Jane Fridlyand

Comparison of Discrimination Methods for the Classification of Tumors Using Gene Expression Data

A reliable and precise classification of tumors is essential for successful diagnosis and treatment of cancer. cDNA microarrays and high-density oligonucleotide chips are novel biotechnologies increasingly used in cancer research. By allowing the monitoring of expression levels in cells for thousands of genes simultaneously, microarray experiments may lead to a more complete understanding of the molecular variations among tumors and hence to a finer and more informative classification. The ability to successfully distinguish between tumor classes (already known or yet to be discovered) using gene expression data is an important aspect of this novel approach to cancer classification. This article compares the performance of different discrimination methods for the classification of tumors based on gene expression data. The methods include nearest-neighbor classifiers, linear discriminant analysis, and classification trees. Recent machine learning approaches, such as bagging and boosting, are also considered. The discrimination methods are applied to datasets from three recently published cancer gene expression studies.

Widespread potential for growth-factor-driven resistance to anticancer kinase inhibitors

Mutationally activated kinases define a clinically validated class of targets for cancer drug therapy. However, the efficacy of kinase inhibitors in patients whose tumours harbour such alleles is invariably limited by innate or acquired drug resistance. The identification of resistance mechanisms has revealed a recurrent theme—the engagement of survival signals redundant to those transduced by the targeted kinase. Cancer cells typically express multiple receptor tyrosine kinases (RTKs) that mediate signals that converge on common critical downstream cell-survival effectors—most notably, phosphatidylinositol-3-OH kinase (PI(3)K) and mitogen-activated protein kinase (MAPK). Consequently, an increase in RTK-ligand levels, through autocrine tumour-cell production, paracrine contribution from tumour stroma or systemic production, could confer resistance to inhibitors of an oncogenic kinase with a similar signalling output. Here, using a panel of kinase-‘addicted’ human cancer cell lines, we found that most cells can be rescued from drug sensitivity by simply exposing them to one or more RTK ligands. Among the findings with clinical implications was the observation that hepatocyte growth factor (HGF) confers resistance to the BRAF inhibitor PLX4032 (vemurafenib) in BRAF-mutant melanoma cells. These observations highlight the extensive redundancy of RTK-transduced signalling in cancer cells and the potentially broad role of widely expressed RTK ligands in innate and acquired resistance to drugs targeting oncogenic kinases.

A prediction-based resampling method for estimating the number of clusters in a dataset

BackgroundMicroarray technology is increasingly being applied in biological and medical research to address a wide range of problems, such as the classification of tumors. An important statistical problem associated with tumor classification is the identification of new tumor classes using gene-expression profiles. Two essential aspects of this clustering problem are: to estimate the number of clusters, if any, in a dataset; and to allocate tumor samples to these clusters, and assess the confidence of cluster assignments for individual samples. Here we address the first of these problems.ResultsWe have developed a new prediction-based resampling method, Clest, to estimate the number of clusters in a dataset. The performance of the new and existing methods were compared using simulated data and gene-expression data from four recently published cancer microarray studies. Clest was generally found to be more accurate and robust than the six existing methods considered in the study.ConclusionsFocusing on prediction accuracy in conjunction with resampling produces accurate and robust estimates of the number of clusters.

Bagging to improve the accuracy of a clustering procedure

MOTIVATION The microarray technology is increasingly being applied in biological and medical research to address a wide range of problems such as the classification of tumors. An important statistical question associated with tumor classification is the identification of new tumor classes using gene expression profiles. Essential aspects of this clustering problem include identifying accurate partitions of the tumor samples into clusters and assessing the confidence of cluster assignments for individual samples. RESULTS Two new resampling methods, inspired from bagging in prediction, are proposed to improve and assess the accuracy of a given clustering procedure. In these ensemble methods, a partitioning clustering procedure is applied to bootstrap learning sets and the resulting multiple partitions are combined by voting or the creation of a new dissimilarity matrix. As in prediction, the motivation behind bagging is to reduce variability in the partitioning results via averaging. The performances of the new and existing methods were compared using simulated data and gene expression data from two recently published cancer microarray studies. The bagged clustering procedures were in general at least as accurate and often substantially more accurate than a single application of the partitioning clustering procedure. A valuable by-product of bagged clustering are the cluster votes which can be used to assess the confidence of cluster assignments for individual observations. SUPPLEMENTARY INFORMATION For supplementary information on datasets, analyses, and software, consult http://www.stat.berkeley.edu/~sandrine and http://www.bioconductor.org.

A comparison study: applying segmentation to array CGH data for downstream analyses

MOTIVATION Array comparative genomic hybridization (CGH) allows detection and mapping of copy number of DNA segments. A challenge is to make inferences about the copy number structure of the genome. Several statistical methods have been proposed to determine genomic segments with different copy number levels. However, to date, no comprehensive comparison of various characteristics of these methods exists. Moreover, the segmentation results have not been utilized in downstream analyses. RESULTS We describe a comparison of three popular and publicly available methods for the analysis of array CGH data and we demonstrate how segmentation results may be utilized in the downstream analyses such as testing and classification, yielding higher power and prediction accuracy. Since the methods operate on individual chromosomes, we also propose a novel procedure for merging segments across the genome, which results in an interpretable set of copy number levels, and thus facilitate identification of copy number alterations in each genome. AVAILABILITY http://www.bioconductor.org

Improving Melanoma Classification by Integrating Genetic and Morphologic Features

Background In melanoma, morphology-based classification systems have not been able to provide relevant information for selecting treatments for patients whose tumors have metastasized. The recent identification of causative genetic alterations has revealed mutations in signaling pathways that offer targets for therapy. Identifying morphologic surrogates that can identify patients whose tumors express such alterations (or functionally equivalent alterations) would be clinically useful for therapy stratification and for retrospective analysis of clinical trial data. Methodology/Principal Findings We defined and assessed a panel of histomorphologic measures and correlated them with the mutation status of the oncogenes BRAF and NRAS in a cohort of 302 archival tissues of primary cutaneous melanomas from an academic comprehensive cancer center. Melanomas with BRAF mutations showed distinct morphological features such as increased upward migration and nest formation of intraepidermal melanocytes, thickening of the involved epidermis, and sharper demarcation to the surrounding skin; and they had larger, rounder, and more pigmented tumor cells (all p-values below 0.0001). By contrast, melanomas with NRAS mutations could not be distinguished based on these morphological features. Using simple combinations of features, BRAF mutation status could be predicted with up to 90.8% accuracy in the entire cohort as well as within the categories of the current World Health Organization (WHO) classification. Among the variables routinely recorded in cancer registries, we identified age < 55 y as the single most predictive factor of BRAF mutation in our cohort. Using age < 55 y as a surrogate for BRAF mutation in an independent cohort of 4,785 patients of the Southern German Tumor Registry, we found a significant survival benefit (p < 0.0001) for patients who, based on their age, were predicted to have BRAF mutant melanomas in 69% of the cases. This group also showed a different pattern of metastasis, more frequently involving regional lymph nodes, compared to the patients predicted to have no BRAF mutation and who more frequently displayed satellite, in-transit metastasis, and visceral metastasis (p < 0.0001). Conclusions Refined morphological classification of primary melanomas can be used to improve existing melanoma classifications by forming subgroups that are genetically more homogeneous and likely to differ in important clinical variables such as outcome and pattern of metastasis. We expect this information to improve classification and facilitate stratification for therapy as well as retrospective analysis of existing trial data.

Breast tumor copy number aberration phenotypes and genomic instability

BackgroundGenomic DNA copy number aberrations are frequent in solid tumors, although the underlying causes of chromosomal instability in tumors remain obscure. Genes likely to have genomic instability phenotypes when mutated (e.g. those involved in mitosis, replication, repair, and telomeres) are rarely mutated in chromosomally unstable sporadic tumors, even though such mutations are associated with some heritable cancer prone syndromes.MethodsWe applied array comparative genomic hybridization (CGH) to the analysis of breast tumors. The variation in the levels of genomic instability amongst tumors prompted us to investigate whether alterations in processes/genes involved in maintenance and/or manipulation of the genome were associated with particular types of genomic instability.ResultsWe discriminated three breast tumor subtypes based on genomic DNA copy number alterations. The subtypes varied with respect to level of genomic instability. We find that shorter telomeres and altered telomere related gene expression are associated with amplification, implicating telomere attrition as a promoter of this type of aberration in breast cancer. On the other hand, the numbers of chromosomal alterations, particularly low level changes, are associated with altered expression of genes in other functional classes (mitosis, cell cycle, DNA replication and repair). Further, although loss of function instability phenotypes have been demonstrated for many of the genes in model systems, we observed enhanced expression of most genes in tumors, indicating that over expression, rather than deficiency underlies instability.ConclusionMany of the genes associated with higher frequency of copy number aberrations are direct targets of E2F, supporting the hypothesis that deregulation of the Rb pathway is a major contributor to chromosomal instability in breast tumors. These observations are consistent with failure to find mutations in sporadic tumors in genes that have roles in maintenance or manipulation of the genome.

Phase I Study of Intraventricular Administration of Rituximab in Patients With Recurrent CNS and Intraocular Lymphoma

PURPOSE We previously determined that intravenous administration of rituximab results in limited penetration of this agent into the leptomeningeal space. Systemic rituximab does not reduce the risk of CNS relapse or dissemination in patients with large cell lymphoma. We therefore conducted a phase I dose-escalation study of intrathecal rituximab monotherapy in patients with recurrent CNS non-Hodgkins lymphoma (NHL). PATIENTS AND METHODS The protocol planned nine injections of rituximab (10 mg, 25 mg, or 50 mg dose levels) through an Ommaya reservoir over 5 weeks. The safety profile of intraventricular rituximab was defined in 10 patients. RESULTS The maximum tolerated dose was determined to be 25 mg and rapid craniospinal axis distribution was demonstrated. Cytologic responses were detected in six patients; four patients exhibited complete response. Two patients experienced improvement in intraocular NHL and one exhibited resolution of parenchymal NHL. High RNA levels of Pim-2 and FoxP1 in meningeal lymphoma cells were associated with disease refractory to rituximab monotherapy. CONCLUSION These results suggest that intrathecal rituximab (10 to 25 mg) is feasible and effective in NHL involving the CNS.

Rare amplicons implicate frequent deregulation of cell fate specification pathways in oral squamous cell carcinoma

Genomes of solid tumors are characterized by gains and losses of regions, which may contribute to tumorigenesis by altering gene expression. Often the aberrations are extensive, encompassing whole chromosome arms, which makes identification of candidate genes in these regions difficult. Here, we focused on narrow regions of gene amplification to facilitate identification of genetic pathways important in oral squamous cell carcinoma (SCC) development. We used array comparative genomic hybridization (array CGH) to define minimum common amplified regions and then used expression analysis to identify candidate driver genes in amplicons that spanned <3 Mb. We found genes involved in integrin signaling (TLN1), survival (YAP1, BIRC2), and adhesion and migration (TLN1, LAMA3, MMP7), as well as members of the hedgehog (GLI2) and notch (JAG1, RBPSUH, FJX1) pathways to be amplified and overexpressed. Deregulation of these and other members of the hedgehog and notch pathways (HHIP, SMO, DLL1, NOTCH4) implicates deregulation of developmental and differentiation pathways, cell fate misspecification, in oral SCC development.

Amplification of PVT1 Contributes to the Pathophysiology of Ovarian and Breast Cancer

Purpose: This study was designed to elucidate the role of amplification at 8q24 in the pathophysiology of ovarian and breast cancer because increased copy number at this locus is one of the most frequent genomic abnormalities in these cancers. Experimental Design: To accomplish this, we assessed the association of amplification at 8q24 with outcome in ovarian cancers using fluorescence in situ hybridization to tissue microarrays and measured responses of ovarian and breast cancer cell lines to specific small interfering RNAs against the oncogene MYC and a putative noncoding RNA, PVT1, both of which map to 8q24. Results: Amplification of 8q24 was associated with significantly reduced survival duration. In addition, small interfering RNA–mediated reduction in either PVT1 or MYC expression inhibited proliferation in breast and ovarian cancer cell lines in which they were both amplified and overexpressed but not in lines in which they were not amplified/overexpressed. Inhibition of PVT1 expression also induced a strong apoptotic response in cell lines in which it was overexpressed but not in lines in which it was not amplified/overexpressed. Inhibition of MYC, on the other hand, did not induce an apoptotic response in cell lines in which MYC was amplified and overexpressed. Conclusions: These results suggest that MYC and PVT1 contribute independently to ovarian and breast pathogenesis when overexpressed because of genomic abnormalities. They also suggest that PVT1-mediated inhibition of apoptosis may explain why amplification of 8q24 is associated with reduced survival duration in patients treated with agents that act through apoptotic mechanisms.