Prognostic impact of gene expression-based classification for neuroblastoma.

PURPOSE: To evaluate the impact of a predefined gene expression-based classifier for clinical risk estimation and cytotoxic treatment decision making in neuroblastoma patients. PATIENTS AND METHODS: Gene expression profiles of 440 internationally collected neuroblastoma specimens were investigated by microarray analysis, 125 of which were examined prospectively. Patients were classified as either favorable or unfavorable by a 144-gene prediction analysis for microarrays (PAM) classifier established previously on a separate set of 77 patients. PAM classification results were compared with those of current prognostic markers and risk estimation strategies. RESULTS: The PAM classifier reliably distinguished patients with contrasting clinical courses (favorable [n = 249] and unfavorable [n = 191]; 5-year event free survival [EFS] 0.84 +/- 0.03 v 0.38 +/- 0.04; 5-year overall survival [OS] 0.98 +/- 0.01 v 0.56 +/- 0.05, respectively; both P < .001). Moreover, patients with divergent outcome were robustly discriminated in both German and international cohorts and in prospectively analyzed samples (P

Cross-study analysis of gene expression data for intermediate neuroblastoma identifies two biological subtypes.

BACKGROUND: Neuroblastoma patients show heterogeneous clinical courses ranging from life-threatening progression to spontaneous regression. Recently, gene expression profiles of neuroblastoma tumours were associated with clinically different phenotypes. However, such data is still rare for important patient subgroups, such as patients with MYCN non-amplified advanced stage disease. Prediction of the individual course of disease and optimal therapy selection in this cohort is challenging. Additional research effort is needed to describe the patterns of gene expression in this cohort and to identify reliable prognostic markers for this subset of patients. METHODS: We combined gene expression data from two studies in a meta-analysis in order to investigate differences in gene expression of advanced stage (3 or 4) tumours without MYCN amplification that show contrasting outcomes (alive or dead) at five years after initial diagnosis. In addition, a predictive model for outcome was generated. Gene expression profiles from 66 patients were included from two studies using different microarray platforms. RESULTS: In the combined data set, 72 genes were identified as differentially expressed by meta-analysis at a false discovery rate (FDR) of 8.33%. Meta-analysis detected 34 differentially expressed genes that were not found as significant in either single study. Outcome prediction based on data of both studies resulted in a predictive accuracy of 77%. Moreover, the genes that were differentially expressed in subgroups of advanced stage patients without MYCN amplification accurately separated MYCN amplified tumours from low stage tumours without MYCN amplification. CONCLUSION: Our findings support the hypothesis that neuroblastoma consists of two biologically distinct subgroups that differ by characteristic gene expression patterns, which are associated with divergent clinical outcome.

Classification of neuroblastoma patients by published gene-expression markers reveals a low sensitivity for unfavorable courses of MYCN non-amplified disease.

Currently, Pubmed lists 385 marker genes for neuroblastoma outcome. Using a customized neuroblastoma-microarray, we evaluated the prognostic impact of the gene-expression pattern of 349 of these candidates (90.6%) in 127 neuroblastoma patients with divergent outcome. By significance analysis of microarrays (SAM) and both uncorrected and Bonferroni-corrected ANOVA, 166/349 (47.5%), 218/349 (62.5%) and 128/349 (36.4%) candidates showed significant differential expression between patients with contrasting outcome. By Prediction Analysis for Microarrays (PAM), a 38-gene-classifier was derived from all markers, which classified patients outcome with an overall accuracy of 78.5%. However, patients with unfavorable outcome of MYCN non-amplified disease were largely misclassified (accuracy: 35%), suggesting that these courses are not identified by current marker genes.

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

PURPOSE: To develop a gene expression-based classifier for neuroblastoma patients that reliably predicts courses of the disease. PATIENTS AND METHODS: Two hundred fifty-one neuroblastoma specimens were analyzed using a customized oligonucleotide microarray comprising 10,163 probes for transcripts with differential expression in clinical subgroups of the disease. Subsequently, the prediction analysis for microarrays (PAM) was applied to a first set of patients with maximally divergent clinical courses (n = 77). The classification accuracy was estimated by a complete 10-times-repeated 10-fold cross validation, and a 144-gene predictor was constructed from this set. This classifier's predictive power was evaluated in an independent second set (n = 174) by comparing results of the gene expression-based classification with those of risk stratification systems of current trials from Germany, Japan, and the United States. RESULTS: The first set of patients was accurately predicted by PAM (cross-validated accuracy, 99%). Within the second set, the PAM classifier significantly separated cohorts with distinct courses (3-year event-free survival [EFS] 0.86 +/- 0.03 [favorable; n = 115] v 0.52 +/- 0.07 [unfavorable; n = 59] and 3-year overall survival 0.99 +/- 0.01 v 0.84 +/- 0.05; both P < .0001) and separated risk groups of current neuroblastoma trials into subgroups with divergent outcome (NB2004: low-risk 3-year EFS 0.86 +/- 0.04 v 0.25 +/- 0.15, P < .0001; intermediate-risk 1.00 v 0.57 +/- 0.19, P = .018; high-risk 0.81 +/- 0.10 v 0.56 +/- 0.08, P = .06). In a multivariate Cox regression model, the PAM predictor classified patients of the second set more accurately than risk stratification of current trials from Germany, Japan, and the United States (P < .001; hazard ratio, 4.756 [95% CI, 2.544 to 8.893]). CONCLUSION: Integration of gene expression-based class prediction of neuroblastoma patients may improve risk estimation of current neuroblastoma trials.

Differential expression of neuronal genes defines subtypes of disseminated neuroblastoma with favorable and unfavorable outcome.

PURPOSE: Identification of molecular characteristics of spontaneously regressing stage IVS and progressing stage IV neuroblastoma to improve discrimination of patients with metastatic disease following favorable and unfavorable clinical courses. EXPERIMENTAL DESIGN: Serial analysis of gene expression profiles were generated from five stage IVS and three stage IV neuroblastoma. Differential expression of candidate genes was evaluated by real-time quantitative reverse transcription-PCR in 76 pretreatment tumor samples (stage IVS n=27 and stage IV n=49). Gene expression-based outcome prediction was determined by Prediction Analysis for Microarrays using 38 tumors as a training set and 38 tumors as a test set. RESULTS: Comparison of serial analysis of gene expression profiles from stage IV and IVS neuroblastoma revealed approximately 500 differentially expressed transcripts. Genes related to neuronal differentiation were observed more frequently in stage IVS tumors as determined by associating transcripts to Gene Ontology annotations. Forty-one candidate genes were evaluated by quantitative reverse transcription-PCR and 18 were confirmed to be differentially expressed (P

Gene expression signature predicting pathologic complete response with gemcitabine, epirubicin, and docetaxel in primary breast cancer.

PURPOSE: Primary systemic therapy (PST) with gemcitabine (G), epirubicin (E), and docetaxel (Doc) has resulted in a pathologic complete response (pCR) in 26% of primary breast cancer patients. This study was aimed at the identification of a gene expression signature in diagnostic core biopsy tissue samples that predicts pCR. PATIENTS AND METHODS: Core biopsy samples from patients with operable primary breast cancer, T2-4N0-2M0, enrolled onto two phase I and II trials evaluating GEDoc (n = 48) and GE sequentially followed by Doc (GEsDoc; n = 52) as PST were snap frozen and subjected to RNA expression profiling. A signature predicting pCR was discovered in the training set (GEsDoc) applying a support vector machine algorithm, and performance of this classifier was validated on the independent test set (GEDoc) by receiver operator characteristics analysis. RESULTS: We identified a signature consisting of 512 genes, which was enriched in genes involved in transforming growth factor beta and RAS-mediated signaling pathways, that predicts pCR with a sensitivity of 78%, a specificity of 90%, and an overall accuracy of 88% (95% CI, 75% to 95%). Apart from our signature, only HER2 overexpression was an independent predictor of pCR in multivariate analysis. CONCLUSION: In conclusion, our gene expression signature allows prediction of pCR to PST containing G, E, and Doc with unprecedented high overall accuracy and robustness.

Cross-platform analysis of cancer microarray data improves gene expression based classification of phenotypes.

BACKGROUND: The extensive use of DNA microarray technology in the characterization of the cell transcriptome is leading to an ever increasing amount of microarray data from cancer studies. Although similar questions for the same type of cancer are addressed in these different studies, a comparative analysis of their results is hampered by the use of heterogeneous microarray platforms and analysis methods. RESULTS: In contrast to a meta-analysis approach where results of different studies are combined on an interpretative level, we investigate here how to directly integrate raw microarray data from different studies for the purpose of supervised classification analysis. We use median rank scores and quantile discretization to derive numerically comparable measures of gene expression from different platforms. These transformed data are then used for training of classifiers based on support vector machines. We apply this approach to six publicly available cancer microarray gene expression data sets, which consist of three pairs of studies, each examining the same type of cancer, i.e. breast cancer, prostate cancer or acute myeloid leukemia. For each pair, one study was performed by means of cDNA microarrays and the other by means of oligonucleotide microarrays. In each pair, high classification accuracies (> 85%) were achieved with training and testing on data instances randomly chosen from both data sets in a cross-validation analysis. To exemplify the potential of this cross-platform classification analysis, we use two leukemia microarray data sets to show that important genes with regard to the biology of leukemia are selected in an integrated analysis, which are missed in either single-set analysis. CONCLUSION: Cross-platform classification of multiple cancer microarray data sets yields discriminative gene expression signatures that are found and validated on a large number of microarray samples, generated by different laboratories and microarray technologies. Predictive models generated by this approach are better validated than those generated on a single data set, while showing high predictive power and improved generalization performance.

Prediction of clinical outcome and biological characterization of neuroblastoma by expression profiling.

Neuroblastoma is a common childhood tumor comprising cases with rapid disease progression as well as spontaneous regression. Although numerous prognostic factors have been identified, risk evaluation in individual patients remains difficult. To define a reliable prognostic predictor and gene signatures characteristic of biological subgroups, we performed mRNA expression profiling of 68 neuroblastomas of all stages. Expression data were analysed using support vector machines (SVM-rbf), prediction analysis of microarrays (PAM), k-nearest neighbors (k-NN) algorithms and multiple decision trees. SVM-rbf performed best of all methods, and predicted recurrence of neuroblastoma with an accuracy of 85% (sensitivity 77%, specificity 94%). PAM identified a classifier of 39 genes reliably predicting outcome with an accuracy of 80%. In comparison, conventional risk stratification based on stage, age and MYCN-status only reached a predictive accuracy of 64%. Kaplan-Meier analysis using the PAM classifier indicated a 5-year survival of 20 versus 78% for patients with unfavorably versus favorably predicted neuroblastomas, respectively (P = 0.0001). Significance analysis of microarrays (SAM) identified additional genes differentially expressed among subgroups. MYCN-amplification and high expression of NTRK1/TrkA demonstrated a strong association with specific gene expression patterns. Our data suggest that microarray-derived data in addition to traditional clinical factors will be useful for risk assessment and defining biological properties of neuroblastoma.

Automatic identification of subcellular phenotypes on human cell arrays.

Light microscopic analysis of cell morphology provides a high-content readout of cell function and protein localization. Cell arrays and microwell transfection assays on cultured cells have made cell phenotype analysis accessible to high-throughput experiments. Both the localization of each protein in the proteome and the effect of RNAi knock-down of individual genes on cell morphology can be assayed by manual inspection of microscopic images. However, the use of morphological readouts for functional genomics requires fast and automatic identification of complex cellular phenotypes. Here, we present a fully automated platform for high-throughput cell phenotype screening combining human live cell arrays, screening microscopy, and machine-learning-based classification methods. Efficiency of this platform is demonstrated by classification of eleven subcellular patterns marked by GFP-tagged proteins. Our classification method can be adapted to virtually any microscopic assay based on cell morphology, opening a wide range of applications including large-scale RNAi screening in human cells.