Human heart tissue specimens are very difficult to acquire. However, we managed to obtain 35 samples from the left ventricles of 35 organ donors of different age, weights and sex, whose hearts could not be used for transplant purposes for various reasons (see Methods). Thus, we can assume that the number and type of samples used was sufficiently high and diverse to ensure their representativity and randomness. Reports of the use of qRT-PCR in human heart tissue are scarce because of the technical difficulties involved in obtaining such samples. However, given that qRT-PCR is especially suitable for determining gene expression in small pieces of tissue, we tried to establish guidelines for accurate data normalisation intended for human heart studies based on qRT-PCR.
Our technical procedure proved to be a valid method of quantifying gene expression, rendering high correlation coefficients (R2) and robust PCR efficiencies.
Through geNorm analysis, PPIA, RPLP and GADPH appeared as the most stable genes and TBP the least stable (Figure 2). It has been reported that in over 90% of cases, gene expression data are normalised using GADPH, ACTB, 18S rRNA or 28S rRNA as single control genes . However, several studies have shown that these reference genes undergo variation according to the experimental conditions, treatment and cell cycle stage [11, 12].
In human adipocytes and preadipocytes, ACTB and 18S rRNA gene expression levels change under hormonal stimulation . Moreover, it has been reported that ACTB is unsuitable as a control for gene expression analysis in interstitial cells derived from sheep heart valves . The 18S rRNA gene has been considered an ideal internal control in qRT-PCR analysis. However, ribosomal RNA accounts for up to 80–90% of total cellular RNA, and several studies have shown that rRNA varies less under conditions that affect the expression of mRNAs  but that possible imbalances in rRNA and mRNA fractions between different samples makes genes encoding ribosomal RNAs unsuitable as references . As far as we are aware, no previous study has tried to identify adequate housekeeping genes for use in human heart tissue. Morgan et al.  examined the expression of genes related to the renin-angiotensin system in human atrial tissue using GADPH as an endogenous control. Other authors  have used RNA extracted from human endomyocardial biopsies and isolated cardiomyocytes for real-time RT-PCR using GADPH, HPRT and the oncogene ABL as housekeeping genes. These endogenous controls showed very low variation in individual gene expression levels across cardiac pathologies, suggesting their suitable use as reference genes for quantitative PCR studies in cardiac tissue. In these previous studies, the specific testing of several candidate reference genes to determine the most suitable reference for use in cardiac tissue was not reported [15, 16]. In contrast, Radonic et al.  determined transcription levels of several housekeeping genes in different human tissues, including heart, and identified TBP as the gene with the lowest range of RNA transcription across tissues. This finding is in agreement with the present results.
We found PPIA, RPLP plus GADPH to be a reliable set of genes for normalising data (Figure 3) according to the geNorm programme. As reported by Vandesompele et al. , geNorm proposes a pairwise variation of 0.15 as the cut-off under which the inclusion of an additional control gene is not required. Using our set of candidate genes, this would mean that adding a fifth gene to the four most stable genes identified would really provide the best results. Notwithstanding, this cut-off of 0.15 should not be considered in a strict sense, but rather as a reference to determine the optimal number of housekeeping genes. Sometimes the observed trend can be equally informative, and using the three best reference genes is, in most cases, a valid strategy for much more accurate and reliable normalisation compared to the use of a single housekeeping gene.
Using NormFinder software  as another tool to validate the expression stability of the ten candidate reference genes; GADPH, PPIA and RPLP also achieved the best stability values. While geNorm detected the two genes whose expression ratios showed least variation from those of the other genes tested, NormFinder was able to identify the single gene with the most stable expression. Hence, the most stable candidate gene was found to be GADPH, and the least stable TBP. Using this programme, we obtained the same results as with GeNorm except for the rank position ascribed to GADPH as the most stable single gene. However, the least stable genes and most stable ones identified by GeNorm and Normfinder were generally well-matched.
As a limitation to our study, we should mention that although we established the purity of our RNA samples, due to the amount of RNA available, we could not run electrophoretic tests to check RNA integrity.
Finally, the set of reference genes determined here as the best endogenous controls to be used in qRT-PCR studies in heart tissue has applications in developing cardiovascular diagnostic tests and therapeutic strategies that will substantially improve human health .