To eliminate non-biological variation, gene expression analysis involving qRT-PCR requires stringent normalization strategies. Among the several approaches proposed, use of reference genes is currently the preferred way of normalization . However, the use of improper reference genes is known to lead to erroneous results . Importantly, the studies of the expression levels for the reference gene themselves, particularly, for GADPH and ACTB, showed considerable variation in different tissues and experimental conditions. Specifically in omental and subcutaneous fat depots, a variation in expression of these reference genes was found to be dependent on the presence of obesity and type 2 diabetes mellitus (T2D) . These findings necessitate the need to validate reference genes for studies of human visceral adipose samples.
Scientifically, the validation of reference genes presents a circular problem: assessing stability of expression of a given gene cannot be achieved without using another gene as a reference. Several algorithms have been proposed to address this conundrum . GeNorm software  is one of the most popular algorithms for validating candidate reference genes with low variability. It utilizes two parameters to quantify the reference gene expression stability: M (average expression stability) and V (pairwise variation). A low M value is indicative of a more stable expression, hence, increasing the suitability of a particular gene as a reference gene. Another feature of GeNorm is that it does not require a normal distribution of data. However, co-regulation of candidate genes does seem to influence the efficiency of this algorithm due to the use of pair-wise comparisons. To minimize this risk, the eight candidate reference genes selected for this analysis were chosen on the basis of the difference in their physiological functions-cytoskeleton (ACTB), carbohydrate metabolism (GAPDH), signaling pathways (YWHAZ), transcription (RPII or POLR2A), metabolic salvaging of nucleotides (HPRT1), protein synthesis (18S) and protein degradation (UBC).
The GeNorm algorithm determines expression stability (M) via a pair-wise comparison of one candidate reference gene and all other candidate genes independent of the level of gene expression for each sample. An identical expression ratio of two reference genes in all samples is used as an indicator of expression stability. Thus GeNorm analysis is independent of variation in amount of starting material between samples. According to this analysis, ACTB and RPII represented the best combination of reference genes for visceral adipose tissue among lean and obese patients (Figure 1a), while HPRT1 and GAPDH were ranked third and fourth, respectively. After completion of this step, a pairwise variation (V) was calculated between two sequential normalization factors (NFn and NFn+1) for all genes. A large variation indicates that the added gene has a significant effect and should be included for calculation for a reliable normalization factor. Figure 2a, b show that further inclusion of additional reference genes did not significantly increase the pair-wise variation and that the use of two reference genes is sufficient for accurate normalization. An advantage to the GeNorm algorithm is that it is minimally affected by expression intensity of the candidate genes. In addition, since the approach is based on multiple pair-wise comparisons, the need for large sample size is mitigated. This was reinforced by the observation that increase in sample size did not dramatically alter the final results. On increasing sample size from 9 to 19, ACTB and RPII were again found to be the most stable genes with GAPDH and HPRT1 being ranked the next two best genes (Figure 1b).
Housekeeping genes, in addition to their basic functions, exert pleiotropic effects on other cellular systems, decreasing the value of the function-based predictions of co-regulation. To overcome this problem Anderson et al., proposed a model based approach incorporated into the software NormFinder. This algorithm ranks candidate reference genes according to the least estimated intra and inter group variation, which serves as an effective method to overcome the influence of co-regulation. Although NormFinder takes into account the heterogeneity in the tested samples, and attempts to distinguish between stability and bias, this model-based approach is self-restricted by the importance it places on overall expression intensity of each candidate gene. A close inspection of the analysis of the results produced by NormFinder showed that it biased towards candidate reference genes that have overall similar expression values (in terms of Ct). Consequently, the robustness of this method is linked to the sample size.
The ultimate objective of NormFinder is to identify candidate reference genes(s) with an inter group variation as close to zero as possible, while at the same time having small intra-group variation. When the genes were ranked solely by their stability values, GAPDH and HPRT1 appeared to be the best combination of endogenous controls (Table 1). Further examination of the results reveals that although UBC was ranked third, its intra-group variation was large (Figure 3a), therefore, UBC gene was eliminated from further consideration leaving the next most stable reference genes: ACTB, B2M and RPII.
The same genes, GAPDH, HPRT1, ACTB and RPII, were ranked as the most stable both by GeNorm and NormFinder softwares. However, the best combinations of two genes proposed by these two algorithms were different. This variation was expected based on the vastly different approaches used by each of the analysis softwares and dependence of robustness of NormFinder on sample size. In GeNorm, gene expression stability (M) based on the expression ratio of the two genes (pairwise comparison) is the most important criteria for evaluating a reference gene, while NormFinder focuses on genes with the least intra and inter group variations. Thus, in cases when two genes show high expression variation while their ratio (M) remains unchanged, there will be discordance in ranking by the two algorithms.
Further, NormFinder gains in robustness as the number of samples is increased. This was confirmed by increasing sample size to 19. Ranking of genes from most stable to least stable revealed - ACTB and RPII as the best combination of genes and this was in accordance with with GeNorm results (Figure 3b).
In contrast to the previous study by Catalan et al., in visceral adipose tissue samples both the algorithms highlighted 18S as one of the least stable gene. This was not unexpected, as several arguments against the use of rRNA as reference genes have been previously put forth. The strongest argument against its use in real time RT-PCR data analysis is its high abundance compared to other target mRNA which hinders accurate subtraction from the baseline value .
In order to compare the GeNorm and NormFinder results with an independent ranking method, the data was also analyzed with the BestKeeper tool . In this approach, ideal reference genes are expected to have stable expression, indicated by low variation in the tissue under consideration . With BestKeeper, stability (SD) and relationship to the BestKeeper index (r and p values) are the two most important criteria for evaluating the stability of reference genes. This algorithm uses a pair-wise correlation analysis for all pairs of candidate genes based on the raw Ct values and calculates the geometric mean of the best suited ones. Based on low CV and high coefficient of correlation (r) to the BestKeeper index, ACTB and RPII followed by GAPDH and HPRT1 were ranked as the top four genes (Table 3). High correlation coefficient is an indicator of stable expression of the reference genes in visceral adipose tissue. Again, 18S was ranked as the least stable and excluded from further analysis. Robustness of the algorithm was assessed by increasing sample size (n = 19). Sample size was found to have minimal effect on the results. The same two genes ACTB and RPII were identified as the most stable followed by YWHAZ and GAPDH (Table 4).
Overall, the BestKeeper results were in line with the NormFinder data, and with minor differences, the GeNorm data, indicating the reliability of the validation for reference genes in the present study (Table 2). Regardless of the algorithm used, all three software ranked the same set of genes as the most stable.
In conclusion, we recommend ACTB and RPII as stable reference genes most suitable for gene expression studies of human visceral adipose tissue. The use of these genes as a reference pair may further enhance the robustness of qRT-PCR in this model system.