Reanalysis of neuroblastoma expression profiling data using improved methodology and extended follow-up increases validity of outcome prediction

doi:10.1016/j.canlet.2009.02.052

Cancer Letters

Volume 282, Issue 1, 8 September 2009, Pages 55-62

https://doi.org/10.1016/j.canlet.2009.02.052 Get rights and content

Abstract

Neuroblastoma is the most common extracranial childhood tumor, comprising 15% of all childhood cancer deaths. In an initial study, we used Affymetrix oligonucleotide microarrays to analyse gene expression in 68 primary neuroblastomas and compared different data mining approaches for prediction of early relapse. Here, we performed re-analyses of the data including prolonged follow-up and applied support vector machine (SVM) algorithms and outer cross-validation strategies to improve reliability of expression profiling based predictors. Accuracy of outcome prediction was significantly improved by the use of innovative SVM algorithms on the updated data. In addition, CASPAR, a hierarchical Bayesian approach, was used to predict survival times for the individual patient based on expression profiling data. CASPAR reliably predicted event-free survival, given a cut-off time of three years. Differential expression of genes used by CASPAR to predict patient outcome was validated in an independent cohort of 117 neuroblastomas. In conclusion, we show here for the first time that reanalysis of microarray data using improved methodology, state-of-the-art performance tests and updated follow-up data improves prognosis prediction, and may further improve risk stratification of individual patients.

Section snippets

Background

Neuroblastoma is the most common extracranial tumor of childhood and accounts for approximately 15% of all childhood cancer deaths [1], [2], [3]. Although numerous prognostic factors have been identified, risk evaluation of individual patients remains difficult due to the clinical heterogeneity of neuroblastoma. This malignancy is unique in its wide spectrum of clinical behaviour, which ranges from spontaneous regression and differentiation to highly aggressive disease that often results in

Affymetrix microarray data

Microarray data was obtained from our initial neuroblastoma expression profiling study. Data were normalized using the MAS5.0 algorithm implemented in R statistical language as previously described [13].

Patient cohort

Updated follow-up information was obtained from the study center of the German neuroblastoma study group. Of 68 patients in our initial study [14], one was lost to follow-up and was therefore excluded from further analysis.

SVM

For SVM analysis, RapidMiner 4.1 (Rapid-I, Dortmund, Germany) was used

Sample annotation and follow-up

We first obtained updates of annotations and follow-up data of our initial patient cohort [14]. Updated follow-up data was available from the German Neuroblastoma Study Data Centre for 67 of 68 patients. One patient was lost to follow-up, and was therefore excluded from further analysis. The median follow-up time for patients in this study was 1.717 days (IQR: 640 days), compared to 1.108 days (IQR: 810 days) at the time of our initial analysis. Since our initial analysis, six additional

Discussion

On the route to clinical application of high-throughput expression profiling, different strategies have been applied to increase validity and practicability. For example, we applied high-throughput real-time PCR to verify our predictive signature using a technique feasible for routine clinical use [9]. Nevertheless, a prerequisite for any array-based diagnostic tool is the validity of predictive patterns obtained from microarray based expression profiling studies. This can be achieved by

Conflicts of interest statement

None declared.

Acknowledgements

We thank Kathy Astrahantseff for the critical reading of the manuscript, Shahab Asgharzadeh for helpful discussions and Barbara Hero and the German Neuroblastoma Study Group for providing clinical data. A.E. and A.S. were supported by Grants from the National Genome Research Network (NGFN) and the EU (Framework 6, EET Pipeline, Grant no. 037260). L.K. received funding from the BMBF through ViroQuant (Grant nr. 0313923).

References (27)

J.M. Maris et al.
Neuroblastoma
Lancet
(2007)
J.R. Park et al.
Neuroblastoma: biology, prognosis, and treatment
Pediatr. Clin. North. Am.
(2008)
M. Schwab et al.
Neuroblastoma: biology and molecular and chromosomal pathology
Lancet Oncol.
(2003)
B. Berwanger et al.
Loss of a FYN-regulated differentiation and growth arrest pathway in advanced stage neuroblastoma
Cancer Cell
(2002)
M. Ohira et al.
Expression profiling using a tumor-specific cDNA microarray predicts the prognosis of intermediate risk neuroblastomas
Cancer Cell
(2005)
G.M. Brodeur
Neuroblastoma: biological insights into a clinical enigma
Nat. Rev. Cancer
(2003)
J. Vandesompele et al.
Unequivocal delineation of clinicogenetic subgroups and development of a new model for improved outcome prediction in neuroblastoma
J. Clin. Oncol.
(2005)
A. Oberthuer et al.
Customized oligonucleotide microarray gene expression-based classification of neuroblastoma patients outperforms current clinical risk stratification
J. Clin. Oncol.
(2006)
A. Schramm et al.
Translating expression profiling into a clinically feasible test to predict neuroblastoma outcome
Clin. Cancer Res.
(2007)
J.H. Schulte et al.
MYCN regulates oncogenic MicroRNAs in neuroblastoma
Int. J. Cancer
(2008)

Q.R. Chen et al.

An integrated cross-platform prognosis study on neuroblastoma patients

Genomics

(2008)

L. Kaderali et al.

CASPAR: a hierarchical bayesian approach to predict survival times in cancer from gene expression data

Bioinformatics

(2006)

J.H. Schulte et al.

Microarray analysis reveals differential gene expression patterns and regulation of single target genes contributing to the opposing phenotype of TrkA- and TrkB-expressing neuroblastomas

Oncogene

(2005)

Cited by (10)

Transcript signatures that predict outcome and identify targetable pathways in MYCN-amplified neuroblastoma
2016, Molecular Oncology
Citation Excerpt :
Recently, high-throughput profiling techniques have enabled the identification of transcriptomic-based predictors, commonly known as gene signatures. Signatures with reported predictive power in NB have been identified using a variety of techniques including manually curated gene sets that represent specific biological processes (Asgharzadeh et al., 2012), experimental perturbations to modulate gene expression (e.g., MYCN knockdown) (Barbieri et al., 2012; Fardin et al. 2009, 2010; Liang et al., 2014; Nevo et al., 2009; Valentijn et al., 2012; van Nes et al., 2013), subtype analysis (Abel et al., 2011), or comparison of gene expression profiles derived from patients previously classified as high and low-risk NB patients (Vermeulen et al., 2009; Oberthuer et al., 2006; De Preter et al., 2010; Ohira et al., 2005; Asgharzadeh et al., 2006; Fischer et al., 2006; Garcia et al., 2012; Schramm et al., 2009, 2005, 2007; von Stedingk et al., 2013; Wei et al., 2004). Reported signatures vary from 3 to 158 genes, comprise relatively few overlapping genes, and are associated with a diverse array of biological processes including differentiation, hypoxia, therapy resistance, migration/invasion, inflammation, and DNA repair.
In the pediatric cancer neuroblastoma (NB), patients are stratified into low, intermediate or high-risk subsets based in part on MYCN amplification status. While MYCN amplification in general predicts unfavorable outcome, no clinical or genomic factors have been identified that predict outcome within these cohorts of high-risk patients. In particular, it is currently not possible at diagnosis to determine which high-risk neuroblastoma patients will ultimately fail upfront therapy.
We analyzed the prognostic potential of most published gene expression signatures for NB and developed a new prognostic signature to predict outcome for patients with MYCN amplification. Network and pathway analyses identified candidate therapeutic targets for this MYCN-amplified patient subset with poor outcome.
Most signatures have a high capacity to predict outcome of unselected NB patients. However, the majority of published signatures, as well as most randomly generated signatures, are highly confounded by MYCN amplification, and fail to predict outcome in subpopulations of high-risk patients with MYCN-amplified NB. We identify a MYCN module signature that predicts patient outcome for those with MYCN-amplified tumors, that also predicts potential tractable therapeutic signaling pathways and targets including the DNA repair enzyme Poly [ADP-ribose] polymerase 1 (PARP1).
Many prognostic signatures for NB are confounded by MYCN amplification and fail to predict outcome for the subset of high-risk patients with MYCN amplification. We report a MYCN module signature that is associated with distinct patient outcomes, and predicts candidate therapeutic targets in DNA repair pathways, including PARP1 in MYCN-amplified NB.
Progress in treatment and risk stratification of neuroblastoma: Impact on future clinical and basic research
2011, Seminars in Cancer Biology
Citation Excerpt :
The treatment stratification concept in LINES is based on the results of recent trials, which have suggested that the risk of relapse in patients who have low-risk tumors but no MYCN amplification may well be defined by the presence versus absence of any structural genetic abnormalities [47,74,75]. Several groups have made efforts to propose mRNA expression-based classifiers for treatment stratification, which are in part combined with CGH or microRNA [76–81]. The lack of overlap in the proposed gene lists of these classifiers may be partly explained by the use of different patient cohorts, technologies and/or bioinformatics approaches in the cited studies.
Close international collaboration between pediatric oncologists has led to marked improvements in the cure of patients, seen as a long-term overall survival rate of about 80%. Despite this progress, neuroblastoma remains a challenging disease for both clinicians and researchers. Major clinical problems include lack of acceptable cure rates in high-risk neuroblastoma and potential overtreatment of subsets of patients at low and intermediate risk of the disease. Many years of intensive international cooperation have recently led to a promising joint effort to further improve risk classification for treatment stratification, the new International Neuroblastoma Risk Group Classification System. This approach will facilitate comparison of the results of clinical trials performed by different international collaborative groups. This, in turn, should accelerate refinement of risk stratification and thereby aid selection of appropriate therapies for individual patients. To be able to identify new therapeutic modalities, it will be necessary to elucidate the pathogenesis of the different subtypes of neuroblastoma. Basic and translational research have provided new tools for molecular characterization of blood and tumor samples including high-throughput technologies for analysis of DNA, mRNAs, microRNAs and other non-coding RNAs, as well as proteins and epigenetic markers. Most of these studies are array-based in design. In neuroblastoma research they aim to refine risk group stratification through incorporation of molecular tumor fingerprints and also to enable personalized treatment modalities by describing the underlying pathogenesis and aberrant signaling pathways in individual tumors. To make optimal use of these new technologies for the benefit of the patient, it is crucial to have a systematic and detailed documentation of both clinical and molecular data from diagnosis through treatment to follow-up. Close collaboration between clinicians and basic scientists will provide access to combined clinical and molecular data sets and will create more efficient steps in response to the remaining treatment challenges. This review describes the current efforts and trends in neuroblastoma research from a clinical perspective in order to highlight the urgent clinical problems we must address together with basic researchers.
Identification of therapy-sensitive and therapy-resistant neuroblastoma subtypes in stages III, IVs and IV
2011, Cancer Letters
Citation Excerpt :
There are still many questions to be answered regarding the pathogenesis of heterogeneous NBs. Recent advances in gene technology may provide much more information and the means to address these questions, such as DNA or RNA micro arrays and genomic or molecular signatures [40,41]. However, these methods are still expensive and require special circumstances to handle properly.
The aim was to present a new model of risk stratification with high predictive sensitivity for non-localized neuroblastomas (NBs). “MYCN amplification”, “unfavorable histology of the International Neuroblastoma Pathology Classification (INPC) system” and “low Ha-ras/trk A expression” could be defined as an independent predictor for high-risk NBs. A risk stratification flow chart was applied to 103 advanced NBs in which all three factors were examined and 69 were grouped as high-risk NBs of which 38 patients died. The predictive sensitivity for poor patient outcome was 86%, which included 38 of the 44 total deaths in this analysis. Using the number of the three independent risk factors in each tumor, the 69 high-risk NBs were classified into three subgroups. NBs with the three risk factors (triple risk) represented the most aggressive character and survival of the affected patients was only 10% (“therapy-resistant NBs”). Survivals of the patients with NBs possessed the two (double) risk factors or the one (single) risk factor were 29% and 66%, respectively. This stratification also elucidated a subgroup in which patient survival was 90% (“therapy-sensitive”). There were 21 NBs with “high Ha-ras/trk A expression”, “favorable INPC histology” and “unamplified MYCN” (no risk NBs). Among the four subgroups without a risk factor, with a single risk factor, with double risk and with triple risk, Kaplan–Meier analysis showed a significant difference in NB patient outcome (p < 0.0001). Risk stratification might improve the therapeutic efficacy for the high-risk NBs and might decrease therapy-related sequelae in the lower risk NBs.
The Application of Bayesian Methods in Cancer Prognosis and Prediction
2022, Cancer Genomics and Proteomics
Use of attribute driven incremental discretization and logic learning machine to build a prognostic classifier for neuroblastoma patients
2014, BMC Bioinformatics
Korarchaeota diversity, biogeography, and abundance in Yellowstone and Great Basin hot springs and ecological niche modeling based on machine learning
2012, PLoS ONE

View all citing articles on Scopus

View full text

Reanalysis of neuroblastoma expression profiling data using improved methodology and extended follow-up increases validity of outcome prediction

Abstract

Section snippets

Background

Affymetrix microarray data

Patient cohort

SVM

Sample annotation and follow-up

Discussion

Conflicts of interest statement

Acknowledgements

Lancet

Pediatr. Clin. North. Am.

Lancet Oncol.

Cancer Cell

Cancer Cell

Neuroblastoma: biological insights into a clinical enigma

Nat. Rev. Cancer

Unequivocal delineation of clinicogenetic subgroups and development of a new model for improved outcome prediction in neuroblastoma

J. Clin. Oncol.

Customized oligonucleotide microarray gene expression-based classification of neuroblastoma patients outperforms current clinical risk stratification

J. Clin. Oncol.

Translating expression profiling into a clinically feasible test to predict neuroblastoma outcome

Clin. Cancer Res.

MYCN regulates oncogenic MicroRNAs in neuroblastoma