Plant material and treatments.

In the chrysanthemum cultivar Reagan, the anthocyanin biosynthesis in the ray florets of the capitulum is light-dependent. Thus, the capitula from three color lines (purple, red, and pink) were selected as the plant material in this study. During the development of the capitulum, a total of six stages were defined: S0, S1, S2, S3, S4, and S5 (Sun et al., 2010). All samples were collected in Sept. 2013 from a growth chamber (the intensity of illumination was 50 ± 5 μmol·m⁻²·s⁻¹, the photoperiod was 16 h in the daytime and 8 h at night, the temperature was 20 ± 1 °C, and relative humidity was 60%) located at the Beijing Forestry University, Beijing, China. To investigate the effects of light on anthocyanin biosynthesis during floral development, two light treatments (illuminated condition and dark condition) were used on the five floral developmental stages. These stages were sampled as follows: L1, L2, L3, L4, and L5 were treated under a fluorescent lamp and were sampled during the capitulum developmental stages S1, S2, S3, S4, and S5, respectively; D1, D2, D3, D4, and D5 were 100% shaded using silver papers at the S0 stage and were sampled once every other 48 h when they developed to the S1, S2, S3, S4, and S5 stages, respectively (Fig. 1; Supplemental Table 1). All of the samples were transported to the laboratory immediately after acquisition and were then frozen in liquid nitrogen and stored at –80 °C.

View Figure

• Download as Powerpoint
• Download Figure

RNA extraction, quality controls, DNase I digestion, and first strand cDNA synthesis.

Total RNA was isolated using a Quick RNA Isolation Kit (Huayueyang Biotechnology Co., Beijing, China) from the ray florets of the three color lines at different floral developmental stages and light conditions. RNA quality was assessed by combining information from several control steps. First, the purity was inferred from the absorption ratios using a spectrophotometer (NanoDrop ND-1000; Thermo Fisher Scientific, Waltham, MA). Only the RNA samples with an A₂₆₀/A₂₈₀ ratio between 1.9 and 2.1 and an A₂₆₀/A₂₃₀ ratio greater than 2.0 were used for further analysis. Then, the integrity of the RNA samples was visualized using 1% (w/v) agarose gel electrophoresis stained with ethidium bromide. Finally, all RNA samples were pretreated with RNase-free DNase I (Promega, Madison, WI) at 37 °C for 30 min to eliminate any DNA contamination. The first strand cDNA was synthesized from 1 μg of total RNA using the M-MLV reverse transcription system (Promega) according to the manufacturer’s protocol.

Selection of candidate reference genes.

In our preliminary study, five Illumina HiSeq transcriptomic libraries of C. ×morifolium ‘Reagan’ were constructed. In total, 103,517 unigenes were identified, of which 60,712 were annotated against four public protein databases (data not shown). Using the DGE technology, the relative expression levels of these unigenes were analyzed in the ray florets of the illuminated L1, L2, and L3 samples and the non-illuminated D2 and D3 samples. The gene expression level was calculated using the fragments per kilobase per million fragments (FPKM) values (Mortazavi et al., 2008). The difference in gene expression between the samples was determined based on the FPKM value. As many as 25,006 unigenes showed little variation in the DGE technology. From these unigenes, nine reference genes that showed the most stable expression (|log₂ ratio|≤ 0.5) were selected as candidates for the normalization of gene expression, including actin-related protein (Actin), elongation factor 1α (Ef1α), beta-tubulin (β-tubulin), calcium-dependent protein kinase (CDPK), ABC transporter (ABC), protein phosphatase 2A (PP2A), F-box protein (F-box), E3 ubiquitin-protein ligase (E3), and TIP41-like protein (TIP41). These genes were evaluated in three color lines of C. ×morifolium ‘Reagan’ at five floral developmental stages and under two light conditions. Among these genes, Actin, β-tubulin, and E3 were traditional reference genes, whereas the other six were new genes we selected (Table 1).

Table 1.Summary of amplicon characteristics of reference genes in chrysanthemum ‘Reagan’, including fragments per kilobase per million fragments (FPKM) value, gene function, primer sequences, amplicon length, optimum annealing temperature (Tm), polymerase chain reaction efficiency (PE), regression coefficient (R²), and quantification cycle (Cq) value.^z

View Fullscreen

Primer design and test of amplification efficiency.

Primers were designed using Primer 3 software (Rozen and Skaletsky, 1999) with the following criteria: melting temperature of 55 to 70 °C, primer lengths of 20 to 25 bp, and amplicon lengths of 100 to 300 bp (Table 1). A melting curve analysis was performed by RT-qPCR after 40 cycles of amplification. The specificity of the primer amplicons was verified by 2% (w/v) gel electrophoresis.

RT-qPCR analysis.

PCR reactions were performed using a Mini Opticon Real-time PCR System (Bio-Rad, Hercules, CA) based on the SYBR Premix Ex Taq (Takara Bio, Shiga, Japan). Reactions were prepared in a total volume of 20 μL containing 2 μL of template (≈50 ng), 0.4 μL of each amplification primer (10 μmol), 10 μL of 2 × SYBR Premix Ex Taq, and 7.2 μL of ddH2O. The following standard thermal profile was used for all PCR reactions: 95 °C for 30 s followed by 40 cycles of 95 °C for 5 s, the optimal annealing temperature for 30 s, and 72 °C for 30 s. Melting curves were recorded after 40 cycles by heating from 65 to 95 °C by increasing the temperature stepwise by 0.5 °C every 5 s. PCR amplification efficiency was derived from a standard curve generated by a 10-fold serial dilution of cDNA. Negative controls were included for each primer pair and each RT-qPCR reaction was performed in triplicate.

Data analysis.

The expression levels of the tested reference genes were determined by quantification cycle (Cq) values, in which the quantification of the fluorescence reached a specific threshold level of detection. The Cq values were converted to relative values with the lowest Cq value set to one and the amplification efficiencies for each primer pair were incorporated (Table 1). To determine which reference genes were best suited for transcript normalization in chrysanthemum, the data obtained for each sample were analyzed using the geNorm [Version 3.4 (Vandesompele et al., 2002)], NormFinder [Version 0.953 (Andersen et al., 2004)], and qBase plus (Hellemans et al., 2007) programs according to their respective protocols. These strategies provided complementary measures of gene expression stability among the cDNA samples.

Assessment of normalization.

CmF3H was previously cloned by the authors and was used as a target gene to demonstrate the reliability of the results based on the set of reference genes identified here. The expression levels of CmF3H in different samples were quantified using stable reference gene(s) or a combination of reference genes determined by geNorm. Primer pairs of forward (5′GGTGCAAGATTGGAGGGAGAT3′) and reverse (5′CAGGTTGAGGACACTTTGGGTAG3′) for CmF3H were used.

Performance of the primers for each reference gene.

The results of the agarose gel electrophoresis demonstrated that all nine primer pairs amplified a single band of the expected size from various cDNA templates. The correlation coefficient (R²) indicated the quality of the fit of the standard curve to the plotted data points. The R² values of all of the primer pairs were greater than 0.9800 and ranged between 0.9800 (ABC) and 0.9990 (β-tubulin), indicating a strong relationship between the detected Cq values and the corresponding relative amount of template in all of the amplification reactions (Table 1). PCR amplification efficiency indicated the amplicon doubling rates of specific primer pairs during amplification, which ranged from 90.5% (F-box) to 107.9% (PP2A) and were obtained from the standard curves generated using a 10-fold serial dilution of the cDNA used in this study (Table 1). These results reflected the high quality and specificity of the RT-qPCR reactions. Therefore, these candidate genes were used in the next assay.

Expression profiles of candidate reference genes.

The Cq values for the nine genes were pooled so that the expression levels of each gene could be compared. The expression profiles of the nine candidate reference genes presented as Cq values are shown in Figure 2. The mean values of the reference genes were between 27.67 (β-tubulin) and 31.21 (ABC), which represented the different expression levels in C. ×morifolium ‘Reagan’. β-tubulin showed the highest expression level in all samples and the lowest Cq value (19.03). E3 presented the lowest expression level and the highest Cq value (37.52). Generally, the expression levels of most genes ranged from 27.89 to 31.75, and the Cq values of all candidate genes were dispersive, which indicated that none of the selected genes was expressed stable in the different samples (Fig. 2). Therefore, it was necessary to evaluate the reference genes for normalization under different experimental conditions.

View Figure

• Download as Powerpoint
• Download Figure

The stability of candidate reference genes.

In the present study, three methods were selected for analyzing the expression stability of genes. The first method, geNorm, calculates the average expression stability (M) based on the average pairwise variation between all genes tested. A lower M value indicates a more stable gene expression, whereas a higher M value indicates less stable gene expression. Stably expressed genes have values below 1.5 (Vandesompele et al., 2002) or 0.5 (Hellemans et al., 2007). The ranking order of genes used in this study according to their M values is depicted in Figure 3A, and the M values for all genes were far lower than 0.5. Because only the ray floret was considered in the present study, a lower threshold of 0.12 was adopted. F-box and TIP41 were the most stably expressed genes with M values of 0.0907; β-tubulin and ABC were the least stable genes in all samples (Fig. 3A). In different flower color lines across various floral developmental stages, F-box and PP2A all performed well with M values of 0.0889, 0.0844, and 0.0839 in the purple, pink, and red lines, respectively, whereas Ef1α and Actin were the least stable genes in the purple line with the respective M values of 0.1706 and 0.1674. TIP41, Ef1α, and E3 had unstable expression patterns in the pink and red lines. For the illuminated and non-illuminated samples, F-box and TIP41 showed the same performance as in the total samples. Generally, F-box and PP2A were the most stable genes under all of the experimental conditions.

View Figure

• Download as Powerpoint
• Download Figure

Another important analysis performed using geNorm was the determination of the optimal number of reference genes or the effects of adding extra genes to the analysis by calculating the pairwise variation (V) between two sequential candidate genes. A V_n/V_n+1 value lower than 0.015 under all experimental conditions indicates that adding an extra gene to obtain a reliable normalization factor (NF) was unnecessary. Based on this parameter, the combination of three reference genes, F-box, TIP41, and CDPK, were recommended for all samples because the pairwise variation value was lower than 0.015 (Fig. 3B). Two reference genes were sufficient for obtaining accurate results in the different flower color lines and illuminated samples because the V_2/3 values were lower than or equal to 0.015. Thus, the combinations of F-box and PP2A or F-box and TIP41 were the best reference selections in these four samples. For the non-illuminated treatment, the V_4/5 value was lower than 0.015, so four reference genes were chosen for normalizing the expression pattern results of the target genes (Fig. 3B).

The stability of reference genes was reanalyzed using NormFinder to confirm the results obtained with geNorm. NormFinder is based on variance estimation and takes intragroup and intergroup variations into account in calculating the NF. Genes were then ranked according to their stability under a given set of experimental conditions. In our study, the ranking order generated by this method was slightly different from the results obtained with geNorm (Table 2). F-box was still the most stable gene in all of the tested samples, whereas CDPK ranked second using this method and was more stable than TIP41. In different flower color lines at various floral developmental stages, the two most stable genes calculated by NormFinder were the same as those determined by geNorm, but according to the NormFinder results, ABC and CDPK were more stable than F-box in the pink and red lines, respectively. For samples under different light conditions, F-box and TIP41 were still the most stable genes under illuminated conditions, whereas in the non-illuminated samples, Actin was more stable than TIP41 (Table 2).

Table 2.Ranking of reference genes and their expression stability values calculated using NormFinder.^z

View Fullscreen

Finally, the Cq values and specific amplification efficiencies for each candidate reference gene were inputted into qBase plus to further evaluate their expression stabilities. qBase converts the data into normalized relative quantities. In this study, F-box and TIP41 were identified as the most stable reference genes in all samples and under different light conditions, as shown in Table 3. F-box and PP2A also performed well in the different flower color lines with the same stabilities as those determined by geNorm.

Table 3.Ranking of reference genes and the mean cv values calculated using qBase plus.^z

View Fullscreen

Reference gene validation.

F3H leads to the formation of dihydroflavonols, which is a required substance for anthocyanin biosynthesis. To verify the reliability of the selected reference genes, we used four normalization approaches according to the results of geNorm to analyze the expression patterns of CmF3H during different floral developmental stages of the three flower color lines under different light conditions. When normalized using the most stable reference genes F-box or PP2A, either singly or in combination, the expression patterns of CmF3H were almost identical with only slight differences (Fig. 4A and B). In contrast, when normalized using the least stable gene, β-tubulin, the expression pattern showed significant differences. The highest expression level was different when the least and most stable reference genes were used for normalization (Fig. 4A and B). The expression level of CmF3H was highest in the S3 stage when normalized to the least stable gene but was highest in the S5 stage when normalized to the two most stable reference genes either separately or in combination. Interestingly, the expression level of CmF3H increased with the floral development under the two light conditions. Compared with the illuminated samples, the expression levels of CmF3H in the non-illuminated samples were significantly lower during each floral developmental stage, indicating that the expression of CmF3H was repressed after shading.

View Figure

• Download as Powerpoint
• Download Figure

The significance of systematic verification for multiple reference genes.

Ideal reference genes should be stably expressed under all experimental conditions and independent of organs, tissues, developmental stages, and different treatments (Huggett et al., 2005). Thus, the use of one or more stable reference genes in RT-qPCR has progressed in tandem with the evolution of mathematical equations for quantification, moving from the one-reference model (Livak and Schmittgen, 2001; Pfaffl, 2001) to modern advanced relative quantification models with multiple reference genes (Hellemans et al., 2007). Therefore, a systematic verification for multiple reference genes under specific experimental conditions is extremely important for gene expression studies using RT-qPCR (Fu et al., 2013).

The high efficiency of candidate reference gene selection using transcriptomic libraries.

The frequently used methods for candidate reference gene selection are literature review and searching public databases of expressed sequence tags for reference genes related to certain biological processes such as glycolysis, cellular metabolism, protein synthesis, and degradation (Yeap et al., 2014). However, traditional reference genes are usually obtained using these methods. Several studies have showed that the expression of traditional reference genes might be unstable under different experimental conditions (Gutierrez et al., 2008), which might affect the expression analysis of target genes to a large extent. For instance, Gutierrez et al. (2008) found that compared with those validated reference genes, the normalization of target gene expression showed a 100-fold deviation using those invalidated reference genes. Using traditional reference genes blindly may result in wrong or even opposite results (Huggett et al., 2005; Vandesompele et al., 2002).

The transcriptome is a necessary link that connects genetic information and functional proteins (Martin and Wang, 2011), and its characteristics of high throughput and efficiency make it an important method for gene expression studies (Lee et al., 2010). Through the transcriptome approach, sequence information for thousands of genes can be analyzed simultaneously, quickly, and accurately. Moreover, based on transcriptomic libraries, the expression of thousands of genes can be detected rapidly using DGE technology (Martin and Wang, 2011). Through the analysis of transcriptomic data, the stable reference genes expressed in different cultivars, organs, and under various experimental conditions can thus be screened out efficiently. This approach has been successfully applied in many studies of plants such as tomato (Coker and Davies, 2003), perennial ryegrass (Lee et al., 2010), and arabidopsis (Czechowski et al., 2005). In the present study, we used DGE technology to screen out nine candidate chrysanthemum reference genes with moderate expression in five transcriptomic libraries. Of these, six were new genes, laying a foundation for the comprehensive evaluation of the expression stability of these candidate reference genes.

The stability of chrysanthemum reference genes used in the expression analysis of anthocyanin biosynthesis genes and floral development genes.

Chrysanthemum is an ideal crop for anthocyanin research because in chrysanthemum, only the cyanidin metabolic pathway is related to the anthocyanin metabolism and the background of pigment metabolism is simple. Understanding the expression patterns of key genes involved in the anthocyanin biosynthetic pathway of chrysanthemum under various conditions will thus help us to understand the molecular mechanism of color generation in chrysanthemum. However, the molecular mechanism underlying this induction remains poorly understood. In chrysanthemum, Ef1α and PP2A have been shown to be stably expressed under different biotic and abiotic stresses (Gu et al., 2011), but no detailed validation has been conducted to test their stability at different floral developmental stages and light conditions.

To our knowledge, this was the first systematic study to simultaneously investigate the expression stability of reference genes involved in flower color, floral development, and light effects in chrysanthemum. Based on the results of the present study, of the nine candidate genes, the traditional housekeeping gene β-tubulin yielded poor values in most of the sample sets, which was consistent with studies performed in cineraria (Jin et al., 2013), chinese iris (Gu et al., 2014), oil palm (Yeap et al., 2014), and cork oak [Quercus variabilis (Marum et al., 2012)]. Another conventional gene, EF1α, was not very stably expressed in the present study, which was inconsistent with the studies in soybean [Glycine max (Hu et al., 2009; Jian et al., 2008)], rice [Oryza sativa (Jain et al., 2006)], tomato (Løvdal and Lillo, 2009), cineraria (Jin et al., 2013), and tobacco [Nicotiana tabacum (Schmidt and Delaney, 2010)]. Actin was the last traditional reference gene selected in the present study, which has been used as a reliable reference gene for many years. However, further studies have revealed that it is not stable when experimental conditions change (Radonić et al., 2004). Therefore, it is necessary to evaluate the stability of this gene under specific experimental conditions. In our study, Actin performed well under different light conditions, which was similarly reported in soybean (Hu et al., 2009; Jian et al., 2008), but is contrary to the findings from studies in rice (Kim et al., 2003), sugarcane [Saccharum officinarum (Iskandar et al., 2004)], potato [Solanum tuberosum (Nicot et al., 2005)], and chamomile (Fu et al., 2013).

In general, the six new reference genes that we selected were more stable than the three conventional genes. In chrysanthemum, CDPK yielded high M values under different light conditions during the floral developmental process. Ullanat and Jayabaskaran (2002) found that in cucumber (Cucumis sativus), light differentially regulates CDPK transcript levels in a tissue-specific manner. Therefore, whether CDPK can perform better in the different tissues of chrysanthemum such as the tubular florets also requires further study. PP2A, which encodes the serine/threonine phosphatase, has been implicated in metabolism, DNA replication, transcription and translation, cell cycle, and signal transduction (Janssens and Goris, 2001). Some studies have found that PP2A is stable and is therefore suitable for use as a reference gene (Artico et al., 2010; Løvdal and Lillo, 2009). Compared with β-tubulin and EF1α, PP2A was stably expressed to some extent in our study and was highly stable in different color lines but unstable under different light conditions at various floral developmental stages. This result is consistent with the results from the different flower color lines and floral developmental stages of cineraria (Jin et al., 2013). F-box, which plays important roles in the signal transduction of plant hormones and the biological processes of self-incompatibility and floral development (Craig and Tyers, 1999), was the most stable gene in this study. Liu et al. (2012) and Mafra et al. (2012) reported that F-box was the most stable reference gene in tobacco and citrus (Citrus sp.) under conditions of viral infections, results that are consistent with the present study. In the six new genes we selected, E3 and ABC had unstable expression in all sample sets, and the M value of E3 remained especially high in all color lines. By contrast, TIP41 expressed constantly in all sample sets, which was consistent with the studies performed in tomato (Exposito-Rodriguez et al., 2008) and arabidopsis (Czechowski et al., 2005), but contrary to a study performed in chamomile (Fu et al., 2013). Therefore, we should choose reference genes that are suitable for normalization according to different experimental conditions.

The superiority of a comprehensive analysis of reference gene expression.

Although RT-qPCR has been widely used over the past few years, normalization of reference genes still attracts some criticism. There is no consensus as to which type of algorithm should be used to determine the stability of gene expression because different statistical methods, which are based on different principles, yield potentially contradictory results from the same data. Several strategies have been introduced to normalize the expression of reference genes, including software such as geNorm (Vandesompele et al., 2002), NormFinder (Andersen et al., 2004), qBase plus (Hellemans et al., 2007), and BestKeeper (Pfaffl, 2001). Among the studies in which geNorm, NormFinder, and qBase plus have been simultaneously used for normalization, some minor differences have been shown among these three programs in terms of evaluating the best sets of reference genes in blueberry (Die and Rowland, 2013), oil palm (Yeap et al., 2014), and citrus (Yan et al., 2012). Moreover, studies in longan [Dimocarpus longan (Lin and Lai, 2010)], tobacco (Schmidt and Delaney, 2010), and rubber tree [Hevea brasiliensis (Li et al., 2011)] have reported substantial differences between geNorm and NormFinder. The inconsistencies between these methods were expected because they are based on distinct statistical algorithms. In this article, all three methods were used to analyze the data to reduce the risk of inconsistent results. Although the ranking orders of gene stability generated by these methods were not exactly the same, the ranks of the most unstable genes were identical in all sample sets, which was similar to the results of a study in wheat [Triticum aestivum (Paolacci et al., 2009)]. In addition, the two most stable genes recommended by the three methods, F-box and PP2A, were the same in all of the samples.

The reliability of comprehensive evaluation on reference genes.

The expression pattern analysis of CmF3H under different light conditions and developmental stages of the capitulum was further performed to validate the usefulness of the two most stable genes selected. When normalized by F-box, the relative expression level of CmF3H showed the same pattern as when using the combination of F-box and PP2A, indicating that the two most stable genes are sufficient for RT-qPCR analysis of gene expression. In addition, Andersen and colleagues noted that geNorm may be insensitive to coregulated reference genes (Andersen et al., 2004). Because there are no coregulated reference genes in our study, the results calculated by geNorm are more convincing.