*3.8. Comprehensive Stability Analysis Using RefFinder*

Since the different algorithms determined different stability rankings for the candidate genes, the program RefFinder was used to calculate the geometric mean of the ranking results from the four methods [16]. The comprehensively ranked candidate genes (Table 4) did not present one or two universal reference genes for the normalization of gene expression data for all samples. However, the consensus for the top two genes in adult trees during leaf development, under cold treatment for 7 days or 24 h, and under heat treatment for 24 h was consistent with the results of the ΔCt and NormFinder analyses. The top two reference genes in seedlings or under temperature stress were consistent with the results of the ΔCt or geNorm analysis, respectively. In the other experimental conditions, the top two most stable genes in the overall ranking appeared in either the top two or top three positions in one of the other four algorithms. In addition, all analyses revealed that *TUA* was the most unstable gene. We then analyzed the top five most stable genes as determined using the ΔCt, geNorm, NormFinder, BestKeeper, and RefFinder algorithms (Figure 6). The top two most stable reference genes selected using RefFinder for the various experimental sets were ranked highly by three or four of the other software programs.

**Figure 6.** Venn diagrams displaying the overlap between the top five most stable reference genes as ranked by ΔCt, geNorm, NormFinder, and BestKeeper for the different datasets. Genes in the overlapping area were confirmed by more than one algorithm. The light purple, purple, light green and pink ellipses contain the top five most stable reference genes selected by ΔCt, geNorm, NormFinder, and BestKeeper, respectively.

