Introduction

Ewing sarcoma is the second most common primary bone sarcoma. It is an orphan state disease with approximately 900 new diagnoses a year in Europe [1]. It is also called the Ewing Sarcoma Family of Tumours (ESFT) and includes Ewing sarcoma of bone, extra-osseous Ewing sarcoma, Primitive Neuroectodermal (PNET) and Askin's tumours. Ewing sarcoma is diagnostically defined by a Ewing sarcoma EWS (chromosome 22) translocation resulting in fusion with an ETS transcription factor, the commonest abnormality (85%) being EWS-FLI1 (chromosome 11). Ewing sarcoma is a disease affecting children and young adults with a peak incidence at age fifteen. With current treatment options the 5 year survival for non-metastatic disease is 60-70%. However, survival for the 25% of patients that present with metastatic disease is approximately 20% [2], and for those who develop relapsed and/or refractory disease, the survival is no more than 10%.

Current patients are subdivided by disease stage, namely non-metastatic, metastatic and recurrence, and patients in each group are treated the same. But apparently this subdivision is not always related to clinical outcome, because of the patients who present with non-metastatic disease, approximately 30% die within 5 years. This group may be currently undertreated while the 70% who survive may be over-treated. It may therefore be important to separate the high risk patients from the low risk patients and to be able to detect chemotherapy resistance and metastases early.

A way of predicting patients' outcome is by using prognostic factors. The most commonly used are clinical features, eg age, gender, metastases. Biomarker is a synonym for biological markers and is defined as "a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processor or pharmacologic responses to a therapeutic intervention" [3]. Biomarkers are currently already being used for screening, diagnosis, prognosis and monitoring of cancer patients. In 2005 the Reporting recommendations for tumour MARKer prognostic studies (REMARK) guidelines were published [4]. The goal of these guidelines is to make the results from clinical prognostic studies transparent and to improve the level of comparison that is possible between studies.

We report an overview of the research which has been done to identify reliable biomarkers for Ewing sarcoma in the past 15 years, where we detail the kind of markers that have been tested, the number of patients involved and the p-value showing the significance of the marker. The results highlight some interesting biomarkers, but they have yet to be validated.

Materials and methods

Search strategy

We report data available in the public domain only. Papers were identified from PubMed searches and from references in the found articles. The search algorithm was: (Ewing sarcoma) AND (prognostic factors) OR (biomarker). Only papers published between 1995 and 2010 are included. The latest search was done in June 2010. Whenever multiple reports from the same study were published, we used only the report with the latest published date to avoid any duplication of information. Papers were eligible if they: (1) described (or cited a paper that described) a Ewing sarcoma study of prognostic factors or biomarkers; (2) were published in English; and (3) came from industrialized countries. All types of evaluation were accepted (full papers, conference abstracts, reports) as long as results (including data) were presented.

Data extraction

Data extraction was conducted independently by the first author (A.M. v. M.). We used a systematic method for the search normally used for meta-analysis [5]. Differences in data extraction were resolved by consensus with a second author (A.B.H). From each eligible trial we recorded authors' names, journal and year of publication and the results from the study.

Results and Discussion

Eligible trials

A flow-chart indicating the identification of reports for inclusion in the analysis is reported for Ewing sarcoma (Figure 1). During the search many reports had to be excluded mainly because no prognostic markers were reported in the article. When we searched the reports using full text, we had to exclude some papers because no Ewing sarcoma patients were included in these reports. We identified 86 articles which were eligible for our search criteria. In these papers a total of 11, 625 patients were reported.

Figure 1
figure 1

Flowchart for the identification of eligible reports.

Full size image

In this report we looked at the published data on the use of biomarkers for the last 15 years. Biomarkers were grouped into phenotypic markers and biological markers. Markers were taken as statistically significant if p < 0.05. For phenotypic markers we reported the outcome for gender, tumour size, presence of metastases and histological response after treatment (Tables 1, 2, 3 & 4). We showed the p-value reported in the eligible articles and the distribution of p correlated to the number of patients (Figures 2). There doesn't seem to be a relationship between the number of patients and the p-value. For example, the distribution of histological response shows that the studies with small patient numbers have the same statistical significance as these with large patient numbers. Throughout this report, the assumption is that the biomarker has a linear relationship to outcome. We know that for many biomarkers, this is not the case. For example, data transformation using either bicubic splines or fractional polynomials is often required to correlate continuous relationships between biomarkers and outcome, as opposed to predefined cutpoints [6]. We can only have limited extrapolation of the reported data to outcome as in most instances these questions have not been addressed.

Table 1 Outcome for phenotypic marker: gender
Full size table
Table 2 Outcome for phenotypic marker: tumour size
Full size table
Table 3 Outcome for phenotypic marker: metastases
Full size table
Table 4 Outcome for phenotypic marker: histological response
Full size table
Figure 2
figure 2

Distribution of p related to patient number for the phenotypic markers: gender, tumour size, metastases and histological response. The red line shows the cut-off point of p = 0.05.

Full size image

Primary outcome

The investigated biomarkers are subdivided in two groups, phenotypic markers and biological markers. For the phenotypic markers gender, tumour size, metastases and histological response are reported in Tables 1, 2, 3 and 4. For all these phenotypic markers we compared the patient number and p-value, in which p < 0.05 was taken as statistically significant. However we weren't able to retrieve the p-value in all articles, sometimes it was only mentioned as being significant or non-significant. For each phenotypic marker we looked at the differences in overall survival between: for gender, men vs women; for tumour size, < 8-10 cm vs > 8- 10 cm; for metastasis presence at initial presentation vs absence and for histological response, > 90% necrosis vs < 90% necrosis. Distributions of p related to patient numbers in these four phenotypic markers are shown in Figures 2. For these four phenotypic markers we show that there is no correlation between the number of patients and the statistical significance of the outcome. More phenotypic markers were reported: fusion type, ethnicity, performance status and margins. However because of the low number of studies which reported these outcomes these results are not shown in detail. In 26 articles the impact of tumour site on the overall survival is shown, but because sites are compared in different ways it is difficult to summarize these findings.

Currently clinical stage is being used to determine whether a patient has a high or low risk for developing metastases or recurrence. However, it seems that clinical stage is not always related to outcome, because of the patients who present with non-metastatic disease, only 70% of them survive for 5-years. Therefore, what is the difference between the 70% of the patients who survive and the 30% who don't? Can one somehow foretell chemotherapy resistance and detect metastases early? One way to predict the outcome of patients apart from clinical stage is to use biomarkers. These are objective measurements which reflect biological processes. The biomarkers currently being used are tumour size and the presence of metastases. Biological markers are not being used, even though they may provide a way to predict a patient's outcome more accurately than phenotypic markers. From the results for phenotypic markers we can see that gender is probably not significant important for patient outcome. In 15 articles we found 11 reports that gender is non-significant. Tumour size > 8 cm seems to be important, with 15 out of 22 articles finding it to be a predictor and significantly related to negative outcome. The presence of metastasis is a strong predictor of negative outcome with 24 articles reporting it as significantly relevant compared to only 3 reporting it as non-significant. For histological response, 12 out of 16 articles found that necrosis > 90% after treatment is a significant predictor for positive outcome.

For some phenotypic markers it is unclear how the cut-off point between predictor of positive or negative outcomes is determined. For tumour size the cut-off point for negative outcome is > 8 cm, but it is undefined how this is selected. It seems more logical that tumour size is a continuous variable with an increasingly negative outcome with increasing size. The same can probably be said for age and surgical margins.

Biological markers are more difficult to compare, because for most of these markers only one or two reports are published. We grouped the biological markers according to their function and we ended up with 5 groups, namely cell cycle, karyotype, immunological, blood products and the remaining biological markers which couldn't be classified in one of the other groups. The results from the biological markers are shown in Tables 5, 6, 7, 8 and 9. The correlation between patient number and statistical significance of the outcome for the five groups is shown in Figures 3. We show that there is no correlation between the patient number and the statistical significance of the outcome. It appears that ki67, an S-phase cell cycle biomarker, may be a biomarker of cell activity in the tumour that significantly correlates with outcome. The mechanism for the activation of cell cycle appears unclear, but is presumably driven by other factors other that EWS-FLI1 translocation. Loss of function of cell cycle dependent kinases (p16, p14, p21) and other regulators of the cell cycle through the p53 pathway (MDM2, p53), also appear deregulate in a proportion of tumours and potentially are useful prognostic markers. Importantly, activity of telomerase appears significantly correlated with outcome as occurs in many other tumours. There appears much interest in secondary copy number changes and mutations in Ewing sarcoma, and in particular, chromosome 1 (Table 6). For example, recent evidence points to gain of 1q and alteration in abundance of a gene product called CDt2 involved in ubiquitination [7]. It is however difficult to objectively say anything about the other reported markers because they may influence each other. This appears most clear for tumour size and metastases, where bigger tumours may correlate with a higher chance of having metastases. For biological markers it is probably the same issue, but less clear because we don't really know their true experimental influence on tumour genesis. For example, LDH levels are probably a reflection of cell turnover in larger tumours, and may be an indirect measure of bulk of disease (comparing Table 2 versus Table 9). It is also more difficult to say anything about biological markers because they haven't been tested as extensively as phenotypic markers, and certainly they have not often been validated independently. Results for most of these markers are only reported in 1 or 2 articles with sometimes small numbers of patients and no statistical validation. To improve this situation it would important to capture high quality clinical material and clinical outcome to develop a bio-bank. We may be able to test the most promising biomarkers from previously run studies and so define their significance. Either a multivariate analysis or data mining analysis should be done to evaluate the way biomarkers affect each other. The easiest way to achieve this objective is by collecting material and outcome data from large phase III trials. It is also important to standardize the way material is collected and how the biomarkers are compared. For example, the phenotypic marker tumour site is the most often tested marker with results published in 26 articles (data not shown). However it is not possible to say anything about these results since different tumour sites are compared in the reports. This is also true for the marker age in which different age groups are compared with each other, for example some articles compare patients < 18 years vs > 18 years, others < 30 years vs > 30 years (data not shown).

Table 5 Outcome for biological markers: cell cycle
Full size table
Table 6 Outcome for biological markers: karyotype
Full size table
Table 7 Outcome for biological markers: immunological
Full size table
Table 8 Outcome for biological markers: blood products
Full size table
Table 9 Outcome for biological markers: remaining
Full size table
Figure 3
figure 3

Distribution of p related to patient number for the biological markers related to cell cycle, karyotype, immunological, blood products and remaining markers. The red line shows the cut-off point of p = 0.05. Note, there is no line for immunological phenotypic markers because for all the results p < 0.05.

Full size image

For markers of tumour growth, angiogenesis if often quantified, but so far biomarker analysis has been predominantly limited to measurement of VEGF pathway (Table 8). The immunological biological markers interleukins and tumour necrosis factors seem very promising (Table 7). However these have all been tested in one institute, with very small patient numbers and the data doesn't seem to be validated. Most of the biological markers mentioned in the blood products group (Table 8) are probably surrogates for tumour size and they should be validated in either a multivariate analysis or machine learning to see if they can be used as an objective biological marker.

At the present time it is no possible to make a definite list of biological biomarkers able to predict patient outcome, mainly because these markers also have to be stratified with respect to the major staging phenotypic features, e.g. presence of metastasis and degree of histological response. It is also unclear what quality control measure were used in the limited patient cohorts. Our recommendation would be continue divide patients according to their disease stage and also to use the phenotypic biomarkers metastasis, tumour size and histological response. For biological biomarkers we would like to validate previous work done on the markers for 9p21 locus and the involved genes and proteins, heat shock proteins, telomerase related markers, interleukins, tumour necrosis factors, VEGF pathway, lymphocyte count, MTA1, STEAP1, CCND1, MDM-2, Ki-67, p53, p27 and cadherin-11. At this time, neither phenotypic (clinical) or biological biomarkers are utilised in stratification of patients in clinical trials.