Tomato Micro-Tom Rg1 (MT-Rg1) fruits were sampled at eight different developmental stages (Figure 1), according to Gillaspy et al. (1993). The collected fruit stages are as follows: (1) 1.5 cm immature green fruit, (2) 2 cm immature green fruit, (3) mature green fruit, (4) breaker, (5) turning, (6) orange, (7) red firm, and (8) red ripe fruit. Fruits (pericarp tissue only) were frozen and stored at –80°C until further use. Total RNA was isolated using Trizol, followed by several precipitation steps, a DNase I treatment, and subsequently, cDNA was prepared, as described in the Materials and Methods section.

Based on previous studies conducted on S. lycopersicum cv. ciliegia (Expósito-Rodríguez et al., 2008), a total of four candidate reference genes were selected for qRT-PCR normalization. These are CAC, SAND, and Expressed (Expósito-Rodríguez et al., 2008), together with the commonly used house-keeping gene ACTIN2 (Bemer et al., 2012) (Table 1). Primers used are listed in Table 2. The primers for the first three genes span intronic regions (Table 2), i.e., a larger amplicon will be seen when DNA contamination is present. Primer melting curves for all genes showed a unique peak corresponding to the expected amplicon (Supplementary Figure 1). The correct size of the amplicons was verified by gel electrophoreses.

Finally, as described at the end of the Results section, relative expression of the transcription factors FUL1 (Bemer et al., 2012) and AP2c (Karlova et al., 2011) was analyzed and compared to previous reports (Karlova et al., 2011; Bemer et al., 2012) in order to validate the normalization procedure using the most stable reference genes identified during fruit development in MT-Rg1.

Data processing is illustrated in Figure 2. To evaluate the individual reaction kinetics, without the need for a standard curve, we used Real-Time PCR Miner software (Zhao and Fernald, 2005). Raw data of the time (cycle) sequence fluorescence values were imported from the real time PCR machine into Miner to calculate primer efficiency and its associated standard deviation, and the cycle threshold (Ct) value. The resulting data was used as input for qBase.

The qBase software (Hellemans et al., 2007) processes data modules as independent experiments. Thus Ct values for each individual reaction, and each tested reference gene, were imported as independent experiments. The first step in the qBase workflow combines raw data from all individual run files of the same experiment into a single data table, where data points with identical sample and gene names are automatically identified as technical replicates. Then, we executed a data quality control, where replicated reactions that differ between them in more than 0.5 Ct were excluded from further analysis as a potential outlier, totaling 19 excluded cases (6.59% of the complete data set). Next, average efficiency of genes and its associated standard deviation were replaced with the values obtained from the Miner software for each gene. Finally, we performed the quantification of relative quantities, which lie in the conversion of quantification cycle values into relative quantities based on the gene specific amplification efficiency.

Then a normalization factor based on the expression levels of the tested reference genes was calculated using both geNorm (Vandesompele et al., 2002) and NormFinder (Andersen et al., 2004), which are Excel based software packages.

geNorm calculates a gene stability measure M as the average pairwise variation of a particular gene against other reference genes (Vandesompele et al., 2002) (Figure 3A). Genes are ranked according to increasing expression stability, with CAC and Exp displaying the most stable expression profiles and ACT and SAND less stable profiles. The CAC and Exp genes were also reported as one of the most stable reference genes in S. lycopersicum cv. ciliegia (Expósito-Rodríguez et al., 2008). Next, normalization factors were calculated for the most stable reference genes. Additionally, the software calculates the pairwise variation (V) between two sequential normalization factors (NF_n and NF_{n + 1}) in order to determine the minimum number of reference genes required for normalization (Figure 3B) (Vandesompele et al., 2002). In other words, pairwise variation estimates how the NF changes if you add another reference gene to the NF calculation. geNorm pairwise comparison shows that adding a third gene has no significant contribution to the newly calculated normalization factor (V2/3), meaning that using two reference genes is sufficient to normalize our qRT-PCR data. Its value is lower than the cutoff of 0.15 proposed by Vandesompele et al. (2002), which suggests the inclusion of additional reference genes are not necessary. From this analysis we concluded the use of the two most stable reference genes CAC and Exp is sufficient for accurate normalization (V2/3 = 0.103) in MT-Rg1 fruits.

NormFinder (Andersen et al., 2004) is an algorithm to identify the optimal normalization gene among a set of candidates. The validity of this approach is related to the number of samples and candidates analyzed because it considers intra- and intergroup variation to estimate a stability value of the tested reference genes. The intragroup refers to the biological replicates and the intergroup refers to the different fruit developmental stages. We used the input file containing the qBase-calculated relative quantity values for each gene, sample and biological replica and ran the analysis. The results revealed CAC as the most stable reference gene with the best stability value (i.e., the lowest value) of 0.049 (Figure 4). The best combination of genes was CAC and Exp. These results match those proposed by the geNorm software.

Consequently, we decided to continue the experiment using the best reference genes CAC and Exp to normalize the gene expression data of FUL1 and AP2c and compare it with normalized expression data with the less stable reference gene SAND in all tested samples.

To assess the validity of the MT-Rg1 genotype as a model tomato species, we calculated the relative expression of AP2c and FUL1 at different developmental stages, and compared them with previous reports (Karlova et al., 2011; Bemer et al., 2012). Accurate measurement of gene expression requires a normalization by multiple, carefully selected reference genes. We therefore decided to normalize gene expression using both the most stable reference gene pair identified by GeNorm and NormFinder, CAC and Exp, but also using the less stable reference gene, SAND.

AP2 is a transcription factor which plays important roles in development, ethylene response and pathogen resistance (Karlova et al., 2011). Five distinct tomato cDNAs encoding AP2 putative tomato orthologs were identified by Karlova et al. (2011) in the Moneymaker tomato cultivar. From this report, we chose the AP2c sequence, which shows the highest expression in early stage fruits, but no significant expression beyond the mature green stage. Our results in MT-Rg1 fruits (Figure 5A) were concordant with the previously reported expression profile in the tomato cultivar Moneymaker (Karlova et al., 2011). In the first three fruit stages, we observed a high expression and then expression levels dropped from the breaker stage on.

Tomato has two orthologs of FUL, a transcription factor involved in fruit ripening (Bemer et al., 2012). We also calculated the relative expression of FUL1 in order to validate our MT-Rg1 qRT-PCR results. FUL1 expression was found to be very low during early stages of fruit development, but rapidly increased from the breaker stage, reaching its maximum in the orange ripe stage (Figure 5B). A comparable trend of expression was reported in tomato MT fruits (Bemer et al., 2012). Normalizing the qRT-PCR results for AP2c and FUL1 with the less stable reference gene SAND did result in differences in expression levels in the fruit stages with high AP2c and FUL1 expression levels, although the trend of the expression profile did not change substantively.

Tomato (S. lycopersicum) MT-Rg1 genotype (Pino et al., 2010) seeds were sown in flat trays containing a 3:1:1:1 mixture of leaf soil, soil, sand, and perlite, respectively. Plants were grown in a greenhouse under local conditions (20°43′1″ N; 101°19′56″) in the summer (around 16 h light/8 h dark) at an average mean temperature of 30°C.

Tomato fruits were sampled at eight different developmental stages, based on the fruit stage division proposed by Gillaspy et al. (1993). In total, four fruits of each developmental stage were collected: 1.5 cm immature green fruits, 2 cm immature green fruit, mature green fruit, breaker, turning, orange, red firm, and red ripe (Figure 1). Fruits devoid of seeds and placental tissue (only pericarp tissue) were frozen in liquid nitrogen and stored at -80°C until RNA extraction.

Total RNA extraction was performed using the TRIzol^® (Ambion) protocol, followed by precipitation using 0.8 M sodium citrate, 1.2 M NaCl and isoproponol, a second precipitation using 8 M LiCl, and a third precipitation with 3 M NaAc pH 5.2 and 100% ethanol. Subsequently, a DNase I (Life Technologies) treatment was performed according to manufacturer’s specifications, and RNA was recovered using phenol/chloroform extraction followed by a precipitation with 3 M NaAc pH 5.2 and 100% ethanol. RNA was resuspended in 10 μl of DEPC water and quantified by measuring its absorbance at 260 nm. RNA integrity was evaluated by the 260/280 and 260/230 ratios, and confirmed by agarose gel electrophoresis.

cDNA was synthesized from 1.3 μg of total RNA using the SuperScript^® III System (Life Technologies). In summary, total RNA was mixed with 1 μl of oligo dT (50 μM), 1 μl of dNTPs (10 μM) and MQ water, giving a total volume of 14 μl, and incubated for 5 min at 65°C and then chilled on ice. Subsequently, 4 μl of First Strand Buffer (5x), 1 μl of DTT (0.1 M) and 1 μl of SuperScript reverse transcriptase III (200 units/ μl) were added, each reaction was incubated for 2 h at 50°C and, finally, inactivated for 5 min at 70°C. cDNA was diluted 1:50 for use in quantitative real-time PCR experiments.

Quantitative real-time PCR amplification reactions were performed and run in technical triplicate on 96-wells plates on an ABI 7500 Fast Real Time PCR System using SYBR^® Green I (Life Technologies). Each PCR reaction mix consisted of 2 μl of SYBR Green (1:10000), 0.4 μl of forward and reverse oligos (10 mM), 2 μl of PCR buffer (10x), 0.05 μl of dNTPs (10 mM), 1.2 μl of MgCl₂ (50 mM) and 0.05 μl of Platinum^® Taq DNA Polymerase (Life Technologies; 2 U/rxn) in a total volume of 10 μl. Finally, 10 μl of 1:50 diluted template cDNA was added, resulting in a total volume of 20 μl per PCR reaction. PCR cycling was performed as follows: 5 min at 94°C followed by 40 rounds of 15 s at 94°C, 10 s at 60°C, 15 s at 72°C, and finally 1 round of 35 s at 60°C. Melting curve cycling consisted of: 15 s at 95°C, 1 min at 60°C, 30 s at 95°C, and 15 s at 60°C.

The qRT-PCR experiment was performed using two biological replicates. For the first one, total RNA extracted from one fruit of each stage was used. cDNA for the second biological replicate was prepared from total RNA extracted from a pool of three fruits of each stage. For each analysis, three technical replicates were performed. The Supplementary Figure 2 shows the results of the comparison between the two biological replicates and the standard deviation of the three technical replicates for each fruit stage. In general, the detected Ct values did not differ much, but slightly more stability was observed in those where the cDNA was prepared from the RNA of the three pooled fruits. No difference in the standard deviation was observed between the two strategies (Supplementary Figure 2).

Raw fluorescence values from quantitative PCR experiments were imported into the Real-Time PCR Miner software (Zhao and Fernald, 2005). Ct values, average efficiency and standard deviation calculated by the Miner program were used as input for the qBase software (Hellemans et al., 2007). The qBase program calculates relative quantities of expression and produced an input file for the geNorm software (Vandesompele et al., 2002). The geNorm program determines the minimum number of genes required to calculate a reliable normalization factor. Finally, the qBase data file, with few modifications, was also used as an input file for the NormFinder program (Andersen et al., 2004), which considers the average expression stability between replicates and also between samples (intra- and inter-group variation, respectively) to suggest the best candidates for reference genes.