Systems biology of Ewing sarcoma: a network model of EWS-FLI1 effect on proliferation and apoptosis

Abstract

Ewing sarcoma is the second most frequent pediatric bone tumor. In most of the patients, a chromosomal translocation leads to the expression of the EWS-FLI1 chimeric transcription factor that is the major oncogene in this pathology. Relative genetic simplicity of Ewing sarcoma makes it particularly attractive for studying cancer in a systemic manner. Silencing EWS-FLI1 induces cell cycle alteration and ultimately leads to apoptosis, but the exact molecular mechanisms underlying this phenotype are unclear. In this study, a network linking EWS-FLI1 to cell cycle and apoptosis phenotypes was constructed through an original method of network reconstruction. Transcriptome time-series after EWS-FLI1 silencing were used to identify core modulated genes by an original scoring method based on fitting expression profile dynamics curves. Literature data mining was then used to connect these modulated genes into a network. The validity of a subpart of this network was assessed by siRNA/RT-QPCR experiments on four additional Ewing cell lines and confirmed most of the links. Based on the network and the transcriptome data, CUL1 was identified as a new potential target of EWS-FLI1. Altogether, using an original methodology of data integration, we provide the first version of EWS-FLI1 network model of cell cycle and apoptosis regulation.

Figure 1.

(A) Flow chart of the article. Gray rectangles are key steps of our analysis. Methods and concepts are described in rounded rectangles. (1) Transcriptome time-series data were obtained from shA673-1C and -2C clones after silencing or silencing and re-expressing EWS-FLI1. (2) An original method based on nonlinear curve fitting was used to perform the analysis of transcriptome time series. (3) EWS-FLI1-modulated genes were selected; this list was restricted to the genes affecting proliferation and apoptosis. (4) A network representation of EWS-FLI1 signaling was chosen; it consists of influences (positive or negative) between genes, proteins and complexes. (5) EWS-FLI1 signaling network model was reconstructed from the above selected genes connected by the influences known from literature. (6) The notion of necessary connection was introduced; it allows to refine a network model when, for instance, additional experimental data are provided. (7) Silencing experiments were performed on several EWS-FLI1-regulated genes; new necessary connections were identified and added to EWS-FLI1 signaling network. (B) Causal relations between data and the influence network.

Figure 2.

(A) RT-QPCR for a panel of EWS-FLI1-modulated genes along time series experiments in shA673-1C cells on DOX addition/removal (solid: inhibition, dashed grey: rescue) and in four Ewing cell lines (A673, EW7, EW24 and SKNMC) on transfection with nontargeting siRNA (siCT) or EWS-FLI1-targeting siRNA (siEF1) after 24, 48 or 72 h. Relative expression level (%) for each gene to the starting point shA673-1C condition or to siCT conditions are displayed on the y axis. Data are presented as mean values and the standard deviations. (B) Western blot for a panel of EWS-FLI1-modulated genes along a time series experiment in shA673-1C cells on DOX addition and in four Ewing cell lines (A673, EW7, EW24 and SKNMC) on transfection with nontargeting siRNA (siCT) or EWS-FLI1 targeting siRNA (siEF1) after 72 h. For PARP western blot, full length protein is indicated by the arrow and cleaved PARP by the arrowhead. Beta-actin was used as loading control.

Figure 2.

(A) RT-QPCR for a panel of EWS-FLI1-modulated genes along time series experiments in shA673-1C cells on DOX addition/removal (solid: inhibition, dashed grey: rescue) and in four Ewing cell lines (A673, EW7, EW24 and SKNMC) on transfection with nontargeting siRNA (siCT) or EWS-FLI1-targeting siRNA (siEF1) after 24, 48 or 72 h. Relative expression level (%) for each gene to the starting point shA673-1C condition or to siCT conditions are displayed on the y axis. Data are presented as mean values and the standard deviations. (B) Western blot for a panel of EWS-FLI1-modulated genes along a time series experiment in shA673-1C cells on DOX addition and in four Ewing cell lines (A673, EW7, EW24 and SKNMC) on transfection with nontargeting siRNA (siCT) or EWS-FLI1 targeting siRNA (siEF1) after 72 h. For PARP western blot, full length protein is indicated by the arrow and cleaved PARP by the arrowhead. Beta-actin was used as loading control.

Figure 3.

(A) Time series corresponding to the first principal modes of gene expression variation in EWS-FLI1 inhibition (solid line) and re-expression experiments (dashed line). (B) Comparison of two methods for selecting modulated genes, one based on switch like (SL) score, the other one based on fold change (FC). For both methods, top 4000 probesets for each clone (shA673-1C and -2C) were selected (ranked by their SL score or by FC between the first and the last time points). The Venn diagram compares these top scored probesets. The intersection of both methods is partial for two reasons: (i) the SL score can be large for a time series tightly following the assumed model of response, even if having a moderate variance, (ii) FC method is not considering intermediate time points. Both CUL1 and CFLAR exhibit temporal expression responses that have a good fit to the proposed switch-like response model. However, only some CFLAR probesets are characterized by significant fold change values. (C) Examples of curve fitting to the time series in microarray experiments. AQP1, E2F2 and CDKN1C expression profiles are shown. Blue curves represent microarray experimental values; red curves correspond to fitted functions. Switch-like scores (SL), pulse-like scores (PL) and transitions parameters (Tr) are listed under each plot. SL and PL scales are not comparable as the fitting procedures are different. It can be noticed that both scores for E2F2 are smaller than those for AQP1 for two reasons: the amplitude of expression variation is smaller for E2F2 and the transition happen at a time point closer to the limits of the time window. The scores for CDKN1C are clearly lower, because the expression level is less smooth. In that case, transition parameters cannot be identified, because the inflections points of the fitted curves are outside of the time window.

Figure 4.

(A) Annotated network of EWS-FLI1 effects on proliferation and apoptosis, derived from literature-based fact sheet. White nodes represent genes or proteins; gray nodes represent protein complexes. EWS-FLI1 (green square) and cell cycle phases/apoptosis (octagons) represent the starting point and the outcome phenotypes of the network. Green and red arrows symbolize respectively positive and negative influence. Nodes with green frame are induced by EWS-FLI1 according to time series expression profile and nodes with red frame are repressed. The network structure shows intensive crosstalk between the pathways used for its construction, up to the point that the individual pathways cannot be easily distinguished. (B) Refined network including new links inferred from experimental data (thick arrows) from transcriptome time series and siRNA/RT-QPCR.

Figure 5.

Illustration of necessary and non-necessary connections within given network models and data. (i) An observed influence from EWS-FLI1 to CUL1 is a necessary connection because no indirect explanation (path with intermediate nodes) can be identified within the network model. (ii) P300 represses IER3, but this can be explained through RELA, thus P300 -| IER3 is not necessary.

Figure 6.

(A) Transcriptional influences between EWS-FLI1, CFLAR, MYC, P300, E2F1, RELA, IER3 and FOXO1 nodes extracted from the literature-based influence network. (B) Interpretation of experiments (siRNA transfection and RT-QPCR) in shA673-1C cells. Thickness of arrows shows the strength of the influence (values given in Supplementary Table S10). Blue arrows are based on RT-QPCR time series. Plain arrows represent transcriptional influences that are necessary for explaining data. Dashed arrows are questionable influences that can be explained through intermediate node. The arrow EWS-FLI1 -| FOXO1 is not necessary, although a recent article has identified it as a direct connection (72). (C) The necessary connections shown in Figure 6B have been compared with a subpart of the influence network (Figure 6A). All connections of this subpart have been confirmed, although two of them display an opposite sign. (D) Example of influences that cannot be interpreted as a necessary connection, because of ambiguity in the choice. Indeed, either RELA → IER3 is necessary and RELA -| P300 is not, or RELA-|P300 is necessary and RELA → IER3 is not. In this case, we decided to consider both connections (RELA → IER3; RELA -| P300) as non-necessary. Within this choice, the set of necessary connections is interpreted as the minimal set of connections that explain the maximum amount of data, with no ambiguity.