Resolving and dating the phylogeny of Cornales – Effects of taxon sampling, data partitions, and fossil calibrations

Abstract

The order Cornales descends from the earliest split in the Asterid clade of flowering plants. Despite a few phylogenetic studies, relationships among families within Cornales remain unclear. In the present study, we increased taxon and character sampling to further resolve the relationships and to date the early diversification events of the order. We conducted phylogenetic analyses of sequence data from 26S rDNA and six chloroplast DNA (cpDNA) regions using parsimony (MP), maximum likelihood (ML), and Bayesian inference (BI) methods with different partition models and different data sets. We employed relaxed, uncorrelated molecular clocks on BEAST to date the phylogeny and examined the effects of different taxon sampling, fossil calibration, and data partitions. Our results from ML and BI analyses of the combined cpDNA sequences and combined cpDNA and 26S rDNA data suggested the monophyly of each family and the following familial relationships ((Cornaceae–Alangiaceae)–(Curtisiaceae–Grubbiaceae))–(((Nyssaceae–Davidiaceae)–Mastixiaceae)–((Hydrostachyaceae–(Hydrangeaceae–Loasaceae))). These relationships were strongly supported by posterior probability and bootstrap values, except for the sister relationship between the N–D–M and H–H–L clades. The 26S rDNA data and some MP trees from cpDNA and total evidence suggested some alternative alignments for Hydrostachyaceae within Cornales, but results of SH tests indicated that these trees were significantly worse explanations of the total data. Phylogenetic dating with simultaneous calibration of multiple nodes suggested that the crown group of Cornales originated around the middle Cretaceous and rapidly radiated into several major clades. The origins of most families dated back to the late Cretaceous except for Curtisiaceae and Grubbiaceae which may have diverged in the very early Tertiary. We found that reducing sampling density within families and analyzing partitioned data sets from coding and noncoding cpDNA, 26S rDNA, and combined data sets produced congruent estimation of divergence times, but reducing the number and changing positions of calibration points resulted in very different estimations.