq) data from A673 and SK-N-MC EwS cells 21,22 revealed strong signals for EWSR1-FLI1 that mapped to a polymorphic GGAAmicrosatellite located~150 kb telomeric of MYBL2 (Fig. 1d). In both cell lines, this GGAA-microsatellite exhibited EWSR1-FLI1dependent epigenetic characteristics of an active enhancer indicated by H3K4me1 and H3K27ac marks (Fig. 1d). The EWSR1-FLI1-dependent enhancer activity of this GGAAmicrosatellite was confirmed in reporter assays, for which we cloned fragments of~880 bp from cell line-derived haplotypes differing in their number of consecutive GGAA-repeats (6, 10, or 12 GGAA-repeats) in the pGL3-Fluc vector. Other regulatory variants in the flanking regions were excluded by whole-genome sequencing (WGS) of the parental cell lines and by Sanger sequencing of the cloned fragments (see Methods). In these assays, we observed a positive correlation of the measured enhancer activity and the number of consecutive GGAA-repeats (Fig. 1e).
To test whether EWSR1-FLI1 prefers haplotypes with more consecutive GGAA-repeats, we carried out ChIP-seq analysis using relatively long reads (single-end 150 bp) for EWSR1-FLI1 in the EwS cell line RDES that is heterozygous at the MYBL2associated GGAA-microsatellite (12 versus 14 consecutive GGAA-repeats). In this analysis, we obtained 31 ChIP-seq reads spanning the entire GGAA-microsatellite. In line with our results from reporter assays (Fig. 1e), 71% of these spanning reads (22/ 31) mapped to the longer haplotype, whereas only 29% (9/31) mapped to the shorter one (P = 0.015).
We next performed expression quantitative trait locus (eQTL) analysis in 35 primary EwS tumors for which matched gene expression and WGS data were available. Prior reports suggested that more than 13 consecutive GGAA-repeats at EWSR1-FLI1 bound GGAA-microsatellites delineate a critical number beyond which very strong EWSR1-FLI1 binding and enhancer activity can be observed 16,19,24 , which is in agreement with our ChIP-seq analysis showing preferential EWSR1-FLI1 binding to the longer haplotype as stated above. Classifying all haplotypes in either "short" (≤13 GGAA-repeats) or "long" (>13 GGAA-repeats), we detected a significantly higher MYBL2 expression in EwS tumors with long/long haplotypes compared to those with short/short haplotypes (Supplementary Data 4, Supplementary Fig. 1h).
Together, these findings indicate that MYBL2 is a clinically relevant direct EWSR1-FLI1 target gene, whose expression can be modulated by EWSR1-FLI1 binding to a polymorphic enhancerlike GGAA-microsatellite.
MYBL2 is critical for proliferation and cell survival of EwS cells. To obtain first clues on the functional role of MYBL2 in primary EwS, we performed gene-set enrichment analysis (GSEA) of MYBL2 co-expressed genes in 166 EwS tumors. GSEA revealed that MYBL2 co-expressed genes were strongly enriched in human orthologs of known MYBL2 targets in zebrafish 29 and in signatures related to proliferation 30 , cell cycle progression 31 , and sensitization to apoptosis mediated by a CDK-inhibiting protein 32 (Fig. 2a, Supplementary Data 5), suggesting that MYBL2 may constitute a key downstream mediator of EWSR1-FLI1induced, evolutionary conserved proliferation programs.
To test this hypothesis, we performed MYBL2 knockdown experiments in A673, SK-N-MC, and RDES EwS cell lines with moderate to high baseline MYBL2 expression ( Supplementary  Fig. 2a-c). Using four different siRNAs, we found that MYBL2 silencing reduced proliferation through blockage of G2/M progression, which was accompanied by increased apoptotic cell death ( Fig. 2b-d).
To further explore the function of MYBL2 in EwS growth, we generated DOX-inducible anti-MYBL2 shRNA expression systems in A673 and SK-N-MC cells using two different shRNAs. In both cell lines, DOX-induced MYBL2 silencing significantly reduced clonogenic growth in vitro and tumor growth in vivo compared to a non-targeting control shRNA ( Fig. 2e-g, Supplementary Fig. 2d-g). In line with our transient knockdown experiments, we observed an increased number of stalled mitoses, indicating G2/M blockage, and more apoptotic tumor cells positive for cleaved caspase 3 in MYBL2-silenced xenografts (Fig. 2h, i). Collectively, these findings indicate that MYBL2 is a critical pro-proliferative downstream effector of EWSR1-FLI1 required for proper G2/M transition and cell survival.
MYBL2 mediates its phenotype via upregulation of CCNF, BIRC5, and AURKB. To identify potential direct MYBL2 targets that could explain its pro-proliferative effect, we sequenced RNA of three EwS cell lines with/without MYBL2 knockdown (Fig. 3a). Consistent with our enrichment analyses in primary EwS and functional experiments, GSEA of the identified differentially expressed genes (DEGs) showed that MYBL2 suppression leads to a strong downregulation of the same gene sets comprising human orthologs of zebrafish MYBL2 targets and identical proliferation, cell cycle, and sensitization to CDK-inhibitor mediated apoptosis gene signatures (Fig. 3b, Supplementary Fig. 3a, Supplementary Data 6).
shControl P = 0.008 Gene-sets with FDR q < 0.05 Normalized enrichment score Overall survival (%) We then focused on the 76 most significantly DEGs (mean log2 FC |≥1.5|, Bonferroni-adjusted P < 0.05), of which representative genes were validated by quantitative real time PCR (qRT-PCR) ( Supplementary Fig. 3b, c, Supplementary Data 7). ChIP-seq analysis using a specific anti-MYBL2 antibody revealed that 50 of these 76 DEGs (66%) showed evidence for MYBL2 promoterbinding (Fig. 3c, Supplementary Data 8). Using microarray data of 166 patient tumors in which 92% of these direct MYBL2 targets were represented enabled correlation of their expression levels with that of MYBL2 and with OS of patients stratified by median expression of the corresponding gene (Supplementary Data 9, 10). Among these genes, CCNF, BIRC5, and AURKB stood out for being highly significantly co-expressed with MYBL2 (Bonferroniadjusted P < 0.05, r Pearson ≥ 0.7) (Fig. 3d), and associated with poor OS (Fig. 3e). To investigate their functional role, we individually knocked down either gene using two specific siRNAs in two different EwS cell lines ( Supplementary Fig. 3d) and assessed proliferation and cell viability in vitro. Strikingly, knockdown of these genes broadly phenocopied the antiproliferative and anti-survival effect of MYBL2 silencing (Fig. 3f, g), suggesting that they may constitute important mediators of the pro-proliferative EWSR1-FLI1/MYBL2 transcriptional program. However, as other functionally relevant genes (e.g. MKI67, KIF20A, PIF1) are also regulated by MYBL2 ( Supplementary Fig. 3c), it is conceivable that other genes may contribute to the phenotype of MYBL2.
High MYBL2 levels sensitize EwS cells toward CDK2 inhibition. As there are-to the best of our knowledge-currently no direct MYBL2 inhibitors available, we reasoned that targeting its major upstream cyclin dependent kinase, CDK2, which activates MYBL2 through phosphorylation 20 , may offer a new therapeutic option for EwS patients with high MYBL2 expression. To test this possibility, we treated EwS cells with two small-molecule CDK2 inhibitors (CVT-313 and NU6140). While both inhibitors strongly reduced growth of A673 EwS cells at the lower micro-molar range, sensitivity toward them was dramatically diminished when MYBL2 was suppressed (Fig. 4a). Such differential effect was not observed in control cells expressing a non-targeting shRNA (Fig. 4a). Notably, NU6140 is a dual inhibitor of CDK2 and the major downstream MYBL2 target AURKB. Since this inhibitor enabled to specifically target EwS cells up-and downstream of MYBL2, we tested its effect on EwS growth in vivo. Treatment of NOD/scid/gamma (NSG) mice with NU6140 significantly (P < 0.05) reduced growth of EwS xenografts compared to vehicle (DMSO) (Fig. 4b), and was accompanied by reduced levels of phosphorylated MYBL2 and increased apoptotic cell death (Fig. 4c). However, this inhibitor had no additional effect on growth of xenografts with silenced MYBL2 expression (Fig. 4b), suggesting that MYBL2 is important for the anti-proliferative effect of CDK2 inhibitors. Consistently, different EwS cell lines with high MYBL2 levels showed higher sensitivity toward NU6140 than a EwS cell line with constitutively low MYBL2 expression ( Supplementary Fig. 4a, b). A similar effect on growth of A673 EwS xenografts was observed using the CDK2 inhibitor CVT-313 ( Supplementary Fig. 4c, e). Since we neither observed significant weight loss (Supplementary Fig. 4d) nor histomorphological changes in inner organs in mice treated for 14 days with up to 40 mg kg −1 of either inhibitor, these results indicated that CDK2 inhibition can safely impair growth of EwS tumors and that MYBL2 may serve as a biomarker to predict its efficacy.
Interestingly, we observed in A673/TR/shEF1 cells that CDK2 appears to be moderately upregulated by EWSR1-FLI1 (Supplementary Fig. 4f), and found evidence for binding of EWSR1-FLI1 at the CDK2 locus in EwS cells (Supplementary Fig. 4g). However, whether EWSR1-FLI1 regulates CDK2 expression directly or indirectly remains to be elucidated in future studies.

Discussion
Collectively, our discoveries made in an aggressive childhood cancer exemplify how oncogenic cooperation between a cancer driver-mutation (here EWSR1-FLI1) and a regulatory germline variant (here a polymorphic enhancer-like GGAA microsatellite) can create a major source of inter-tumor heterogeneity determining clinical outcome and drug response through modulation of a druggable key downstream player (Fig. 4d).
To explore the possibility of such oncogenic cooperation in EwS beyond MYBL2, we analyzed the top five additional hits of our initial screen whose high expression was associated with worse patient overall survival (EXO1, C1ORF112, ESPL1, HJURP, RAD54L; Supplementary Data 3) for the presence of EWSR1-FLI1 bound GGAA-microsatellites or ETS-like binding motifs in the vicinity of these genes, and in that case also possible eQTL effects. While no EWSR1-FLI1 binding was observed at the ESPL1 locus, we found evidence for EWSR1-FLI1 binding at GGAA-microsatellites or ETS-like single GGAA-motifs at the other loci. However, most of these EWSR1-FLI1-binding sites did not show genetic variability in WGS data from primary EwS samples, and if so, they appeared to have no eQTL properties ( Supplementary Fig. 5), which may further support the special role of MYBL2 in EwS.
Our results suggest that cooperation between diseasepromoting somatic mutations and regulatory germline variants could constitute a general mechanism to explain diversity of disease phenotypes, possibly beyond cancer. In line with this idea, recent reports for neurodegenerative and metabolic diseases showed that the same disease-causing somatic event/mutation can induce distinct phenotypes depending on (inherited) variations in regulatory elements 7,33,34 . We anticipate that our findings made in the EwS model are translatable to other malignancies, and propose that integration of the regulatory genome in the process of developing new predictive markers and therapeutic strategies is necessary to refine and fully exploit "omics"-based precision medicine.
DNA/RNA extraction, reverse transcription, and qRT-PCR. DNA was extracted with the NucleoSpin Tissue kit (Macherey-Nagel); plasmid DNA was extracted from bacteria with the PureYield kit (Promega). RNA extraction was performed with the NucleoSpin RNA kit (Macherey-Nagel) and RNA was reverse transcribed using the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems). qRT-PCRs were performed using SYBR Select Master Mix (Applied Biosystems) and reactions were run on a Bio-Rad CFX Connect instrument and analyzed using the Bio-Rad CFX Manager 3.1 software. All oligonucleotides were purchased from MWG Eurofins Genomics. For primer sequences see Supplementary Data 11. Transient transfection. For siRNA transfection, cells were seeded in a six-well plate at a density of 1.5 × 10 5 per well in 1.6 ml of growth medium. The cells were transfected with either a negative control non-targeting siRNA (Sigma-Aldrich MISSION siRNA Universal Negative Control #1) or specific siRNAs (25-65 nM, depending on the cell line and the siRNA) and HiPerfect (Qiagen). Cells were retransfected 48 h after the first transfection and harvested 96 h after the first transfection. siRNA sequences are given in Supplementary Data 11. For plasmid transfection, cells were seeded in a six-well plate at a density of 2 × 10 5 per well in 1.8 ml of growth medium. Plasmids were transfected with Lipofectamine LTX and Plus Reagent (Invitrogen). The pGL3 vector used for reporter assays has been described before 16 .
DNA constructs and reporter assays. MYBL2-associated GGAA-microsatellites (with~440 bp 5′ and 3′ flanking regions) from three EwS cell lines were PCRcloned upstream of the SV40 minimal promoter into the pGL3-Fluc vector (Promega) 16 . Primer sequences are given in Supplementary Data 11. The presence of additional variants devoid of the GGAA-microsatellite was ruled out by WGS of the parental cell lines and Sanger sequencing of the cloned fragments. A673/TR/ shEF1 cells (2 × 10 5 per well) were transfected with the Firefly pGL3-Fluc vector containing respective microsatellites and the Renilla pGL3-Rluc vector (Promega) (ratio 100:1) in a six-well plate with 1.8 ml of growth medium. Four hours after transfection, transfection medium was replaced by medium with/without DOX (1 µg ml −1 ; Sigma-Aldrich). Cells were lysed and assayed with a dual luciferase assay system (Berthold) after 72 h. Firefly luciferase activity was normalized to that of Renilla.
CRISPR interference (CRISPRi) and analysis of cell growth. Due to the lack of functional DNAse, CRISPRi does not cause a knockout of the targeted DNA sequence, but blocks protein binding to it 26,28 . For the reported experiments, a DNAse-dead CAS9 (dCAS9) fused to the KRAB effector domain, which promotes an inhibiting chromatin state, is targeted to the genomic region of interest by specific gRNAs to silence the activity of a given enhancer 26,28 . To achieve this, we used a pHAGE TRE dCas9-KRAB vector (Addgene #50917) and a pLKO.1-puro U6 sgRNA BfuAI large stuffer vector (Addgene #52628), the latter containing either two gRNAs, targeting sequences adjacent to the MYBL2-associated GGAAmicrosatellite, or a scrambled control (Supplementary Data 11). Lentivirus production was performed in HEK293T cells. RDES EwS cells were infected with the respective lentiviruses and selected with 1 µg ml −1 puromycin and 1.5 µg ml −1 G418 (both Invivogen). The cells were induced with DOX (1 μg ml −1 ; Sigma-Aldrich) for 5 days, after which MYBL2 and EWSR1-FLI1 levels were measured by qRT-PCR.
For measurement of cell growth, cells were grown in medium containing selection antibiotics and DOX (2 µg ml −1 ) for 14 days as described 37 . Thereafter, 8 × 10 4 cells/well were plated in quadruplicate wells of 24-well plates in the presence of DOX. After four additional days, cells were washed and fixed with trichloroacetic acid for 1 h at 4°C. Then, plates were washed with phosphatebuffered saline (PBS), air dried, and cells were stained with crystal violet (Sigma-Aldrich) for 30 min. Surplus crystal violet was removed by rinsing the plates with PBS. Cell-bound crystal violet was dissolved in 10% acetic acid, and optical density was measured at 595 nm in a DS-11 spectrophotometer (DeNovix Inc.).  Western blot. Protein from A673/TR/shEF1 cells was extracted at d0, d7, d11, d14, and d17 with RIPA and anti-protease cocktail (Roche). Western blots were performed following routine protocols and specific band detection was achieved by the use of rabbit monoclonal anti-FLI1 antibody (1:1000, ab133485; Abcam) 38 , rabbit polyclonal anti-MYBL2 antibody (1:500, sc-725; Santa Cruz) 39 , and mouse monoclonal anti-ß-actin (1:10,000, A-5316; Sigma-Aldrich). Anti-rabbit IgG horseradish peroxidase-coupled antibody (1:3000, Amersham Bioscience) and antimouse IgG horseradish peroxidase coupled antibody (1:3,000; Amersham Bioscience) was used as secondary antibody. Proteins were visualized using chemiluminescence (Pierce ECL Western blot chemiluminescent substrate; Thermo Fisher Scientific).
Proliferation assays. Cells were seeded in a six-well plate at a density of 1.5 × 10 5 per well in 1.6 ml of growth medium. The cells were transfected with either a negative control non-targeting siRNA in duplicate wells (Sigma-Aldrich MISSION siRNA Universal Negative Control #1) or up to four specific siRNAs (MWG Eurofins Genomics) (25-65 nM, depending on the cell line and siRNA) using HiPerfect (Qiagen). siRNA sequences are given in Supplementary Data 11. Retransfection was performed 48 h after the first transfection. Ninety-six hours after the first transfection, cells were harvested (including supernatant), stained with Trypan blue (Sigma-Aldrich), and counted in a standardized hemocytometer (C-Chip, NanoEnTek).
Analysis of cell cycle and apoptosis. Analysis of cell cycle phases was performed by propidium iodide (PI) (Sigma-Aldrich) staining. Cells were transfected with siRNAs equivalently to the proliferation assays (see above), harvested after 96 h (including supernatant), fixed in ethanol (70%) at 4°C, and stained with PI solution (50 µg ml −1 , with 20 µg ml −1 RNAse A (Invitrogen)). Analysis of apoptosis has been performed by combined Annexin V-FITC/PI staining (BD Pharmingen FITC Annexin V Apoptosis Detection Kit II; BD Biosciences). Cells were transfected with siRNAs equivalently to the proliferation assays (see above) and harvested after 96 h (including supernatant). The samples were assayed on an Accuri C6 flow cytometer and analyzed with the Accuri C6 CFlow Plus software. An example of the gating strategy is given in Supplementary Fig. 6.
Colony-forming assays. A673 and SK-N-MC cells containing either a DOXinducible non-targeting control shRNA or MYBL2-targeting specific shRNAs were seeded in triplicate wells of a 12-well plate at a density of 500 cells (A673) or 1000 cells (SK-N-MC) per well in 2 ml of growth medium. Cells were grown with/ without DOX (1 µg ml −1 ; Sigma-Aldrich) for 10-14 days depending on the cell line and afterwards stained with crystal violet (Sigma-Aldrich). Colony number was determined on scanned plates using Fiji (ImageJ) 40,41 .
cDNA library and RNA sequencing (RNA-seq). A673, SK-N-MC, and RDES EwS cell lines were transfected in triplicates with either a negative control nontargeting siRNA or a specific siRNA targeting MYBL2 (siMYBL2_1). Total RNA was extracted using the NucleoSpin II kit (Macherey-Nagel). Complementary DNA libraries were sequenced with an Illumina HiSeq2500 instrument using 150 bp paired-end sequencing. Obtained reads were aligned on the human genome (hg19) using TopHat (version 2.0.6) 42 . Counting of reads on annotated genes from the GRCh37 gene build was done using htseq-count (v. HTSeq-0.5.3p9) 43 with the following parameters: htseq-count -a 10 -q -s no -m union. Sample-to-sample normalization and differential expression analyses were performed using the R package DESeq2 (v.1.18.0) 44 . RNA-seq data were deposited at the Gene Expression Omnibus (GEO; accession code GSE119972).
Chromatin immunoprecipitation and sequencing (ChIP-seq). DNA-protein cross-linking was done in the presence of 1% of paraformaldehyde on 12 × 10 6 A673 or 4 × 10 6 RDES cells, respectively, for each condition for 10 min. Cell lysis, chromatin shearing, immunoprecipitation, and DNA purification were performed with reagents from iDeal ChIP-seq kit for  46 ). Peaks were called with MACS2 with option narrow 47 . To normalize, we took the input dataset from the same cell line. PAVIS was used for peak annotation and visualization 48 . For analysis of the number of EWSR1-FLI1 ChIP-seq reads spanning the different haplotypes of the MYBL2associated GGAA-microsatellite in RDES cells, exclusively the spanning reads were extracted from the BAM file with SAMtools, and mapped to the corresponding haplotype according to CIGAR scores 49 . Significance levels were calculated using a binomial test (p = 0.5). ChIP-seq data concerning MYBL2 were deposited at the GEO (accession code GSE119972).
CDK2 inhibitor assays in vitro. Cells were seeded in a 96-well plate at a density of 5 × 10 3 per well. In case of cells containing DOX-inducible constructs, cells were treated with/without DOX (1 µg ml −1 ; Sigma-Aldrich). Twenty-four hours after seeding or pre-incubation with DOX, respectively, CDK2 inhibitors (CVT-313 or NU6140; Merck and Tocris) were added in serially diluted concentrations ranging from 0.001 to 100 µM. Each well contained an equal concentration of 0.5% DMSO (Sigma-Aldrich). Cells only treated with 0.5% of DMSO served as a control. After 72 h of inhibitor treatment, the plates were assayed on a Thermo Fisher Varioskan plate reader after incubation with Resazurin (20 µg ml −1 ; Sigma-Aldrich) for 6 h.
Xenotransplantation and CDK2 inhibitor treatment in vivo. All mouse experiments were approved by the local authorities in compliance with all relevant ethical regulations (including, but not limited to, tumor size). Sample size was predetermined using power calculations with β = 0.8 and α = 0.05 based on preliminary data and in compliance with the 3 R system (replacement, reduction, refinement). 3 × 10 6 A673 and SK-N-MC cells, containing either a DOX-inducible negative control shRNA or specific shRNAs against EWSR1-FLI1 or MYBL2, were injected subcutaneously with a 1:1 mix of PBS (Biochrom) and Geltrex (LDEV-Free Reduced Growth Factor Basement Membrane Matrix, Thermo Fisher Scientific; max volume 100 µl) in the right flanks of 3-9 months old female or male NSG mice (Charles River Laboratories). For shRNA sequences see Supplementary Data 11. When tumors were palpable, mice were randomized to the control group (17.5 mg ml −1 sucrose (Sigma-Aldrich) in drinking water) or the treatment group (2 mg ml −1 DOX (Beladox, bela-pharm) and 50 mg ml −1 sucrose (Sigma-Aldrich) in drinking water).
Tumor size was measured with a caliper every 2 days and tumor volume was calculated as V = a × b 2 /2 with a being the largest diameter and b being the smallest diameter. Once the tumors reached a volume of 1500 mm 3 respective mice were sacrificed by cervical dislocation. Other humane endpoints were determined as follows: Ulcerated tumors, loss of 20% body weight, constant curved or crouched body posture, bloody diarrhea or rectal prolapse, abnormal breathing, severe dehydration, visible abdominal distention, obese Body Condition Scores (BCS), apathy, and selfisolation. For CDK2 inhibitor treatment in vivo, cells were injected as described above. When tumors were palpable, mice were assigned to either the vehicle (DSMO) or a treatment group (20 or 40 mg kg −1 ), each with or without addition of DOX to the drinking water (2 mg ml −1 DOX; Beladox, bela-pharm). In case of CVT-313 (Tocris) treatment, DOX was not applied. The CDK2 inhibitors NU6140 or CVT-313 (both Tocris) were administered i.p. for 12 days, with a break of 1 day every 4 days of treatment. The experimental endpoint was predetermined as 14 days after first injection of either inhibitor, or if humane endpoints as described above were reached before. To check histomorphological changes of inner organs upon CDK2 inhibitor treatment, we examined hematoxylin and eosin (HE) stained slides of heart, lungs, liver, stomach, pancreas, intestines, kidneys, adrenal glands, bone marrow, and spleen from treated and non-treated mice. Tumor tissues were subjected to HE staining, as well as immunohistochemical staining for p-MYBL2 and cleaved caspase 3 (as described below). Animal experiments were approved by the government of Upper Bavaria and conducted in accordance with ARRIVE guidelines, recommendations of the European Community (86/609/EEC), and UKCCCR (guidelines for the welfare and use of animals in cancer research).
Survival analysis. Microarray data of 166 primary EwS tumors (GSE63157, GSE34620, GSE12102, GSE17618) for which clinical annotations were available were downloaded from the GEO. Data were either generated on Affymetrix HG-U133Plus2.0 or on Affymetrix HuEx-1.0-st microarray chips and separately normalized by RMA using custom brainarray chip description files (CDF, v20).
ComBat was used to remove batch effects 50,51 . Patients were stratified by their quintile or median intra-tumoral gene expression levels. Mantel-Haenszel test was performed to calculate significance levels, using either a custom code (GenEx) for batch queries or GraphPad PRISM version 5 for individual genes (GraphPad Software Inc., CA, USA). P values < 0.05 were considered as statistically significant. Survival data were crossed with gene expression microarray data (Affymetrix HG-U133A2.0) generated in A673/TR/shEF1 cells (GSE27524; 53 h DOX-treatment), which were normalized as described above (RMA with brainarray CDF, v19).
Gene-set enrichment analysis. Using the Affymetrix gene expression dataset comprising 166 primary EwS patients, enrichment of gene-sets that are among MYBL2 co-regulated genes were identified by ranking of Pearson´s correlation coefficient of the expression of every gene with MYBL2 expression and performance of a pre-ranked GSEA with 1000 permutations 52 56 . To call the genotypes of the MYBL2-associated GGAA-microsatellite, we applied HipSTR (v.0.6.2) 23 on the WGS data using a minimum threshold of ten reads. All genotypes passed the following HipSTR default filters: --min-call-qual 0.9; --max-call-flank-indel 0.15; --max-call-stutter 0.15; --min-call-allele-bias -2; --min-call-strand-bias -2.
Human samples and ethics approval. Archived human tissue samples were retrieved from the Institute of Pathology of the LMU Munich (Germany) and the Gerhard-Domagk Institute of Pathology of the University Hospital of Münster (Germany). All patients provided written informed consent. Retrospective and blinded analysis of anonymized samples was carried out upon ethical approval of LMU Munich's ethics committee (approval no. 550-16 UE).
Tissue microarrays and immunohistochemistry. Formalin-fixed paraffinembedded samples were collected at the Institute of Pathology of the LMU Munich 57 . We harvested at least two cores per sample with a core-diameter of 1 mm from all blocks to construct tissue microarrays. All EwS samples showed cytogenetic evidence for a translocation of the EWSR1 gene either as determined by fluorescence in situ hybridization and/or qRT-PCR. The samples were reviewed by a reference pathologist. Four-micrometer sections were cut for immunohistochemistry and antigen retrieval was performed with microwave treatment using the antigen retrieval ProTaqs I Antigen-Enhancer (Quartett) for p-MYBL2 or the Target Retrieval Solution (Agilent Technologies) for cleaved caspase 3. In total, 7.5% aqueous H 2 O 2 solution (room temperature) and blocking serum from the corresponding kits were used for 20 min for blockage of endogenous peroxidase. Then slides were incubated for 60 min with the primary antibodies anti-p-MYBL2 (1:100 dilution; Abcam, ab76009) and anti-cleaved caspase 3 (1:100 dilution, Cell Signaling, #9661). Afterwards slides were incubated with a secondary anti-rabbit IgG antibody (MP-7401, ImmPress Reagent Kit, Peroxidase-conjugated) followed by subsequent target detection using DAB+chromogen (Agilent Technologies). Slides were counterstained with hematoxylin Gill's Formula (H-3401; Vector).
Evaluation of immunoreactivity and quantification of mitoses. Evaluation of p-MYBL2 immunostaining was carried out semi-quantitatively by a blinded observer in analogy to the Immune Reactive Score (IRS), which is used routinely by pathologists for quantification of hormone receptor expression in mammary carcinoma, ranging from 0 to 12 as described 58 . The intensity of p-MYBL2 immunoreactivity (score 0 = none, score 1 = low, score 2 = intermediate, and score 3 = strong) and the percentage of cells stained with each intensity (score 0 = 0%, score 1 = 0-9%, score 2 = 10-50%, score 3 = 51-80%, and score 4 = 81-100%) was determined per high-power field (×40). The product of the predominant intensity score and its percentage score defined the final IRS. For cleaved caspase 3 immunostaining, automated quantification of the percentage of positive highpower field area was performed using Fiji (ImageJ) 40,41 . Mitoses were quantified in HE-stained slides by a blinded observer per high-power field. Final scores/quantifications were determined by examination of 4-16 high-power fields of at least one section for each sample.
Statistical analysis and software. Statistical data analysis was performed using GraphPad PRISM 5 (GraphPad Software Inc., CA, USA) on the raw data. If not otherwise specified in the figure legends comparison of two groups in functional in vitro experiments was carried out using a two-tailed Mann-Whitney test; ***P < 0.001, **P < 0.01, *P < 0.05. Comparison of three groups with data in ordinal scale was performed using Kruskal-Wallis test. If not otherwise specified in the figure legends, data are presented as dot plots with horizontal bars representing means, and whiskers representing the standard error of the mean (SEM). Sample size for all in vitro experiments was chosen empirically. In Kaplan-Meier survival analyses, curves were calculated from all individual survival times of patients or mice, respectively. Curves were compared by Mantel-Haenszel test to detect significant differences between the groups. For in vivo experiments, sample size was predetermined using power calculations with β = 0.8 and α = 0.05 based on preliminary data and in compliance with the 3R system (replacement, reduction, refinement).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Data availability