Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles

Failure to prolyl hydroxylate hypoxia-inducible factor alpha phenocopies VHL inactivation in vivo

Many functions have been assigned to the von Hippel-Lindau tumor suppressor gene product (pVHL), including targeting the alpha subunits of the heterodimeric transcription factor HIF (hypoxia-inducible factor) for destruction. The binding of pVHL to HIFalpha requires that HIFalpha be hydroxylated on one of two prolyl residues. We introduced HIF1alpha and HIF2alpha variants that cannot be hydroxylated on these sites into the ubiquitously expressed ROSA26 locus along with a Lox-stop-Lox cassette that renders their expression Cre-dependent. Expression of the HIF2alpha variant in the skin and liver induced changes that were highly similar to those seen when pVHL is lost in these organs. Dual expression of the HIF1alpha and HIF2alpha variants in liver, however, more closely phenocopied the changes seen after pVHL inactivation than did the HIF2alpha variant alone. Moreover, gene expression profiling confirmed that the genes regulated by HIF1alpha and HIF2alpha in the liver are overlapping but non-identical. Therefore, the pathological changes caused by pVHL inactivation in skin and liver are due largely to dysregulation of HIF target genes.

Transcriptional genomics associates FOX transcription factors with human heart failure

BACKGROUND

Specific transcription factors (TFs) modulate cardiac gene expression in murine models of heart failure, but their relevance in human subjects remains untested. We developed and applied a computational approach called transcriptional genomics to test the hypothesis that a discrete set of cardiac TFs is associated with human heart failure.

METHODS AND RESULTS

RNA isolates from failing (n=196) and nonfailing (n=16) human hearts were hybridized with Affymetrix HU133A arrays, and differentially expressed heart failure genes were determined. TF binding sites overrepresented in the -5-kb promoter sequences of these heart failure genes were then determined with the use of public genome sequence databases. Binding sites for TFs identified in murine heart failure models (MEF2, NKX, NF-AT, and GATA) were significantly overrepresented in promoters of human heart failure genes (P<0.002; false discovery rate 2% to 4%). In addition, binding sites for FOX TFs showed substantial overrepresentation in both advanced human and early murine heart failure (P<0.002 and false discovery rate <4% for each). A role for FOX TFs was supported further by expression of FOXC1, C2, P1, P4, and O1A in failing human cardiac myocytes at levels similar to established hypertrophic TFs and by abundant FOXP1 protein in failing human cardiac myocyte nuclei.

CONCLUSIONS

Our results provide the first evidence that specific TFs identified in murine models (MEF2, NKX, NFAT, and GATA) are associated with human heart failure. Moreover, these data implicate specific members of the FOX family of TFs (FOXC1, C2, P1, P4, and O1A) not previously suggested in heart failure pathogenesis. These findings provide a crucial link between animal models and human disease and suggest a specific role for FOX signaling in modulating the hypertrophic response of the heart to stress in humans.

Linking oncogenic pathways with therapeutic opportunities

The accumulation of multiple mutations and alterations in the cancer genome underlies the complexity of cancer phenotypes. A consequence of these alterations is the deregulation of various cell-signalling pathways that control cell function. Molecular-profiling studies, particularly DNA microarray analyses, have the potential to describe this complexity. These studies also provide an opportunity to link pathway deregulation with potential therapeutic strategies. This approach, when coupled with other methods for identifying pathway activation, provides an opportunity to both match individual patients with the most appropriate therapeutic strategy and identify potential options for combination therapy.

BABELOMICS: a systems biology perspective in the functional annotation of genome-scale experiments

We present a new version of Babelomics, a complete suite of web tools for functional analysis of genome-scale experiments, with new and improved tools. New functionally relevant terms have been included such as CisRed motifs or bioentities obtained by text-mining procedures. An improved indexing has considerably speeded up several of the modules. An improved version of the FatiScan method for studying the coordinate behaviour of groups of functionally related genes is presented, along with a similar tool, the Gene Set Enrichment Analysis. Babelomics is now more oriented to test systems biology inspired hypotheses. Babelomics can be found at .

A dominant function of IKK/NF-kappaB signaling in global lipopolysaccharide-induced gene expression

Porphyromonas gingivalis is an etiologic pathogen of periodontitis that is one of the most common inflammatory diseases. Recently, we found that P. gingivalis LPS activated the transcription factor nuclear factor-kappaB (NF-kappaB) through the IkappaB kinase complex (IKK). NF-kappaB is a transcription factor that controls inflammation and host responses. In this study, we examined the role of IKK/NF-kappaBin P. gingivalis LPS-induced gene expression on a genome-wide basis using a combination of microarray and biochemical approaches. A total of 88 early response genes were found to be induced by P. gingivalis LPS in a human THP.1 monocytic cell lines. Interestingly, the induction of most of these genes was abolished or attenuated under the inactivation of IKK/NF-kappaB. Among those IKK/NF-kappaB-dependent genes, 20 genes were NF-kappaB-inducible genes reported previously, and 59 genes represented putative novel NF-kappaB target genes. Using transcription factor binding analysis, we found that most of these putative NF-kappaB target genes contained one or multiple NF-kappaB-binding sites. Also, some transcription factor-binding motifs were overrepresented in the promoter of both known and putative NF-kappaB-dependent genes, indicating that these genes may be regulated in a similar fashion. Furthermore, we found that several transcription factors associated with metabolic and inflammatory responses, including nuclear receptors, activator of protein-1, and early growth responses, were induced by P. gingivalis LPS through IKK/NF-kappaB, indicating that IKK/NF-kappaB may utilize these transcription factors to mediate secondary responses. Taken together, our results demonstrate that IKK/NF-kappaB signaling plays a dominant role in P. gingivalis LPS-induced early response gene expression, suggesting that IKK/NF-kappaB is a therapeutic target for periodontitis.

Heterogeneity of Breast Cancer among Patients and Implications for Patient Selection for Adjuvant Chemotherapy

Although the benefits of adjuvant chemotherapy are not controversial, the absolute effect of such therapy is small. Therefore, there is a need to identify biomarkers that can help select patients with localized breast cancer for treatment. Despite intense research in this field, no biomarker has been shown to be useful to predict benefit of adjuvant chemotherapy in daily practice. This can partially be explained by the fact that breast cancer is composed of several distinct subclasses, as shown by large-scale genomic analyses. In this review, we discuss why the current research approach based on a single biomarker is limited by the heterogeneity of cancer among patients. We then propose three solutions to improve the research strategies in this field: investigate one biomarker in a single homogeneous subclass to improve its predictive value; study the predictive value of multibiomarker assays in larger populations; and use functional pathways to predict the efficacy of a given drug.

Defining aggressive prostate cancer using a 12-gene model

The critical clinical question in prostate cancer research is: How do we develop means of distinguishing aggressive disease from indolent disease? Using a combination of proteomic and expression array data, we identified a set of 36 genes with concordant dysregulation of protein products that could be evaluated in situ by quantitative immunohistochemistry. Another five prostate cancer biomarkers were included using linear discriminant analysis, we determined that the optimal model used to predict prostate cancer progression consisted of 12 proteins. Using a separate patient population, transcriptional levels of the 12 genes encoding for these proteins predicted prostate-specific antigen failure in 79 men following surgery for clinically localized prostate cancer (P = .0015). This study demonstrates that cross-platform models can lead to predictive models with the possible advantage of being more robust through this selection process.

A rapid method for microarray cross platform comparisons using gene expression signatures

Microarray technology has become highly valuable for identifying complex changes in global gene expression patterns. The inevitable use of a variety of different platforms has compounded the difficulty of effectively comparing data between projects, laboratories, and public access databases. The need for consistent, believable results across platforms is fundamental and methods for comparing results across platforms should be as straightforward as possible. We present the results of a study comparing three major, commercially available, microarray platforms (Affymetrix, Agilent, and Illumina). Concordance estimates between platforms was based on mapping of probes to Human Gene Organization (HUGO) gene names. Appropriate data normalization procedures were applied to each dataset followed by the generation of lists of regulated genes using a common significance threshold for all three platforms. As expected, concordance measured by directly comparing gene lists was relatively low (an average 22.8% for all platforms across all possible comparisons). However, when statistical tests (gene set enrichment analysis--GSEA, parametric analysis of gene enrichment--PAGE) which align gene lists with continuous measures of differential gene expression were applied to the cross platform datasets using significant gene lists to poll entire datasets, the relatedness of the results from all three platforms was specific, obvious, and profound.

DNA arrays as predictors of efficacy of adjuvant/neoadjuvant chemotherapy in breast cancer patients: current data and issues on study design

Chemotherapy provides variable benefit to patients with breast cancer, with usually modest but occasionally severe side effects. Hence, there is a need to identify predictive biomarkers for its efficacy. DNA arrays have been used in this setting as potential novel predictive diagnostic tools. Several gene signatures and single gene markers were proposed to predict response to chemotherapy. Although this technology offers interesting perspectives through large-scale analysis of the transcriptome, its ability to identify clinically relevant predictors is highly dependent on study design. In the present manuscript, we will review currently available results of breast cancer pharmacogenomics and focus on aspects of study design that are critical to reliably identify predictive biomarkers using DNA array technology. We will discuss whether studies should be done in the overall, unselected breast cancer population or in specific homogeneous molecular subclasses. Next, we will compare advantages and limitations of cohort-based and case-control studies. The choice of end-point to discriminate between sensitive and resistant patients will also be examined.

Group testing for pathway analysis improves comparability of different microarray datasets

MOTIVATION

The wide use of DNA microarrays for the investigation of the cell transcriptome triggered the invention of numerous methods for the processing of microarray data and lead to a growing number of microarray studies that examine the same biological conditions. However, comparisons made on the level of gene lists obtained by different statistical methods or from different datasets hardly converge. We aimed at examining such discrepancies on the level of apparently affected biologically related groups of genes, e.g. metabolic or signalling pathways. This can be achieved by group testing procedures, e.g. over-representation analysis, functional class scoring (FCS), or global tests.

RESULTS

Three public prostate cancer datasets obtained with the same microarray platform (HGU95A/HGU95Av2) were analyzed. Each dataset was subjected to normalization by either variance stabilizing normalization (vsn) or mixed model normalization (MMN). Then, statistical analysis of microarrays was applied to the vsn-normalized data and mixed model analysis to the data normalized by MMN. For multiple testing adjustment the false discovery rate was calculated and the threshold was set to 0.05. Gene lists from the same method applied to different datasets showed overlaps between 42 and 52%, while lists from different methods applied to the same dataset had between 63 and 85% of genes in common. A number of six gene lists obtained by the two statistical methods applied to the three datasets was then subjected to group testing by Fisher's exact test. Group testing by GSEA and global test was applied to the three datasets, as well. Fisher's exact test followed by global test showed more consistent results with respect to the concordance between analyses on gene lists obtained by different methods and different datasets than the GSEA. However, all group testing methods identified pathways that had already been described to be involved in the pathogenesis of prostate cancer. Moreover, pathways recurrently identified in these analyses are more likely to be reliable than those from a single analysis on a single dataset.

Analysis of microarray experiments of gene expression profiling

The study of gene expression profiling of cells and tissue has become a major tool for discovery in medicine. Microarray experiments allow description of genome-wide expression changes in health and disease. The results of such experiments are expected to change the methods employed in the diagnosis and prognosis of disease in obstetrics and gynecology. Moreover, an unbiased and systematic study of gene expression profiling should allow the establishment of a new taxonomy of disease for obstetric and gynecologic syndromes. Thus, a new era is emerging in which reproductive processes and disorders could be characterized using molecular tools and fingerprinting. The design, analysis, and interpretation of microarray experiments require specialized knowledge that is not part of the standard curriculum of our discipline. This article describes the types of studies that can be conducted with microarray experiments (class comparison, class prediction, class discovery). We discuss key issues pertaining to experimental design, data preprocessing, and gene selection methods. Common types of data representation are illustrated. Potential pitfalls in the interpretation of microarray experiments, as well as the strengths and limitations of this technology, are highlighted. This article is intended to assist clinicians in appraising the quality of the scientific evidence now reported in the obstetric and gynecologic literature.

Transcription Association of VHL and SDH Mutations Link Hypoxia and Oxidoreductase Signals in Pheochromocytomas

Pheochromocytomas and paragangliomas are neural-crest-derived tumors that arise from mutations in RET, VHL, NF1, and in the genes-encoding succinate dehydrogenase (SDH) subunits B (SDHB), C (SDHC), and D (SDHD). Despite their genetic diversity, these tumors cannot be clearly distinguished on the basis of their primary mutation. We recently identified two major transcriptional programs embedded within familial and sporadic pheochromocytomas and paragangliomas using global expression profiling. This review will summarize the major results of these studies and discuss their implications. The transcription data revealed that: (a) tumors with mutations in VHL, SDHB, and SDHD genes share a transcription signature of hypoxia, angiogenesis, and oxidoreductase imbalance; (b) SDHB protein is suppressed in tumors with mutations in SDHB and SDHD, and also in a subset of tumors with VHL mutations; and (c) HIF1alpha is involved in the SDHB downregulation observed in these tumors. These results are consistent with the existence of a close interconnection between the VHL and SDH pathways mediated predominantly by hypoxia and oxidoreductase signals. It further suggests that low SDHB levels indicative of impaired mitochondrial complex II function may be a shared element of these pheochromocytomas. SDHB may thus constitute a marker for tumors with abnormal hypoxic profile.

A multivariate approach for integrating genome-wide expression data and biological knowledge

MOTIVATION

Several statistical methods that combine analysis of differential gene expression with biological knowledge databases have been proposed for a more rapid interpretation of expression data. However, most such methods are based on a series of univariate statistical tests and do not properly account for the complex structure of gene interactions.

RESULTS

We present a simple yet effective multivariate statistical procedure for assessing the correlation between a subspace defined by a group of genes and a binary phenotype. A subspace is deemed significant if the samples corresponding to different phenotypes are well separated in that subspace. The separation is measured using Hotelling's T(2) statistic, which captures the covariance structure of the subspace. When the dimension of the subspace is larger than that of the sample space, we project the original data to a smaller orthonormal subspace. We use this method to search through functional pathway subspaces defined by Reactome, KEGG, BioCarta and Gene Ontology. To demonstrate its performance, we apply this method to the data from two published studies, and visualize the results in the principal component space.

Microarray-assisted pathway analysis identifies mitogen-activated protein kinase signaling as a mediator of resistance to the green tea polyphenol epigallocatechin 3-gallate in her-2/neu-overexpressing breast cancer cells

Overexpression of the epidermal growth factor receptor family member Her-2/neu in breast cancer leads to autophosphorylation of the receptor and induction of multiple downstream signaling pathways, including the Akt kinase to nuclear factor-kappaB (NF-kappaB) cascade that is associated with poor prognosis. Previously, we showed that the green tea polyphenol epigallocatechin 3-gallate (EGCG) inhibits growth of NF639 Her-2/neu-driven breast cancer cells via reducing receptor autophosphorylation and downstream Akt and NF-kappaB activities. Interestingly, upon prolonged culture in the presence of EGCG, cells resistant to the polyphenol could be isolated. Here, we report that resistant cells have lost tyrosine phosphorylation on the Her-2/neu receptor. Surprisingly, they displayed elevated NF-kappaB activity, and inhibition of this activity sensitized cells to EGCG. Data from microarray studies of the original and resistant NF639 populations of cells were subjected to Gene Set Enrichment Analysis pathway assessment, which revealed that the mitogen activated protein kinase (MAPK) pathway was activated in the resistant cells. Treatment of the resistant cells with the MAPK inhibitor U0216 reduced growth in soft agar and invasive phenotype, whereas the combination of EGCG and U0216 resulted in cells with a cobblestone epithelial phenotype. Thus, activation of the MAPK pathway mediates resistance to EGCG.

Transformation from committed progenitor to leukaemia stem cell initiated by MLL-AF9

Leukaemias and other cancers possess a rare population of cells capable of the limitless self-renewal necessary for cancer initiation and maintenance. Eradication of these cancer stem cells is probably a critical part of any successful anti-cancer therapy, and may explain why conventional cancer therapies are often effective in reducing tumour burden, but are only rarely curative. Given that both normal and cancer stem cells are capable of self-renewal, the extent to which cancer stem cells resemble normal tissue stem cells is a critical issue if targeted therapies are to be developed. However, it remains unclear whether cancer stem cells must be phenotypically similar to normal tissue stem cells or whether they can retain the identity of committed progenitors. Here we show that leukaemia stem cells (LSC) can maintain the global identity of the progenitor from which they arose while activating a limited stem-cell- or self-renewal-associated programme. We isolated LSC from leukaemias initiated in committed granulocyte macrophage progenitors through introduction of the MLL-AF9 fusion protein encoded by the t(9;11)(p22;q23). The LSC were capable of transferring leukaemia to secondary recipient mice when only four cells were transferred, and possessed an immunophenotype and global gene expression profile very similar to that of normal granulocyte macrophage progenitors. However, a subset of genes highly expressed in normal haematopoietic stem cells was re-activated in LSC. LSC can thus be generated from committed progenitors without widespread reprogramming of gene expression, and a leukaemia self-renewal-associated signature is activated in the process. Our findings define progression from normal progenitor to cancer stem cell, and suggest that targeting a self-renewal programme expressed in an abnormal context may be possible.

A subtype of multiple sclerosis defined by an activated immune defense program

Given the heterogeneous nature of multiple sclerosis (MS), we applied DNA microarray technology to determine whether variability is reflected in peripheral blood (PB) cells. In this study, we studied whole-blood gene expression profiles of 29 patients with relapsing-remitting MS (RRMS) and 25 age- and sex-matched healthy controls. We used microarrays with a complexity of 43K cDNAs. The data were analyzed using sophisticated pathway-level analysis in order to provide insight into the deregulated peripheral immune response programs in MS. We found a remarkable elevated expression of a spectrum of genes known to be involved in immune defense in the PB of MS patients compared to healthy individuals. Cluster analysis revealed that the increased expression of these genes was characteristic for approximately half of the patients. In addition, the gene signature in this group of patients was comparable with a virus response program. We conclude that the transcriptional signature of the PB cells reflects the heterogeneity of MS and defines a sub-population of RRMS patients, who exhibit an activated immune defense program that resembles a virus response program, which is supportive for a link between viruses and MS.

ADGO: analysis of differentially expressed gene sets using composite GO annotation

MOTIVATION

Genes are typically expressed in modular manners in biological processes. Recent studies reflect such features in analyzing gene expression patterns by directly scoring gene sets. Gene annotations have been used to define the gene sets, which have served to reveal specific biological themes from expression data. However, current annotations have limited analytical power, because they are classified by single categories providing only unary information for the gene sets.

RESULTS

Here we propose a method for discovering composite biological themes from expression data. We intersected two annotated gene sets from different categories of Gene Ontology (GO). We then scored the expression changes of all the single and intersected sets. In this way, we were able to uncover, for example, a gene set with the molecular function F and the cellular component C that showed significant expression change, while the changes in individual gene sets were not significant. We provided an exemplary analysis for HIV-1 immune response. In addition, we tested the method on 20 public datasets where we found many 'filtered' composite terms the number of which reached approximately 34% (a strong criterion, 5% significance) of the number of significant unary terms on average. By using composite annotation, we can derive new and improved information about disease and biological processes from expression data.

AVAILABILITY

We provide a web application (ADGO: http://array.kobic.re.kr/ADGO) for the analysis of differentially expressed gene sets with composite GO annotations. The user can analyze Affymetrix and dual channel array (spotted cDNA and spotted oligo microarray) data for four species: human, mouse, rat and yeast.

CONTACT

chu@kribb.re.kr

SUPPLEMENTARY INFORMATION

http://array.kobic.re.kr/ADGO.

Genomic approaches to drug discovery

Considerable progress has been made in exploiting the enormous amount of genomic and genetic information for the identification of potential targets for drug discovery and development. New tools that incorporate pathway information have been developed for gene expression data mining to reflect differences in pathways in normal and disease states. In addition, forward and reverse genetics used in a high-throughput mode with full-length cDNA and RNAi libraries enable the direct identification of components of signaling pathways. The discovery of the regulatory function of microRNAs highlights the importance of continuing the investigation of the genome with sophisticated tools. Furthermore, epigenetic information including DNA methylation and histone modifications that mediate important biological processes add to the possibilities to identify novel drug targets and patient populations that will benefit from new therapies.

The application of systems biology to drug discovery

Recent advances in the 'omics' technologies, scientific computing and mathematical modeling of biological processes have started to fundamentally impact the way we approach drug discovery. Recent years have witnessed the development of genome-scale functional screens, large collections of reagents, protein microarrays, databases and algorithms for data and text mining. Taken together, they enable the unprecedented descriptions of complex biological systems, which are testable by mathematical modeling and simulation. While the methods and tools are advancing, it is their iterative and combinatorial application that defines the systems biology approach.

The Arabidopsis root hair cell wall formation mutant lrx1 is suppressed by mutations in the RHM1 gene encoding a UDP-L-rhamnose synthase

Cell and cell wall growth are mutually dependent processes that must be tightly coordinated and controlled. LRR-extensin1 (LRX1) of Arabidopsis thaliana is a potential regulator of cell wall development, consisting of an N-terminal leucine-rich repeat domain and a C-terminal extensin-like domain typical for structural cell wall proteins. LRX1 is expressed in root hairs, and lrx1 mutant plants develop distorted root hairs that often swell, branch, or collapse. The aberrant cell wall structures found in lrx1 mutants point toward a function of LRX1 during the establishment of the extracellular matrix. To identify genes that are involved in an LRX1-dependent developmental pathway, a suppressor screen was performed on the lrx1 mutant, and two independent rol1 (for repressor of lrx1) alleles were isolated. ROL1 is allelic to Rhamnose Biosynthesis1, which codes for a protein involved in the biosynthesis of rhamnose, a major monosaccharide component of pectin. The rol1 mutations modify the pectic polysaccharide rhamnogalacturonan I and, for one allele, rhamnogalacturonan II. Furthermore, the rol1 mutations cause a change in the expression of a number of cell wall-related genes. Thus, the lrx1 mutant phenotype is likely to be suppressed by changes in pectic polysaccharides or other cell wall components.

JProGO: a novel tool for the functional interpretation of prokaryotic microarray data using Gene Ontology information

A novel program suite was implemented for the functional interpretation of high-throughput gene expression data based on the identification of Gene Ontology (GO) nodes. The focus of the analysis lies on the interpretation of microarray data from prokaryotes. The three well established statistical methods of the threshold value-based Fisher's exact test, as well as the threshold value-independent Kolmogorov-Smirnov and Student's t-test were employed in order to identify the groups of genes with a significantly altered expression profile. Furthermore, we provide the application of the rank-based unpaired Wilcoxon's test for a GO-based microarray data interpretation. Further features of the program include recognition of the alternative gene names and the correction for multiple testing. Obtained results are visualized interactively both as a table and as a GO subgraph including all significant nodes. Currently, JProGO enables the analysis of microarray data from more than 20 different prokaryotic species, including all important model organisms, and thus constitutes a useful web service for the microbial research community. JProGO is freely accessible via the web at the following address: http://www.jprogo.de.

Cellular transcriptomics -- the next phase of endocrine expression profiling

Transcriptome analysis, or global gene expression profiling, has become a commonly used and valuable tool in both basic and clinical endocrine research. Novel endocrine regulators have 'surfaced' and greater consideration is now given to understanding function at the level of gene networks. Recent developments have shown that the transcriptome is considerably larger and more divergently expressed than was previously thought. Endocrine cells express a great variety of coding and noncoding RNAs in a highly cell-specific manner. If further value is to be taken from this research area, then steps towards defined cellular transcriptomics must be taken. New sampling techniques that utilize novel genetic models are a key first step.

Next station in microarray data analysis: GEPAS

The Gene Expression Profile Analysis Suite (GEPAS) has been running for more than four years. During this time it has evolved to keep pace with the new interests and trends in the still changing world of microarray data analysis. GEPAS has been designed to provide an intuitive although powerful web-based interface that offers diverse analysis options from the early step of preprocessing (normalization of Affymetrix and two-colour microarray experiments and other preprocessing options), to the final step of the functional annotation of the experiment (using Gene Ontology, pathways, PubMed abstracts etc.), and include different possibilities for clustering, gene selection, class prediction and array-comparative genomic hybridization management. GEPAS is extensively used by researchers of many countries and its records indicate an average usage rate of 400 experiments per day. The web-based pipeline for microarray gene expression data, GEPAS, is available at http://www.gepas.org.

Pathway analysis using random forests classification and regression

MOTIVATION

Although numerous methods have been developed to better capture biological information from microarray data, commonly used single gene-based methods neglect interactions among genes and leave room for other novel approaches. For example, most classification and regression methods for microarray data are based on the whole set of genes and have not made use of pathway information. Pathway-based analysis in microarray studies may lead to more informative and relevant knowledge for biological researchers.

RESULTS

In this paper, we describe a pathway-based classification and regression method using Random Forests to analyze gene expression data. The proposed methods allow researchers to rank important pathways from externally available databases, discover important genes, find pathway-based outlying cases and make full use of a continuous outcome variable in the regression setting. We also compared Random Forests with other machine learning methods using several datasets and found that Random Forests classification error rates were either the lowest or the second-lowest. By combining pathway information and novel statistical methods, this procedure represents a promising computational strategy in dissecting pathways and can provide biological insight into the study of microarray data.

AVAILABILITY

Source code written in R is available from http://bioinformatics.med.yale.edu/pathway-analysis/rf.htm.

Inference of disease-related molecular logic from systems-based microarray analysis

Computational analysis of gene expression data from microarrays has been useful for medical diagnosis and prognosis. The ability to analyze such data at the level of biological modules, rather than individual genes, has been recognized as important for improving our understanding of disease-related pathways. It has proved difficult, however, to infer pathways from microarray data by deriving modules of multiple synergistically interrelated genes, rather than individual genes. Here we propose a systems-based approach called Entropy Minimization and Boolean Parsimony (EMBP) that identifies, directly from gene expression data, modules of genes that are jointly associated with disease. Furthermore, the technique provides insight into the underlying biomolecular logic by inferring a logic function connecting the joint expression levels in a gene module with the outcome of disease. Coupled with biological knowledge, this information can be useful for identifying disease-related pathways, suggesting potential therapeutic approaches for interfering with the functions of such pathways. We present an example providing such gene modules associated with prostate cancer from publicly available gene expression data, and we successfully validate the results on additional independently derived data. Our results indicate a link between prostate cancer and cellular damage from oxidative stress combined with inhibition of apoptotic mechanisms normally triggered by such damage.

Identification of QTLs controlling gene expression networks defined a priori

Background

Gene expression microarrays allow the quantification of transcript accumulation for many or all genes in a genome. This technology has been utilized for a range of investigations, from assessments of gene regulation in response to genetic or environmental fluctuation to global expression QTL (eQTL) analyses of natural variation. Current analysis techniques facilitate the statistical querying of individual genes to evaluate the significance of a change in response, also known as differential expression. Since genes are also known to respond as groups due to their membership in networks, effective approaches are needed to investigate transcriptome variation as related to gene network responses.

Results

We describe a statistical approach that is capable of assessing higher-order a priori defined gene network response, as measured by microarrays. This analysis detected significant network variation between two Arabidopsis thaliana accessions, Bay-0 and Shahdara. By extending this approach, we were able to identify eQTLs controlling network responses for 18 out of 20 a priori-defined gene networks in a recombinant inbred line population derived from accessions Bay-0 and Shahdara.

Conclusion

This approach has the potential to be expanded to facilitate direct tests of the relationship between phenotypic trait and transcript genetic architecture. The use of a priori definitions for network eQTL identification has enormous potential for providing direction toward future eQTL analyses.

A review of statistical methods for expression quantitative trait loci mapping

With high-throughput technologies now widely available, investigators can easily measure thousands of phenotypes for quantitative trait loci (QTL) mapping. Microarray measurements are particularly amenable to QTL mapping, as evidenced by a number of recent studies demonstrating utility across a broad range of biological endeavors. The early success stories have impelled a rapid increase in both the number and complexity of expression QTL (eQTL) experiments. Consequently, there is a need to consider the statistical principles involved in the design and analysis of these experiments and the methods currently being used. In this article we review these principles and methods and discuss the open questions most likely to yield significant progress toward increasing the amount of meaningful information obtained from eQTL mapping experiments.

Improved prediction of treatment response using microarrays and existing biological knowledge

A desired application for microarrays in the clinic is to predict treatment response from an often diverse patient population. We present a method for analyzing microarray data that is predicated on biological pathway and function knowledge as opposed to a purely data-driven initial analysis. From an analysis perspective, this methodology takes advantage of information that is available across genes on a single array, as well as differences in those patterns across measurements. By using biological knowledge in the initial analysis, the accuracy and robustness of microarray profile classification is enhanced, especially when low numbers of samples are available. For clinical studies, particularly Phase I or I/II studies, this technique is exceptionally advantageous.

Integrating genetic and gene expression data: application to cardiovascular and metabolic traits in mice

The millions of common DNA variations that occur in the human population, or among inbred strains of mice and rats, perturb the expression (transcript levels) of a large fraction of the genes expressed in a particular tissue. The hundreds or thousands of common cis-acting variations that occur in the population may in turn affect the expression of thousands of other genes by affecting transcription factors, signaling molecules, RNA processing, and other processes that act in trans. The levels of transcripts are conveniently quantitated using expression arrays, and the cis- and trans-acting loci can be mapped using quantitative trait locus (QTL) analysis, in the same manner as loci for physiologic or clinical traits. Thousands of such expression QTL (eQTL) have been mapped in various crosses in mice, as well as other experimental organisms, and less detailed maps have been produced in studies of cells from human pedigrees. Such an integrative genetics approach (sometimes referred to as “genetical genomics”) is proving useful for identifying genes and pathways that contribute to complex clinical traits. The coincidence of clinical trait QTL and eQTL can help in the prioritization of positional candidate genes. More importantly, mathematical modeling of correlations between levels of transcripts and clinical traits in genetic crosses can allow prediction of causal interactions and the identification of “key driver” genes. An important objective of such studies will be to model biological networks in physiologic processes. When combined with high-density single nucleotide polymorphism (SNP) mapping, it should be feasible to identify genes that contribute to transcript levels using association analysis in outbred populations. In this review we discuss the basic concepts and applications of this integrative genomic approach to cardiovascular and metabolic diseases.

Transcriptional profiling of aging in human muscle reveals a common aging signature

We analyzed expression of 81 normal muscle samples from humans of varying ages, and have identified a molecular profile for aging consisting of 250 age-regulated genes. This molecular profile correlates not only with chronological age but also with a measure of physiological age. We compared the transcriptional profile of muscle aging to previous transcriptional profiles of aging in the kidney and the brain, and found a common signature for aging in these diverse human tissues. The common aging signature consists of six genetic pathways; four pathways increase expression with age (genes in the extracellular matrix, genes involved in cell growth, genes encoding factors involved in complement activation, and genes encoding components of the cytosolic ribosome), while two pathways decrease expression with age (genes involved in chloride transport and genes encoding subunits of the mitochondrial electron transport chain). We also compared transcriptional profiles of aging in humans to those of the mouse and fly, and found that the electron transport chain pathway decreases expression with age in all three organisms, suggesting that this may be a public marker for aging across species.

Transcriptional profiling of aging in human muscle reveals a common aging signature

We analyzed expression of 81 normal muscle samples from humans of varying ages, and have identified a molecular profile for aging consisting of 250 age-regulated genes. This molecular profile correlates not only with chronological age but also with a measure of physiological age. We compared the transcriptional profile of muscle aging to previous transcriptional profiles of aging in the kidney and the brain, and found a common signature for aging in these diverse human tissues. The common aging signature consists of six genetic pathways; four pathways increase expression with age (genes in the extracellular matrix, genes involved in cell growth, genes encoding factors involved in complement activation, and genes encoding components of the cytosolic ribosome), while two pathways decrease expression with age (genes involved in chloride transport and genes encoding subunits of the mitochondrial electron transport chain). We also compared transcriptional profiles of aging in humans to those of the mouse and fly, and found that the electron transport chain pathway decreases expression with age in all three organisms, suggesting that this may be a public marker for aging across species.

Genome-Wide Expression Studies of Atherosclerosis

During the past 6 years, gene expression profiling of atherosclerosis has been used to identify genes and pathways relevant in vascular (patho)physiology. This review discusses some critical issues in the methodology, analysis, and interpretation of the data of gene expression studies that have made use of vascular specimens from animal models and humans. Analysis of gene expression studies has evolved toward the genome-wide expression profiling of large series of individual samples of well-characterized donors. Despite the advances in statistical and bioinformatical analysis of expression data sets, studies have not yet fully exploited the potential of gene expression data sets to obtain novel insights into the molecular mechanisms underlying atherosclerosis. To assess the potential of published expression data, we compared the data of a CC chemokine gene cluster between 18 murine and human gene expression profiling articles. Our analysis revealed that an adequate comparison is mainly hindered by the incompleteness of available data sets. The challenge for future vascular genomic profiling studies will be to further improve the experimental design, statistical, and bioinformatical analysis and to make data sets freely accessible.

Downregulation of electron transport chain genes in visceral adipose tissue in type 2 diabetes independent of obesity and possibly involving tumor necrosis factor-alpha

Impaired oxidative phosphorylation is suggested as a factor behind insulin resistance of skeletal muscle in type 2 diabetes. The role of oxidative phosphorylation in adipose tissue was elucidated from results of Affymetrix gene profiling in subcutaneous and visceral adipose tissue of eight nonobese healthy, eight obese healthy, and eight obese type 2 diabetic women. Downregulation of several genes in the electron transport chain was the most prominent finding in visceral fat of type 2 diabetic women independent of obesity, but the gene pattern was distinct from that previously reported in skeletal muscle in type 2 diabetes. A similar but much weaker effect was observed in subcutaneous fat. Tumor necrosis factor-alpha (TNF-alpha) is a major factor behind inflammation and insulin resistance in adipose tissue. TNF-alpha treatment decreased mRNA expression of electron transport chain genes and also inhibited fatty acid oxidation when differentiated human preadipocytes were treated with the cytokine for 48 h. Thus, type 2 diabetes is associated with a tissue- and region-specific downregulation of oxidative phosphorylation genes that is independent of obesity and at least in part mediated by TNF-alpha, suggesting that impaired oxidative phosphorylation of visceral adipose tissue has pathogenic importance for development of type 2 diabetes.

Bioinformatics and cancer: an essential alliance

Modern research in cancer has been revolutionized by the introduction of new high-throughput methodologies such as DNA microarrays. Keeping the pace with these technologies, the bioinformatics offer new solutions for data analysis and, what is more important, it permits to formulate a new class of hypothesis inspired in systems biology, more oriented to blocks of functionally-related genes. Although software implementations for this new methodologies is new there are some options already available. Bioinformatic solutions for other high-throughput techniques such as array-CGH of large-scale genotyping is also revised.

Expression-guided in silico evaluation of candidate cis regulatory codes for Drosophila muscle founder cells

While combinatorial models of transcriptional regulation can be inferred for metazoan systems from a priori biological knowledge, validation requires extensive and time-consuming experimental work. Thus, there is a need for computational methods that can evaluate hypothesized cis regulatory codes before the difficult task of experimental verification is undertaken. We have developed a novel computational framework (termed “CodeFinder”) that integrates transcription factor binding site and gene expression information to evaluate whether a hypothesized transcriptional regulatory model (TRM; i.e., a set of co-regulating transcription factors) is likely to target a given set of co-expressed genes. Our basic approach is to simultaneously predict cis regulatory modules (CRMs) associated with a given gene set and quantify the enrichment for combinatorial subsets of transcription factor binding site motifs comprising the hypothesized TRM within these predicted CRMs. As a model system, we have examined a TRM experimentally demonstrated to drive the expression of two genes in a sub-population of cells in the developing Drosophila mesoderm, the somatic muscle founder cells. This TRM was previously hypothesized to be a general mode of regulation for genes expressed in this cell population. In contrast, the present analyses suggest that a modified form of this cis regulatory code applies to only a subset of founder cell genes, those whose gene expression responds to specific genetic perturbations in a similar manner to the gene on which the original model was based. We have confirmed this hypothesis by experimentally discovering six (out of 12 tested) new CRMs driving expression in the embryonic mesoderm, four of which drive expression in founder cells.

Although genome sequences and much gene expression data are readily available, the determination of sets of transcription factors regulating particular gene expression patterns remains a problem of fundamental importance. Tissue-specific gene expression in developing animals is regulated through the combinatorial interactions of transcription factors with DNA regulatory elements termed cis regulatory modules (CRMs). Although genetic and biochemical experiments allow the identification of transcription factors and CRMs, those experiments are laborious and time-consuming. Philippakis et al. introduce a new approach (termed “CodeFinder”) for quantifying the enrichment for particular combinations of transcription factor binding site motifs within predicted CRMs associated with a given gene set of interest, identified from gene expression data. The authors' analyses allowed them to discover a specific combination of transcription factor binding site motifs that constitute a core cis regulatory code for expression of a particular subset of genes in muscle founder cells, an embryonic cell population in the developing fruit fly (Drosophila melanogaster) mesoderm, and also led them to the discovery and subsequent experimental validation of novel, tissue-specific CRMs. Importantly, the CodeFinder approach is generally applicable, and thus could be used to support, refute, or refine a known or hypothesized cis regulatory code for any biological system or genome of interest.

Characterizing disease states from topological properties of transcriptional regulatory networks

Background

High throughput gene expression experiments yield large amounts of data that can augment our understanding of disease processes, in addition to classifying samples. Here we present new paradigms of data Separation based on construction of transcriptional regulatory networks for normal and abnormal cells using sequence predictions, literature based data and gene expression studies. We analyzed expression datasets from a number of diseased and normal cells, including different types of acute leukemia, and breast cancer with variable clinical outcome.

Results

We constructed sample-specific regulatory networks to identify links between transcription factors (TFs) and regulated genes that differentiate between healthy and diseased states. This approach carries the advantage of identifying key transcription factor-gene pairs with differential activity between healthy and diseased states rather than merely using gene expression profiles, thus alluding to processes that may be involved in gene deregulation. We then generalized this approach by studying simultaneous changes in functionality of multiple regulatory links pointing to a regulated gene or emanating from one TF (or changes in gene centrality defined by its in-degree or out-degree measures, respectively). We found that samples can often be separated based on these measures of gene centrality more robustly than using individual links.

We examined distributions of distances (the number of links needed to traverse the path between each pair of genes) in the transcriptional networks for gene subsets whose collective expression profiles could best separate each dataset into predefined groups. We found that genes that optimally classify samples are concentrated in neighborhoods in the gene regulatory networks. This suggests that genes that are deregulated in diseased states exhibit a remarkable degree of connectivity.

Conclusion

Transcription factor-regulated gene links and centrality of genes on transcriptional networks can be used to differentiate between cell types. Transcriptional network blueprints can be used as a basis for further research into gene deregulation in diseased states.

The challenges of gene expression microarrays for the study of human cancer

Large-scale genomic studies promise to advance our understanding of the biology of human cancers and to improve their diagnosis, prognostication, and treatment. The analysis and interpretation of genomics studies have faced challenges. The retrospective and observational design of many studies has rendered them susceptible to confounding and bias. Technological variations and advances have impacted on reproducibility. Statistical hurdles in relating a large number of variables to a small number of observations have added further constraints. This review considers the promise and challenge associated with the large-scale clinically oriented genomic analysis of human cancer and attempts to emphasize potential solutions.

Gene expression changes and molecular pathways mediating activity-dependent plasticity in visual cortex

Two key models for examining activity-dependent development of primary visual cortex (V1) involve either reduction of activity in both eyes via dark-rearing (DR) or imbalance of activity between the two eyes via monocular deprivation (MD). Combining DNA microarray analysis with computational approaches, RT-PCR, immunohistochemistry and physiological imaging, we find that DR leads to (i) upregulation of genes subserving synaptic transmission and electrical activity, consistent with a coordinated response of cortical neurons to reduction of visual drive, and (ii) downregulation of parvalbumin expression, implicating parvalbumin-expressing interneurons as underlying the delay in cortical maturation after DR. MD partially activates homeostatic mechanisms but differentially upregulates molecular pathways related to growth factors and neuronal degeneration, consistent with reorganization of connections after MD. Expression of a binding protein of insulin-like growth factor-1 (IGF1) is highly upregulated after MD, and exogenous application of IGF1 prevents the physiological effects of MD on ocular dominance plasticity examined in vivo.

Deregulated expression of fat and muscle genes in B-cell chronic lymphocytic leukemia with high lipoprotein lipase expression

Lipoprotein lipase (LPL) is a prognostic marker in B-cell chronic lymphocytic leukemia (B-CLL) related to immunoglobulin V(H) gene (IgV(H))mutational status. We determined gene expression profiles using Affymetrix U133A GeneChips in two groups of B-CLLs selected for either high ('LPL+', n=10) or low ('LPL-', n=10) LPL mRNA expression. Selected genes were verified by real-time PCR in an extended patient cohort (n=42). A total of 111 genes discriminated LPL+ from LPL- B-CLLs. Of these, the top three genes associated with time to first treatment were Septin10, DMD and Gravin (P</=0.01). The relationship of LPL+ and LPL- B-CLL gene expression signatures to 52 tissues was statistically analyzed. The LPL+ B-CLL expression signature, represented by 64 genes was significantly related to fat, muscle and PB dendritic cells (P<0.001). Exploration of microarray data to define functional alterations related to the biology of LPL+ CLL identified two functional modules, fatty acid degradation and MTA3 signaling, as being altered with higher statistical significance. Our data show that LPL+ B-CLL cells have not only acquired gene expression changes in fat and muscle-associated genes but also in functional pathways related to fatty acid degradation and signaling which may ultimately influence CLL biology and clinical outcome.

Improved scoring of functional groups from gene expression data by decorrelating GO graph structure

MOTIVATION

The result of a typical microarray experiment is a long list of genes with corresponding expression measurements. This list is only the starting point for a meaningful biological interpretation. Modern methods identify relevant biological processes or functions from gene expression data by scoring the statistical significance of predefined functional gene groups, e.g. based on Gene Ontology (GO). We develop methods that increase the explanatory power of this approach by integrating knowledge about relationships between the GO terms into the calculation of the statistical significance.

RESULTS

We present two novel algorithms that improve GO group scoring using the underlying GO graph topology. The algorithms are evaluated on real and simulated gene expression data. We show that both methods eliminate local dependencies between GO terms and point to relevant areas in the GO graph that remain undetected with state-of-the-art algorithms for scoring functional terms. A simulation study demonstrates that the new methods exhibit a higher level of detecting relevant biological terms than competing methods.

Reactive oxygen species have a causal role in multiple forms of insulin resistance

Insulin resistance is a cardinal feature of type 2 diabetes and is characteristic of a wide range of other clinical and experimental settings. Little is known about why insulin resistance occurs in so many contexts. Do the various insults that trigger insulin resistance act through a common mechanism? Or, as has been suggested, do they use distinct cellular pathways? Here we report a genomic analysis of two cellular models of insulin resistance, one induced by treatment with the cytokine tumour-necrosis factor-alpha and the other with the glucocorticoid dexamethasone. Gene expression analysis suggests that reactive oxygen species (ROS) levels are increased in both models, and we confirmed this through measures of cellular redox state. ROS have previously been proposed to be involved in insulin resistance, although evidence for a causal role has been scant. We tested this hypothesis in cell culture using six treatments designed to alter ROS levels, including two small molecules and four transgenes; all ameliorated insulin resistance to varying degrees. One of these treatments was tested in obese, insulin-resistant mice and was shown to improve insulin sensitivity and glucose homeostasis. Together, our findings suggest that increased ROS levels are an important trigger for insulin resistance in numerous settings.

Identification of distinct molecular phenotypes in acute megakaryoblastic leukemia by gene expression profiling

Individuals with Down syndrome (DS) are predisposed to develop acute megakaryoblastic leukemia (AMKL), characterized by expression of truncated GATA1 transcription factor protein (GATA1s) due to somatic mutation. The treatment outcome for DS-AMKL is more favorable than for AMKL in non-DS patients. To gain insight into gene expression differences in AMKL, we compared 24 DS and 39 non-DS AMKL samples. We found that non-DS-AMKL samples cluster in two groups, characterized by differences in expression of HOX/TALE family members. Both of these groups are distinct from DS-AMKL, independent of chromosome 21 gene expression. To explore alterations of the GATA1 transcriptome, we used cross-species comparison with genes regulated by GATA1 expression in murine erythroid precursors. Genes repressed after GATA1 induction in the murine system, most notably GATA-2, MYC, and KIT, show increased expression in DS-AMKL, suggesting that GATA1s fail to repress this class of genes. Only a subset of genes that are up-regulated upon GATA1 induction in the murine system show increased expression in DS-AMKL, including GATA1 and BACH1, a probable negative regulator of megakaryocytic differentiation located on chromosome 21. Surprisingly, expression of the chromosome 21 gene RUNX1, a known regulator of megakaryopoiesis, was not elevated in DS-AMKL. Our results identify relevant signatures for distinct AMKL entities and provide insight into gene expression changes associated with these related leukemias.

An estrogen receptor-negative breast cancer subset characterized by a hormonally regulated transcriptional program and response to androgen

Little is known of the underlying biology of estrogen receptor-negative, progesterone receptor-negative (ER(-)/PR(-)) breast cancer (BC), and few targeted therapies are available. Clinical heterogeneity of ER(-)/PR(-) tumors suggests that molecular subsets exist. We performed genome-wide expression analysis of 99 primary BC samples and eight BC cell lines in an effort to reveal distinct subsets, provide insight into their biology and potentially identify new therapeutic targets. We identified a subset of ER(-)/PR(-) tumors with paradoxical expression of genes known to be either direct targets of ER, responsive to estrogen, or typically expressed in ER(+) BC. Differentially expressed genes included SPDEF, FOXA1, XBP1, CYB5, TFF3, NAT1, APOD, ALCAM and AR (P<0.001). A classification model based on the expression signature of this tumor class identified molecularly similar BCs in an independent human BC data set and among BC cell lines (MDA-MB-453). This cell line demonstrated a proliferative response to androgen in an androgen receptor-dependent and ER-independent manner. In addition, the androgen-induced transcriptional program of MDA-MB-453 significantly overlapped the molecular signature of the unique ER(-)/PR(-) subclass of human tumors. This subset of BCs, characterized by a hormonally regulated transcriptional program and response to androgen, suggests the potential for therapeutic strategies targeting the androgen signaling pathway.

Common markers of proliferation

When normal tissue and tumour samples are compared by microarray analysis, the biggest differences most often occur in the expression levels of genes that control cell proliferation. However, this difference is detected whenever mRNA samples that are taken from two cell populations with different proliferation rates are compared. Although the exact genes that comprise this 'proliferation signature' often differ, they are almost always genes that are involved in the fundamental process of cell proliferation. Can the proliferation signature be used to improve our understanding of the cell cycle and cancer pathogenesis, as well as being used as a biomarker for cancer diagnosis and prognosis?

Application of Oligonucleotide Microarrays to Assess the Biological Effects of Neoadjuvant Imatinib Mesylate Treatment for Localized Prostate Cancer

PURPOSE

Neoadjuvant administration of antineoplastic therapies is used to rapidly assess the clinical and biological activity of novel systemic treatments. To assess the feasibility of using microarrays to assess molecular end points following targeted treatment in a heterogeneous tumor, we measured global gene expression in localized prostate cancer before and following neoadjuvant treatment with imatinib mesylate.

PATIENTS AND METHODS

Patients with intermediate-risk to high-risk prostate cancer were treated for 6 weeks with 200 to 300 mg of oral imatinib mesylate. Frozen tissue was obtained from pretreatment ultrasound-guided biopsies and posttreatment radical prostatectomy specimens. Oligonucleotide microarray analysis following laser capture microdissection (LCM) and RNA amplification was used to assess gene expression changes associated with imatinib mesylate therapy. Immunohistochemistry was used to measure protein expression of MKP1 and CD31 and to assess cellular apoptosis.

RESULTS

Of the 11 patients enrolled, high-quality microarray data was obtained from both biopsies (n = 7) and radical prostatectomy specimens (n = 9). Technically introduced intrasample gene expression variability was found to be significantly less than intertumor biological variability. Large gene expression differences were observed, and the gene with the most consistent differential expression (MKP1) was validated by immunohistochemistry. Gene set enrichment analysis suggests that imatinib mesylate therapy results in apoptosis of microvascular endothelial cells, an observation anecdotally supported by immunohistochemistry.

CONCLUSIONS

This study shows that high-quality microarray data can be generated using LCM and RNA amplification to discover potential mechanisms of targeted therapy in cancer.

Integrative Analysis of Gene Expression Data Including an Assessment of Pathway Enrichment for Predicting Prostate Cancer

Background

Microarray technology has been previously used to identify genes that are differentially expressed between tumour and normal samples in a single study, as well as in syntheses involving multiple studies. When integrating results from several Affymetrix microarray datasets, previous studies summarized probeset-level data, which may potentially lead to a loss of information available at the probe-level. In this paper, we present an approach for integrating results across studies while taking probe-level data into account. Additionally, we follow a new direction in the analysis of microarray expression data, namely to focus on the variation of expression phenotypes in predefined gene sets, such as pathways. This targeted approach can be helpful for revealing information that is not easily visible from the changes in the individual genes.

Results

We used a recently developed method to integrate Affymetrix expression data across studies. The idea is based on a probe-level based test statistic developed for testing for differentially expressed genes in individual studies. We incorporated this test statistic into a classic random-effects model for integrating data across studies. Subsequently, we used a gene set enrichment test to evaluate the significance of enriched biological pathways in the differentially expressed genes identified from the integrative analysis. We compared statistical and biological significance of the prognostic gene expression signatures and pathways identified in the probe-level model (PLM) with those in the probeset-level model (PSLM). Our integrative analysis of Affymetrix microarray data from 110 prostate cancer samples obtained from three studies reveals thousands of genes significantly correlated with tumour cell differentiation. The bioinformatics analysis, mapping these genes to the publicly available KEGG database, reveals evidence that tumour cell differentiation is significantly associated with many biological pathways. In particular, we observed that by integrating information from the insulin signalling pathway into our prediction model, we achieved better prediction of prostate cancer.

Conclusions

Our data integration methodology provides an efficient way to identify biologically sound and statistically significant pathways from gene expression data. The significant gene expression phenotypes identified in our study have the potential to characterize complex genetic alterations in prostate cancer.

The effect of GeneChip gene definitions on the microarray study of cancers

The Affymetrix GeneChip is a popular microarray platform for genome-wide expression profiling and has been widely used in functional genomics especially in the classification of cancers. Due to the updating of genome data, much of the genome information with which the chips were designed is out-of-date and it has been reported that many of the genes/transcripts on the chips differ from their original definition when mapping the probes to the new genome information. Dai et al. have reported that the updated definition can cause as much as 30-50% discrepancy in the genes selected as differentially expressed on a heart tissue expression profiling dataset. Understanding the nature of this difference is therefore very important for the utilization of the data. In this work, with a large cancer dataset as an example, we compared two major definitions and investigated their effects on classification, clustering, discovery of differentially expressed genes and gene-set-based analysis. Results show that the two definitions agree well on clustering and classification results but genes and gene sets discovered as differentially expressed or enriched can be very different. Discoveries based on the Affymetrix definition can cover most of those based on the new definition, but tend to have more false positives.

Establishment of the epithelial-specific transcriptome of normal and malignant human breast cells based on MPSS and array expression data

INTRODUCTION

Diverse microarray and sequencing technologies have been widely used to characterise the molecular changes in malignant epithelial cells in breast cancers. Such gene expression studies to identify markers and targets in tumour cells are, however, compromised by the cellular heterogeneity of solid breast tumours and by the lack of appropriate counterparts representing normal breast epithelial cells.

METHODS

Malignant neoplastic epithelial cells from primary breast cancers and luminal and myoepithelial cells isolated from normal human breast tissue were isolated by immunomagnetic separation methods. Pools of RNA from highly enriched preparations of these cell types were subjected to expression profiling using massively parallel signature sequencing (MPSS) and four different genome wide microarray platforms. Functional related transcripts of the differential tumour epithelial transcriptome were used for gene set enrichment analysis to identify enrichment of luminal and myoepithelial type genes. Clinical pathological validation of a small number of genes was performed on tissue microarrays.

RESULTS

MPSS identified 6,553 differentially expressed genes between the pool of normal luminal cells and that of primary tumours substantially enriched for epithelial cells, of which 98% were represented and 60% were confirmed by microarray profiling. Significant expression level changes between these two samples detected only by microarray technology were shown by 4,149 transcripts, resulting in a combined differential tumour epithelial transcriptome of 8,051 genes. Microarray gene signatures identified a comprehensive list of 907 and 955 transcripts whose expression differed between luminal epithelial cells and myoepithelial cells, respectively. Functional annotation and gene set enrichment analysis highlighted a group of genes related to skeletal development that were associated with the myoepithelial/basal cells and upregulated in the tumour sample. One of the most highly overexpressed genes in this category, that encoding periostin, was analysed immunohistochemically on breast cancer tissue microarrays and its expression in neoplastic cells correlated with poor outcome in a cohort of poor prognosis estrogen receptor-positive tumours.

CONCLUSION

Using highly enriched cell populations in combination with multiplatform gene expression profiling studies, a comprehensive analysis of molecular changes between the normal and malignant breast tissue was established. This study provides a basis for the identification of novel and potentially important targets for diagnosis, prognosis and therapy in breast cancer.