**4. Discussion**

Changes in plant secondary metabolism are closely related to the transcriptional activities of key genes, and gene expression analysis is a key technique for understanding the mechanisms involved in these processes [32,49]. RT-qPCR is the most accurate technique to obtain gene expression profiles that relate to biological function and regulatory networks [50–52]. However, the accuracy of RT-qPCR results mainly depends on using optimal internal reference genes that are stably expressed in samples across different experimental conditions. Reference genes are crucial for normalization of gene expression data and avoiding experimental errors by minimizing non-biological variation between different samples [24,33,35]. To ensure the accuracy of experiments, it is important to select suitable reference genes for each species that similar transcript levels under different experimental conditions.

*Lindera megaphylla* is an ecologically important and dominant broad-leaved evergreen tree species that is naturally distributed in the warm-temperate and subtropical zones of China [1]. This tree contributes to the seasonal landscape and produces volatile compounds with strong effects on bacteria and toxic gases. For example, the terpenes produced have a strong antibacterial effect [6,10,53]. *L. megaphylla* is also used as a medicinal plant [2]. However, few studies have focused on the molecular biology of *L. megaphylla* due to limited genomic information, and to date, no reference genes have been reported. In this study, we obtained transcriptome databases of different tissues of *L. megaphylla* and identified appropriate internal reference genes for use when studying the expression of genes. We initially used 40 internal reference genes from published model plants such as *Arabidopsis thaliana* and other species in the Lauraceae family to screen for constitutively expressed reference genes from the transcriptome database of *L. megaphylla*. We obtained 20 candidate genes based on an FPKM value > 50. As shown in Table 2, the final 14 candidate genes were

selected based on their E-values in the range of 91.04–107.17% and R2 values in the range of 0.991–0.999, which indicated that the primer pairs for standardized evaluation by RT-qPCR had high sensitivity and accuracy. In addition, the average Ct values of the candidate genes ranged from 17.94 (*PAB2*) to 27.51 (*UBC36*), indicating different expression levels (Figure 1). The results obtained are similar to those of many previous studies, such as on *Cryptomeria fortune* [33], *Gerbera hybrid* [19], and *Piper species* [48]. The results indicate that none of the reference genes had constant expression levels under all tested experimental conditions or in different species. Thus, it is necessary to carefully select the most appropriate reference gene to ensure the accuracy of RT-qPCR analysis for specific conditions and with specific materials. In this study, we combined four statistical algorithms (ΔCt, geNorm, NormFinder, and BestKeeper) to assess the expression stability of 14 candidate genes (*TUA*, *PPR*, *EIF4A-3*, *CYP26-2*, *helicase-15*, *TCTP*, *ACT7*, *PAB2*, *GAPDH*, *UBC28*, *EF2*, *UBC7*, *UBC36,* and *ubiquinone*) in different tissues across 16 different leaf developmental stages, and under different temperature stresses. The results demonstrated that the optimal reference genes were not the same under different conditions (Table 4).

There were slight differences in the rankings of candidate reference genes between the different algorithms. However, analysis by the ΔCt and NormFinder algorithms consistently identified the most stable or unstable candidate reference genes for most experimental sets, while in a few experimental groups, the expression patterns of similar genes were the most or least stable in geNorm and NormFinder. The ranking of candidate genes by BestKeeper suggested some differences compared to the other algorithms. For example, for different leaf developmental stages, the ΔCt and NormFinder platforms indicated that the *ACT7* and *UBC36* genes were the most stable. geNorm placed these genes in fourth and first place, while the BestKeeper program placed these genes in sixth and seventh place. Although the rankings of candidate genes produced by the different algorithms were slightly different, the top five stable candidate genes selected by the algorithms were similar for each group of experimental conditions (Table 3). For instance, *ACT7*, *UBC36*, *TCTP*, *UBC7,* and *ubiquinone* were the top five most stable genes based on the geNorm and NormFinder analyses across the leaf development stages, and the ΔCt analysis showed similar results, except for *ubiquinone*. BestKeeper analysis identified two stable genes: *UBC7* and *TCTP*. Numerous other studies have found similar differences between the outputs of geNorm and NormFinder [32,54], and many studies also demonstrated that these subtle differences result from the use of different algorithm models [33,34,55].

To comprehensively synthesize the results of the four algorithms, RefFinder was utilized to rank the identified candidate genes in *L. megaphylla*. This analysis plays an important role in integrating the screening results of reference genes from other algorithms by assigning an appropriate weight to each gene and calculating the geometric mean of its weights to produce a final ranking [32,34,35]. Fortunately, we found that the results from RefFinder were similar to those of the different algorithms in each experimental set, proving that RefFinder can assess and screen the optimal reference genes [56], as shown in Figure 6 and Table 4.

The results also indicate that we screened and identified the optimal reference gene combinations for use in *L. megaphylla* samples generated under different experimental conditions. To compare expression in different seedling tissues, *helicase-15* and *PAB2* were the most suitable, whereas *UBC28* and *UBC7* were most stably expressed in different adult trees tissues. When two different groups of tissues were analyzed, *helicase-15* and *UBC28* emerged as the most stable gene combination. For different leaf developmental stages, *ACT7* and *UBC36* were best, whereas *ubiquinone* and *UBC7* were best when analyzing samples over the entire growth cycle. Interestingly, for cold stress of 7 days and 24 h, the optimal reference genes were *TCTP* + *ubiquinone* and *GAPDH* + *UBC36*, respectively. For heat treatment, the best reference genes were *PAB2* and *CYP20-2*, while for overall temperature stress, *PAB2* and *PPR* were most stable. When all samples were tested, *ubiquinone*, *EF2*, *UBC7,* and *GAPDH* were the optimal candidate reference genes overall for the normalization of gene expression in *L. megaphylla* (Table 4). These analyses are sufficient to

demonstrate the necessity of screening suitable internal reference genes under different experimental conditions for each species.

With increasing demand for accurate scientific data, it has become important to screen for the best internal reference genes in a greater number of plant species and for different experimental treatments [19,32,49]. Some housekeeping genes involved in cytoskeleton structure or primary metabolism, including *ACT*, *TUA,* and *EIF4α*, are extensively used as reference genes in many plant species. For example, *ACT2/7* and *TUA* are the three most stable genes across different developmental stages of *Glycine max* [57]. *AhyACT*, *AhyMDH*, and *AhyEF-1a* are the most stable genes in different tissues of amaranth [56]. Research on bamboo revealed that *eIF4α* was most stable in different organs, while *CYP*, *eEF1α,* and *UBQ5* were found to be the optimal reference genes for different developmental stages of *Bambusa tulda* [18]. In *Litsea cubeba*, *F-BOX*, *EF1α,* and *EIF4α* were the most stable reference genes across different tissues and developmental stages [40]. In the present study, *TUA* was the least stable candidate gene in *L. megaphylla,* as calculated by the different programs, while the best combinations of genes were *helicase-15* + *UBC28* and *ACT7* + *UBC36* in different tissues and developmental stages.

Zhong et al. [24] found that *ACT* was stably expressed in high-temperature-stressed *Psoralea corylifolia*. Chen et al. [22] indicated that *EF-1α* was the most stably expressed and suitable reference gene under heat and cold treatments. In this study, *PAB2* + *CYP20-2* and *UBC36* +*TCTP* were identified as the most stable reference genes under heat and cold treatments, respectively. The experimental results once again show the importance of screening the best reference genes under different experimental conditions for different species.
