Quantitative analysis of gene expression array data from TCDD treated mouse hepatoma cells. PM Saama, MR Fielden and TR Zacharewski.

cDNA array technology represents an efficient means to analyze the expression profile of hundreds of genes. However, many current methodologies are not reproducible and appropriate quantitative analysis has not been described. Using a commercially available cDNA array, gene expression profiles of mouse Hepa 1c1c7 cells were compared following treatment with DMSO (solvent) or 100 nM 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD). Hybridization signal intensities (HSI) from two replicate experiments were quantitated on a PhosphoImager. Results from conventional non-parametric methods of analysis and a more powerful parametric approach using a General Linear Model (GLM) were then compared. The non-parametric analysis of HSI data used global means of the background corrected HSI values to normalize each data point for direct comparison. Treatment-related differences in gene expression were then ranked for each replicate. Rank correlation between replicates was 0.23 (P <.05). Using a GLM, the variation in HSI was partitioned into treatment, replicate, group, and interactions between them as classification effects and random effects of genes within group. The model was analyzed using the GLM procedure of SAS. Least Squares Means (LSM) for the genes within group were computed using Tukey adjustments. Genes were ranked according to the magnitude of the LSM. This model accounted for 94% of the variation in HSI. All of the effects in the model were significant (P < .0001). The two methods gave different rankings for the genes. The probability of observing a treatment effect on each gene could not be calculated using the non-parametric analysis. However, the GLM approach provided a mechanism for hypothesis testing on constitutive genes which was more consistent with the experimental design.