The overall variance, a basis of a simple, yet highly effective, tool for the identification of ‘reference genes’ , was calculated for the normalized C values of each of the 3 technical replicates.The overall variances were compared across the 16 ‘reference genes’ to identify optimal RT-PCR controls with the most stable gene expression (smallest overall variance) across all kidney samples (fig. For example, 18S r RNA and Hmbs showed the largest variation for all 3 replicates, indicating that these genes represented suboptimal endogenous controls in this experimental setup.Even more accurate RT-PCR controls could be obtained by accounting for the expression of multiple genes, e.g.using a method of a geometric mean that is the basis of the Ge Norm  and Best Keeper programs .In the case of renal tissues, 18S r RNA and the cyclophilin A-encoding gene Ppia (but not Gapdh) have been recommended as the preferred RT-PCR controls for studies of the renal tubulointerstitial compartment, based on analyses of these 3 genes in renal biopsies .In the current study, we used Taq Man® and Affymetrix Gene Chip® assays to compare the expression stability of 16 commonly used RT-PCR controls in whole kidneys affected by mild and severe cystic kidney disease from an extensively characterized animal model [13, 14]. The genome-wide expression data were generated previously with 14 Affymetrix Gene Chip® Mouse Genome 430 2.0 arrays (Affymetrix Inc., Santa Clara, Calif., USA; 7 biological replicates for mild and 7 for severe cystic kidney phenotype) .In contrast, Ppia, Gapdh and Pgk1 showed the smallest variation and would serve as the more appropriate endogenous internal RT-PCR controls.However, the overall variance method does not separate for the between-group variance (i.e.
The probe sets were excluded from further analyses if they matched alternative transcripts or represented ‘rare EST events’ according to the UCSC Genome Browser. The p values were derived based on the normal distribution assumption.Comparisons of data generated by these 2 distinct gene expression platforms allowed an independent validation of these 2 approaches.First, normalization of the Taq Man®-based RT-PCR data was performed for all samples by subtracting the mean of the C values, this approach successfully reduced variances in the expression of all studied genes, although to various extents (fig. Then, we ranked the appropriateness of the 16 ‘reference genes’ as endogenous internal RT-PCR controls based on their overall variance, with smaller variance indicating higher gene expression stability.RNA and c DNA were prepared previously  from whole kidneys harvested from 7 mildly and 7 severely affected 10-day-old F2 mice generated in a (C57BL/6J-cpk/ × CAST)F1 intercross (cystic disease severity was defined by kidney length, weight and volume; e.g. Large-scale validation of these gene expression data with 14 low-density Affymetrix U74Av2 arrays provided a formal technical validation of the 430 2.0 data (gene expression correlated strongly across these 2 array platforms; r = 0.72 for all genes, r = 0.90 for differentially expressed genes with p values was performed by subtracting the sample mean from each value.average kidney volume was more than 8 times higher among the 7 highly vs. These 14 samples were used in triplicates for gene expression analyses using 16 Taq Man assays arranged into a mouse endogenous control low-density array [Applied Biosystems, Foster City, Calif., USA (online suppl. This mean-centering adjustment removes potential RNA loading differences from sample to sample.In contrast to the previously utilized 2-fold change threshold for the equivalence test , we used a more stringent 1.5-fold change threshold since a 2-fold change is often assumed to be differentially expressed and not appropriate to consider as a ‘reference gene’.The overall variance analysis was conducted on the same data in the same fashion as the equivalence test except that the overall variances of all samples were computed for each gene.For the microarray data, each probe set was analyzed separately.The median p value from each gene was used for overall gene ranking.Their use, however, relies on the premise that these genes are expressed at consistently stable levels across all experimental conditions under investigation.Unfortunately, none of the endogenous controls was found to be constantly expressed across different tissues, developmental stages, or pathological and study conditions [1, 2].