Gene expression-based predicition classifier in melanoma

In 2005, around 66,000 Americans have developed cutaneous melanoma, and around 10,000 have died from the disease. The NCI-SEER database documents a 619% increase in the annual incidence of melanoma and a 165% increase in the annual mortality from 1950-2000. Proclivity for metastasis and therapeutic resistance are hallmarks of melanoma; after metastatic spread to vital organs, the average life span of patients is less than a year. Once diagnosed, the criteria for prognosis of melanoma patients of clinical stages I and II are insufficient, because they are based merely on morphological features of the primary tumor. These features do not allow providing patients with meaningful and accurate individual prognostic information at the time of diagnosis. Around 50% of clinical stage II melanoma patients will develop metastatic disease later on and are being included into adjuvant treatment trials as state-of-the-art procedure. As a consequence, the lack of valuable individualized prognostic information results in (i) sustained overtreatment of several thousands of melanoma patients with all its negative socioeconomic aspects and (ii) significant interference with drug development in adjuvant treatment trials. So far, no molecular parameters such as genes or signal transduction pathways have been described to prediction of disease outcome in melanoma patients. A central, essential feature of the proposal is to develop individualized prediction of disease outcome from a merely morphology-based into a state-of-the-art molecular approach. Over the last 14 years, we have established a unique collection of cryo-asserved tissue samples representing all stages of melanoma development with extensive clinical annotation (i.a., up to 14 years clinical follow-up). This collection has been subjected to gene expression profiling. By use of "weighted voting" algorithm and pathway and gene set enrichment analysis we have identified in silico genes and pathways being differentially expressed and enriched, respectively, between melanomas with good and bad disease outcome. Here, we propose to use state-of-the-art technologies to (i) generate a candidate gene expression-based classifier best predicting disease outcome of melanoma patients at time of diagnosis; (ii) validate a candidate prediction classifier in an independent (validation) sample set; and (iii) to develop a robust technology platform for clinical translation. In addition, the project may help to describe cellular events being associated with disease progression and metastasis in melanoma and thus, to identify candidate target genes or proteins for drug discovery. Our unique collection of gene expression data with extensive clinical annotation provides the opportunity to break new ground not only to disease subclassification and provision of accurate individualized prognostic information in melanoma patients by molecular means, but also to the genetic identification of cellular events presumably being associated with tumor progression and metastasis on a genetic basis.

In 2005, around 66,000 Americans have developed cutaneous melanoma, and around 10,000 have died from the disease. The NCI-SEER database documents a 619% increase in the annual incidence of melanoma and a 165% increase in the annual mortality from 1950-2000. Proclivity for metastasis and therapeutic resistance are hallmarks of melanoma; after metastatic spread to vital organs, the average life span of patients is less than a year. Once diagnosed, the criteria for prognosis of melanoma patients of clinical stages I and II are insufficient, because they are based merely on morphological features of the primary tumor. These features do not allow providing patients with meaningful and accurate individual prognostic information at the time of diagnosis. Around 50% of clinical stage II melanoma patients will develop metastatic disease later on and are being included into adjuvant treatment trials as state-of-the-art procedure. As a consequence, the lack of valuable individualized prognostic information results in (i) sustained overtreatment of several thousands of melanoma patients with all its negative socioeconomic aspects and (ii) significant interference with drug development in adjuvant treatment trials. So far, no molecular parameters such as genes or signal transduction pathways have been described to prediction of disease outcome in melanoma patients. A central, essential feature of the project was to develop individualized prediction of disease outcome from a merely morphology-based into a state-of-the-art molecular approach. Over the last 14 years, we have established a unique collection of cryo-asserved tissue samples representing all stages of melanoma development with extensive clinical annotation (i.a., up to 14 years clinical follow-up). This collection has been subjected to gene expression profiling and biomathematical analysis. In cooperation with the Center for Medical Statistics, Informatics and Intelligent Systems at the Medical University of Vienna we performed GSEA and weighted voting analyses. This led to: 1. The identification of 25 different gene classifiers, that predicted the clinical course of patients with 100% accuracy in our data set. 2. By integration of our data set with a published data set (Winnepenninckx V et al., J Clin Oncol 98:472; 2006) we could identify further 3 different "integrated" gene classifiers, that predicted the clinical course of patients with > 95% accuracy in our data set. These gene prediction classifiers provide a promising tool for further validation in / or integration with additional data sets from different patient cohorts and could be of fundamental importance for future studies aimed at accurate individualized prediction of the clinical course of melanoma patients.