Patient stratification using global gene expression signatures

Abstract

This thesis describes three different studies that have used gene expression profiling to distinguish different disease phenotypes.

The first study involves the transition of a previously published gene expression signature to a new technological platform and the subsequent validation of the classifier in its updated form. The study is performed using two large sets of patient samples from nearly all academic medical centers in the Netherlands. The classifier’s relevance lies in the fact that it can detect lymph node metastasis in head and neck cancer, an ability that could significantly improve the assignment of appropriate treatment. It does so, not in isolation, but in combination with techniques for patient assessment that are already part of clinical practice. A necessary next step will be to include the assay and the proposed treatment strategy in a prospective study, to assess whether the predicted clinical benefit is real and measurable.

A second study aimed to stratify patients with a primary tumor of the colon that had already metastasised to the liver, metastases that were subsequently removed and expression profiled. However, an effective classifier remained elusive even when different groups of patients and different risk definitions were used during signature discovery and validation. Although this study does not definitively prove that there is no such gene signature, it is possible that the differences between the patients are too small to detect using whole tumor gene expression. Notwithstanding the negative result, this study does provide an interesting insight into the possible limits of stratifying patient using gene expression profiling.

The last study involves the successful identification and validation of a gene signature that can identify women who experience recurrent implantation failure (RIF) during IVF treatment and for whom an explanatory factor is lacking. This study is a relatively rare example of classification in an area where the role of gene expression profiling has so far been minor, and where its specific use for classification purposes has been virtually non-existent.