Molecular clock dating

The divergence time of Musaceae was estimated using BEAST v2.6.4 [102]. To incorporate multiple fossil calibration points and reduce the bias imported from a single calibration point, the divergence time was estimated by including the whole Zingiberales. SortaDate [50] was used to choose genes suitable for divergence-time estimation. This package determines which gene trees are clock-like, have the least topological conflict with the species tree, and have informative branch lengths. The ML tree generated from the complete plastome sequence dataset was used as an input species tree. As the result of SortaDate, the final screened genes were ccsA, matK, ndhF, rpoC1, and rpoC2. We selected optimal nucleotide substitution models for each of the five genes using Modeltest-NG 0.1.6 [96] under the AICc. These were identified as GTR+G4 for ccsA, matK, rpoC1, rpoC2, and GTR+I+G4 for ndhF.

In BEAST, the newick ML tree of Zingiberales inferred from complete plastome sequences was used as a starting tree due to its more robust phylogenetic resolution. Clock models were linked, while site models were unlinked for each gene. The uncorrelated log-normal distribution relaxed molecular clock model was selected with the Yule model as the tree prior. MCMC run was set to 100 million generations, sampling every 10,000 generations. BEAST 2 output was assessed in Tracer 1.7.2 [103] to evaluate convergence and ensure an effective sample size for all parameters surpassing 200. TreeAnnotator v2.6.4 was used to annotate the maximum clade credibility tree after removing the first 20% of samples as burn-in.

Three fossil records and one secondary calibration point were used in this divergence time estimation. Spirematospermum chandlerae [104] was used to calibrate the crown age of order Zingiberales with a mean age of 83.5 Ma. Zingiberopsis attenuate [105] was applied as a mean age of 65 Ma for the crown node of the Zingiberaceae family. Then Ensete oregonense [106] was used to calibrate the crown age of Ensete and Musella clade with a mean age 43 Ma. Each fossil calibration point was assumed to follow a normal distribution with a standard deviation of 2 and an offset of 2, resulting in 81.6–89.4, 63.1–70.9, and 41.1–48.9 Ma 95% intervals, respectively. The secondary calibration point was generated based on previous studies on Monocots [107, 108]. It was placed on the stem node of Zingiberales with a normal distribution as a mean age of 100 Ma and a broad standard deviation of 5 (95% intervals 90.2 – 110 Ma).