Sources of variation in baseline gene expression levels from toxicogenomics study control animals across multiple laboratories

Michael Boedigheimer, Russell D. Wolfinger, Michael Bass, Pierre R. Bushel, Jeff W. Chou, Matthew N. Cooper, J. Christopher Corton, Jennifer Fostel, Susan Hester, Janice S. Lee, Fenglong Liu, Jie Liu, Hui-Rong Qian, John Quackenbush, Syril Pettit, Karol L. ThompsonBMC 2008 · DOI

Abstract

The study factors that emerged as key sources of variability included gender, organ section, strain, and fasting state. These and other study factors were identified as key descriptors that should be included in the minimal information about a toxicogenomics study needed for interpretation of results by an independent source. Genes that are the most and least variable, gender-selective, or altered by fasting were also identified and functionally categorized. Better characterization of gene expression variability in control animals will aid in the design of toxicogenomics studies and in the interpretation of their results.

Introduction: The Baseline Noise Problem

In toxicogenomics, researchers measure changes in gene expression in animal models to predict human toxicity. To distinguish a chemical's effect from natural biological variation, a reliable baseline from untreated "control" animals is essential. However, baseline gene expression is not static; it fluctuates due to genetics, environmental conditions (like diet and fasting), and technical factors. Without characterizing this "noise," researchers risk misattributing natural variations to chemical toxicity, leading to false positives and irreproducible results.

Methods and Datasets

The researchers conducted a meta-analysis of 536 Affymetrix microarrays (483 high-quality arrays post-QC) from control arms across 16 institutions.

  • Tissues: Rat liver (n=396) and kidney (n=140).
  • Strains: Sprague-Dawley, Wistar, and F344/N.
  • Analysis: Data was normalized using Robust Multi-array Average (RMA). Principal Component Analysis (PCA), Hotelling-Lawley (HL) statistics, and Variance Components (VC) models were used to partition variability among 38 study factors.

Key Results: Primary Drivers of Variance

Statistical analysis revealed that gender, fasting state, rat strain, and organ section are the most significant sources of baseline variability. For instance, in kidney samples, the specific section (cortex vs. medulla) creates distinct expression profiles that can overshadow other factors.

PCA plot illustrating the distinct clustering of gene expression profiles based on kidney organ sections (Cortex vs. Medulla).

The team identified hundreds of gender-selective genes (854 in liver, 863 in kidney) and isolated 163 "housekeeping" genes with consistently low variance across all conditions, validating their use as reliable internal controls in mammalian expression studies.

Key Insights: Fasting and Toxicity Targets

Fasting as a Major Confounder

Fasting state, even overnight, dramatically alters gene expression, particularly in pathways related to lipid utilization. This factor alone can make cross-laboratory comparisons nearly impossible if protocols are not standardized.

PCA plot demonstrating the impact of fasting state and laboratory site on baseline gene expression in liver samples.

High Variance Genes and Toxicity

Crucially, genes with the highest natural baseline variance are often those involved in drug metabolism and defense responses—the very targets of toxicological studies. Approximately 20% of genes linked to kidney injury naturally exhibit high baseline noise, emphasizing the need for caution when interpreting these as biomarkers in small-sample studies.

Limitations and Future Recommendations

A limitation of this meta-analysis is the confounding of variables; specific diets or fasting protocols were often unique to certain labs, making it difficult to isolate "lab effect" from "diet effect."

The authors recommend that gender, strain, diet, and tissue section be adopted as mandatory "key descriptors" in toxicogenomics reporting standards. Better characterization of this variability is vital for improving study design and the accuracy of toxicological interpretations.

Summary updated

Abstract

The study factors that emerged as key sources of variability included gender, organ section, strain, and fasting state. These and other study factors were identified as key descriptors that should be included in the minimal information about a toxicogenomics study needed for interpretation of results by an independent source. Genes that are the most and least variable, gender-selective, or altered by fasting were also identified and functionally categorized. Better characterization of gene expression variability in control animals will aid in the design of toxicogenomics studies and in the interpretation of their results.
BibTeX
@inproceedings{boedigheimer2008sources,
  author = {Michael Boedigheimer and Russell D. Wolfinger and Michael Bass and Pierre R. Bushel and Jeff W. Chou and Matthew N. Cooper and J. Christopher Corton and Jennifer Fostel and Susan Hester and Janice S. Lee and Fenglong Liu and Jie Liu and Hui-Rong Qian and John Quackenbush and Syril Pettit and Karol L. Thompson},
  title = {{Sources of variation in baseline gene expression levels from toxicogenomics study control animals across multiple laboratories}},
  booktitle = {BMC},
  year = {2008},
  doi = {10.1186/1471-2164-9-285},
}
Figure