Lynn M. Crosby

Normalization and analysis of DNA microarray data by self-consistency and local regression

BackgroundWith the advent of DNA hybridization microarrays comes the remarkable ability, in principle, to simultaneously monitor the expression levels of thousands of genes. The quantiative comparison of two or more microarrays can reveal, for example, the distinct patterns of gene expression that define different cellular phenotypes or the genes induced in the cellular response to insult or changing environmental conditions. Normalization of the measured intensities is a prerequisite of such comparisons, and indeed, of any statistical analysis, yet insufficient attention has been paid to its systematic study. The most straightforward normalization techniques in use rest on the implicit assumption of linear response between true expression level and output intensity. We find that these assumptions are not generally met, and that these simple methods can be improved.ResultsWe have developed a robust semi-parametric normalization technique based on the assumption that the large majority of genes will not have their relative expression levels changed from one treatment group to the next, and on the assumption that departures of the response from linearity are small and slowly varying. We use local regression to estimate the normalized expression levels as well as the expression level-dependent error variance.ConclusionsWe illustrate the use of this technique in a comparison of the expression profiles of cultured rat mesothelioma cells under control and under treatment with potassium bromate, validated using quantitative PCR on a selected set of genes. We tested the method using data simulated under various error models and find that it performs well.

Application of cDNA Microarray Technology to In Vitro Toxicology and the Selection of Genes for a Real-Time RT-PCR-Based Screen for Oxidative Stress in Hep-G2 Cells

Large-scale analysis of gene expression using cDNA microarrays promises the rapid detection of the mode of toxicity for drugs and other chemicals. cDNA microarrays were used to examine chemically induced alterations of gene expression in HepG2 cells exposed to a diverse group of toxicants at an equitoxic exposure concentration. The treatments were ouabain (43 μM), lauryl sulfate (260 μ M), dimethylsulfoxide (1.28 M), cycloheximide (62.5 μM), tolbutamide (12.8 mM), sodium fluoride (3 mM), diethyl maleate (1.25 mM), buthionine sulfoximine (30 mM), potassium bromate (2.5 mM), sodium selenite (30 μM), alloxan (130 mM), adriamycin (40 μM), hydrogen peroxide (4 mM), and heat stress (45•C × 30 minutes). Patterns of gene expression were correlated with morphologic and biochemical indicators of toxicity. Gene expression responses were characteristically different for each treatment. Patterns of expression were consistent with cell cycle arrest, DNA damage, diminished protein synthesis, and oxidative stress. Based upon these results, we concluded that gene expression changes provide auseful indicator of oxidative stress, as assessed by the GSH:GSSG ratio. Under the conditions of this cell culture test system, oxidative stress upregulated 5 genes, HMOX1, p21waf1/cip1, GCLM, GR, TXNR1 while downregulating CYP1A1 and TOPO2A. Primers and probes for these genes were incorporated into the design of a 7-gene plate for RT-PCR. The plate design permitted statistical analysis and allowed clear discrimination between chemicals inducing oxidative vs nonoxidative stress. A simple oxidative stress score (0—1), based on the responses by the 7 genes (including p-value) on the RT-PCR plate, was correlated with the GSH:GSSG ratio using linear regression and ranking (Pearson product) procedures. These analyses yielded correlation coefficients of 0.74 and 0.87, respectively, for the treatments tested (when 1 outlier was excluded), indicating a good correlation between the biochemical and transcriptional measures of oxidative stress. We conclude that it is essential to measure the mechanism of interest directly in the test system being used when assessing gene expression as a tool for toxicology. Tables 1—15, referenced in this paper, are not printed in this issue of Toxicologic Pathology. They are available as downloadable text files at http://taylorandfrancis.metapress.com/openurl.asp?genre=journal&issn=0192-6233. To access them, click on the issue link for 30(4), then select this article. A download option appears at the bottom of this abstract. In order to access the full article online, you must either have an individual subscription or a member subscription accessed through www.toxpath.org.

Time- and Dose-Dependent Development of Potassium Bromate-Induced Tumors in Male Fischer 344 Rats

Potassium bromate (KBrO 3) is a rodent carcinogen and a nephro- and neurotoxicant in humans. KBrO3 is used in cosmetics and food products and is a by-product of water disinfection by ozonization. KBrO3 is carcinogenic in the rat kidney, thyroid, and mesothelium and is a renal carcinogen in the male mouse. The present study was designed to investigate the relationship of time and dose to bromate-induced tumors in male Fischer 344 (F344) rats and to provide some insight into the development of these tumors. KBrO 3 was dissolved in drinking water at nominal concentrations of 0, 0.02, 0.1, 0.2, and 0.4 g/L and administered to male F344 rats as the sole water source for 12, 26, 52, 78, or 100 wk. Renal cell tumors were present after 52 wk of treatment only in the high-dose group. Mesotheliomas developed after 52 wk of treatment on the tunica vaginalis. Mesotheliomas were present at sites other than the testicle after 78 wk of treatment, indicating that their origin was the testicular tunic. Thyroid follicular tumors were present as early as 26 wk in 1 rat each from the 0.1- and 0.2-g/L groups. The present study can be used as a basis for the determination of dose-time relationships of tumor development for a better understanding of KBrO3-induced cancer.

Toxicogenomics, Drug Discovery, and the Pathologist

The field of toxicogenomics, which currently focuses on the application of large-scale differential gene expression (DGE) data to toxicology, is starting to influence drug discovery and development in the pharmaceutical industry. Toxicological pathologists, who play key roles in the development of therapeutic agents, have much to contribute to DGE studies, especially in the experimental design and interpretation phases. The intelligent application of DGE to drug discovery can reveal the potential for both desired (therapeutic) and undesired (toxic) responses. The pathologists understanding of anatomic, physiologic, biochemical, immune, and other underlying factors that drive mechanisms of tissue responses to noxious agents turns a bewildering array of gene expression data into focused research programs. The latter process is critical for the successful application of DGE to toxicology. Pattern recognition is a useful first step, but mechanistically based DGE interpretation is where the long-term future of these new technologies lies. Pathologists trained to carry out such interpretations will become important members of the research teams needed to successfully apply these technologies to drug discovery and safety assessment. As a pathologist using DGE, you will need to learn to read DGE data in the same way you learned to read glass slides, patiently and with a desire to learn and, later, to teach. In return, you will gain a greater depth of understanding of cell and tissue function, both in health and disease.

Origin and Distribution of Potassium Bromate-Induced Testicular and Peritoneal Mesotheliomas in Rats

Tissue sections were examined from a 2-year bioassay of male Fischer 344 rats treated with potassium bromate administered in drinking water. All animals exhibiting peritoneal mesotheliomas also had mesotheliomas of the tunica vaginalis testis mesorchium (the reverse was not true), and the correlation of these 2 types of mesotheliomas was highly significant (r2 = 0.98). Mapping of the tunica vaginalis tumors at all time points and at all bromate concentrations revealed a pattern of increasing incidence of tumor formation on the mesothelium of the tunica vaginalis testis as a function of proximity to the mesorchial ligament. Thus, the mesorchium appears to be the major mesothelial target site for potassium bromate-mediated carcinogenesis. The frequency of occurrence of mesotheliomas by location was tunica vaginalis testis (25%), mesosplenium (20%), mesentery (10%), mesojejunum/mesocolon (8%), bladder (6.5%), mesogastrium (13%), liver serosa (5%), and kidney, small intestine, and rectum (1% each). A complete cross-section of the rat testis was prepared and used to construct a complete map of the mesothelium. Any attempt to determine the role of local dose and tissue susceptibility for the purpose of interspecies risk extrapolation must take into account the complex anatomy and physiology of this region of the visceral and testicular suspensory apparatus. Improved histologic approaches are needed for adequate assessment of this delicate suspensory system.

Complementary Roles for Toxicologic Pathology and Mathematics in Toxicogenomics, With Special Reference to Data Interpretation and Oscillatory Dynamics

Toxicogenomics is an emerging multidisciplinary science that will profoundly impact the practice of toxicology. New generations of biologists, using evolving toxicogenomics tools, will generate massive data sets in need of interpretation. Mathematical tools are necessary to cluster and otherwise find meaningful structure in such data. The linking of this structure to gene functions and disease processes, and finally the generation of useful data interpretation remains a significant challenge. The training and background of pathologists make them ideally suited to contribute to the field of toxicogenomics, from experimental design to data interpretation. Toxicologic pathology, a discipline based on pattern recognition, requires familiarity with the dynamics of disease processes and interactions between organs, tissues, and cell populations. Optimal involvement of toxicologic pathologists in toxicogenomics requires that they communicate effectively with the many other scientists critical for the effective application of this complex discipline to societal problems. As noted by Petricoin III et al (Nature Genetics 32, 474-479, 2002), cooperation among regulators, sponsors and experts will be essential for realizing the potential of microarrays for public health. Following a brief introduction to the role of mathematics in toxicogenomics, “data interpretation” from the perspective of a pathologist is briefly discussed. Based on oscillatory behavior in the liver, the importance of an understanding of mathematics is addressed, and an approach to learning mathematics “later in life” is provided. An understanding of pathology by mathematicians involved in toxicogenomics is equally critical, as both mathematics and pathology are essential for transforming toxicogenomics data sets into useful knowledge.

Toxicogenomics and Human Disease Risk Assessment

Complete sequencing of human and other genomes, availability of large-scale gene expression arrays with ever-increasing numbers of genes displayed, and steady improvements in protein expression technology can have a great impact on the field of toxicology. However, we are a long way from devising effective standards for human risk assessments based upon these technologies. Current impediments to effective application of these technologies include appropriate normalization procedures (as “there is no fixed point in transcript space”), confirmation of data quality and demonstration of the functional significance of responses observed. Providing risk assessors with statistically and functionally unconfirmed, large-scale gene expression data sets that generally defy interpretation is not an appropriate approach. We propose that a logical process of data generation be developed, with risk assessment in mind from the outset. The basic principles of toxicology should be applied to selection of experimental systems, dose and duration of exposure, along with appropriate statistical analyses and biological interpretation. If mechanistically based interspecies extrapolation of risk is to be undertaken, suitable biochemical or other follow-up studies should be completed to confirm functional significance of transcriptional changes.