Analysis of the raw expression levels across all samples identified some variation amongst reference genes (Figure 1). Quantification cycle (C_q) 36 values for the 11 genes studied ranged from 8.2 to 30.9, while the majority of these values were between 18.6 and 24.6. The gene encoding 18S rRNA was highly expressed compared to the protein coding genes, reaching threshold fluorescence after only 8.2 amplification cycles, whereas the C_q average of all reference genes within the datasets was approximately 20.7 cycles. As a result, the 18S rRNA transcript levels were around 5700-fold more abundant than the dataset's average. The least abundant transcripts were PLA2 and RPL13, with C_q values of 27.6 and 30.9, respectively. The individual reference gene had different expression ranges across all studied RNA samples. ACT, TUA and TUB showed smaller gene expression variation (below 7 cycles) among studied reference genes, while PLA2, CYP2 and RPL13 had much higher expression variations (above 10 cycles). The wide expression ranges of the eleven tested reference genes confirmed that no single reference gene had a constant expression in different peach samples. Therefore, it is of utmost importance to select a reliable reference gene to normalize gene expression under a certain condition. Due to low expressed genes where C_qs were obtained around cycles 30–35 can lead to large variability 34, the candidate RPL13 was discarded from subsequent tests.

Gene expression stability (M) of these ten reference genes studied was calculated using the software geNorm 3337. The program is a Visual Basic application tool for Microsoft Excel and relies on the principle that the expression ratio of two perfect reference genes should be constant throughout the different experimental conditions or cell types. The M value is defined as the average pair-wise variation of a certain gene with all other tested reference genes, whereas the variation of this certain reference gene to another is determined as the standard deviation of the log2-transformed expression level ratios. The average expression M values of the 10 reference genes were plotted in Figure 2. The gene with the lowest M value is considered as the most stable expression, while the highest M value has the least stable expression. When all the samples were taken together, as shown in Figure 2A, the average expression stability value (M) of TEF2 and UBQ10 was the lowest, and that of CYP2 was the highest, indicating that TEF2 and UBQ10 had the most stable expression and that CYP2 was expressed most variably. TEF2 and UBQ10 were still the most stable genes, while 18S rRNA was the one with the highest M value, suggesting that it was variably expressed in fruit developmental stages (Figure 2B). In the different genotype samples, TEF2 and RP II were the most stable genes, while 18S rRNA was the least stable one (Figure 2C). The results remained very similar in the different storage time series, with the lowest M value for RP II and TEF2 and the highest M value for 18S rRNA (Figure 2D). TEF2 and RP II were still expressed much more stably than the other reference genes in different exogenous regulator treatments, while CYP2 was the least stable reference gene as in all samples (Figure 2E). CYP2 and TEF2 were two best genes among the ten tested reference genes, while 18S rRNA was the most variable one in different tissue samples of peach (Figure 2F).

Although most authors agree in using only one single gene as an internal control for normalization, it has been suggested that the use of two or more reference genes for RT-qPCR studies might generate more reliable results 3338. To evaluate the optimal number of genes required for accurate normalization, pairwise variations Vn/Vn+1 between consecutively ranked normalization factors are calculated to determine the effect of adding the next reference gene in normalization. The normalization factors are defined by calculating the geometric mean of the 3 most stable gene relative quantities and stepwise inclusion of the other genes in the order of their expression stability. A large pairwise variation implies that the added gene has a significant effect on normalization and should be included for calculation of a reliable normalization factor 33. As shown in Figure 3(B, C, D, E, F), the inclusion of the third reference gene did not contribute significantly to the variation of the normalization factor (V_3/4 < 0.15). Based on the cut-off value of 0.15 proposed by geNorm program, below which the inclusion of an additional reference gene is not required, so the two most stable reference genes of each series subset would be sufficient for accurate normalization. When all the samples were taken together, the pairwise variation V_2/3 was 0.213, higher than 0.15, while V_3/4 was 0.146 (Figure 3A), indicating that the addition of the third reference gene was necessary to normalize gene expression. The 3 reference genes were TEF2, UBQ10 and TUB for this group samples. The recommended combinations of control genes of each sample series had mean stability values, M ≤ 1.0 and M ≤ 0.5, which are acceptable for heterogeneous and homogeneous sample panels, respectively 39. Important to note is that the pairwise variation V and mean M values, calculated by the method of Exposito-Rodriguez et al. 27, for the combinations of TEF2/UBQ10/RP II and TEF2/RP II in all the sample pools and the series of fruit developmental and different tissue samples respectively, were all inside the ranges described as above (data not shown).

NormFinder, another VBA applet, uses a model-based approach for identifying the optimal normalization gene(s) among a set of candidates. More stable gene expression is indicated by lower average expression stability values. In this mathematical model, estimation of both intra- and inter-group variation and a separate analysis of the sample subgroups in expression levels are included into the calculation of a gene expression stability value 35. In this sense, five sample-subgroups were established as geNorm analysis. Moreover, expression data were also combined into "vegetative" (stems, roots, leaves, flowers and fruit at different developmental stages) and "mature" (including fruit obtained from different exogenous regulator treatments and two cultivars at different storage time) sample-subgroups. At the same time, all samples with no subgroups and the other five series were analyzed using this approach as well. The results of the NormFinder analysis were shown in Table 3 and Additional file 1. It is noteworthy that definition of sample-subgroups had a notable effect on NormFinder output. However, the NormFinder output with different sample-subgroups and no subgroups exhibited two common features: 1) TEF2, RP II and UBQ10 showed a remarkable stability of their expression levels and were always classified among the top four positions; 2) GAPDH, PLA2, ACT and 18S rRNA exhibited unstable expression profiles and were always included among the least stable reference genes. When it came to the other sample series, TEF2 was calculated to be the most stable single gene with a stability value of 0.007, while 18S rRNA was the most variable one in different exogenous regulator treatments. The results were broadly similar to the series of different tissues, with the highest stability value for TEF2 and lowest stability value for 18S rRNA. RP II was the most reliable gene in fruit developmental and different storage time series, while 18S rRNA remained the most variable one. In different genotype samples, TUB was identified as the most stable gene and PLA2 was the least one, with stability values of 0.002 and 0.053, respectively.

BestKeeper, an Excel-based tool, estimates inter-gene relations of possible reference gene pairs by performing numerous pairwise correlation analyses using raw C_q values of each gene. More important, all genes may be included in the calculation of the BestKeeper index, which can be used to rank the best reference genes because of stable reference genes showing a strong correlation with the BestKeeper index 34. The results of the method analysis of the same data set were presented in Table 4 and Additional file 2. The 10 reference genes studied in our analysis correlated well one with another, if also compared with the BestKeeper index, except for 18S rRNA in different tissue samples. Particularly strong inter-gene correlations were found for the four most stable reference genes in all the sample pools (r > 0.89), especially in the other five series (r > 0.95). The high Pearson's coefficients of correlation indicated that these gene pairs had very similar overall expression patterns. When the complete data set was analyzed, TEF2, UBQ10, and RP II had strong correlation with the BestKeeper index (r > 0.95), and ranked among the top four genes, in accordance with the corresponding NormFinder output, thus identifying these three genes as the most reliable reference genes for normalization. The result was identical to the series of different tissues, different genotypes, different regulator treatments and fruit developmental stages, because of TEF2, UBQ10, and RP II were still included among the 4 top-ranked reference genes. In different storage time series, only TEF2 and RP II were classified among the top four genes. GAPDH, PLA2, and 18S rRNA consistently ranked poorly in the six series, and were identified as the least reliable reference genes.

Tissues of Yuhua 1, a melting flesh peach genotype, were sampled from 8-year-old trees growing in national germplasm orchard of Institute of Horticulture of Jiangsu Academy of Agricultural Sciences, Nanjing, Jiangsu, China. Vegetative tissue samples, such as root, leaf and stem, were taken from young tissue; flowers were harvested at full bloom; and fruit at different developmental stages were taken at 2-week intervals after anthesis over the growing season. At each sampling time, plant materials except for fruits were frozen in dry ice after immediately harvesting transported to the laboratory at Nanjing Agricultural University and then stored at -70°C until total RNA was isolated.

Fruit of Jingyu peach, a non-melting flesh peach genotype, was from Institute of Forestry and Pomology, Beijing Academy of Agricultural and Forestry Sciences, China. At a stage equivalent to commercial ripeness, at about 9.56% and 12.16% soluble solids concentration (SSC) for fruits of Jingyu and Yuhua 1, respectively, free from visual symptoms of any disease or blemishes were chosen and directly stored at 25 ± 1°C for 6 days. For 1-methylcyclopropene (1-MCP) and ethylene treatments, Yuhua 1 fruits were sealed in two closed airtight containers, and 1 μl/L 1-MCP and 100 μl/L ethylene were injected into the two containers through a rubber septum, respectively. Fruits were incubated with 1-MCP or ethylene for 24 h at 25 ± 1°C and then containers were open to allow ripening in air in the same temperature conditions. Fruits of Yuhua 1 were also treated with ABA and IAA by dipping in 100 μM solutions in 0.2% Tween 80 for 20 min, respectively. In the JA treatment, Yuhua 1 fruits were dipped for 10 min in 50 mM methyl-jasmonate (MeJA) solution which also contained 0.2% Tween 80. After being dipped, fruits were dried and then allowed ripening at 25 ± 1°C in the air. Sampling from fruits of the two cultivars in normal room temperature was carried out daily for 6 days. Samples of Yuhua 1 fruit from different storage were used to evaluate the stability of 11 reference genes, while Jingyu fruit samples were only used to study the effect of different genotypes on stability values of 11 reference genes. Samples for different regulator treatments were taken for day 3, including one sample taken before treatment. All experiments were replicated three times with ten fruit as an experiment unit for gene expression studies. For all fruit samples epicarp and endocarp were excluded and the mesocarp was frozen in liquid nitrogen and stored at -70°C until further use.

Total RNA was isolated according to the method described by Meisel et al. 62. Genomic DNA was eliminated by treating each sample with RNase-free DNase I (TaKaRa, Japan) according to the instructions manual. The concentration of isolated total RNA was calculated from absorbance at 260 nm with BioPhotometer (Eppendorf, Hamburg, Germany), the purity was verified by optical density (OD) absorption ratio OD_{260 nm}/OD_{280 nm} between 1.80 and 2.05, and OD_{260 nm}/OD_{230 nm} ranging from 2.00 to 2.60 and the integrity was evaluated by electrophoresis on ethidium bromide-stained 1.0% agarose gels. Intact rRNA subunits of 18S and 28S were observed on the gel and absence of smears indicating minimal degradation of the RNA.

One microgram RNA was reverse-transcribed using the SYBR PrimeScript RT-PCR kit II (TaKaRa, Japan) for first-strand cDNA synthesis with 2.5 μM oligonucleotide dT primer and 5 μM random hexamer priming method according to the manufacturer's recommendations. Before transcription, RNA and primers were mixed and incubated at 70°C for 5 min followed by cooling on ice immediately. The first strand cDNA synthesis was started after adding transcription mixture at 37°C lasting 15 min for reverse transcriptase reaction. Finally, the PrimeScript Reverse Transcriptase was inactivated by heating the reaction mixture for 5 sec at 85°C. Each RNA sample was controlled for genomic DNA contamination without reverse transcriptase addition into cDNA synthesis mixture. All cDNA samples were stored at -20°C and diluted 1:10 with RNase-free water before being used as template in RT-qPCR analysis.

Eleven genes were selected for investigation to identify the most stably expressed reference gene(s) to be used in RT-qPCR studies. This group of genes comprised several classical reference genes which are the most commonly used as internal control for expression studies, such as GAPDH, 18S rRNA and ACT, the others based on previous reports 2543. The peach EST database 63 was queried with Arabidopsis protein sequences using TBLASTN to select peach homologs of genes commonly used as internal controls for gene expression analysis. The chosen peach ESTs were then used to query the Arabidopsis protein database using BLASTX 64 to obtain the description of peach reference genes. The reference genes evaluated are listed on Table 1, as are the corresponding accession numbers, Arabidopsis homolog locus, Arabidopsis locus description and main functions.

Primer pairs for RT-qPCR amplification were designed based on selected sequences using Beacon Designer 7.0 software (Premier Biosoft International, Palo Alto, California, USA) with a melting temperature between 60–62°C, 20–26 bp and about 50% GC content. Amplicon lengths were optimized to 103–146 bp to ensure optimal polymerization efficiency and minimize the impact of RNA integrity on relative quantification of gene expression 65. MFOLD software 66 was subsequently used to evaluate the target sequences amplified by the primer pairs to avoid the formation of secondary structures at the site of primer binding. The primers were further used to query peach EST database with BLASTN to confirm the identity of the genes. Before RT-qPCR, each primer pair was tested via standard RT-PCR to check for size specificity of the amplicon by 2.5% agarose gel electrophoresis and ethidium bromide staining. In addition, target amplicons were sequenced to confirm specificity of the PCR products. The primer sequences, amplicon sizes, and melting temperatures of all PCR products were indicated in Table 2.

RT-qPCR was performed using a Rotor-Gene 3000 (Corbett Robotics, Australia) and the SYBR Green Real-time Master Mix (TOYOBO, Japan). The PCR reaction volume was 20 μL containing 1.5 μL of diluted cDNA and 0.2 μM of each primer. Thermocycling conditions were set as an initial polymerase activation step for 2 min at 95°C, followed by 45 cycles of 15 sec at 94°C for template denaturation, 15 sec at 60°C for annealing and 20 sec at 72°C for extension and fluorescence measuration. Afterwards, a dissociation protocol with a gradient from 57°C to 95°C was used for each primer pair to verify the specificity of the RT-qPCR reaction and the absence of primer dimer. In addition, each PCR reaction included a reverse transcription negative control to check for potential genomic DNA contamination. Reagent contamination was also detected by a reaction mix without template. All samples were amplified in triplicates and the mean was used for RT-qPCR analysis.

Expression levels of the tested reference genes were determined by the number of amplification cycles (C_q) needed to reach a specific threshold level of detection. All amplification plots were analyzed with a threshold fluorescence value of 0.1 to obtain C_q values using the Rotor-Gene software version 6.1 (Corbett Research). The PCR efficiency showed in Table 2 was calculated for each gene with LinRegPCR program 67 from raw fluorescence data taken from the Rotor-Gene 3000 detection system. Results from the LinRegPCR and Rotor-Gene software were imported into Microsoft Excel and transformed to relative quantities using the comparative C_q method and specific efficiencies for each gene 68. The data obtained were converted into correct input files, according to the requirements of the software, and analyzed using three different VBA applets, geNorm (version 3.4) 3337, NormFinder (version 0.953) 3569 and BestKeeper (version 1.0) 3470.