Melon (Cucumis melo L. var. inodorus cv. ‘Elizabeth’) plants were grown in a plastic greenhouse under commercial production conditions at the Huazhong Agricultural University (East Longitude 113°41′ 115°05′, North Latitude 29°58′ 31°22′). This cultivar is monoecious and without hermaphroditic flowers. To get fruits, two different fruit set methods were used, i.e., CPPU treatment and artificial pollination. For CPPU treatment, female flowers were covered with paper bags 1 day before anthesis to prevent natural pollination. These paper bags were removed from the female flowers on the day of anthesis, and the unpollinated ovaries were sprayed with CPPU (Shiteyou, China) at a concentration of 10 μM in the morning. The female flowers were subsequently covered again after spraying till the onset of fruit development. Simultaneously, on other female flowers, artificial pollination was done by hand. Three biological replicates were adopted for each treatment. Fruits were harvested at 1, 3, 5, 7, 10, 15, 20, 25, 28, 30, 32, and 34 days after anthesis (DAA). Two fruits were randomly harvested at each sampling time from each biological replication and mixed together. Ovaries were sampled at 1 and 3 DAA. The mesocarp tissues in the center-equatorial portion of fruits were collected as samples after 3 DAA. All the samples were immediately frozen in liquid nitrogen and stored at -80°C for the subsequent RNA extraction, enzyme assay, and sucrose measurement.

The qRT-PCR protocol (Nolan et al., 2006) and 11 golden rules (Udvardi et al., 2008) were used as guidelines in the experiments. Total RNA was extracted from the ovaries or mesocarp tissues with the TransZol (TransGen, China) according to its instruction. RNA quality and quantity were measured by a NanoDrop 2000 spectrophotometer (Thermo, China), and RNA integrity was confirmed in a 2% agarose-gel electrophoresis. A PrimeScript RT Reagent Kit with gDNA Eraser (Perfect Real Time; TaKaRa, China) was used to eliminate the genomic DNA (gDNA) in the RNA samples and synthesize the cDNA. gDNA was extracted from the mesocarp tissues with a Plant Genomic DNA Kit (Tiangen, China) and amplified using 2× PCR Reagent (Tiangen, China).

Fourteen genes were initially selected as the candidate reference genes. Table 1 lists the information on these candidate reference genes. For each gene, BLASTN was performed via the Melonomics database¹ against the Melon transcripts CM_3.5. The Arabidopsis homologs were used as the query sequences. The sequence that best matched each Arabidopsis query was downloaded with its respective structure information. Primers that covered exon–exon junction or flanked an intron were designed using Primer3Plus². The product size was set as 80–150 bp. The specificity of the designed primers were manually verified and confirmed by running BLASTN against the Melon transcripts CM_3.5. The 2% agarose-gel electrophoresis was further used to determine the PCR amplification specificity for each gene, with gDNA and cDNA as templets. The melon species name abbreviation “Cm” was adopted as a prefix to specify the orthologous melon genes. To make the results comparable, six best reference genes previously reported in melon, including ribosomal protein L2 (L2), actin^∗, CYP, ADP-ribosylation factor 1 (ADP), ribosomal protein L (RPL), and ubiquitin extension protein (UBI) that identified in stems, roots and leaves (Kong et al., 2014b; Sestili et al., 2014), were also selected. Their primer sequences were also used in the present study. The unigene accession numbers of actin^∗ and CYP provided in the reference (Sestili et al., 2014) were used to retrieve the Melonomics database. The results showed that the two genes were annotated as actin7 and CYP2 on melon genome, respectively. Their annotations and gene IDs on genome were used to replace the gene names and unigene accession numbers supplied in the previous reference (Sestili et al., 2014). The prefix of “Cm” was also added before names of the six reported reference genes.

Quantitative reverse transcription PCR reactions were performed on a QuantStudio 7 Flex Real-Time PCR System (The Applied Biosystems, America) using a total volume of 10 μL, which contained 0.2 μM of each primer, 1× Top Green qPCR SuperMix (TransGen, China), and100 ng of cDNA. The amplification conditions were 30 s at 94°C, 40 cycles of 5 s at 95°C, 15 s at 58°C and 10 s at 72°C, followed by a melting curve analysis by heating the PCR products from 65 to 95°C. qRT-PCR reactions were performed in two technical replicates with a negative control without template. The PCR amplification efficiency for each gene was determined by analyzing fivefold serial dilutions of pooled cDNA in the concentrations of 800, 160, 32, 6.4, and 1.28 ng μL^-1.

The cycle threshold (Ct) value was recorded for each qRT-PCR reaction. The R statistical package³ was used to draw the boxplot to display the expression variation for each gene. The amplification efficiency (E) was calculated using the equation:

for each gene, in which the slope is the standard curve slope generated by the QuantStudio 7 Flex Real-Time PCR System based on the fivefold serial dilutions of pooled cDNA samples. The algorithms of geNorm (Vandesompele et al., 2002) and NormFinder (Andersen et al., 2004) were used to evaluate the expression variation. Prior to data entry, the raw Ct values were corrected by PCR efficiency and transformed into relative expression quantities using the equation (1 + E)^ΔCt, in which ΔCt is the difference of the lowest Ct value of the calibrator and the Ct value of the sample being tested.

Pollinated fruits at 10, 15, 20, 25, 28, and 32 DAA were used to determine the sucrose contents and enzyme activities. Extraction and measurement of sucrose were performed as previously described (Liu et al., 2012). The gas chromatograph Agilent 7890A (Agilent Technologies, USA) coupled with a HP-5 capillary column (30 m × 0.32 mm × 0.25 μm) and a CTC PAL autosampler (CTC Analytics, Switzerland) were used for sucrose detection and quantification. Extractions and activity measurements of AI and SPS were conducted as previously reported (Hubbard et al., 1989). CmAIN2 and CmSPS1 were selected to test the effectiveness of the identified reference genes. Primers of the two genes were designed according to the aforementioned methods and listed in Table 1. To demonstrate the transcriptional regulation of sucrose metabolism, an earlier sampling point at 7 DAA was added and used as control to analyze the expression levels of CmAIN2 and CmSPS1. qRT-PCR reactions were performed according to the aforementioned methods. The 2^-ΔΔCt method was used to calculate the relative expression level. The best reference genes identified by geNorm and NormFinder were used for normalization. Geometric means were calculated for the reference gene combinations and used for normalization. The widely used reference gene CmCYP7 in melon fruits and the least stable reference gene identified in this study were additionally used for normalization. Three biological and two technical replicates were adopted for the aforementioned measurements at each sampling point.

A total of 20 genes were selected as the candidate reference genes in this study, which included the six reference genes that previously identified in melon roots, leaves, and stems (Kong et al., 2014b; Sestili et al., 2014). Information on these 20 candidate reference genes is listed in Table 1. Primer sequences of the six reference genes previously identified in melon (Kong et al., 2014b; Sestili et al., 2014) were also used in this study. While, primer sequences of the other 14 candidate reference genes were designed.

To improve the specificity of qRT-PCR analysis, one primer of each gene was initially designed to match an exon–exon junction. However, a suitable primer binding site was difficult to find in the exon–exon junction regions of some genes. Primers of the said genes were then located on the exons separated by an intron. Amplification specificity of the 20 genes was determined by a 2% agarose-gel electrophoresis with cDNA and genomic DNA (gDNA) as templates, respectively (Figure 1). The results showed that only the expected products were amplified with cDNA templates and no products were amplified with gDNA templates for CmACT, CmCAC, CmPP2A, CmRP2, CmRPS15, CmSAND, CmTBP2, and CmTIP41. Meanwhile, CmCYP7, CmEF1α, CmGAPDH, CmRAN, CmYLS8, and CmTUA5 amplified specific products from both cDNA and gDNA templates. However, the amplicons from gDNA templates containing an intron are larger than those from cDNA templates, demonstrating the success of primer design and the absence of gDNA contaminations in the cDNA samples. Among the six previously identified reference genes in melon, CmACT7, CmCYP2, CmL2, and CmRPL amplified the same products on both gDNA and cDNA templates, whereas, CmADP had no products and CmUBI generated a larger product on gDNA templates, respectively. Meanwhile, a single peak in the melting curve analysis further supported the specific amplification of each gene. Amplification efficiencies of the 20 genes varied from 93.1% (CmEF1α) to 109.3% (CmL2). Table 1 summarizes the primer sequences and amplification characteristics of the 20 genes.

Expression variations of the 20 genes were examined during the developments of pollinated and CPPU treated fruits and represented as boxplot in Figure 2. The mean Ct values for the 20 genes across the 24 samples ranged from 19.0 (CmCYP2) to 29.4 (CmSAND). CmADP exhibited the least expression variation, whereas CmRP2, CmRAN, CmGAPDH, and CmEF1a showed the highest expression variations. Given the presences of non-biological variations in the experiments, gene expression stability cannot be accurately estimated by direct comparison of the raw Ct values. Therefore, geNorm and NormFinder were used to assess the expression stability.

GeNorm calculates the expression stability value (M) for each gene and considers that gene with lower M value has higher expression stability. Taking into account the effects of fruit set methods and developmental phases, geNorm ranked CmRPL and CmRPS15 as the pair of best reference genes, and CmUBI as the least stable gene (Figure 3A). Furthermore, geNorm determines the optimum number of reference genes required for reliable normalization by calculating the pairwise variation values (V). When the pairwise variation (V_n/n+1) is less than 0.15, it is recommended that no additional genes are required for the normalization. Pairwise variation analysis demonstrated that at least five reference genes were required for more reliable normalization, namely, CmRPL, CmRPS15, CmTIP41, CmCYP7, and CmADP (Figure 3B).

NormFinder uses a model-based approach and considers variations across groups to calculate the expression stability for each gene. The gene with lower stability value is top ranked. The samples were divided into fertilized and parthenocarpic groups according to the fruit set methods. The plots of inter- and intragroup variations with respect to fruit set methods for each gene showed that CmTIP41 had the lowest intergroup variation, whereas CmACT7 exhibited the lowest intragroup variation (Figure 4). CmRPS15 was determined as the best reference gene with the lowest stability value of 0.268. Meanwhile, NormFinder identified CmRAN and CmTBP2 as the combination of two best reference genes, with minimal combined inter- and intragroup variations. CmEF1α was ranked as the least stable reference gene (Figure 4).

Parallel changes of sucrose contents, activities of AI and SPS, and expression patterns of CmAIN2 and CmSPS1 were analyzed during fruit ripening (Figure 5). The AI activity was highest at 10 DAA and then decreased sharply with fruit ripening (Figure 5A). The SPS activity gradually increased from 10 DAA and reached the highest level at 25 DAA, then decreased and kept stable till fruit matured (Figure 5B). The sucrose contents were very low before 20 DAA and then gradually increased with fruit ripening (Figure 5C). The best genes that determined by geNorm and NormFinder were used to normalize the expression levels of CmAIN2 and CmSPS1, respectively. These genes included the single gene CmRPS15, combination of CmRPS15 and CmRPL, combination of CmRAN and CmTBP2, and multiple genes of CmRPL, CmRPS15, CmTIP41, CmCYP7, and CmADP. Meanwhile, CmCYP7 and the least stable gene CmUBI determined by geNorm were also used for normalization. The relative expression levels of CmAIN2 gradually decreased after 7 DAA, and were nearly undetectable after 20 DAA when the best reference genes were used for normalization (Figure 5A), which were in agreement with the changing patterns of AI activities. However, when CmUBI or CmCYP7 was used for normalization, the expression levels of CmAIN2 increased from 7 to 10 DAA and then decreased as fruit ripening. CmSPS1 was upregulated from 7 to 15 DAA, then gradually downregulated as fruit matured, no matter the stable or less stable genes were used for normalization (Figure 5B). However, compared with the expression levels normalized by the best genes, the expression levels of CmSPS1 were obviously overestimated from 15 to 28 DAA when the less stable gene CmUBI or CmCYP7 was used for normalization.